Prevalence and Characterization of Escherichia coli and Salmonella Strains Isolated from Stray Dog and Coyote Feces in a Major Leafy Greens Production Region at the United States-Mexico Border

In 2010, Romaine lettuce grown in southern Arizona was implicated in a multi-state outbreak of Escherichia coli O145:H28 infections. This was the first known Shiga toxin-producing E. coli (STEC) outbreak traced to the southwest desert leafy green vegetable production region along the United States-Mexico border. Limited information exists on sources of STEC and other enteric zoonotic pathogens in domestic and wild animals in this region. According to local vegetable growers, unleashed or stray domestic dogs and free-roaming coyotes are a significant problem due to intrusions into their crop fields. During the 2010–2011 leafy greens growing season, we conducted a prevalence survey of STEC and Salmonella presence in stray dog and coyote feces. Fresh fecal samples from impounded dogs and coyotes from lands near produce fields were collected and cultured using extended enrichment and serogroup-specific immunomagnetic separation (IMS) followed by serotyping, pulsed-field gel electrophoresis (PFGE), and antimicrobial susceptibility testing. A total of 461 fecal samples were analyzed including 358 domestic dog and 103 coyote fecals. STEC was not detected, but atypical enteropathogenic E. coli (aEPEC) strains comprising 14 different serotypes were isolated from 13 (3.6%) dog and 5 (4.9%) coyote samples. Salmonella was cultured from 33 (9.2%) dog and 33 (32%) coyote samples comprising 29 serovars with 58% from dogs belonging to Senftenberg or Typhimurium. PFGE analysis revealed 17 aEPEC and 27 Salmonella distinct pulsotypes. Four (22.2%) of 18 aEPEC and 4 (6.1%) of 66 Salmonella isolates were resistant to two or more antibiotic classes. Our findings suggest that stray dogs and coyotes in the desert southwest may not be significant sources of STEC, but are potential reservoirs of other pathogenic E. coli and Salmonella. These results underscore the importance of good agriculture practices relating to mitigation of microbial risks from animal fecal deposits in the produce production area.


Introduction
Foodborne disease illnesses caused by pathogen contamination of fresh produce are being recognized in greater numbers in the United States (U.S.) and abroad [1], [2]. An analysis of Centers for Disease Control and Prevention (CDC) data on reported foodborne illnesses from 1973 to 1997 indicated that outbreaks associated with fresh produce accounted for 6% of all reported foodborne disease outbreaks in the 1990s compared with just 0.7% in the 1970s [3], [4]. A more recent survey of CDC outbreak data from 1998-2008 showed these numbers are still rising, with 46% of foodborne illnesses being attributed to produce and 22% specifically attributed to leafy greens [5]. While norovirus infections transmitted downstream during post-harvest handling are likely the major driver of these statistics, reports of fresh produce-associated outbreaks from zoonotic agents potentially spread by domestic and wild animal reservoirs in the pre-harvest environment are clearly contributing to this disease burden [6], [7].
Approximately 90% of commercial lettuce produced for the U.S. market is grown in two major produce production regions that rotate seasonally [8]: the Salinas Valley in the central California coast (April through October) and the desert southwest at the U.S.-Mexico border (November through March). The desert southwest growing region includes Yuma, Arizona, California's Imperial Valley, and northern Mexico. The role of domestic animals and wildlife as potential sources and transmitters of zoonotic bacterial pathogens to lettuce and other leafy greens and agriculture water has been studied at length in the central California coast [9], [10], [11], [12], [13]. There is limited information, however, on the importance of animal reservoirs in the pre-harvest bacterial contamination of fresh produce in other parts of the country [14]. The desert presents unique pre-harvest food safety challenges including urban encroachment where produce fields and irrigation canals may be adjacent to housing developments and recreation vehicle (RV) parks. In addition to concerns about human sources of foodborne pathogens near leafy green production areas of the desert, growers report problems with unleashed, free-roaming domestic dogs (Canis familiaris) entering their fields. Off-leash or stray dog intrusions into produce fields and the surrounding production area may result in damage to crops and destruction of potentially contaminated plants (Figure 1). Growers also report frequent coyote (Canis latrans) sightings and signs (tracks, scat, feces) on roads adjacent to produce fields where tractors and other equipment are used.
In spring 2010, an outbreak of Escherichia coli O145:H28 infections involving 27 confirmed and 4 probable case-patients from 5 states was linked to Romaine lettuce grown in southern Arizona [15], [16]. This was the first known leafy green-related Shiga toxin-producing E. coli (STEC) outbreak traced to the desert growing region. Based on investigations by the U.S. Food and Drug Administration (FDA), pre-harvest contamination of irrigation canals possibly due to sewage runoff from a nearby RV park could have caused the contamination, although no laboratory-confirmed environmental source of the outbreak was determined [17]. Following this outbreak, the present study was funded as a Center for Produce Safety ''Rapid Response Project'' with the purpose to determine prevalence and characterize pathogenic Escherichia coli and Salmonella strains isolated from dog and coyote fecal samples collected in the southwest desert during the 2010 to 2011 leafy green vegetable growing season.

Ethics Statement
Permission to access privately owned lands was obtained from the produce companies enrolled in the study. Animal shelter administrative directors in the U.S. and Mexico approved participation in the study. Dog fecal samples from the shelter in Mexico were transported by vehicle across the Mexico-US border by one of our industry collaborators. A permit for importation of dog feces was not required per the United States Department of Agriculture (USDA) Animal and Plant Health Inspection Service (APHIS) ''Animal and Animal Product'' import guidelines (#1102 Feline and Canine Material). Wildlife scientific collection permits and university animal care and use approval were not necessary in this study because fecal samples were collected from the ground and no animals were handled.

Sampling
Three animal shelters were enrolled in the study, one each in Yuma, Arizona, Imperial Valley, California, and northern Mexico. These facilities were chosen because animal control officers had worked historically with leafy green growers in the region to remove stray dogs from agriculture fields. We aimed to sample once monthly during the desert southwest leafy greens growing season (November to March) with a goal to collect ,300 samples based on sample size calculations. Due to limits in the number of impounded dogs available each month and logistics with the shelter personnel, each facility was sampled six times spread variably from November 3, 2010 to May 5, 2011 (Table 1).
A standardized questionnaire was used to collect demographic data (location and date found, breed, sex, age, and reason the dog was impounded) from records at the facilities. Dog fecal samples (n = 358) were collected by industry cooperators after training by University of California, Davis (UC Davis) veterinarians in aseptic fecal sample collection, storage and shipping. Freshly deposited feces from animals caged individually at the U.S. shelters were taken from the kennel floor using a sterile tongue depressor and placed in a sterile 227 gram fecal cup with a snap-cap lid (National Scientific Supply Co., Claremont, CA). Dogs at the shelter in Mexico were caged in groups, thus there was potential for crosscontamination. To minimize the risk of cross-contamination, care was taken to collect only freshly deposited feces from individual dogs impounded in the past 24 hours.
Fresh coyote fecal material (n = 103) found on the roads in and up to 1 mile from leafy green vegetable fields located in Yuma, Arizona and Imperial Valley, California were collected by industry cooperators using a sterile tongue depressor and fecal cup as described above. Industry personnel routinely survey fields for animal intrusions and have training in the identification of sign (sightings, tracks, feces) of wildlife species including coyotes common in agriculture areas. In order to ensure the fecal material was fresh, the sites were walked by grower personnel the evening prior to sampling, and any existing coyote feces was removed. The fields were then surveyed again at dawn the following morning to collect fresh feces. Wildlife trail cameras (Cuddeback Digital, Green Bay, WI), physical sightings, and other sign were used to confirm the presence of coyotes at the study areas.
Samples were shipped to UC Davis overnight on blue-ice on the day of collection and processed on the day of arrival at the laboratory (approximately 24 hours).

Laboratory
Our overall goal was to culture Shiga toxin-producing E. coli belonging to serogroups STEC O103, O145, O157, O26, and non-typhoidal Salmonella enterica using a combined pre-enrichment step followed by immunomagnetic separation (IMS), selective plating, latex agglutination, and PCR or biochemical confirmation as described below. Isolates were then characterized by presence of virulence factors, genetic relatedness, and antibiotic resistance.  Pre-enrichment. Initially, non-selective pre-enrichment for the simultaneous culture of STEC and Salmonella was performed by adding 10 grams of feces to 100 mL of universal preenrichment broth (UPB; Difco, Becton Dickinson, Sparks, MD) and incubating for 20 hours at 35uC using a protocol modified in our laboratory for dog fecal material [18]. One milliliter of enriched UPB was then transferred to 9 mL tryptic soy broth (TSB; Becton Dickinson, Sparks, MD) for STEC detection and incubated for 2 hours at 25uC with shaking at 100 rpm (Innova 44, Eppendorf North America, Hauppauge, NY), followed by 8 hours at 42uC with shaking at 100 rpm, then at 6uC without shaking until processing the following day [10], [19]. One milliliter of enriched UPB was also transferred to 9 mL of buffered peptone water (BPW; Hardy Diagnostics, Santa Maria, CA) for Salmonella detection and incubated for 24 hours at 37uC with shaking at 50 rpm [11]. In 2011, we discontinued the use of the UPB preenrichment step to streamline the protocol. Instead, pre-enrichment was performed by adding 10 grams of feces directly into a WhirlPak bag containing 100 mL TSB followed by the same incubation parameters as just described [10], [19]. Spiking experiments comparing the UPB and TSB pre-enrichment methods revealed no statistical difference (p = 0.32) in recovery of STEC or Salmonella (data not shown). As such, we completed this study using the streamlined protocol without the UPB step.
Aliquots of the primary enrichment broths were mixed with sterile glycerol to a final concentration of 14.3%, and stored at 220uC [11].
Escherichia coli. IMS using Dynal anti-E. coli O157, O26, O103, and O145 beads (Invitrogen, Grand Island, NY) was performed on TSB enrichment broths with the automated Dynal BeadRetriever (Invitrogen) per the manufacturer's instructions. Following incubation and washing, 50 mL of the resuspended beads were plated onto Rainbow agar (Biolog, Hayward, CA) with novobiocin (20 mg/L) and tellurite (0.8 mg/L) (MP Biomedicals, Solon, OH) and streaked for isolation [10], [19]. Another 50 mL were plated onto Sorbitol MacConkey Agar (BD Becton, Sparks, MD) with cefixime (0.05 mg/L) (USP, Rockville, MD) and tellurite (2.5 mg/L), streaked for isolation, and incubated at 37uC overnight. E. coli O157:H7 RM1484 and three non-O157 STEC strains (O103, O145, O26) were used as positive controls to observe the expected phenotype on the selective agars. Up to 10 colonies exhibiting characteristic morphology were subcultured to both agar types and incubated again at 37uC overnight for purification. Up to four pure colonies were then streaked onto Luria-Bertani (LB; Fisher Scientific, Waltham, MA) agar and incubated at 37uC for 24 hours. Suspect colonies from LB agar were screened using ImmuLex commercial latex slide agglutination assays (Statens Serum Institut, Denmark) with pooled STEC antisera (E. coli OK O antiserum to detect EPEC and STEC) followed by O-group specific antisera (O103, O145, O157, O26) according to the manufacturer's instructions. All assays included a negative control (saline) to check for non-specific agglutination. Two bacterial colonies from each positive plate were banked onto Cryobeads (ProLab Diagnostics, Round Rock, TX) and stored at 280uC until characterized further.
All isolates (n = 278) positive for STEC O-groups by latex agglutination were submitted to the Pennsylvania State University E. coli Reference Center to confirm O-type using a multiplex PCR that detects eight major STEC O-groups (O26, O45, O103, O111, O113, O121, O145, O157) as described previously [20]. H-antigens were identified by the same lab using PCR-RFLP [21]. The isolates were also tested by PCR at the reference laboratory for the presence of virulence factors including stx1, stx2, eaeA, and hlyA genes [22]. A subset of aEPEC isolates (n = 12) positive for the eaeA gene, but not belonging to the 8 STEC O-groups identified by the reference laboratory's multiplex PCR, were serotyped conventionally by agglutination reactions against antisera developed for each of the O serogroups [20].
Because all isolates were Shiga toxin-negative despite many belonging to STEC-associated O-groups, a retrospective analysis of banked TSB enrichment broths (n = 461) frozen at -20uC with glycerol was performed at the United States Department of Agriculture's Agricultural Research Service (ARS) Western Regional Research Center laboratory. Briefly, template DNA was prepared by boiling followed by detection of stx1 and stx2 virulence genes by multiplex qPCR; the method was validated previously for screening TSB pre-enrichment broths from environmental samples including coyote feces [19].
Salmonella. IMS using anti-Salmonella Dynabeads (Invitrogen, Grand Island, NY) was performed on BPW broths as described previously using the Dynal BeadRetriever (Invitrogen) [11]. Following incubation and washing, 100 mL of separated broth was further enriched in 3 mL Rappaport-Vassiliadis Soya Peptone (RVS; Difco, Becton Dickinson, Franklin Lakes, NJ) broth for 48 hours at 42uC [23]. Ten microliters of RVS broth were then plated on xylose lysine deoxycholate (XLD; Difco, Becton Dickinson, Franklin Lakes, NJ) agar and incubated at 37uC overnight. Samples with growth of hydrogen sulfide positive colonies were confirmed for Salmonella by performing biochemical profiles (triple sugar iron, urea, citrate, and lysine decarboxylase) on up to six individual colonies from each XLD agar. Salmonella Enteritidis ATCC BAA1045 was used as a positive control to observe the expected phenotype. The same colonies used for biochemical profiling were also streaked onto LB agar and incubated at 37uC overnight. Two bacterial colonies from each positive plate were banked onto Cryobeads (ProLab Diagnostics, Round Rock, TX) and stored at 280uC. Serotyping using the Kauffmann-White scheme was conducted by the United States Department of Agriculture (USDA) National Veterinary Services Laboratories in Ames, Iowa [24].
Pulse field Gel Electrophoresis. E. coli and Salmonella isolates were retrieved from frozen storage at 280uC and clonal relationships were assessed by pulsed-field gel electrophoresis (PFGE) according to the CDC's PulseNet standard procedure using Salmonella Braenderup ATCC BAA664 as the molecular size standard [25]. Briefly, bacterial isolates were suspended in buffer containing 10 mM Tris pH 8 and 10 mM EDTA for DNA isolation. DNA was digested in enzyme buffer with restriction enzyme XbaI. Images were analyzed, and the similarity among different strains was characterized using Bionumerics version 7.1 software (Applied Maths, Austin, TX). Pattern comparisons were made using the software cluster analysis tool and confirmed by visual examination to assign pulsotypes [26], [27].
Antimicrobial Susceptibility testing. Frozen E. coli and Salmonella isolates were thawed and streaked onto trypticase soy agar (TSA; Difco, Becton Dickinson, Franklin Lakes, NJ) with 5% sheep blood agar, then incubated at 37uC for 24 hours. The broth microdilution method for antimicrobial susceptibility testing was performed in accordance with the Clinical and Laboratory Standards Institute (CLSI) guidelines [28], [29]. Isolates were evaluated for susceptibility to 12 antimicrobial drugs (ampicillin, amoxicillin/clavulanic acid, ceftriaxonem azithromycin, chloramphenicol, sulfisoxazole, cefoxitin, kanamycin, streptomycin, trimethoprim/sulfamethoxazole, tetracycline, ceftiofur) using the National Antimicrobial Resistance Monitoring System (NARMS) Gram negative tray (Trek Diagnostic Systems, Westlake, OH). E. coli ATCC 25922, E. coli ATCC 35218, and Pseudomonas aeruginosa ATCC 27853 were used as quality control organisms for MIC determination in accordance with CLSI guidelines. Breakpoints guidelines were adopted from the NARMS report, with the exception of azithromycin, for which no breakpoints have been published [30]. Based on a recent publication proposing an epidemiologic cut-off for wild-type Salmonella of #16 mg/ml, isolates with MIC values .16 mg/ml were considered resistant to azithromycin for the purpose of this study [31].

Statistical Analysis
WinEpi online software (http://www.winepi.net/uk/index.htm) was used to calculate sample size with a confidence level set at 95% and a population size of 1,000-10,000. Prevalence estimates were based on data from a longitudinal study of coyote populations in the central California coast in 2008-2010 [11], [19], with 1% for E. coli O157, 5% for non-O157 STEC, and 10% for Salmonella enterica; using these estimates, the required number of samples would be 258-294 for E. coli O157, 57-59 for non-O157 STEC, and 29 for Salmonella detection. Based on the expected low prevalence of E. coli O157, our goal was to collect at least 300 fecal samples.
Data were entered in Microsoft Excel 2007 spreadsheets and exported for analysis in STATA (Stata 11.0, College Station, TX).
McNemar's chi-square test was used to compare the sensitivities of pre-enrichment methods (UPB and TSB) and O-group testing methods (latex agglutination, multiplex PCR). Univariate logistic regression was conducted to identify statistical associations between enteric pathogen status and covariates (age, sex, etc.). Covariates with p-values equal or less than 0.20 were considered for inclusion in an exact logistic regression model. A p-value # 0.05 was used to detect significantly associated factors for pathogen presence in the regression model.  (Table 1). Descriptive characteristics of the shelter dog population are shown in Table 2. Shiga toxin-producing E. coli was not detected in any fecal samples, but aEPEC (eaeA+) strains were isolated from 13 (3.6%) dog and 5 (4.9%) coyote fecal samples. Salmonella was detected in 33 (9.2%) dog and 33 (32.0%) coyote fecal samples.

Prevalence and risk factors
Univariate analysis revealed no significant association between the relative number of aEPEC or Salmonella positive and negative dogs by age, gender, breed, or reason that the dog was impounded. There was a significant difference in the number of dog fecal samples positive for Salmonella by shelter location. Exact logistic regression revealed that, after adjusting for age, the odds of a dog from the shelter in Mexico being Salmonella positive was 4.88 (95% CI: 1.60, 20.31, P ,0.01) times higher than for dogs at the Arizona shelter. In November 2010, a higher seasonal prevalence of Salmonella was observed at the California and Mexico shelters, primarily due to a specific serovar (Senftenberg) as described below.
For coyote fecal samples, univariate logistic regression showed no significant difference in the odds of aEPEC or Salmonella isolation in fecal samples collected in Arizona compared with California (OR = 1.10, 95% CI: 0.43, 2.78, P = 0.99). There was a significantly lower odds of Salmonella isolation from coyote scat sampled in February (OR = 0.16, 95% CI: 0.04, 0.70, P = 0.02) and March (OR = 0.1, 95% CI: 0.02, 0.80, P = 0.03) compared with November; but, unlike dogs, this seasonal observation was not due to any specific serovar.

Phenotypic and molecular characterization of E. coli isolates
We screened initially for suspect STEC isolates by using Ospecific (O103, O145, O157, O26) IMS and commercial latex agglutination. A total of 278 isolates presumptively belonging to these four serogroups were submitted to the Pennsylvania State University E. coli reference laboratory for confirmation using their multiplex PCR that detects 8 major STEC O-groups [20]. We found discordance between the IMS-latex agglutination classification of O-groups compared with the multiplex PCR (Table 3). Specifically, if the multiplex PCR at the reference laboratory is used as the standard, the sensitivities of O-specific latex agglutination (31.2% for O103, 8.5% for O145, and 26.3% for O26) were significantly different from the multiplex PCR method (p ,0.01).
All 278 isolates were negative by PCR for genes encoding stx1 and stx2. Additionally, 461 frozen fecal TSB-enrichment broths were negative using a multiplex qPCR assay to detect stx1 and stx2 genes [19]. Isolates (n = 187) lacking any virulence factors and not belonging to one of the eight serogroups identified by the reference laboratory's multiplex PCR were not further characterized. Among the remaining 91 isolates, a total of 29 different E. coli serotypes were identified (Table 4). Excluding clones from the same samples, there were 18 isolates comprising 14 serotypes with genes encoding eaeA, including one (O26:H11) with both eaeA and hlyA genes; these isolates were classified as aEPEC (Table 4) [32], [33]. There was more diversity among serotypes from coyotes compared with dogs. Specifically, 11 of 15 coyote fecal isolates were different serotypes with none being dominant. In contrast, almost half of the dog samples contained two dominant serotypes that were negative for virulence markers, O103:H16 and O103:H49 (Table 4).
A total of 17 pulsotypes (PT) were found among 18 aEPEC strains from dogs and coyote feces (Figure 2). Two non-pathogenic E. coli O145:H11 isolates were included in the dendogram for comparison with the E. coli O145:H28 human clinical strain (PT-8) associated with the 2010 outbreak linked to Romaine lettuce. Of note, E. coli O145 strains isolated from dog feces during the study had different H types (O145:H11 and O145:H34) and were genetically unrelated to the 2010 outbreak strain based on PFGE analysis (Table 4, Figure 2).
All aEPEC isolates were tested for antibiotic resistance, and 6 (33.3%) were found to be pansusceptible to the antimicrobial drugs we used in the NARMS panel (Table 5). One isolate from coyote and 3 isolates from dog feces were resistant to two or more antibiotics. Two aEPEC isolates, serotypes O167:H9 and O114:H8, from dog feces collected in Arizona and Mexico, respectively, were resistant to four antibiotics including ampicillin, ceftriaxone, chloramphenicol, tetracycline (O167:H9), and chloramphenicol, sulfisoxazole, streptomycin, and trimethoprim/sulfamethoxazole (O114:H8).

Phenotypic and molecular characterization of Salmonella isolates
Overall, 29 different Salmonella enterica serovars were identified with 46 (70%) of 66 isolates belonging to subspecies Group I ( Table 6). Two dominant serovars, Senftenberg and Typhimurium comprised 58% of the isolates from dog samples. In contrast, no predominant Salmonella serovars were identified among strains isolated from coyotes. A significant association was observed between serovar Senftenberg and the date of sample collection (P = 0.03), with 10 of the 14 S. Senftenberg isolations occurring in the month of November, primarily from the shelter in Mexico. Figure 3, PFGE analysis of Salmonella isolates revealed 27 distinct pulsotypes. S. Senftenberg isolates (n = 8) from dog feces collected on two dates at the shelter in Mexico belonged to four different but closely related pulsotypes (PT-9, 10, 11, 12). In contrast, S. Senftenberg isolates (n = 6) from California shelter dogs collected on three sampling dates belonged to a single pulsotype (PT-13). S. Typhimurium PT-18 and PT-21 were the only shared pulsotypes among samples from different locations and species including 2 California coyote and an Arizona dog isolate (PT-18), and two dogs from Mexico and a California coyote (PT-21).

As shown in
Antibiotic resistance testing revealed that 58 (87.9%) of the 66 Salmonella isolates, evenly distributed between dogs and coyotes, were susceptible to the antibiotics tested (Table 5). Of the 8 Salmonella isolates with resistance to at least one antibiotic, 4 were resistant to 2 or more drugs. An S. Newport isolate from coyote feces collected in California showed the most antibiotic resistant phenotype in this study including resistance to ampicillin, amoxicillin/clavulanic acid, cefriaxone, chloramphenicol, and trimethoprim/sulfamethoxazole.

Discussion
In this study, we show that stray dogs and free-roaming coyotes in the southwest desert leafy greens production region at the U.S.-Mexico border do not appear to be significant reservoirs of E. coli O157 and other STEC, but aEPEC and Salmonella were prevalent in fecal samples using the methods described herein.

E. coli detection and characterization
Previous studies of STEC occurrence in domestic dogs in the U.S. have focused on detection of Shiga toxin genes among animals with and without gastroenteritis. In one survey, a higher prevalence of stx1 (3% and 15%) and stx2 (36% and 23%) was found in diarrheic and non-diarrheic greyhounds, respectively [34]. In another study, there was no occurrence of stx1 or stx2 in 52 healthy Midwestern research colony dogs [35]. For logistical reasons, we were not able to sample unhealthy dogs with diarrhea in isolation wards at the U.S. shelters, and health status information was not available at the shelter in Mexico; thus, we collected only fresh, normally formed fecal samples from dogs. Outside the U.S., a survey in Japan revealed an extremely low prevalence of E. coli O157:H7 in dogs and cats, with only 1 of 614 (0.2%) fecal samples testing positive [36]. If we over-estimated E. coli O157:H7 prevalence at 1% in our sample size calculations, it is possible we would have needed to collect more than 300 fecal samples to detect E. coli O157:H7 in the shelter dog population. In another Japanese survey, a positive association between the presence of dogs or cats on beef cattle farms and prevalence of O157 in cattle was found [37]. This association, however, could be attributed to the hygiene practices of the farm (e.g., farms that allow dogs to run loose may have poorer hygiene and biosecurity practices than farms that do not), rather than evidence of colonization. In Argentina, stx1 and stx2 were detected in 3.7% and 4.2% of dog samples, respectively, and STEC was culture confirmed in 4% of the samples [38]. While our results suggest that STEC is rare in southwest dog populations, there are caveats to consider when comparing our results with prevalence surveys in other locations. For example, in Argentina, dog rectal swabs were taken and screened for Shiga toxin genes, which may have lead to improved detection compared with our pen floor samples. Data on STEC occurrence in coyotes is even sparser in the literature. E. coli O157:H7 prevalence surveys in the Midwest and Washington State found the pathogen in 0/100 and 0/7 coyote fecal samples, respectively [39], [40]. We tested a comparable number of coyote samples (n = 103), which may have been too few to detect E. coli O157:H7. In contrast, during a longitudinal study of various domestic and wild animals in the central California coast, E. coli O157:H7 was isolated from 2 of 145 (1.4%) colonicfecal samples from hunted coyotes [19]. Interestingly, 2 (1.9%) of 103 coyote fecal samples in our study contained Shiga toxinnegative E. coli O157:H+ isolates positive for eaeA. Bentancor and colleagues (2010) have characterized non-Shiga toxin producing E. coli O157 strains from dogs in Argentina and speculated that strains with the eae gene may represent a potential human health threat [41].
We also endeavored to determine if the E. coli O145:H28 human clinical outbreak strain linked to Romaine lettuce grown in southern Arizona in spring 2010 was present in our fecal samples, but found that all strains in the E. coli O145 serogroup isolated during this study were phenotypically and genetically unrelated to the 2010 outbreak strain (Figure 2). The public health significance of Shiga toxin-negative E.coli strains belonging to serogroup O145, as well as the other Shiga toxin-negative isolates belonging to ''top 6'' STEC-associated serogroups found in our study, is still somewhat unclear. Ultimately, additional molecular studies are needed to better understand the human virulence potential of these strains. The most clinically significant strain we isolated may be Shiga toxin-negative E. coli O26:H11 eaeA+/hlyA+ cultured from dog feces at the Mexican shelter in February (Table 4). Loss and gain of stx genes by STEC O26 have been described, and aEPEC O26 may be ancestral to STEC O26 [46], [47]. We believe it is unlikely this strain or others lost Shiga toxin genes during passage through enrichment since stx1/stx2 was not detected in TSB pre-enrichment broths by qPCR, provided the concentration of bacteria was within our level of detection. However, aEPEC strains could potentially re-acquire Shiga toxin genes and become STEC during passage through human or animal hosts [47].
We encountered several methodological challenges during this study related to culture and identification of STEC serogroups. For example, non-specific binding of STEC O-antigens (O103, O145, O157, O26) to IMS beads and method of capture may have caused isolation of multiple O-groups not originally targeted (Table 3). We also found discordant results between O-group specific latex agglutination and confirmatory tests, which has been described previously [48]. The utility of using a commercial latex agglutination screen to identify presumptive STEC from environ- Table 3. mental samples such as feces needs further assessment and comparison with other methods including the serology gold standard. Nevertheless, we concluded that screening for Shiga toxin genes by qPCR would be a more efficient approach to detect STEC colonies in the future, rather than screening for STEC Ogroups.

Salmonella detection and characterization
Reported Salmonella prevalence in dogs varies greatly between studies, from 5% to over 70% [49], [50], [51], [52]. The predominance of two Salmonella serovars in the shelter dog population, but not in the sampled coyote population, is interesting. According to a USDA NVSL 2011 annual report, Typhimurium and Senftenberg, followed by Muenchen, Newport, and Javiana, were the most common serovars isolated nationally from dogs and cats during 2011 [53]. We speculate that the dominant serovars among shelter dog samples could indicate a common source of exposure to Salmonella in the environment. For example, contaminated dog food is increasingly recognized as a risk factor for Salmonella infections in dogs [54], [55], [56]. The findings could also be indicative of poor sanitation practices and overcrowding. Staffing at the facility in Mexico was noticeably limited and animals were typically kenneled in large groups of ten or more dogs, making the transmission of Salmonella between  individuals more likely to occur. We controlled for this variable by limiting fecal collections to fresh feces from individual animals impounded within the last 24 hours. Often, however, it was noted that most of the individual animals sampled on a given day from the Mexican shelter had been collected from the same location and were likely living in groups together. Thus, dogs shedding genetically related strains may have shared common exposures prior to or during impound. The fact that nearly one in three (32%) coyote fecal samples collected near leafy green fields were positive for Salmonella was surprising. In contrast, a survey from the central California coast found that the organism was recovered from only 3 (7.5%) of 40 coyote colonic fecal sample enrichment broths [11]. Variations in sampling and laboratory culture methods between the two studies could, in part, explain these differences. It is also possible that individual coyotes were re-sampled over the course of this study. However, given the wide geographic and temporal distribution of sampling locations in combination with the high diversity of serovars and PFGE subtypes (Figure 3), we do not believe that repeat sampling of individual coyotes contributed significantly to our overall prevalence. Even in the event of re-sampling, the apparent high prevalence of Salmonella in coyote fecal material found in or near the production area is of importance to growers given the potential to contaminate the plants directly, or indirectly via agriculture water sources and farm equipment, Additional studies using trapping or hunting techniques are needed to determine actual prevalence of foodborne pathogens in the southwest desert coyote population.
There was some seasonality observed in Salmonella recovery from both dog and coyote samples including a significantly higher prevalence in November compared with samples collected in late winter and spring months. More long-term studies, however, are indicated to reveal any true seasonality or temporal patterns of Salmonella occurrence. The relatively high prevalence of Salmonella shedding in dogs and coyotes may be due in part to the hunting/scavenging behaviors of canids in the region. In desert regions where prey and water resources are limited, both coyotes and stray dogs opportunistically forage on fresh and rotten nutritional sources, including garbage and other refuse, vegetable and fruit matter, and the meat of dead animals. Such scavenging behaviors may put dogs and coyotes at a higher risk of exposure to Salmonella. High prevalence of Salmonella has previously been found in other scavenging species of the southwest, such as turkey vultures (Cathartes aura) [57]. Additionally, these animals may be obtaining water from anthropogenic sources, such as irrigation ditches, sediment basins, and camp-sites in the absence of natural water sources. Water samples, from both static sources and flowing streams, rivers, and creeks, often yield high percentages of Salmonella positive samples [10], [58], [59].
We found more antibiotic resistance among aEPEC isolates compared with Salmonella isolates (Table 5). Four (22.2%) of 18 aEPEC and 4 (6.1%) of 66 Salmonella isolates were resistant to two or more antibiotic classes; two dog aEPEC isolates (O114:H8 and O167:H9), a dog S. Senftenberg, and a coyote S. Newport displayed resistance to 3 or more antibiotic classes. Of note, Newport and Senftenberg have been identified as emerging multidrug-resistant serovars worldwide [60], [61]. In a wildlife study conducted during the same time period in the central California coast, a majority of Salmonella enterica subspecies Group IIIa and IIIb isolates from wild-caught amphibians and reptiles captured near produce fields were resistant to at least one antibiotic [12]. It appears that antibiotic resistance among Salmonella isolates is less

Prevention and Control Recommendations
The produce industry currently addresses foodborne pathogen hazards from domestic and wild animal sources through adherence to best practices established by the Arizona and California Leafy Green Marketing Agreements [62], [63]. For example, fecal material in the production area is removed, and a minimal 5-foot radius no-harvest buffer zone ( Figure 1D) is used to prevent contaminated plants from entering the food supply. Repeated intrusions into produce fields by dogs and coyotes can be managed by use of fencing, control of strays, and depredation (coyotes). In our study region, animal control officers on both sides of the border work closely with the produce growers to assist with stray animal problems in fields. In the Imperial-Yuma region, loose dogs can be particularly problematic in the fall-winter season-which is also the leafy greens growing season-when the area becomes a popular tourist (''snowbird'') destination for recreational vehicle enthusiasts who often travel with their dogs. In Mexico, stray dog control is more challenging because of limited resources, large numbers of un-owned dogs, and cultural barriers to dog population control.
It is worth noting that although intrusions by stray dogs may represent a food safety risk for fresh produce, trained working dogs if used properly can actually be an asset and should not be discouraged. For example, dogs have been used to help disperse and deter large flocks of nuisance birds, and scent detection dogs have been used experimentally to identify in-field fecal contamination of raw produce [64].

Conclusions
In summary, results from this study will assist the produce industry by providing baseline information on the occurrence of zoonotic enteric pathogens found in fecal material from common domestic and wild canid populations in the desert southwest produce production region. The findings underscore the importance of good agriculture practices for leafy greens and other produce, especially those relating to animal intrusions and preand post-harvest environmental assessments. Follow-up surveys are warranted to determine pathogenic E. coli and Salmonella prevalence in other potential domestic and wildlife reservoirs in this region, and comparison with other possible environmental sources of microbial contamination (e.g., canals, irrigation water, and soil amendments).