Impact of Season, Demographic and Environmental Factors on Salmonella Occurrence in Raccoons (Procyon lotor) from Swine Farms and Conservation Areas in Southern Ontario

Salmonella has been detected in the feces of many wildlife species, including raccoons (Procyon lotor), but little is known about the epidemiology of Salmonella in wildlife living in different habitat types. Our objective was to investigate demographic, temporal, and climatic factors associated with the carriage of Salmonella in raccoons and their environment on swine farms and conservation areas. Using a repeated cross-sectional study design, we collected fecal samples from raccoons and environmental samples (soil, manure pits, dumpsters) on 5 swine farms and 5 conservation areas in Ontario, Canada once every five weeks from May to November, 2011–2013. Salmonella was detected in 26% (279/1093; 95% CI 22.9–28.2) of raccoon fecal samples, 6% (88/1609; 95% CI 4.5–6.8) of soil samples, 30% (21/69; 95% CI 20.0–42.7) of manure pit samples, and 23% (7/31; 95% CI 9.6–41.0) of dumpster samples. Of samples testing positive for Salmonella, antimicrobial resistance was detected in 5% (14/279; 95% CI 2.8–8.3) of raccoon fecal, 8% (7/89; 95% CI 3.2–15.5) of soil, 10% (2/21; 95% CI 1.2–30.4) of manure pit, and 0/7 dumpster samples. Using multi-level multivariable logistic regression analyses, we found location type (swine farm or conservation area) was not a significant explanatory variable for Salmonella occurrence in raccoon feces or soil (p > 0.05). However, detection of Salmonella in raccoon feces was associated with rainfall, season, and sex with various interaction effects among these variables. We detected a variety of Salmonella serovars that infect humans and livestock in the feces of raccoons indicating that raccoons living near humans, regardless of location type, may play a role in the epidemiology of salmonellosis in livestock and humans in southwestern Ontario.


Introduction
tributaries [39]. Each site was trapped once every 5 weeks during this period, and environmental samples were collected at each site during the same week that animals were trapped.
All of the farm sites selected for this study were part of FoodNet Canada, a sentinel site based enteric pathogen surveillance program [40], and identified themselves as being primarily swine farms. The farms were chosen based on their proximity to the University of Guelph, and included farrow-to-young grower and farrow-to-finish operations. Up to 2500 pigs were housed indoors on each farm. In addition to pigs, one farm had 100-110 dairy cattle on site, another farm had approximately 100 layer chickens housed indoors, and another swine farm had a poultry operation and horses and ponies on site. Only waste from pigs was stored in the manure pits sampled in this study and poultry manure was not spread on the farms' fields. Three of the five farms did not administer injectable antimicrobials or antimicrobials in-feed to swine. The other two farms administered both injectable and in-feed antimicrobials to swine.
The conservation areas ranged in size from 75-1608 ha and were all located primarily in peri-urban areas. Habitat types within the conservation areas included mixed hardwood and coniferous forests, ponds, lakes, and wetlands. Recreational activities in many of these areas included hiking, fishing, picnicking, camping, and swimming.

Sample Collection
Raccoons were live-trapped and processed as described previously [41]. Briefly, at each site, 20-40 Tomahawk live traps (Tomahawk Live Trap Co.; Tomahawk, Wisconsin, USA) were set 3-4 nights/week at each site in areas with limited public access, but where raccoons were known to be present, including around dumpsters and buildings. Upon capture, raccoons were anesthetized using an intramuscular injection of 0.025 mg/kg dexmedetomidine hydrochloride (Dexdomitor 0.5 mg/ml; Pfizer Animal Health, Kirkland, Quebec, Canada) and 5 mg/kg ketamine hydrochloride (Vetalar 100 mg/ml; Bioniche Animal Health, Belleville, ON, Canada). For subsequent identification, a numbered metal ear tag (1005-3, National Band and Tag Co.; Newport, Kentucky, USA) was placed in one ear and a passive integrated transponder tag (GPT12 Pre-Load Sterile; Biomark, Boise, Idaho, USA) was injected subcutaneously between the shoulder blades. Sex, age class (adult or juvenile, on the basis of animal size and teeth wear/ staining), and body mass were recorded for each animal. Fecal swabs were collected per rectum using Cary-Blair applicators (BBL CultureSwab, BD; Becton, Dickinson and Company, Annapolis, Maryland, USA). Although individuals were only sampled once per trapping week, multiple samples were collected from the same individual if they were caught in subsequent trapping sessions.
Ten to twenty soil samples were collected within a 2-m radius of the traps on the first day of each trapping week at each study site. Approximately 10 g of soil, free of obvious fecal contamination, was collected into sterile containers. At each swine farm, one manure pit sample was collected on the first day of each trapping week. To collect manure pit samples, a 24' Nasco Swing Sampler (Conbar, Monroeville, NJ, USA) was used to collect three sub-samples from three locations around the pit, and up to two depths (i.e., the top 1/3, and mid depth of the storage), for a total of 6 sub-samples. The sub-samples were poured into a sampling bucket with a clean plastic liner and mixed to create the pooled sample and then transferred into a sterile vial. During 2013, dumpster samples were collected, when available, each month, from 2-3 dumpsters/site from three conservation areas. The other two conservation areas were not sampled because dumpsters were not present on these sites. Dumpster samples were collected < 1 day after dumpsters were emptied by swabbing the bottom of the dumpster using a Swiffer1 wipe (Armstrong, Proctor and Gamble, Cincinnati, Ohio, USA) attached to a 1-2 m extension pole (Bennett, Concord, Ontario, Canada). Swiffer1 wipes were kept moist by placing them in Whirl-pak bags (Nasco, Fort Atkinson, Wisconsin, United States) containing approximately 20 ml of sterile 0.85% saline solution kept at 4°C before and after sampling.

Laboratory Work
All sample types were labeled and placed in a cooler in the field and refrigerated upon return to the laboratory before being submitted for Salmonella isolation. The methods used have been described previously [35]. To prepare the dumpster samples for assay, 50 ml of buffered peptone water was added to the Swiffer1 sample, mixed and incubated at 37°C for 24hrs. Then 0.1 ml of liquid was inoculated into modified semi-solid Rappaport vassiliadis and incubated at 42°C for 24-72hrs.
Salmonella isolation was performed at the McEwen Lab, Canadian Research Institute for Food Safety, University of Guelph. One Salmonella isolate from each positive sample was submitted for serotyping, phagetyping (for serotypes Typhimurium, Enteritidis, and Heidelberg), and antimicrobial susceptibility testing at the National Microbiology Laboratory (formerly the Laboratory for Foodborne Zoonoses), Public Health Agency of Canada (Guelph, Ontario, Canada).

Statistical Modeling
The primary outcome variable for raccoon fecal, soil, manure pit, and dumpster samples was Salmonella (positive vs. negative). To compare the prevalence of Salmonella in raccoon fecal, soil, manure pit, and dumpster samples, type of sample was included as a fixed effect. For all other statistical modeling, samples were analyzed by sample type. Explanatory variables included, if applicable, raccoon sex (male or female), raccoon age (adult or juvenile), location type (swine farm or conservation area), year (2011-2013), sum of rainfall over 30 days prior to sample collection, mean temperature over 30 days prior to sample collection, and season (May-July or August-November). Two distinct seasons were considered: rearing (May-July) and pre-denning/dispersal (August-October) as defined by Rosatte et al. [38]. Sum of rainfall and mean temperature prior to 30 days of sample collection were used because this time period for rainfall was found to be significantly associated with Salmonella carriage in free-ranging wildlife in other studies [29,43]. Mean daily temperature and total rainfall per day were downloaded from Environment Canada from the nearest weather station with complete data (Fergus Ministry of the Environment (MOE), ON) from 2011 to 2013. Missing values were filled in using data from the next nearest weather station in Guelph, Ontario.
Raccoon fecal and soil samples were modelled using multi-level univariable and multi-variable logistic regression. Univariable models were initially constructed and then variables were retained in the final model if they were significant, part of a significant interaction term or acted as a confounding variable. A variable was considered to be a confounding variable if it was a non-intervening variable and its removal from the model resulted in ! 20% change in the coefficients of any statistically significant variable [44]. The significance level was set at α = 0.05. Additional details regarding the building of multi-variable models are described in the S1 Appendix.
Manure pit and dumpster samples were analyzed using univariable exact logistic regression due to a small effective sample size for these dependent variables. No statistical tests were conducted on antimicrobial resistant results due to the rarity of this outcome being detected. To determine the association between the presence of the most common Salmonella serovars in raccoon feces (Oranienburg, Newport, I:4, [5],12:b:-, and Thompson analyzed separately) in relation to all other Salmonella serovars, univariable logistic regression models with random effects were also used. Factors examined for association with the most common Salmonella serovars included year, location type, season, age, and sex of the raccoon.
For each model, the odds ratio and 95% confidence interval of each variable were reported. The significance of main effects, categorical variables with more than two categories and interaction terms were assessed using Wald's χ 2 test. Independent continuous variables that had nonlinear relationships with the outcome variable based on a significant quadratic term and on visual assessment of the lowess curve were categorized or modelled as a quadratic relationship if appropriate.
Random effects were included to account for autocorrelation among samples taken from the same site, animal, manure pit, or dumpster. A random effect was excluded from the final model if the variance components were very small and if the model fit was not improved based on assessing changes in Akaike's Information Criterion (AIC) and Bayesian Information Criterion (BIC) values to the removal of the random effect. If the model fit was not improved by the inclusion of random effects, then ordinary logistic regression was used to analyze the data. To determine the amount of variation explained by each organizational level (i.e., site and sample levels), variance partition coefficients (VPCs) were estimated from the variance components of the final multi-level logistic regression models using the latent variable technique [44]. All statistical tests were conducted using STATA (STATA Intercooled 13.0; StataCorp, College Station, Texas, USA).

Assessing model fit
For multi-level models, Pearson and deviance residuals were used to determine if there were any outlying observations, and best linear unbiased predictors (BLUPs) of the random effects were examined to assess overall model fit. Specifically, we graphically assessed if the BLUPs fulfilled the assumptions of normality and homogeneity of variance. If the BLUPS did not meet these model assumptions, we compared models with and without the random intercept(s) using AIC and BIC to confirm that the addition of these terms improved model fit.
For ordinary logistic regression models with continuous independent variables, the Hosmer-Lemeshow goodness-of-fit test was used to assess model fit and for those without continuous variables, Pearson's goodness-of-fit test was used. In addition, for regular logistic regression models, leverage, delta-beta, Pearson, standardized Pearson and deviance residuals were used to visually assess if there were any outlying observations for these models. If any outliers were found in either the multi-level or regular logistic regression models, they were investigated for recording errors; otherwise, they were left in the model.

Results
We collected 1096 fecal samples from 627 individual raccoons, 1609 soil samples, 69 manure pit samples, and 31 dumpster samples. Three individual raccoons were caught at two different swine farms during the study, so we randomly selected and excluded one result each from the analyses. Sex and age were undetermined for one and three raccoons, respectively. Salmonella was detected in 26% raccoon fecal, 6% soil, 30% manure pit, and 23% dumpster samples ( Table 1). The proportion of all sample types testing positive for Salmonella by age, sex, location type, season, and year categories are also presented in Table 1.

Raccoon Fecal Samples
In the univariable models, Salmonella occurrence in raccoon feces was significantly associated with sex, season and rainfall (S1 Table). The final multivariable model for raccoon fecal samples included sex, season, rainfall and season-sex and season-rainfall interaction terms ( Table 2).
The contrasts derived from the models showed that male raccoons were more likely to carry Salmonella during May to July than female raccoons (Table 3). In addition, male and female raccoons were more likely to carry Salmonella during August to November than May to July (Table 3). Higher rainfall was associated with a lower predicted probability of Salmonella in raccoon feces in both seasons; however, the relationship was much more pronounced August to November than May to July (Fig 2).

Environmental samples
In the univariable models, Salmonella detection in soil samples was significantly associated with season and year, and there was a significant quadratic relationship between Salmonella occurrence in soil samples and mean temperature over 30 days prior to sample collection (S1 Table). The final multivariable model for Salmonella occurrence in soil samples included year and season. In this model, soil samples collected between August and November were more likely to be Salmonella positive than samples collected between May to July ( Table 2). In addition, soil samples collected in 2013 were less likely to be Salmonella positive than samples collected in 2011 and 2012 (Table 2).
Manure pit samples were significantly less likely to be Salmonella positive in 2013 than in 2011 and 2012 in the univariable analysis (S1 Table). In the main effects model for manure pit samples, Salmonella was significantly more likely to occur in samples collected in August to November than May to July and less likely to occur in year 2013 than in 2011 and 2012 ( Table 2). In the univariable analysis for dumpster samples for season, there was no significant difference in Salmonella occurrence in dumpster samples August to November compared to May to July (OR = 0.2; 95% CI, 0.02-1.64, p = 0.099). Salmonella serovars Salmonella serovars detected in raccoon fecal, soil, manure pit, and dumpster samples are presented by sample type and location type (Table 4). Thirty and 19 Salmonella serovars were detected in raccoon feces and soil samples, respectively. Salmonella serovars Newport, Oranienburg, I,4,[5],12:b-, and Thompson were the most commonly detected in raccoon fecal samples, and Thompson, Newport, Typhimurium, Oranienburg and Agona were among the most common in soil samples (Table 4).
In the univariable analysis for the most common serovars detected in raccoon fecal samples, Salmonella Newport was more likely to occur in raccoon feces than other serovars on swine farms than conservation areas (S2 Table), and Salmonella Oranienburg was more likely to occur than other serovars in female than male raccoon feces, less likely for juveniles than    Table). There were no significant associations in any of the univariable analyses for Salmonella I, 4,[5],12:b-or Thompson (S2 Table).

Diagnostics for Residual Analyses
There were no outlying observations with recording errors in any of the multi-level or logistic regression models. Although the BLUPS were not normally distributed in all of the multi-level b Other serovars are those that were detected once and in only one sample type. They include the following:  Impact of Season, Demographic and Environmental Factors on Salmonella Occurrence in Raccoons models, the random effects were included in the models since model fit was improved based on the reduction in AIC and BIC when these effects were included. For models constructed using regular logistic regression (i.e., no random effects), all models fit the data based on non-significant Hosmer-Lemeshow (binary data) and Pearson goodness-of-fit (binomial data) tests.
Of the Salmonella positive samples, antimicrobial resistance was detected in 5.2% (10/187; 95% CI 2.6-9.6) of raccoon fecal and 9.5% (4/42; 95% CI 2.6-22.6) of soil samples from conservation areas as compared to 4.3% (4/92; 95% CI 1. The most common resistance observed was to AMP (10/23), TCY (8/23) and STR (8/23); resistance to antimicrobials of very high importance [45] was observed in only 2/23 isolates and from only conservation areas (Table 5). Resistance to more than one antimicrobial was observed in 56% (13/23) of resistant isolates. Resistance to three or more classes of antimicrobials were detected in four Salmonella isolates of three serovars: Heidelberg (PT 19) from a soil sample from a conservation area, Typhimurium (DT104) from a raccoon fecal and manure pit sample from a swine farm, and Schwarzengrund from a raccoon fecal sample from a swine farm. Although all isolates with intermediate susceptibility were considered to be resistant for the purpose of this study, intermediate susceptibility was found in 26% (6/23) of the isolates, and in all occasions it was found to FOX or CHL (Table 5).

Discussion
We predicted that raccoons and soil samples on swine farms would have greater exposure to environmental sources of Salmonella than raccoons living on conservation areas. Location type did not influence Salmonella occurrence in raccoon fecal or soil samples, which was in contrast to Pedersen et al. [46] who found higher Salmonella prevalence in Rock Pigeons (Columba livia) on dairy farms than urban areas and to Navarro-Gonzalez et al. [28] who found higher Salmonella prevalence in free-ranging wild boars (Sus scrofa) in cattle grazed areas than those in areas with no cattle exposure. Although Salmonella prevalence in manure pit samples was 30% and was consistent with what has been detected in fecal material from swine pens on swine farms (33%; 95% CI, 20-48; [25], the prevalence of Salmonella in soil samples from swine farms (6%; 95% CI, [4][5][6][7][8] was lower than what has been previously detected in soil samples from swine farm (23%; 95% CI, 12-37) environments from 18 different farms across five states in the United States [25]. Interestingly, the prevalence of Salmonella in soil samples on swine farms was more consistent with what was detected in soil samples from conservation areas in this study (5%; 95% CI, 4-7) and in sand on beaches used for human bathing in the United Kingdom (6%; 95% CI, 3-10; [47], which had no direct livestock exposure. Salmonella prevalence in raccoon fecal and soil samples on farms and conservation areas may not have differed in our study because not only did all of the study sites have some level of human use, but they all were located within the boundaries of the Grand River watershed in Ontario, which is heavily impacted by agriculture and humans [39]. Although manure spreading [48], human refuse [49], or pet dog feces [50] are possible sources of Salmonella exposure to raccoons, Salmonella occurrence in natural water has been well documented [21,[51][52], and transmission to terrestrial animals may occur through contact with or consumption of contaminated water [53]. In a major produce region of California, Salmonella was detected most frequently in water samples (7%; 95% CI, 4-11) compared to samples from soil/sediment (3%; 95% CI, 1-4), wildlife feces/fecal swabs (4.2%; 95% CI, 3-6), beef cattle feces (0.1%; 95% CI, 0-1), and pre-harvest produce (0/271) [54]. Because some of the same Salmonella serovars were detected in both wildlife and water samples, it was suggested that Salmonella might be transported between wildlife and water [54].
In the Grand River watershed, Salmonella was detected in 78% of water samples collected from agricultural/rural and urban tributaries [21]. This is quite high relative to other watersheds in Canada and was more comparable to Salmonella levels found in rivers in Georgia, U.S. A, which are heavily affected by agriculture [21]. The profile of Salmonella serovars detected from aquatic samples from the Grand River Watershed [21] appeared to be similar to what we detected in raccoon fecal samples in this study. Similarly, in a study of free-ranging urban and rural raccoons in Indiana, Morse et al. [55] found that the Salmonella serovars detected in raccoon feces and lymph nodes reflected the profile of Salmonella serovars found in water, fish, and mussel samples collected from the environment the raccoons were trapped two years prior. Because raccoons frequently use and feed in aquatic habitats [56] and water can act as a vehicle for Salmonella transport [57], raccoons may be exposed to Salmonella from contaminated rivers, streams, and tributaries.
Many of the most common serovars found in raccoon fecal samples in this study were also among the most common serovars found in soil samples. The similarity in Salmonella serovars found in raccoons feces and their local environment suggests that raccoons are exchanging Salmonella with their environment and that raccoons and/or soil may be used as sentinels for environmental Salmonella. For example, Salmonella Enteritidis, which was the second most common serovar detected in raccoons from Pennsylvania in 2011 [36], was detected infrequently in raccoon fecal samples and soil samples in this study as well as water samples from the Grand River watershed [21]. In addition, Salmonella Oranienburg is not historically a common serotype infecting human populations in Ontario; however, in 2010 it was listed as one of the top ten serovars affecting humans in Ontario for the first time in many years [58]. Likewise in this study, Oranienburg was detected commonly in raccoon and environmental samples in 2011 and in 2012, but was not detected at all in 2013.
Some Salmonella serovars, such as Salmonella Agona and Typhimurium var. DT 104 Copenhagen, may have been detected primarily in raccoon and environmental samples from swine farms due to contamination of the environment with livestock manure. FoodNet Canada found that among Salmonella in Ontario livestock in 2010, Agona and Typhimurium var. Copenhagen were among the most common serovars in swine [59]. Salmonella Typhimurium var. DT 104 Copenhagen phagetype DT 10 has also been detected in retail pork samples from Ontario [60]. Some Salmonella serovars detected in the manure pit samples in this study (e.g., Livingstone and Livingstone var. 14) were not commonly found in swine in Ontario [59], so it is possible that some isolates represented serovars better able to survive and persist in the manure pit environment. Also, three of the five farms had other domestic animal species present on the farm, and although the farms reported the waste of other animals was not present in the sampled manure pits, we cannot definitively rule that out. Although we detected some significant differences in the models for the most common serovars for raccoon fecal samples, caution should be taken when interpreting the results because the effective sample size was small and consequently we were unable to fit multivariable models.
We found a significant interaction between sex and season for the carriage of Salmonella in raccoon feces. In previous analyses, we also found significant interactions between sex and season for carriage of Salmonella on paws and in feces of raccoons [41] and prevalence and number of Baylisascaris procyonis in raccoon feces [35]. Females may have a lower prevalence of Salmonella and Baylisascaris than males earlier in the season due to behavioral differences associated with females provisioning young and maintaining smaller home-ranges from May to July [61][62][63]. If female raccoons are not moving around as often or as far as males during provisioning of their young, this might result in them having less exposure to sources of Salmonella and other pathogens in their environment. It also has been suggested that behavior may play a role in higher prevalence and transmission of Salmonella observed between men and women [64] and male and female wild boars [65].
Both male and female raccoons were more likely to carry Salmonella in their feces from August to November than May to July. These differences may be related to seasonal changes in diet [66] similar to the effect of feed composition and structure on survival of Salmonella in the gastrointestinal tracts of swine [67]. The higher prevalence of Salmonella detected in raccoons from August to November may have been related to the increased prevalence of Salmonella found in soil samples during the same time period; however, the intensity of transmission between raccoons and the environment is difficult to determine because the soil could also be contaminated from raccoon or other wildlife feces. Interestingly, Clostridium difficile was more likely to occur in raccoon fecal samples from May to July than August to November in the same population of raccoons examined in this study [68], suggesting the epidemiology and microbiology of carriage of these organisms in raccoon feces may be quite different.
Although Salmonella occurrence in raccoon fecal samples was negatively associated with rainfall during both seasons, the relationship was more pronounced from August to November than May to July. Higher rainfall was also associated with lower carriage of Salmonella in mongoose feces [29], and a negative correlation was found between rainfall the previous 30 days prior to sample collection and Salmonella excreted by quokkas (Setonix brachyurus), an Australian marsupial [69]. Dietary shifts associated with rainfall have been hypothesized to alter the intestinal microbiota and influence fecal shedding of Salmonella in free-ranging kangaroos [43]. It has also been suggested that rainfall might help disseminate Salmonella throughout the environment [29]. No associations were detected between Salmonella occurrence in soil and rainfall, which suggests raccoons, might be exposed to Salmonella through environmental sources other than the soil. Contaminated water is one possible exposure of Salmonella to raccoons because during drier periods, water availability is more limited, and animals might aggregate at common water sources where pathogens might be concentrated [29].
Salmonella prevalence was higher in soil and manure pit samples later than earlier in the season, which is similar to the findings of other studies [25,70]. Although Salmonella can contaminate soil through animal waste, human wastewater, or contaminated water, Salmonella prevalence and survival in the soil depends on a variety of factors including temperature, moisture, soil type, presence of plants, exposure to sunlight (ultraviolet), protozoan predation, and the initial number of organisms present [71]. A combination of these factors might explain the seasonal and yearly variation detected in Salmonella prevalence in soil samples in this study. In general, higher temperatures increase mortality of Salmonella in soil [72] and decrease survival of Salmonella in stored swine manure [73]. There is also a risk of under-estimating the prevalence of Salmonella in environmental samples when using culture-based methods because bacterial cells can enter a viable, but non-culturable state [74].
The ability of Salmonella to survive in the environment may play an important role in its transmission between host animals, dissemination, and persistence in animal and human populations [21]. In Michigan, the feces of raccoons were found to contribute to as much as 60% of fecal material in urban storm water sewers that drain into streams and rivers [75]. Once raccoons are colonized and shedding the bacteria, they have the potential to disseminate Salmonella throughout the environment. Although the duration that raccoons shed the same Salmonella serovar in their feces has been reported to be up to 30 days, raccoons have been found to carry different Salmonella serovars in their feces each month they were sampled [35]. This further suggests that raccoons are not maintaining long-term colonization of these bacteria without re-exposure from the environment [35].
We detected many Salmonella serovars in raccoon feces that were among the most commonly reported in humans from Ontario in 2011-2012, including Typhimurium, Heidelberg, Newport, I 4, [5],12:b:-, Thompson, and Infantis [58, 60]. Raccoons can shed Salmonella intermittently [35,55] and have been found to carry multiple Salmonella [55,[76][77], so all of the Salmonella serovars in raccoon feces may not have been detected. Salmonella prevalence in raccoon feces in this study was 26% (95% CI, [23][24][25][26][27][28], which is consistent with what has previously been found in rural raccoons living in southwestern Ontario (27%; 95% CI, 12-46; [35]. Although the sensitivity of Salmonella testing could be increased by testing whole fecal specimens rather than fecal swabs and by sampling over multiple consecutive days [78], those methods were not used in this study because we were unable to obtain whole fecal samples from the majority of the raccoons captured and raccoons were not trapped and resampled consecutively over multiple days. Antimicrobial resistance was rarely detected in Salmonella isolates from raccoon feces in this study, and the prevalence was similar to what has been previously reported from raccoons from Ontario and Pennsylvania (4%; [35]; 2%; [36]), but lower than what has been reported from rural and urban raccoons from Indiana (16%; 7/43; 95% CI 6.8-30.7; [55]. The observed prevalence was much lower than that reported for swine (62%; 38/61; 95% CI 49.0-74.4) from on-farm surveillance, chicken (47%; 180/382; 95% CI 42.0-52.3) and pork samples from retail meat surveillance (69%, 25/36; 95% CI 51.9-83.7), and humans (26%; 950/3601; 95% CI 24.9-27.9) in Ontario [60]. The source of antimicrobial resistance on swine farms that did not report antimicrobial use and in conservation areas is unclear. A meta-analysis comparing the prevalence of antimicrobial resistance in organic and conventional poultry, swine, and beef concluded that bacterial isolates from conventional production systems exhibited more antimicrobial resistance than isolates from organic production, but that some resistant isolates were still detected from samples from organic farms [79]. Agricultural and surface run-off may play a role in the dissemination of Salmonella [21] and resistant bacteria [80], but it is also possible that raccoons and other wildlife such as gulls and waterfowl may acquire Salmonella and antimicrobial resistant bacteria from contaminated areas and transport it to more natural areas. Other potential sources of Salmonella and antimicrobial resistant bacteria for raccoons include contaminated pet [81,82] or human food [83] and/or waste [84].
Fluoroquinolones and cephalosporins are the antimicrobials likely to be used in clinical treatment of human salmonellosis [85]; only 2 Salmonella isolates were resistant to cephalosporins and none were resistant to fluoroquinolones. Although raccoons have been found to carry many Salmonella serovars of public health significance and have the potential to transport them to humans, their role in disseminating resistant Salmonella is less clear.

Conclusion
Our results indicate raccoons living on farms and conservation areas in southwestern Ontario may be important hosts of Salmonella. Occurrence of Salmonella in raccoons and soil was affected by year and climatic variables, but demographic factors were also important predictors for Salmonella occurrence in raccoon feces. Although location type did not affect the frequency of Salmonella occurrence in either sample type, indicating Salmonella in the environment is widespread, more sites would need to be sampled in order to determine if our results hold true across the entire Grand River Watershed. Different serovars and antimicrobial resistance phenotypes dominated on farms versus conservation sites, suggesting raccoons locally acquire Salmonella and that they may be a useful sentinel species for Salmonella and associated antimicrobial resistance in the environment. Because many of the most common Salmonella serovars causing illness in humans and impacting livestock health were also found in raccoon fecal samples, raccoons may have the potential to transport Salmonella to humans, livestock, and the environment regardless of the original source of infection.
Supporting Information S1 Appendix. Multivariable models. (DOCX) S1 Table. Results from univariable multi-level and exact a logistic regression models showing associations between the occurrence of Salmonella in raccoon fecal, soil, and manure pit samples with respect to raccoon age and sex for raccoon samples, location type, year, season, and climatic variables in Ontario, Canada. (DOCX) S2 Table. Results from univariable multi-level logistic regression models showing associations between the most common Salmonella serovars compared to all others found in raccoon samples with respect to raccoon age and sex, location type, year, and season in Ontario, Canada. (DOCX)