Genomic Screening of Antimicrobial Resistance Markers in UK and US Campylobacter Isolates Highlights Stability of Resistance over an 18-Year Period

ABSTRACT Campylobacter jejuni and Campylobacter coli are important bacterial causes of human foodborne illness. Despite several years of reduced antibiotics usage in livestock production in the United Kingdom (UK) and United States (US), a high prevalence of antimicrobial resistance (AMR) persists in Campylobacter. Both countries have instigated genome sequencing-based surveillance programs for Campylobacter, and in this study, we have identified AMR genes in 32,256 C. jejuni and 8,776 C. coli publicly available genome sequences to compare the prevalence and trends of AMR in Campylobacter isolated in the UK and US between 2001 and 2018. AMR markers were detected in 68% of C. coli and 53% of C. jejuni isolates, with 15% of C. coli isolates being multidrug resistant (MDR), compared to only 2% of C. jejuni isolates. The prevalence of aminoglycoside, macrolide, quinolone, and tetracycline resistance remained fairly stable from 2001 to 2018 in both C. jejuni and C. coli, but statistically significant differences were observed between the UK and US. There was a statistically significant higher prevalence of aminoglycoside and tetracycline resistance for US C. coli and C. jejuni isolates and macrolide resistance for US C. coli isolates. In contrast, UK C. coli and C. jejuni isolates showed a significantly higher prevalence of quinolone resistance. Specific multilocus sequence type (MLST) clonal complexes (e.g., ST-353/464) showed >95% quinolone resistance. This large-scale comparison of AMR prevalence has shown that the prevalence of AMR remains stable for Campylobacter in the UK and the US. This suggests that antimicrobial stewardship and restricted antibiotic usage may help contain further expansion of AMR prevalence in Campylobacter but are unlikely to reduce it in the short term.


RESULTS
Comparison of genome assembly-and sequence read-based AMR marker detection. To assess the efficacy of the NCBI AMRFinder software tool with Campylobacter genome sequences, we compared it with a recent AMR analysis of 381 UK C. jejuni and C. coli isolates (13). Painset et al. (13) performed AMR marker detection directly on the Illumina sequencing reads using in-house scripts, whereas we used the NCBI AMRFinder software tool, which requires genomes to be assembled first. There were no false positives detected using the NCBI AMRFinder software for any of the antibiotic resistance classes (Table 1), and there was 100% agreement for detection of quinolone resistance (GyrA D87 mutation) and aminoglycoside resistance markers. Screening for tetracycline resistance resulted in 2/161 tetO-positive isolates that were initially reported negative for the tetO gene using NCBI AMRFinder. Further investigation of these two samples showed the tetO gene to be split over two contigs, which was detectable by secondary screening with BLAST using the Abricate software tool, and this was subsequently done as standard (see Table S1 posted at https://doi.org/10.6084/ m9.figshare.19292126.v1). Finally, macrolide resistance based on 23S rRNA gene mutations resulted in three negative samples previously reported positive (13). However, all three samples contained both the wild-type and mutated 23S rRNA alleles, suggesting that only one or two of the three C. jejuni 23S rRNA genes mutated. Genome assembly algorithms are likely to ignore the minority gene variants, leading to these samples testing negative. Finally, there were five genome assemblies that contained an aph(39)-IIIa aminoglycoside resistance gene (Table S1) not included in the work of Painset et al. (13). Overall, there was very good concordance between the two tests, and we considered this a validation for testing Campylobacter genome assemblies with the NCBI AMRFinder software tool.
Characteristics of a 2001-2018 C. jejuni and C. coli genome assembly database from public sources. A total of 32,256 C. jejuni and 8,776 C. coli genomes were included in our study. The distribution over the year categories is shown in Fig. 1A; for C. coli, approximately 400 UK samples were added for each of the years from 2015 to 2018, whereas the number of US C. coli samples increased from approximately 500 samples in 2015 to just over 2,000 in 2018. Similar trends were observed for C. jejuni, where the number of UK samples varied between approximately 2,000 and 4,000 per year, while the number of US samples increased from approximately 1,000 in 2015 to just under 4,000 in 2018. The samples from 2001 to 2014 were combined, as these did not have consistent availability of samples, with yearly numbers varying between 0 and 350 (C. coli, UK), 7 and 704 (C. coli, US), 10 and 1,881 (C. jejuni, UK) and 0 and 277 (C. jejuni, US) (see Fig. S1 and S2 at https://doi.org/10.6084/m9.figshare.19292126.v1).
The C. jejuni and C. coli samples were split into several groups: UK-and US-derived samples, and within each, historical samples from 2001 to 2014 versus the individual years from 2015 to 2018, and the three source categories (farm animals, humans, and poultry). Comparison of these categories (Fig. 1B) showed that UK samples were dominated by human isolates (approximately 70% for C. coli, versus 5% for C. coli from the US), with farm animal samples poorly represented. In contrast, US samples showed a better representation of farm animal samples for both C. jejuni and C. coli, especially in samples derived from 2001 to 2014.
Distribution of antimicrobial resistance markers in C. jejuni and C. coli. The antimicrobial resistance profiles were determined for all the C. jejuni and C. coli genomes using NCBI AMRFinder. The presence of any of the 44 resistance markers was translated into aminoglycoside resistance (20 markers), macrolide resistance (9 markers), quinolone resistance (10 markers), or tetracycline resistance (5 markers). Table 2 shows the breakdown of the number of isolates predicted to be resistant for both C. coli and C. jejuni. a Detection of AMR markers using NCBI AMRFinder version 3.6.7 on genomes assembled using Shovill version 1.0.9, using samples described by Painset et al. (13). The predictions from NCBI AMRFinder were compared to the results from Painset et al. (13) and the percent agreement reported for AMR marker-positive and AMR marker-negative samples. b Five samples contained an additional aph(39)-IIIa aminoglycoside resistance gene not reported in reference 13 and were not included in the percentage calculation. Two of these samples were phenotypically resistant to aminoglycosides (13). c Three negative samples were previously reported as erythromycin susceptible in resistance testing and showed inconclusive mutation detection in the study by Painset et al. (13). These were excluded from the percentage calculations. d The two negative samples had the tetO resistance gene divided over two contigs, which resulted in NCBI AMRFinder being unable to detect the tetO resistance gene. Additional screening using Abricate version 0.9.8 with the "-mincov 30" setting (allowing screening for gene fragments) allowed the detection of this AMR marker (Table S1).
Comparison of AMR levels between the UK and US (2001 to 2018). The UK and US are currently the two countries which do large-scale genome sequencing-based surveillance of Campylobacter (14,26), and this allows for insight in the dynamics of antibiotic resistance in these two industrialized countries over time. The years 2015 to 2018 were used individually while combining the samples from 2001 to 2014 to function as a possible baseline for comparative purposes. The predicted resistances to aminoglycosides, macrolides, quinolones, and tetracycline were plotted out for the UK and US isolates and compared for trends ( Fig. 2), with the individual years from 2001 to 2018 shown in Fig. S1 and S2 at https://doi.org/10.6084/m9.figshare.19292126.v1.
The general trend for both C. coli and C. jejuni was that the proportion of aminoglycoside-and tetracycline-resistant isolates was significantly higher in the US than in the UK between 2015 and 2018, while the prevalence of quinolone resistance was significantly lower in the US (Fig. 2), prevalences of macrolide resistance were similar for both countries for C. jejuni, and the prevalence was higher for US C. coli for three of the four years but lower in 2015 (Fig. 2). Approximately 40% of the US C. coli isolates were predicted to be resistant to aminoglycosides (range, 39 to 45%), whereas in the UK C. coli isolates, the value was approximately 20% (range, 12 to 26%). For C. jejuni, the prevalence of aminoglycoside resistance was between 12 and 14% for the US isolates, compared to 1 to 4% for the UK isolates. Differences were less pronounced for tetracycline resistance for C. jejuni (range, 39 to 46% for UK samples, versus 48 to 56% for US samples), while in C. coli this varied between 36 and 56% for the UK isolates and between 50 and 65% for the US isolates. Macrolide resistance was present in much higher levels in C. coli in both the UK and the US isolates, ranging from 3 to 19% in the UK isolates and 6 to 14% in the US C. coli isolates, while it was between 0 and 1% in the UK and the US C. jejuni isolates. Finally, the trend was reversed for quinolone resistance, with a higher proportion of the UK isolates predicted to be quinolone resistant. For C. coli, the proportions of quinolone-resistant isolates were 31 to 44% for UK isolates and 25 to 39% for US isolates, while for C. jejuni the proportions were 39 to 44% for UK isolates and 22 to 23% for US isolates. For each of the Campylobacter species and antibiotic class, statistically significant differences were generally observed between countries and year (see Table S2 at https://doi.org/10.6084/m9.figshare.19292126.v1).
Besides, the differences between the two countries, we also observed that the only antibiotic class for which there was a clear decrease in resistance from 2015 to 2018 was macrolides (Fig. 2), with decreases from 15% to 6% in the US isolates and from 19% to 5% in the UK isolates. For the other three antibiotic classes, there was little change from 2015 to 2018, and especially in C. coli, the proportions of resistant isolates remained high for aminoglycosides (US), quinolones (UK and US), and tetracycline (UK and US). For C. jejuni, the levels of quinolone resistance are especially of concern in the UK.
Associations between isolation source and multilocus sequence type (MLST) genotypes and AMR markers in C. coli and C. jejuni. Due to the pronounced differences in isolation source between the UK and US (Fig. 1B), we investigated whether there were differential contributions to the proportions of antibiotic-resistant isolates for C. coli and C. jejuni isolates from 2015 to 2018 (Fig. 3A). There was no clear pattern of overrepresentation of any isolation source for the AMR isolates in C. coli. However, the low number of farm animal isolates from the UK may hide some effects. Similarly, for C. jejuni there was no clear link with any of the three source categories (Fig. 3A).
We also investigated whether specific multilocus sequence type (MLST) clonal complexes (CC) were associated with specific antibiotic resistances in the UK and US isolates. For C. coli, 97.3% of the isolates are from CC-828. For C. jejuni, we looked at the clonal complexes for which for both the UK and US there were at least .100 samples available for 2015 to 2018 (Fig. 3B). For aminoglycoside resistance, there was a spread over all the major clonal complexes for US isolates, with CC-21 and CC-353 having the highest proportion of resistant isolates (17 to 20%). In contrast, for the UK isolates, aminoglycoside resistance was primarily found in CC-353 and CC-42. The CC-353 clonal complex was also involved in macrolide resistance, although this played a minor role in the C. jejuni isolates, with CC-45 showing the highest proportion of resistant isolates in the United States. For both quinolone resistance and tetracycline resistance, all major MLST genotypes contributed to resistance, but the major contribution to quinolone resistance in UK isolates came from CC-353, CC-354, and CC-464, where almost all isolates were predicted to be quinolone resistant, consistent with earlier reports (26,29). Finally, the highest proportions of tetracycline resistance were associated with CC-354 and CC-464 for UK isolates and CC-257 and CC-42 for US isolates. Most of the clonal complexes reported here were primarily associated with poultry and human infections, except for CC-42, associated with cattle.

DISCUSSION
In this study, we have exploited the possibilities afforded by the active genome sequencing-based surveillance for Campylobacter in the UK and US, to determine the trends in AMR over recent years and compare these trends within and between the two countries. Similar studies have previously been conducted in individual countries, but often at a smaller scale than employed in this study or focusing on a shorter period. For instance, a recent UK study investigated 528 human isolates from 2015 to 2016 (13), while a 2018 US-based study focused on 589 isolates from 2015 (14). These and other genome sequencing-based studies have all shown that identifying resistance markers in genome sequences matches well with phenotypic resistance. Hence, we are confident that our study appropriately represents AMR trends detected in 32,256 C. jejuni and 8,876 C. coli samples, as listed in Table S1 at https://doi.org/10.6084/m9.figshare.19292126.v1.
Monitoring of AMR on this scale requires active surveillance, with sequencing data and metadata being released in the public domain. We have combined the information present in the GenBank database, which contains both US and UK samples, and the Campylobacter PubMLST database, which includes a large number of UK isolates. Our data set is based on a deduplicated, quality-controlled set of genome sequences with strict requirements for available metadata. The criteria for isolate source, year, and country data allowed direct comparison between the UK and US Campylobacter samples for C. coli and C. jejuni (Fig. 1B). The UK data set leans heavily on the Oxfordshire Sentinel Surveillance (25,26), which explains the dominance of human samples in the UK data set in each individual year. This contrasts with the US data set, which has an equal distribution over the three source categories. While this is not ideal for comparative purposes, we do note that for C. jejuni the distribution over the dominant MLST clonal complexes was very similar (Fig. 3), with the major differences being that CC-257 and CC-464 were prevalent in higher proportions in UK samples (5% and 6% more of the total number of samples). In comparison, CC-353 was more prevalent in US samples (7% more). All three clonal complexes are primarily associated with poultry, and hence, for C. jejuni, we concluded that the data sets could still be compared. For C. coli, this was more difficult to assess, as most samples clustered in the agricultural clade 1A (30,31), which contains only CC-828 and a minority of samples not part of this clonal complex. The individual MLSTs may primarily represent the STs prevalent in the two countries and may be affected by the differences in isolation sources between the samples from the two countries (Fig. 1B).
In this study, we have focused on four classes of antibiotics, due to their relevance in Campylobacter and their usage in the categories of hosts investigated. The macrolide erythromycin and the fluoroquinolone ciprofloxacin are used to treat human infections with Campylobacter (22). In contrast, all four classes of antibiotics are used therapeutically in the veterinary sector for livestock, albeit not for control of Campylobacter, as the usage is associated with other infectious diseases in livestock (32). As highlighted earlier, the WHO has listed Campylobacter as a high-priority antibiotic-resistant pathogen (2), especially linked to fluoroquinolone resistance in Campylobacter. Antimicrobial resistance in Campylobacter spp. is a combination of de novo point mutations associated with quinolone and macrolide resistance and genetic exchange on plasmids and other mobile elements for aminoglycoside and tetracycline resistance genes (33). In addition, the natural competence of C. jejuni and C. coli (34,35) may allow further dissemination of resistance through populations. Although the usage of antibiotics has been limited in the UK and the US for a considerable period of time (32,36,37), there was no apparent reduction visible in the prevalence of resistance in the large numbers of C. jejuni and C. coli samples investigated in this study (Fig. 2). Our statistical analyses suggest that there may be small increases in C. jejuni resistance to some antibiotics in recent years, compared to the levels of 2014 and earlier (see Table S2 at https://doi.org/ 10.6084/m9.figshare.19292126.v1). However, due to the composition of the data set and its retrospective nature, this is something that requires a prospective study to further investigate, which falls outside the aims and scope of the work presented here. A possible hypothesis for the observations is that resistance to especially tetracyclines, aminoglycosides, and quinolones is not detrimental to Campylobacter or not sufficiently detrimental to allow outcompetition by susceptible isolates. Conversely, resistance may even be associated with advantageous phenotypes, as the gyrase mutations associated with quinolone resistance gave rise to increases in virulence phenotypes in C. jejuni (38,39), while tetO-containing plasmids are ubiquitous and stable in C. jejuni and C. coli (40,41). Aminoglycoside resistance genes are especially widespread in C. coli, and their association with mobile elements (33,42) may assist their stability in the populations.
The similarities between the UK and US data sets suggest that the situation in other industrialized countries may be similar, but this will require more intensive surveillance programs and will need to balance clinical, food, and agricultural samples, including sewage and water sources. A recent study reported that fluoroquinolone usage in pigs and poultry in France from 2011 to 2018 decreased by .70% in poultry and almost 90% in pigs. However, resistance to ciprofloxacin increased in poultry Campylobacter isolates from 50% to approximately 60%, whereas in Escherichia coli, prevalence of ciprofloxacin resistance was stable in broiler and pig isolates but decreased in turkey isolates (43). Similar studies have been done in other countries but often are based on small or limited data sets where selection bias may affect the results.
The lack of reduction of the proportion of AMR isolates in Campylobacter from both countries does suggest that reduced usage of antimicrobials, such as mandated by antimicrobial stewardship, will not be sufficient to reduce the incidence of AMR in Campylobacter, although it may well contribute positively to reining in further increases of AMR. For clinical purposes, macrolides such as erythromycin are still useful for treating C. jejuni infections, but their efficacy may be reduced for C. coli. In contrast, the high level of quinolone resistance in both C. jejuni and C. coli does not bode well for the future efficacy of ciprofloxacin treatment of CampylobacterCampylobacter infections in humans. With regard to aminoglycosides and tetracyclines, the latter is not recommended due to the high level of resistance, whereas with aminoglycosides, gentamicin can still be used for now, as prevalence of gentamicin resistance is still low.
Taken together, the data presented here strongly suggest that reduced usage of antibiotics has not resulted in a significant reduction of antimicrobial resistance in Campylobacter, which is of considerable public health and economic concern. Changes in agricultural practices, slaughter, and retail will need to be substantial to reduce the overall prevalence of Campylobacter. These efforts should lessen the need for antibiotic usage to achieve the goals of antimicrobial stewardship. Similar comparative studies could be done in other countries and for other foodborne zoonotic infections to assess whether this situation is unique for Campylobacter or mirrored in other pathogenic bacteria.

MATERIALS AND METHODS
Campylobacter genome assemblies and metadata categories included in this study. A total of 44,751 C. jejuni and 12,709 C. coli genome assemblies were collected from the GenBank and Campylobacter PubMLST databases and were coupled to metadata from the GenBank files and from PubMLST. Genome assemblies were obtained from the NCBI database using ncbi-genome-download version 0.2.11 (https://github .com/kblin/ncbi-genome-download/) and supplemented with genome sequences from the Campylobacter PubMLST website (http://pubmlst.org/campylobacter/) (44). Metadata were extracted from GenBank flat files and the NCBI Pathogens database (https://www.ncbi.nlm.nih.gov/pathogens). All genome assemblies were screened for assembly statistics using Quast version 4.5 (45), and genome assemblies were excluded if failing two or more of the following criteria: number of contigs # 200, N 50 $ 25 kb, L 50 # 25 contigs, largest contig $ 50 kb, or number of Ns per 100 kb $ 50. Genomes with a total size outside 1.4 Mbp and 2.1 Mbp were automatically excluded. Duplicate entries were removed by comparing sample names, assembly statistics such as N 50 , L 50 , and genome size and metadata such as year, source, and country. After deduplication, samples were subsequently only included if the following metadata were available: isolation source, year of isolation, and isolation in the UK or US. Isolation sources were combined to give three main categories: farm animals (pigs, cattle, sheep, and goats, including milk and meat samples), human (clinical isolates), and poultry (chicken and turkey, including meat samples). Samples representing environmental, farm, generic "food," and wild bird samples were excluded, as were samples lacking other metadata. This resulted in the exclusion of 6,327 C. jejuni and 877 C. coli samples with missing information on country, year, or isolation source; 1,634 C. jejuni and 142 C. coli samples were not from the UK or US, 3,116 C. jejuni and 2,086 C. coli samples were from outside the period from 2001 to 2018, and 1,418 C. jejuni and 828 C. coli samples were not from animal, ruminant, poultry, or human sources. We focused on 2015 to 2018 as individual years because both the UK and US were doing surveillance projects, while historical samples were categorized as 2001 to 2014. Overall, we analyzed a total of 41,032 genome sequences, represented by 32,256 C. jejuni and 8,776 C. coli genomes (see Table S1 at https:// doi.org/10.6084/m9.figshare.19292126.v1).
Comparison of genome assembly-based screening for AMR genes with sequencing read-based screening. The supplementary data presented by Painset et al. (13) contained 381 accession numbers to FASTQ read files of UK Campylobacter samples used for screening for antimicrobial resistance genes. The FASTQ files were downloaded from the Sequence Read Archive using fastq-dump from the SRA toolkit (https://github.com/ncbi/sra-tools), and genomes were assembled using Spades version 3.14 (46) via the Shovill version 1.0.9 tool and standard settings (https://github.com/tseemann/shovill/). All genomes passed the Quast quality control (QC) and were screened using the NCBI AMRFinder tool version 3.6.7 (17) with the nucleotide setting (-n) and the organism switch (-O Campylobacter) as described above. All genome assemblies were also screened using Abricate version 0.9.8 (https://github.com/tseemann/ abricate/) and the NCBI database with a minimum coverage of 30% (-mincov 30) to check for the tetO tetracycline resistance gene being split over two or more contigs. These 381 genome assemblies have not been included in the other analyses presented here.
Statistical analysis of association between resistance markers and descriptor variables (country, year, and isolation source). The presence or absence of resistance for each of the four classes of antibiotics was assessed using a generalized linear model with a logit link function (binomial family). Country, year, and isolation source of each sample were considered explanatory variables. Interactions between predictors were not considered. Bootstrapping with 500 repeats was carried out to estimate the 95% confidence interval (CI) of the prevalence of resistance for each country and year. All statistical analyses were carried out in R (version 4.0.3) (47), and the data are presented in Table S2 at https://doi.org/10 .6084/m9.figshare.19292126.v1.
Data availability. All genome sequences used in this study are available from the GenBank/EMBL/DDBJ databases or the Campylobacter PubMLST website (https://pubmlst.org/organisms/Campylobacter-jejunicoli/). The assembly accession numbers (NCBI Genome) or genome identifier (ID) numbers (Campylobacter PubMLST) are listed in Table S1 at https://doi.org/10.6084/m9.figshare.19292126.v1, together with the metadata used and the AMR gene data.

ACKNOWLEDGMENTS
We gratefully acknowledge the efforts of the surveillance programs in the US and UK and their continued contribution to data sharing by making the genome sequences and partial metadata publicly available.
Part of this work was supported by funding from the European Union's Horizon 2020 Research and Innovation program under grant agreement no. 773830: One Health European Joint Programme (WorldCOM). The US whole-genome sequencing work was supported by the National Institutes of Health/Food and Drug Administration under award number 5U19FD007113-02, whereas UK Campylobacter surveillance was supported by the UK Food Standards Agency. This publication made use of the PubMLST website developed by Keith Jolley and sited at the University of Oxford. The development of that website was funded by the Wellcome Trust.
We declare no conflict of interest.