Four databases, PubAg, Web of Science, PubMed, and CAB Abstracts, were screened using different search strings (Supplementary File 1) on 13 April 2017 and were updated on 18 October 2021. Search strings were designed to identify publications on cephalosporin, colistin, quinolone, and carbapenem resistance in swine. Data on fluoroquinolone and colistin resistance was extracted to conduct a separate systematic review (Hayer et al., 2022).

Records were retrieved, and duplicate articles were removed using Zotero version 5.0.96.3. In addition to the database searches, reports containing information on AMR in E. coli from swine were retrieved from the websites of national and international agencies coordinating surveillance programs for AMR (EFSA, Europe; RESAPATH, France; DANMAP, Denmark; MARAN, Netherlands; NORMVET, Norway; SWEDRES, Sweden; FINRES, Finland; NARMS, United States; and CIPARS, Canada).

Retrieved records were first screened for inclusion if there was a mention of AMR, E. coli or Enterobacteriaceae, and livestock in title and/or abstract. The selected full-length articles were then downloaded and data was collected from articles if there was data on occurrence of phenotypic AMR against 3GCs or carbapenems in E. coli collected from pigs. Articles were excluded if:

A predesigned datasheet was developed using Microsoft Excel 2013 for collecting data such as publication year, country and year of bacterial isolation, health status of pigs, phenotypic resistance tested, microbiological methods employed for AMR determination, criteria for classifying bacteria as resistant/non-resistant, total number of isolates tested and number of isolates resistant against each antimicrobial and whether information on sampling design, antimicrobial use and husbandry was provided. Extracted phenotypic and genotypic data are provided in Supplementary File 2.

Arguably, the biggest issue in aggregating AMR data across studies was the use of different breakpoints to classify bacteria as “resistant” in these studies. We harmonized the AMR prevalence data in a multi-step manner. First, for those studies where MIC distributions were available, the number of resistant bacteria were re-calculated as if the current CLSI breakpoints (CLSI, 2021) were used for defining resistance. If MIC distributions were not available or disk diffusion method was used for estimating resistance, the number of resistant isolates in a particular study was multiplied by a harmonizing factor. These harmonization factors were calculated using the following formulae (Van Boeckel et al., 2019):

MIC or disc diffusion reference distributions were downloaded from EUCAST (2021). Breakpoints were determined for each article based on the performance methodology (e.g., CLSI-M100 series) document provided in the bibliography. If this information was not provided, it was assumed that authors used CLSI breakpoints for interpretation as these are the most commonly used guidelines to differentiate resistant bacteria globally (Van Boeckel et al., 2019).

Considering that isolates collected from healthy pigs were likely to have different phenotypic resistance profiles compared to isolates from diseased pigs, pooled prevalences of resistance were estimated separately for these health categories at a country level. Studies that reported AMR levels in mixed bacterial populations isolated from both healthy and diseased pigs were not included in the meta-analyses, but the results from these studies are available in Supplementary File 2. Studies that did not provide data on health status were assumed to have been conducted on healthy pigs because the authors compared their results to those from other healthy pig populations in the discussion section and these isolates were not based on diagnostic submissions or collected during outbreak investigations. Removing data from such articles did not impact the pooled prevalence estimates (data not shown).

Another issue with pooling resistance data was the detection of resistance to different 3GCs and carbapenems across studies. To address this issue, we selected three 3GC resistances as markers: ceftiofur, cefotaxime, and ceftriaxone resistance. At least one of these cephalosporin resistances was estimated in all studies except one, and for studies estimating resistance to more than one of these 3GCs, the correlation between the proportions of isolates resistant to these 3GCs was high (correlation coefficients > 0.9). For those studies providing data on more than one of these three cephalosporin resistances (e.g., both ceftiofur and cefotaxime resistance measured on same isolates), we used the highest number of isolates estimated to be resistant after harmonization for any of these 3GCs in a study for downstream analyses. For example, if the number of ceftiofur and cefotaxime resistant isolates after harmonization were 30 and 35, respectively in a study, then we used 35 as the number of isolates resistant to 3GCs for statistical analysis. In addition, when available, ceftazidime resistance was analyzed separately from the other 3GCs because the genetic basis of ceftazidime resistance and ceftiofur, cefotaxime, or ceftriaxone resistance are different (Hiki et al., 2017). Similarly, if a study provided data on multiple carbapenem resistances (e.g., doripenem and ertapenem resistance for same set of isolates in the same article), the highest number of isolates after harmonization for any of these carbapenems were used as a proxy for number of isolates resistant to carbapenem on a per article basis.

The pooled proportions of 3GC, ceftazidime and carbapenem resistances were calculated at a country-level by transforming the proportions of resistant isolates using the Freeman–Tukey double arcsine transformation method (Miller, 1978), followed by a random effects model using inverse method and back-transforming to get the final pooled proportion estimates (in percent) across studies using library meta version 5.1-0 (Schwarzer, 2021). Statistical heterogeneity (I²) was retrieved from the output of this model and pooled estimates at a country-level were considered to be heterogenous and homogenous if p-values were <0.05 and >0.05, respectively. Choropleth maps were created to graph the pooled proportions of 3GC and ceftazidime resistances at a global scale using QGis version 3.20.3.

Similarly, pooled fecal carriage rates of 3GC and carbapenem resistance were also estimated and mapped. Fecal carriage rate is defined as the number of fecal samples collected from individual animals that test positive for carrying E. coli with specific resistance phenotype after screening on selective media. Herd level prevalence was not estimated and are not presented in the results section but raw data to do so has been provided in Supplementary File 2. Finally, choropleth maps for estimates for “sampling index” were also generated. For this study, sampling index was defined as the number of unique isolates tested for phenotypic 3GC, ceftazidime or carbapenem resistance divided by the number of specific pig population at a country level. For ease of interpretation, this index was generated per 1,000 pigs. Country-level pig population data (as of 2018) was downloaded from FAO (Food and Agriculture Organisation, 2022).

Data on genotypic prevalence of 3GC and carbapenem resistance was also extracted from the articles selected above and is available in Supplementary File 2. Pooled proportions of ESBL/ampC genes in E. coli isolated on selective media or having unique resistance phenotypes (i.e., carbapenem resistant, ESBL/ampC producers, and 3GC resistant) were estimated as described above.

In order to extract more data on the genomic basis of resistance to these antimicrobials, we downloaded information on genomes of 6,167 E. coli isolates of swine origin from Enterobase webserver (Zhou et al., 2020) (last accessed on 24 July 2021) and analyzed these genomes. The advantage of this data over the PCR data extracted from the literature search is that information regarding several genomic characteristics can be obtained at the isolate level. However, the disadvantage of this dataset is a lack of critical metadata such as microbiological methods employed, health status of the animal, etc.

Short read sequences of these isolates were downloaded from NCBI database based on their accession numbers from Enterobase (Zhou et al., 2020). Raw reads were checked for quality control using FastQC version 0.11.9 (Andrews, 2021) and subsequently trimmed using Trimmomatic version 0.39 (settings: sliding window mode; number of bases to average across = 4; average quality required = 20) (Bolger et al., 2014). Trimmed reads were then assembled using Shovill version 1.1.0 (default settings) (Seemann, 2021b). When short reads were not available for download, assembled genomes were directly downloaded from the Enterobase webserver. Genomes were characterized for virulence genes (Virulence Finder) (Malberg Tetzschner et al., 2020), resistance genes (ResFinder) (Bortolaia et al., 2020), chromosomal mutations that can confer resistance (PointFinder) (Zankari et al., 2017) and plasmid replicons (Plasmidfinder) (Carattoli et al., 2014) using Abricate version 1.0.0 (Seemann, 2021a). All databases were last updated on 24 July 2021. Isolates carrying ESBL/ampC and carbapenemase genes were identified and described in the sections below. However, the number of the isolates carrying carbapenemase resistance genes were too few to do the analyses described in the following paragraphs.

Relative abundance of ESBL/ampC genes were estimated as the ratio of isolates carrying a particular bla gene family (e.g., bla_CMY) divided by total number of isolates carrying bla_CMY, bla_CTX–M and/or bla_SHV. Furthermore, relative abundance of subclasses of bla_CTX–M genes were estimated as the ratio of isolates carrying a particular bla_CTX–M subclass gene family (e.g., bla_CTX–M–15) divided by total number of isolates carrying bla_CTX–M genes. Relative abundances of ESBL/ampC genes (bla_CMY, bla_CTX–M, and bla_SHV) and bla_CTX–M subclasses at a global and continent level were plotted using library ggplot2 version 3.3.5 (Wickham et al., 2021).

Random Forest models were built to characterize genetic features (virulence factors, plasmid replicons, chromosomal mutations, acquired resistance genes, and plasmid replicon types) that could predict the presence of bla_CMY and bla_CTX–M genes. Two models were built separately for bla_CMY and bla_CTX–M genes (dependent variables). Models were not built for bla_SHV isolates as they were too few in numbers to make reasonable models that can be generalized at a local or global scale. Only one of the perfectly correlated genetic features (r > 0.99) was retained for further analysis using library caret version 6.0-90 (Kuhn et al., 2021). The dataset was divided into training (65% of the dataset) and testing (35% of the dataset) data based on stratified random sampling in order to retain the same proportion of results for the dependent variables in these split datasets (presence of bla_CMY or bla_CTX–M depending on the model) (library caTools version 1.18.2) (Tuszynski, 2021). These models were first built for imbalanced training datasets using quantile-classifier algorithm (O’Brien and Ishwaran, 2019) in library RandomForestSRC (version 2.14.0) with the following settings: mtry = 25, number of trees = 12,000, split rule = AUC, block size = 100. Model performances were evaluated using geometric means. Training models were used to identify the top variables (i.e., variables of importance) that can predict bla_CMY or bla_CTX–M genes in testing data using default settings (library RandomForestSRC version 2.14.0) (Ishwaran and Kogalur, 2021). The top 10 variables of importance were extracted and plotted using library ggplot2 version 3.3.5 (Wickham et al., 2021). Unadjusted odds ratios were also estimated for the top 10 variables of importance to estimate an effect size and directionality and for ease of interpretation.

The stepwise process followed for article selection is presented in Figure 1. Ultimately, data from 394 articles that fulfilled the inclusion criteria was extracted and evaluated in this review. Description of these articles is presented in Table 1. The majority of the articles contained data from European (196 articles; 49.7%) or Asian (125 articles; 31.7%) countries. The studies included were published from 1998 to 2021, although there was a clear increase in the number of articles published after 2010 (70.1% of all articles). EUCAST epidemiological cut-offs (n = 83) and CLSI breakpoints (n = 174) were used to classify isolates as susceptible/resistant in 222 articles. The majority of the articles utilizing EUCAST epidemiological cut-offs were from European countries, whereas CLSI breakpoints were predominantly used in studies from other countries. Overall, data on resistance against 3GCs or carbapenem in E. coli of swine origin from 61 countries were available.

Information on health characteristics of the pigs, time of sample collection and specific geographic location was available in 88.8, 89.8, and 87.8% of the articles, respectively. However, information regarding the age of pigs, sampling scheme employed for isolate collection, husbandry characteristics and previous antibiotic use was only available in 74.4, 52.5, 17.0, and 39.1% of the articles, respectively. These figures on reporting characteristics further decrease if surveillance reports are not considered. For example, if we only include data published in journal articles, then only 19.4% of the studies reported previous antibiotic use. Although we did not analyze this formally, there were very few studies that provided data regarding the clinical history, tissues collected and strain information of the E. coli collected from diseased pigs.

There was a clear disparity between number of isolates characterized phenotypically for 3GC, ceftazidime or carbapenem resistance across continents (Table 1). Overall, there were 175,190 isolates evaluated for phenotypic resistance globally, of which 51.7 and 28.7% of the isolates were collected from Europe and North America, respectively. After factoring in the pig populations by calculating sampling indices, this disparity was even more evident (Figure 2 and Supplementary File 2). Out of 15 countries with the highest sampling indices, 13 were European and rest of the countries were Canada and Grenada. On the other hand, 13 out of 15 countries with lowest sampling indices were Asian (6), African (4), or South American (3) and the rest of the countries were Cuba and Serbia. The exact number of isolates per study are provide in Supplementary File 2.

Pooled proportion estimates of 3GC resistance in E. coli from healthy pigs at the country level are mapped in Figure 3 and provided in Supplementary File 3. Briefly, 3GC resistance was very low (0–1%) and low (1–2%) in isolates collected from 24 and 8 European countries, respectively, with isolates from Portugal having highest proportion of this resistance (2%). These estimates were statistically homogenous for data from 26 out of the 32 European countries. Similarly, the pooled proportion of 3GC resistance ranged between very low and low in countries from North America (Grenada, 0%; Canada, 1%; and United States, 3%) and Oceania (Australia, 2%). In South America, the pooled proportion estimates were also very low to low (Chile, 0%; Argentina, 0%; and Brazil, 8%). However, the estimates from North American, South American, and Oceanic countries were either based on one data-point or were statistically heterogenous, with the exception of pooled proportion of 3GC resistance in Argentina, which was statistically homogenous.

In contrast to the results above, the pooled proportion estimates from African and Asian countries were highly variable. Very low (<1%) pooled proportion estimates of 3GC resistance were estimated for Japan, South Korea, Rwanda, and Uganda. Low percentage of isolates were resistant to 3GCs in four countries (Cambodia, 2%; Philippines, 5%; Vietnam, 8%; and Ghana, 7%). In contrast, for other countries estimates of pooled proportion of 3GC resistance ranged between moderate (Laos, 12%; Indonesia, 12%; China, 16%; Myanmar, 18%; Thailand, 20%; and South Africa, 19%), high (India, 24% and Tanzania, 37%), and extremely high (Nigeria, 91%). These estimates were either statistically heterogeneous or based on a single data-point, except for Japan, Laos, and Uganda.

Data on isolates from diseased pigs was available from 30 countries. Pooled proportion estimates of 3GC resistances at the country level are mapped in Figure 4 and provided in Supplementary File 3. In general, the pooled proportion estimates of 3GC resistance were higher in isolates from diseased pigs compared to healthy pigs. In Europe, North America, and Australia, pooled proportion in these subsets of isolates ranged between very low (<1% in 8 countries) and low (1–6% in 10 countries), with higher (moderate) pooled proportion estimated from Spain (14%), Austria (14%), and United States (19%). In Asian and South American countries, pooled proportion ranged from low (between 1 and 4% in South Korea, Brazil, Vietnam, and Japan) to high (between 24 and 47% in Argentina, India, China, Thailand, and Taiwan). No estimates from diseased pigs in African countries were available. The estimates were statistically homogenous for 10 of the 30 countries (8 European countries, Australia, and Taiwan).

Detailed pooled proportion estimates of ceftazidime resistance are presented in Supplementary File 3. Estimates of resistance to ceftazidime largely mirrored the estimates for frequency of 3GC resistances. Based on the data from 46 countries, the correlation between proportions of ceftazidime and 3GC resistance in isolates from healthy pigs were strong (r = 0.98, Pearson correlation). However, the pooled proportion estimates of 3GC resistance were 6.54 (95% CI: 2.57–10.5) times higher as compared to ceftazidime resistance in isolates collected from healthy animals. There was a moderate correlation between proportion estimates of 3GC and ceftazidime resistance in isolates from diseased pigs (r = 0.58, 10 countries, Pearson correlation), and proportion estimates of 3GC resistance were 1.93 times higher (95% CI: 0.07–3.79) than those of ceftazidime.

Data on carbapenem resistance in isolates from healthy and diseased pigs was available from 47 countries (32 European, 7 Asian, 5 African, and 3 North American) and 9 countries (2 European, 6 Asian, 1 Australia), respectively. Proportions of carbapenem resistance in isolates from healthy pigs were estimated to be zero in 38 of these countries (Table 2). Pooled proportion of carbapenem resistance was 0% in isolates from healthy pigs from the European countries, except Portugal (0.1%). In Asia, these proportions ranged from 0% in Cambodia, Japan, and Laos, to 0.8–3.6% in China, Philippines, Thailand, and Vietnam. Estimates were highly variable in African countries ranging from 0% (Uganda and Rwanda), low (2% South Africa and 5 Tanzania) to 100% [Ghana, on a study performed on 43 isolates (Larbi et al., 2021)].

Pooled proportion estimates of carbapenem resistance were also 0% based on data from isolates from diseased pigs in Australia, Austria, Japan, Switzerland, and Taiwan (Table 2). In addition, a multi-country study in Europe estimated this proportion at 0.1% in isolates from diseased pigs in six countries (Germany, Belgium, United Kingdom, Denmark, Italy, and Netherlands) (Pulss et al., 2017). In this study, resistant isolates were isolated only from Germany and Italy. In China, India, and South Korea, this prevalence ranged between 0.6 and 4.62%, whereas in Thailand this prevalence was 59.1% based on a single study (Pokhrel et al., 2018; Table 2).

Data on pooled fecal carriage rates of 3GC resistance was available from 26 countries (Figure 5 and Supplementary File 3). Fecal carriage rates were lower in European countries as compared to most of the non-European countries. In general, fecal carriage rates were 433 times higher than the pooled proportion estimates. In Europe, pooled fecal carriage rates ranged were <10% in Finland, France, and Norway, between 10 and 20% in Switzerland, Denmark, Belgium, Poland, and Netherlands and between 20 and 45% in Latvia, United Kingdom, Portugal, Austria, Spain, and Germany. In Asia, pooled fecal carriage rates ranged between 5 and 22% in Japan, South Korea, and Taiwan, whereas this rate ranged between 55 and 80% in samples from Vietnam, China, Lebanon, and India. These rates ranged between 44 and 80% in other countries (Cuba, Rwanda, Argentina, Uruguay, and United States), with the exception of Brazil (3%). It should be noted that data from 17 of these 26 countries was based on a single study, and pooled estimates were statistically heterogenous for all but 3 countries (Finland, Japan, and Poland). Still, there was a moderate correlation (r = 0.53, Pearson correlation) between pooled proportion estimates and fecal carriage rates of 3GC resistance. Fecal carriage rate of carbapenem resistance was found to be zero in 6,791 fecal samples collected from Norway, Sweden, Denmark, Finland, and Switzerland. Pooled fecal carriage rate of carbapenem resistance was estimated to be 1.36% from China and India.

The biggest confounding factor in estimation of accurate fecal carriage rates was the use of different antibiotics and concentrations thereof in the media used for isolating 3GC or carbapenem E. coli. Of the 55 articles reporting animal-level or herd-level prevalence of 3GC resistance, 16 employed plates that were commercially available for 3GC detection and rest (39) added various 3GCs to routinely used lab media for 3GC detection (Supplementary File 2). Out of these 39 studies, 32 and 5 used cefotaxime and ceftriaxone at the concentration of 1–4 mg/L to supplement media, respectively. Rest of the studies used cefoxitin and ceftiofur (Supplementary File 2). Similarly, of the 14 articles that provided estimates of carbapenem resistance at herd- or animal-level, 9 used commercial plates and 5 supplemented routine laboratory media with ertapenem, meropenem and imipenem (Supplementary File 2).

Most of the studies focused on identifying genes of interest after selecting isolates on a media containing carbapenems or 3GCs or after identifying individual isolates as ESBL/ampC-producers. We have provided the data on isolates collected without aforementioned selection procedures in the Supplementary File 2, but we will only discuss data on those isolates that were identified based on selection procedures in the rest of this section.

bla_CTX–M were the dominant genes in ESBL/ampC or 3GC-resistant E. coli isolates in Asia and Europe and were present in 84.3% (95% CI: 76.3–91.1%) and 65.3% (95% CI: 55.1–74.9%) of such isolates in Asia and Europe, respectively (Figure 6A). bla_SHV genes were present in 3.42% (95% CI: 0.08–9.55%) and 3.36% (95% CI: 1.09–6.44%) of ESBL/ampC or 3GC-resistant E. coli in Asia and Europe, respectively. bla_CMY genes were present in only 3.58% (95% CI: 0.19–9.41%) and 1.99% (95% CI: 0.55–3.99%) of such isolates in Asia and Europe, respectively (Figure 6A). In contrast, bla_CMY was the most prevalent gene group in 3GC-resistant E. coli in North America and these genes were present in 73.1% (95% CI: 41.9–95.4%) of the isolates; whereas bla_CTX–M and bla_SHV were present in 22.6% (95% CI: 0–65.7%) and 0% (95% CI: 0–1.41%) of such isolates, respectively (Figure 6A).

There were also differences in the distribution of bla_CTX–M groups across different continents (Figure 6B). For example, in Europe bla_CTX–M–1G (consisting primarily of bla_{CTX–M–1, –15}, _–55) was the predominant bla_CTX–M group and was present in 50.5% (95% CI: 40.2–60.7%) of the ESBL/ampC or 3GC-resistant E. coli isolates; whereas in Asia, both bla_CTX–M–1G (54.8%, 95% CI: 44.3–65.2%) and bla_CTX–M–9G (consisting of bla_{CTX–M–14, –27}, _{and –65}) (39.5%, 95% CI: 29.7–49.5%) were the predominant groups.

The frequency of carbapenem resistance genes in carbapenem-resistant and/or ESBL-E. coli isolates are presented in Table 3. In brief, a wide variety of carbapenem resistance bla genes were found to be circulating in pig populations in Asia and Europe. These included bla_{NDM–1, NDM–5}, _{VIM–1, KPC–2}, _{and OXA–48} in several Asian countries (China, India, and South Korea) (Table 3). bla_VIM–1 and bla_OXA–181 were present in pig populations in Germany and Italy (as part of a pan-European study) (Table 3). Carbapenem resistance genes were found to be circulating in E. coli isolates collected between 2001 and 2011 in South Korea and Germany (Table 3), hinting that these genes have been present in pig populations globally since a long time.

Overall, there were 1,223 genomes out of the 6,167 sequences analyzed that carried ESBL/ampC genes (either bla_CTX–M, bla_CMY, or bla_SHV genes) globally. Details of these genomes are available in Supplementary File 4. bla_CTX–M genes were more prevalent compared to bla_CMY and bla_SHV genes in ESBL/ampC isolates from Asia (89.0% bla_CTX–M, 10.5% bla_CMY, and 1.61% bla_SHV) and Europe (77.1% bla_CTX–M, 17.9% bla_CMY, and 6.04% bla_SHV) (Figure 7A). In contrast, bla_CMY were the most prevalent genes in ESBL/ampC isolates in North America (77.9% bla_CMY, 21.3% bla_CTX–M, and 1.15% bla_SHV) (Figure 7A).

Furthermore, there were several different subtypes of bla_CTX–M genes distributed globally in the sampled pig populations. bla_CTX–M–1, bla_CTX–M–55, and bla_CTX–M–14 were the most prevalent bla_CTX–M genes present in 27.8, 27.5, and 20.8% of bla_CTX–M carrying isolates, respectively (Figure 7B). However, the distribution patterns of these bla_CTX–M subtypes varied across different continents (Figure 7B). For example, 40.5 and 50.3% of bla_CTX–M-positive isolates carried bla_CTX–M–55 genes in North America and Asia, respectively, whereas only 4.25% of bla_CTX–M-positive isolates carried this gene in Europe (Figure 7B). Similarly, distributions of bla_CTX–M–1, _–14, _{and –15} genes also varied across continents (Figure 7B).

Only 22 of the E. coli genomes carried a carbapenemase gene. The most common type of carbapenemase gene was bla_NDM–5, that was present in 13 Chinese isolates. Other carbapenemase genes were bla_NDM–1 (China, n = 3), bla_IMP–27 (Canada, United States, n = 3), bla_IMP–38 (China, n = 1), bla_VIM–5 (Belgium, n = 1) and bla_GES–5 (Germany, n = 1). Since carbapenemase genes were so rare, we were not able to generalize their genetic characteristics and global distribution, but this information can be extracted from the Supplemental File 4.

Extended spectrum beta-lactamase/ampC isolates belonged to 319 unique sequence types (STs). Top five genetic characteristics of these isolates are presented in Table 4. Nearly 40% of the ESBL/ampC isolates belonged to clonal complex (CC) 10 (21.9%, including ST10 and ST744), CC23 (10.1%, including ST88 and ST410), and CC155 (7.3%, including ST58); and 75% of ESBL/ampC isolates were either phylotype A (43%) or B1 (32%).

The results of random forest models are displayed in Figure 8. Specific plasmid replicon types (IncI1-alpha, IncC, and IncI2-delta), virulence factors (iro operon) and genes that can confer resistance to aminoglycosides [aph(3′′)-Ib] and florfenicol (floR) were the most common predictors of presence of bla_CMY gene in these E. coli genomes. Four genetic determinants of (fluoro)quinolone resistance (qnrS1, gyrA-S83L, parC-S80I, and gyrA-D87N) were among the top 10 predictors of bla_CTX–M genes in these genomes, with other AMR markers [sul2, floR, aac (3)-II, and mph(A)], plasmid replicon (IncI11-alpha) and virulence factor (clpV/tssH) being the other top 10 predictors of bla_CTX–M presence in isolates in swine populations.