ABSTRACT
Biological soil amendments of animal origin (BSAAO) increase nutrient levels in soils to support the production of fruits and vegetables. BSAAOs may introduce or extend the survival of bacterial pathogens which can be transferred to fruits and vegetables to cause foodborne illness. Escherichia coli survival over 120 days in soil plots (3 m2) covered with (mulched) or without plastic mulch (not mulched), amended with either poultry litter, composted poultry litter, heat-treated poultry pellets, or chemical fertilizer, and transfer to cucumbers in 2 years (2018 and 2019) were evaluated. Plots were inoculated with E. coli (8.5 log CFU/m2) and planted with cucumber seedlings (Supremo). The number of days needed to reduce E. coli levels by 4 log CFU (dpi4log) was determined using a sigmoidal decline model. Random forest regression and one-way analysis of variance (ANOVA; P < 0.05) identified predictors (soil properties, nutrients, and weather factors) of dpi4log of E. coli and transfer to cucumbers. The combination of year, amendment, and mulch (25.0% increase in the mean square error [IncMSE]) and year (9.75% IncMSE) were the most prominent predictors of dpi4log and transfer to cucumbers, respectively. Nitrate levels at 30 days and soil moisture at 40 days were also impactful predictors of dpi4log. Differing rainfall amounts in 2018 (24.9 in.) and 2019 (12.6 in.) affected E. coli survival in soils and transfer to cucumbers. Salmonella spp. were recovered sporadically from various plots but were not recovered from cucumbers in either year. Greater transfer of E. coli to cucumbers was also shown to be partially dependent on dpi4log of E. coli in plots containing BSAAO.
IMPORTANCE Poultry litter and other biological soil amendments are commonly used fertilizers in fruit and vegetable production and can introduce enteric pathogens such as Escherichia coli O157:H7 or Salmonella previously associated with outbreaks of illness linked to contaminated produce. E. coli survival duration in soils covered with plastic mulch or uncovered and containing poultry litter or heat-treated poultry litter pellets were evaluated. Nitrate levels on day 30 and moisture content in soils on day 40 on specific days were good predictors of E. coli survival in soils; however, the combination of year, amendment, and mulch type was a better predictor. Different cumulative rainfall totals from year to year most likely affected the transfer of E. coli from soils to cucumbers and survival durations in soil. E. coli survival in soils can be extended by the addition of several poultry litter-based soil amendments commonly used in organic production of fruits and vegetables and is highly dependent on temporal variation in rainfall.
INTRODUCTION
In the United States, the poultry industry in the Delmarva peninsula (Mid-Atlantic region, including Delaware and portions of Maryland and Virginia) produces approximately 609 million chickens annually, generating approximately $3.5 billion in annual revenue since 2017 (1). The region currently has 5,114 poultry houses, which produces approximately 552,599 tons of poultry litter (manure and bedding material) annually (2). Poultry litter is mostly collected and piled in sheds and then applied to agricultural fields as an untreated or composted biological soil amendment(s) of animal origin (BSAAO) to improve soil fertility and health. BSAAO provides essential nutrients, primarily nitrogen and phosphorus, along with secondary nutrients (calcium, magnesium, and sulfur) and various micronutrients to enhance crop production (3–5). Soil amendments replenish organic material, enhance the density, structure, and water holding capacity of the soil, and increase crop yield (5–8). However, raw poultry litter and other untreated BSAAOs can potentially carry foodborne pathogens such as Shiga-toxigenic Escherichia coli, Salmonella enterica, Listeria monocytogenes, Campylobacter spp., and Cryptosporidium parvum (9).
Previous field and laboratory work has shown that the survival of foodborne pathogens in the soil or animal manure varies depending on spatiotemporal (site, year, and season), agricultural (amendment type, management, and depth), soil composition (organic matter, humidity, and carbon content), and weather effects (rainfall, air temperature, sunlight, relative humidity, and season) (10–13). Manure processing methods such as composting, heat treatment, and pelletization have been utilized to reduce pathogens in these amendments (8, 14, 15). Pires et al. (12) surveyed 666 farms from eight U.S. states (California, Wisconsin, Pennsylvania, Washington, Ohio, Vermont, Maine, and Minnesota) to determine agricultural practices associated with the application of BSAAO on organic farms. Authors reported that 47% to 92% of growers used BSAAO, and 85% of them used raw manure as BSAAO. Moreover, 46.8% of farmers reported using BSAAO in fields for produce which is typically consumed fresh and 50.1% used BSAAO for produce that is consumed cooked or processed. The study reported that 56.8% of the farmers growing fresh produce use composted soil amendments, and 74.6% of them used on-site composted manure. Of the organic farmers growing fruits and vegetables in this survey, 25% used untreated BSAAO (raw manure) as an organic fertilizer, with poultry litter, followed by dairy manure, horse manure, and small ruminant manure as the most commonly used manure types. The authors also reported that 90% of the respondents of the survey harvested crops after 90 or 120 days after raw manure application as per the U.S. Department of Agriculture (USDA) National Organic Program (NOP) guidelines. The USDA NOP regulations for composting include establishing a carbon-to-nitrogen ratio (C:N) between 24:1 and 40:1 and maintaining temperatures (55 to 76.6°C) for 3 days in a static pile system (16). USDA NOP guidelines also propose applying untreated BSAAO (manure) to soils intended to grow produce 90 to 120 days before harvest of the crop to minimize the likelihood of contamination from soils by providing a sufficient period of time for foodborne pathogens to die off. These guidelines state that when produce is grown close to the ground and comes into direct contact with the soils, a 120-day interval between application and harvest is used; if the crop is not likely to contact the soil, then a 90-day interval can be used. Pathogen survival in BSAAO-amended soil may vary depending upon geographical location, environmental conditions, and local farming practices (11).
The produce safety rule (PSR) of the Food Safety Modernization Act (FSMA) has been proposed by the U.S. Food and Drug Association (FDA) to regulate fresh produce production. Subpart F defines BSAAO and associated practices to reduce food safety risks of public health significance. It states that if treated (processed) or untreated, BSAAO must be handled in such a manner that it does not contact the edible portion of produce/crop after application (17). While the FDA currently has no objection to the use of the USDA National Organic Program standards (90/120-day interval), risk assessment to determine the appropriate interval between the application of untreated BSAAO to soils and the harvest of fruits and vegetables is ongoing (17). Apart from determining the appropriate interval between manure application and harvest of fruits and vegetables, data are lacking to characterize extrinsic and intrinsic parameters of soils along with environmental factors in pathogen persistence in BSAAO-amended soils. While increasing amounts of data are emerging to help evaluate the spatial, temporal, and biological effects of amended soils on pathogen persistence, limited scientific data are available on assessing direct microbial transfer from contaminated amended soils to fruits or vegetables. Research presented here attempts to characterize specific weather, physical, and chemical factors affecting pathogen survival in soils and transfer to cucumbers. In 2013, 2014, and 2015, three large multistate outbreaks of salmonellosis associated with cucumbers grown in the field either in Mexico or in the United States resulted in 1,266 illnesses, 270 hospitalizations, and 7 deaths (18–20).
The objectives of the study described here were (i) to evaluate nonpathogenic E. coli as a surrogate for enteric pathogen survival in soils amended with poultry litter-based BSAAO and (ii) to examine the influence of extrinsic and intrinsic factors of soil on E. coli survival and transfer to cucumbers.
RESULTS
The combination of year (2018 or 2019), amendment (consisting of either composted poultry litter [CPL], heat-treated poultry litter pellets [HTPP], poultry litter [PL], or unamended [UN; containing chemical fertilizer]), and mulch (M; covered with black plastic mulch; NoM, not mulched [not covered with black plastic mulch]) was the strongest predictor of E. coli survival as measured by the number of days postinoculation needed to achieve a 4-log decrease in log CFU per gram (dry weight) (dpi4log). The combination of year, amendment, and mulch had a larger percent increase in the mean square error [% IncMSE] value (25%) than combinations of amendment and mulch (9.0% IncMSE) and of year and amendment (5.5% IncMSE) in the random forest (RF) model (Fig. 1). The RF model accounted for 98% of variance in predicting E. coli population declines (dpi4log) in soils.
Variable importance plots from random forest models for factors predicting survival of E. coli (dpi4log) in soils containing various amendments (CPL, HTPP, PL, or UN) (A) or transfer of E. coli to cucumbers (log MPN/cucumber) (B).
The combination of year, biological soil amendment type, and plastic mulch affected E. coli levels in soils.Our results show that it is difficult to disentangle the influences of temporal (year) and agricultural effects (amendment and mulch type) on the survival duration of E. coli in soils containing BSAAO as measured by dpi4log values. In general, longer survival durations were observed in 2018 than in 2019. Table 1 shows that in 2018, 50% (4/8) of the amendment-mulch combinations in replicate plots had dpi4log values >90 days, while in 2019, no amendment-mulch combination (0/8) had a dpi4log value of >90 days. The combination of mulching with biological soil amendments extended survival durations of E. coli but decreased survival durations in unamended (UN; chemical fertilizer) plots in 2018. In 2018, mulched plots containing composted poultry litter (97.5 days) and heat-treated poultry pellets (106 days) had significantly (P < 0.05) greater mean dpi4log values than not-mulched plots containing composted poultry litter (49 days) and heat-treated poultry pellets (28.5 days), respectively. In 2018, dpi4log values in mulched unamended plots (49 days) were significantly lower than in not-mulched unamended plots (100 days). The dpi4log values for mulched amended plots in 2018, including mulched heat-treated poultry pellets (106 days), poultry litter (99.5 days), and composted poultry litter (97.5 days), were statistically similar to unamended, not mulched dpi4log values (100 days). The influence of these biological amendments and plastic mulch may amplify the temporal effects of any particular season or year on E. coli survival duration.
Mean dpi4log values for each combination of year, amendment type, and mulch type
In 2019, the trend of mulched amended plots supporting longer E. coli survival durations (larger dpi4log values) than not-mulched plots was observed for composted poultry litter and poultry litter treatments. E. coli survival (dpi4log values) was significantly greater in mulched composted poultry litter (50 days) and mulched poultry litter (85.5 days) than in not-mulched composted poultry litter (28.5 days) and not-mulched poultry litter plots (46 days), respectively. However, the opposite trend was observed, where mulched heat-treated poultry pellet plots (65.5 days) supported significantly shorter survival durations for E. coli than not-mulched heat-treated poultry pellet plots (85.5 days). In 2018 and 2019, 3 of 8 not-mulched plot types (2018, unamended [100 days]; 2018, poultry litter; 2019, heat-treated poultry pellets [85.5 days]) had greater dpi4log values for E. coli than their mulched counterparts.
Rainfall patterns and soil moisture affected E. coli survival in amended soils.The differences in cumulative rainfall may explain the longer survival durations of E. coli in 2018 than in 2019. More rainfall (24.94 in.) in 2018 than in 2019 (12.63 in.) was observed over the 120-day field trial period. Figure 2 shows the cumulative rainfall every 10 days over the 120-day periods in the 2018 and 2019 studies. In 2018, there were 4.17 in., 4.28 in., 3.83 in., and 3.16 in. of rainfall between the periods 31 to 40 days, 41 to 50 days, 51 to 60 days, and 61 to 70 days, respectively, compared to 1.82 in., 0.3 in., 0.17 in., and 1.65 in. in the same time periods in 2019. The rainfall amounts and patterns likely affected the moisture content of the soil on specific days of analysis. The RF model identified the soil moisture content at day 40 (SM40) to have a greater predictive value (8.25% IncMSE) of E. coli survival than the combination of year and mulch type (5.5% IncMSE). The mean SM40 value for 2018 mulched poultry litter-amended plots (26.2%) was significantly greater than every other mulched or nonmulched amended plot in 2018 and 2019 except three plot types: 2018 mulched heat-treated poultry pellets (19.7%), 2019 mulched composted poultry litter (18.9%), and 2019 mulched heat-treated poultry pellets (22%) (Table 2). Of these four combinations of year, amendment, and mulch which had statistically similar SM40 values, the sigmoidal model estimated the greatest decline in E. coli levels between 31 and 70 days occurred in the 2019 mulched composted poultry litter (2.35 log CFU/g [dry weight]), followed by 2019 mulched heat-treated poultry pellets (1.83 log CFU/g [dry weight]), 2018 mulched heat-treated poultry pellets (0.91 log CFU/g [dry weight]), and 2018 mulched poultry litter-amended plots (0.002 log CFU/g [dry weight]).
Cumulative rainfall (inches) for every 10-day interval between days 0 and 120 in years 2018 (dashed line) and 2019 (solid line).
Mean soil moisture values on day 40 for each combination of year, amendment type, and mulch type
Levels of soluble carbon (dry weight basis, total water extractable carbon) measured on day 40 (DWBTWEC 40) may have been affected by the differing rainfall amounts and patterns along with soil moisture content, preceding day 40. In 2018, the values of soluble carbon on day 40 in plots for three combinations of year, amendment, and mulch—2018 mulched heat-treated poultry pellets (217 mg/kg), 2018 mulched poultry litter (202 mg/kg), and 2018 not mulched heat-treated poultry pellets (133 mg/kg)—were greater than all other combinations in the same year (Table 3). As with previously evaluated predictors, the role of soluble carbon in extending E. coli survival in amended soils may be dependent on multiple temporal and agricultural factors. In 2018, both mulched heat-treated poultry pellets (106 days) and mulched poultry litter (99.5 days) had high dpi4log values, potentially indicating that higher soluble carbon levels may extend E. coli survival in plots. However, in 2018, not-mulched plots containing heat-treated poultry pellets had a high soluble carbon level at day 40 but lower dpi4log values (28.5 days), potentially indicating that other conditions in the soil may affect E. coli survival. In 2018, plots containing mulched composted poultry litter (71.4 mg/kg) and not-mulched unamended plots (56.6 mg/kg) had significantly lower soluble carbon values at day 40 than plots containing mulched poultry litter and heat-treated poultry pellets but had statistically similar dpi4log values. In 2019, the soluble carbon at day 40 value (175 mg/kg) of mulched plots containing heat-treated poultry pellets was significantly greater than those of mulched composted poultry litter plots (57.9 mg/kg), not-mulched composted poultry litter plots (38.2 mg/kg), or mulched unamended (45.6 mg/kg) plots.
Mean soluble carbon values on day 40 for each combination of year, amendment type, and mulch type
Nitrate levels on day 30 (NO3-N30) also had a high %IncMSE value (9.25%) in the RF model in predicting dpi4log values for E. coli. In general, NO3-N30 values were greater in 2018 than in 2019 for most combinations of year, amendment, and mulch. As shown in Table 4, in 2018, NO3-N30 values for plots containing mulched poultry litter (520.6 mg/kg) and unamended (577.2 mg/kg) plots had significantly higher levels of nitrate than not-mulched composted poultry litter plots (26.8 mg/kg) and not-mulched heat-treated poultry pellets (15.9 mg/kg).
Mean nitrate content on day 30 for each combination of year, amendment type, and mulch type
Transfer of E. coli to cucumbers from BSAAO.After the application of E. coli on day 0 to plots, cucumber seedlings were planted on day 0 or on day 5. Upon maturation, cucumbers (160 to 180) were harvested between 54 and 74 days postinoculation (dpi) in both 2018 and 2019. In 2018, E. coli were recovered from 97.5% (156/160) of cucumbers tested, with levels ranging from 1.23 to 6.03 log most probable number (MPN)/cucumber (Fig. 3); in 2019, 35% (63/180) of cucumbers examined contained E. coli, with levels ranging from 0.01 to 2.63 log MPN/cucumber.
E. coli populations (log10 MPN/cucumber) on cucumbers harvested from each combination of amendment (composted poultry litter [CPL], heat-treated poultry pellet [HTTP], poultry litter [PL], or no biological amendment [UN]) and mulch type (black plastic [M] or no mulch [NoM]) within each year (2018 or 2019). The bottom and top of the boxes represent the first and the third quartiles, respectively, whereas the bars inside the whiskers represent the median values. The upper and lower ends of the whiskers represent the largest and smallest values, and open circles represent individual log10 MPN per cucumber values.
As with dpi4log values for E. coli in soil, a random forest model was used to identify variables which predicted the transfer of E. coli to cucumbers (log MPN E. coli/cucumber). The factors year (9.75%), combination of year and amendment (9.5%), and combination of year, amendment type, and mulch (9.25%) had essentially equivalent %IncMSE values (Fig. 1). The predictive value of the dpi4log (4% IncMSE) was lower than the aforementioned combinations but still useful to our study. Although soil moisture at day 100 (SM100) was also relatively highly ranked as a predictor (6.5% IncMSE), its practical value in predicting transfer of E. coli to cucumbers was limited, since cucumbers were harvested before 100 dpi of the study. As with predicting dpi4log values, the differing rainfall amounts and patterns in 2018 and 2019 most likely affected the transfer of E. coli to cucumbers (Fig. 2 and 3).
The RF model accounted for approximately 63% of the variance in the model to predict transfer of E. coli to cucumbers, lower than the 98% of variance explained by the RF model used to predict dpi4log values, indicating that overall transfer of E. coli from soils to cucumbers may be less predictable than E. coli survival in soils.
Regression decision trees were compiled based on dpi4log values from results using 70% of the data (cucumbers) in the “training” set (Fig. 4). Year was highly valued as a predictor for E. coli transfer to cucumbers (log MPN/cucumber) and, in tandem with dpi4log values, provide distinctive factors to evaluate E. coli transfer to cucumbers. Further node analysis of a selected regression tree model was based on dpi4log values. This regression tree showed that 35% (27/78) of cucumbers containing E. coli were from test plots with dpi4log values of less than 47 days, while the remaining 65% (51/78) originated from plots with dpi4log of >47 days (Fig. 4). For the 35% of cucumbers originating from plots with a dpi4log of <47 days, 18% (14/78) of cucumbers were from plots with dpi4log values less than 35 days (mean, 0.017 log MPN/cucumber): 2019 mulched and not-mulched unamended plots and 2019 not-mulched poultry litter plots. The remaining 17% (13/78) of cucumbers were grown in plots with dpi4log values between 35 and 47 days: 2018 not-mulched heat-treated poultry pellets plots and 2019 not-mulched composted poultry litter (1.6 log MPN/cucumber). These results suggest that plots with smaller dpi4log values transferred lower levels of E. coli from soils to cucumbers than plots with larger dpi4log values. For the 65% (51/78) of cucumbers originating from plots with dpi4log values of >47 days, 15% (12/78) were from 2018 mulched plots containing heat-treated poultry pellets which had a dpi4log value of 106 days (2.8 log MPN/cucumber). Of the 50% (39/78) of cucumbers originating from plots with dpi4log values between 47 and 100 days, 26% (20/78) of cucumbers were grown in 2018 not-mulched plots containing poultry litter, 2019 not-mulched plots containing heat-treated poultry pellets, 2019 mulched plots containing heat-treated poultry pellets, and 2019 mulched plots containing poultry litter, with dpi4log values between 50 and 93 days (mean E. coli of 1.1 log MPN/cucumber). The 13% (10/78) of cucumbers originating from plots which had dpi4log values greater than 93 days were from 2018 plots containing mulched composted poultry litter, mulched poultry litter, and not-mulched unamended plots (mean E. coli level, 2.2 log MPN/cucumber). The 12% (9/78) of cucumbers originating from plots with dpi4log values less than 50 days originated from 2018 not-mulched plots containing composted poultry litter and unamended mulched plots (2.9 log MPN/cucumber).
Regression for log10 MPN per cucumber values using dpi4log values (days) as the sole predictor in the decision tree. In boxes below decision nodes, log10 MPN per cucumber values are listed on top, followed by number of total cucumbers (n = 78) and percentage of cucumbers in the decision path.
Sporadic Salmonella species detection in soil and on cucumbers.In 2018, Salmonella spp. were recovered on day 60 from soil samples collected from mulched plots containing heat-treated poultry pellets or not-mulched unamended plots. Although Salmonella isolates were recovered from these specific plots in 2018, Salmonella spp. were not recovered from cucumbers grown in these same plots. However, Salmonella spp. were recovered from cucumbers grown in the following 2018 plots: mulched plots containing composted poultry litter and mulched and not-mulched plots containing poultry litter. In 2019, Salmonella was recovered sporadically from specific plots: not mulched containing heat-treated poultry pellets on day 10; mulched plots containing poultry litter on days 60 and 70; mulched plots containing heat-treated poultry pellets on day 90; and mulched unamended plots on day 110. No Salmonella spp. were recovered from plots amended with composted poultry litter. No Salmonella spp. were recovered from cucumbers grown in 2019.
DISCUSSION
Our employment of the RF regression to extract factors or variables that affected E. coli survival in amended soils was an opportunity to identify practical and agricultural predictors which are commonly utilized or relatively easily measured in the course of soil testing and monitoring activities. Our study prioritized predictors measured at times early in the 120-day study which would allow farmers or other stakeholders to make informed decisions on the potential survival of enteric pathogens in soils and transfer to fruits and vegetables.
E. coli survival in soils is affected by a combination of amendment type, rainfall pattern, and soil moisture content.Previous work has shown that the inactivation (decline) rate (kmax) of Salmonella enterica subsp. enterica serovar Newport in sandy soils amended with heat-treated poultry pellets and planted with spinach in a controlled environment was approximately 3-fold lower (0.10 log CFU/g [dry weight]/day) than in soils not containing heat-treated poultry pellets (0.35 log CFU/g [dry weight]/day) (21). Those findings agree with those presented here showing that soils containing heat-treated poultry pellets can support longer durations of enteric bacterial survival than soils containing no biological or poultry litter-based amendment.
While total cumulative rainfall over each 10-day interval was not a strong predictor of E. coli survival in the RF model, the increase in rainfall between 31 and 40 days (4.17 in.) compared to that in the preceding interval (21 to 30 days, 0.03 in.) in 2018 led to the increase in soil moisture content on day 40 (SM40) for specific combinations of year, amendment, and mulch.
The longer survival of E. coli in the 2018 mulched composted poultry litter plots and, to a lesser extent, in 2018 mulched plots containing poultry litter than in the 2019 mulched plots containing composted poultry or heat-treated poultry pellets may have been due to higher soil moisture values from the rainfall occurring between 31 and 40 days in 2018. That survival trend was maintained by rainfall events which occurred in 2018 between days 41 to 50, 51 to 60, and 61 to 70, resulting in 4.28 in., 3.83 in., and 3.16 in., respectively. The decline of E. coli levels observed in 2019 between 31 and 70 days in 2019 CPL M and HTTP M may have also been affected by the lack of rainfall during this period compared to that in 2018. In 2019, the largest 10-day cumulative rainfall amount observed was between 0 and 10 dpi (2.84 in.) and was greater than the amount observed in 2018 (0.31 in.) in the same period of time. However, the lack of rainfall exceeding 2.84 in. after day 10 in a 10-day period between days 11 and 120 of the study may have resulted in the overall shorter survival durations (smaller dpi4log values) of E. coli in 2019 than in 2018. We hypothesize that the higher soil moisture levels observed at day 40 in 2018 led to nutrient, chemical, and microbial community changes which slowed the decline of E. coli in mulched plots containing biological soil amendments. Other studies evaluating mulched soils (no BSAAO added) used to grow lettuce reported E. coli levels that were significantly (P < 0.05) greater than on bare (not mulched) soils in one of two spring seasons evaluated; however, bacterial fecal coliform counts were consistently greater in mulched soils than in bare soils (22). Those authors postulated that mulching may have increased moisture and temperatures of soils, increasing the likelihood of E. coli and fecal coliform survival. In previous growth chamber studies where soil moisture was relatively constant throughout the duration of the study, higher soil moisture levels slowed Salmonella Newport declines in soils containing heat-treated poultry pellets compared to that in unamended soils (21). Given sufficient nutrients and moisture, E. coli was observed to grow by 3 to 4 log CFU/ml in soil/composted poultry litter extracts within 24 h with limited microbial competition (10). Similarly, levels of Salmonella Newport were able to increase by 3 log CFU/ml in the presence of indigenous soil microorganisms in extracts of soil amended with heat-treated poultry pellets (23). Increasing soil moisture content through surface irrigation of soils contained in large pots in a greenhouse significantly increased E. coli levels in soils containing poultry litter compared to that in soils with no biological amendments or chemical fertilizer or in soils amended with horse manure (5). In that study, E. coli levels increased by larger amounts in week 2 than in weeks 1, 4, and 8, indicating E. coli can be influenced by periods of intermittent growth during their overall decline in soils containing BSAAO. Salmonella Newport levels in soils changed in response to the combination of time, amendment, and irrigation events when soils amended with poultry litter, heat-treated poultry pellets, or urea were irrigated on a weekly basis for 29 days (13). In that study, the mean increase in S. Newport levels across all amended soils after irrigation during week 1 was significantly greater than the mean increases observed at weeks 2, 4, and 8. Soil moisture percentages for all three amended soil types were statistically similar after irrigation events in week 3 and 4 but supported different levels of S. Newport survival by week 4, with heat-treated poultry pellet- and urea-amended soils supporting significantly greater S. Newport levels than poultry litter-amended soils. Those results are analogous to results observed in our present study in 2018 and 2019, in which statistically similar relatively large soil moisture values on day 40 (SM40) were observed among several different amendment treatments. Findings from both the previous study (13) and our present study indicate that use of soil moisture as the sole determinant of E. coli or Salmonella survival in amended soils without including specific BSAAO effects may not appropriately predict pathogen survival in soils.
Effect of temporal variation and rainfall patterns on E. coli levels in soils.The effect of rainfall patterns and temporal variation on the survival of E. coli in soils containing BSAAO has been noted in previous studies. In a large three-site experiment involving 12 separate field trials in different seasons examining the survival of nonpathogenic E. coli (including E. coli TVS 355) in soils containing various untreated BSAAOs, the year of the trial was estimated to account for 66% of the variance of E. coli survival in a spatiotemporal model (11). In that study, in one specific year (2013), the survival durations (days) of both nonpathogenic and attenuated O157 E. coli in studies started in either spring or fall were significantly longer than those of studies started in 2011 or 2012. Other studies have shown variable survival durations of Shiga-toxigenic E. coli (STEC) and Salmonella enterica in bovine and swine manure-amended soils in spring seasons of different years in the same location (24), influenced by changes in soil moisture and temperature between the years.
Other workers analyzing the same large data set from Sharma et al. (11) showed that soil moisture was highly predictive of E. coli levels as well. A number of factors related to rainfall were also relatively highly predictive of E. coli levels in soils containing BSAAO, including the number of days or rainfall since the previous day of analysis for E. coli levels (sampling day), average daily precipitation since the previous sampling day, average daily precipitation over the week prior to the sampling day, and days of rain over the week prior to the sampling day (25). Our work here analyzed cumulative rainfall amounts over 10 days and changes in rainfall amounts over 10 days, both of which had low predictive values for dpi4log values for E. coli in the RF regression model. However, the combination of the effects of large amounts of rainfall between days 30 and 40 increasing soil moisture in soils containing specific amendments may have affected levels of other factors which had a high predictive value for dpi4log values for E. coli, including soil moisture percentage at day 40 (SM40; 8.25% IncMSE) and nitrate levels at day 30 (NO3-N30; 9.25% IncMSE).
Nutrient content of soil and E. coli survival.In general, plots containing poultry litter and heat-treated poultry pellets had larger soluble carbon (dry weight basis, total water extractable carbon) values at day 40 than unamended or composted poultry litter-amended plots. The microorganisms involved in the composting process of PL may have utilized the soluble carbon, while the chemical fertilizer used in the unamended plots may not have originally contained any soluble carbon, accounting for the lower values of soluble carbon at day 40 in these plots. Previous authors have stated the difficulty in separating the effect soil nutrients and the microbial community (10) but showed carbon availability in soils amended with either poultry litter or composted poultry litter was not a limiting factor for microbial growth or survival for E. coli or other microorganisms. Those results also indicate that lower soluble carbon levels in specific plots (such as those plots containing composted poultry litter or unamended) may be sufficient for E. coli survival. Therefore, it is unclear if soluble carbon values directly or indirectly impact E. coli survival in soils. The longer survival durations of E. coli observed in 2018 mulched plots containing composted poultry litter and in not-mulched unamended plots, both of which had relatively low soluble carbon values on day 40, indicate the potential for other factors, such as amendment-specific microbial community changes, to promote E. coli survival.
Plots with lower nitrate values on day 30 (NO3-N30) supported shorter E. coli survival durations (smaller dpi4log values). In 2018, not-mulched plots containing composted poultry litter had a dpi4log value of 49 days, while 2018 not-mulched plots containing heat-treated poultry pellets had a dpi4log value of 28.5 days; 2018 plots which were mulched and contained poultry litter had a dpi4log value of 99.5 days but were not significantly different from the dpi4log value (49 days) of 2018 mulched unamended plots. The trend of plots with lower NO3-N30 levels supporting smaller dpi4log values was also observed in 2019 with nitrate levels in mulched (242.9 mg/kg) or not-mulched (120.8 mg/kg) plots containing heat-treated poultry pellets. As with those in 2018, the 2019 plots containing lower nitrate levels at day 30 had significantly smaller dpi4log values (2019 not-mulched plots with poultry litter, 46 days; 2019 not-mulched unamended plots, 39.5 days) than plots with significantly greater nitrate levels at day 30 (2019 mulched plots containing heat-treated poultry pellets, 65.5 days; 2019 not-mulched plots containing heat-treated poultry pellets, 85.5 days).
Previous work has shown that moist soil supported the conversion of ammoniacal nitrogen (NH3-N) to NO3 (26). Other studies have shown that poultry litter compost or heat-treated poultry pellets provided greater levels of NH3-N to support either E. coli or Salmonella Newport survival in soils, respectively (10, 21). In the 2018 portion of the study, where more rainfall was observed than in 2019, much of the ammoniacal nitrogen would have been converted to nitrate, increasing nitrate levels and potentially providing additional nutrients for E. coli survival (27) or for other soil microbes beneficial to E. coli survival. This potential scenario may have occurred in wet soils under anaerobic conditions, promoting the bacterial utilization of nitrate (28).
Effect of biological soil amendment type on E. coli survival in amended soils.In a previous study (11), poultry litter applied to surfaces of soils (not tilled) supported longer survival durations of nonpathogenic or O157 E. coli in 60% of the comparisons, while soils containing no biological amendment or chemical fertilizer supported the longest survival durations in 20% of comparisons. The observation that soils containing no additional nutrients supported E. coli survival for equivalent or longer durations than soils containing untreated BSAAOs shows that temporal and weather conditions can potentially be as influential as biological soil amendments.
Several findings from previous studies (11, 21) support the results of our present study. Added nitrogen and soluble carbon (water extractable carbon) were shown to significantly influence Salmonella Newport survival (21) in soils amended with heat-treated poultry pellets. Similarly, in the present study, the poultry litter-based amendments (composted poultry litter, heat-treated poultry pellets, and poultry litter) may have provided nutrients to extend the survival of E. coli in soils. An examination of survival durations of a multistrain inoculum of nonpathogenic E. coli (including strain TVS 355, as used in the present study) showed that soils amended with composted poultry litter supported longer survival durations and slower declines of E. coli levels than soils amended with composted dairy manure, dairy manure-based compost (vermicomposted), compost from food scraps, or left unamended (no biological or chemical fertilizer) (10). Those authors postulated that E. coli survival in plots containing composted poultry litter was due to additional nitrogen and changes in the microbial community that were difficult to decouple. Our present study also shows the ability of chemically fertilized soils (unamended in our study) to support E. coli survival for durations similar to those supported by the biological amendments, as shown by statistically similar dpi4log values for the 2018 plots that were not mulched and unamended and those containing mulched heat-treated poultry pellets or mulched poultry litter.
Soils with different nutrient contents supporting equivalent durations of E. coli survival may point to the different routes of survival that enteric bacteria can utilize. Previous studies examining survival of E. coli in soils amended with different composted feedstocks revealed that the microbial community present in soils amended with poultry litter compost was distinct from that in either unamended soils or soils containing dairy manure windrow or vermicompost (10), which may impact the survival duration of enteric pathogens in soil. In that study, soils containing poultry litter composts had a greater relative abundance of Sphingobacteriia (Bacteroidetes) and Acidobacteria (orders RB41 or iii1-15) along with a greater relative abundance of the fungal taxon Ascobolus (Ascomycota).
Transfer of E. coli to cucumbers.Prior work suggests that if pathogens are present in soils amended with manure, compost, or heat-treated amendments for long durations, they can be transferred to fruit and vegetable surfaces through dry manure dust (29, 30). Oni et al. (29) showed that Salmonella populations transferred from manure dust particles to spinach leaves declined more slowly than Salmonella cells transferred from water to leaves. Islam et al. (30) showed the transfer of E. coli O157:H7 to lettuce and parsley leaves from contaminated soil amendments and irrigation water. In the present study, cucumbers originating from plots which supported longer survival durations of E. coli (year 2018) had comparatively higher levels of E. coli on harvested cucumbers than plots from 2019 where survival durations were shorter.
These results indicated that bacterial transfer was strongly but not completely influenced by E. coli survival duration in the soil. Micallef et al. (31) evaluated dispersion of nonpathogenic E. coli populations to cucumbers and their leaves grown on different mulches under horizontal and vertical production systems over 2 years. Similar to our present study, those authors found that transfer of E. coli to the fruit varied between years. In that study, the production system did not affect E. coli levels on cucumbers (0.6 log CFU/cucumber), even if E. coli was present in low numbers in soil. These results are in agreement with our present study showing that dpi4log, in conjunction with year, can be used as a predictor in determining bacterial survival in soil and transmission to cucumbers. In the present study, longer survival durations of E. coli were recorded in amended soils in 2018 than in 2019. In the present study, a distinct pattern was not shown for a specific combination of amendment and mulch supporting higher levels of E. coli transfer to cucumbers, suggesting that other extrinsic factors besides amendment type may influence transfer of E. coli to fruit.
Several extrinsic factors may contribute to the transfer to produce from soils or other matrices. For example, air or soil temperatures, wind, rainfall, and rainfall-induced splash or runoff events can lead to long-term persistence of microbial pathogens in soil and on plant surfaces (10, 11, 21, 22, 31). Increased rainfall during the growing season can influence soil moisture content, affecting the microbial load and survival in soil and transfer to edible portion of the crop. An increase of 0 to 1 log CFU in bacterial populations on lettuce surfaces grown on different mulches and in different seasons after a rainfall event (20 mm) was observed (22). Rainfall events may also induce a change in the microbial community on produce surfaces, attributed to changes in the soil microbial community (32). In the present study, cumulative rainfall in any 10-day interval was not identified directly as a prominent predictor of the transfer of E. coli to cucumbers. Indirectly, the more frequent rainfall events in 2018 might have led to a higher likelihood of transfer of E. coli from soils to cucumbers due to the probable increased frequency of rain splash events compared to that in 2019. Specifically, the cumulative rainfall amounts recorded between days 51 to 60 (3.83 in.) and 61 to 70 (3.16 in.) in 2018 when cucumbers were harvested (days 54, 60, 69, and 74) may have promoted E. coli transfer from soils to cucumbers through rain splash, as has been reported in previous studies. Cumulative rainfall between days 51 to 60 (0.17 in.) and days 61 to 70 (1.65 in) in 2019 was lower than in 2018, potentially decreasing the number of transfer events of E. coli to cucumbers. A previous study using simulated rainfall showed transfer of approximately 2 log CFU/g of S. enterica serovar Typhimurium onto tomato plants when a rainfall event lasted between 0 and 10 min (33). Shah et al. (21) mimicked a splash event by performing manual transfer of heat-treated poultry pellet-amended or unamended soil contaminated with Salmonella Newport onto spinach leaves. Higher S. Newport populations (ca. 2 log MPN/plant) were recorded on spinach leaves from plants grown in amended soil than on those in unamended soil (−0.2 log MPN/plant), suggesting that the longer survival duration of S. Newport in amended soils may be a major factor in the transfer to edible portions of crops.
Lack of Salmonella species recovery from cucumbers.No Salmonella spp. were recovered from any cucumbers grown in any type of amended plots in 2019. In the present study, Salmonella spp. recovered from soils were thought to be introduced from wildlife observed around the farm (deer and birds), and were not inoculated to soils as for E. coli TV 355. Previous work has shown that soil moisture may play an important role in the survival of Salmonella spp. in soils. A correlation between the rate of Salmonella Newport decline and the soil moisture content in heat-treated poultry pellet-amended and unamended soils in a growth chamber study was observed (23). A 20% increase in soil moisture levels after rainfall events increased Salmonella levels in soil amended with biosolids (34). Irrigation events were shown to increase inoculated Salmonella levels in heat-treated poultry pellets and raw poultry litter amendments without soil (13). Salmonella can grow in poultry litter-based soil extracts (13, 23). S. Newport populations increased by 4 to 5 log CFU/ml in 96 h in soil extracts containing heat-treated poultry pellets, indicating that sufficient nutrients and moisture can support relatively rapid Salmonella growth between the 10-day sampling intervals used in our present study.
In the present study, rainfall events before day 60 (2018) and before day 10 (2019) led to higher moisture levels in soils containing unamended plots or those containing heat-treated poultry pellets, indicating that these plots may have provided sufficient nutrients for Salmonella to either grow from levels below the detection limit of our Salmonella assay to detectable levels or to be transferred to the plots from another environmental source. The limited amount of nutrients such as nitrogen, carbon, and potassium in composted poultry litter may not have supported the survival of Salmonella spp. during the composting process (35), subsequently making potential nutrients for Salmonella growth unavailable in composted poultry litter plots in the present study.
Previous studies examining the survival of enteric pathogens in soils or soils containing BSAAO have focused on specific factors and their effect on E. coli survival; our study identifies specific time points and associates specific levels of physical and chemical factors to emphasize the fluctuation in these factors associated with E. coli survival. Results presented here show that fluctuations in specific soil properties, potentially induced by rainfall increasing soil moisture content, affect E. coli levels over short periods of time but with relatively long-lasting effects regarding the transfer of E. coli to the surface of fruits and vegetables. Overall, our work shows that BSAAO-amended soils can support survival of E. coli and that the duration can be extended or shortened by temporal and weather effects on soil properties.
MATERIALS AND METHODS
Litter and plots.Three types of poultry litter-based soil amendments were used: poultry litter total cleanout (PL), composted poultry litter (CPL; obtained from the University of Delaware, Georgetown, DE), and heat-treated poultry litter pellets (HTPP; Replenish 3-4-3, New Country Organics, Waynesboro, VA). The unamended plots were fertilized with a commercial inorganic fertilizer (10 N 10 P 10 K; Quaker Gap Fertilizer, Four Oaks, NC) purchased at Home Depot (Newark, DE). The field plot layout for both years is depicted in Fig. 5. Amounts of amendments added to the soil were 6 lb PL or 4 lb CPL and HTPP per 3-m2 plot to achieve 125 lb/acre soil nitrogen content, as per Mid-Atlantic commercial vegetable production recommendations.
Field plot layout in 2018 and 2019 to examine the survival of E. coli TVS 355 in soils containing BSAAO and their transfer to cucumbers grown in these plots. Plots (rectangles) in two rows were not covered (no plastic mulch, white), while in two other rows, plots were covered with black plastic mulch (plastic mulch, gray). Plots (1-m by 3-m rectangles) were either unamended (UN, chemical fertilizer) or amended with composted poultry litter (CPL), heat-treated poultry pellets (HTPP), or poultry litter (PL). Within each row, plots were separated by 3 m, and each row was separated from every other row by 3 m. Plots at the end of each of the four rows represented an uninoculated plot to determine if runoff containing inoculated E. coli was contaminating these plots.
Bacterial culture preparation.Rifampicin-resistant nonpathogenic E. coli strain TVS 355 was applied to plots in this study (32). The culture was revived from frozen stocks stored at −80°C on MacConkey agar supplemented with 80 μg/ml rifampicin (MACR [Remel Inc., Lenexa, KS], rifampicin [Fisher Scientific, Nazareth, PA]) and incubated at 37°C for 24 h. One day prior to the experiment, three to five colonies from a MACR plate were inoculated into 200 ml tryptic soy broth with 80 μg/ml rifampicin (TSBR; Difco, BD, Sparks, MD) and incubated at 37°C for 18 h. On the day of application to soil plots, the overnight culture was centrifuged (2,000 × g for 10 min), and cells were resuspended in sterile phosphate-buffered saline (PBS; pH 7.4) (Fisher Scientific). The suspension was diluted in PBS to obtain an inoculum level of 106 CFU/ml for application to plots. The E. coli inoculum was poured into a 12-liter battery-powered backpack sprayer (H.D. Hudson Manufacturing Company, Chicago, IL) immediately prior to spray inoculation of the experimental plots.
Persistence of E. coli in field plots containing different biological soil amendments.The study was conducted during the summer months (June to September) of 2018 and 2019. Each year, a total of 20 field plots (3 m2; four plots per treatment) were prepared in a complete randomized design in quadruplicates with a 3-m buffer gap between plots. Soils were classified as coarse, mixed, well-drained Greenwich loam soil. Four plots containing each biological soil amendment (PL, CPL, HTPP, or UN), were constructed. Two plots containing each amendment type were covered with mulch (M), 1-mm thick black plastic (Nolt Supply, Leola, PA), while the remaining plots containing each amendment type were left uncovered (not mulched [NoM]). Amendments or inorganic fertilizer was applied on the soil surface without tilling. Each 3-m2 plot received 1 liter of E. coli TVS 355 (ca. 6 log CFU/ml) sprayed close to the ground with a backpack sprayer to minimize wind drift, resulting in 8.5 log CFU/m2. Soil samples were collected every 10 days up to 120 days postinoculation (dpi) from each plot using sterile plastic spoons (Walgreens, DE) and placed in a sterile 24-oz Whirl-Pak bag (Nasco, Fort Atkinson, WI). Soil from the top 5 cm of the plot was collected with a sterile scoop. Soil from different locations within each plot was collected as composite soil samples (150 g). The composite soil samples were divided into three separate 30-g soil samples (i) for quantification of inoculated E. coli populations, (ii) for determination of the presence of Salmonella, and (iii) for physical and chemical soil properties. One of the 30-g soil samples (from the composite soil sample) in the sterile Whirl-Pak bags was hand massaged for 2 min before being diluted in 120 ml of sterile TSBR (1:5 dilution) and mixed to quantify E. coli populations. Sample suspensions were serially diluted in 0.1% buffered peptone water (BPW; Acumedia, Neogen Corp., MI) and plated (0.1 ml, in duplicates) on MACR. When E. coli levels were low enough to be quantified without serial dilution, 1 ml of the suspension was plated on 4 MACR plates (0.25 ml/plate). CFU of E. coli were counted after 24 h of incubation at 37°C. When colony counts fell below the level of enumeration (0.7 log CFU/g), E. coli populations were quantified by following a modified most probable number (MPN) method (32). Briefly, homogenized soil samples were diluted 1:1 in double strength (2×) TSBR, followed by serial dilution in 1× TSBR, in a 48-well deep-well plate (VWR, Radnor, PA), with four replicate dilutions, and incubated at 37°C for 24 h. The next day, 5-μl aliquots from each well were streaked individually onto MACR and incubated for 24 h at 37°C. The presence of E. coli in each replicate well in each streak on MACR was then evaluated.
Transfer of E. coli on cucumbers from soils containing BSAAO.At 0 or 5 days after E. coli application to the plots, 3-week-old cucumber (Supremo; Seedway, NY) seedlings/plants were transplanted into the plot. Each plot contained 14 plants in 2 rows (7 per row), with 0.5 m between rows and 0.2 m separating each plant in a row. Upon maturation of cucumbers, i.e., between days 54 and 74 in both trials, a total of 160 cucumbers (16 per treatment) were harvested in year 1 (on days 54, 60, 69, and 74) and 168 cucumbers (18 per treatment) harvested in year 2 (on days 54, 59, and 64) to determine E. coli populations on cucumbers. Each whole cucumber was suspended in 60 ml of sterile TSBR and hand massaged for 2 min, with the resulting suspension used for the previously described MPN assay to quantify E. coli populations.
Detection of Salmonella spp.Samples (soil or cucumber) were also analyzed for the presence or absence of Salmonella spp. as described by Davies et al. (36). Thirty grams of soil or a whole cucumber was diluted 1:5 in 120 ml of sterile universal preenrichment broth (UPB; Difco, BD, Sparks, MD) in Whirl-Pak bag and incubated at 37°C for 24 h. That suspension was then transferred (1:5 dilution) to tetrathionate broth (TET; Remel Inc., Lenexa, KS) and incubated (42°C for 24 h), followed by another transfer (1:5 dilution) to Rappaport Vassiliadis broth (RV; Remel Inc.) and incubation for 24 h at 42°C. After incubation, 10 μl was streaked on xylose-lysine-Tergitol 4 (XLT4) agar (Remel Inc.) using a sterile loop (Fisher Scientific, Nazareth, PA) and incubated for 48 h at 37°C. Plates were observed for dark black-centered colonies (presumptive Salmonella) and selected to prepare frozen stocks. Isolates were stored at −80°C until further use. Presumptive Salmonella isolates were tested by extracting DNA and performing PCR amplification of the invA gene using forward (5′-GTGAAATTATCGCCACGTTCGGGCAA-3′) and reverse (5′-TCATCGCACCGTCAAAGGAACC-3′) primers with an amplicon product size of a 284 bp. The following reaction parameters were used: initial denaturation at 95°C for 4 min, denaturation at 95°C for 30 s, annealing at 67.2°C for 30 s, and elongation at 72°C for 60 s, repeated for 30 cycles; final extension at 72°C for 10 min (37). Another Salmonella-specific PCR assay (38) was performed to test invA-negative samples. PCR mixture was prepared in a total volume of 25 μl consisting of 2 μl of template DNA, 12 μl of PCR master mix (DreamTaq; Thermo Scientific, Lafayette, CO), and 1 μM TS-11 (5′-GTCACGGAAGAAGAGAAATCCGTACG-3′) and TS-5 (5′-GGGAGTCCAGGTTGACGGAAAATTT-3′), resulting in a 375-bp PCR product. The reaction parameters were an initial denaturation (95°C, 10 min), 40 cycles of denaturation (95°C, 20 s), annealing (60°C, 30 s), and extension (72°C, 30 s), and a final extension of 7 min at 72°C (32).
Soil extrinsic and intrinsic factors analysis.Physical and chemical properties of soil and amendments were tested by standard procedures established by the University of Delaware Soil Testing Program (Newark, DE). Soil samples (30 g of composite soil sample) collected during the trial (every 10 days up to 120 days) were evaluated as described for moisture, temperature, soluble carbon (water extractable carbon) content (39), carbon-nitrogen ratio (40, 41), nitrate (42), total nitrogen (41), and electrical conductivity (43). Soil temperature was measured using a smart temperature sensor and HOBO ONSET data logger (Onset Computer Corporation, Pocasset, MA, USA). Air temperature and total rainfall data for the duration of trials were obtained from the University of Delaware Weather Station (http://www.deos.udel.edu/data/monthly_retrieval.php) and are organized in the supplemental file “Soil temperature, air temperature, and cumulative rainfall in 2018 and 2019” available at https://doi.org/10.15482/USDA.ADC/1520517.
Statistical analysis.Statistical analyses consisted of a two-stage process. First, the replicate field plot data were used to obtain a nonlinear regression model, specific to each observed experimental condition (year, amendment, and mulch), that accurately predicted and characterized E. coli survival relative to days postinoculation (dpi). Subsequently, predictions from the nonlinear regression models were used to examine influence of year, amendment, and mulch factors on survival of E. coli.
Surviving E. coli populations, recovered every 10 days from 0 to 120 dpi, were calculated as log CFU per gram (dry weight) from each of the duplicate plots for each combination of mulch and amendment type in each year. The observed relationship between the log CFU per gram (dry weight) and the days postinoculation for each combination of mulch and amendment was accurately characterized using a nonlinear sigmoidal regression model (SAS PROC NLIN), with number of sigmoids ranging from zero (i.e., constant log CFU/g [dry weight] across dpi) to a double, triple, or quadruple sigmoid (44, 45). Subsequently, each model was used to estimate the days postinoculation by which the combination of year, amendment type, and mulch achieved a 4-log reduction (dpi4log) from the log CFU per gram (dry weight) estimated by the model at dpi 0. Examples of declines are shown in Fig. 6 and in the supplemental file (“Graph of E. coli levels over 120 days in soils under various conditions” available at https://doi.org/10.15482/USDA.ADC/1520517).
Examples of sigmoidal decline models for 4 of 24 combinations of year, amendment, and mulch plots used in this study for 2018 and 2019. In 2018, mulched plots containing heat-treated poultry pellets (HTPP) had the largest dpi4log value (days) for E. coli (A), while not-mulched composted poultry litter (CPL) plots had the smallest dpi4log value (B). In 2019, mulched plots containing poultry litter (PL) had the largest dpi4log value (C), while not-mulched CPL plots had the smallest dpi4log value (D). gdw, grams (dry weight).
To identify combinations of year, amendment, and mulch factors most important in facilitating accurate prediction of dpi4log and the transfer of E. coli to cucumbers (log MPN/cucumber), the randomForest R package (46) was used to fit two random forest (RF) models. Predictor variables specified in the RF models were (i) a factor whose unique levels were the combinations of all observed levels of the year, amendment, and mulch factors (i.e., year.amend.mulch), (ii) three different factors whose unique levels were combinations of the levels of 2 of the 3 year/amendment/mulch factors (i.e., year.amend, year.mulch, and amend.mulch), and (iii) all individual factors—year, amend, and mulch—as well as total percent carbon, total percent nitrogen, C/N ratio, nitrate concentrations (at dpi of 0, 30, 60, 90, and 120), DWBTWEC (dry weight basis, total water extractable carbon), soil moisture, soluble salts, and changes in DWBTWEC, soil moisture (SM), and soluble salts from the previous 10-day measurement (e.g., soil moisture at 20 dpi − soil moisture at 10 dpi). The random forest algorithm fit 500 individual regression tree models, each based on a different random sample of predictors (34 of 102 predictors for dpi4log and 40 of 120 predictors for log MPN/cucumber), and compiled consensus measures (e.g., %IncMSE) identifying predictors having the greatest ability to improve the RF model’s accuracy in predicting dpi4log and log MPN per cucumber. For each RF model, the same random sample of 70% (78 of 112) of the data set observations were randomly chosen to train the model, and the remaining 30% (34 of 120) of the data set was used to assess the model’s fit.
To correctly align the values of all predictor variables (listed above) used to fit the RF models, the original data set, consisting of 32 observations (2 years by 4 amendments by 2 mulches by 2 plots) needed to be restructured (i.e., consolidated). Cucumbers were sampled from each of the field plots at dpi 54, 60, 69, and 74 in 2018 and at dpi 54, 59, and 64 in 2019. Hence, the E. coli data for cucumber (log MPN/cucumber) consisted of 4 amendments by 2 mulches by 2 plots by 4 dpi (in 2018) plus 4 amendments by 2 mulches by 2 plots by 3 dpi (in 2019) for 112 total observations. The 112 values needed in the consolidated data set for each of the 120 predictor variables were populated by simply repeating the values observed at each associated plot combination of year, amendment, and mulch. Values of dpi4log estimates were specified in the consolidated data set by expanding the single estimate obtained from each sigmoidal regression model as equally spaced values by beginning at the earliest day postinoculation at which log CFU per gram (dry weight) achieved a 4-log reduction, ranging up to the next later day postinoculation at which data values were observed. For example, dpi4log was estimated by the sigmoidal model for 2018 CPL with mulch to be 95 dpi. The next latest observed day postinoculation was 100. So, the 8 estimates needed to align with the 8 observed cucumber measurements for 2018 CPL with mulch were obtained as 95 + i · (100 − 95)/7 for i = 0, …, 7 rounded to the nearest decimal place (i.e., 95, 95.7, 96.4, 97.1, 97.9, 98.6, 99.3, 100). Each of these dpi4log estimates was then randomly associated with 1 of the 8 observed cucumber observations. To prevent any potential preference of the random forest algorithm for predictors with values on a larger scale than others, all predictor variables were rescaled to a standardized range of 0 to 100 (see supplemental files “Data set used in random forest model to identify variables and factors which predict dpi4log values of E. coli in soils containing biological soil amendment of animal origin” and “Data set used in random forest model to predict transfer of E. coli from soils to cucumber fruits” available at https://doi.org/10.15482/USDA.ADC/1520517).
Predictors identified by the RF results as having the greatest importance in accurately predicting dpi4log or log MPN per cucumber, and which were easily measured early in the growing season, were further examined to understand reasons for their importance. Analysis of variance (ANOVA) and regression tree models were constructed. The selected predictors were subjected to generalized linear (47) one-way ANOVA models, using the observed 32-observation data set and conducting Sidak-adjusted pairwise means comparisons among the 16-level combinations of year, amendment, and mulch. The respective statistical distributions and link function specified in the ANOVA models for each predictor variable were gamma and log for dpi4log, beta and logit for percent soil moisture at day 40 (SM40), and negative binomial and log for soluble carbon (DWBTWEC) at day 40. Additionally, to better understand how the selected predictor variables contributed to accurately estimating dpi4log or log MPN per cucumber, an individual rpart regression tree model was fit to each of several 70%-sized training data sets, randomly sampled from the 112-observation consolidated data set used to fit the RF models.
ACKNOWLEDGMENTS
Funding for this project was provided by an interagency agreement (IAA-224-17-2023S) between the U.S. Food and Drug Administration and the U.S. Department of Agriculture, Agricultural Research Service (USDA ARS), a non-assistance cooperative agreement (NACA 58-8042-7-034, “Characterization and mitigation of bacterial pathogens in the fresh produce production and processing continuum”) between the USDA ARS and the University of Delaware, and the USDA ARS Project Plan Characterization and Mitigation of Bacterial Pathogens in the Fresh Produce Production and Processing Continuum (8042-32420-006-00-D).
We thank Scott Hopkins for preparing the plots and Quinn Riley, Kyle McCaughan, Micah Greenzweig, Eddi Blanco, Samantha Gartley, Brienna Anderson-Coughlin, and Shani Craighead at the University of Delaware for their technical assistance. We also thank several scientists from the FDA Center for Food Safety and Applied Nutrition: David T. Ingram and Michael Mahovic from the Office of Food Safety, Division of Produce Safety, Fresh Produce Branch, for logistical and technical support; Jane Van Doren, Hao Pang, Yuhuan Chen, and David Oryang from the FDA Office of Analytics and Outreach, Division of Risk and Decision Analysis, Risk Analysis Branch, for technical input.
We have no conflict of interest.
FOOTNOTES
- Received 30 September 2020.
- Accepted 5 January 2021.
- Accepted manuscript posted online 22 January 2021.
This is a work of the U.S. Government and is not subject to copyright protection in the United States. Foreign copyrights may apply.
