Data reuse
Collection
4 items
•
Updated
filename
stringlengths 13
34
| markdown
stringlengths 0
359k
|
|---|---|
10.1371_journal.pone.0301029
|
RESEARCH ARTICLE
A systematic review and meta-analysis on the
efficacy of vaccination against colibacillosis in
broiler production
Surya PaudelID
C. de Carvalho FerreiraID
7, Alessandra PiccirilloID
6*
1,2☯, Ilias Apostolakos3☯, Ronald Vougat Ngom4,5, Giuditta TilliID
6, Helena
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
1 Department of Infectious Diseases and Public Health, Jockey Club College of Veterinary Medicine and Life
Sciences, City University of Hong Kong, Kowloon, Hong Kong SAR, 2 Clinic for Poultry and Fish Medicine,
Department for Farm Animals and Veterinary Public Health, University of Veterinary Medicine Vienna,
Vienna, Austria, 3 Veterinary Research Institute, Hellenic Agricultural Organization “DIMITRA”, Thessaloniki,
Greece, 4 Department of Animal Production, School of Veterinary Medicine and Sciences, University of
Ngaoundere, Ngaounde´ re´, Cameroon, 5 Veterinary Public Health Institute, Vetsuisse Faculty, University of
Bern, Bern, Switzerland, 6 Department of Comparative Biomedicine and Food Science, University of Padua,
Legnaro, Italy, 7 Flanders Research Institute for Agriculture, Fisheries and Food, Merelbeke, Belgium
☯ These authors contributed equally to this work.
* [email protected]
Abstract
OPEN ACCESS
Citation: Paudel S, Apostolakos I, Vougat Ngom R,
Tilli G, de Carvalho Ferreira HC, Piccirillo A (2024)
A systematic review and meta-analysis on the
efficacy of vaccination against colibacillosis in
broiler production. PLoS ONE 19(3): e0301029.
https://doi.org/10.1371/journal.pone.0301029
Editor: Mohamed Ezzat Abd El-Hack, Zagazig
University Faculty of Agriculture, EGYPT
Received: December 20, 2023
Accepted: March 8, 2024
Published: March 22, 2024
Copyright: © 2024 Paudel et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: All relevant data are
within the manuscript and its Supporting
Information files.
Funding: The author(s) received no specific
funding for this work.
Competing interests: The authors have declared
that no competing interests exist.
Colibacillosis, a disease caused by Escherichia coli in broiler chickens has serious implica-
tions on food safety, security, and economic sustainability. Antibiotics are required for treat-
ing the disease, while vaccination and biosecurity are used for its prevention. This
systematic review and meta-analysis, conducted under the COST Action CA18217—Euro-
pean Network for Optimization of Veterinary Antimicrobial Treatment (ENOVAT), aimed to
assess the efficacy of E. coli vaccination in broiler production and provide evidence-based
recommendations. A comprehensive search of bibliographic databases, including, PubMed,
CAB Abstracts, Web of Science and Agricola, yielded 2,722 articles. Following a defined
protocol, 39 studies were selected for data extraction. Most of the studies were experimental
infection trials, with only three field studies identified, underscoring the need for more field-
based research. The selected studies reported various types of vaccines, including killed
(n = 5), subunit (n = 8), outer membrane vesicles/protein-based (n = 4), live/live-attenuated
(n = 16), and CpG oligodeoxynucleotides (ODN) (n = 6) vaccines. The risk of bias assess-
ment revealed that a significant proportion of studies reporting mortality (92.3%) or feed con-
version ratio (94.8%) as outcomes, had “unclear” regarding bias. The meta-analysis,
focused on live-attenuated and CpG ODN vaccines, demonstrated a significant trend favor-
ing both vaccination types in reducing mortality. However, the review also highlighted the
challenges in reproducing colibacillosis in experimental setups, due to considerable varia-
tion in challenge models involving different routes of infection, predisposing factors, and
challenge doses. This highlights the need for standardizing the challenge model to facilitate
comparisons between studies and ensure consistent evaluation of vaccine candidates.
While progress has been made in the development of E. coli vaccines for broilers, further
research is needed to address concerns such as limited heterologous protection,
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
1 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
practicability for application, evaluation of efficacy in field conditions and adoption of novel
approaches.
Introduction
Poultry meat is expected to hold a substantial share in global meat consumption, accounting
for approximately half of the growth in meat production by 2032 [1]. As the poultry industry
intensifies, ensuring optimum food safety and animal welfare becomes a top priority. How-
ever, bacterial pathogens, such as Escherichia coli pose a major challenge to the poultry indus-
try [2]. E. coli is a gram-negative bacterium in the family of Enterobacteriaceae that normally
resides in the healthy gut of chickens as a commensal [3, 4]. However, infection with patho-
genic strains can lead to colibacillosis, a syndrome that affects chickens of all ages [5]. Coliba-
cillosis in broilers can manifest in various clinical forms, including omphalitis, airsacculitis,
femoral head necrosis and cellulitis, resulting in high condemnation rates and mortality [6–8].
The avian E. coli isolates are highly heterogenous, with pathogenicity likely involving coordi-
nation among several virulence genes, host factors or transfer of genetic elements among E.
coli populations [9, 10]. Thus, understanding the pathogenesis of colibacillosis remains a chal-
lenge [11].
Colibacillosis is primarily treated with antibiotics. However, recent studies have shown the
emergence of antibiotic-resistant strains of E. coli on a global scale [12]. Multidrug resistance
in E. coli has become a concerning threat, even in flocks where no antibiotics are used [13–15].
In Europe, E. coli is identified as one of the most relevant antimicrobial resistant bacterial path-
ogens from poultry [16]. Consequently, there is a growing need to explore preventive strategies
such as vaccination and biosecurity, as opposed to relying solely on antimicrobial treatments.
In the past, numerous studies have reported a range of potential vaccine candidates as summa-
rized in previous reviews and book chapters [5, 17–20]. However, there have been no system-
atic reviews conducted to assess the efficacy of E. coli vaccines in chickens. Such reviews are
essential for providing clear, comprehensive evidence that can inform evidence-based recom-
mendations. Consequently, this study, conducted under the framework of the COST Action
CA18217—European Network for Optimization of Veterinary Antimicrobial Treatment
(ENOVAT), aimed to carry out a systematic review and meta-analysis to understand the cur-
rent evidence regarding the efficacy of vaccination in preventing colibacillosis in broiler
chickens.
Methods
This review was performed according to the Cochrane Handbook for Systematic Reviews of
Interventions [21] and adheres to the structured and reporting guidelines outlined in the Pre-
ferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 [22].
Protocol registration
A systematic review protocol was developed, registered in the University of Padua Research
Archive institutional repository (https://hdl.handle.net/11577/3439974), and published on the
Systematic Reviews for Animals and Food (SYREAF) website (https://syreaf.org).
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
2 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Eligibility criteria
The primary focus of this systematic review was to include controlled trials with natural dis-
ease exposure. However, disease challenge studies and observational studies were also consid-
ered. The studies had to be conducted in broiler production chain (Population) and assess the
protective efficacy of vaccine candidates (Intervention). The vaccine intervention was com-
pared to either an infected and untreated control group or a group that received placebo treat-
ment (Comparator). The selected Outcomes of vaccine efficacy for this review were mortality,
feed conversion ratio (FCR) and condemnation rate at the slaughterhouse. Articles written in
English or Spanish were included, and no restrictions were imposed on publication date or
geographical location of the studies.
Sources of information
To ensure comprehensive coverage of articles, the following bibliographic databases for litera-
ture search were used that provide a high level of article recall in the biomedical field [23]: i)
MEDLINE (via PubMed, https://pubmed.ncbi.nlm.nih.gov/), ii) CAB Abstracts (via Ovid,
https://www.wolterskluwer.com/en/solutions/ovid/cab-abstracts-31), iii) Web of Science
(WoS, http://webofknowledge.com/), and iv) Agricola (via ProQuest, https://www.proquest.
com/). Searches in CAB Abstracts and Agricola were conducted through the University of
Bern (Switzerland), while those in PubMed and WoS through the University of Padova (Italy).
All databases of WoS were used, including WoS core collection, BIOSIS Citation Index,
KCI-Korean Journal Database, Medline, Russian Science Citation Index and SciELO Citation
Index. However, because of their research scopes, certain editions were excluded, namely Arts
& Humanities Citation Index (A&HCI), Conference Proceedings Citation Index-Science
(CPCI-S), Conference Proceedings Citation Index-Social Science & Humanities (CPCI-SSH)
and Social Sciences Citation Index (SSCI). The initial search was conducted in September
2021 and a second search covering the period from September 2021 to October 2023 was per-
formed in October 2023. The databases and search string were the same in both the search
events.
Search strategy and study selection
The search strategy employed a multi-stranded approach, utilizing various combinations of
concepts to ensure comprehensive retrieval of relevant research and achieve high sensitivity
[21]. The concept and the corresponding search strings are presented in Table 1.
Table 1. Bibliographic search strategy to identify studies examining the effect of vaccines against colibacillosis in
broiler chickens.
Major terms
#1 Broilers
#2 Vaccination
#3 Colibacillosis
Key words
chicken* OR poultry* OR gallus OR broiler* OR flock
vaccination* OR vaccine* OR bacterin* OR sub-unit* OR "killed vaccine*" OR "live vaccine*"
OR "autogenous vaccine*"
colibacillosis OR colisepticaemia OR peritonitis OR coli OR Escherichia OR coliform OR
colisepticemia OR coligranuloma OR Hjarre’s OR "air sac disease" OR cellulitis OR
osteomyelitis OR "brittle bone disease" OR salpingitis OR synovitis OR omphalitis OR enteritis
OR "hemorrhagic septicemia" OR "chronic respiratory disease" OR "swollen head syndrome"
OR "venereal colibacillosis" OR "coliform cellulitis" OR "yolk sac infection" OR APEC OR
"pathogenic E. coli" OR "primary infection" OR "secondary infection" OR multifactorial OR
multicausal
#1 AND #2 AND
#3
records screened
https://doi.org/10.1371/journal.pone.0301029.t001
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
3 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
The bibliographic records of the identified articles were downloaded in BibTeX format and
imported to Rayyan [24]. To ensure data accuracy, deduplication process was conducted using
the built-in function of Rayyan. The screening and evaluation of studies were conducted in
two steps. In the first step, at least two independent reviewers screened titles and abstracts.
Any conflicts or disagreements were resolved through discussion or with the involvement of a
third reviewer. To maintain consistency among reviewers, a calibration exercise was first con-
ducted by screening 25 randomly selected studies. The eligibility of studies was evaluated
using a set of questions adapted from a previously published protocol [25] as:
1. Is the study original research assessing the use of vaccine(s) to prevent or control colibacil-
losis in broilers? YES [PASS], NO [EXCLUDE], UNCLEAR [PASS]
2. Does the study include an eligible comparator via a controlled trial, disease challenge study
or observational study? YES [PASS], NO [EXCLUDE], UNCLEAR [PASS]
Studies were excluded only if all reviewers unanimously agreed that the answer to any of
the screening questions was “no”. The studies that passed the first screening step proceeded to
the next, where the full-text articles were retrieved and assessed for eligibility. In the second
phase, the following set of questions was applied:
i. Is a full text of more than 500 words available? YES [PASS], NO [EXCLUDE]
ii. Is a full text available in English and/or Spanish? YES [PASS], NO [EXCLUDE]
iii. Is the population of the study broilers? YES [PASS], NO [EXCLUDE]
iv. Is the intervention of the study the use of vaccine(s) to prevent or control colibacillosis in
broilers? YES [PASS], NO [EXCLUDE]
v. Is at least one of mortality, FCR, or condemnations at slaughter due to colibacillosis the out-
come(s) described? YES [PASS], NO [EXCLUDE]
vi. Is the study design a controlled trial with natural disease exposure or a disease challenge
study or an observational field study? YES [PASS], NO [EXCLUDE]
Data extraction
A Microsoft Excel (2019 version) standardized spreadsheet, developed and validated by the
authors, was used for data extraction. Relevant study characteristics, population type (broilers
or broiler breeders), group size, year of the study, age of the birds during intervention and out-
come assessment, and duration of observation were collected. Detailed data on the interven-
tion were also extracted, including the vaccine type and commercial name, route and dose of
administration, comparator group, unit of population, and total number of birds included. For
studies involving disease challenge, information on challenge day, duration, strain, and admin-
istration route were collected.
Data extraction focused on mortality, FCR and condemnations at slaughterhouse due to
colibacillosis. For mortality, the unit of measurement and assessment period were recorded.
For studies reporting FCR and/or condemnations at slaughter, values such as FCR value and/
or age/weight of slaughtered birds were extracted.
Risk of bias
The risk of bias (RoB) assessment deviated from the original protocol and used instead a
recently reported poultry-specific method [26]. Five domains of bias were evaluated, including
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
4 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
bias from randomization (Domain 1), deviations in interventions (Domain 2), missing out-
come data (Domain 3), measurement of the outcome (Domain 4), and selection of reported
results (Domain 5). Briefly, each domain of bias is composed of several signaling questions
that guide the overall risk of bias for each domain. This overall risk of each domain can then
be reported as ‘‘low risk” ‘‘unclear”, or ‘‘high risk”. At the end, for each included study, a final
overall risk of bias judgment is provided to each outcome based on the results from the five
domains. Therefore, a ‘‘low risk of bias” outcome would result from all five domains being
classified as ‘‘low risk”; ‘‘unclear” would result when either one or two domains for that out-
come have been classified as ‘‘unclear”; and ‘‘high risk of bias” would result from at least three
domains being classified as ‘‘unclear” or if at least one domain is classified as ‘‘high risk”. The
RoB assessment was conducted only for mortality and FCR outcomes.
Data synthesis and statistical analysis
The results of the literature search and selection were reported, and descriptive analysis of
extracted data was done by using Microsoft Excel (version 2019). After data extraction,
included studies were narratively summarized according to the type of vaccines used and
study’s setting. The meta-analysis was performed using Revman version 5.4.1 according to
Higgins et al. [21]. Considering the data of the selected studies, the meta-analysis was per-
formed for two groups of vaccines (i.e. live attenuated vaccine and CpG oligodeoxynucleo-
tides) with “mortality” as outcome. Some differences between included studies (age and route
of inoculation, E. coli strain, dosage, etc.) were not considered during meta-analysis. The com-
parison concerned “infected and vaccinated” and “infected and no vaccinated” groups. Data
used consisted of the number of dead animals per each group. The effect measure for outcome
were odds ratios (with the 95% confidence levels) with a fixed model visualized through Forest
plots. The heterogeneity among studies was evaluated with Cochrane test based on Chi-
Squared. Significant heterogeneity was considered when the I2 value was greater than 50% and
the p-value was less than 0.05. The sources of heterogeneity between studies were not
explored.
Reporting bias assessment
As recommended by the Cochrane methodology [21], funnel plots followed by the Egger’s test
were used to assess publication bias for the outcome ‘‘mortality” using MedCalc version
22.019. This was performed only when sufficient data were available (>10 studies). Certainty
was not assessed.
Results and discussion
Number of eligible studies
The results of the selected studies, based on the inclusion criteria, are summarized in Fig 1. Ini-
tially, 2,722 studies were identified from the selected databases. After removing non-eligible
studies based on the inclusion criteria, 39 studies were deemed suitable for data extraction.
Studies characteristics
In total, 39 studies were selected for data extraction. Various types of E. coli vaccines with dif-
ferent efficacy were reported in the selected studies (Fig 2), including killed (n = 5), subunit
(n = 8), outer membrane vesicles/protein based (n = 4), live/live-attenuated (n = 16) and
CpG-ODN (n = 6) vaccines. Ten studies evaluated the efficacy of a commercially available vac-
cine, while others aimed to assess the suitability of newly reported vaccine candidates.
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
5 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Fig 1. Methodological PRISMA (Preferred Reporting Items for Systematic Review and Meta-analysis) flowchart
for the selection of studies.
https://doi.org/10.1371/journal.pone.0301029.g001
Fig 2. E. coli vaccine types, vaccination routes and challenge routes used to assess vaccination efficacy in broilers.
“n" refers to number of studies.
https://doi.org/10.1371/journal.pone.0301029.g002
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
6 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Mortality was considered as an assessment parameter in all studies. Additionally, FCR was
included in most studies, except for five.
Vaccine types
Characteristics of E. coli vaccine types in broiler chickens are described below.
Killed vaccines
Vaccines using inactivated bacteria can elicit an immune response, mainly through humoral
immunity. These vaccines are considered safe as they do not replicate within the host. Hyper-
immunization of chickens with intravenous injection of a heat-killed E. coli J5 strain was
shown to be effective in preventing mortality and pathological lesions following a challenge
[27] (Table 2, S1 Table). In another study formalin treatment, irradiation and ultrasonication
methods were employed for bacterial inactivation, all of which were shown effective in pre-
venting lesions [28]. Sayed et al. (2021) also showed that formalin-killed E. coli vaccination sig-
nificantly reduced mortality after challenge and could be combined with an avian influenza
vaccine prepared using the same method, which can also induce a higher antibody response in
birds [29]. Recently, autogenous vaccines have been used as a potential solution to address the
heterogeneity of E. coli isolates. However, there is limited evidence regarding their effective-
ness. In a study conducted by Keita et al. (2022), passive immunization using a bivalent forma-
lin-killed autogenous vaccine was found to be effective in reducing mortality when chicks
were challenged with one of two E. coli strains [30]. This indicates strain-specific protection
provided by the vaccine. Additionally, a combination approach involving parent stock vacci-
nation with an autogenous vaccine, along with the supplementation of feed with Enterococcus
faecium DSM 7134 and fructo-oligosaccharides in the progeny, showed benefits in terms of
improving body weight and gut health. However, FCR was not affected by this combination
approach [31]. All of the above-mentioned vaccines have not yet reached to the commercial
market.
Subunit vaccines
These vaccines consist of purified antigenic parts, such as proteins or protein fragments, rather
than the entire pathogen [32]. Table 3 provides a summary of the key characteristics of subunit
E. coli vaccines in broiler chickens. Details of each study are provided in S1 Table. Vandemaele
et al. (2006) investigated the impact of immunization with the biologically active lectin domain
Table 2. Important characteristics of killed vaccines against colibacillosis in broiler chickens.
Reference
Abdul-Aziz & El-
Sukhon, 1998 [27]
Ibrahim et al., 1997
[28]
Day of
vaccination
5, 14 & 20
14
Sayed et al., 2021 [29]
21 & 42
Keita et al., 2022 [30]
20 & 22 weeks
Fuhrmann et al., 2022
[31]
12 & 17 weeks
Route of
vaccination
IV
IM
SC
IM1
IM1
Dose and route of
challenge
0.2 ml of 6x108
CFU/ml (IV)
108 CFU/bird
0.2 ml of 107 CFU
(IM)
0.1 ml of 3x108
CFU/ml (SC)2
0.6 ml of 3.2x107
CFU/ml (oral)2
Important findings
Chickens hyperimmunized with E. coli J5 showed protection based on mortality,
clinical signs and pathological lesions.
Formalin-killed, irradiated and ultrasonicated E. coli induced protection.
Formalin-killed E. coli prevented mortality from experimental E. coli infection
and can be combined with inactivated avian influenza vaccine.
Breeders received a bivalent autogenous vaccine; passive immunization was
effective in chicks against challenges with one out two E. coli stains only
Passive immunization together with administration of pre-and probiotics have
beneficial effects on body weight and gut health.
IV: intravenous, IM: intramuscular, SC: subcutaneous, CFU: colony forming unit; 1vaccination of breeders for passive immunization; 2 challenge of progenies
https://doi.org/10.1371/journal.pone.0301029.t002
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
7 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Table 3. Important characteristics of subunit vaccines against colibacillosis in broiler chickens.
Reference
Day of vaccination Route of
Dose and route of challenge
Important findings
Vandemaele
et al., 2005 [34]
10 (single) or 10 &
30 (with booster)
vaccination
IM or IN
Vandemaele
et al., 2006 [33]
10
Lynne et al., 2012
[35]
14
Chaudhari et al.,
2013 [36]
1 (single) or 1 & 14
(with booster)
IM
IM
Oral
10 ml/group, 3x1010 CFU/ml; prior
infection with NDV (nebulization)
10 ml/group, 2.7x1010 CFU/ml (prior
infection with NDV; nebulization) or 0.2
ml 104 (intra-airsac)
0.1 ml of 107 CFU (intra-airsac)
50 μl of 107 CFU (intra-airsac)
Chaudhari & Lee,
2013 [37]
1
Oral
0.1 ml of 106 CFU (intra-airsac)
Ma et al., 2018
[38]
Dissanayake
et al., 2010 [39]
1–3 & 14–16
Oral
5x1011 CFU (Oral)
7 & 21
IM
106 CFU (subcutaneous)
Tuntufye et al.,
2012 [40]
10 & 24
Intranasally &
IM
2x106 CFU (intra airsac)
IM: intramuscular, IN: intra nasal, CFU: colony forming unit
https://doi.org/10.1371/journal.pone.0301029.t003
Immunization with sugar-binding domain of FimH (FimH156)
effectively induced high levels of adhesion-inhibiting antibodies
but did not provide protection against APEC O78 infection.
Immunization with the biologically active lectin domain of
PapGII could effectively induce high levels of adhesion-
inhibiting antibodies but did not provide protection against
APEC O78 infection delivered via coarse spray or via intra-air
sac challenge.
The Iss antigen provided significant protection against
challenges with three different APEC strains (O1, O2 and O78).
Prime and boost vaccination with an attenuated Salmonella
strain carrying P-fimbriae (papA, papG), aerobactin receptor
(iutA) and CS31A surface antigen (clpG) genes of E. coli induced
immune response and provided protection against E. coli
challenge.
Coadministration of live attenuated Salmonella strain expressing
the heat-liable toxin of E. coli B subunit (LTB) increased the
efficacy of the Salmonella-delivered APEC vaccine developed by
Chaudhari et al. (2013).
Immunization of birds with a recombinant Lactobacillus
saerimneri expressing FimA and OmpC antigen of O78 APEC
provided protection.
Liposome-encapsulated mixture of rough LPSs significantly
lowered lesion scores and increased body weight but no
difference was observed in mortality.
Four ferri-siderophore receptors (FuhE, FepA, IroN, IutA) were
expressed in live or bacterial ghost cells; none of the two
recombinants were protective.
of PapGII on the avian immune response to APEC O78 challenge [33]. The results showed
that while immunization effectively stimulated the production of adhesion-inhibiting antibod-
ies, it did not provide protection against APEC O78 infection through coarse spray or intra-air
sac challenge. Similar results were observed when the sugar-binding domain of FimH
(FimH156) was used [34]. However, Lynne et al. (2012) demonstrated that vaccinating birds
with increased serum survival gene (iss) fusion proteins provoked the serum and mucosal anti-
body response and consequently resulted in broad protection against bacterial challenges with
three different APEC strains: O1, O2 and O78 [35]. Similarly, a modified strain of Salmonella
carrying multiple genes from E. coli, including P-fimbriae (papA, papG), aerobactin receptor
(iutA) and CS31A surface antigen (clpG), elicited mucosal and systemic antibody responses,
and stimulated lymphocytic proliferation [36]. Although a single vaccination with this attenu-
ated Salmonella strain only provided partial protection against E. coli challenge, repeated vacci-
nation significantly enhanced the protective response. Co-administration of this vaccine
candidate with a live attenuated Salmonella expressing the heat-labile toxin of E. coli B subunit
(LTB) as an adjuvant proved to be more effective in reducing mortality and morbidity rates in
challenged birds [37]. In a separate study, a probiotic bacterium called Lactobacillus saerimneri
was used as a delivery system to create a recombinant vaccine that expressed fimbrial subunit
A (FimA) and outer membrane protein C (OmpC) of O78 APEC. Oral administration of the
recombinant L. saerimneri effectively induced an antigen-specific immune response and pro-
vided protection, as 70% of vaccinated birds survived while 100% mortality was observed in
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
8 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
the non-vaccinated challenge control group [38]. Additionally, a non-adjuvanted liposome-
encapsulated mixture of rough LPSs exhibited a positive dose-dependent effect, especially in
terms of antibody level in birds. Immunization with the highest dose (5μg) resulted in lower
lesion scores and increased body weight, although the mortality rate did not show a significant
difference [39].
Iron uptake system genes play a crucial role in the virulence mechanism of APEC. Four
ferri-siderophore receptors namely FuhE, FepA, IroN, IutA were expressed in recombinant
live or bacterial ghost cells [40]. However, despite increased IgG titers in birds, neither the
intranasal administration of recombinant live E. coli nor the intramuscular infection of recom-
binant ghost cells was able to reduce mortality and lesion scores, leading to the conclusion that
both vaccine candidates were non-protective.
Outer membrane vesicles/proteins-based vaccines
Outer membrane vesicles (OMVs) are naturally derived spherical nanovesicles originating
from the bacterial outer membrane which contains various bacterial components, such as lipo-
polysaccharides, proteins, and other antigens [41]. Their protective efficacy was evaluated in
broiler chickens as shown in Table 4 and S1 Table. Immunization of birds with a nanosized
OMV-based vaccine derived from APEC O2 demonstrated no adverse effects. Moreover, it
significantly increased the survival rate, reduced bacterial loads, and suppressed the produc-
tion of proinflammatory cytokines [42]. Immunologically, the vaccine primarily stimulated
antigen-specific antibody responses and IFN-γ mediated immune responses in the host. Tak-
ing a step further, a combination of multi-serogroup OMVs derived from O1, O2 and O78 E.
coli strains induced a robust non-specific and antigen specific immune responses. This was
evident from the production of IgG antibodies specific to APEC antigens and resulted in a 90–
100% increase in protection against challenges with APEC O1, O2 or O78 strains compared to
the control group [43]. It is difficult to attribute the observed protection solely to specific pro-
teins or polysaccharides within the OMVs due to their complex composition.
To enhance vaccine uptake and improve pharmacokinetic and pharmacodynamic proper-
ties, nanotechnology-based vaccines are beneficial. These vaccines are designed to boost the
immune response by providing antigenic targets in a way that mimics natural infection,
improving stability and targeting specific immune cells [44, 45]. Mohammed et al. (2021)
investigated the potential of chitosan nanoparticles in enhancing the immune response of
chickens after vaccination with the outer membrane proteins (OMPs) and flagellar antigens
from O1 and O78 serogroups [46]. The study utilized two types of chitosan nanoparticles,
Table 4. Important characteristics of outer membrane vesicles/protein-based vaccines against colibacillosis in broiler chickens.
Reference
Day of
vaccination
Route of
vaccination
Hu et al., 2020 [43]
7 & 14
IM
Hu et al., 2020 [42]
7 & 14 & 21
IM
Mohammed et al.,
2020 [46]
Abd El-Aziz et al.,
2022 [47]
21
14
SC
SC
Dose and route of
challenge
5x108 CFU (intra-
airsac)
5x108 CFU (intra-
tracheal)
107 CFU (IM)
107 CFU (IM)
IM: intramuscular, SC: sub cutaneous, CFU: colony forming unit
https://doi.org/10.1371/journal.pone.0301029.t004
Important findings
Vaccination with nanosized OMVs had no side effects and efficiently protected chicks
against homologous infection with APEC O2. It provoked antibody and IFN-γ
mediated immune responses.
Combined OMVs from O1, O2 and O78 strains provided robust and broad protection
against E. coli challenges with all three strains.
Addition of chitosan and ascorbate chitosan nanoparticles improved the immune
response induced by outer membrane proteins and flagellin.
Chitosan loaded nanoparticles with Montanide adjuvant enhanced immunity for a
longer time period.
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
9 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
namely the characterized chitosan (CS) and ascorbate chitosan (AsCS), in both loaded and
encapsulated forms. The results demonstrated that both forms of chitosan nanoparticles
improved the immune response, in terms of antibody production in chickens and provided
protection against infections induced by E. coli O1 and O78, compared to the control group.
Subsequent research by Abd El-Aziz et al. (2022) corroborated these findings and further dem-
onstrated that the addition of Montanide as an adjuvant to the chitosan nanoparticles pro-
longed the humoral and cell-mediated immunological responses, thereby enhancing
immunity for an extended period [47].
Live and live attenuated vaccines
Live and live attenuated vaccines have advantages for mass application, as they can be deliv-
ered as spray or via drinking water. Most of the studies in this review evaluated the efficacy of
these vaccines (Table 5, S1 Table). A study by Frommer et al. (1994) found that a non-patho-
genic piliated E. coli strain provided broad protection against colibacillosis-induced mortality
caused by O1:K1, O2:K1 and O78:K80 [48]. Immunization at 14 or 21 days of age was more
effective than at an early age (1 or 7 days) and drinking water or intramuscular administration
showed better efficacy than the spray method. Kariyawasam et al. (2002) demonstrated that
administering a live E. coli strain of O78 serotype via aerosol route at 18 days of age reduced
pathological lesions and systemic bacterial colonization when challenged with the same strain
[49].
Advancements in understanding genetic characteristics of E. coli led to the development of
live attenuated vaccine candidates by targeting essential genes required for multiplication in
the host. Peighambari et al. (2002) tested the efficacy of double mutants created by deletion of
cya and crp genes in O2 and O78 E. coli strains [50]. The mutant O2 strain provided moderate
protection against air sac lesions when administered via spray, while the mutant O78 strain
was ineffective. It was also observed that antibody response was not stimulated in vaccinated
birds, indicating the importance of innate or adaptive immunity for protection against coliba-
cillosis. Other mutants with galE, purA or aroA deletions showed similar immunogenicity and
serogroup-specific protection but they did not provide effective cross-protection against heter-
ologous challenge [51].
Currently, there are two licensed live attenuated vaccines against E. coli infection. The crp
deletion mutant of E. coli-O78 (Nisseiken Co., Ltd., Tokyo, Japan) is marketed in Japan and is
recommended to be administered via fine spray (particle size <20 μm) in day-old chickens.
The vaccine has shown effectiveness in reducing lesions following the challenge with homo-
and heterologous strains [52, 53], although it may be ineffective against heterologous challenge
based on mortality pattern, clinical signs, pathology and bacterial re-isolation [53]. The second
live attenuated vaccine is developed by deleting aroA gene of E. coli-O78 strain (Poulvac1, E.
coli, Zoetis), and is usually administered via coarse spray in day-old chicks. In a field trial, the
Poulvac vaccine did not affect the weight gain in broiler chickens [54]. In experimental condi-
tions, several studies have suggested its efficacy against homologous challenge in reducing coli-
bacillosis-associated pathological lesions [54–58]. Oral vaccination of birds at day 5 of age
reduced morbidity [59] but drinking water application was found to be ineffective in inducing
protection [56]. The findings regarding heterologous protection were not consistent among
studies with some showing effectiveness against certain strains [54, 60], but not others [56, 57,
60]. The efficacy of the vaccine against homologous challenge was enhanced by supplementa-
tion of probiotics Enterococcus faecalis [58] and pre-treatment with Lincospectin improved the
vaccine’s response [61]. However, the immune response elicited by the live attenuated vaccine
was reported to be interfered by the prior application of ceftiofur sodium antibiotic in layer
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
10 / 22
PLOS ONE Table 5. Important characteristics of live and live attenuated vaccines against colibacillosis in broiler chickens.
Reference
Day of vaccination
Route of
vaccination
Dose and route of
challenge
Important findings
Efficay of vaccination against colibacillosis in broilers
Live vaccine
Frommer et al.,
1994 [48]
1, 7, 14 or 21
IM, spray or per os
1x108 CFU (IM)
Kariyawasam et al.,
2002 [49]
18
Live attenuated vaccine
Peighambari et al.,
2002 [50]
14 or 10 and 14
Aerosol
108 CFU (Intra
airsac); together with
IBV
Coase spray
0.1x109 CFU (prior
infection with IBV)
1 and 14
Coarse spray
Kariyawasam et al.,
2004 [51]
Asaad et al., 2019
[52]
Abd El-Mawgoud
et al., 2020 [53]
Sadeghi et al., 2018
[54]
Galal et al., 2018
[55]
Rawiwet et al., 2009
[59]
Mohammed et al.,
2016 [56]
1
1
1
1
5
1
Fine spray (crp-
deletion mutant)
Eye drop (aroA
deletion mutant)
Fine spray
100 ml/group of 109
CFU/ml (aerosol)
6x108 CFU (IT)
0.5 ml of 108 CFU/ml
(SC)
Coarse spray
108 (IT)
Coarse spray
109 CFU (IT)
Oral
Coarse spray or
drinking water
0.5 ml of 1.2x109
CFU/ml (IT)
6x108 CFU (IT)
Gharib et al., 2017
[57]
1 or 1 and 14
Elbestawy et al.,
2021 [60]
1
Tarabees et al., 2019
[58]
1 and 15
Coarse spray
9x108 CFU (IT)
Coarse spray
Coarse spray
0.5 ml of 1.2x108 CFU
(IT)
0.5 ml of 1x108 CFU
(oral)
Galal et al., 2021
[61]
1 or 7
Coarse spray
0.1 ml of 109 CFU/ml
(IT)
Li et al., 2017 [63]
1 day and 12 weeks
Coarse spray
0.1 ml of 5x106 CFU/
ml (intrauterine)
Vaccination at the age of 14 or 21, but not at 1 or 7 days, via IM
or per os elicited protection; mortality was reduced after
challenge with O1, O2 and O78 E. coli challenges; spray
vaccination provided inadequate protection.
Vaccination reduced pathological lesions and systemic bacterial
colonization after homologous challenge.
Double mutant was created deleting cya and crp genes of O2 and
o78 strains; moderate protection was observed with mutant O2
strain as it reduced air sac lesions but the mutant O78 was not
effective.
Mutants of galE, purA or aroA from O78 strain provided
homologous but not heterologous (O2) protection.
Both vaccines were effective to minimize the pathological lesions
following homo-and heterologous (O1) challenges.
The crp deletion mutant vaccine was efficacious to reduce
mortality and bacterial colonization after homologous challenge
but was not effective against heterologous challenge (O125).
aroA deletion mutant (Poulvac) reduced clinical signs and lesions
due to O78 and untypeable E. coli strains; no difference was
observed in mortality.
aroA deletion mutant (Poulvac) provided protection against the
homologous challenge; the vaccine did not interfere with humoral
immune response induced by other vaccines such as AI, NDV,
IBV or IBD.
aroA deletion mutant (Poulvac) reduced morbidity (pathological
lesions) but no difference was seen following homologous
challenge.
Spray vaccination of aroA deletion mutant (Poulvac) led to
significant reduction in pathological lesions but drinking water
application was not effective; homologous protection was
observed but not the heterologous protection against O1
challenge
aroA deletion mutant (Poulvac) was effective against O78 but not
against O125 challenge, protection was associated with cell
mediated immunity.
aroA deletion mutant (Poulvac) provided protection against O27
and O8 assessed with FCR, mortality, lesions, clinical signs and
bacterial re-isolation; protection against O115 was not significant.
aroA deletion mutant (Poulvac) decreased the mortality rate and
bacterial colonization after homologous challenge; vaccine
response was improved by supplementation of probiotics
Enterococcus faecalis.
aroA deletion mutant (Poulvac) given at 7 days in birds with prior
treatment with lincospectin 100 for 3 days was the most effective
to prevent mortality and loss of performance due to challenge
with O78 strain.
aroA deletion mutant (Poulvac) alone was not protective; Poulvac
followed by autogenous vaccine delayed the onset of clinical signs
for 3–4 days but no signs of protection against homo-and
heterologous challenges.
(Continued )
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
11 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Table 5. (Continued)
Reference
Day of vaccination
Route of
vaccination
Dose and route of
challenge
Important findings
Sˇenk et al., 2022
[73]
Poulvac (12 & 20 weeks) with/
without autogenous vaccine
(18 weeks)
Poulvac: Spray
ND
Autogenous: IM
Vaccinating birds with both commercial live-attenuated
(Poulvac) and autogenous vaccines showed some benefits
compared to using only the live-attenuated vaccine.
IM: intramuscular, CFU: colony forming unit, IT: intra tracheal, ND: not done
https://doi.org/10.1371/journal.pone.0301029.t005
birds [62]. Recently, Li et al. (2017) reported that Poulvac administered alone at 1 and 12
weeks did not protect birds against intrauterine challenge despite high antibody titers [63].
However, when the live attenuated vaccine was followed by an autogenous vaccine, the onset
of disease was delayed, but there was no evidence of protection against homologous or heterol-
ogous challenges.
CpG-ODN vaccines
CpG oligodeoxynucleotides (CpG ODN) are synthetic DNA molecules that contain specific
patterns of cytosine and guanine bases (CpG) recognized by the immune system. They have
been used as vaccine adjuvants to enhance the immune response to antigens, including viral or
bacterial protein [64]. Several studies investigated the effectiveness of CpG ODN vaccines in
reducing mortality caused by colibacillosis (Table 6, S1 Table). Gunawardana et al. (2019)
explored the use of CpG ODN administered in ovo to stimulate the immune system of newly
hatched chicks and protect them against subcutaneous (SC) bacterial challenge with E. coli ser-
ogroup O2 [65]. The study revealed that the administration of synthetic CpG-ODN in freshly
hatched chicks led to a rapid increase in immune cells, such as macrophages and dendritic
cells, as well as cytokine responses in spleen and lungs. The authors also observed enhanced
protection against bacterial challenge in the chickens treated with synthetic CpG-ODN, as
indicated by reduced bacterial loads in various tissues and increased survival rates. Similar
findings were reported in the studies conducted by Taghavi et al. (2009) and Gomis et al.
(2004), which employed similar study designs involving in ovo vaccination, SC challenge with
E. coli O2 at comparable doses, and similar observation period for mortality [66, 67]. Two
additional studies examined the effects of CpG-ODN vaccines in newly hatched chicks but
with different study designs. Allan et al. (2018) also employed in ovo delivery, however the
challenge with E. coli O2 was done via the intranavel route and at a much lower dosage (25
CFUs vs 105 CFUs in the aforementioned studies) [68]. Nevertheless, the authors observed
increased survival rates in chicks compared to the control group. In a study by Sarfraz et al.
(2022) using a similar vaccination route and challenge model, the effectiveness of different
innate immune stimulants and their combination was compared [69]. The results showed that
the in ovo administration of CpG-ODN in conjunction with polyinosinic-polycytidylic acid
was the most efficient in protecting chicks when they were challenged via intranasal route. In
another study, one-day-old chicks received intrapulmonary delivery of CpG-ODN followed by
SC challenge with E. coli O2, resulting in approximately half the relative risk of mortality com-
pared to birds that received saline [70].
Field studies
Only a limited number of studies (Table 7, S1 Table) examined the efficacy of E. coli vaccines
in field settings compared to the experimental studies mentioned above. The first two studies
[71, 72] were not considered for data extraction, as they did not meet the selection criteria.
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
12 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Table 6. Important characteristics of CpG-ODN vaccines against colibacillosis in broiler chickens.
Reference
CpG-ODN
Day of
vaccination
Route of
vaccination
Dose and route of
challenge
Important findings
Gomis et al., 2004 [67]
in ovo
in ovo
105 CFU (SC)
Taghavi et al., 2009
[66]
in ovo
in ovo
104.5 CFU (SC)
Goonewardene et al.,
2017 [70]
1
intrapulmonary
104.5 CFU (SC)
Chickens treated with synthetic CpG-ODN showed enhanced protection against
bacterial challenge, indicated by reduced bacterial loads in various tissues and
increased survival rates.
Chickens treated with synthetic CpG-ODN showed enhanced protection against
bacterial challenge, indicated by reduced bacterial loads in various tissues and
increased survival rates.
SC challenge with E. coli O2, resulting in about half of the relative risk of mortality
as did the birds that received saline
Allan et al., 2018 [68]
in ovo
in ovo
25 CFU
(intranavel)
Challenge with E. coli O2 was done via the intranavel route and at a much lower
dosage; increased survival rates of chicks in their experiments compared to the
control group.
Gunawardana et al.,
2019 [65]
Sarfraz et al., 2022
[69]
in ovo
in ovo
104.5 CFU (SC)
In rapid increase of immune cells such as macrophages and dendritic cells;
enhanced protection against bacterial challenge, indicated by reduced bacterial
loads in various tissues and increased survival rates.
in ovo
in ovo
25–30 CFU
(intranavel)
Coadministration of CpG (10μg/embryo) and poly I:C 15μg/embryo provided
100% protection against experimental yolk sac infection.
SC: subcutaneous, CFU: colony forming unit
https://doi.org/10.1371/journal.pone.0301029.t006
One of them evaluated the effectiveness of a commercially available inactivated subunit vac-
cine, which contains E. coli fimbrial antigen and flagellar toxin (Nobilis1, MSD Animal
Health). The vaccine was administered intramuscularly to commercial broiler breeders [71].
The results showed that the vaccinated flocks experienced lower mortality potentially associ-
ated with natural E. coli infection. However, there were no significant differences observed in
terms of first week mortality in chicks, slaughterhouse condemnation rates and FCR between
birds from vaccinated or non-vaccinated breeder flocks. Another study investigated the effi-
cacy of the live attenuated Poulvac1 E. coli vaccine, which was administered in day old chicks
as recommended [72]. The findings demonstrated that colibacillosis-like lesions were less fre-
quent in vaccinated flocks compared to non-vaccinated flocks. However, no differences were
observed in FCR between the two groups. Another study showed that colibacillosis-related
lesions were observed less frequently in the flock of birds vaccinated with both live attenuated
Poulvac1 and autogenous vaccines compared to the group vaccinated only with live attenu-
ated vaccine, indicating some benefits of combining both vaccines in the field [73].
Risk of bias
In the evaluation of 39 papers reporting mortality data, the overall RoB revealed ‘‘some con-
cerns” for the majority (n = 36, 92.3%) and ‘‘high risk” for a small fraction (n = 3, 7.7%)
Table 7. Important characteristics of field studies vaccination against colibacillosis in broiler chickens.
Reference
Gregersen et al.,
2010 [71]
Mombarg et al.,
2014 [72]
Age of
vaccination
Route of
vaccination
Dose and route of
challenge
Important findings
12 & 18 weeks
IM
day old
Spray
ND
ND
Vaccinated breeder flock experienced less mortality due to E. coli natural infection but
the vaccination no beneficial impact on the first week mortality of chicks.
Colibacillosis associated lesions recorded in slaughterhouse were less frequent in
vaccinated flocks compared to non-vaccinated flocks.
IM: intramuscular, ND: not done
https://doi.org/10.1371/journal.pone.0301029.t007
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
13 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Fig 3. Risk of Bias (RoB) in the E. coli vaccination studies that reported mortality (A) or feed conversion ratio (B) as
one of the assessment parameters.
https://doi.org/10.1371/journal.pone.0301029.g003
(Fig 3A). When examining seven studies that included FCR data, the overall RoB was evalu-
ated as ‘‘unclear’ (Fig 3B). For mortality outcome, one study (2.6%) exhibited “high risk” of
bias, one (2.6%) ‘‘low risk”, and 37 (94.8%) were assessed to have ‘‘unclear” (Fig 3B). The over-
all risk of bias of domains 3 to 5 was assessed as ‘‘low” for FCR. In both mortality and FCR,
domains 1 and 2 emerged as the primary sources of bias. The overall bias arising from the ran-
domization process (domain 1) recorded is mainly related to the lack of information concern-
ing concealed allocation sequence of animals in the groups. For example, for mortality as
outcome, only seven studies (n = 17.9%) provided this information. The concern with the
result of the domain 2 (bias due to deviation from the intended intervention) is due to the
absence of information concerning the awareness or not of animal caregivers/researchers
about the assigned interventions and whether there were deviations from the intended inter-
vention that arose because of the trial. Again, only one study (n = 2.6%) provided these details
in the papers with mortality as outcome.
Meta-analysis
For the live attenuated vaccine, a total of twelve studies were included in the meta-analysis (Fig
4). All studies were performed after the year 2000 and the majority (83.3%) had an overall RoB
assessed as “unclear”. When considering the effects on mortality, the comparison between the
non-vaccinated and vaccinated groups showed a significant (P < 0.00001) trend favoring vac-
cination with a pooled odds ratio of 0.30 (95% CI: 0.19–0.48). A minimal but non-significant
level of heterogeneity (P = 0.21; I2 = 24%) among studies was recorded. Due to the absence of a
significant heterogeneity among included studies, the subgroup and meta-regression analysis
were not performed. As presented in Fig 5, the six studies included in the meta-analysis of the
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
14 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Fig 4. Forest plot of live attenuated vaccine efficacy considering the “mortality” assessment parameter.
https://doi.org/10.1371/journal.pone.0301029.g004
CpG-ODN vaccine had an overall RoB assessed as ‘unclear’ and a high level of heterogeneity
(P < 0.00001; I2 = 91%). The pooled odds ratio for mortality was 0.64 (95% CI: 0.45–0.90).
Due to the limited number of studies included, the sources of heterogeneity were not assessed.
The reporting bias was assessed only for the live attenuated vaccine efficacy considering the
“mortality” as outcome because more than ten studies were included in the meta-analysis. The
funnel plot (Fig 6) and the Egger’s test results (Intercept = 0.43, 95% CI [-1.94 to 2.80],
P = 0.69) showed a symmetry of the studies and a non-significant regression test, respectively,
indicating an absence of publication bias (studies with no significant results) and validating
the analysis as reasonable and reliable.
Outlook and conclusion
Colibacillosis in broiler chickens poses challenges for animal health and welfare, with serious
economic consequences, impacting food safety and security, which can have a clear effect on
consumers’ wellbeing and livelihoods. This systematic review examines the efficacy of various
vaccines in preventing colibacillosis in broilers, revealing that, while some vaccine candidates
have shown promising results, challenges and limitations remain that need to be addressed.
Killed and subunit vaccines, while being safe due to only exposing animals to fragments of
the pathogen, have limited range of protection and require injections in birds. This ultimately
Fig 5. Forest plot of CpG oligodeoxynucleotides (CpG-ODN) vaccine efficacy considering the “mortality”
assessment parameter.
https://doi.org/10.1371/journal.pone.0301029.g005
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
15 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Fig 6. Funnel plot of live attenuated vaccine efficacy considering the “mortality” assessment parameter.
https://doi.org/10.1371/journal.pone.0301029.g006
raises concerns about the practicality and cost-effectiveness in commercial broiler production.
On the other hand, even though live attenuated vaccines can be mass-applied through practical
routes, such as drinking water or spray, rigorous safety assessments and investigations into
potential long-term effects are still necessary.
In this study, meta-analysis was possible only for live-attenuated and CpG ODN vaccines,
reflecting great heterogenicity among studies. While the analysis demonstrated a significant
trend favoring these vaccination types in reducing mortality, there is a significant variation in
challenge models used in vaccine research for E. coli infections, involving various routes of
infection, predisposing factors, and a wide range of challenge doses. It shows the difficulty in
reproducing the disease experimentally. Nevertheless, standardizing the challenge model is
essential for consistent evaluation of vaccine candidates and comparison between studies [74].
Therefore, expanding the portfolio of E. coli vaccines, considering practical feasibility and
serotype-independent protection, as well as establishing a robust infection model, are crucial.
Field studies offer insights into the real-world vaccine effectiveness, yet the limited number of
studies found in this systematic review highlights the need for more research, namely to evalu-
ate vaccine efficacy in field conditions and assess additional parameters such as pathological
consequences, economic impact and long-term protection.
Developing an effective vaccine against colibacillosis in chickens is complex due to the high
heterogenicity of E. coli isolates, elusive disease mechanisms, and absence of definitive markers
for pathogenic isolates [17, 20]. This complexity is evident in the limited number of vaccines
reaching the commercial market, with conflicting reports about their effectiveness [53, 60].
Future vaccine development requires a multi-dimensional approach, focusing on identify-
ing conserved antigens that confer broad protection across different APEC serotypes or incor-
porate antigens that confer broad protection across different APEC serogroups. Multivalent
vaccines targeting multiple serogroups or incorporating diverse antigens may offer enhanced
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
16 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
efficacy and broader coverage. Exploring innovative technology, such as irradiation [75] or
glycoconjugate vaccines [76], may hold promise for improving vaccine delivery and bird pro-
tection against colibacillosis. Increased investment in research and development along with
collaborative efforts between academic, industry and regulatory agencies, can expedite the
translation of promising vaccine candidates into commercial products. Public-private partner-
ships and funding initiative can also incentivize vaccine development for diseases, such as coli-
bacillosis, with significant impacts on animal health and economic sustainability. Despite
providing valuable insights, this review has limitations. The focus was primarily on broiler pro-
duction chain, excluding vaccine types and studies related to the E. coli vaccination in layer
birds. Subgroup analysis was challenging due to variations in challenge models and experi-
mental designs. The meta-analysis results should be interpreted with caution considering the
diversity of influencing factors and the reduced number of studies considered. Additionally,
studies that did not include mortality, FCR and condemnation at slaughter as assessment crite-
ria were excluded, although reproducing colibacillosis in experimentally infected birds is chal-
lenging, leading to exclusion of some vaccination studies in broilers.
In conclusion, while significant progress has been made in the development of E. coli vac-
cines for broilers, challenges persist. The benefits of vaccination have been demonstrated in
several studies, with meta-analysis showing a positive effect of live attenuated and CpG-ODN
vaccination in reducing mortality. However, further research is needed to enhance under-
standing of effective vaccines against colibacillosis, considering factors such as antigen selec-
tion, adjuvant choice, delivery method, and use of novel approaches.
Supporting information
S1 Checklist. PRISMA 2020 checklist. Checklist for the Preferred Reporting Items for Sys-
tematic Reviews and Meta-Analyses workflow.
(DOCX)
S1 Table. Extracted metadata of studies included in the analyses of this paper.
(XLSX)
Acknowledgments
This study was carried out within the COST Action CA18217 European Network for Optimi-
zation of Veterinary Antimicrobial Treatment (ENOVAT), supported by the European Coop-
eration in Science and Technology (COST) (www.enovat.eu and www.cost.eu). The authors
wish to thank all participants supporting the COST Action CA18217—ENOVAT, particularly
the members of the Working Group 4 (Antimicrobial treatment guidelines; https://enovat.eu/
wg4/). The authors would also like to thank the whole ENOVAT Drafting Group “Veterinary
guidelines on antimicrobial use in poultry colibacillosis” for the wise support and valuable
expertise provided. A Swiss Government Excellent Scholarship was granted to RVN for the
years 2020–21 when he started working in the Drafting Group.
Author Contributions
Conceptualization: Surya Paudel, Ilias Apostolakos, Alessandra Piccirillo.
Data curation: Surya Paudel, Ilias Apostolakos, Ronald Vougat Ngom, Giuditta Tilli, Helena
C. de Carvalho Ferreira.
Formal analysis: Surya Paudel, Ilias Apostolakos, Ronald Vougat Ngom, Helena C. de Car-
valho Ferreira.
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
17 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
Investigation: Surya Paudel, Ilias Apostolakos, Ronald Vougat Ngom, Alessandra Piccirillo.
Methodology: Ronald Vougat Ngom, Alessandra Piccirillo.
Software: Ilias Apostolakos, Ronald Vougat Ngom.
Supervision: Helena C. de Carvalho Ferreira, Alessandra Piccirillo.
Validation: Helena C. de Carvalho Ferreira, Alessandra Piccirillo.
Visualization: Surya Paudel, Ilias Apostolakos.
Writing – original draft: Surya Paudel, Ilias Apostolakos.
Writing – review & editing: Surya Paudel, Ilias Apostolakos, Ronald Vougat Ngom, Giuditta
Tilli, Helena C. de Carvalho Ferreira, Alessandra Piccirillo.
References
1. OECD/FAO. OECD-FAO Agricultural Outlook 2023–2032. Paris 2023. https://doi.org/10.1787/
08801ab7-en.
2. Paudel S, Hess M, Hess C. Advances in understanding bacterial diseases in poultry: challenges and
perspectives In: Wit Sd, editor. Optimising poultry flock health First ed. Philadelphia, USA: Burleigh
Dodds Science Publishing 2022. p. 35–66.
3. Abdelhamid MK, Quijada NM, Dzieciol M, Hatfaludi T, Bilic I, Selberherr E, et al. Co-infection of Chicken
Layers With Histomonas meleagridis and Avian Pathogenic Escherichia coli Is Associated With Dysbio-
sis, Cecal Colonization and Translocation of the Bacteria From the Gut Lumen. Front Microbiol. 2020;
11:586437. Epub 20201030. https://doi.org/10.3389/fmicb.2020.586437 PMID: 33193238; PubMed
Central PMCID: PMC7661551.
4.
Thomson NM, Gilroy R, Getino M, Foster-Nyarko E, van Vliet AHM, La Ragione RM, et al. Remarkable
genomic diversity among Escherichia isolates recovered from healthy chickens. PeerJ. 2022; 10:
e12935. Epub 20220301. https://doi.org/10.7717/peerj.12935 PMID: 35251780; PubMed Central
PMCID: PMC8896058.
5. Nolan LK, Vaillancourt JP, Barbieri NL, Logue CM. Colibacillosis. In: (Editor-in-Chief) DES, editor. Dis-
eases of Poultry. 14th Edition ed: John Wiley & Sons, Inc.; 2020. p. 770–830.
6.
Landman WJ, van Eck JH. The incidence and economic impact of the Escherichia coli peritonitis syn-
drome in Dutch poultry farming. Avian Pathol. 2015; 44(5):370–8. https://doi.org/10.1080/03079457.
2015.1060584 PMID: 26083823.
7. Apostolakos I, Laconi A, Mughini-Gras L, Yapicier OS, Piccirillo A. Occurrence of Colibacillosis in Broil-
ers and Its Relationship With Avian Pathogenic Escherichia coli (APEC) Population Structure and
Molecular Characteristics. Front Vet Sci. 2021; 8:737720. Epub 20210908. https://doi.org/10.3389/
fvets.2021.737720 PMID: 34568479; PubMed Central PMCID: PMC8456121.
8. Gaussmann B, Hess C, Grafl B, Kovacs M, Troxler S, Stessl B, et al. Escherichia coli isolates from fem-
oral bone marrow of broilers exhibit diverse pheno- and genotypic characteristics that do not correlate
with macroscopic lesions of bacterial chondronecrosis with osteomyelitis. Avian Pathol. 2018; 47
(3):271–80. Epub 20180314. https://doi.org/10.1080/03079457.2018.1440065 PMID: 29451003.
9. Mageiros L, Meric G, Bayliss SC, Pensar J, Pascoe B, Mourkas E, et al. Genome evolution and the
emergence of pathogenicity in avian Escherichia coli. Nat Commun. 2021; 12(1):765. Epub 20210203.
https://doi.org/10.1038/s41467-021-20988-w PMID: 33536414; PubMed Central PMCID:
PMC7858641.
10. Palmieri N, Apostolakos I, Paudel S, Hess M. The genetic network underlying the evolution of pathoge-
nicity in avian Escherichia coli. Front Vet Sci. 2023; 10:1195585. Epub 20230621. https://doi.org/10.
3389/fvets.2023.1195585 PMID: 37415967; PubMed Central PMCID: PMC10321414.
11. Dziva F, Stevens MP. Colibacillosis in poultry: unravelling the molecular basis of virulence of avian path-
ogenic Escherichia coli in their natural hosts. Avian Pathol. 2008; 37(4):355–66. https://doi.org/10.1080/
03079450802216652 PMID: 18622850.
12. Pires J, Huisman JS, Bonhoeffer S, Van Boeckel TP. Increase in antimicrobial resistance in Escherichia
coli in food animals between 1980 and 2018 assessed using genomes from public databases. J Antimi-
crob Chemother. 2022; 77(3):646–55. https://doi.org/10.1093/jac/dkab451 PMID: 34894245.
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
18 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
13.
Zhang P, Shen Z, Zhang C, Song L, Wang B, Shang J, et al. Surveillance of antimicrobial resistance
among Escherichia coli from chicken and swine, China, 2008–2015. Vet Microbiol. 2017; 203:49–55.
Epub 20170220. https://doi.org/10.1016/j.vetmic.2017.02.008 PMID: 28619166.
14. Hess C, Troxler S, Jandreski-Cvetkovic D, Zloch A, Hess M. Escherichia coli Isolated from Organic Lay-
ing Hens Reveal a High Level of Antimicrobial Resistance despite No Antimicrobial Treatments. Antibi-
otics (Basel). 2022; 11(4). Epub 20220330. https://doi.org/10.3390/antibiotics11040467 PMID:
35453218; PubMed Central PMCID: PMC9027956.
15. Han T, Zhang Q, Liu N, Wang J, Li Y, Huang X, et al. Changes in antibiotic resistance of Escherichia
coli during the broiler feeding cycle. Poult Sci. 2020; 99(12):6983–9. Epub 20200801. https://doi.org/10.
1016/j.psj.2020.06.068 PMID: 33248614; PubMed Central PMCID: PMC7704736.
16. EFSA, Health EPoA, Welfare., Nielsen SS, Bicout DJ, Calistri P, et al. Assessment of animal diseases
caused by bacteria resistant to antimicrobials: Poultry. EFSA J. 2021; 19(12):e07114. Epub 20211224.
https://doi.org/10.2903/j.efsa.2021.7114 PMID: 34987629; PubMed Central PMCID: PMC8703241.
17. Ghunaim H, Abu-Madi MA, Kariyawasam S. Advances in vaccination against avian pathogenic Escheri-
chia coli respiratory disease: potentials and limitations. Vet Microbiol. 2014; 172(1–2):13–22. Epub
20140505. https://doi.org/10.1016/j.vetmic.2014.04.019 PMID: 24878325.
18. Swelum AA, Elbestawy AR, El-Saadony MT, Hussein EOS, Alhotan R, Suliman GM, et al. Ways to min-
imize bacterial infections, with special reference to Escherichia coli, to cope with the first-week mortality
in chicks: an updated overview. Poult Sci. 2021; 100(5):101039. Epub 20210211. https://doi.org/10.
1016/j.psj.2021.101039 PMID: 33752065; PubMed Central PMCID: PMC8010699.
19. Christensen H, Bachmeier J, Bisgaard M. New strategies to prevent and control avian pathogenic
Escherichia coli (APEC). Avian Pathol. 2021; 50(5):370–81. Epub 20210203. https://doi.org/10.1080/
03079457.2020.1845300 PMID: 33146543.
20. Kathayat D, Lokesh D, Ranjit S, Rajashekara G. Avian Pathogenic Escherichia coli (APEC): An Over-
view of Virulence and Pathogenesis Factors, Zoonotic Potential, and Control Strategies. Pathogens.
2021; 10(4). Epub 20210412. https://doi.org/10.3390/pathogens10040467 PMID: 33921518; PubMed
Central PMCID: PMC8069529.
21. Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, et al. Cochrane handbook for system-
atic reviews of interventions version 6.3 (available at http://www.training.cochrane.org/handbook).
2022.
22. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020
statement: an updated guideline for reporting systematic reviews. BMJ. 2021; 372:n71. Epub
20210329. https://doi.org/10.1136/bmj.n71 PMID: 33782057; PubMed Central PMCID: PMC8005924.
23. Bramer WM, Rethlefsen ML, Kleijnen J, Franco OH. Optimal database combinations for literature
searches in systematic reviews: a prospective exploratory study. Syst Rev. 2017; 6(1):245. Epub
20171206. https://doi.org/10.1186/s13643-017-0644-y PMID: 29208034; PubMed Central PMCID:
PMC5718002.
24. Ouzzani M, Hammady H, Fedorowicz Z, Elmagarmid A. Rayyan-a web and mobile app for systematic
reviews. Syst Rev. 2016; 5(1):210. Epub 20161205. https://doi.org/10.1186/s13643-016-0384-4 PMID:
27919275; PubMed Central PMCID: PMC5139140.
25. Sargeant JM, Bergevin MD, Churchill K, Dawkins K, Deb B, Dunn J, et al. The efficacy of antibiotics to
control colibacillosis in broiler poultry: a systematic review. Anim Health Res Rev. 2019; 20(2):263–73.
https://doi.org/10.1017/S1466252319000264 PMID: 32081126.
26. Bueno I, Ricke I, Hwang H, Smith E, Nault A, Johnson TJ, et al. Efficacy of Antibiotic and Non-antibiotic
Interventions in Preventing and Treating Necrotic Enteritis in Broiler Chickens: A Systematic Review.
Avian Dis. 2023; 67(1):20–32. https://doi.org/10.1637/aviandiseases-D-22-00069 PMID: 37140108.
27. Abdul-Aziz TA, el-Sukhon SN. Chickens hyperimmunized with Escherichia coli J5 strain are protected
against experimental challenge with Escherichia coli O78 serotype. Vet Res Commun. 1998; 22(1):7–9.
https://doi.org/10.1023/a:1005927026159 PMID: 9541985.
28.
Ibrahim AI, Elattar AA, E—Shahidy MS. Studies on E. coli isolates from respiratory affected broilers and
protection evaluation of different prepared bacterins Assiut Veterinary Medical Journal. 1997; 37
(74):152–62.
29. Sayed MFE, Soliman RA, Ghanem HM, Khedr MMS, Mohamed GM, Safty M. Trials for preparation and
evaluation of a combined inactivated reassorted H5N1 and Escherichia coli O157 vaccine in poultry.
Vet World. 2021; 14(6):1677–81. Epub 20210628. https://doi.org/10.14202/vetworld.2021.1677-1681
PMID: 34316218; PubMed Central PMCID: PMC8304417.
30. Keita A, Le Devendec L, Amelot M, Puterflam J, Lucas C, Bougeard S, et al. Efficacy of passive immuni-
zation in broiler chicks via an inactivated Escherichia coli autogenous vaccine administered to broiler
breeder hens. Avian Pathol. 2022; 51(5):445–56. Epub 20220805. https://doi.org/10.1080/03079457.
2022.2084362 PMID: 35634647.
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
19 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
31.
Fuhrmann L, Zentek J, Vahjen W, Gunther R, Saliu EM. Effects of a Specific Pre- and Probiotic Combi-
nation and Parent Stock Vaccination on Performance and Bacterial Communities in Broilers Challenged
with a Multidrug-Resistant Escherichia coli. Antibiotics (Basel). 2022; 11(12). Epub 20221126. https://
doi.org/10.3390/antibiotics11121703 PMID: 36551360; PubMed Central PMCID: PMC9774208.
32. Clark TG, Cassidy-Hanley D. Recombinant subunit vaccines: potentials and constraints. Dev Biol
(Basel). 2005; 121:153–63. PMID: 15962478.
33. Vandemaele F, Bleyen N, Abuaboud O, vanderMeer E, Jacobs A, Goddeeris BM. Immunization with
the biologically active lectin domain of PapGII induces strong adhesion-inhibiting antibody responses
but not protection against avian pathogenic Escherichia coli. Avian Pathol. 2006; 35(3):238–49. https://
doi.org/10.1080/03079450600710997 PMID: 16753616.
34. Vandemaele F, Ververken C, Bleyen N, Geys J, D’Hulst C, Addwebi T, et al. Immunization with the bind-
ing domain of FimH, the adhesin of type 1 fimbriae, does not protect chickens against avian pathogenic
Escherichia coli. Avian Pathol. 2005; 34(3):264–72. https://doi.org/10.1080/03079450500112682
PMID: 16191712.
35.
Lynne AM, Kariyawasam S, Wannemuehler Y, Johnson TJ, Johnson SJ, Sinha AS, et al. Recombinant
Iss as a potential vaccine for avian colibacillosis. Avian Dis. 2012; 56(1):192–9. https://doi.org/10.1637/
9861-072111-Reg.1 PMID: 22545546.
36. Chaudhari AA, Matsuda K, Lee JH. Construction of an attenuated Salmonella delivery system harboring
genes encoding various virulence factors of avian pathogenic Escherichia coli and its potential as a can-
didate vaccine for chicken colibacillosis. Avian Dis. 2013; 57(1):88–96. https://doi.org/10.1637/10277-
061312-Reg.1 PMID: 23678735.
37. Chaudhari AA, Lee JH. Evaluation of the adjuvant effect of Salmonella-based Escherichia coli heat-
labile toxin B subunits on the efficacy of a live Salmonella-delivered avian pathogenic Escherichia coli
vaccine. Avian Pathol. 2013; 42(4):365–72. Epub 20130701. https://doi.org/10.1080/03079457.2013.
811466 PMID: 23815619.
38. Ma ST, Ding GJ, Huang XW, Wang ZW, Wang L, Yu ML, et al. Immunogenicity in chickens with orally
administered recombinant chicken-borne Lactobacillus saerimneri expressing FimA and OmpC antigen
of O78 avian pathogenic Escherichia coli. J Med Microbiol. 2018; 67(3):441–51. Epub 20180119.
https://doi.org/10.1099/jmm.0.000679 PMID: 29458539.
39. Dissanayake DRA, Wijewardana TG, Gunawardena GA, Poxton IR. Potential use of a liposome-encap-
sulated mixture of lipopolysaccharide core types (R1, R2, R3 and R4) of Escherichia coli in controlling
colisepticaemia in chickens. J Med Microbiol. 2010; 59(Pt 1):100–7. https://doi.org/10.1099/jmm.0.
014720-0 PMID: 19797465.
40.
Tuntufye HN, Ons E, Pham AD, Luyten T, Van Gerven N, Bleyen N, et al. Escherichia coli ghosts or live
E. coli expressing the ferri-siderophore receptors FepA, FhuE, IroN and IutA do not protect broiler chick-
ens against avian pathogenic E. coli (APEC). Vet Microbiol. 2012; 159(3–4):470–8. Epub 20120509.
https://doi.org/10.1016/j.vetmic.2012.04.037 PMID: 22633153.
41. Micoli F, MacLennan CA. Outer membrane vesicle vaccines. Semin Immunol. 2020; 50:101433. Epub
20201209. https://doi.org/10.1016/j.smim.2020.101433 PMID: 33309166.
42. Hu R, Liu H, Wang M, Li J, Lin H, Liang M, et al. An OMV-Based Nanovaccine Confers Safety and Pro-
tection against Pathogenic Escherichia coli via Both Humoral and Predominantly Th1 Immune
Responses in Poultry. Nanomaterials (Basel). 2020; 10(11). Epub 20201120. https://doi.org/10.3390/
nano10112293 PMID: 33233490; PubMed Central PMCID: PMC7699605.
43. Hu R, Li J, Zhao Y, Lin H, Liang L, Wang M, et al. Exploiting bacterial outer membrane vesicles as a
cross-protective vaccine candidate against avian pathogenic Escherichia coli (APEC). Microb Cell Fact.
2020; 19(1):119. Epub 20200603. https://doi.org/10.1186/s12934-020-01372-7 PMID: 32493405;
PubMed Central PMCID: PMC7268718.
44. Barbieri NL, de Oliveira AL, Tejkowski TM, Pavanelo DB, Rocha DA, Matter LB, et al. Genotypes and
pathogenicity of cellulitis isolates reveal traits that modulate APEC virulence. PLoS One. 2013; 8(8):
e72322. Epub 20130819. https://doi.org/10.1371/journal.pone.0072322 PMID: 23977279; PubMed
Central PMCID: PMC3747128.
45. Pati R, Shevtsov M, Sonawane A. Nanoparticle Vaccines Against Infectious Diseases. Front Immunol.
2018; 9:2224. Epub 20181004. https://doi.org/10.3389/fimmu.2018.02224 PMID: 30337923; PubMed
Central PMCID: PMC6180194.
46. Mohammed GM, ElZorkany HE, Farroh KY, Abd El-Aziz WR, Elshoky HA. Potential improvement of the
immune response of chickens against E. coli vaccine by using two forms of chitosan nanoparticles. Int J
Biol Macromol. 2021; 167:395–404. Epub 20201201. https://doi.org/10.1016/j.ijbiomac.2020.11.200
PMID: 33275976.
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
20 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
47. Abd El-Aziz WR, Ibrahim HM, Elzorkany HE, Mohammed GM, Mikhael CA, Fathy NA, et al. Evaluation
of cell-mediated immunity of E.coli nanovaccines in chickens. J Immunol Methods. 2022; 506:113280.
Epub 20220513. https://doi.org/10.1016/j.jim.2022.113280 PMID: 35577101.
48.
Frommer A, Freidlin PJ, Bock RR, Leitner G, Chaffer M, Heller ED. Experimental vaccination of young
chickens with a live, non-pathogenic strain of Escherichia coli. Avian Pathol. 1994; 23(3):425–33.
https://doi.org/10.1080/03079459408419013 PMID: 18671110.
49. Kariyawasam S, Wilkie BN, Hunter DB, Gyles CL. Systemic and mucosal antibody responses to
selected cell surface antigens of avian pathogenic Escherichia coli in experimentally infected chickens.
Avian Dis. 2002; 46(3):668–78. https://doi.org/10.1637/0005-2086(2002)046[0668:SAMART]2.0.CO;2
PMID: 12243531.
50. Peighambari SM, Hunter DB, Shewen PE, Gyles CL. Safety, immunogenicity, and efficacy of two
Escherichia coli cya crp mutants as vaccines for broilers. Avian Dis. 2002; 46(2):287–97. https://doi.org/
10.1637/0005-2086(2002)046[0287:SIAEOT]2.0.CO;2 PMID: 12061637.
51. Kariyawasam S, Wilkie BN, Gyles CL. Construction, characterization, and evaluation of the vaccine
potential of three genetically defined mutants of avian pathogenic Escherichia coli. Avian Dis. 2004; 48
(2):287–99. https://doi.org/10.1637/7093 PMID: 15283416.
52. Asaad HM, Amen O, Elazeem MA, Saif-Edin M. Comparative study on commercial vaccines against E.
coli in broiler chickens Assiut Veterinary Medical Journal 2019; 65:22–9.
53. El-Mawgoud AIA, El-Nahass ES, Shany SAS, El-Sawah AA, Dahshan AM, Nasef SA, et al. Efficacy of
Live Attenuated Vaccine and Commercially Available Lectin Against Avian Pathogenic E. coli Infection
in Broiler Chickens. Vet Sci. 2020; 7(2). Epub 20200513. https://doi.org/10.3390/vetsci7020065 PMID:
32414109; PubMed Central PMCID: PMC7355798.
54. Sadeghi M, Tavakkoli H, Golchin M, Ghanbarpour R, Amanollahi S. Efficacy and safety of Poulvac E.
coli vaccine in broiler chickens challenged with E. coli serotype O78 and an acute field isolate. Compar-
ative Clinical Pathology 2018; 27:1629–36. https://doi.org/10.1007/s00580-018-2784-4.
55. Galal HM, Tawfek AM, Abdrabou MI, Hessain AM, Alhaaji JH, Kabli SA, et al. Recent approaches for
control of E. coli and respiratory complex in Middle East. Saudi J Biol Sci. 2018; 25(7):1302–7. Epub
20180403. https://doi.org/10.1016/j.sjbs.2018.04.004 PMID: 30505174; PubMed Central PMCID:
PMC6252003.
56. Mohammed MA, Bakhit BM, Ibrahim AA, Saleh M. Evaluation of the living Escherichia coli-O78 deleted
aroA vaccine against homologous and heterologous E. coli challenge in broiler chickens. Journal of
Advanced Veterinary Research 2016; 6(3):89–92.
57. Gharib AA, Hamouda AM, Abdel-Wahab AAM, Fawzy MF. Protective Efficacy of a Commercial Live
Attenuated aroA mutant Vaccine Against Avian Pathogenic Escherichia coli Challenge in Broilers Zaga-
zig Veterinary Journal 2017; 45(4).
58.
Tarabees R, El-Sayed MS, Shehata AA, Diab MS. Effects of the Probiotic Candidate E. faecalis-1, the
Poulvac E. coli Vaccine, and their Combination on Growth Performance, Caecal Microbial Composition,
Immune Response, and Protection against E. coli O78 Challenge in Broiler Chickens. Probiotics Antimi-
crob Proteins. 2020; 12(3):860–72. https://doi.org/10.1007/s12602-019-09588-9 PMID: 31650414.
59. Rawiwet V, Chansiripornchai N. The Efficacy of Escherichia coli AroA-Live Vaccine inBroilers against
Avian E. coli Serotype O78 Infection. Thai Journal of Veterinary Medicine 2009; 39(4):337–42.
60. Elbestawy AR, Ellakany HF, Abd El-Hamid HS, Ibrahim MS, Gado AR, Mustafa NS, et al. Comparative
evaluation of a live E. coli vaccine and cefotaxime treatment against three E. coli serotypes in broilers.
Journal of King Saud University—Science. 2021; 33(2). https://doi.org/10.1016/j.jksus.2021.101353.
61. Galal HM, Abdrabou MI, Faraag AHI, Mah CK, Tawfek AM. Evaluation of commercially available aroA
delated gene E. coli O78 vaccine in commercial broiler chickens under Middle East simulating field con-
ditions. Sci Rep. 2021; 11(1):1938. Epub 20210121. https://doi.org/10.1038/s41598-021-81523-x
PMID: 33479449; PubMed Central PMCID: PMC7820230.
62.
63.
Fernandes Filho T, Favaro C Jr., Ingberman M, Beirao BC, Inoue A, Gomes L, et al. Effect of spray
Escherichia coli vaccine on the immunity of poultry. Avian Dis. 2013; 57(3):671–6. https://doi.org/10.
1637/10456-112612-ResNote.1 PMID: 24283136.
Li L, Thofner I, Christensen JP, Ronco T, Pedersen K, Olsen RH. Evaluation of the efficacy of an autog-
enous Escherichia coli vaccine in broiler breeders. Avian Pathol. 2017; 46(3):300–8. Epub 20170320.
https://doi.org/10.1080/03079457.2016.1267857 PMID: 27982712.
64. Klinman DM, Klaschik S, Sato T, Tross D. CpG oligonucleotides as adjuvants for vaccines targeting
infectious diseases. Adv Drug Deliv Rev. 2009; 61(3):248–55. Epub 20090114. https://doi.org/10.1016/
j.addr.2008.12.012 PMID: 19272313.
65. Gunawardana T, Ahmed KA, Goonewardene K, Popowich S, Kurukulasuriya S, Karunarathna R, et al.
Synthetic CpG-ODN rapidly enriches immune compartments in neonatal chicks to induce protective
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
21 / 22
PLOS ONE Efficay of vaccination against colibacillosis in broilers
immunity against bacterial infections. Sci Rep. 2019; 9(1):341. Epub 20190123. https://doi.org/10.1038/
s41598-018-36588-6 PMID: 30674918; PubMed Central PMCID: PMC6344490.
66.
Taghavi A, Allan B, Mutwiri G, Foldvari M, Van Kessel A, Willson P, et al. Enhancement of immunopro-
tective effect of CpG-ODN by formulation with polyphosphazenes against E. coli septicemia in neonatal
chickens. Curr Drug Deliv. 2009; 6(1):76–82. https://doi.org/10.2174/156720109787048221 PMID:
19418959.
67. Gomis S, Babiuk L, Allan B, Willson P, Waters E, Ambrose N, et al. Protection of neonatal chicks
against a lethal challenge of Escherichia coli using DNA containing cytosine-phosphodiester-guanine
motifs. Avian Dis. 2004; 48(4):813–22. https://doi.org/10.1637/7194-041204R PMID: 15666862.
68. Allan B, Wheler C, Koster W, Sarfraz M, Potter A, Gerdts V, et al. In Ovo Administration of Innate
Immune Stimulants and Protection from Early Chick Mortalities due to Yolk Sac Infection. Avian Dis.
2018; 62(3):316–21. https://doi.org/10.1637/11840-041218-Reg.1 PMID: 30339510.
69. Sarfraz M, Nguyen TTT, Wheler C, Koster W, Gerdts V, Dar A. Characterization of Dosage Levels for In
Ovo Administration of Innate Immune Stimulants for Prevention of Yolk Sac Infection in Chicks. Vet Sci.
2022; 9(5). Epub 20220422. https://doi.org/10.3390/vetsci9050203 PMID: 35622731; PubMed Central
PMCID: PMC9142911.
70. Goonewardene KB, Popowich S, Gunawardana T, Gupta A, Kurukulasuriya S, Karunarathna R, et al.
Intrapulmonary Delivery of CpG-ODN Microdroplets Provides Protection Against Escherichia coli Septi-
cemia in Neonatal Broiler Chickens. Avian Dis. 2017; 61(4):503–11. https://doi.org/10.1637/11684-
060617-Reg.1 PMID: 29337617.
71. Gregersen RH, Christensen H, Ewers C, Bisgaard M. Impact of Escherichia coli vaccine on parent
stock mortality, first week mortality of broilers and population diversity of E. coli in vaccinated flocks.
Avian Pathol. 2010; 39(4):287–95. https://doi.org/10.1080/03079457.2010.495744 PMID: 20706885.
72. Mombarg M, Bouzoubaa K, Andrews S, Vanimisetti HB, Rodenberg J, Karaca K. Safety and efficacy of
an aroA-deleted live vaccine against avian colibacillosis in a multicentre field trial in broilers in Morocco.
Avian Pathol. 2014; 43(3):276–81. https://doi.org/10.1080/03079457.2014.917760 PMID: 24824589.
73. Sˇ enk D, Papousˇkova´ A, Masařı´kova´ M, Palkovičova´ J, Čı´zˇek A. Impact of commercial and autogenous
Escherichia coli vaccine combination on broiler breeder stock performance, gross pathology, and diver-
sity of Escherichia coli isolates. ACTA VET BRNO. 2022; 91:383–90. https://doi.org/10.2754/
avb202291040383.
74. Paudel S, Fink D, Abdelhamid MK, Zo¨ggeler A, Liebhart D, Hess M, et al. Aerosol is the optimal route of
respiratory tract infection to induce pathological lesions of colibacillosis by a lux-tagged avian patho-
genic Escherichia coli in chickens. Avian Pathology. 2021; 50(5):417–26. https://doi.org/10.1080/
03079457.2021.1978392 WOS:000703401400001. PMID: 34505551
75. Paudel S, Hess C, Kamal Abdelhamid M, Lyrakis M, Wijewardana V, Thiga Kangethe R, et al. Aerosol
delivered irradiated Escherichia coli confers serotype-independent protection and prevents colibacillo-
sis in young chickens. Vaccine. 2023; 41(7):1342–53. Epub 20230113. https://doi.org/10.1016/j.
vaccine.2022.12.002 PMID: 36642629.
76. Smith AA, Corona-Torres R, Hewitt RE, Stevens MP, Grant AJ, Glycoengineering of Veterinary Vac-
cines C. Modification of avian pathogenic Escherichia coli chi7122 lipopolysaccharide increases acces-
sibility to glycoconjugate antigens. Microb Cell Fact. 2022; 21(1):181. Epub 20220907. https://doi.org/
10.1186/s12934-022-01903-4 PMID: 36071433; PubMed Central PMCID: PMC9449299.
PLOS ONE | https://doi.org/10.1371/journal.pone.0301029 March 22, 2024
22 / 22
PLOS ONE
|
10.1371_journal.pone.0299141
|
RESEARCH ARTICLE
Symptoms 6 months following SARS-CoV-2
infection in Nepali women
1☯, Sajani Manandhar2, Bimal Sharma Chalise3, Sagar
Deepak S. ShresthaID
Kumar Rajbhandari3, Anup Bastola3, Parmananda Bhandari3, Santa Kumar Das4,
Pankaj Pant4, Sangita Sharma4, Hari Prasad Kattel4, Roshan Kumar Jha5, Mahendra
Raj ShresthaID
5, Anil Shrestha5, Richard R. LoveID
6☯*
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Shrestha DS, Manandhar S, Chalise BS,
Rajbhandari SK, Bastola A, Bhandari P, et al.
(2024) Symptoms 6 months following SARS-CoV-
2 infection in Nepali women. PLoS ONE 19(3):
e0299141. https://doi.org/10.1371/journal.
pone.0299141
Editor: Babatunde Olanrewaju Motayo, Federal
Medical Centre Abeokuta, NIGERIA
Received: July 19, 2023
Accepted: February 5, 2024
Published: March 11, 2024
Copyright: © 2024 Shrestha et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: Data collected as part
of this research will be available to established
investigators presenting appropriate scientific and
ethical proposal approval documentation. All de-
identified subject data and the oral informed
consent text will be available with publication of
this communication. The contact person for data
access committee, ethics committee, or other
institutional requests may be sent is: Sanju
Maharjan Programme Manager Health Unit Email:
[email protected].
1 Department of Internal Medicine, People’s Dental College and Hospital, Kathmandu, Nepal, 2 New ERA,
Kathmandu, Nepal, 3 Sukraraj Tropical and Infectious Disease Hospital, Kathmandu, Nepal, 4 Tribhuvan
University Teaching Hospital, Kathmandu, Nepal, 5 Nepal Armed Police Forces Hospital, Kathmandu, Nepal,
6 Independent researcher, Madison, United States
☯ These authors contributed equally to this work.
* [email protected]
Abstract
In Nepal, over 1 million individuals have tested positive for SARS-CoV-2. We sought to
describe the frequency of nonrecovery from this infection at 6 months and associated symp-
toms. We conducted a retrospective cohort study of 6142 women who had positive and neg-
ative PCR tests for this infection 6 months previously at 3 institutions in Kathmandu. In
telephone interviews women provided information on 22 symptoms and their intensities,
health status and history, and functional status. Of 3732 women who had tested PCR posi-
tive, 630 (16.9%) reported that they were unrecovered. These 630 unrecovered women
were distinguished statistically from the 3102 recovered women by more frequent histories
of allergies, rheumatoid disease, BCG immunization, Covid vaccination, strep throat and
recent URIs, and both weight gain and weight losses of more than 5 kg in the 6 months fol-
lowing testing, and stressful events in the preceding year. Fatigue, pain, difficulty remember-
ing, shortness of breath, heat and cold intolerance and unrefreshing sleep were reported in
41.9% to 10.5% of these 630 unrecovered women. Six months after confirmed SARS-CoV-
2 infection 16.9% of Nepali women have long-COVID manifested as an immune, metabolic,
and hormonal systems disruptive and dysfunction syndrome
Introduction
In Nepal, over 1 million individuals have tested positive for SARS-CoV-2, and this figure may
underestimate actual numbers of cases because of limited testing. Data in patients beyond 3
months from diagnosis of this infection about symptoms, their severities, and timelines of
these, are limited, particularly in low- and middle-income countries. In high-income coun-
tries, investigative journal reports have suggested that 10 to 30 percent of infected individuals,
more commonly middle-aged women, have persistent functional capacity-limiting symptoms,
6 months and beyond the time of initial diagnosis [1–4].
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
1 / 12
PLOS ONE Funding: The senior author, Dr. Richard Love,
provided financial support for the collection of the
data reported here. He was involved in all aspects
of the study and this manuscript as indicated in the
manuscript.
Competing interests: The senior author, Dr.
Richard Love, provided financial support for the
data collection in this report. This does not alter our
adherence to PLOS ONE policies on sharing data
and materials.
Symptoms 6 months following SARS-CoV2 infectio
Persistent symptoms following Covid infection, mimic post-infectious disease syndromes
reported for multiple viral illnesses, Lyme disease, infectious mononucleosis, viral hepatitis, Q
fever and SARS-1, as well as those of chronic fatigue syndrome (CFS)- Myalgic Encephalomy-
elitis (ME), a poorly understood, complex and chronic clinical syndrome affecting women
four times more often than men [5–14]. CFS/ME is characterized by at least 6 months of men-
tal and physical fatigue, muscle weakness exacerbated by physical and social/mental exertion,
malaise, pain, non-restorative sleep, and cognitive impairment [14]. CFS/ME is a clinical diag-
nosis, with neuroinflammatory, metabolic, and hormonal physiological features [15].
Because of the significant health, social and economic consequences of persistent symptoms
following SARS-CoV-2 infection, we designed a study to describe the frequencies and intensi-
ties of the commonest reported symptoms, health history correlates and functional status in a
convenience sample of women tested for this infection because of suggestive symptoms 6
months previously, in order to evaluate similarities of suggested non-recovery from this infec-
tion to Chronic Fatigue Syndrome/Myalgic Encephalomyelitis [14].
Methods
We conducted a retrospective cohort study of women who, between August 3, 2020, and Sep-
tember 29, 2021, had self-referred themselves and then underwent PCR tests for SARS-CoV-2
done at three referral institutions in Kathmandu (Sukraraj Tropical and Infectious Disease
Hospital, Tribhuvan University Teaching Hospital and Nepal Armed Police Forces Hospital)
because they were symptomatic with fever, shortness of breath, cough, or anosmia. Between
March 21, 2021, and March 24, 2022, trained Nepali interviewers who were unaffiliated with
the testing institutions, called 6481 consecutive test-positive case women approximately 6
months (minimum 4 months) following their PCR test. If a family member answered the call
and reported that the individual had died, this was recorded. Three attempts were made to
contact these women. Participants were briefed about the study, as explained in the ethics
committee-approved protocol, informed verbal consent sought with no incentive offered for
participating in the study, before enrolling them in the study.
During the same study period, attempts were made to call 5940 randomly selected age (in
same 5-year age group) and test date-matched women who had tested negative for SARS-
CoV-2 at the same institutions 6 months previously (Fig 1).
All successfully contacted and consented case (test-positive) and control (test-negative)
women were first screened for histories of reinfections (test positive cases) or infections (test
negative controls) with SARS-CoV-2, cancer, tuberculosis, pregnancy, mental illness, and
HIV. Individuals who did not report these medical conditions were study eligible. Interviewers
then asked participants about their current symptoms, health status history and functional sta-
tus. The development and validation of the symptom questionnaire has been reported in
another publication [16]. The functional status questions were selected from among 90 items
of the Common data elements for evaluation of Chronic Fatigue Syndrome/Myalgic Encepha-
lomyelitis by the United States National Institutes of Health [17]. A question about Covid-19
vaccination asked whether the patient had ever had such immunization, details of which were
not ascertained; The Chinese Sinopharm BIBP, AstraZeneca, and Johnson and Johnson
immunizations were available to these women. Any positive response would very likely have
indicated receipt of a vaccine after the testing date report which anchored entry into the study.
For the patients who had tested positive, we then asked about their recovery. “Do you feel
that you have not completely recovered from your Covid-19 infection?” followed by other
questions seeking confirmatory or clarifying responses about “feeling unwell” attributed to
Covid-19 infection and feeling unreturned to usual level of health following Covid-19
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
2 / 12
PLOS ONE Symptoms 6 months following SARS-CoV2 infectio
Fig 1. Consort diagram of study populations.
https://doi.org/10.1371/journal.pone.0299141.g001
infection. We defined a subgroup of test-positive cases as “unrecovered from SARS-CoV-2
infection” based on affirmative response to this questioning.
We used Pearson’s chi square or Fisher’s exact test for categorical data, and mood’s median
test for continuous data in comparisons of characteristics between groups. All p values were
two sided; p values of <0.05 were considered statistically significant. Statistical analyses were
done with SPSS version 25.
The study was approved by the Ethics Committee of the Nepal Health Research Council I.
D. # l8l / 202l, on 12 March 2021, and amended on 24th August 2021, and subsequently the
Institutional Review Board at Marquette University in the United States. The senior author
(RRL) provided the funding for the research and was involved in all aspects of this activity as
indicated in the section detailing authors’ contributions.
Results
The test positive (42.4%) and test negative (59%) potential study subjects who were not study
eligible occurred because for the majority of these no contact could be made after 3 phone
calls, attributed to: no answer, wrong number, left the country, number change and new num-
ber unknown, or among significant numbers of potential subjects, the consent and data-
obtaining processes were unimplementable because of the associated time burdens. The
greater percentage of test negative individuals who were not study eligible occurred because
7.5% of identified subjects had subsequently developed SARS-CoV-2 infection, and a larger
percentage of these individuals refused to acknowledge the subject of the call. The age, test site,
test date characteristics of the study eligible and ineligible subjects were similar.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
3 / 12
PLOS ONE Symptoms 6 months following SARS-CoV2 infectio
Among the 3732 test-positive study eligible women (Median age 35 years, age range: 18–96
years; interview timepoint from diagnosis range 4–8 months, median 6.0 months, mean 5.7
months); 630 (16.9%) (Median age 37 years, age range: 18–77 years, interview timepoint from
diagnosis range 4–8 months, median 6.0 months, mean 5.7 months) reported that they had
not completely recovered from their SARS-CoV-2 infection.
The 2436 test negative study eligible women had similar demographic characteristics:
(Median age 35 years, age range: 8–82 years, interview timepoint from diagnosis range 3.3–9.6
months, median 6.0 months, mean 6.4 months). The 630 unrecovered women differed statisti-
cally from the recovered and the test negative women with greater median age (37 versus 35
and 35 years) (p = 0.002, and p = 0.03).
Table 1 shows the characteristics of the three study populations. Unrecovered women are
statistically different from the recovered women with more frequent histories of allergies,
asthma, BCG immunization, rheumatoid disease, strep throat, and URIs in the preceding year.
Unrecovered women were statistically different also from recovered women in reporting both
weight gain and weight losses of more than 5 kg in the 6 months after SARS-CoV2 testing,
higher usual (before illness) level physical activity which may suppress hormonal fluctuation,
and more frequent stressful events in the preceding year. Age, diabetes, tobacco abuse, hyper-
tension history, and BMI statistically significant differences were not observed between these
two groups.
Unrecovered women are statistically different from the test-negative and reportedly SARS-
CoV-2 infection history-free women for all of the characteristics listed in the preceding para-
graph except asthma, and in having had more frequent dengue, alcohol consumption in the
last 30 days and hypertension histories, and fewer pregnancies in the previous year, more fre-
quent history of menstrual cycles in the preceding 3 months, and less frequent surgical and
general anesthetic procedures in the last year (Table 1). Diabetes, asthma, and three BMI mea-
sures did not differ between these groups.
Covid-19 immunization was more frequent in unrecovered women than in either recov-
ered or test negative women. Table 2 shows the detailed data about symptoms in the 630 test-
positive unrecovered women. Higher frequencies of worst pain, fatigue, shortness of breath,
poor sleep and difficulty remembering are seen. The low frequencies of self-reported depres-
sion, anxiety, chills, or fever, light-headedness or dizziness, cough, and changes in taste and
smell are notable.
Table 3 shows the frequencies of the 6 most common symptoms in the three study groups
of women. The percentages reported for each of these symptoms among the unrecovered
women are significantly different from those reported by both the recovered and test-negative
women at p = 0.0001. 423 (67%) of the 630 unrecovered women had at least one of the 3 most
common symptoms—fatigue, pain or shortness of breath.
Table 4 shows the responses to questions about activities of daily living for the 3 studied
groups of women. The unrecovered women report more frequent health problem- interfer-
ence with their usual activities, but the percentages reporting these problems in the unrecov-
ered group seem remarkably low compared with the percentages reporting symptoms in
Table 3.
Discussion
The principal findings from this study are:
• 16.9% or 1/6th of middle-aged Nepali women reported themselves as being unrecovered/
unwell/unhealthy following their PCR test confirmed SARS-CoV-2 infections at a median of
6 months, and reported symptoms and functional status information consistent with these
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
4 / 12
PLOS ONE Table 1. Demographic and Medical History Characteristics among 3 groups of women: 1) Test positive women self-assessed as unrecovered, 2) Test positive women
self-assessed as recovered, and 3) Test negative women who had remained Covid-19 symptoms free.
Symptoms 6 months following SARS-CoV2 infectio
Group 1
Test +
Unrecovered
630
5.7
37.4
37
15.7
18–77
Group 2
Test +
Recovered
3102
5.7
37
35
16.9
18–96
Group 3
Test–and
Disease-Free
2463
6.4
37.1
35
15.6
18–92
25.3
13.1
50.4
2.3
1.4
77.3
16.8
10.6
12.7
4.4
2.3
6.2
0.2
4.0
0.2
11.6
20.5
5.4
15.2
91.6
0.8
75.7
2.9
5.1
3.2
0.8
12.3
24.9
25
13.1
48.2
3.3
1.4
73.9
8.9
4.5
10.2
4.7
1.1
3.2
0
3.4
0.9
4.2
11.4
3.9
12.9
88.1
1.3
71.4
2.2
3.6
4.3
1.5
3.9
39.7
24.9
11.8
46.2
5.9
1.5
71.4
10.6
6.2
9.9
4.2
1.8
4.1
0
5.1
0.6
6.7
12.5
2.7
18.1
87.8
2.6
60.9
1.4
4.0
15.0
6.3
5.1
38.1
Demographic and Medical History Characteristics
Total
Time from PCR test (mean months)
Mean age (years)
Median age b, c (years)
Patients aged 50 and > (%)
Age range (years)
Calculated BMI (kg/m2)
BMI > = 30 (%)
BMI> = 25 (%)
Pregnancy in year before Covid test (% yes)f
Pregnancies (mean #)
Menstrual period in preceding 3 months before interview (% yes) d
Weight increases by 5 or more kg since testing (% yes) e, f
Weight decreases by 5 or more kg since testing (% yes) e, f
Hypertension history (%) c
Diabetes history (%)
Asthma history (%) a
Allergies history (%) b, d
Hepatitis B infection history (%)
Family history of serious infectious diseases (% yes)
Tobacco smoker (% yes)
Stressful event in last year (%) b, d
2 or more URIs in last year (%) b, d
Alcohol in last 30 days (% yes) f
Child under 5 in home (% yes) c
BCG vaccination in past (% yes) a, c
Influenza vaccination in in year4 (% yes)
Covid vaccination e, f
Dengue history d
Influenza in last year (% yes)
Surgical procedure in last year f (% yes)
General anaesthetic in last year f (% yes)
Strep throat last year (%) e, f
Physical activity <3h/week (usual level before Covid infection) e, f
a Statistically significant difference between group 1 and 2 @ p = 0.05.
b Statistically significant difference between group 1 and 2 @ p = 0.01.
c Statistically significant difference between group 1 and 3 @ p = 0.05.
d Statistically significant difference between group 1 and 3 @ p = 0.01.
e Statistically significant difference between group 1 and 2 @p = 0.001.
f Statistically significant difference between group 1 and 3 @p = 0.001.
https://doi.org/10.1371/journal.pone.0299141.t001
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
5 / 12
PLOS ONE Symptoms 6 months following SARS-CoV2 infectio
Table 2. Presence and intensity of 22 symptoms at a median of 6 months from diagnosis in 630 women self-
assessed to be incompletely recovered from SARS-CoV-2 infection.
Symptoms
Worst pain/ache
Pain/ache locations:
Muscles = 25
Back/Whole body = 82
Head = 55
Joints = 20
Chest = 75
Feeling sad or depressed
Difficulty in word-finding
Light-headedness or dizziness on standing
Lack of motivation
Mental and physical fatigue/tiredness
Poor, unrefreshing sleep
Fever and/or chills
Mental confusion or disorientation
Difficulty thinking and concentrating
Shortness of breath
Reduced physical activity
Increased sensitivity to sound or light
Rapid or irregular heartbeat
Cough
Anxious or worried
Difficulty remembering
Change in sense of smell
Increased fatigue the day after more-than-usual physical or social activity
Change in sense of taste
Numbness in fingers or toes
Heat or cold intolerance
https://doi.org/10.1371/journal.pone.0299141.t002
0
392
1
46
2
84
3
63
4
31
5
14
594
600
572
587
366
564
618
594
586
454
576
583
581
593
571
523
603
592
615
597
560
4
8
9
6
25
10
2
2
5
35
7
7
8
6
10
13
6
2
3
9
12
11
7
27
11
116
27
3
13
18
71
25
8
14
17
16
44
11
16
4
11
28
7
7
13
18
79
18
5
15
13
45
11
15
16
10
15
36
6
9
6
9
19
10
7
7
7
4
1
2
1
31
13
7
1
3
7
19
6
14
7
3
12
6
2
6
0
4
9
4
1
3
1
6
5
3
4
1
6
8
2
5
2
0
2
Table 3. Frequencies at a median of 6 months after SARS-CoV-2 PCR testing of the 6 most commonly reported symptoms in successfully interviewed women in 3
groups: 1) Test positive women self-assessed as unrecovered, 2) Test positive women self-assessed as recovered, and 3) Test negative women who had remained
Covid-19 symptoms free.
Demographic and Medical History Characteristics
Number and % reporting symptoms
Total Cases
Fatigue
Pain
Shortness of breath
Difficulty remembering
Heat or cold intolerance
Poor, unrefreshing sleep
https://doi.org/10.1371/journal.pone.0299141.t003
Group 1
Test +
Unrecovered
630
264 (41.9)
238 (37.8)
176 (27.9)
107 (17.0)
70 (11.1)
66 (10.5)
Group 2
Test +
Recovered
3102
39 (1.3)
39 (1.3)
22 (0.7)
25 (0.8)
9 (0.3)
15 (0.5)
Group 3
Test–and
Disease-Free
2436
86 (3.5)
96 (3.9)
40 (1.6)
37 (1.5)
20 (0.8)
37 (1.5)
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
6 / 12
PLOS ONE Table 4. Functional status for activities of daily living at a median of 6 months after Covid-PCR testing in successfully interviewed women in 3 groups: 1) Test posi-
tive women self-assessed as unrecovered, 2) Test positive women self-assessed as recovered, and 3) Test negative women who had remained Covid-19 symptoms
free.
Demographic and Medical History Characteristics
Number and % reporting symptoms
Symptoms 6 months following SARS-CoV2 infectio
Total Cases
Health Problems interfering with:
Doing the usual work done before your Covid-19 testing 6 months ago a, b
Walking upstairs a, b
Doing household tasks involving lifting, carrying, or cleaning
Taking care of children or adults with health problems a
Group 1
Test +
Unrecovered
630
13 (2.4)
17 (2.1)
75 (11.9)
15 (3.2)
Group 2
Test +
Recovered
3102
9 (0.3)
18 (0.7)
128 (4.1)
35 (1.4)
Group 3
Test–and
Disease-Free
2436
36 (1.7)
31 (1.4)
143 (5.9)
37 (2.2)
a Statistically significant difference between groups 1 and 2 @ p<0.05.
b Statistically significant difference between groups 1 and 3 @ p<0.01Data about symptoms for test positive recovered and test negative women show no striking
excesses of symptoms in the previously infected patients.
https://doi.org/10.1371/journal.pone.0299141.t004
assessments. Both groups of demographically matched recovered and never-infected
women, interviewed contemporaneously, reported low frequencies of the same major symp-
toms. The recovered and uninfected groups of women were very similar in their frequencies
of health characteristics—hypertension, diabetes, asthma, allergies, tobacco abuse, strongly
suggesting that the specific health characteristics that were different in the unrecovered
women are genuinely associated with this condition. These data indicate that non-recovery
and chronic illness after 6 months is a consequence of SARS-CoV-2 infection. The 6-month
timepoint defining such illness is the metric used to diagnose chronic fatigue syndrome [14].
• The health characteristics data suggest a rich picture of immune, metabolic, and hormonal
factors associated with persistence of symptoms and unrecovered status. Previously sug-
gested increased frequencies of histories of asthma and allergies were found, but also greater
immune system activation or susceptibility histories of dengue, and more frequent strep
throat and URIs in the last year, and more frequent BCG vaccination were reported. While
BCG vaccination has been suggested to produce protective “trained immunity” beneficial in
reducing the severity of Covid-19 illness, a randomized trial of BCG vaccination in health
workers to protect against Covid-19 found no evidence of benefit, with a trend suggesting
increased risk of infection from this vaccination [18,19]. In the data reported here, BCG vac-
cination was associated with increased risk for non-recovery at 6 months. Covid vaccination
history was more commonly reported in unrecovered women, but absent further details, a
cause-and-effect relationship cannot be suggested.
Important and significant metabolic factor differences were reported by the unrecovered
women with more frequent reported weight losses and gains of greater than 5 kg. in the period
since diagnosis, and more frequent major stressful events in the previous year. Further, it is
notable that in these Nepali populations, hypertension, diabetes, and particularly BMI/obesity
differences between recovered and unrecovered patients were not observed.
• Some of the differences between unrecovered women and women who never developed
SARS-CoV-2 infection are notable. The unrecovered women had less frequent pregnancies
in the previous year, more frequent history of menstrual cycles in the preceding 3 months,
and less frequent surgical and general anesthetic procedures in the last year, as well as more
frequent reported weight losses and gains of greater than 5 kg. in the period since diagnosis,
and more frequent major stressful events in the previous year.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
7 / 12
PLOS ONE Symptoms 6 months following SARS-CoV2 infectio
These conclusions should be interpreted in the following contexts. First, the women studied
were a convenience sample of individuals living in the Kathmandu valley, who self-referred
themselves for SARS-CoV-2 PCR blood testing because of symptoms of infection. Thus, this
study group is not randomly selected from the total Nepali population of symptom and symp-
tomless individuals and individuals capable and incapable of seeking testing. Further, few of
the studied women were likely to have been hospitalized with Covid and have incurred greater
specific organ injuries of lung, heart, brain, and blood coagulation tissues and systems. Severity
of illness was not otherwise assessed. During this period, the local hospitalization rates for
women with Covid-19 were under 5% and limited population variant testing suggested that
the prominent variants were alpha variant and delta variant [20]. The studied women were
mostly urban area residents, in a country where 80% of inhabitants live in rural areas, and
were likely better educated, generally younger, and healthier at the time of diagnosis than their
fellow women countrywomen. These considerations signify that the studied populations are
selected and are not representative of the total population of Nepal.
The details of how items are phrased in Nepali may be critical in some circumstances.
Finally, we were able to contact and successfully obtain data from 59% of test positive and 42%
of test negative cases or family members. While fractions of the non-recruited women were
because of specific study eligibility criteria, larger proportions were non-recruited because of
the specific operational conditions of the study—interviews solicited and conducted by tele-
phone, the process of acquiring informed consent, the absence of incentives, and the time
commitment. These non-interviewed women were demographically similar to those who were
successfully interviewed.
The strengths of this study lie in the large numbers of women from three testing sites stud-
ied, and in the facts and circumstances that: 1. A control group of SARS-CoV-2 test negative
and by history never-affected women was studied; these women reported low levels of any
symptoms, with much lower frequencies than the unrecovered women, which data offer com-
pelling evidence that the association of perceived unrecovered/unwell/unhealthy status and
symptoms with SARS-CoV-2 infection is strong. 2. Assessment of recovered status was made
by probing questioning about non-recovery (implying a chronic situation), presence of
unwellness, and persistent perceived adverse change in health status (unhealthy) following
SARS-CoV-2 infection, and symptom data were obtained directly from the patients them-
selves. 3. The case status was defined by a laboratory PCR test. 4. The symptoms’ descriptions
were for periods of 3 days. And: 5. The symptom questionnaire had reliability and validity
information suggesting reasonable credibility for the study population investigated, and the
findings are internally consistent—the non-recovered patients clearly report more specific
symptoms of important intensities and associated interference with activities of daily living
(Tables 2 and 4) [16]. These strengths all support an argument that these observational data for
the populations studied are not biased and are of high quality [21]. Finally, these data are
important because they describe illness experience in women from a low-middle income coun-
try where the frequency of symptomatic and serious illness with SARS-CoV-2 has been sug-
gested to be significantly lower than has been observed in high-income countries, and multiple
confounding factors such as lower self-reported levels of depression and anxiety, less mood-
altering and aspirin drug use, and lower alcohol consumption are not present [22].
Comparing the literature regarding nonrecovery and symptoms 6 months following con-
firmed Covid infection is problematic because the majority of reports address patients from
western countries (only 2 of 9 studies in a systematic review -reference #4- had patients from
non-western countries), concern previously hospitalized patients with severe acute disease, are
of older patient groups, and with the exception of a recent study in 3762 volunteers, are small
[1,4,23]. While the current report is not of a population-based sample, it comes closer to
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
8 / 12
PLOS ONE Symptoms 6 months following SARS-CoV2 infectio
describing what a representative group of low- and middle-income country Covid sufferers
are likely to be experiencing 6 months after their diagnoses than has been reported to date, spe-
cifically in addressing the issue of patient perceptions of their recovery status. The long term
most common symptoms’ picture in the reported literature is very similar to that reported
here: fatigue, CNS functional problems, and sleep disturbances [1,4,23]. Further what has been
the subject of limited study to date regarding long Covid sufferers is the associated co-morbid
conditions and metabolic and endocrine changes. The data reported here but begin to untan-
gle the full spectrum of these associated factors.
Long Covid symptoms and underlying physiologic mechanisms
Despite the richness of the reported data, the specific symptom frequencies reported by the
non-recovered women do not provide themselves adequate information to propose a defini-
tion of “long-COVID” and suggest that somehow our questions missed identification of criti-
cal symptoms. Malaise, a defining symptom in chronic fatigue syndrome, was not included
because of its problematic translation into Nepali [16]. Questions about energy levels, muscle
weakness, appetite, and details of exercise capacities were also omitted. The assignment of
non-recovery case status however, along with the specific symptoms data, validate this status
assignment, and suggest metabolic and endocrine hormonal systems disruption and that
future research should explore details of symptoms associated with these systems.
Importantly, these data suggest the parameters of a physiological model for the develop-
ment of long- COVID similar to that proposed for chronic fatigue syndrome [14,15]. Long-
COVID present at 6 months from time of infection, as seen in these Nepali women, is charac-
terized by:
1. Immune system dysfunctional responses associated with allergies, asthma, and BCG vacci-
nation histories; increased frequency of URIs, strep throat and dengue with neuroinflam-
matory symptoms of pain, difficulty remembering (suggested to be reflective of microglial
or dendritic damage), and poor sleep, in the absence of fever and specific signs of active
infection.
2. Metabolic and hormonal dysfunction with feelings of being unrecovered, unwell, and per-
sistently unhealthy, which are incompletely described by usual specific symptom
assessments.
3. Metabolic disturbances with specific symptoms of mental and physical fatigue, shortness of
breath (without fever or cough, suggesting exercise capacity loss), pain, poor sleep, difficulty
remembering, and heat and cold intolerance; and associated significant weight changes,
decreased physical activity, and history of stressful events.
4. Endocrine-hormonal change hypersensitivities associated with recent pregnancy and men-
strual cycling, and heat and cold intolerance.
Investigations of CFS/ME have suggested that it is a hypometabolic syndrome, and long-
COVID has been hypothesized to be a one carbon stress syndrome [24,25]. Together the cur-
rent and these reports suggest that long-COVID patients should be investigated for serum ser-
ine and markers of oxidative stress such as glutathione, as well as total serum B12, holo-
transcobalamin (holoTC), the metabolic markers methylmalonic acid and homocysteine, and
plasma formate. Physiological and dynamic assessment of multiple hormones are also sug-
gested: ACTH, cortisol, TRH, TSH, thyroid, insulin, epinephrine, serotonin, melatonin,
growth hormone, and aldosterone as examples [26]. A recent rigorous small study suggests
that Covid infection and type 1 interferon-driven inflammation decrease serotonin levels, and
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
9 / 12
PLOS ONE Symptoms 6 months following SARS-CoV2 infectio
that this change explains many of the major long Covid symptoms. [27] Interventions with
ACTH, Vitamin B-12, folate, glutathione, and serine, directed to re-setting metabolic and hor-
monal systems are suggested by these interpretations and models, as have been suggested for
CFS/ME [15,28].
Conclusion
Six months after PCR test-confirmed SARS-CoV-2, 16.9% of Nepali women reported being
unrecovered, and/or unwell, and/or unhealthy, with associated dominantly metabolic and hor-
monal systems symptoms, defining for them long-COVID. For these women immune system
over-activation factors and dysfunction were associated with this metabolic and endocrine-
hormonal disruptive condition.
Supporting information
S1 Dataset.
(CSV)
Author Contributions
Conceptualization: Deepak S. Shrestha, Richard R. Love.
Data curation: Deepak S. Shrestha, Sajani Manandhar, Richard R. Love.
Formal analysis: Sajani Manandhar, Richard R. Love.
Funding acquisition: Richard R. Love.
Methodology: Deepak S. Shrestha, Richard R. Love.
Project administration: Deepak S. Shrestha, Bimal Sharma Chalise, Pankaj Pant, Roshan
Kumar Jha, Richard R. Love.
Resources: Bimal Sharma Chalise, Sagar Kumar Rajbhandari, Anup Bastola, Parmananda
Bhandari, Santa Kumar Das, Pankaj Pant, Sangita Sharma, Hari Prasad Kattel, Roshan
Kumar Jha, Mahendra Raj Shrestha, Anil Shrestha.
Supervision: Deepak S. Shrestha, Richard R. Love.
Validation: Deepak S. Shrestha.
Visualization: Richard R. Love.
Writing – original draft: Richard R. Love.
Writing – review & editing: Deepak S. Shrestha, Sajani Manandhar, Richard R. Love.
References
1.
ndrrma.gov.np[internet]. Government of Nepal: Ministry of Home Affairs, National Disaster Risk Reduc-
tion and Management Authority (NDRRMA); [cited 2023 Nov 26]. Available from: https://covid19.
ndrrma.gov.np/.
2. Assaf G, Davis H, McCorkell L, Wei H, Brooke O, Akrami A, et al. What does COVID-19 recovery actu-
ally look like? An analysis of the prolonged COVID-19 symptoms survey by patient-led research team.
London, UK: The COVID-19 Body Politic Slack Group. 2020.
3. Velasquez-Manoff M. What If You Never Get Better From Covid-19? The New York Times. 2021 Jan 21
[cited 2021 March 27]. Available from: https://www.nytimes.com/2021/01/21/magazine/covid-
aftereffects.html.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
10 / 12
PLOS ONE Symptoms 6 months following SARS-CoV2 infectio
4. Groff D, Sun A, Ssentongo AE, Ba DM, Parsons N, Poudel GR, et al. Short-term and Long-term Rates
of Postacute Sequelae of SARS-CoV-2 Infection: A Systematic Review. JAMA Network Open. 2021; 4
(10):e2128568–e. https://doi.org/10.1001/jamanetworkopen.2021.28568 PMID: 34643720
5. Ayres JG, Flint N, Smith EG, Tunnicliffe WS, Fletcher TJ, Hammond K, et al. Post-infection fatigue syn-
drome following Q fever. Qjm. 1998; 91(2):105–23. https://doi.org/10.1093/qjmed/91.2.105 PMID:
9578893.
6. Berelowitz GJ, Burgess AP, Thanabalasingham T, Murray-Lyon IM, Wright DJ. Post-hepatitis syn-
drome revisited. J Viral Hepat. 1995; 2:133–8. https://doi.org/10.1111/j.1365-2893.1995.tb00018.x
PMID: 7493307.
7. Buchwald DS, Rea TD, Katon WJ, Russo JE, Ashley RL. Acute infectious mononucleosis: characteris-
tics of patients who report failure to recover. Am J Med. 2000; 109(7):531–7. https://doi.org/10.1016/
s0002-9343(00)00560-x PMID: 11063953.
8. Hickie I, Davenport T, Wakefield D, Vollmer-Conna U, Cameron B, Vernon SD, et al. Post-infective and
chronic fatigue syndromes precipitated by viral and non-viral pathogens: prospective cohort study. Bmj.
2006; 333(7568):575. Epub 20060901. https://doi.org/10.1136/bmj.38933.585764.AE PMID:
16950834; PubMed Central PMCID: PMC1569956.
9.
Lam MH-B, Wing Y-K, Yu MW-M, Leung C-M, Ma RCW, Kong APS, et al. Mental Morbidities and
Chronic Fatigue in Severe Acute Respiratory Syndrome Survivors: Long-term Follow-up. Arch Intern
Med. 2009; 169(22):2142–7. https://doi.org/10.1001/archinternmed.2009.384 PMID: 20008700.
10. Moldofsky H, Patcai J. Chronic widespread musculoskeletal pain, fatigue, depression and disordered
sleep in chronic post-SARS syndrome; a case-controlled study. BMC Neurology. 2011; 11(1):37.
https://doi.org/10.1186/1471-2377-11-37 PMID: 21435231
11. Ścieszka J, Dąbek J, Cieślik P. Post-Lyme disease syndrome. Reumatologia. 2015; 53(1):46–8. Epub
20150410. https://doi.org/10.5114/reum.2015.50557 PMID: 27407225; PubMed Central PMCID:
PMC4847307.
12. White PD, Thomas JM, Amess J, Crawford DH, Grover SA, Kangro HO, et al. Incidence, risk and prog-
nosis of acute and chronic fatigue syndromes and psychiatric disorders after glandular fever. Br J Psy-
chiatry. 1998; 173:475–81. https://doi.org/10.1192/bjp.173.6.475 PMID: 9926075.
13. White PD, Thomas JM, Kangro HO, Bruce-Jones WD, Amess J, Crawford DH, et al. Predictions and
associations of fatigue syndromes and mood disorders that occur after infectious mononucleosis. Lan-
cet. 2001; 358(9297):1946–54. https://doi.org/10.1016/S0140-6736(01)06961-6 PMID: 11747919.
14.
Institute of Medicine. Beyond Myalgic Encephalomyelitis/Chronic Fatigue Syndrome: redefining an ill-
ness. A report guide for clinicians. The National Academies Press. 2015. https://doi.org/10.17226/
19012.
15. Craddock TJA, Fritsch P, Rice MA, del Rosario RM, Miller DB, Fletcher MA, et al. A Role for Homeo-
static Drive in the Perpetuation of Complex Chronic Illness: Gulf War Illness and Chronic Fatigue Syn-
drome. PloS one. 2014; 9(1):e84839. https://doi.org/10.1371/journal.pone.0084839 PMID: 24416298
16. Shrestha DS, Rahat AM, Sridevi P. A consecutive series study of the frequencies, intensities, and natu-
ral history of symptoms following Covid-19 infection in Nepali men. J Nepal Health Res Counc. 2023; 21
(1):40–5. Epub 20230908. https://doi.org/10.33314/jnhrc.v21i1.4371 PMID: 37742147.
17.
commondataelements.ninds.nih.gov[internet]. NINDS Common Data Elements [cited 2021 Mar 27].
Available from: https://www.commondataelements.ninds.nih.gov/Myalgic%20Encephalomyelitis/
Chronic%20Fatigue%20Syndrome#pane-138.
18. Cirovic B, de Bree LCJ, Groh L, Blok BA, Chan J, van der Velden WJFM, et al. BCG Vaccination in
Humans Elicits Trained Immunity via the Hematopoietic Progenitor Compartment. Cell Host Microbe.
2020; 28(2):322–34.e5. Epub 20200615. https://doi.org/10.1016/j.chom.2020.05.014 PMID: 32544459;
PubMed Central PMCID: PMC7295478.
19. Pittet LF, Messina NL, Orsini F. et al. Randomized trial of BCG vaccine to protect against Covid-19 in
health care workers. N Engl J Med. 2023; 388(17):1582–96. https://doi.org/10.1056/NEJMoa2212616
PMID: 37099341; PubMed Central PMCID: PMC10497190.
20. Paudel S, Dahal A, Bhattarai HK. Temporal Analysis of SARS-CoV-2 Variants during the COVID-19
Pandemic in Nepal. COVID. 2021; 1(2):423–34. https://doi.org/10.3390/covid1020036
21. Hoy D, Brooks P, Woolf A, Blyth F, March L, Bain C, et al. Assessing risk of bias in prevalence studies:
modification of an existing tool and evidence of interrater agreement. J Clin Epidemiol. 2012; 65
(9):934–9. Epub 20120627. https://doi.org/10.1016/j.jclinepi.2011.11.014 PMID: 22742910.
22. Mukherjee S. Why does the pandemic seem to be hitting some countries harder than others? The New
Yorker. 2021 Feb 22 [cited 2021 March 27]. Available from: https://www.newyorker.com/magazine/
2021/03/01/why-does-the-pandemic-seem-to-be-hitting-some-countries-harder-than-others.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
11 / 12
PLOS ONE Symptoms 6 months following SARS-CoV2 infectio
23. Davisa H.E., Assafa G.S., McCorkell L., et al.: Characterizing long COVID in an international cohort: 7
months of symptoms and their impact. EClinicalMedicine. 2021; 38:101019. Epub 20210715. https://
doi.org/10.1016/j.eclinm.2021.101019 PMID: 34308300; PubMed Central PMCID: PMC8280690.
24. Balnis J, Madrid A, Hogan KJ, Drake LA, Adhikari A, Vancavage R, et al. Persistent blood DNA methyla-
tion changes one year after SARS-CoV-2 infection. Clinical Epigenetics. 2022; 14(1):94. https://doi.org/
10.1186/s13148-022-01313-8 PMID: 35871090
25. McCaddon A, Regland B. COVID-19: A methyl-group assault? Med Hypotheses. 2021; 149:110543.
Epub 20210218. https://doi.org/10.1016/j.mehy.2021.110543 PMID: 33657459; PubMed Central
PMCID: PMC7890339.
26. Naviaux RK, Naviaux JC, Li K, Bright AT, Alaynick WA, Wang L, et al. Metabolic features of chronic
fatigue syndrome. Proc Natl Acad Sci U S A. 2016; 113(37):E5472–80. Epub 20160829. https://doi.org/
10.1073/pnas.1607571113 PMID: 27573827; PubMed Central PMCID: PMC5027464.
27. Wong A.C., Devason A.S., Umana I.C., et al.: Serotonin reduction in post-acute sequelae of viral infec-
tion. Cell. 2023; 186(22):4851–67.e20. Epub 20231016. https://doi.org/10.1016/j.cell.2023.09.013
PMID: 37848036.
28. Klein J, Wood J, Jaycox J, Lu P, Dhodapkar RM, Gehlhausen JR, et al. Distinguishing features of Long
COVID identified through immune profiling. medRxiv. 2022. Epub 20220810. https://doi.org/10.1101/
2022.08.09.22278592 PMID: 35982667; PubMed Central PMCID: PMC9387160.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299141 March 11, 2024
12 / 12
PLOS ONE
|
10.3389_fmolb.2022.1074714
|
B
A
C
Figure S1. Molecular Dynamic simulation of sacsin-ATP complex. (A) Cluster analysis of the
loop; the most populated cluster is shown in red, the second in orange, the third in yellow, the
fourth in green and the least populated in blue. (B) RMSD analysis of protein C-alpha (blue) and
ligand (magenta). (C) Protein-ligand interaction diagram. Interactions are categorized into four
types: Hydrogen Bonds (green), Hydrophobic (lilac), Ionic (pink) and Water Bridges (blue). The
stacked bar charts are normalized over the course of the entire trajectory. Values over 1.0 are
possible as some protein residue may make multiple contacts of same subtype with the ligand.
Figure S2. Hsp90 middle and C-terminal regions and corresponding segments of sacsin that
do not superimpose. PyMol cartoon of Hsp90 (yellow and blue, residues 386 to 677) and sacsin
(green and red, residues 487 to 772). The blue long helix of the Hsp90 middle domain and that of
sacsin are superimposed. The rest of the structure shows no immediately recognisable common
topology.
B
A
C
Figure S3. Molecular Dynamic simulation of sacsin-geldanamycin complex. (A) Cluster
analysis of the loop; the most populated cluster is shown in red, the second in orange, the third in
yellow, the fourth in green and the least populated in blue. (B) RMSD analysis of protein C-alpha
(blue) and ligand (magenta). (C) Protein-ligand interaction diagram. Interactions are categorized
into four types: Hydrogen Bonds (green), Hydrophobic (lilac), Ionic (pink) and Water Bridges (blue).
The stacked bar charts are normalized over the course of the entire trajectory. Values over 1.0 are
possible as some protein residue may make multiple contacts of same subtype with the ligand.
B
A
C
Figure S4. Molecular Dynamic simulation of sacsin-AUY922 complex. (A) Cluster analysis of
the loop; themost populated cluster is shown in red, the second in orange, the third in yellow, the
fourth in green and the least in blue. (B) RMSD analysis of protein C-alpha (blue) and ligand
(magenta). (C) Protein-ligand interaction diagram. Interactions are categorized into four types:
Hydrogen Bonds (green), Hydrophobic (lilac), Ionic (pink) and Water Bridges (blue). The stacked
bar charts are normalized over the course of the entire trajectory. Values over 1.0 are possible as
some protein residue may make multiple contacts of same subtype with the ligand.
|
10.1371_journal.pone.0299456
|
RESEARCH ARTICLE
Automated code development based on
genetic programming in graphical
programming language: A pilot study
Pavel Kodytek, Alexandra BodzasID*, Jan Zidek
Department of Cybernetics and Biomedical Engineering, VSB-Technical University of Ostrava, Ostrava,
Czech Republic
* [email protected]
Abstract
Continual technological advances associated with the recent automation revolution have
tremendously increased the impact of computer technology in the industry. Software devel-
opment and testing are time-consuming processes, and the current market faces a lack of
specialized experts. Introducing automation to this field could, therefore, improve software
engineers’ common workflow and decrease the time to market. Even though many code-
generating algorithms have been proposed in textual-based programming languages, to the
best of the authors’ knowledge, none of the studies deals with the implementation of such
algorithms in graphical programming environments, especially LabVIEW. Due to this fact,
the main goal of this study is to conduct a proof-of-concept for a requirement-based auto-
mated code-developing system within the graphical programming environment LabVIEW.
The proposed framework was evaluated on four basic benchmark problems, encompassing
a string model, a numeric model, a boolean model and a mixed-type problem model, which
covers fundamental programming scenarios. In all tested cases, the algorithm demon-
strated an ability to create satisfying functional and errorless solutions that met all user-
defined requirements. Even though the generated programs were burdened with redundant
objects and were much more complex compared to programmer-developed codes, this fact
has no effect on the code’s execution speed or accuracy. Based on the achieved results, we
can conclude that this pilot study not only proved the feasibility and viability of the proposed
concept, but also showed promising results in solving linear and binary programming tasks.
Furthermore, the results revealed that with further research, this poorly explored field could
become a powerful tool not only for application developers but also for non-programmers
and low-skilled users.
Introduction
Graphical programming refers to a category of programming languages that use visual repre-
sentations, such as icons, symbols, diagrams, or other graphical elements, to facilitate the
design and creation of software applications. Unlike traditional text-based programming
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Kodytek P, Bodzas A, Zidek J (2024)
Automated code development based on genetic
programming in graphical programming language:
A pilot study. PLoS ONE 19(3): e0299456. https://
doi.org/10.1371/journal.pone.0299456
Editor: Govind Vashishtha, Wroclaw University of
Science and Technology: Politechnika Wroclawska,
POLAND
Received: December 15, 2023
Accepted: February 10, 2024
Published: March 7, 2024
Copyright: © 2024 Kodytek et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: The code supporting
this study is available from https://zenodo.org/
records/10542753.
Funding: This work was supported by the “Student
Grant System” of VSB-TU Ostrava, project number
SP2022/88. The funders had no role in study
design, data collection and analysis, decision to
publish, or preparation of the manuscript.
Competing interests: The authors have declared
that no competing interests exist.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
1 / 20
PLOS ONE A pilot study
languages, where the code, i.e., textual commands, are written in text editors or integrated
development environments, graphical programming allows users to interactively create pro-
grams by manipulating and connecting graphical elements. Since graphical programming
does not require a strong understanding of the language and its syntax, these languages are
often designed to make programming more intuitive and accessible to non-programmers.
Automated code development in LabVIEW or any other graphical programming environ-
ment is inspired by reversing a standard software development model. This engineering design
process can be perceived as a methodical series of steps that allow programmers to create func-
tional products and processes [1]. This process can be highly repetitive, and certain stages
often require multiple iterations before proceeding to the next step. Since requirements-based
testing and validation, also known as test-driven development, is a common and essential part
of software development [2] and a standard procedure for programmers who must verify the
code’s functionality, by reversing this process, we can automatically generate code instead of
developing programs or unit testing frameworks. In this reverse scenario, we can automatically
create programs based on the predefined input requirements, and by backpropagating the
input- output differences, we can modify the generated code until all requirements are satis-
fied. By transforming this task into a fully automated process, we can therefore fundamentally
reshape the development principles for basic programs, and instead of employing human
experts for code development and test report validation, we can utilize computers to generate
programs and evaluate test reports.
Automated code generation in textual-based environments has been used in the software
industry for decades [3], and especially in recent years, many novel program generation
approaches have been proposed and evaluated on common benchmark problems [4]. These
approaches to code generation employ various techniques, including artificial intelligence,
machine learning, or genetic evolution methods, to repair or generate efficient and error-free
codes. A significant research direction in this field involves the use of machine learning, espe-
cially neural network models. Most of these studies employed recurrent neural networks [5–
7], transformer models [8,9], or convolutional neural networks [10,11], which are able to learn
patterns and structures from given code samples. By training these models on large code
repositories, they can capture syntax, semantics, and even higher-level programming con-
structs, which enables them to generate usable code [12].
Another popular direction for automated code generation includes evolutionary algo-
rithms, particularly genetic algorithms [13] and genetic programming. Genetic programming
(GP) is technically regarded as a special evolutionary algorithm inspired by Darwin’s evolu-
tionary theory, where algorithms are characterized by the existence of a population of individ-
uals exposed to various environmental circumstances that lead to natural selection. [14]
Unlike genetic algorithms, in genetic programming, the individuals in the population are
computational programs, which are typically represented as sequences of instructions or
expression trees. [15] These populations are iteratively transformed and evolved over genera-
tions into other populations by applying genetic operations to aproximate or find a solution to
a specific problem. [16] To measure the degree of adaptation of individuals to the environ-
ment, usually a fitness function is employed [16]. However, also other alternative measures
could be utilized, for example, Wasserstein distance [17] for probability distribution outputs,
or Single-valued Neutrosophic Cross-Entropy [18], which measures the dissimilarity in cases
of uncertain or incomplete information. This approach to code generation is primarily useful
for optimizing codes for specific tasks or constraints and usually require user-defined input/
output examples. However, studies using combinations of input/output examples with natural
language descriptions can also be found [19]. One of the biggest advantages of using genetic
algorithms based on user-defined requirements, which is the primary focus of this study, is the
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
2 / 20
PLOS ONE A pilot study
potential usage of such a system by non-programmers. A well-known case of such a program
is, for example, Flash Fill, which is one of the most used data tools integrated into Microsoft
Excel that is able to automatically fill out the data in sheets by using predictive technology. [20]
According to the comprehensive survey on program synthesis with evolutionary algorithms
conducted by Sobania et al., the most frequently used approaches for code generation involve
stack-based GP, using mostly Push as a representation language, grammar-guided GP (includ-
ing tree-based and linearized grammar-based approaches), and linear GP [21] Although the
stack-based GP approach makes up the largest proportion of the identified studies (37 in-
scope papers), due to its most common language representation (Forth, Push, or Postscript
programming language), it is not considered relevant in real-world software projects, espe-
cially from the perspective of software development [21]. Other meta-heuristic algorithms
introduced in specific fields of automated programming, especially regarding optimization
tasks, may include the particle swarm optimizer, gravitational search algorithm, artificial bee
colony algorithm, grey wolf optimizer [22], or differential evolution [23] and slime mould
algorithm [24,25].
Even though numerous code generation methods have been proposed for textual-based lan-
guages in the last few decades [26,27], to the best of the authors’ knowledge, none of these
methods have been implemented in graphical-based programming languages, especially Lab-
VIEW. Moreover, the implementation of the actual state-of-the-art methods in a graphical lan-
guage is rather inefficient and almost impossible without using text-to-object converters since
all existing algorithms are primarily designed for text-based languages. Due to the fact that
most of the graphical programming environments, including LabVIEW, do not even have
such converters or do not support textual compilers, the main aim of this study is to create a
proof of concept for a yet unexplored automated GP-based framework for code generation in
the graphical programming environment LabVIEW. The proposed code generation frame-
work fully depends on the input requirements, which can be defined even by users without
any prior programming knowledge. The entire framework was tested on four basic benchmark
problems encompassing fundamental data types, such as string, boolean, numeric, and their
combinations. The achieved evaluation results not only demonstrated the algorithm’s ability
to generate functional programs in string, numeric and boolean domains, but also proved that
the algorithm is able to work in a solution space that isn’t strongly typed, and therefore, can
lead to universal solutions. Although the generated solutions showed a significant degree of
complexity in comparison to programs written by SW developers, the outcomes of this study
prove the feasibility of this idea, where even non-programmers and low-skilled users could
efficiently generate programs. The proposed approach in this study is, therefore, the first of its
kind in this research field and may serve as a good starting point and inspiration for research-
ers and programmers working with graphical programming languages. Since graphically ori-
ented programming is recently on the rise with the growing industry 4.0, where PLC-based
systems and fast test-developing environments such as LabVIEW or Teststand play a signifi-
cant role, introducing such algorithms to these environments would help to solve many
human-restricted problems.
Materials and methods
LabVIEW, as a graphical development environment, utilizes a different code representation in
comparison to traditional text-based programming languages. This representation involves
indexing functions, which cannot be translated directly into human-readable and understand-
able text as in typical text-based environments. This difference in representation complicates
the usage of text-based language prediction models like GPT-3 [28] or other commonly used
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
3 / 20
PLOS ONE A pilot study
models for generating code. Hence, in this scenario, genetic programming appears to be the
most suitable approach, which evolves and optimizes graphical structures toward user-defined
requirements. The following chapters are devoted to the LabVIEW programming environ-
ment and its language syntax and provide a detailed description of the proposed code genera-
tion framework.
Labview development environment language syntax
LabVIEW, short for Laboratory Virtual Instrument Engineering Workbench, is a visual pro-
gramming environment commonly used in measurement, automation, or control applica-
tions. Unlike traditional text-based programming languages, LabVIEW utilizes a graphical
data flow programming paradigm where the code is represented in the form of interconnected
graphical elements called virtual instruments, denoted as VIs, that can be perceived as func-
tions or subroutines in conventional programming languages. The abstract syntax is typically
represented as a data flow graph or a block diagram, where each VI or block is a self-contained
unit of code or a native function. The execution of the program is then conceptualized as a
flow of data, where variables are propagated via the wires through a sequential series of con-
nected functions, which execute as soon as all data is available on the inputs.
The core of each VI is divided into two interdependent parts: the front panel, which repre-
sents the user interface, and the block diagram, i.e., the code, responsible for the program’s
logic and functionality. The visual representation of both parts of a simple part of a code is
demonstrated in Fig 1.
If we analyze this part of the code from the programmer’s viewpoint, the created method in
LabVIEW carries the name add_pi and has one numeric input as a parameter. The output of
this function is then an input value increased by the value of π. However, in a much deeper
sense of the language, the created program contains four basic objects (two input objects, a
function, and an output), where each object is represented by a specific icon in the block dia-
gram. These objects can be considered instances of objects in object-oriented programming,
and therefore each of the four elements contains its own private data and methods (such as
Fig 1. The created front panel and block diagram for a virtual device called "add_pi".
https://doi.org/10.1371/journal.pone.0299456.g001
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
4 / 20
PLOS ONE A pilot study
captions, labels, or set-value methods). Since every single object inserted in the block diagram
is a child of a prime class called LabVIEW Object, each element created in the block diagram
and also on the front panel is a child of this class. Hence, if we are able to refer to any object by
using a pointer, we can also programmatically change its publicly accessible data or invoke its
publicly accessible methods. It is also possible to get this reference through all objects con-
tained in the data of our "main" object, which is our program. Through the reference, we are
therefore able to obtain references to all front panel or block diagram objects.
Additionally, apart from the four main objects, there are three secondary objects in the
given diagram that are presented as connections or wires between the functions, which ensure
the data flow of the program. The main difference between a main object and a secondary
object is that a secondary object reference can only be obtained from the main object reference
to which the secondary object is assigned (connected). Even though for most LabVIEW pro-
grammers, this object is just a simple wire connecting two blocks or functions, it is a sophisti-
cated class that, in its private parameters, stores information about its description, the
connected terminals, state, program pauses, connection points, or references to the main
objects. Moreover, this class allows navigating the program from one place to another by using
a set of obtained references to various block diagrams or front panel objects.
Although most programmers do not use this information and do not need to understand
these concepts, it is important knowledge that allows performing automated program develop-
ment tasks in the LabVIEW programming environment.
LabVIEW scripting. An essential LabVIEW feature that was used in this work is a Lab-
VIEW VI Scripting software add-on, which provides a set of functions used to access advanced
private methods and information that is normally not available to the user. This includes func-
tions that are used to perform code analysis, editing, or even code creation. Nowadays, many
leading developers use these features to create templates or automatically generate frameworks
for other developers; however, this work deals only with methods allowing to programmati-
cally insert and connect objects within the block diagram. The particular LabVIEW scripting
processes employed in this work are depicted in Table 1.
As it might be seen, scripting is a diverse and powerful tool for modifying the final program,
and by combining these functions with built-in LabVIEW functions responsible for the run of
the program, assignment of the values to inputs, and their reading or evaluation, we can obtain
a tool that is an essential part of automatic program development, and which enables us to cre-
ate or modify the particular programs.
The proposed framework
In this study, we approach the problem of automated code development in a similar way as
human evolution works. Each generated VI, which is a final representation of a program, can
Table 1. Basic types of LabVIEW scripting functions used for the purpose of automated code generation.
Process
Type
Navigation
Detail
Function
Between function and
wire
Between wire and
function
New VI
Trace dependencies and connections from the function to the wire
Trace dependencies and connections from the wire to the function
Creates (insert) new method/function
Creation
Object on a
front panel/block diagram
Creates (insert) new objects (native LV functions, controls) in the
program
Wire
Creates (insert) connections between functions
Object positioning
Changes the position of objects inside a block diagram
https://doi.org/10.1371/journal.pone.0299456.t001
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
5 / 20
PLOS ONE A pilot study
Fig 2. The entire sequence of the proposed code generation framework.
https://doi.org/10.1371/journal.pone.0299456.g002
be seen as a human phenotype that represents the complete characteristics of an individual from
the generation. Since only the creation and navigation scripting methods are utilized for the code
generation, the whole code information can be obtained in two sets, the Wirer and the Creator,
which are thoroughly described in the chapter Genetic structure of the program. The process of
the proposed code generation approach, from requirement definition to the formation of a new
generation is depicted in Fig 2 and described step by step in the subsequent chapters.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
6 / 20
PLOS ONE A pilot study
Genetic structure of the program. The Wirer and the Creator can be perceived as chro-
mosomes in human cells, and just as human chromosomes, the Wirer and the Creator set con-
tain a collection of genes, where each gene, in our case, carries specific information about the
structure (the created functions or other objects such as constants) or binding (connections
between the program building blocks). Therefore, whereas the creator is responsible for the
insertion of functions on the block diagram by using the creation scripting method, the Wir-
erer is responsible for tracing the functions’ inputs and outputs and for creating connections
between the particular functions on the block diagram (performed by utilizing the navigation
and creation LV scripting methods). The proposed complete genetic structure is demonstrated
in Fig 3.
Similarly, as the individual genes create cell characteristics by using sub-alleles, the genes
contained in the Creator and the Wirer create the final form and the behavior of the generated
code. Each gene in both sets, therefore, contains an identifier in the form of an unsigned 8-bit
integer that is used to assign the gene a specific function (such as a mathematical operation,
equal function, or string function) and data in the form of a byte array, which contains all nec-
essary information about the particular element defined by the identifier. For instance, if the
gene is mapped as a string constant (defined by the identifier number 0), then the
Fig 3. The proposed genetic structure of programs.
https://doi.org/10.1371/journal.pone.0299456.g003
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
7 / 20
PLOS ONE A pilot study
corresponding data within the gene is converted from a byte array back to a string. On the
other hand, in the case of a mathematical function (identifier 3), only the first byte of the data
carries information about the pointer to a specific numeric function, such as addition (value
0), subtraction (value 1), etc. A detailed explanation of how genetic identifiers are mapped into
specific functions is provided in Chapter Initialization process, Table 4. By employing byte
array data representation for the data within the Wirer and the Creator, we can furthermore
meet the possible requirements for the infinite number of genes in both sets, which ensures no
limitations by unnecessary conditions during the program development.
Although the Wirerer contains the same genetic structure as a creator, its only function is
to create connections between the existing functions. To connect an output of one function
with an unconnected input of another function, the navigation LabVIEW scripting method
only requires references to the corresponding input and output terminals of the involved func-
tions. Due to this fact, the Wirer does not use the information obtained in the identifier, and
only the connector’s information is extracted from the data, where the first and second ele-
ments of a byte array are converted to an integer value representing the indexes of the corre-
sponding output and input terminals.
Definition of system requirements. To ensure that the system functions in accordance
with the abovementioned goals, it is necessary for the user to have the possibility of defining
the input and output variables of the system. According to the requirement definition, a funda-
mental requirement for our system was to maintain human readability and be easy to under-
stand so that even non-programmers could generate codes. The proposed system in this work
supports four basic data types, namely boolean (Bool), string (Str), double (Dbl), and integer
(Int). All the user’s functionality requirements are stored in the form of an array of elements,
where each element is, like each gene, represented by a control name and an array of values for
a given parameter. These parameters are then, at the beginning, loaded into the system and
stored in a functional global variable (a frequently used LabVIEW design pattern allowing con-
trolled access to data) so they can be reused in the evaluation process. Since we used a byte
array for storing the data within the genes, there is no need to address the issue of different
functions’ data type compatibility requirements at this program level. The evaluation function
then accepts an array of inputs and outputs.
The input interface for a common user is realized by using the freely accessible library
CLAUDIE_XLSX (Compact Library and Universal Data Import Export xlsx), which allows
writing and reading data from Microsoft Excel [29]. The file structure was selected in a way
that facilitates the user’s ability to enter his requests in the form of input combinations and
their required outputs. An example of an implemented structure for a specific task is demon-
strated in Table 2.
The first file line defines whether the element is an input (control) or an output (indica-
tor), the second line defines its data type (bool, dbl, int, str), the third line declares the name
Table 2. Input values example for a numerical problem.
in
dbl
Num_1
5
6
15
17
36
in
dbl
Num_2
10
11
42
43
44
https://doi.org/10.1371/journal.pone.0299456.t002
out
Dbl
Num_Ind
15
7
57
60
80
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
8 / 20
PLOS ONE A pilot study
Table 3. The used population parameters with the corresponding parameter values.
Parameter
Population size
Complexity
Identifier
Value
Fixed value
Randomized with Gaussian distribution–mean selected and sigma 0.1
Randomized in a value range of 0–8 according to Table 4 with a uniform distribution
Data within the Creator
Generated byte array of U8 values with uniform distribution.
Data within the Wirerer
Two random U8 values with a uniform distribution in an array
https://doi.org/10.1371/journal.pone.0299456.t003
of the element and the following lines create the inputs together with the desired output
values.
Initialization process. The initialization of default parameters is a critical and challenging
problem in evolutionary algorithms. Within this step, the algorithm creates the initial popula-
tion by generating individuals with constrained random parameters for the subsequent evolu-
tionary process. All population parameters with particular value ranges are depicted in Table 3.
Since it is usually not very common to control the size of the population [30], this parame-
ter is in this work set to be optional for the users, and for all performed tests, it was set to a con-
stant value, usually 1000. The second adjustable population parameter used in this study is the
complexity parameter (set in the range from 3 to 20, depending on the problem’s complexity).
This number represents the mean value of a Gaussian curve, which is formed by the number
of genes generated for each descendant in a population. In other words, the number of genes
in a generation likely corresponds to a Gaussian curve with a defined standard deviation of 0.1
and a mean equal to the selected complexity value. This "randomness" ensures a better distri-
bution of the generated code possibilities already in the first iteration and prevents the algo-
rithm from getting stuck at the local minimum [31].
To assign the gene a LabVIEW object (function, control, or a constant), we proposed a
mapping table (refer to Table 4), which takes a randomly generated number of an identifier in
a defined range and, based on its value, assigns the identifier a specific LabVIEW building
block represented by a native LabVIEW ID class number. These LabVIEW ID numbers can be
perceived as inner environment identifiers for particular parent classes, where each LabVIEW
building block belongs to a specific parent class. Even though the knowledge of the predefined
LabVIEW classes is not required for common programming tasks, it is part of the basic con-
cept of the environment.
The problem that we encountered with this LabVIEW inner categorization was a multiple
occurrence of the class with an identifier of 16400 for significantly different functions. Even
Table 4. An overview of the used identifiers and their relation to specific block diagram objects and data usage.
Gene identificator
Num ID in LV
Building block/Object
Data usage
0
1
2
3
4
5
6
7
8
16392
16395
16476
16400
16390
16422
16400_1
16429
16400_2
String Constant
Control Terminal
Bundler
Math. Function
Converts byte array to string
Gene data not used
Gene data not used
Maps the first byte of the data array to a mathematic function
Digital Numeric Constant
Converts byte array to a number
Boolean constant
Bool Function
Equal Function
Select Function
Gene data not used
Maps the first byte of the data array to a boolean function
Gene data not used
Gene data not used
https://doi.org/10.1371/journal.pone.0299456.t004
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
9 / 20
PLOS ONE A pilot study
though the parent is identical for different descendants, such as mathematical functions, bool-
ean functions, and the function select, which is our case, these functions do not have the same
inputs, outputs, or meaning. To solve this problem, we divided these descendants into individ-
ual subclasses. This means that the gene identifier doesn’t contain the inner LabVIEW class
number but instead contains an extended unique identifier that differentiates the problematic
classes. The following Table 4 shows the relation between the identification numbers and the
corresponding LabVIEW functions and explains the usage of the gene data for the particular
identifiers.
Another concept that we employed during the initialization process in this work is strongly
typed genetic evolution. Unlike classical programming, where any function can be inserted
into the code, STGE limits the number of available functions. This means that only functions
with a corresponding data type to the selected data type are available. An exception is a bool-
ean data type, where, for instance, the "equal" function is a dynamic function that can be used
by all data types. It is important to note that since this approach reduces the available func-
tions, the mapping table (Table 4) dynamically changes according to the allowed data types.
Introducing this procedure directly affects the values that can be written into the gene’s identi-
fiers, and so by limiting these values, STGE helps speed up the evolution process and signifi-
cantly increases the chances for a successful evolution [32].
Wirer and Creator data generation. Simultaneously with the generation of the identifi-
ers for the Wirer and Creator sets, another parallel process produces, for each created iden-
tifier, gene data in the form of a U8-byte array. For the generation of the data within Wirer
genes, we applied an initialization rule where two numbers from the range of <0, Gauss
(complexity; 0,1)> were selected on the basis of the uniform probability. This rule ensures
that each element of the program has an equal chance of being joined with any other ele-
ment. These generated numbers are then inserted into a U8-byte array that represents the
particular gene data. On the other hand, for the data in the Creator, we had to consider all
the possibilities that may arise on a theoretical level. Since the Creator assigns information
(the generated data) to a created function on the basis of the function’s identifier, there is a
probability of assigning data to functions that do not need that information, such as select
or equal functions (refer to Table 4 for functions where gene data are not applicable). In a
practical application, this case doesn’t seem to be a big problem, but in a worst-case sce-
nario, the Creator might assign data to a string constant where the user is expecting a spe-
cific input (the infinite monkey theorem). To prevent this occurrence, the algorithm for
generating Creator data generates a sequence of numbers in a range of 0–255 based on an
even distribution of probabilities until the algorithm meets the defined stopping condition.
Unlike in the case of the Wirer, this process, therefore, continues to a potential infinity as
long as another randomly generated number from the interval <0; 1> is greater than 0.95i,
where i is the length of the currently generated string. This approach theoretically enables
infinitely large text and ensures maximal variability.
Creation of new child. The creation of a new child from the individual Creator and
Wirer genes is implemented in this work sequentially, where each gene is processed separately.
At first, all genes from the Creator are processed, which means that all program building
blocks, i.e., functions or constants, are created based on the gene identifier and data by calling
the LV scripting creation function. All created functions are, during this process, placed on a
block diagram at a random place, so the current code doesn’t have to meet the programming
standards yet (clean code without unnecessary bends in block diagram wires, top-down and
left-right data flow layout, and many more). In the case of block diagram constants, the algo-
rithm also handles empty gene data by generating a random constant value. Although this case
doesn’t apply to the initialization phase, during the evolution, the genes might lose the genetic
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
10 / 20
PLOS ONE A pilot study
material within the data. After creating all the building blocks on the block diagram, the algo-
rithm processes the Wirer genes and creates links between the block diagram functions, i.e.,
between the output of the first object and the input of the second object. During this pro-
cess, the algorithm extracts all objects’ input and output terminals by using the LabVIEW
scripting navigation method, and from the extracted information (references to terminals),
it creates two arrays representing block diagram objects’ inputs and outputs. The final con-
nection of two functions or objects on the block diagram is then realized by creating wires
between the indexed elements from the above-mentioned arrays, where the array index
numbers are obtained from the gene data. To increase the chance of finding a suitable con-
nection, we complemented this process with a controlled selection, where the whole array
of input terminals is filtered based on the output terminal data type. Since in the LabVIEW
programming environment it is possible to connect two different terminals with different
data types, which results in a broken wire and non-executable programs, by employing this
filtration, only input terminals with a corresponding data type to a selected output terminal
are preserved.
In the last phase of the process of creating a new child, the generated section of code is eval-
uated for its functionality. During this evaluation, the program might not pass for several rea-
sons. The main reason for failure is a poor logical interconnection between the functions. This
can be caused by unconnected terminals, broken wires, introduced feedback loops (where a
function input is linked to the same function output, causing a delay in the output of the exe-
cution), or by connecting multiple outputs to one input (which is a problem specific to graphi-
cal programming, unlike text-based languages). To eliminate this problem, we implemented
an additional sequential algorithm that traces all wires in a block diagram by using a built-in
LabVIEW method and then deletes all the incorrectly connected wires. Afterward, for all
unconnected user-defined inputs, the algorithm establishes a random connection between a
particular input and an existing output with a corresponding data type. Additionally, in the
case of any missing required function input, the algorithm creates an appropriate constant
with a corresponding data type and connects the constant to a desired input. This process
ensures a higher success rate for creating valid and executable code and improves the chances
of algorithm convergence.
Evaluation of child. A key component that significantly improved the algorithm’s results
was the implementation of a back-analysis of the created child. Due to the fact that the above-
mentioned process generates random connections and possible constants, it was essential to
store this information in a gene pool of the Wirerer and the Creator. Storing this information
in the form of newly created genes primarily prevented the descendants with the best results
from the development deterioration. The proposed back-analysis algorithm validates the func-
tionality of the particular code samples against the user-defined requirements and is based on
the following formula:
Err ¼
Xinf
i¼0
Pð�xiÞ (cid:0) Sð �xi; W; CÞ
ð1Þ
where the resulting error value Err is the sum of all partial differences between the desired
value (output) P for the selected combination of inputs and the obtained values S (actual out-
puts), dependent on the values of W and C representing the sets Wirer and Creator.
Although this task is not challenging according to the development, the performance of this
process is the most time-consuming. Since the program’s user-defined inputs and outputs are
in the form of arrays, the evaluation process has to be repeated for each child as many times as
many input combinations the user defined. Within the evaluation, we implemented a different
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
11 / 20
PLOS ONE A pilot study
evaluation logic for each data type in the program. Operations working with a boolean data
type result in a "binary" value of a deviation, where 0 stands for a case of a match (T = T,
F = F), and the value of 1 indicates a mismatch. On the other hand, the resulting deviation in
numeric data type is equal to the absolute value of the difference between the input and output
values. Text strings use an equivalent logic, but the deviation is calculated as the sum of all par-
tial results for each ASCII character converted to a U8 data type. All error results for each
descendant are then stored in an array for further processing.
Creation of a new generation. At the beginning of the evolutionary process, we have to
select suitable parents for the new descendants. This task is achieved by sorting the evaluation
results by the smallest number and selecting the best available results with the lowest score, i.e.,
the lowest difference between the achieved and the desired output. Descendants with the best
results then become parents. The number of selected descendants for the next evolution was
determined by the trial-and-error process and, for the majority of the experiments, was set to
6. To avoid generating similar local maxima due to identical evaluation results, only the first
occurrence of the duplicate values is selected. In this study, one-to-one inheritance was
employed, so each child has exactly one parent [33].
After selecting the best parents, the algorithms finally proceed to mutation-based evolution
by applying mutation to the Creator and Wirer genes. In the case of the mutation of Creator
genes, a random value from a range of <0, 100> is generated for each gene in the set. The indi-
vidual genes are then modified only if the generated value is less than 25. This modification
applies to the identifier, which affects the change of the function, as well as to partial values in
the gene data (every single element in the byte array). In the case of the gene modification with
a 1/4 probability, there is an additional 25% chance that the current identifier or data element
value is decreased by a randomly generated number in a range of 1–3, a 25% chance that the
current value is increased by the same number, and a 50% chance that the particular gene
value is not changed. The overall probability of changing the gene within the Creator set is,
therefore, 12.5%. A similar logic applies also to the evolution of the Wirerer genes, only with
different modification probability values. The Wirer genes, in case of a mutation (also a 25%
probability), are in one-third of cases incremented by a number from 1 to 100% of their origi-
nal value; in 1/3 of cases, the genes are decreased by this number, and there is the same proba-
bility that the genes are not changed. The total probability of the mutation of Wirer genes then
amounts to approximately 16.6%. This approach to mutation within the Wirer genes ensured
higher variability and induced sufficiently large changes in values. Since the Wirer does not
use the information obtained in the identifier, this operation affects only the gene data ele-
ments, specifically the output and input indexes. The whole process of evolution is depicted in
a diagram in Fig 4.
The mutation of genes within the Wirer and Creator is followed by the potential creation
and elimination of individual genes, which turned out to be a key feature of a functional code-
generating system. While the above-mentioned steps completely imitate the human process of
mutation (the shrink mutation operator), the creation and elimination of genes slightly deviate
from human evolution. Even though we can observe this process in humans, its manifestation
is rather physiological (the creation of a new phenotype by combining the changed alleles). On
the other hand, the process of code creation is about creating or destroying an existing gene,
which prevents future generations from degradation. The elimination process is performed
with a probability of 15% for each gene and results in the removal of a particular gene from the
gene pool of the specific set. The same probability is also assigned to the process of creating a
new gene, in which a full new gene is added to the gene pool of existing genes. This step is per-
formed for both the Wirerer and Creator genes.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
12 / 20
PLOS ONE A pilot study
Fig 4. The sequence of processing genes during the evolution.
https://doi.org/10.1371/journal.pone.0299456.g004
Experimental evaluation
To verify the proposed framework, we formulated some basic tasks and benchmark problems.
Since the proposed study is designed as a pilot study mainly focused on the feasibility of code
generation in a graphical programming environment, the proposed evaluation consists of only
simple benchmark problems, dealing mostly with linear and binary tasks. Although there are
many benchmark problems available for this purpose, all of them introduce loops and cycles
that are not included in this pilot study. Due to this reason, we have chosen simple benchmark
problems that do not require more complex loops or cycles. The employment of simpler tasks
furthermore allowed us to focus more on the performance evaluation of our algorithms instead
of analyzing problems arising from the code’s complexity. Within the evaluation procedure,
we defined three main prototype problems, including a string model, a numeric model, and a
boolean model.
The string prototype problem verification was performed by selecting a function with one
input and one output. The particular values for this task are given in Table 5. The desired result
of the algorithm is then the addition of a string "ms" to the input value in a numeric format
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
13 / 20
PLOS ONE A pilot study
Table 5. Input and output values for a string prototype problem.
Input Value
Required (Output) Values
100
10
1
500
https://doi.org/10.1371/journal.pone.0299456.t005
100 ms
10 ms
1 ms
500 ms
(string formatting). This task is relatively challenging from the perspective of genetic evolution
and model testing since the function for concatenating strings has to be evolved, and a string
constant with an expected value has to be created at the same time. As we mentioned in previ-
ous chapters, the user-defined string can be of any size, and thus it is necessary to mutate and
evolve the creation, as well as the destruction of the genes within the data.
The second prototype problem refers to a numeric problem and represents a multi-input
task where the required output is the creation of a mathematical function between the two
inputs. Both the selected input values and the expected output values are listed in Table 6. The
most interesting task was the addition of the inputs; hence, this problem was chosen as a typi-
cal task of the system.
The last prototype problem we defined in this study is the boolean model with two boolean
input variables and one boolean output. Within this model, we employed two boolean tasks,
including a logical function OR and an EQUAL function. The assignment for the selected task
is depicted in Table 7. In this case, the main point was to examine the evolution with a limited
ability to verify the outcomes.
Results
The first problem, which was evaluated to confirm the algorithm’s functionality, was a string
problem. The proposed test model involves two problematic key points, which are related to
the conversion of the values inside a data element. This process requires size adjustments as
well as the evolution of the values within the data. The results of multiple algorithm runs are
shown in Fig 5.
Table 6. Input and output values for the numeric prototype problem.
Input Value 1
Input Value 2
Required output
5
6
15
17
36
10
11
42
43
44
https://doi.org/10.1371/journal.pone.0299456.t006
15
17
57
60
80
Table 7. Input and output values for the boolean prototype problem.
Input Value 1
Input Value
2
Required output (Task OR)
Required output
(Task EQUAL)
F
T
F
T
F
F
T
T
https://doi.org/10.1371/journal.pone.0299456.t007
F
T
T
T
T
F
F
T
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
14 / 20
PLOS ONE A pilot study
Fig 5. Final solutions of a string problem for multiple algorithms runs.
https://doi.org/10.1371/journal.pone.0299456.g005
From these results, we can deduce that we can achieve repeatable solutions even with
changes in input parameters. The generated results clearly show the ability of the algorithm to
find the "optimal" program. In solution (A), the generated code is burdened with an extra
function. Nonetheless, this doesn’t influence the code functionality, and the unused parts are
removed during the program compilation. In the case of solution (C), we changed the maxi-
mum complexity input parameter. This parameter was set for the basic operations in a range
of <0; 10> due to higher computational complexity, but in the case of C, we set this parameter
to a value of 30. It is apparent that the resulting solution is already out of the optimal and read-
able code; however, the solution still meets the user’s requirements.
A key element during the evaluation of the program’s evolution was the verification of sim-
ple addition function behavior. Generated solutions, i.e., programs, during the experimental
verification led to clear conclusions and findings that weren’t revealed in more complex math-
ematical functions. The results of this process are demonstrated in Fig 6.
The first problem that can be observed in this solution is the occurrence of different data
types in the code, caused by the disabled usage of strict data types. In the first generation, the
program included many more objects that were related to the boolean or string data type, but
Fig 6. Numeric problem solution. The original form of the generated code (A), the cleaned-up form of the result (B).
https://doi.org/10.1371/journal.pone.0299456.g006
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
15 / 20
PLOS ONE A pilot study
Fig 7. Solution of OR (A) and EQUAL (B) functions of the boolean problem.
https://doi.org/10.1371/journal.pone.0299456.g007
with subsequent generations, the number of these elements decreased until only elements
related to the required data type remained. This was caused by the fact that redundant units of
code (creator genes) are strongly dominated by useful functions. Genetic evolution, therefore,
leads to a selection of genes that do not include these functions. This effect is not strictly domi-
nant, but a positive trend has been observed. Another curious behavior we noticed was the cre-
ation of a function Y = A + B by using three subtraction functions that resulted in an equation
Y = A—((A-B) -B). This result is considered the correct solution to a problem; however, the
solution seems to be extremely complex according to its purpose. This occurrence can be
restricted by limiting the complexity parameter.
In the case of the boolean model, the algorithm was able to achieve satisfying results accord-
ing to the program’s functionality. As can be seen in Fig 7, due to a higher complexity parame-
ter (a value of 10) and the disabled usage of strict data types, the result of the logical OR
function is burdened with a relatively large amount of useless function blocks. On the other
hand, for the second test case, including the EQUAL function, we enabled the usage of strict
data types while preserving the high complexity parameter. The final solution to this task (B)
contains multiple random connections of several equal functions. The result was created
within the first generation, where we set a high complexity (10) and enabled strict data types.
The additional last step of the evaluation comprises testing the usage of all basic data types.
The output of this task should be a function that converts the input numeric value to a logical
value on the basis of the required string output value. The proposed input and output values
are listed in Table 8.
These results (see Fig 8) proved that the proposed algorithm is able to work in a solution
space that isn’t strongly typed, and therefore, it can lead to universal solutions. To improve the
readability of some codes in this work, we additionally cleaned the code by using a built-in
tool that automatically reroutes wires and rearranges block diagram objects. As mentioned
Table 8. Input and output values for testing the combination of data types.
Input value
0
1
2
3
https://doi.org/10.1371/journal.pone.0299456.t008
Required output
Equal to zero
Not equal to zero
Not equal to zero
Not equal to zero
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
16 / 20
PLOS ONE A pilot study
Fig 8. Solution of a combinatory problem. (A) the original code of the solution, (B) cleaned-up solution of the task.
https://doi.org/10.1371/journal.pone.0299456.g008
before, graphical programming in LabVIEW is based on placing and connecting objects, such
as functions, constants, or terminals, on a program’s block diagram. Since during automatic
code generation, the user has no control over the functions’ positions or their connections
(bends in wires), such automatically generated code does not meet the best coding practices or
readability standards, which have to follow the top-to-bottom and left-to-right dataflow para-
digm. Therefore, by using the embedded automated cleanup functionality, which is able to
adjust the spacing, remove bends in wires, or logically rearrange block diagram objects, the
code can meet at least the essential requirements and programming standards.
Another important aspect we analyzed during the technical evaluation was the algorithm’s
speed and efficiency. This estimation was realized by measuring the independent processing
times of the Creator, Wirer, as well as the evaluation process. To get a good estimation, the
processing time for each part was averaged over multiple generations with a population size of
1000 and a set complexity value of 10. The estimated processing times for all phases can be
seen in Table 9.
Discussion
The main goal of this pilot study was to create a proof-of-concept for an automated code gen-
eration approach within the graphical-based programming language LabVIEW. Since, to the
best of the authors’ knowledge, none of the automated code generation methods have been
implemented in graphical-based programming languages, especially LabVIEW, this study
aims to prove the feasibility and practical potential of this proposed concept. For this purpose,
we designed and developed a requirement-based automated code generation algorithm that is
Table 9. Average processing times for the individual processes within the evolution.
Tested benchmark model
Creator process duration (ms)
Wirer process duration (ms)
Evaluation process duration (ms)
String
Numeric
Boolean
Mixed type
20.08
19.40
24.25
28.05
https://doi.org/10.1371/journal.pone.0299456.t009
86.00
63.28
85.92
62.68
19.16
28.86
45.20
30.29
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
17 / 20
PLOS ONE A pilot study
able to provide functional solutions on the basis of user-defined requirements. These input
requirements can be defined by experienced software developers but also by non-program-
mers or users with little programming experience. The proposed framework in this study was
tested on four different benchmark problems that were designed to assess the framework’s
ability to generate error-free, functional, and efficient codes across various data types. The pro-
posed string problem model tested the framework’s capability in text manipulation and string
operation tasks, the numeric problem model focused on arithmetic operations and handling
numerical data, the boolean model dealt with logical conditions and provided insight into the
framework’s decision-making processes, and the last mixed-type problem model tested the
framework’s versatility in handling multiple data types. The performance of the proposed
framework was then evaluated by assessing the generated codes’ accuracy, complexity, execu-
tion speed, or adherence to user requirements.
During the experimental evaluation, the designed code generation system achieved success
not only in “hill-climbing” tasks, where we were able to find appropriate solutions with a gra-
dient ascent algorithm, but also in one-point search problems, such as the boolean problem.
Moreover, the mutation genetic operator, in combination with the proposed approach has
been identified as a proper strategy for creating connections in graphical programs. Based on
the achieved results, we have to point out that even though the algorithm was, in all tasks, able
to find a functional and errorless solution that met all input requirements, these solutions were
much more complex and were burdened with redundant objects in comparison to program-
mer-developed codes. The complexity of the code, in this case, can be interpreted as a higher
number of required connected objects that create the final solution. This is mainly caused by
the natural behavior of evolutionary algorithms and genetic programming, which focus more
on finding the best solution than the optimal solution. Due to this fact, the algorithm proposed
in this study might find multiple different solutions that are not optimal, even if they fully sat-
isfy the defined requirements. However, this level of complexity did not affect the accuracy or
execution speed of the generated code compared to manually written codes.
Although we are able to demonstrate the satisfactory functionality of the proposed method
and prove the feasibility of this concept, we cannot fully declare that the problem is solved.
Furthermore, additional research in this field has to be conducted, especially regarding the
optimization and finding optimal solutions. Another remarkable fact revealed by the experi-
ments is that a significant portion of the computation time for creating a single child is taken
by the Wirer, i.e., by creating interconnections. To optimize this process in the future, it would
be beneficial to develop a more sophisticated algorithm that better reflects the actual input and
output requirements of the created functions instead of only processing the genetic material in
genes and randomizing connections in cases of non-valid solutions. Future research should
also be devoted to the implementation of more complex structures, such as cycles or loops, so
we can fully exploit all the strengths of the proposed solution and test the methodology on
more complex benchmark problems. Moreover, this would allow researchers to compare the
proposed approach with existing methods. In addition, exploring multi-parental genetic pro-
gramming or some types of polygamy-based algorithms can become an important area for
future research. Future research should also consider the potential benefits of using cloud
computing since the search space could reach enormous dimensions. This would enable
researchers to create thousands or even millions of programs in a second, which could be vali-
dated and iterated over the best of the best results to find the final solution to much more com-
plex problems.Based on these findings, we can conclude that this pilot study not only proved
the feasibility of automated code development in graphically oriented programming languages
but also built a strong foundation for further research in this relatively unexplored domain.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
18 / 20
PLOS ONE A pilot study
Author Contributions
Conceptualization: Pavel Kodytek.
Formal analysis: Alexandra Bodzas.
Methodology: Pavel Kodytek.
Project administration: Jan Zidek.
Software: Pavel Kodytek.
Validation: Alexandra Bodzas, Jan Zidek.
Writing – original draft: Pavel Kodytek.
Writing – review & editing: Alexandra Bodzas.
References
1. Gurung G, Shah R, Jaiswal DP. Software Development Life Cycle Models-A Comparative Study. int. j.
sci. res. comput. sci. eng. inf. technol. 2020; 6: 30–37.
2. Ghafari M, Gross T, Fucci D, Felderer M. Why Research on Test-Driven Development is Inconclusive?
Proceedings of the 14th ACM / IEEE International Symposium on ESEM. 2020; 1–10.
3. Basin D, Deville Y, Flener P, Hamfelt A, Fischer Nilsson J. Synthesis of programs in computational
logic. In: Bruynooghe M, Lau K, editors. Program Development in Computational Logic: A Decade of
Research Advances in Logic-Based Program Development. Heidelberg: Springer Berlin; 2004. pp.
30–65.
4.
Liu Y, Tantithamthavorn CK, Liu Y, Li L. On the reliability and explainability of automated code genera-
tion approaches. ACM Trans. Softw. Eng. Methodol.2024. https://doi.org/10.1145/3641540
5. Hu X, Li G, Xia X, Lo D, Jin Z. Deep code comment generation. Proceedings of the 26th ICPC. 2018;
200–210.
6.
Zhu Z, Xue Z, Yuan Z. Automatic graphics program generation using attention-based hierarchical
decoder. ACCV. 2018; 181–196.
7. Watson C, Tufano M, Moran K, Bavota G, Poshyvanyk D. On learning meaningful assert statements for
unit test cases. Proc. ACM/IEEE 42nd Int. Conf. Softw. Eng. 2020; 1398–1409.
8. Gemmell C, Rossetto F, Dalton J. Relevance transformer: Generating concise code snippets with rele-
vance feedback. Proc. 43rd Int. ACM SIGIR Conf. Res. Develop. Inf. Retr. 2020; 2005–2008.
9. Yang G, Chen X, Liu K, Chen Y. Fine-grained Pseudo-code Generation Method via Code Feature
Extraction and Transformer. 2021 28th APSEC. 2021; 213–222.
10. Asiroglu B, et al. Automatic HTML code generation from mock-up images using machine learning tech-
niques. 2019 Proc. Sci. Meeting Elect.-Electron. Biomed. Eng. Comput. Sci. (EBBT). 2019; 1–4.
11.
LeClair A, Haque S, Wu L, McMillan C. Improved code summarization via a graph neural network. Proc.
28th Int. Conf. Program Comprehension. 2020; 184–195.
12. Priya R, Wang X, Hu Y, Sun Y. A deep dive into automatic code generation using character based
recurrent neural networks. 2017 International Conference CSCI. 2017; 369–374.
13. Vashishtha G, Kumar R. An effective health indicator for the Pelton wheel using a Levy flight mutated
genetic algorithm. Meas. Sci. Technol. 2021; 32: 094003.
14. Caˆmara D. Bio-inspired networking. 1st ed. London: Iste Press Elsevier; 2015.
15. Banzhaf W, Francon DF, Keller RE, Nordin P. Genetic programming: An introduction: on the automatic
evolution of computer programs and its applications. San Francisco: Morgan Kaufmann Publishers
Inc.; 1998.
16. Banzhaf W. Evolutionary Computation and Genetic Programming. In: Lakhtakia A, Martı´n-Palma RJ,
editors. Engineered Biomimicry. Amsterdam: Elsevier; 2013. pp. 429–447.
17. Vashishtha G, Kumar R. Unsupervised Learning Model of Sparse Filtering Enhanced Using Wasser-
stein Distance for Intelligent Fault Diagnosis. J. Vib. Eng. Technol. 2023; 11: 2985–3002.
18. Vashishtha G, Chauhan S, Yadav N, Kumar A, Kumar R. A two-level adaptive chirp mode decomposi-
tion and tangent entropy in estimation of single-valued neutrosophic cross-entropy for detecting impeller
defects in centrifugal pump. Appl. Acoust. 2022; 197: 108905.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
19 / 20
PLOS ONE A pilot study
19. Hemberg E, Kelly JW, O’Reilly Una-May. On domain knowledge and novelty to improve program syn-
thesis performance with grammatical evolution. Proceeding GECCO. 2019; 1039–1046.
20. Gulwani S, Herna´ ndez-Orallo J, Kitzelmann E, Muggleton SH, Schmid U, Zorn B. Inductive program-
ming meets the real world. Commun. ACM. 2015; 58: 90–99.
21. Sobania D, Schweim D, Rothlauf F. A comprehensive survey on program synthesis with evolutionary
algorithms. IEEE Trans. Evol. Comput. 2022; 27: 82–97.
22. Chauhan S, Vashishtha G. Mutation-based Arithmetic Optimization Algorithm for global optimization.
2021 CONIT. 2021; 1–6.
23. Chauhan S, Vashishtha G, Kumar A, Abualigah L. Conglomeration of Reptile Search Algorithm and Dif-
ferential Evolution Algorithm for Optimal Designing of FIR Filter. 2023; 42: 2986–3007.
24. Chauhan S, Vashishtha G, Kumar A. A symbiosis of arithmetic optimizer with slime mould algorithm for
improving global optimization and conventional design problem. J Supercomput. 2022; 78: 6234–6274.
25. Vashishtha G, Chauhan S, Singh M, Kumar R. Bearing defect identification by swarm decomposition
considering permutation entropy measure and opposition-based slime mould algorithm. Measurement.
2021; 178: 109389.
26. Ahmed A., Azab S, Abdelhamid Y. Source-Code Generation Using Deep Learning: A Survey. In: Moniz
N, Vale Z, Cascalho J, Silva C, Sebastião R, editors. Progress in Artificial Intelligence. Cham: Springer
Nature Switzerland; 2023. pp. 467–482.
27. Dehaerne E, Dey B, Halder S, De Gendt S, Meert W. Code generation using machine learning: A sys-
tematic review. IEEE Access. 2022; 10: 82434–82455.
28. Kalyan KS. A survey of GPT-3 family large language models including ChatGPT and GPT-4. J. Nat.
Lang. Process. 2024; 6: 100048.
29. ATE Systems. Compact Library and Universal Data Import Export Excel Files Toolkit Download. [cited
10 December 2023]. In: NI [Internet]. Available from: https://www.ni.com/cs-cz/support/downloads/
tools-network/download.compact-library-and-universal-data-import-export-excel-files-toolkit.html?
fbclid=IwAR1IqKGrYRokNFEQcRbhZielVC-6f4BV3R7JL9fwqNG98XQV92R05JovH1Y#374309
(2023).
30.
Jassadapakorn C, Chongstitvatana P. Self-adaptation mechanism to control the diversity of the popula-
tion in genetic algorithm. Int. J. Comput. Sci. Inf. Technol. 2011; 3: 111–127.
31. Kazimipour B, Li X, Qin A. A review of population initialization techniques for evolutionary algorithms.
2014 IEEE Congress on Evol. Comput. 2014; 2585–2592.
32. Montana DJ. Strongly Typed Genetic Programming. Evol. Comput. 1995; 3: 199–230.
33. Ashlock W, Ashlock D. Single Parent Genetic Programming. 2005 IEEE Congress on Evol. Comput.
2005; 2: 1172–1179.
PLOS ONE | https://doi.org/10.1371/journal.pone.0299456 March 7, 2024
20 / 20
PLOS ONE
|
10.1371_journal.pwat.0000210
| "RESEARCH ARTICLE\n\nPerformance evaluation and application of\nhost-specific Bacteroidales and mito(...TRUNCATED)
|
10.1371_journal.pwat.0000127
| "RESEARCH ARTICLE\n\nHousehold, neighbourhood and service\nprovider risk factors for piped drinking-(...TRUNCATED)
|
10.1371_journal.pone.0300377
| |
10.1371_journal.pstr.0000095
| "RESEARCH ARTICLE\n\nThe macroeconomic money-nature nexus:\nAre growing money supplies a relevant\no(...TRUNCATED)
|
10.3389_fimmu.2021.689397
| "Supplementary Material for:\n\nCrosstalk between CD11b and Piezo1 mediates macrophage responses\nto(...TRUNCATED)
|
10.1371_journal.pone.0299837
| "RESEARCH ARTICLE\n\nHas sentiment returned to the pre-pandemic\nlevel? A sentiment analysis using U(...TRUNCATED)
|
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 16
Size of downloaded dataset files:
54.4 MB
Size of the auto-converted Parquet files:
54.4 MB
Number of rows:
1,470