A 21-month longitudinal, multi-regional field study was conducted on 19 USDA-NOP certified farms located in four USA regions (California, Minnesota, Maine, and Maryland) during two growing seasons (March 2017–December 2018). Details on the recruitment, study design and farm enrollment can be found in a previous publication (Ramos et al., 2021). Briefly, farms were enrolled based on the following criteria: (1) certified USDA-NOP fresh produce farm, (2) use of untreated manure (i.e., raw or untreated manure), (3) grow at least one of the following fresh crops: leafy greens, root vegetables, and/or fruits, (4) willingness to participate and provide biological samples (raw manure, soil, irrigation water, and fresh produce) and information about their farm management practices and location. Farms were sampled periodically from March 2017 to December 2018 using the following scheme in each growing season: (1) untreated manure, compost and soil were sampled before manure application, (2) soil samples were collected after manure incorporation and monthly for a total of seven times, (3) fresh produce, and (4) irrigation water. Sample collection and preparation are described below.

The following samples (a total of 3,260 samples) were collected from the farms during the two growing seasons: 233 untreated manure samples (cattle, horse, small ruminants, or poultry), 2,461 soil samples, 527 harvested produce samples [leafy greens (e.g., lettuce, spinach, kale), root vegetables (e.g., carrots, radishes) and/or fruits (e.g., tomatoes, cucumbers, peppers)], and 39 irrigation water samples.

A questionnaire was developed and used on each farm to obtain data on farm-level management practices and environmental characteristics. The questionnaire included: (i) farm demographics and information (e.g., farm size, number of staff, third-party certifications, marketing outlets, worker training); (ii) crop production (e.g., crop types, previous plant/soil diseases), (iii) soil characteristics and surrounding landscape (e.g., soil type, soil testing, previous use, use for livestock/poultry, surrounding landscape use), (iv) fertilizer use [e.g., manure and compost use, manure type, source, storage, application method, rate and frequency, in-house vs. commercial sources, other animal biological soil amendments, location of BSAAO relative to the crops (down/up-hill)]; (v) agricultural water (e.g., source, irrigation type, testing, irrigation frequency), and (vi) management practices (e.g., tillage practices, crop rotation, livestock/poultry in property and proximity to livestock/feedlots, integration of livestock, presence of domestic/companion animals, presence of wildlife, evidence of animal feces in fields). In addition, time-variable factors were recorded for each month visited, including intrusion of wild and domestic animals in the fields, abnormal field conditions, irrigation frequency, type of produce crop, and growth stage. The full questionnaire can be accessed on request.

Samples were collected and tested as previously described (Ramos et al., 2021). Farms were sampled monthly for a period of up to 180 days over two growing seasons. On each site, a total of five composite untreated manure samples was collected from manure piles using a shovel and a sterile scoop (Fisher Scientific, Hampton, NH). Before manure application (d0A, baseline), each composite sample was collected from five separate spots and different depths (not surface) in the manure pile with at least 90–100 g total per composite sample. For each manure and produce type, four composite soil samples (five subsamples each) were collected after manure incorporation (d0B) and monthly, for a total of seven times (30, 60, 90, 120, 150, and 180-days post-application (dpa) of manure) for each crop cycle as described by Ramos et al. (2021). Soil samples were collected using a stainless-steel soil core sampling probe (2.54 cm diam) disinfected with 70% ethanol between samples and 15.2 cm deep from the soil surface (Ramos et al., 2021). Agricultural water was collected once at each farm during each growing season as previously described (Ramos et al., 2021). Briefly, water samples were collected directly into a sterile Nalgene bottle (Thermo Scientific, Rochester, NY) by hand or with a sampling pole. A maximum of two irrigation water samples (1 L each) were collected from different sources (e.g., agricultural well, pond, reservoir, lagoon, or creek). All samples were collected by one of the research team members, transported or shipped to the laboratory in an insulated cooler with ice packs, and processed within 48 h post-collection.

All samples were prepared as previously described by Ramos et al. (2021) for detection, isolation, and identification using three separate enrichment schemes for non-O157 STEC, L. monocytogenes, E. coli O157:H7, Salmonella spp., and generic E. coli. Here we describe details of the enrichment and isolation of non-O157 STEC. Details on the isolation and identification of other three pathogens can be found in Ramos et al. (2021). For each manure, compost, and soil sample collected, 30 g of sample were weighed and transferred to 24 oz Whirl-Pak bags (Nasco, Modesto, CA) containing 270 ml of tryptic soy broth (TSB; 1:10 dilution), then TSB bags were manually massaged for 1 min before incubation at 25°C for 2 h, followed by a 42°C incubation for 8 h shaking at 100 rpm in a Multitron programmable shaking incubator (Eppendorf, Hauppauge, NY, USA). Then samples were held at 6°C with no shaking until used for further enrichment and isolation of non-O157 STEC.

Irrigation water samples were processed as detailed previously (Partyka et al., 2018; Ramos et al., 2021). Briefly, water samples were analyzed for bacterial indicators (i.e., generic E. coli and non-E. coli fecal coliforms) and pathogens (non-O157 STEC, E. coli O157:H7, L. monocytogenes, and Salmonella) using standard membrane filtration techniques. Water samples were vacuum-filtered through a sterile 0.45 mm nitrocellulose filter assembly.

Application of manure can introduce foodborne pathogens into the soil and increases the risk of soil and fresh produce contamination (Harris et al., 2013; Strawn et al., 2013b; Park et al., 2015). Management practices and environmental factors, such as the season of application, soil characteristics, and weather-related factors, can influence such presence of foodborne pathogens in the soil (Strawn et al., 2013a,b; Marine et al., 2015; Pagadala et al., 2015; Weller et al., 2015); however, the NOP requirements are based on a single metric: time-interval criteria (USDA-NOP, 2011). In the present study, foodborne pathogen presence in manure and soil (d0A, before application) and manure management-related factors (animal species source, age, storage, application manure rate, season, and method) were significantly associated with non-O157 STEC presence in the soil in the univariate model (Table 1). However, only manure application season (summer and fall) and application method were statistically significant in the final multivariable model (Table 2). Application of manure in the summer increased the odds of detecting non-O157 STEC in the soil, whereas fall application decreased the odds compared to spring application. Most of the farms apply manure during the fall, either because of state and local regulatory requirements (Midwest and East coast) (Pires et al., 2018). Moreover, fall application allows for a greater time of manure breakdown, while allowing to be in compliance with NOP required time-interval of 90–120 days between application and harvest of fresh produce (Pires et al., 2018). These findings have particular importance because it requires defined planning for the planting and harvesting of crops covered by FSMA-PSR.

Seasonal variations in the presence of foodborne pathogens in soil and contamination of fresh produce have been described in several studies (Moriarty et al., 2011; Marine et al., 2015; Sharma et al., 2019). In a Mid-Atlantic USA study, depending on site and initial soil moisture, season contributed to the survival variability of gEC and attO157 in soils amended with manure (Sharma et al., 2019). Moreover, Strawn et al. (2013b), reported that the application of manure in the prior year was associated with the risk of Salmonella-positive field samples (Strawn et al., 2013b). However, the farms in the present study applied manure every year as a soil amendment, and the manure-related factors (livestock source, application rate, storage, etc.) were not significant in the final model.

Farms applying untreated manure followed by rototilling/disking were less likely to have a non-O157 STEC-positive soil sample. Other manure incorporation methods were less commonly used by organic farmers, and most of the farms would use a combination of two or more methods (Pires et al., 2018). Entry et al. (2010) reported that the survival of fecal indicators organisms (i.e., E. coli and Enterococcus sp.) in radish in rhizosphere soil and radish roots were influenced by tillage method and depth of incorporation of dairy manure into the soil (Entry et al., 2010). Sharma et al. (2019) described that the survival of attenuated pathogens (gEc and attO157) in manure-amended soils was affected by application method (i.e., surface-applied manure, tilled-in manure), and the effect depended on the inoculum level and type of amendment, with poultry manure supporting the longest survival. In contrast, Hutchison et al. (2004) showed that the decline of three foodborne pathogens (i.e., E. coli O157:H7, Salmonella spp., and Campylobacter spp.) was significantly more rapid for livestock manure when it was left on the soil surface, potentially due to desiccation and UV effect (Hutchison et al., 2004). Other environmental conditions such as ambient temperature affected the rate of die-off of Salmonella and E. coli in soil amended with bovine manure (Natvig et al., 2002). Moreover, a recent meta-analysis study reported that temperature (soil and air combined) was the most significant factor affecting persistence of E. coli and Salmonella spp, in addition to the application methods, with manure incorporation into the soil with longer survival compared to surface application (Tran et al., 2020). The inconsistent findings among studies may show that there are factors other than application method affecting the persistence and survival of pathogens, such as microbial competition, soil characteristics, and environmental factors.

Previous use of the field for livestock and the presence of animal feces on the field (as reported by the farmer) were associated with a higher risk of a non-O157 STEC-positive sample in the soil. Livestock can carry foodborne pathogens intermittently, and the direct fecal deposit of feces can be a source of the pathogens, which, depending on the pathogen, can survive in the soil for long periods (Alegbeleye et al., 2018). Adjacent land use for cow-calf rearing and the presence of feral pigs were implicated in the 2007 spinach outbreak due to E. coli O157:H7 in Central Coast California (Jay et al., 2007). More recently, the 2018 multi-state E. coli O157:H7 outbreak associated with romaine lettuce and potential contamination of water due to proximity to a large animal feeding operation highlights the need to investigate the proximity of livestock operations to fresh produce fields (Waltenburg et al., 2021). Berry et al. (2015, 2019) reported that 400 feet may not be adequate to limit the transmission of E. coli O157:H7 to produce crops close to a Concentrated Animal Feeding Operation (CAFO), and flies may have a role in leafy greens contamination (Berry et al., 2015, 2019). Actually, the prevalence of enterohemorrhagic E. coli was highest on fresh produce fields near areas suitable for livestock grazing on California Central Coast farms (Karp et al., 2015).

Integration of crop-livestock has been increasing in different regions in the US. Despite many benefits from the presence of livestock, it increases the risk of cross-contamination of vegetable and nut crops (Salaheen et al., 2015; Patterson et al., 2018). A study conducted in sustainable farming systems in Tennessee and North Carolina suggested the potential transmission of antimicrobial-resistant non-O157 E. coli between animal feeding operations and fresh produce (Glaize et al., 2020). In order to reduce contamination, it is important to assess the risk of contamination from the surrounding landscape and adjacent land use, such as proximity of livestock operations and potential wildlife intrusions. As part of the Leafy Greens STEC Action Plan, FDA is conducting a multi-year California longitudinal study to assess the pathogen dynamics in the environment, particularly landscape and adjacent land use effects on contamination (FDA, 2022).

Presence of fecal material increased the odds of non-O157 STEC-positive samples in the soil. Although no differentiation in the analysis was made between fecal material from domestic animals and wildlife, the majority of those observations were linked to wildlife presence (research team observations). Wildlife such as birds, reptiles, feral pigs, and deer are well-known carriers of foodborne pathogens such as Salmonella, Campylobacter, and pathogenic E. coli (Jay et al., 2007; Jay-Russell et al., 2012; Gorski et al., 2013; Alegbeleye et al., 2018; Navarro-Gonzalez et al., 2020). Measuring the direct risks of wildlife intrusion and fresh produce production contamination is complex; however, several outbreaks have been linked to wildlife intrusion. Deer were identified as the causal source of an E. coli O157:H7 outbreak in strawberries in Oregon (Laidler et al., 2013), and cranes were the environmental source for a campylobacterosis outbreak in peas (Gardner et al., 2011). The FSMA-PSR includes language for prevention of contamination and risk mitigation; however, this is based on individual farm assessments, which may vary from grower to grower and may be landscape-specific (FDA, 2015). Determining adequate no-harvest buffers around in-field fecal material is necessary in order to avoid the contamination of fresh produce by foodborne pathogens (Salmonella, E. coli, and E. coli O157:H7), which depends on fecal sources, irrigation, and splash events (Atwill et al., 2015; Weller et al., 2017; Topalcengiz et al., 2020). An experimental study conducted in the Northeastern United States showed E. coli levels on lettuce and lettuce-feces distance were negatively correlated, when using a buffer smaller or equal to 1 meter during foliar regulation (Weller et al., 2017). The current recommendations of the California Leafy Greens Market Agreement (LGMA) if there is evidence of fecal material presence is not to harvest any crop within a minimum of five-foot radius buffer distance from the contamination site (LGMA, 2021). Therefore, evaluation is needed for the risk of wildlife intrusion, either by fecal material presence or crop destruction, and implementation of mitigation strategies, such as buffer zones and crop destruction.

A multitude of biotic and abiotic factors affect the persistence of foodborne pathogens in soil such as soil type (e.g., clay, sandy), soil composition (e.g., organic matter, moisture, nutrients), soil temperature, and soil texture (Ingham et al., 2005; Alegbeleye et al., 2018). The effect of these factors may be altered by agronomic practices and environmental factors, such as precipitation and ambient temperature (Entry et al., 2010; Sharma et al., 2019; Pang et al., 2020). In the present study, the presence of generic E. coli, soil moisture, and sodium were associated with likelihood of non-O157 presence in the manure-amended soil. In model 2, generic E. coli was significantly associated with the presence of non-O157 in the soil. Given that non-O157 STEC would contribute to the generic E. coli, this association would be expected, however, we did not quantify the pathogen, and direct correlation was not assessed. Generic E. coli may be a good predictor of non-O157 STEC presence and a fecal contamination indicator. Because detection of STEC and E. coli O157:H7 in soil is infrequent, and due to the prohibitive use of pathogenic strains in privately operated farms in experimental studies, generic E. coli has been widely used as a proxy for fecal contamination in studies investigating the survival of pathogenic E. coli in soil amended with untreated animal manure in produce fields (Reed-Jones et al., 2016; Sharma et al., 2016, 2019; Patterson et al., 2018). A recent meta-analysis reported that temperature (soil and air combined) was the most prominent factor affecting the persistence of E. coli in manure-amended soils (Tran et al., 2020). Moreover, Sharma et al. (2019) reported that moisture content was the individual factor that accounts for the greatest variability in survival duration of generic E. coli in manure-amended soils in the Mid-Atlantic USA (Sharma et al., 2019). In the current study, soil moisture was significantly associated with the presence of non-O157 STEC in the soil in the final multivariable model (model 2), and the odds of identifying a non-O157 STEC positive sample increased by 1.37 for each 10% in the soil moisture. This is in disagreement with previous studies (Park et al., 2016; Sharma et al., 2016). Lower initial soil moisture content (11.2 to 12.1%) supported longer durations of survival for certain seasons and sites in a Mid-Atlantic study (Sharma et al., 2019). Soil moisture depends on precipitation, with irrigation being crop-specific and crop-maturity–related. In the present study, we did not record irrigation events, which were adjusted according to the crop; instead, we used soil moisture at the sample level, accounting for between-sample variability which combines weather factors (precipitation and temperature) and irrigation, because it was measured at soil-sample level, as opposed to the one unique measurement at plot level by the moisture sensor (at field level measured by HOBO).

Interestingly, in this study, multiple weather-related factors (temperature, precipitation, wind, and snowfall) at different lag times were significantly associated with non-O157 STEC presence in the univariable analysis, but only snowfall was retained in the final multivariable analysis (model 2) with lower odds of a non-O157:H7 STEC positive soil sample. These findings contrast with other studies on the association between prevalence of foodborne pathogens in the fresh produce farm environment and weather factors. Temperature and precipitation influenced the likelihood of L. monocytogenes and Salmonella in fruit and vegetable farms from New York State (Strawn et al., 2013a). Events such as rain and precipitation were associated with the presence of Listeria species and L. monocytogenes in New York State farm environments (Weller et al., 2015). In both models, there was lower likelihood of identifying a non-O157 STEC positive soil sample in the second year of the study, even accounting for other factors. This is in agreement with inoculation experimental studies, in which spatio-temporal factors (site, year, and season) influenced survival duration of generic E. coli and attenuated O157 E. coli more than weather effects (i.e., average daily temperature and rainfall), with the year being the greatest contributor to variability (Sharma et al., 2019). This shows the importance of multi-year and multi-site studies over several growing seasons when investigating foodborne pathogen survival and persistence in soil amended with untreated manure.

Sampling day had very similar results in both models. There was a trend that the odds of a positive sample were lower up to 90 days post-manure application (compared to d0B), but not statistically significant, with the exception of sampling day 60 post-manure application. Interestingly, this trend changed direction after 120 days post-application, although not statistically significant. Part of the explanation could be due to the fact that the total number of samples per farm started to decrease after this time point and most of the farms had finished harvesting produce, only keeping the plots for the research sampling. The presence of vegetable decay and potential intrusion of wildlife could have contributed to this shift as well as to the potential introduction of foodborne pathogens. Certified organic farms follow USDA-NOP standards for wait-times (90/120 days) between the application of raw manure and crop harvest. The multivariable models show that the odds of finding a non-O157 STEC positive soil sample decreases with time. The time-interval as a mitigation strategy is based on the assumption that an appropriate amount of time has passed to reduce the risks of foodborne pathogen survival in the soil (Phan-Thien et al., 2020). However, in the present study, several on-farm practices (i.e., manure management), soil factors, and weather contributed to the non-O157 STEC presence in the soil after the application of untreated manure. Understanding how these factors interact and what the key time-variant factors are is crucial to creating a customized risk-based assessment of the farm and implementing mitigation practices for the reduction of farm environment contamination of foodborne pathogens when untreated manure is used as a soil amendment.

There are some limitations in observational studies and predictive models to assess the risks factors associated with the likelihood of pathogen presence in soil amended with untreated manure. First, we were unable to directly link the presence of non-O157 STEC in the manure and soil, as it was not part of the current study to investigate genetic relatedness for the isolates; however, no association was found between the presence of pathogens in manure and soil in the final models. Another limitation is that the study was conducted on commercial farms without interventions, in which the exact contamination load in the manure was unknown (as opposed to inoculation trials). Information collected in observational studies with multiple variables is relevant as representing the natural stage and diversity of agricultural environments, which allows us to investigate simultaneously different predictors. Moreover, data collection suffered from multiple issues that could potentially introduce bias. First, there was a large number of missing data for a number of variables. Some of these were removed but we chose to retain others due to their biological relevance to the study objectives. This particularly affected multivariable model 1 potentially leading to a loss of some external validity for this model. Next, different data groups were sampled with different temporal resolutions (sampling day vs. once or twice a year) and spatial resolutions (farm vs. plot). This was partially accounted for by using random effects but could still lead to some bias, as well as, causing a loss of analytical power due to the need to repeat identical values of predictors with lower resolution to match the resolution of the outcome variable. Finally, some variables (notably soil composition) suffered from an unbalanced sampling frame with some farms having much less samples collected, leading to the need to aggregate values at the farm level with a loss of resolution for these variables.