Isolates examined within this study were obtained from three different collections of Salmonella cultures that were previously sampled through (1) a wildlife research study (21) and public health surveillance programs led by (2) the Public Health Agency of Canada, FoodNet Canada (https://www.canada.ca/en/public-health/services/surveillance/foodnet-canada.html), and (3) Public Health Ontario. Salmonella isolates collected through public health surveillance of human, livestock, and environmental sources were identified and included if the samples originated from the same geographic region and time period as the previous wildlife study (Wellington-Dufferin and Region of Waterloo Counties in southern Ontario, 2011–2013). The study region, which includes the cities of Waterloo, Kitchener, Guelph, Cambridge and surrounding areas, has a population of ~1 million people and is situated within a region of intensive agriculture. It is also located within the Grand River watershed (6,800 km²), the largest watershed in southern Ontario (21).

Salmonella isolates for this study component were obtained from a previous three-year repeated cross-sectional study of raccoons on swine farms and in conservation areas (21); only the subset of Salmonella isolates collected on swine farms were selected for sequencing. These samples included fecal swabs from live-trapped raccoons, soil samples collected adjacent to trapped raccoons, and swine manure pit samples. The sampling sites and methods used for live-trapping and processing of raccoons have been described previously (21).

The recruitment and selection of farms for FoodNet Canada has previously been described (26). Livestock samples included pooled samples from swine, chickens (broilers and layers), and cattle (dairy and beef). Samples included both fresh fecal samples and manure pit samples for all species except for layers, from which only fresh fecal samples were obtained. Additional details about the sampling approach used are available in Flockhart et al. (26).

River water samples were obtained from five core water sites upstream of regional drinking water intake in the Grand River watershed. Beach water samples were obtained from local swimming sites in three different recreational areas in southern Ontario (Shade's Mills, Laurel Creek, Elora Gorge). Details regarding the approach and methods used to sample water from rivers and beaches have previously been described by Kadykalo et al. (27).

Salmonella isolates from human clinical cases collected by Public Health Ontario as part of the reportable disease surveillance program were cultured from blood, stool, and urine specimens (https://www.publichealthontario.ca/en/diseases-and-conditions/infectious-diseases/enteric-foodborne-diseases/salmonellosis).

Isolation of Salmonella was previously performed according to methods described in 'National Integrated Enteric Disease Surveillance Program: Sample Collection, Preparation and Laboratory Methodologies, January 2010' (https://www.phac-aspc.gc.ca/foodnetcanada/pdf/lab_sop-eng.pdf). One Salmonella isolate from each sample was sub-cultured and tested further. Presumptive-positive colonies were confirmed via biochemical reactions on triple sugar iron slant and urea slant agars (Becton, Dickinson). Additional confirmation was performed using Salmonella O Poly A-I & Vi antiserum (Becton, Dickinson). All isolates were previously tested for susceptibility to the following 15 antimicrobials by broth microdilution (Sensititre, CMV3AGNF panel, NARMS; Thermo Scientific): gentamicin (GEN), kanamycin (KAN), streptomycin (STR), amoxicillin-clavulanic acid (AMC), cefoxitin (FOX), ceftiofur (TIO), ceftriaxone (CRO), ampicillin (AMP), chloramphenicol (CHL), sulfisoxazole (SOX), trimethoprim-sulfamethoxazole (SXT), tetracycline (TCY), nalidixic acid (NAL), ciprofloxacin (CIP), and azithromycin (AZM). Susceptibility testing of livestock, wildlife, and environmental isolates was previously performed by the Antimicrobial Resistance Reference Laboratory, at the National Microbiology Laboratory (NML) at Guelph (Public Health Agency of Canada, Guelph, ON, Canada) in accordance with methods outlined by The Canadian Integrated Program for Antimicrobial Resistance Surveillance [CIPARS; (28)]. Isolates with intermediate resistance were classified as susceptible.

All raccoon fecal, soil, and manure pit samples originating from swine farms in the previous wildlife study (21) were selected for sequencing. Previously sequenced human Salmonella isolates originating from the Wellington-Dufferin-Guelph or Region of Waterloo Public Health Counties in Ontario between 2011 and 2013 were included in the present study. Previous selection of human isolates for sequencing as part of CIPARS surveillance included routine surveillance of 11 serovars (4, [5], 12,i:-, Dublin, Enteritidis, Heidelberg, Infantis, Kentucky, Newport, Paratyphi A, Paratyphi B, Typhi, and Typhimurium), in addition to supplementary sequencing by request or for specific research projects investigating AMR [e.g., ciprofloxacin resistance, extended-spectrum beta-lactamases, bla_CMY−2 in S. Heidelberg; (29–31)]. For further details regarding CIPARS methods, please see the Design and Methods section of the annual report (32). Previously sequenced Salmonella from water and livestock collected by FoodNet Canada were included in the present study if they had been collected within the Region of Waterloo sentinel site between 2011 and 2013. Previous selection for sequencing of livestock isolates collected by FoodNet Canada was done on the basis of metadata; up to five isolates with identical metadata were sequenced (i.e., same submitter, collection date, commodity, serotype, and phage-type, if applicable); contaminated isolates and isolates without a serotype were not eligible for sequencing.

Genomic DNA extractions were performed at the University of Guelph, or at the NML in Winnipeg, Manitoba. Briefly, Salmonella cultures were grown on Mueller Hinton agar and incubated at 35°C overnight. Cultures were then sent to the NML in Winnipeg for DNA extraction and sequencing, or these steps were performed on site, at the University of Guelph and the NML at Guelph, Ontario, respectively. Cultures of 1 ml Salmonella were used as input to the Qiagen DNEasy plant and tissue 96 kit, following manufacturer protocols (Qiagen, Hilden, Germany). Sequencing was performed using the Nextera XT libraries and Illumina MiSeq version 3 (600-cycle kit) or NextSeq550 platforms, according to the manufacturer's protocols. Raw reads were assembled using SPAdes (33), as part of the Shovill pipeline (version 1.0.1; https://github.com/tseeman/shovill) with the following settings: “–minlen 200 –mincov 2; –assembler spades; –trim”.

Prediction of legacy multi-locus sequence types was performed using MLST (version 2.19.0; https://github.com/tseemann/mlst), which uses the Achtman 7-loci scheme for Salmonella enterica (https://pubmlst.org/mlst/). Core genome multi-locus sequence typing (cgMLST) of all isolates was performed using the “fairly simple allele calling” tool fsac (version 1.2.0; https://github.com/dorbarker/fsac) with the 3,002-loci Enterobase scheme (https://enterobase.warwick.ac.uk/). Isolates with 20 or more missing loci were considered poor quality and excluded from any further analyses. Isolates with greater than two missing genotypes for resistant phenotypic test results (by antimicrobial class), and a sequencing depth of <30 using Mash (34) were also excluded from further analyses. A threshold of n = 10 was used as the maximum number of allelic differences present for Salmonella to be considered as possibly belonging to the same strain, consistent with the threshold used by PulseNet (35). Population structure was visualized using cgMLST data and minimum spanning trees generated by the standalone GrapeTree software package [version 1.5; (36)], using the “MSTreeV2” algorithm, which accounts for missing data. For the minimum spanning tree visualizing the overall population of Salmonella, a lenient clustering threshold was used (k = 30) to provide a qualitative assessment of overlap between isolates from different sources, while minimizing unnecessary noise. An additional minimum spanning tree was constructed to examine isolate overlap for only serovars present in both human and raccoon isolates; a stricter clustering threshold (k = 10) was used to provide additional resolution and to minimize clustering of isolates from potentially different strains, consistent with the PulseNet threshold (35). Allelic differences between cgMLST profiles from different sources were calculated using R (version 3.6.3). R code is available at https://github.com/nadinevogt21/salmonella-human-raccoon-WGS-project. In silico identification of serovars was performed using the SISTR command-line tool (version 1.1.1; https://github.com/phac-nml/sistr_cmd), and default settings with the “centroid” allele database.

Acquired resistance genes were identified using CARD-RGI (database version 3.0.8; https://github.com/arpcard/rgi) which uses a curated database that includes optimized identity and coverage settings for each gene (37). The following setting was used: “–exclude nudge” (i.e., loose hits were not nudged to strict). Only acquired resistance genes identified with perfect and strict hits under the “protein homolog model” were reported; regulatory genes, genes requiring mutation for expression, and genes representing chromosomally encoded resistance genes were not reported. In silico identification of plasmid incompatibility groups was performed using Abricate (version 1.0.1; https://github.com/tseemann/abricate) and the PlasmidFinder database (updated May-17 2020) which uses the database from the Center for Genomic Epidemiology, Technical University of Denmark, DTU (https://cge.cbs.dtu.dk/services/PlasmidFinder/). Settings included nucleotide coverage of 70% and percent identity of 98%.

For raccoon and human isolates with fewer than 10 allelic differences based on the cgMLST scheme, Snippy (version 4.4.0; https://github.com/tseemann/snippy) was used to further distinguish genetic differences based on single-nucleotide polymorphisms (SNPs) using the entire genome, with default settings. Salmonella enterica subsp. enterica serovar Heidelberg str. SL476 (Accession No.: NC_011083.1) was used as a reference genome.

Multi-level logistic regression was used to model the odds of the presence of select plasmid incompatibility groups and acquired resistance genes found in Salmonella from different animal, human, and environmental sources. All statistical tests were performed using STATA (STATA Intercooled 14.2; StataCorp, College Station, Texas, USA). Only incompatibility groups or resistance genes with an overall prevalence of at least 10% and at most 90% were analyzed. Incompatibility types that were present only or predominantly (>95%) in one host-restricted serovar were reported descriptively only, due to the potential for complete confounding by serovar. Human and raccoon sources were retained as categories for analyses, and additional sources were grouped together into the following categories: livestock (cattle, swine, chicken), environmental (water, soil). Logistic regression models with source type as an independent variable were initially constructed using the “melogit” command, with a random intercept to account for clustering of isolates obtained from the same raccoon or swine manure pit (applicable to isolates from the wildlife research study). Site-level data for isolates obtained through FoodNet Canada were unavailable, thus, isolates that may have originated from the same farm or water source were treated as independent observations. For multi-level models that did not converge using “melogit”, the model was fitted using QR decomposition of parameters, using the command “meqrlogit”. All models were adjusted for the potential confounding effect of sampling year, if deemed appropriate. Sampling year was retained in the model based on the following criteria: it was statistically significant (p ≤ 0.05), or its removal resulted in a >20% change in the coefficient of the source type variable (38). To examine the fit of multi-level models, Pearson's residuals were assessed for outliers, and the best linear unbiased predictions (BLUPs) were assessed for normality and homoscedasticity. The random intercept was removed if the variance components were very small (<1 ×10⁻³) and if model fit was not improved based on changes to the Bayesian Information Criterion (BIC) value (38). If the random intercept was not retained in the final model, the model was fit using ordinary logistic regression. For ordinary logistic regression models, model fit was assessed using the Pearson's goodness-of-fit test, in addition to visual assessment of leverage, delta-beta, Pearson, and deviance residuals to identify any potential outliers. Outliers were investigated for recording errors, but otherwise left in the model. Exact logistic regression was used if low effective sample sizes posed estimation issues or to obtain median unbiased estimates if certain categories had zero positive observations; the score method was used to calculate p-values for these models. All tests were two-tailed, and a significance level of α = 0.05 was used.

A total of 628 isolates were identified using the study criteria. After excluding isolates with 20 or more missing loci based on the 3,002-loci cgMLST scheme, 608 isolates were available for subsequent analyses from the following sources: human (n = 58), raccoon (n = 92), chicken (n = 214), cattle (n = 60), swine (from FoodNet Canada, n = 34), swine manure pit (from previous wildlife study; n = 21), soil (n = 45), water (n = 76), and beach (n = 8). Among isolates derived from cattle samples, these were roughly equally represented by beef (n = 29/60) and dairy animals (n = 31/60). The majority of chicken isolates were obtained from broilers (n = 204/214), with a handful of isolates from laying hens (n = 10/214). The subset of isolates obtained as part of the previous wildlife study (i.e., raccoon, soil, and swine manure pit samples) totaled 158; accession numbers for these isolates are available in Supplementary File 1. Isolates from all sources were evenly distributed across different sampling years: 2011 (36%), 2012 (34%), 2013 (30%). Since all human isolates that were included in this study exhibited phenotypic resistance to at least one antimicrobial (as part of the selection criteria for previous research projects), the prevalence of phenotypic multiclass resistance (to at least three drug classes) was highest in this source (65.5%, 95%CI: 51.8–77.5%). The prevalence of multiclass resistant and non-resistant isolates in the remaining source categories (all of which included phenotypically resistant and susceptible isolates, and were not selected based on phenotypic resistance) was as follows: livestock (10.6% multiclass resistant, 95%CI: 7.5–14.5%, 57.4% non-resistant, 95%CI: 51.9–62.8%), raccoon (2.2% multiclass resistant, 95%CI: 0.2–7.6%, 96.7% non-resistant, 95%CI: 90.8–99.3%), and environmental (3.1% multiclass resistant, 95%CI: 0.9–7.7%, 85.3% non-resistant, 95%CI: 77.9–90.9%). Results from previous phenotypic susceptibility testing of the 158 isolates from the swine farm subset is available in Bondo et al. (21).

A total of 45 different serovars were identified, and the most commonly identified serovars among all isolates were S. Kentucky (n = 132), S. Heidelberg (n = 99), S. Newport (n = 55), S. Typhimurium (n = 39), S. Agona (n = 38), S. Infantis (n = 33) and S. Enteritidis (n = 30; Figure 1A). Population structure based on cgMLST and the distribution of source types for all isolates is illustrated in Figure 1B. For population structure of all isolates by detailed sampling source, see Supplementary Figure 1. Among raccoon isolates, the most frequently identified serovars were S. Newport (n = 33/92, 36%), S. Agona (n = 16/92, 17%), S. Infantis (n = 9/92, 9.8%), and S. Paratyphi var. Java (n = 9/92, 9.8%). The most prevalent serovars in human isolates were S. Enteritidis (n = 14/58, 24%), S. Heidelberg (n = 14/58, 24%), and S. Kentucky (n = 11/58, 19%); the latter two serovars were infrequently represented, or not present, among environmental and raccoon isolates, but were commonly identified in samples from chickens and cattle (Figure 1B). A heatmap of the distribution of serovars identified in humans in this study that were also identified in other sources is provided in Supplementary Table 1. Salmonella Enteritidis was isolated from all sources except for swine isolates. Several serovars that were infrequently identified in human isolates (<5%) were not identified in any other sample in our study (i.e., S. Litchfield, S. Dublin, S. Chester; Supplementary Table 1). Serovars present in both humans and raccoons, in decreasing frequency relative to raccoons, included: S. Newport, S. Infantis, S. Typhimurium, S. Heidelberg, and S. Enteritidis. The genetic relationships among isolates of S. Newport, S. Infantis, and S. Typhimurium are provided in Figure 2; since only one isolate each of S. Heidelberg and S. Enteritidis was identified in raccoons, these serovars were reported descriptively only. For detailed information regarding the distribution of all serovars by source, see Table 1.

A total of 54 sequence types were identified, and the distribution of the 11 most common sequence types by source is provided in Table 2. Seven isolates were not typeable by MLST (two raccoon, one soil, two water, one swine, one chicken) due to a missing allele, or a partial match. The following internationally recognized sequence types (39) were identified in our study population: S. Heidelberg ST15 (n = 95), S. Typhimurium ST19 (n = 33), S. Infantis ST32 (n = 32), ST198 S. Kentucky (n = 15), ST96 S. Schwarzengrund (n = 7), ST10 S. Dublin (n = 2), ST45 S. Newport (n = 2), and ST65 S. Brandenburg (n = 2). As chickens were the most common source type in this study, sequence types from poultry-adapted serovars such as S. Kentucky ST152, and S. Heidelberg ST15 were most common (Table 2). Salmonella Kentucky isolates identified in humans (exclusively ST198) were infrequently identified in chicken isolates (n = 4; Table 2). Among S. Kentucky isolates from chickens, the vast majority (n = 109/114) of these were ST152, and the remaining isolates were ST198 (n = 4/114); one sample could not be typed due to a partial or missing allele. Most S. Newport isolates from raccoons and environmental sources were ST350, whereas the two human S. Newport isolates identified in our study were both ST45. Certain sequence types, including S. Infantis ST32 and S. Typhimurium ST19, were identified in every source. Other sequence types were identified only or predominantly in certain animal sources, but could also be found in water samples: S. Cerro ST367 and S. Give in cattle. For detailed information regarding the distribution of sequence types by source, see Supplementary Table 2.

Serovars identified among both human and raccoon isolates were: S. Enteritidis, S. Heidelberg, S. Infantis, S. Newport, and S. Typhimurium (Figure 2). Based on our previously specified cluster threshold of 10 loci, none of the human and raccoon isolates clustered together for any of the serovars present in both host species: S. Infantis, S. Newport, and S. Typhimurium isolates with a minimum of 73, 359 and 76 different loci, respectively (Figure 2). Single antimicrobial susceptible isolates of S. Enteritidis ST11 and S. Heidelberg ST15 were identified in raccoons and differed from human isolates by a minimum of 61 and four loci, respectively. Further analysis of the single raccoon S. Heidelberg ST15 isolate compared to the five closely related human isolates (<10 cgMLST loci different) revealed a minimum of 15 and a maximum of 21 SNPs that differed between these isolates.

Salmonella Infantis was identified in all source types; all 33 isolates were within 107 loci of one another, whereas all non-human sources containing S. Infantis could be grouped together at a lower threshold of 86 loci. The minimum number of allelic differences to the most closely related S. Infantis human isolate was similar for raccoons (73 loci) and livestock (72 loci) but was lower for environmental samples (56 loci). In contrast, the 30 S. Enteritidis isolates which were found in all source types (but predominantly humans and livestock) clustered together at higher threshold of 441 loci. Among Salmonella Enteritidis isolates (n = 30), livestock and environmental isolates displayed greater similarity to human isolates (a minimum of 12 and 16 allelic differences, respectively) compared to the single raccoon S. Enteritidis isolate (at least 61 allelic differences). Salmonella Typhimurium, also isolated from all sources (n = 39 isolates), displayed a high clustering threshold for all isolates, with all but two isolates grouped together at a threshold of 377 loci. The minimum numbers of allelic differences for S. Typhimurium isolated from raccoons, livestock, and environmental samples compared with human isolates were comparable (76, 79, and 69 loci, respectively). Serovars such as S. Heidelberg (n = 99) that were predominantly isolated from chickens displayed a considerably lower clustering threshold, together with isolates from humans, water, cattle, and the single raccoon S. Heidelberg isolate (n < 37 allelic differences for 85% of isolates). Identical S. Heidelberg were identified between livestock (chicken) and human samples (0 allelic differences), whereas a minimum of at least 16 and four allelic differences were noted for environmental isolates and the single raccoon isolate, respectively.

Two distantly related clusters of S. Kentucky were identified (>2,700 allelic differences). Salmonella Kentucky (n = 132) identified in environmental samples differed from the nearest human isolates by 2,758 loci, compared to the minimum of 139 loci that differed between livestock and human isolates (S. Kentucky was not identified among raccoon isolates). Finally, for serovars such as S. Newport (n = 55) that were mainly identified in environmental sources (e.g., soil) and raccoon isolates, many had identical cgMLST types (0 allelic differences). The two human S. Newport isolates differed from the most closely related raccoon S. Newport isolate at 359 and 2,592 loci, respectively. In contrast, the minimum number of allelic differences between S. Newport from environmental isolates and human isolates was lower (126 loci).

A total of 65 acquired resistance genes and 43 incompatibility groups were identified among all isolates. Commonly identified resistance genes with an overall prevalence >10% were: fosA7, tet(B), aph(6)-Id, aph(3”)-Ib, aac(6')-Iaa, and bla_CMY−2 (Table 3). Human isolates contained the greatest diversity of resistance genes, with 58/65 genes identified. Other sources contained fewer resistance genes, in decreasing order: livestock (n = 29/65 genes), environmental (n = 27/65 genes), and raccoon isolates (n = 9/65 genes). A number of human and livestock isolates, along with three water isolates, contained bla_CMY−2, which confers the cephamycinase-producing phenotype. Class A and C beta-lactamase genes responsible for conferring the extended-spectrum beta-lactamase (ESBL) phenotype (e.g., bla_CTX−M−14, bla_CTX−M−65) were only identified among human isolates. One human isolate was contained a mcr-1 gene conferring resistance to colistin. Two human isolates with the ACSSuT (ampicillin, chloramphenicol, streptomycin, sulfamethoxazole, and tetracycline) resistance pattern were identified, one in a ciprofloxacin-resistant S. Typhimurium, and another in a S. Infantis resistant to third-generation cephalosporins.

Few resistance genes were identified among raccoon isolates (Table 3), corresponding with the n = 3/92 isolates that demonstrated phenotypic resistance (one isolate with STR-TCY, and two isolates with SOX-STR-TCY). Most of the resistance genes identified in raccoon isolates conferred resistance to aminoglycosides (n = 11/35 genes) and fosfomycin (n = 19/35 genes); no beta-lactam, phenicol, or quinolone resistance genes were identified in isolates from any of these animals (Table 3). In contrast, environmental isolates contained resistance genes to several major classes of antimicrobials (including beta-lactamases, phenicols, and quinolones; Table 3). The most commonly identified plasmid incompatibility groups are presented in Table 4. Some incompatibility groups were only identified in certain sources (e.g., ColE10 in swine), or appeared predominantly in certain sources (e.g., IncX1-3 in chicken isolates, IncFiip96a in raccoon and environmental isolates). Other incompatibility groups were identified in all source types (e.g., IncI1-Igamma, IncFIIS). For a summary of test sensitivity and specificity of in silico AMR prediction using phenotypic susceptibility testing results as the gold standard, see Supplementary Table 3. The overall sensitivity and specificity of in silico identification of AMR genes were 96.3 and 97.0%, respectively.

Except for IncI1-Igamma, all incompatibility groups and resistance genes examined had a significant association with source type. Certain incompatibility groups were significantly more likely to be identified in human or livestock than in raccoon or environmental isolates (i.e., IncX1-1, colRNAI, colpVC, IncI1-Igamma), whereas others were significantly more likely to be identified in raccoon and environmental isolates compared to human and livestock isolates (i.e., IncFIIS, IncFiip96a; Tables 5, 6). Incompatibility groups IncX1-3 and IncFIB(AP001918) were not modeled since they were exclusively, or almost exclusively (>95%), identified in one serovar (Salmonella Kentucky), and the vast majority (>85%) of these isolates originated from broiler chickens. Most resistance genes analyzed were significantly more likely to be identified in livestock and/or human isolates compared with environmental and/or raccoon isolates (i.e., tetB, aac(6′)-Iaa,aph(6)-Id,aph(3″)-Ib,blaCMY-2). In particular, the odds of identifying tet(B) and aph(3”)-Ib were highest in livestock isolates, and the odds of identifying bla_CMY−2 were highest in human isolates (Tables 5, 6). The odds of identifying bla_CMY−2 were also significantly greater in livestock isolates compared to raccoon or environmental isolates (Table 6). Random intercepts were not retained in several models since variance components were extremely small (i.e., ColpVC, tetB, aac6'-Iaa, aph-6-Id, fosA7) or exact logistic regression was used (i.e., bla_CMY−2, aph3”-Ib, IncFiip96a).