A longitudinal field study was performed on three Virginia farms; each farm represented a different growing region (i.e., Blue Ridge Highlands, Piedmont, Coastal Plain; distance between farms ranged from 338 to 591 km). Within each farm, 20 unique 0.2 ha (5 m x 5 m) sites were selected based on the produce crops grown (e.g., crops that potentially or frequently contact the ground), the feasibility of soil sample collection, and diversity in management practices and topography (Supplementary Table S1). Global Positioning System (GPS) coordinates for each sample site center were recorded and used to ensure the same site was sampled at each visit. Each farm was sampled four times between October 2021 and November 2022 capturing a full growing season. In total, 240 composite soil samples were collected for this study (4 sampling visits, 3 farms, 20 sample sites).

An initial questionnaire, informed by prior field studies (Park et al., 2013; Strawn et al., 2013b) was modified and administered orally to farm management to collect information regarding historical land-use and field-level management practices (e.g., irrigation, worker/equipment presence) for each sample site, prior to the first sampling event on each farm (Supplementary Table S2). Additionally, an observational survey detailing characteristics and management practices was completed for each sampling site during each visit (Supplementary Table S3).

Sampling was performed following previously described protocols (Strawn et al., 2013a; Weller et al., 2015). Briefly, nitrile gloves were worn for sample collection and changed between each field sample site. Within each sample site, sub-samples from five locations (each of the four corners, and centroid) were collected using sterile scoops (Fisher Scientific, Hampton, NH, United States). Soil was collected from the rhizosphere layer (i.e., the zone surrounding the plant roots), up to approximately 15.2 cm below the surface, and deposited into sterile re-closable 3.79 L storage bags (Fisher Scientific, Hampton, NH). All samples were transported on ice and then stored at 4 ± 2°C for processing within 24 h.

For foodborne pathogen testing, two 25 g composite soil samples were prepared, one for each pathogen, by combining 5 g portions of each of the 5 sub-samples into a sterile filtered Whirl-Pak bag (Nasco, Madison, WI, United States). An additional 1 g from each sub-sample (5 g total) were pooled into a Whirl-Pak bag for enumeration of generic E. coli and total coliforms. Composite samples (150 g) were prepared by combining 30 g portions of each of the five sub-soil samples into a brown paper bag (Fisher Scientific, Hampton, NH, United States). Soil samples were held at 21 ± 2°C until visibly dry (approximately 2 weeks) to determine soil moisture content (Maguire et al., 2005). Air-dried soil samples were shipped and analyzed at Waters Agricultural Laboratories (Camilla, GA, United States) for macro- and micro-nutrients. Samples were analyzed under the Mehlich 3 extraction method (Mehlich, 1984). A list of soil nutrients and properties that were collected (and transformed) can be found in Table 1.

Listeria detection and isolation was performed using a modified version of the United States Food and Drug Administration’s (FDA) Bacteriological Analytical Manual (BAM) (Hitchins et al., 2022). Briefly, soil samples were diluted 1:10 by adding 225 mL of buffered Listeria enrichment broth (BLEB; Oxoid, Basingstoke, Hampshire) to each sample (25 g) and incubated at 30 ± 2°C for 4 h. After 4 h of incubation, Listeria selective enrichment supplement (Oxoid, Basingstoke, Hampshire) was added to each sample and incubated at 30 ± 2°C. At 24 and 48 h of incubation, 50 μL of each enrichment was streaked for isolation onto Modified Oxford Agar (MOX; Becton Dickinson, Franklin Lakes, NJ, United States) and L. monocytogenes Plating Medium (LMPM; R&F Products, Downers Grove, IL, United States) and incubated at 30 ± 2°C for 48 h and 35 ± 2°C for 48 h, respectively. Up to three presumptive Listeria colonies per time and plating agar combination (e.g., MOX at 24 h and LMPM at 48 h) were sub-streaked onto reversed agar combinations (e.g., MOX plated onto LMPM and LMPM plated onto MOX) and incubated for 48 h at each agar’s respective temperatures. Presumptive positive Listeria colonies were confirmed by a sigB PCR assay and species identified by partial sigB gene sequencing, as previously described (Nightingale et al., 2007).

Salmonella detection and isolation were performed using a modified version of the procedure outlined in the United States FDA BAM (Andrews et al., 2011). Briefly, soil samples were diluted 1:10 by adding 225 mL of Buffered Peptone Water (BPW) to each 25 g composited sample. Samples were incubated at 35 ± 2°C for 24 h. Following incubation, 1.0 and 0.1 mL were transferred to 9.0 and 9.9 mL of Tetrathionate (TT; Oxoid; Basingstoke, Hampshire) and Rappaport Vassiliadis (RV: Oxoid, Basingstoke, Hampshire) broth, respectively. Enrichment broths were incubated at 35 ± 2°C and 42 ± 2°C for 24 and 48 h, respectively. From each broth, 50 μL was placed onto Xylose Lysine Tergitol-4 (XLT-4; Neogen, Lansing, MI, United States) and Hektoen Enteric (HE; Neogen Lansing, MI, United States) agars and incubated at 35 ± 2°C for 24 h. Up to three presumptive Salmonella colonies per enrichment and plating agar combination (e.g., TT-XLT-4, RV-XLT-4) were sub-streaked onto reversed agar combinations (e.g., XLT-4 plated onto HE and HE plated onto XLT-4) and incubated at 35 ± 2°C for 24 h. Presumptive positive Salmonella colonies were confirmed by a PCR assay for the invA gene (Kim et al., 2007). One isolate per Salmonella-positive sample was serotyped by Kauffmann–White classification at the National Veterinary Laboratory Services (Ames, IA, United States).

Generic E. coli and total coliform enumeration were performed as previously described (Berry and Miller, 2005). Briefly, a 5 g composite sample was placed in a filtered Whirl-Pak bag diluted in 45 mL of 2% BPW. Following serial dilution, 1 mL of each dilution was plated onto E. coli/coliform Petrifilm (3 M, Saint Paul, MN, United States) in duplicate. Petrifilm was incubated at 37 ± 2°C for 48 h, and anlyzed according to manufacturers instructions to enumerate E. coli and total coliforms. Counts were log transformed (log₁₀ CFU/g).

Meteorological data including average daily air temperature (°C), total precipitation (mm), and relative humidity (%) for the 5 days prior to sampling were obtained for each farm and sampling visit. Meteorological data were collected from the nearest Virginia Agricultural Research and Extension Center to each farm using the Virginia Tech WeatherSTEM platform.¹

Analyzes were performed in RStudio version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria). Descriptive statistics were performed on all variables to characterize trends across farms and sampling visits, and summary statistics were calculated for soil nutrient data. Extreme outliers in soil nutrient data were identified using the Rosner Test (Rosner, 1975, 1983), and biologically implausible outliers were removed. Due to strong skew, some soil properties were log₁₀ transformed (Table 1).

Bayesian mixed models were implemented using the brms package (Bürkner, 2017a,b) to identify factors associated with: (i) the isolation of Listeria spp., L. monocytogenes, and Salmonella; (ii) if log₁₀ generic E. coli levels were above versus below the limit of detection; and (iii) E. coli concentration in samples with enumerable levels. For the logistic models, a Bernoulli distribution with a logit link function was used. For the linear models, an identity link was used. The model outcome was a function of a random effect of site nested in farm, a random effect of month, and a fixed effect of year in addition to a fixed effect for the factor of interest (Supplementary Table S4).

All models were fit using the brms package with uninformative priors, 5,001 iterations per chain, thinning set to 10, and 3 chains (Bürkner, 2017a,b). The Maximum a posteriori estimate (MAP) is the most probable value of the posterior distribution for the effect estimate for the linear models and odds ratio for the logistic models. The presence and strength of an association between the outcome and factor of interest were assessed using the probability of direction (PD), and region of practical equivalence (ROPE) values (Makowski et al., 2019a,b). In addition to assessing associations between soil microbial quality and environmental parameters, models were also implemented to determine if there were substantial differences in soil microbial quality between farms. Since there were no Listeria spp. positive samples from Farm C, and no Salmonella positive samples from Farm A, models were run (i) using data from all farms, and (ii) using only data from farms where there was at least one positive sample for the microbial target. A factor was considered associated with the odds of detecting a microbial target if ROPE was ≤0.025 and the probability of direction (PD) was ≥0.95; for all factors where these thresholds were not met, this was considered as failing to find evidence of an association.

For each farm and sampling visit, values for calcium (mg/kg), magnesium (mg/kg), phosphorous (mg/kg), potassium (mg/kg), sodium (mg/kg), sulfur (mg/kg), aluminum (mg/kg), copper (mg/kg), iron (mg/kg), manganese (mg/kg), molybdenum (mg/kg), zinc (mg/kg), total nitrogen (%), total carbon (%), organic matter (%), and pH were assessed. Summary statistics for all soil nutrients and properties across farms and sampling visits can be found in Table 1. Furthermore, values for air temperature (°C), dew point (°C), relative humidity (%), and cumulative rainfall (mm) for the 5 days prior to each sampling visit by farm can be found in Supplementary Table S5, or soil type in Supplementary Table S7.

The overall prevalence of Listeria spp. and L. monocytogenes in soil across the three farms were 10.0% (24/240) and 2.5% (6/240), respectively (Table 2). Farm B had the highest prevalence of both Listeria spp. (18.8%; 15/80) and L. monocytogenes (6.3%; 5/80), while no samples from Farm C were positive for Listeria. On Farm B, the five soil samples positive for L. monocytogenes represented four unique sampling sites, with one site testing positive during two separate sampling visits (visit 1 and 4; both during the fall separated by approximately 1 year). The overall prevalence of Salmonella in soil across the three farms was 4.2% (10/240; Table 2). Ninety percent (9/10) of Salmonella-positive soil samples were recovered on Farm C, with the majority recovered during the first sampling visit (6 of 9). These nine Salmonella-positive soil samples from Farm C represented eight unique sampling sites, with one site testing positive during two separate sampling visits (visit 1 and 2). The site with repeated Salmonella positive soil samples was directly adjacent to a pond used for irrigation. Irrigation on Farm C was weather dependent and applied by drip under plastic mulch. One soil sample was positive for Salmonella from Farm B during sampling visit 4. Farm A yielded no Salmonella- positive samples across all four visits.

Generic E. coli levels fell below the lower limit of detection (1 log₁₀ CFU/g) in over half of the soil samples collected in this study (55%; 133/240). The greatest prevalence of generic E. coli was found from soils collected on Farm B (67.5%; 54/80); followed by Farm A (43.8%; 35/80) and Farm C (22.5%; 18/80; Table 2). For the 107 soil samples where generic E. coli were above the limit of detection, the average concentration across all farms and sampling visits was 1.53 ± 0.77 log₁₀ CFU/g (Supplementary Table S6). While not significantly different by farm, Farm B had the highest generic E. coli levels among positive samples at 1.61 ± 0.86 log₁₀ CFU/g, followed by Farm A (1.48 ± 0.70 log₁₀ CFU/g) and Farm C (1.43 ± 0.62 log₁₀ CFU/g), respectively. Total coliform levels were detectable in all 240 samples collected and had an average concentration of 4.21 ± 0.94 log₁₀ CFU/g (Supplementary Table S6). Similar to generic E. coli, Farm B had the highest total coliform levels (4.66 ± 0.63 log₁₀ CFU/g), followed by Farm A (4.55 ± 0.78 log₁₀ CFU/g) and Farm C (3.42 ± 0.85 log₁₀ CFU/g), respectively.

Since no soil sample from Farm C was positive for Listeria, only data from Farms A and B were used when assessing how odds of isolating Listeria spp. and L. monocytogenes differed across farms. There was no evidence (PD = 0.92; ROPE =0.11) of an association between Listeria spp. and farms when using data from Farms A and B. Only two factors evaluated in the present study were substantially associated with a change in odds of Listeria spp. detection (PD ≥ 0.95 and ROPE ≤0.025); (i) if farm equipment and (ii) if farm workers were present in the field during sampling (Table 3). When farm equipment (e.g., tillage equipment, tractors) and farm workers (employees of the operation) were present during the time of sampling, the odds of detecting Listeria spp. was approximately 9 [MAP = 8.98, 95% Credibility Interval (CrI) =1.63, 38.75] and 8 (MAP = 7.72, CrI = 2.03, 64.17) times more likely, respectively (Table 3). As for L. monocytogenes, soil pH was the only evaluated factor substantially associated with a change in L. monocytogenes detection odds, with each one-unit increase in pH, there was a 5.5 times increase in the odds of L. monocytogenes detection (MAP = 5.48, CrI = 0.75, 345.39; Table 3).

Since no Salmonella was detected on Farm A, only data from Farms B and C were used when assessing the odds of detection. Salmonella was 11.6 (CrI = 1.36, 1155.27) times more likely to be recovered on Farm C, compared to Farm B (Table 3). The average concentration of generic E. coli in samples positive and negative for Salmonella were 1.35 ± 0.73 and 1.54 ± 0.78 log₁₀ CFU/g, respectively. There was no evidence of an association between the odds of Salmonella detection and all other factors considered in this study (e.g., weather, soil nutrients, management factors), including both the presence and concentration of generic E. coli or total coliforms.

Since E. coli levels fell below the lower limit of detection (1 log₁₀ CFU/g) for 55% (133/240) of soil samples, a hurdle model was implemented in order to separately understand factors associated with the presence and concentration of generic E. coli. Odds of generic E. coli detection differed by farm, with the odds of detection being 0.29 and 0.05 times less likely in Farm A and Farm C, compared to Farm B (Table 4). However, when the concentration of E. coli was examined, farm did not substantially impact levels (PD: 0.69–0.79; ROPE: 0.15–0.24). Three soil nutrients and two soil properties examined in this study were substantially associated with increased odds of detecting generic E. coli: calcium (MAP = 4.18, CrI = 0.89, 29.10), iron (MAP = 3.44, CrI = 0.93, 45.29), magnesium (MAP = 7.12, CrI = 1.53, 104.32), organic matter (MAP = 7.81, CrI = 0.75, 97.69), and pH (MAP = 4.87, CrI = 2.31, 12.22; Table 4). Organic matter had the greatest effect on generic E. coli detection with a one-unit change in organic matter (%) resulting in a 7.81 increase in odds of generic E. coli detection. Three soil nutrients were substantially associated with decreased levels of generic E. coli: iron (MAP = 1.71, CrI = −3.22, −0.15), sodium (MAP = −1.31, CrI = −2.21, −0.30), and sulfur (MAP = −0.73, CrI = −1.66, −0.04). A one-unit change in iron (mg/kg) resulted in the largest decrease in generic E. coli levels with a − 1.71 log₁₀ CFU/g change (Table 4).

Total coliform concentrations were 1.51 log₁₀ CFU/g lower on Farm C, as compared to Farm B (CrI = −2.04, −1.05; Table 4). No substantial difference in total coliform levels were observed between Farm B and Farm A (PD = 0.88, ROPE = 0.27). Four soil nutrients and two soil properties were substantially positively associated with total coliform levels: calcium (MAP = 0.49, CrI = 0.06, 1.02), iron (MAP = 0.38, CrI = 0.05, 0.78), manganese (MAP = 0.75, CrI = 0.41, 1.04), magnesium (MAP = 0.80, CrI = 0.23, 1.33), organic matter (MAP = 1.09, CrI = 0.48, 1.61), and soil pH (MAP = 0.36, CrI = 0.14, 0.55; Table 4). Similar to E. coli detection, organic matter had the greatest effect with a one-unit change in organic matter (%) resulting in an increase in total coliforms of 1.09 log₁₀ CFU/g.

From the 10 Salmonella-positive samples, five different Salmonella serovars were identified. Salmonella isolates from Farm B (visit 4) were identified as III 42:z10:e,n,x,z15 (Table 5). The remaining Salmonella isolates, all from Farm C, yielded the following serovars: Enteritidis (1), Florida (2), Javiana (3), and Newport (3). S. enteritidis, S. Florida, and S. Javiana were all isolated from visit 1 on Farm C, while S. Newport was only isolated during visits 2 and 3 (summer timeframe).

When farm equipment (e.g., tillage equipment, tractors, etc.) and farm workers were present in the field at the time of sampling, the odds of isolating Listeria spp. increased substantially. While we failed to find an association between soil disturbances by animal intrusions and Listeria spp. prevalence, these results suggest that soil disturbances by field management practices may impact Listeria spp. prevalence. This aligns with previous studies which have demonstrated that management practices that disturb the soil and allow the subsurface to become exposed increase the probability of microorganism detection (Vivant et al., 2013; Strawn et al., 2013b). In contrast, an additional study found that within an organic field in Maryland (United States), Listeria spp. populations decreased following tilling events (Reed-Jones et al., 2016). However, since tillage has been shown to affect the microbial structure of soil, leading to the loss of total carbon, total nitrogen, and the breakdown of organic matter (Jackson et al., 2003), more research is needed on the effects of soil disturbances on bacterial detection, specifically Listeria.

The study presented here demonstrated that intrastate region influenced Salmonella prevalence in Virginia soils with Salmonella detection more likely to occur in soils from the Coastal Plain region (Farm C). Additionally, the Salmonella serovar isolated from Farm B was not isolated from Farm C. These results suggest that diversity of Salmonella may differ by region; however, this finding may be an artifact of the small number of positive Salmonella samples in the reported study. This is supported by previous work that has shown Salmonella serovar diversity differs across regions, including interstate regions (Gorski et al., 2011; Strawn et al., 2014; Murphy et al., 2023). These findings are of interest as produce grown and packed in the Coastal Plain region of Virginia have been implicated in Salmonella outbreaks associated with fresh produce (Greene et al., 2008; Bell et al., 2015). Previous research in this region of Virginia has also shown repeated isolation of Salmonella from environmental sources including sediment, water, soil, and wildlife (Greene et al., 2008; Gruszynski et al., 2014; Angelo et al., 2015; Bell et al., 2015; Truitt et al., 2018; Gu et al., 2019; Murphy et al., 2023). It has been hypothesized that Salmonella may be introduced to Coastal Plain soils by wildlife within the production environment (Greene et al., 2008; Gruszynski et al., 2014; Bell et al., 2015; Truitt et al., 2018; Gu et al., 2019). A 2013 study conducted within the Coastal Plain region of Virginia revealed that 8.8% (23/262) of wildlife samples were positive for Salmonella, with the majority originating from avian species (65.2%; 15/23) followed by reptiles (30.4%; 7/23) and mammalian species (4.3%; 1/23) (Gruszynski et al., 2014). Surveys in this study collected from Farm C (costal plain region) listed expected animal intrusion from avian, rodent, and reptile species. While all farms in this study mentioned wildlife control measures were in place, it was noted the difficulty to maintain these barriers and/or controls. Therefore, a continued understanding of how regional factors, such as wildlife in the surrounding area, contribute to Salmonella prevalence could be of value to growers when evaluating the risks associated with their production environments.

The prevalence of both L. monocytogenes and generic E. coli; as well as, the concentration of total coliforms were positively associated with increased pH levels. Farm B soil, which had the highest prevalence of both L. monocytogenes and generic E. coli, and the highest concentration of total coliforms, had the highest pH values at 6.34 ± 0.57 (range 5.30–7.60). Conversely, Farm C had the lowest pH values (5.97 ± 0.44), with no L. monocytogenes-positive samples, the lowest generic E. coli prevalence, and the lowest total coliform concentrations. While L. monocytogenes can grow in both neutral and acidic soil conditions (Locatelli et al., 2013; Linke et al., 2014; Ferguson et al., 2023), a previous study conducted across 2 years from 12 areas in Austria found that from 467 soil samples collected, Listeria isolation was significantly associated with soils that had a neutral pH (7.44; 3.43–9.90; Linke et al., 2014). As for E. coli, greater survival rates have been observed in soils with a moderately neutral to alkaline pH (6.0–8.5), compared to acidic (Reddy et al., 1981; Sjogren, 1994; Wang et al., 2014; Emch et al., 2020). In Virginia, soils tend to be acidic and require lime additions every few years as the pH slowly drops, with lime applications based on soil testing with a target pH of 6.2 for most produce fields (Maguire and Heckendorn, 2022). As this practice may contribute to changes in soil pH, more research is needed to better understand the relationship between soil pH, management practices, and pathogen prevalence.

However, one limitation of the study presented here is data was not collected for when fertilization of sites occurred (e.g., when were additions/applications of nitrogen, phosphorous); as well as, information on exact fertilization sources. This additional data could assist with elucidating potential associations between foodborne pathogens and macro- and micro-nutrients. It is known that crops will uptake soil nutrients for their metabolisms (Näsholm et al., 2009; Bindraban et al., 2015); thus understanding how and at what concentrations foodborne pathogens may require macro-and micro-nutrients could assist in understanding survival and persistence behavior within agricultural soils.

Total coliforms and generic E. coli are frequently used as indicators for sanitary conditions and the presence of fecal bacteria; however, since both have known ecological and environmental niches other than fecal sources, their presence is not always indicative of fecal contamination (Cox et al., 1988; Luo et al., 2011). Testing for foodborne pathogens is often more labor, time, and cost-intensive, so testing for E. coli and total coliforms is often still beneficial, under certain circumstances (e.g., flooding). Previous studies have investigated the use of indicator bacteria, including generic E. coli and total coliforms, as indicators of foodborne pathogens in soils, and have found varying results (Natvig et al., 2002; Holvoet et al., 2014; Reed-Jones et al., 2016; Hruby et al., 2018; Emch et al., 2020). Results from this study showed E. coli, both prevalence and concentration, and total coliforms were not associated with Salmonella, Listeria spp., or L. monocytogenes prevalence. Therefore, the use of indicator bacteria for predicting times and locations of increased likelihood of pathogen prevalence in soil may not be appropriate. This result may be due to the limited sample size, complexities in soil properties, and/or environmental conditions, but in the present study indicator bacteria were not associated with Salmonella or L. monocytogenes prevalence.

However, the use of indicator organisms for pathogen prediction under specific soil conditions and situations may be useful, such as after a flooding event (Castro-Ibáñez et al., 2015; Bergholz et al., 2016; Callahan et al., 2017). Flooding events pose a large risk for growers both from a food safety, quality, and operational perspective. A previous study found that, following a flooding event, bacteria could be recovered from soils up to 9 m from the threshold of the flood (Callahan et al., 2017), while an additional study observed irrigation water samples collected 1 week after a flooding event had elevated levels of both coliforms and generic E. coli (Castro-Ibáñez et al., 2015). These findings demonstrate a flood event may lead to increased levels of biological contamination in the environment (e.g., soil, irrigation water). Recent flooding events in other major growing regions of the United States have emphasized this risk, attracting attention from the United States FDA; as well as, trade organizations, such as Western Growers Association and the California Leafy Greens Marketing Agreement (LGMA) (California Leafy Greens Marketing Agreement, 2023; Food and Drug Administration, 2023; Timmins, 2023). Specifically, the California LGMA requires soil testing following a flooding event that suggests growers should wait a minimum of 60 days after a flooding event to begin planting and that fecal coliform levels should be <100 MPN/g in soils (California Leafy Greens Marketing Agreement, 2021; California Leafy Greens Marketing Agreement, 2023). While the research presented here demonstrates that routine testing of soils for indicator organisms may not be able to provide insight into pathogen prevalence, it may be helpful for growers in mitigating risks during unforeseen events or conditions (e.g., flooding), or trending/monitoring when indicator organism levels adjust to baseline levels.