Temporal, Spatial, and Genomic Analyses of Enterobacteriaceae Clinical Antimicrobial Resistance in Companion Animals Reveals Phenotypes and Genotypes of One Health Concern

Background Antimicrobial resistance (AMR) is a globally important one health threat. The impact of resistant infections on companion animals, and the potential public health implications of such infections, has not been widely explored, largely due to an absence of structured population-level data. Objectives We aimed to efficiently capture and repurpose antimicrobial susceptibility test (AST) results data from several veterinary diagnostic laboratories (VDLs) across the United Kingdom to facilitate national companion animal clinical AMR surveillance. We also sought to harness and genotypically characterize isolates of potential AMR importance from these laboratories. Methods We summarized AST results for 29,330 canine and 8,279 feline Enterobacteriaceae isolates originating from companion animal clinical practice, performed between April 2016 and July 2018 from four VDLs, with submissions from 2,237 United Kingdom veterinary practice sites. Results Escherichia coli (E. coli) was the most commonly isolated Enterobacteriaceae in dogs (69.4% of AST results, 95% confidence interval, CI, 68.7–70.0) and cats (90.5%, CI 89.8–91.3). Multi-drug resistance was reported in 14.1% (CI 13.5–14.8) of canine and 12.0% (CI 11.1–12.9) of feline E. coli isolates. Referral practices were associated with increased E. coli 3rd generation ≤ cephalosporin resistance odds (dogs: odds ratio 2.0, CI 1.2–3.4). We selected 95 E. coli isolates for whole genome analyses, of which seven belonged to sequence type 131, also carrying the plasmid-associated extended spectrum β-lactamase gene blaCTX–M–15. The plasmid-mediated colistin resistance gene mcr-9 was also identified for the first time in companion animals. Conclusions Linking clinical AMR data with genotypic characterization represents an efficient means of identifying important resistance trends in companion animals on a national scale.

To address these issues, national AMR surveillance in humans and production animals has been established (VMD, 2019;PHE, 2020). However, companion animal species have been arguably neglected, despite resistance developing in companion animals exposed to antimicrobial therapy (Schmidt et al., 2018), and transmission of pathogens and resistance between humans and companion animals (Lei et al., 2017). This oversight has partly been due to the companion animal veterinary diagnostic laboratory (VDL) sector being largely comprised of independent commercial laboratories, each utilizing often bespoke information management systems (O'Neill et al., 2014), and different protocols for antimicrobial susceptibility testing (AST) (Marques et al., 2016). Such protocols largely focus on relatively time-consuming techniques for demonstrating phenotypic resistance. Recently, genomics-based methodologies have unlocked a wealth of information pertaining to the development and dissemination of resistance genes in and between bacterial species (McCarthy et al., 2015). Though this has only begun to be explored in companion animals (Zhang, 2016), bacterial isolates disposed of by VDLs remain an untapped resource.
Hence, the aim of this study was, using Enterobacteriaceae isolated from clinical samples of dogs and cats as an exemplar, to define national patterns of AST in companion animals, and factors associated with resistance. Further, we sought to develop a method by which Enterobacteriaceae isolates expressing resistance patterns of potential public health importance could be efficiently captured from VDLs for genotypic characterization. We also sought to investigate the AST method and interpretation variation between VDLs, and to outline how such variation might impact on more detailed genotypic investigations as performed here.

Antimicrobial Susceptibility Test Data
Clinical AST data were collated by the Small Animal Veterinary Surveillance Network (SAVSNET) from four United Kingdom VDLs (A-D) between 1st April 2016 and 31st July 2018, including a unique sample identifier; submitting veterinary practice site postcode; report date; animal species; breed; sex and neuter status; bacterial species isolated, and AST results. Sample type/site (supplied as free text) was manually categorized into urine, ear(s), oro-nasopharyngeal and respiratory sites, feces, anal region including anal sacs, and other/mixed sites. SAVSNET holds ethical approval to collect such data from the University of Liverpool Ethics Committee (RETH0000964).
The manner with which bacterial species were described varied between VDLs, ranging from genus-level (e.g., Escherichia) alone to species-level (e.g., Escherichia coli). Similarly, bacteria were speciated according to varying methodologies, ranging from commercial biochemical tests (VDLs B-D) to MALDI-TOF (VDL A); Enterobacteriaceae were selected for further analysis. Laboratories utilized disk diffusion (VDL B-D) or minimum inhibitory concentration (MIC) (VDL A) approaches, and used different interpretation guidelines (CLSI, EUCAST, and BSAC). As some VDLs (B-D) only provided clinically interpreted AST results, analysis was restricted to interpreted results alone. Only antimicrobials associated with species-specific acquired resistance were summarized. Tested antimicrobials were grouped to class-level (Supplementary Material 1); 'intermediate' results were classified as 'sensitive, ' with multi-drug resistance (MDR) defined as resistance to three or more classes (Magiorakos et al., 2012).

Veterinary Practice Site Demographic Data
The Royal College of Veterinary Surgeons (RCVS) Veterinary Practice Directory (RCVS, 2018) was interrogated on 12th October 2018, using R packages 'rvest' (Wickham, 2016) and 'purrr' (Henry, 2018), to extract information for all United Kingdom veterinary practice sites (Table 1); postcodes containing more than one site treating companion animals (n = 148) were designated 'multi-site postcodes.' The National Statistics Postcode Look-up was used to provide geographical information, including urban/rural status (NSPL, 2015).

Animal Demographic Data
Dog and cat breeds were summarized to standardized breed terms, which were then grouped into genotypically similar groups. Three further breed groups were defined: crossbreeds, breeds not yet genetically classified ('unclassified'), and breeds not recorded or recognizable ('unknown') (Singleton et al., 2020).

Multidrug Resistant Isolate Collection and Storage
Six VDLs (A-F) were approached regarding collection of Enterobacteriaceae isolates, of which three (A, E, F) participated. Submitted isolates needed to comply with one or more of the following criteria: (1) Phenotypic ESBL-producer, (2) MDR, or (3) non-susceptibility to one or more of the 3rd/4th generation cephalosporin, fluoroquinolone, polymyxin, or carbapenem classes. On receipt at the University of Liverpool, isolates (n = 148) were incubated under aerobic conditions for 18-24 h at 37 • C on nutrient agar, and pure colonies stored at −80 • C (Microbank TM cryovial; Pro-Lab Diagnostics, United Kingdom).

Phenotypic Analyses
To mitigate for protocol variation between VDLs, disk diffusion ASTs were undertaken on all received isolates according to CLSI guidelines (CLSI, 2008

Statistical Analyses
All statistical analyses were completed using R (v3.4.4). Descriptive proportions and 95% confidence intervals were adjusted for clustering within practice sites (bootstrap method, n = 5,000 samples) (AOD, 2016). Mixed effects logistic regression models exploring odds of E. coli resistance were fitted separately in dogs and cats (LME4, 2016), considering 3rd/4th generation cephalosporin resistance as a binary outcome variable. MDR, fluoroquinolone and potentiated penicillin resistance were also modeled (Supplementary Materials 5-13). Both practice site and VDL were included as random effects if a likelihood ratio test (LRT) suggested improved fit. Initial univariable screening considered a range of categorical explanatory variables, including sample type/site and animal or practice site demographic features (Supplementary Material 8). Explanatory variables were retained for multivariable analysis if a LRT indicated P ≤ 0.20 against a null model.
Multivariable models underwent step-wise backward elimination to minimize Akaike Information Criterion (AIC). Two-way interaction terms were retained if inclusion resulted in an AIC decrease. Multicollinearity was assessed using the Variance Inflation Factor (CAR, 2018). Statistical significance was defined as P < 0.05.

AST Data -Laboratories A-D
Data were obtained for 29,330 canine and 8,279 feline Enterobacteriaceae isolates from 2,237 practice sites, corresponding to 49.1% (95% confidence interval, CI, 47.7-50.6) of registered companion animal-treating sites (Supplementary Material 14), contributing a median of seven canine (range 1-156) and three feline (range 1-166) isolates each. Though practice site details could not be ascertained in 6.7% (CI 5.3-8.1) of canine and 5.8% (CI 4.1-7.4) of feline AST reports, site demographic and animal population features of the remainder are summarized in Table 1 and Supplementary Material 15, respectively.

3rd/4th Generation Cephalosporin Resistant E. coli
While prevalence remained relatively stable over the 2 years summarized (Figure 1), geographical variation was observed (Figure 2). Male dogs were associated with reduced odds of 3rd/4th generation cephalosporin resistance, whereas male cats were associated with increased odds ( Table 3). Practice sites examining referral cases only were associated with increased odds, as were RCVS accredited sites. Compared to urine samples, the anal region and oronasopharyngeal/respiratory samples were associated with increased odds in dogs and cats, respectively (univariable findings, Supplementary Materials 18,19).

DISCUSSION
Here we have adopted an integrated health informatics and bioinformatics approach to conduct nationally representative surveillance of important clinical AMR trends in companion animals. We have also demonstrated that clinical AMR isolates can be retained by VDLs and further characterized to reveal key AMR insights of one health importance.
Resistance trends in Enterobacteriaceae are of critical importance to human health (Tacconelli et al., 2018). Considering the inherent one health nature of AMR (O'Neill, 2016), it is crucial to gain further understanding of AMR in our companion animal populations. Through collating AST results from half of companion animal-treating practice sites in the United Kingdom, with data being broadly comparative with both the estimated companion animal population (Aegerter et al., 2017) and demographic features of the RCVS practice register (RCVS, 2018), we believe these data are nationally representative. However, mixed species practice sites were under-represented in our dataset. Veterinary practitioners employed in different sectors (companion animal, farm etc.) have reported varied levels of support for ASTs (De Briyne et al., 2013), a disparity perhaps reflected in actual AST results here. Such findings can serve as a stimulus for localized resistance trend monitoring, potentially providing a powerful tool and motivator for antimicrobial stewardship.
Antimicrobial classes rarely (or never) prescribed to companion animals, for example carbapenems (Singleton et al., 2017), were only rarely reported as being tested here. Critically important antimicrobials currently only used in medical practice should not be recommended for veterinary use. However, there would be value in routinely reporting resistance trends to such antimicrobials in veterinary isolates for public health purposes, as demonstrated by our findings relating to colistin resistance here. It should also be noted that differences between AST interpretation guidelines (e.g.,  Lipinski et al., 2008. For both models, veterinary practice site (dogs, variance = 0.22, standard deviation, SD, 0.47; cats, variance = 0.74, SD = 0.86) was modeled as a random effect; for dogs, laboratory site (variance = 0.99, SD = 1.00) was also included. Significant categories within a variable are emboldened.
FIGURE 3 | Maximum likelihood phylogenetic tree constructed by core genome SNPs of 91 E. coli isolates using the genome of E. coli K12 as a reference. Colored bands and numbers in the first column indicate clonal complex (CC) of the isolate (gray color indicates unassigned CC). A heatmap represents antimicrobial susceptibility (red = resistance; yellow = intermediate; green = susceptible). Brown circles in columns six and eleven represent the presence of beta-lactamase genes and the mcr-9 gene, respectively. The seventh -ninth columns represent QRDR mutations in gyrA, parC, and parE, respectively. The tenth column shows the presence/absence of the plasmid-mediated quinolone resistance gene, qnr, in each isolate. Purple squares in the last column indicate plasmid replicons identified in each isolate. Abbreviations for antibiotics tested here are as follows: AP, ampicillin; AUG, clavulanic acid potentiated amoxicillin; KF, cephalothin; CPD, cefpodoxime; C, chloramphenicol; ENF, enrofloxacin; GM, gentamicin; T, tetracycline; TS, trimethoprim potentiated sulfamethoxazole.
EUCAST and CLSI) limited our ability to compare laboratories in this study, re-affirming a need for greater harmonization between laboratories. AST data summarized here also does not include an individual animal identifier, hence it was not possible to designate multiple submissions originating from the same animal. To ameliorate this limitation, we would support efforts at increasing integration between practice and laboratory information management systems.
Considering the above, individual class-level resistance prevalence was generally lower than reported in other studies, though MDR prevalence was broadly comparative (Marques et al., 2016). While others have classified intermediate susceptibility results as resistant, potentially increasing resistance prevalence (VMD, 2019), doing so here would not have elevated findings to levels comparative with previous study (for example, re-classifying intermediate susceptibility findings as resistant would have increased E. coli 3rd generation cephalosporin resistance prevalence in dogs from 8.4 to 10.6%). Previous surveys have largely utilized data supplied by university-based diagnostic laboratories analyzing referral populations (Hernandez et al., 2014;Marques et al., 2016), and here we found increased odds of resistance in isolates originating from referral-only practice sites for dogs. Antimicrobial use prior to culture has been shown to increase odds of isolating resistant bacteria (Hernandez et al., 2014). It is possible that antibiotic therapy prior to referral might in part explain these findings, suggesting a need to stratify surveillance into first opinion and referral populations, broadly in line with human surveillance (PHE, 2020).
Considering animal-level features, genetic breed group, sex and neuter status were associated with 3rd/4th generation cephalosporin resistance odds variation in both species. These findings potentially reflect anatomical (Hernandez et al., 2014) or behavioral differences (Chhetri et al., 2015) between animals, possibly resulting in increased likelihood of bacterial infection or resistance-selecting empirical antimicrobial prescription in some cases. Indeed, we have recently identified such variation in prescribing practices (Singleton et al., 2020). Regarding sample type/site, the anal region in dogs and, oronasopharyngeal and respiratory samples in cats were associated with increased odds. Though resistant E. coli isolated from canine nasal or respiratory sites have been described (Morrissey et al., 2016), it has only rarely been implicated as a causative pathogen in human intensive care settings (de Lastours et al., 2015). In dogs and cats, Enterobacteriaceae are considered minor components of oral, nasopharyngeal and respiratory microbiota, and their role as respiratory pathogens remains debatable (Dorn et al., 2017). Similarly, it is difficult to determine whether anal region samples reflect intestinal carriage or clinical infection, though could be considered to pose a public health risk, nonetheless. Several bacterial species usually considered contaminants were also reported within AST data (e.g., Kluyvera spp.; Pantoea spp. etc.); doing so without thorough evaluation of their association with infection is likely to encourage potentially unnecessary antimicrobial prescription.
There are several Enterobacteriaceae resistance trends of critical one health importance, not least ESBL-or AmpCproduction, fluoroquinolone resistance, carbapenemase producing Enterobacteriaceae, and colistin resistance (MacNair et al., 2018). However, due to the primarily clinical purpose behind AST results analyzed here, some phenotypic tests were only rarely reported. Hence, here we aimed to investigate these isolates to gain further insight into dynamics of resistance development and transmission in the companion animal clinical population, utilizing isolates supplied by multiple independent diagnostic laboratories.
Regarding WGS findings, although ST372 E. coli has been previously primarily associated with canine infections and relatively rare resistance (Valat et al., 2020), here we identified an MDR ST372 also possessing mcr-9, a recently identified plasmid-associated colistin resistance gene (Carroll et al., 2019). In medical practice, colistin is considered an antimicrobial of 'last resort, ' and reported emergence of plasmid-mediated colistin resistance in 2015 instigated widespread concern (Liu et al., 2016). Although there is some debate regarding the clinical relevance of mcr-9 (Tyson et al., 2020) and when it first originated (Elbediwi et al., 2021), our finding represents its first identification in a companion animal, and the first identification of a plasmid-associated colistin resistance gene in companion animals in the United Kingdom. Although companion animals are not prescribed colistin in the United Kingdom, they are relatively frequent recipients of another member of the polymyxin class, polymyxin B (Singleton et al., 2017). Concern has been raised over whether such use might be selecting for colistin resistance (Scott et al., 2019), though this remains to be determined in Enterobacteriaceae. Nevertheless, these findings serve to further reinforce the role of companion animals as carriers of resistance mechanisms of public health importance, and the need for surveillance to highlight the emergence of such trends.
A diverse collection of ESBL genes were found in E. coli isolates undergoing WGS, with the bla CTX−M (1, 9, 14, 15, 27, and 65) family dominating. In people, bla CTX−M−14 and bla CTX−M−15 are of greatest clinical importance (Canton et al., 2012), whereas agricultural species are more commonly associated with bla CTX−M−1 carriage (Abraham et al., 2018). In companion animals, bla CTX−M−1 (Wedley et al., 2017), bla CTX−M−14 (Sun et al., 2016), and bla CTX−M−15 (Timofte et al., 2016) have each been identified. Here bla CTX−M−15 was most prevalent; concerningly, the human pandemic clone ST131 predominated, further demonstrating links between people and companion animals for important resistance and pathogenic clones (Zhang, 2016). This clone has been detected in companion animal clinical isolates previously, and is considered a driver of bla CTX−M−15 dissemination in human and animal populations (Timofte et al., 2016). F-type Inc plasmids were also frequently identified, this group of plasmids (particularly IncFII) being commonly implicated with global dissemination of bla CTX−M (Canton et al., 2012). Inhibitor resistant ESBL genes were also detected at low frequency, the bla TEM−158 gene having been previously reported in a companion animal referral hospital in the United Kingdom (Tuerena et al., 2016).
Although less frequently identified than ESBL genes, AmpC genes were also identified in these isolates. Despite bla CMY−2 being the most well distributed bla CMY allele worldwide (Jacoby, 2009), including in companion animals (Wedley et al., 2011), it was not detected here. Instead, the comparatively recently identified bla CMY−111 (Kao et al., 2018), representing the first identification in companion animals, and bla DHA−1 , this gene only recently reported in European companion animals (Belas et al., 2014), were found to predominate. Although the bla CMY gene has many variants (Jacoby, 2009), and as such a measured view of this novel finding should be taken, our results do suggest increased AmpC diversity in companion animals. We also detected ST410 in this population, this ST being previously implicated in human and companion animal infection (Timofte et al., 2016), including harboring a carbapenemase resistance gene in a sample of canine origin (Reynolds et al., 2019).
Regarding treatment of ESBL and/or AmpC-mediated infections in companion animals, the appropriateness of using alternative agents, such as the carbapenems and piperacillintazobactam, is a topic of ongoing debate in the medical field due to concerns surrounding developing resistance to these 'last report' agents (Tamma and Mathers, 2021). Indeed the prescription of such agents to companion animals is not recommended (BSAVA, 2018), severely limiting treatment options for such cases. Accordingly, whilst emergence of these troubling clinical resistance trends in companion animals is often viewed primarily through the lens of associated public health risk, it should also be remembered that enhanced treatment failure risk may well also carry an increased risk to animal welfare.
Quinolone resistance mechanisms are relatively complex, and consist of interactions between chromosomal mutations and plasmid-mediated genes (Hamed et al., 2018). In this study, chromosomal mutations were commonplace, contrasting with the relative scarcity of plasmid-mediated genes. This finding suggests a greater role of chromosomal mutations in conferring clinical resistance in companion animals, in agreement with previous study (de Jong et al., 2018). Considering AMR genes identified more widely, although many of the AMR genes identified here are commonly associated with plasmid types frequently identified in this study, specific, in-depth investigation was out of the scope of this study. Hence, though unlikely, it is possible that some of the assumed transmissible AMR genes discussed here might be chromosomal in origin.
Although this study was successful in identifying clinical isolates of one health importance, we revealed further evidence of AST method and interpretation variation between VDLs. When confirmatory additional testing was performed at the University of Liverpool, although a similar percentage of isolates were reported as MDR by referring laboratories (82.8%) to that found at Liverpool (83.6%), ESBL testing was only reported for a minority of referring VDLs, potentially systematically under-estimating the impact of ESBLs on companion animal practice, and the potential risk posed to public health. This is likely due notification of ESBL presence being of an arguably greater importance for public health/surveillance than for clinical decision making. Hence, we suggest an urgent need to harmonize standard AST testing procedures, methods, and interpretation more closely, including tests of primary public health/surveillance relevance, if companion animal clinical bacterial isolates are to be used for wide scale surveillance in the future.

CONCLUSION
This study demonstrated a method by which clinical AST results could be repurposed for near real-time passive and active AMR clinical surveillance at a nationally representative scale in companion animals, providing insights of importance and use to both veterinary practitioners and public health. Using Enterobacteriaceae as an exemplar, we found MDR to be common, with prevalence remaining relatively static between 2016 and 2018. We also conducted a pilot study to determine whether MDR E. coli isolates could be efficiently retrieved from VDLs and further characterized at the University of Liverpool, revealing presence of resistance genes and sequence types of importance to human and animal health.

DATA AVAILABILITY STATEMENT
Data utilized during this study is available upon written request to DS (d.a.singleton@liverpool.ac.uk). Whole genome sequence data is freely available via GenBank (accession numbers available, Supplementary Material 4).

AUTHOR CONTRIBUTIONS
NW, AR, GP, P-JN, SD, and FS-V conceived the study. DT, StS, MJ, and LR contributed data and isolates to this project. DS and ShS completed laboratory work. PP, RB, and KB completed or supervised bioinformatic components of this work. DS completed primary analyses and wrote the manuscript under the supervision of NW. All authors contributed to the draft review.

FUNDING
This work was funded by The Veterinary Medicines Directorate (VM0520), the University of Liverpool and SAVSNET. We are extremely grateful for the support and major funding from BBSRC (BB/N019547/1) and BSAVA. KB was funded through a