Distribution and Characterization of Salmonella enterica Isolates from Irrigation Ponds in the Southeastern United States

Abstract

Irrigation water has been implicated as a likely source of produce contamination by Salmonella enterica. Therefore, the distribution of S. enterica was surveyed monthly in irrigation ponds (n = 10) located within a prime agricultural region in southern Georgia and northern Florida. All ponds and 28.2% of all samples (n = 635) were positive for Salmonella, with an overall geometric mean concentration (0.26 most probable number [MPN]/liter) that was relatively low compared to prior reports for rivers in this region. Salmonella peaks were seasonal; the levels correlated with increased temperature and rainfall (P < 0.05). The numbers and occurrence were significantly higher in water (0.32 MPN/liter and 37% of samples) than in sediment (0.22 MPN/liter and 17% of samples) but did not vary with depth. Representative isolates (n = 185) from different ponds, sample types, and seasons were examined for resistance to 15 different antibiotics; most strains were resistant to streptomycin (98.9%), while 20% were multidrug resistant (MDR) for 2 to 6 antibiotics. DiversiLab repetitive extragenic palindromic-element sequence-based PCR (rep-PCR) revealed genetic diversity and showed 43 genotypes among 191 isolates, as defined by >95% similarity. The genotypes did not partition by pond, season, or sample type. Genetic similarity to known serotypes indicated Hadar, Montevideo, and Newport as the most prevalent. All ponds achieved the current safety standards for generic Escherichia coli in agricultural water, and regression modeling showed that the E. coli level was a significant predictor for the probability of Salmonella occurrence. However, persistent populations of Salmonella were widely distributed in irrigation ponds, and the associated risks for produce contamination and subsequent human exposure are unknown, supporting continued surveillance of this pathogen in agricultural settings.

INTRODUCTION

Salmonella enterica is the leading cause of bacterial food-borne illnesses and accounts for approximately 42,000 cases of infections annually in the United States (1). Traditionally, salmonellosis has been considered a zoonotic disease due to its frequent association with poultry and eggs (2, 3); however, recent outbreaks are increasingly attributed to fresh fruits and produce (4), and the number of disease cases per outbreak is sometimes greater for vegetables than for other food products (5). These outbreaks confirm that environmental transmission of Salmonella from produce can lead to human illness (6, 7). Recent investigations have focused on irrigation water as a potential environmental source of Salmonella for preharvest contamination of produce (8–11, 12). Furthermore, the static nature of some irrigation ponds may sustain persistent populations of Salmonella (13, 14).

Assessment of the microbial quality of agricultural water is now required under the proposed Produce Safety Rule (PSR) of the Food Safety Modernization Act (FSMA) in order to ensure the safety of fresh produce (15). The recommended microbial standards for the quality of agricultural water that comes into direct contact with preharvest produce (other than sprouts) requires the threshold of generic Escherichia coli to be <410 most probable number (MPN)/100 ml for any single sample, with a geometric mean of <126 MPN/100 ml for all samples. Although this policy change will provide greater assurance of food safety, it may not be sufficient to eliminate risks from pathogens, such as Salmonella, that have shown little or no correlation between the level of the pathogen and that of generic E. coli (14, 16–22).

The present study focused on irrigation ponds at mixed-produce farms located in a prime agricultural production region of Georgia that is a hot spot for environmental sources of salmonellosis. The case rates within this region were 50/100,000 in 2007, 1.5 times the national average (8, 23). A 12-month survey of river water in this area showed Salmonella in 79.2% of samples, with an increased prevalence during summer months (62%) that was positively correlated with the water temperature (r = 0.49, P < 0.05) and precipitation levels (r = 0.68, P < 0.05) (8). Even higher prevalence (96% of samples) and levels (up to 5,400 MPN/100 ml) of Salmonella were reported in surface water samples of the lower portion of the Suwannee River watershed (9). Both surveys identified serotypes that were associated with human disease in this region. However, genetic analysis of 110 strains collected from the Suwannee River study found that only a small proportion (12%) of strains showed genetic similarity to strains from clinical origins, while most isolates clustered into genogroups that were unique to strains from the Suwannee River.

The connection between Salmonella contamination of river water and human disease outbreaks is ambiguous because river water is generally not used for crop irrigation or drinking water in the Suwannee River watershed. Vegetable crops are commonly irrigated by open-source water from irrigation ponds in the region. Several recent studies have focused on a group of ponds distributed throughout this area in southern Georgia. The levels of Escherichia coli O157 (18) and Campylobacter jejuni (19) were examined, and another independent study surveyed the prevalence, antibiotic resistance, and diversity of Salmonella in the same ponds (24). The objectives of this study were to quantify the spatial and temporal distribution, antibiotic resistance, and genetic diversity of Salmonella populations in both pond water and sediment samples in the context of the utility of fecal indicators and/or other environmental parameters for prediction of Salmonella.

MATERIALS AND METHODS

Sample collection from irrigation ponds.

This study examined 10 irrigation ponds on farms located within the upper Suwannee River watershed in the southeastern United States. Ponds were selected for their diversity, characteristics, and surroundings, including size, depth, drainage area, associated vegetation and wildlife, water intake, irrigation types, and crops irrigated (Table 1). This information was obtained through a survey in September 2010. All 10 ponds were sampled monthly for a continuous 12-month period from March 2011 to February 2012. Beginning from March 2012 through December 2012, more intensive sampling efforts focused on 5 ponds (LV, MD1, NP, CC2, and SC). Water samples (n = 170, 10 liters each) were collected monthly from each pond at two locations, either at the surface or 50 cm below the surface in an area near the irrigation water intake. Two wet sediment samples (1 kg) were collected concurrently from two locations, either at the pond perimeter (n = 170) or from a benthic area under the water intake about 8 to 10 feet from the surface (n = 125), using a benthic dredge (Wildco Fieldmaster mighty grab II dredge). The collection vessels were sanitized with 10% bleach and rinsed with sterile water between samplings. All samples were stored on ice in coolers in the field and transported to the laboratory for refrigeration until analyses were conducted within 24 h of collection. Aliquots of water and sediment samples were used to inoculate enrichment cultures for microbial evaluation as described below.

TABLE 1.

Description of the 10 irrigation ponds surveyed for Salmonella within the upper Suwannee River watershed^a

Pond	Size of pond; size of drainage (acres)	Irrigation type	Bottom pipe	Cropsb	Buffer zone (% of drainage area)	Stream fed	Source(s) of added water	Wildlife survey	In-pond vegetation
BB	7; 500,000	Drip sprinkler	No	FR	No wetlands	No	Well, canals	Deer, sometimes hogs, beaver, birds, alligator	Duckweed
CC1	5; 1,000	Pivot drip	No	LG, V, FV, CO, O	Pines, hardwood (100%)	Seasonal	Pond	Raccoon, deer, beaver, alligator, birds	None
RT1	4; 40	Pivot	Yes	L, M, CO, O	Grass, shrub (30%)	Seasonal	Well	Geese, deer, birds alligator	None
RT2	2; 40	Plastic or pivot drip	Yes	L, M, CO, O	Grass (<10%)	No	None	Birds, deer	None
VH	20; 1,000	Pivot	Yes	LG, L, O	Grass (10%)	No	None	Beavers, deer, alligators, birds, fox	Water hyacinth
CC2	23; 500 (3 ponds)	Drip	Yes	LG, V, FV, O	Pines, hardwood (100%)	Seasonal	Pond	Beavers, deer, raccoon, opossum, birds	None
LV	2.5; 5	Sprinkler	No	LG, CO, V	Grass (10%)	No	Well	Deer, turtle, hogs, birds, raccoon, bobcat	None
MD1	12; 35	Pivot	Yes	LG, CO, O	Grass (<10%)	Seasonal	None	Birds, geese, hogs, raccoon, fox, squirrel	Grass
NP	22; 50	Drip	No	V, CO, M, O	Bushes, shrub (40%)	No	None	Deer, turtle, hogs, birds, raccoon, bobcat	Varies
SC	20; 100	Drip	No	V, CO	Pine (50%)	No	None	Deer, hogs, birds, raccoon, bobcat	None

^aPonds are located within the Suwannee River watershed in the region indicated in Fig. 1. The various characteristics were surveyed approximately 6 months before the study.

^bFR, fruit; FV, fruit and vegetables; LG, leafy greens; V, vegetables; M, melons; L, legumes; CO, cole crops (e.g., broccoli, Brussels sprouts, cabbage, and cauliflower); V, vegetables; O, other (e.g., peanuts, cotton, and corn).

Salmonella MPN determination.

The prevalence of Salmonella was evaluated as the percentage of positive samples from triplicate aliquots of all sample types, i.e., surface and subsurface water (500 ml) and perimeter and benthic sediments (100 g). Salmonella was enumerated by a 3-tube × 3-dilution most probable number (MPN) method as previously described (25). Briefly, triplicate enrichment cultures of lactose broth (2×; Fisher Scientific, Inc.) were inoculated with an equal volume (500, 100, or 10 ml) or weight (100, 10, or 1 g) of water or wet sediment, respectively. Cultures were incubated for 24 h at 37°C and transferred (1 ml) for selective enrichment in tetrathionate (TT) broth (9 ml) for 24 h at 37°C. Colonies were isolated from TT cultures on xylose-lysine-Tergitol 4 (XLT4; Remel, USA) agar and CHROMagar Salmonella plus (CHROMagar Microbiology, Paris, France), followed by incubation for 24 h at 37°C. Up to 5 representative isolates per sample with the typical appearance of Salmonella from each type of agar were inoculated into Luria Burtani (LB) broth for subsequent confirmation by invA gene-based PCR (26). Confirmed isolates were streaked to LB agar (LA) plates for preparation of frozen stocks in LB with 50% glycerol and stored in duplicate at −80°C in two freezers. S. enterica serovar Typhimurium LT2 (ABC Research Laboratories, Gainesville, FL) was used as the positive control for enrichment culture and for PCR. Salmonella concentrations (MPN/liter) were determined by using MPN Calculator (http://mpn-calculator.software.informer.com/). The limits of detection were 0.6 to >110 MPN/liter for water and 3.6 to >1,100 MPN/g for sediment, as determined by the range of the MPN analysis.

Indicator bacterium analyses.

As described in Gu et al. (18, 19), water and sediment samples were screened for the presence or absence of generic E. coli as an indicator of fecal contamination, using Quanti-Tray (Idexx Laboratories, Inc., Westbrook, ME). Positive samples were subsequently tested for generic E. coli by determining the MPN per 100 ml (27). Fecal coliform bacteria (CFU/100 ml) were only monitored for water samples from March 2011 to February 2012. They were assessed from two dilutions of each sample using membrane filtration techniques as previously described (28).

Weather information.

The water temperature (°C) was recorded during sampling from the irrigation ponds. The air temperature (°C) and total rainfall (mm) data of the closest weather station to each irrigation pond (10 to 40 km distance) were collected from the Georgia Automated Environmental Monitoring network (http://www.georgiaweather.net/). The mean temperatures near the ponds between samplings, as well as in the second-to-last week and last week before sampling, were calculated based on the daily weather conditions. Similarly, the total rainfall amounts between samplings and in the second-to-last week and last week before sampling were calculated.

Antibiotic susceptibility of Salmonella from irrigation ponds.

A representative subset of strains (n = 185) were selected, reflecting the distribution among different ponds and sampling time points, and assessed for antibiotic susceptibility. Resistance to 16 antibiotics was assessed as described in previous research (29); the antibiotics included amikacin (30 μg/liter), amoxicillin-clavulanic acid (30 μg/liter), ampicillin (25 μg/liter), cefoxitin (30 μg/liter), ceftriaxone (5 μg/liter), cephalothin (30 μg/liter), chloramphenicol (30 μg/liter), ciprofloxacin (2.5 μg/liter), gentamicin (10 μg/liter), imipenem (10 μg/liter), kanamycin (50 μg/liter), nalidixic acid (30 μg/liter), streptomycin (25 μg/liter), sulfamethoxazole-trimethoprim (23.75/1.25 μg/liter), and tetracycline (10 μg/liter). The antimicrobial susceptibilities of Salmonella isolates were determined by the calibrated dichotomous sensitivity (CDS) test standard (30), using susceptibility test discs according to the manufacturer's protocol (BBL Sensi-Disc). No more than five equidistant antibiotic discs were used per plate. Zones of inhibition were measured after incubation at 35 to 37°C for 18 h, and susceptibilities were classified according to the standards for Enterobacteriaceae, using E. coli ATCC 25922 as the positive control.

Rep-PCR analysis of Salmonella isolates from irrigation ponds.

Representative Salmonella strains (n = 191) from each of the 10 irrigation ponds, different time points, and sample types were screened for genetic diversity by the DiversiLab (bioMérieux) repetitive extragenic palindromic-element sequence-based PCR (rep-PCR) system, according to the manufacturer's protocol. Briefly, genomic DNA was extracted using the UltraClean microbial DNA isolation kit (MoBio Laboratories, Inc.) and quantified using the Qubit double-stranded DNA (dsDNA) broad-range (BR) assay kit (Life Technologies). In a total PCR reaction mixture volume of 25 μl, DNA extract (2 μl) was mixed with 18 μl of master mix, 2 μl of primer mix from the Salmonella rep-PCR DNA fingerprinting kit (bioMérieux), 2.5 μl of 10× buffer I with MgCl₂ (Applied Biosystems, Inc.), and 0.5 μl of AmpliTaq DNA polymerase (Applied Biosystems, Inc.). The PCR conditions included an initial denaturation at 94°C for 2 min, followed by 35 cycles of DNA amplification under the following conditions: 94°C for 30s, 50°C for 30s, and 70°C for 30s, with a final extension at 70°C for 3 min and a hold at 4°C. DNA analysis by microfluidics capillary electrophoresis using lab-on-a-chips was performed in a 2100 Bioanalyzer (Agilent Technologies), and electrophoretograms were automatically uploaded into the DiversiLab system and analyzed online.

Replicate analysis (n = 2) of 23 different Salmonella isolates was performed to evaluate the reproducibility of DiversiLab rep-PCR system results. S. enterica serovar Typhimurium strain LT2 (ABC Research, Inc.) and Vibrio vulnificus strain CMCP6 (Wright laboratory strain collection) were used as quality controls for each rep-PCR assay. The reference Salmonella strains (n = 485) included isolates from the Suwannee River (n = 110), Florida lakes (n = 14), and other environmental (n = 20) or clinical (n = 28) sources, as previously described by Rajabi et al. (9). DiversiLab rep-PCR online profiles from its Salmonella strain library (n = 313) were also used for comparison. The rep-PCR profiles of strains were compared by DiversiLab software, which uses the Pearson correlation coefficient and unweighted pair-group method using average linkages (UPGMA) to generate dendrograms (31). DiversiLab type (DTs) were defined by the DiversiLab analytical protocol and included strains that clustered with >95% similarity in banding pattern and had no obvious missing peaks, as determined by overlay of electrophoretograms (32). DTs from pond strains were evaluated using the Top Match module in the DiversiLab classification report to identify the highest values for similarity of reference strains to pond strains. Reference strains included those from the DiversiLab strain library and the publication of Rajabi et al. (9).

Statistics.

A value of zero occurrence was given to any samples under the detection limits. Upper limit values were given to any samples over the detection limits. All bacterial counts were log transformed using the formula log₁₀(MPN + 1) or log₁₀(CFU + 1). The log-transformed values were used for statistical analysis. Geometric means were calculated for presentation of spatial and temporal distributions in the text, tables, and figures. Pearson linear correlation coefficients were calculated to evaluate the correlations between Salmonella levels and occurrence, weather condition parameters (temperature and rainfall), fecal indicators (generic E. coli and fecal coliforms), or measurable pond characteristics (drainage size [acres] and buffer zone size and vegetation). Spatial and temporal effects on Salmonella mean population and occurrence were evaluated by analysis of variance (ANOVA) and logistic regression models (19). Differences in Salmonella mean populations for different categorical pond characteristics were determined using generalized linear models (GLM). Moreover, multiple linear regressions were run to determine the relationships among environmental parameters and Salmonella density, and binomial logistic regressions were done to predict the probability of Salmonella occurrence (16). For both regression analyses, Salmonella level/occurrence was the dependent variable, while all other parameters (fecal indicator bacteria, temperatures, and total rainfall amounts) were independent variables. A P value of <0.05 was set as the criterion for variable selection in regression models, and it was also considered the significance level in all statistical tests. All statistical analyses were performed using SAS (SAS release 9.3; SAS Institute, Inc., Cary, NC).

RESULTS

Spatial distribution of Salmonella from irrigation ponds.

The 10 ponds examined in this study (18, 19, 24) were located within the watersheds of the Upper Suwannee, Withlacoochee, and Ochlocknee Rivers located in the southern and southern coastal plains of Georgia and Florida (Fig. 1). All 10 ponds were sampled for 1 year from March 2011 to February 2012, and 5 ponds (CC2, LV, MD1, NP, and SC) were sampled for almost 2 years from March 2011 to December 2012. Therefore, pond effects are presented separately for both years (Table 2). Overall, Salmonella was detected in both water and sediment samples for all ponds at some time point (28.2% positive samples [n = 635]; geometric mean = 0.26 MPN/liter), with higher occurrence and levels (P < 0.05) in water (37.4% and 0.29 log MPN/liter) than in sediment samples (17.0% and 0.22 log MPN/liter). The levels and occurrence over time were highly correlated (r > 0.93) for all ponds, and no significant differences were observed between levels from surface versus subsurface water or from pond perimeter versus benthic sediment types. As sample depth did not influence Salmonella distribution, the results for water and sediment samples were pooled per pond and sampling time for further analysis. Significant differences in the levels and occurrence of Salmonella were observed among ponds for water samples collected in year 1; however, these differences were not sustained for the 5 ponds (LV, MD1, NP, SC, and CC2) sampled in year 2 (Fig. 2). For example, ponds LV and MD1 showed significant reductions in year 2 compared to year 1. However, no significant differences in occurrence and levels in sediment were seen for years 1 versus 2, and pond CC2 showed the highest levels in sediment samples for both years.

FIG 1.

Approximate locations of irrigation ponds in southern Georgia and northern Florida. Ponds (n = 10) were located within the watersheds of the upper Suwannee, Withlacoochee, and Ochlocknee Rivers between the Southern Plains and Southern Coast Plains of Georgia. Map created by Casey Harris in QGIS 2.4.0 (www.qgis.org) using data layers from the USGS National Hydrography Dataset (nhd.usgs.gov) and National Map Viewer (www.nationalmap.gov).

TABLE 2.

Levels and prevalence of bacteria in water and sediment samples at irrigation ponds in southern Georgia

Yr, ponda	Value for indicated bacteria in indicated sample type
	Salmonellab						Generic E. colic				Fecal coliforms in waterd
	Water			Sediment			Water		Sediment
	Meane MPN/liter	±SE	% positive samples	Meane MPN/liter	±SE	% positive samples	Mean MPN/100 ml	±SE	Mean MPN/100 ml	±SE	Mean CFU/100 ml	±SE
Yr 1
BB	0.21 cd	0.05	29.2	0.14 bc	0.04	10.5	4.96	0.66	3.42	0.86	9.12	2.39
CC1	0.37 bc	0.11	50.0	0.12 c	0.02	5.0	7.16	2.96	3.15	1.36	15.56	5.74
RT1	0.19 cd	0.04	29.0	0.25 abc	0.10	20.0	12.68	4.10	7.68	3.76	16.17	7.18
RT2	0.19 cd	0.05	25.0	0.15 bc	0.04	10.0	3.62	1.02	3.31	1.00	5.99	1.70
VH	1.06 a	0.41	70.8	0.32 abc	0.19	21.1	4.37	0.67	2.93	1.06	8.41	3.02
CC2	0.29 cd	0.08	41.7	0.49 a	0.26	35.0	7.86	2.03	6.03	1.96	14.53	4.63
LV	0.82 ab	0.26	70.8	0.24 abc	0.10	21.1	7.43	1.64	2.73	0.65	12.41	3.60
MD1	0.73 ab	0.27	58.3	0.40 ab	0.24	25.0	14.95	7.10	5.50	3.06	16.26	7.55
NP	0.13 d	0.02	12.5	0.13 c	0.03	5.3	2.93	0.57	4.11	1.71	3.81	0.86
SC	0.19 cd	0.04	29.2	0.12 c	0.02	5.3	5.09	1.44	2.79	0.71	10.40	3.64
Total	0.32	0.03	41.7	0.21	0.03	15.9	6.26	0.60	3.94	0.47	10.30	1.14
Yr 2
CC2	0.51 a	0.23	50.0	1.10 a	0.66	50.0	9.34	3.25	32.48	19.28	NTf
LV	0.21 ab	0.06	25.0	0.14 b	0.03	10.0	3.83	0.93	2.17	0.83	NT
MD1	0.28 ab	0.10	35.0	0.17 b	0.05	15.0	4.89	1.61	1.65	0.30	NT
NP	0.17 b	0.04	25.0	0.18 b	0.08	10.0	3.07	0.79	5.47	2.25	NT
SC	0.22 ab	0.07	25.0	0.12 b	0.03	5.0	3.01	0.95	4.61	1.65	NT
Total	0.26	0.04	32.0	0.23	0.04	18.0	6.26	0.60	4.94	1.02	NT
Total for both yrs	0.29	0.04	37.4	0.22	0.04	17.0	6.26	0.6	4.44	0.75	NA

^aPonds are described in Table 1. Ponds BB, CC1, RT1, RT2, and VH were sampled from March 2011 to February 2012 only in year 1, while ponds CC2, LV, MD1, NP, and SC were sampled from March 2011 to December 2012 through year 2.

^bSalmonella geometric mean values for MPN/liter with standard errors and the rates of prevalence (% positive) were determined as described in the text for water and sediment samples.

^cGeneric E. coli geometric mean values for MPN/100 ml with standard errors and the rates of prevalence (% positive) were determined as described in the text for water and sediment samples.

^dFecal coliform geometric mean values for CFU/100 ml with standard errors and the rates of prevalence (% positive) were determined as described in text for water samples in year 1 but not year 2.

^eValues in each column with different letters are significantly different from each other (P <0.05).

^fNT, not tested.

FIG 2.

Spatial distribution of Salmonella in irrigation ponds in southern Georgia. Levels (geometric mean MPN/liter ± standard errors) of Salmonella are shown for water (A) and sediment (B) samples. Ponds CC2, LV, MD1, NP, and SC (described in Table 1) were sampled from March 2011 to February 2012 in year 1 and from March 2012 to December 2012 in year 2.

A survey of pond characteristics was conducted immediately prior to this study (Table 1) and showed a significant positive correlation between Salmonella levels from peripheral sediments and pond size (r = 0.65, P < 0.05). Furthermore, ponds with irrigation pipes on the bottom (VH, MD1, CC2, RT1, and RT2) had significantly higher levels than those with higher irrigation pipes (data not shown). However, detailed analysis of the relationship of pond characteristics to Salmonella is beyond the scope of this paper and will be discussed elsewhere.

Seasonal distribution of Salmonella.

Salmonella was observed every month in water samples but was not detected in 6 of the 22 months in sediment samples (Fig. 3; see also Table S1 in the supplemental material). The geometric means for both water and sediment samples varied significantly among months in both year 1 and 2. For example, the Salmonella levels in water samples ranged from a low of 0.15 MPN/liter (March) to a high of 1.24 MPN/liter (September) in year 1 and from 0.12 MPN/liter (November) to 0.97 MPN/liter (October) in year 2. The levels in one water sample (from CC2 in October 2012) and two sediment samples (from VH in May 2011 and from MD1 in July 2011) were actually over the detection limit (1,100 MPN/liter). Prevalence ranged from 15 to 85% in samples from year 1 and from 10 to 50% for the 5 ponds sampled in year 2 (see Table S1 in the supplemental material). Elevated levels and prevalence of Salmonella in water coincided with elevated temperatures for 1 or 2 months preceding sampling (Fig. 3). Conversely, the highest levels and prevalence in sediments occurred earlier (May and July 2011 and May 2012) than those in water samples by 2 or 3 months. Significant correlations between Salmonella and temperature (0.41 ≤ r ≤ 0.53, P < 0.05) were only observed for ponds RT2 and VH. The correlation between Salmonella level and. temperature for all ponds together was only 0.14 ≤ r ≤ 0.18 (P < 0.05). Total rainfall 1 week before sampling had a significant positive correlation with overall Salmonella levels (0.27≤ r ≤ 0.18, P < 0.05) in all ponds, although the correlation was negative for RT2.

FIG 3.

Temporal distribution of Salmonella from irrigation ponds. Levels (geometric mean MPN/liter ± standard errors) of Salmonella (bars) and mean air temperatures (lines) are shown for water (A) and sediment (B) samples. ND, not detectable. Values were averaged for 10 ponds for March 2011 to February 2012 and for 5 ponds from March 2012 to December 2012, as described in the text.

Indicator bacteria and Salmonella.

New proposed FSMA rules for agricultural water quality require the level of generic E. coli to be <410 MPN/100 ml for any one sample, with a geometric mean of <126 MPN/100 ml for all samples. The generic E. coli levels in these pond water samples ranged from nondetectable to 1,046.2 MPN/100 ml, with a geometric mean of 6.26 MPN/100 ml for all water samples (Table 2). Although the rolling geometric means for any individual pond did not exceed proposed thresholds, 6 water samples from ponds CC1 (n = 2) and MD1 (n = 4) had generic E. coli levels of >410 MPN/100 ml. The generic E. coli levels were significantly correlated with the Salmonella levels in both water (r = 0.34) and sediment (r = 0.30) samples (Table 2). Stronger correlations were observed for specific ponds, e.g., 0.46 ≤ r ≤ 0.53 for water and sediment; however, no correlation was observed for these parameters in 5 of 10 individual ponds. Fecal coliforms were also examined in water samples in year 1, and had an overall geometric mean of 10.3 CFU/100 ml (Table 2). A good correlation was found between generic E. coli and fecal coliforms (r = 0.69; data not shown), and the correlation of Salmonella with fecal coliforms was similar to that with E. coli.

Regression analysis.

The relative effects of several parameters (generic E. coli, fecal coliforms, temperature, and rainfall) on Salmonella levels/occurrence in water and sediment samples were examined by multiple linear regression and binomial logistic regression. As fecal coliforms were only monitored in year 1 and only in water, modeling was performed either with (n = 340) or without (n = 240) fecal coliforms in water as an independent variable. Multiple linear regression modeling revealed that generic E. coli levels and temperatures in both water and sediment were significantly related to the Salmonella densities (P < 0.05). However, the R² values of these models were low (<0.2), indicating that the models did not reveal strong linear relationships (16). Therefore, logistic regression models were also used to examine Salmonella occurrence with respect to these parameters, and a predictive relationship (P < 0.05) was observed for both generic E. coli densities and water temperatures (not shown) relative to Salmonella densities in water and sediment samples (Fig. 4). This relationship was maintained regardless of the inclusion of fecal coliform densities in the model.

FIG 4.

Predicted probabilities of Salmonella occurrence using binomial logistic regression models. Models were used to predict the occurrence (percentage of positive samples) of Salmonella with 95% confidence limits (gray shading) relative to the levels of generic E. coli [log(MPN/100 ml)] in water samples when water temperature was held at the constant mean value (A) and in sediment samples when all other independent variables (mean air temperature 1 month before sampling and total rainfall 1 month and 1 week before sampling) were held at the constant mean values (B).

Antibiotic resistance profiles.

Among the 1,360 confirmed Salmonella isolates that were recovered during this study, antimicrobial susceptibility was examined in a group of strains (n = 185) that was selected for representative distribution among pond sources (10 ponds), sample types (4 types), and sample time points (over 22 months). Overall, only two isolates (1.1%) were susceptible to all antibiotics. Nearly all isolates (98.9%) were resistant to streptomycin, and these strains were distributed over all ponds and time points (Table 3). Twenty isolates (10.8%) were resistant to kanamycin, and these strains were recovered from 8 ponds. Twenty-six isolates (14.1%) were resistant to two antibiotics, mostly streptomycin and kanamycin. Only 11 Salmonella isolates (5.9%) were resistant to more than 2 antibiotics, and these strains, with resistance to between 3 and 6 antibiotics, were recovered from 6 ponds between May and October (Table 4). Two isolates (760 and 1101) isolated from the same pond at different time points showed similar resistance profiles (streptomycin, kanamycin, ceftriaxone, imipenem, and trimethoprim-sulfamethoxazole), and one of the isolates also had nalidixic acid resistance.

TABLE 3.

Antibiotic resistance of Salmonella isolates

Antibiotic	Symbol	Dose (μg)	% (no.) of isolates resistant to druga	Pond(s) containing resistant isolatesb
Amikacin	AN	30	0 (0)	None
Amoxicillin-clavulanic acid	AMC	30	1.1 (4)	CC2, NP
Ampicillin	AM	25	3.8 (7)	SC, RT2, RT1, MD1, LV
Cefoxitin	FOX	30	1.1 (2)	NP, CC2
Ceftriaxone	CRO	5	1.1 (2)	SC
Cephalothin	CF	30	2.7 (5)	VH1, RT1, NP, MD1, CC2
Chloramphenicol	C	30	0 (0)	None
Ciprofloxacin	CIP	2.5	2.2 (4)	CC2, NP, SC
Gentamicin	G	10	0 (0)	None
Imipenem	IMP	10	1.6 (3)	SC, RT1
Kanamycin	K	50	10.8 (20)	BB, CC1, LV, MD1, RT1, RT2, SC, VH1
Nalidixic acid	NAA	30	2.7 (5)	SC, MD1, LV
Streptomycin	S	25	98.9 (183)	All ponds
Trimethoprim-sulfamethoxazole	STX	23.75/1.25	3.8 (7)	SC, RT2, RT1, CC2
Tetracycline	TE	10	2.7 (5)	LV, NP, RT2, SC

^aTotal = 185 isolates.

^bPonds are described in Table 1.

TABLE 4.

Profiles of multidrug-resistant Salmonella isolates^a

Strain	Date of isolation	Pond	Sample type	MDR profileb	Deduced serotypec	DTd
1166	August 2012	LV	Water	S, K, AM, NAA	No match	13
1130	July 2012	MD1	Water	S, K, NAA	No match	17
1095	May 2012	NP	Sediment	S, AMC, CIP	Hadar	14
829	October 2011	NP	Water	S, CRO, C	Infantis	23
541	July 2011	RT1	Sediment	S, K, STX, CF	Infantis	23
705	September 2011	RT1	Water	S, K, AM, IMP	No match	42
360	May 2011	RT2	Water	S, K, STX	No match	1
556	July 2011	RT2	Sediment	S, K, AM, TE	No match	28
462	July 2011	SC	Water	K, S, STX, TE, NAA	No match	40
760	October 2011	SC	Water	S, K, CRO, STX, IMP	No match	8
1101	May 2012	SC	Sediment	S, K, CRO, STX, IMP, NAA	Agona	23

^aStrains are characterized by collection date, pond, and sample type (water or sediment). Strains with resistance to >3 antibiotics are shown.

^bAbbreviations and concentrations for antibiotics are defined in Table 4.

^cDeduced serotypes based on ≥95% similarity to reference strains of known serotype in DiversiLab library (see Table S3 in the supplemental material).

^dDT, DiversiLab type, based on the data in Fig. 5.

Genetic diversity of Salmonella in irrigation ponds.

The genetic diversity of Salmonella populations (n = 191) was examined in a group of strains that was representative of sources (6 to 14 strains for each pond), sample types (143 from water and 48 from sediments), and sample time points (5 to 16 for each month), including 85% of the strains in the antibiotic resistance study. The DiversiLab rep-PCR molecular typing system was used to define DiversiLab types (DTs) among the strains. Replicate analyses (n = 2 or more) were performed with 23 different Salmonella strains and demonstrated reproducibility of >95% similarity within the same strain (data not shown). Thus, each DT included strains whose profiles clustered with ≥95% similarity with no obvious missing peaks, as determined by overlay of electrophoretograms (32). The results showed >60% similarity among 43 DTs for all pond isolates. DT profiles were condensed, using only a single representative strain in each DT, in order to generate a collapsed dendrogram (Fig. 5). The number of strains within a genotype was not evenly distributed and varied from 1 to 35 for each DT. The most frequent DTs (14 and 24) comprised 28.8% of all strains examined. Seven DTs contained only strains from sediment samples, while 26 DTs were unique to water, and 10 DTs were isolated from both water and sediments. Among multidrug-resistant (MDR) strains (resistant to >2 antibiotics), genotype DT 23 was overrepresented, accounting for 27% (3/11) of this group, although the 3 strains were collected from different ponds and showed different resistance profiles.

FIG 5.

Distribution of DiversiLab genotypes for strains from irrigation ponds in southern Georgia. DiversiLab types (DTs) are defined as clusters of strains among 191 pond isolates with >95% similarity in their rep-PCR profiles. The dendrogram shown used one representative strain profile for each DiversiLab type (DT), which is identified by a number. The number of strains within each DT (Frequency) and the number derived from water samples versus sediment samples (W:S), as well as the number of months (listed as 1 to 12) and ponds (n = 10 total) represented by each DT, are also shown. DTs of pond strains that were >95% similar to strains from the reference library (Non-Pond Sources) are also shown and include reference strains from the Suwannee River (SR), the DiversiLab library (DL), other pond sites in Florida (FL), and produce (other), as described by Rajabi et al. (9).

The DiversiLab profiles in this study were compared to a DT reference library comprised of 313 entries provided by DiversiLab, which was derived from the 55 most frequently identified serotypes from clinical and nonclinical sources; they were also compared to DT profiles from our previous study with Suwannee river isolates (9). Based on the Top Match module results in the DiversiLab classification report, unique DTs were identified for 22% of the strains in this study that did not match (i.e., had <95% similarity) any reference profile (Fig. 5). Matches of DT profiles of pond strains to reference profiles were evenly divided between profiles of strains from the DiversiLab library (37%) and profiles of strains from the Suwannee River study (36%). The remaining strains matched those from other sources, such as lakes and produce from Florida. It was also noted that 10% of the pond strains in the present study had profiles with >95% similarity to more than one other source. Examination of the genetic diversity did not reveal specific genotypes that were associated with particular ponds (see Table S2 in the supplemental material) or seasons (see Table S3). The number of DTs per pond ranged from 5 (BB) to 13 (NP), and the number of DTs over time ranged between 5 (January and March) and 16 (October). Furthermore, only two DTs with more than 1 strain were unique for any specific month or pond. Thus, the results illustrated the wide spatial and temporal distribution of genetic diversity of Salmonella strains derived from these irrigation ponds.

Genetic similarity to known Salmonella serotypes.

Prior research reported that DiversiLab rep-PCR was capable of predicting the serotypes of Salmonella enterica isolates (33, 34). Therefore, the genetic similarity of pond strains to reference strains with known serotypes was determined based on >95% similarity, and this analysis assigned 95 strains (49.7%) to 24 serotypes (see Fig. S1 in the supplemental material). The most frequent serotypes identified by rep-PCR were Montevideo (13.8%), Newport (12.8%), and Hadar (11.7%). Several serotypes (Blockley, I4[5]12:i:−, Montevideo, Newport, Rubislaw, Saintpaul, and Stanley) were represented by multiple rep-PCR profiles within one serotype (Table 5). Conversely, some DT profiles were distributed among more than one serotype. Serotypes represented by multiple strains were recovered from multiple ponds, and all ponds supported multiple serotypes (see Table S4). Pond LV had the largest number of serotypes (8 serotypes), while BB, RT2, and MD1 had the fewest (4 serotypes). The diversity of serotypes in each pond did not show a significant correlation with measurable pond characteristics (data not shown). A fairly strong correlation (r = 0.75) was found between the number of deduced serotypes and the number of strains from each pond, indicating a wide distribution of serotypes, in agreement with the genetic diversity revealed by rep-PCR.

TABLE 5.

Distribution of serotypes based on similarity to DiversiLab types with known serotypes

Serotypea	DT(s) (no. of strains in DT)b	Pond(s) containing isolates with the serotypec	% of strains
Agona	23 (1)	SC	1.1
Anatum	26 (1)	VH1	1.1
Bareilly	27 (2)	RT2, NP	2.1
Blockley	31 (2), 33 (1)	LV, MD1	3.2
Braenderup	18 (5)	BB, RT1, CC2	5.3
Hadar	14 (11)	CC1, VH1, LV, NP, SC	11.7
Hartford	19 (1)	NP	1.1
I 4 [5] 12:i:-	14 (2), 15 (1), 27 (1)	BB, CC1, RT2	4.3
Infantis	23 (4)	RT1, NP, SC	4.3
Inverness	29 (5)	VH1, CC2	5.3
Javiana	5 (1)	8.3	1.1
Kentucky	14 (1)	CC2	1.1
Kiambu	41 (1)	RT1	1.1
Litchfield	19 (3)	RT2	3.2
Meleagridis	35 (1)	CC1	1.1
Montevideo	20 (1), 21 (2), 24 (9)	BB, CC2, LV, SC	13.8
Muenchen	14 (1)	BB, LV	2.1
Newport	31 (1), 33(11)	CC1, CC2, LV, MD1	12.8
Oranienburg	22 (1)	NP	1.1
Rubislaw	24 (1), 25 (5)	RT1, VH1, CC2, LV	6.4
Saintpaul	13 (4), 14 (3)	VH1, LV, MD1, SC	6.4
Stanley	14 (3), 20 (3)	CC1, VH1, CC2, LV, MD1	7.4
Thompson	14 (2)	RT1, RT2	2.1
Typhimurium	15 (1)	NP	1.1

^aSerotype determinations are based on >95% similarity to DiversiLab rep-PCR types (DTs) of reference strains with known serotypes.

^bThe DTs within each serotype are also shown in Fig. S1 in the supplemental material.

^cThe ponds are described in Table 1.

DISCUSSION

Irrigation water has been suggested as a potentially important source for Salmonella strains that cause illnesses attributed to fresh produce consumption (11, 12), including those involved in multistate outbreaks (6, 35–37). Therefore, Salmonella occurrence and levels in irrigation ponds, as well as possible environmental factors that might influence these levels, were investigated. This study provides a systematic survey of 10 irrigation ponds located within an important agricultural region in southern Georgia and northern Florida (38, 39) and represents the most comprehensive research (i.e., the largest number of samples and strains examined) to date on the risk of Salmonella contamination in agricultural water or pond sediments in the southeastern United States. The overall prevalence (38%) of Salmonella in surface water samples was similar to that in a prior report (39%) for the same ponds during an overlapping time period, using different methodologies (24). The present study also included quantitative analysis, which indicated that the Salmonella values for water samples were relatively low (overall geometric mean = 0.3 MPN/liter) compared to the values found in investigations of river samples within this region, which showed 79 to 96% occurrence of Salmonella and levels from nondetectable up to 5,400 MPN/100 ml (8, 9). Although direct comparisons are difficult, as the studies differed in methodology, geography, hydrology, and time of sampling, the results suggest that flowing rivers may sustain larger populations of Salmonella than more static ponds. In the present study, the overall Salmonella presence and/or density was higher than that reported in other watersheds in the United States (10, 40, 41). Also, in the present study, the Salmonella occurrence and levels were significantly higher for water samples than for sediment samples for both years, which was consistent with observations elsewhere (10, 14, 42).

For the most part, the Salmonella levels were relatively constant and consistently low (mean of <1 MPN per liter or gram) throughout the study. Seasonal variation in the detection of Salmonella showed peak values appearing in the fall (water) and summer (sediment) in both years; thus, increases were not sustained over a whole season. The correlations between Salmonella populations and temperatures or total rainfall were not strong (r < 0.2) and did not indicate a linear relationship. The results of prior research regarding seasonality of Salmonella have been somewhat inconsistent. For example, the Salmonella levels in rivers in the southeastern United States showed much stronger correlations (r = 0.44 to 0.77) with elevated temperatures and rainfall measurements preceding sampling, respectively (8). Conversely, a survey of Canadian rivers reported significantly higher prevalence in spring than in summer, although seasonal precipitation was also correlated with increased Salmonella presence (43). Moreover, results showing small but significant positive correlations with temperatures and rainfall are consistent with prior research for central Florida surface water during an overlapping time period (16). Although regression analysis provided predictive models for effects of rainfall and temperature on Salmonella densities, further study is needed to refine these models for evaluating Salmonella risk. The interaction of temperature and rainfall is complicated by numerous factors, including geographic location, drought, flooding, and various agricultural practices. For example, rainfall may either dilute pathogen concentrations or suspend bacteria from sediments or pond banks into the pond water.

The FSMA now recommends standards for generic E. coli in agricultural water that comes in contact with produce. Unfortunately, the densities of indicator bacteria generally show little or no correlation with the presence/densities of pathogens in environmental samples (14, 16–19). Therefore, we examined whether or not the ponds in this study would meet the FSMA criteria. The results demonstrated that all ponds met the updated thresholds (<26 CFU or MPN per 100 ml, with 90% of samples under 410 CFU or MPN per 100 ml), and significant correlations between indicator bacteria (both generic E. coli and fecal coliforms) and Salmonella were observed. Furthermore, regression models revealed that generic E. coli was the strongest predictor among all parameters for Salmonella presence in both water and sediment samples. However, it should be noted that Salmonella was identified in all ponds over the entire 22 months, even when E. coli was below actionable levels, which suggests that the current limits for generic E. coli are not sufficient to exclude the presence of Salmonella.

Another indication of human impact is the presence of MDR Salmonella strains in agricultural settings, as resistant isolates are generally thought to be derived from livestock exposed to a broad range of antibiotics (44). Relatively high numbers of antibiotic-resistant strains were isolated from these irrigation ponds (20% with resistance to 2 or more antibiotics), which was consistent with a smaller-scale investigation of Salmonella from the same ponds in 2011 to 2013 showing 33% MDR among 51 isolates tested (24). Interestingly, the most common profile of 10 antibiotics reported in the previous study was not observed in our survey. Furthermore, resistance to streptomycin was less frequent in their study than in ours (31% versus 98%). Other surveys in different regions of the United States reported a lower percentage of MDR isolates from agricultural water. For example, only one MDR isolate (serotype Miami) was isolated from agricultural water in North Carolina (45), and a survey of water samples from 14 tomato farms in mid-Atlantic coast states found only three (4.8%) MDR Salmonella strains (42). However, none of our MDR patterns overlapped with the MDR patterns in those studies, except that streptomycin resistance was widely distributed among the MDR strains in most cases, possibly due to the widespread streptomycin exposure of bacteria in soil (46).

The genetic diversity of Salmonella from irrigation ponds was examined by the DiversiLab rep-PCR typing system. This platform offers reduced labor requirements and better assessment of horizontal gene transfer and serotyping compared to the traditional PFGE (currently the gold standard in outbreak source tracking) (34, 47). The 191 Salmonella isolates from 10 irrigation ponds comprised a diverse population that was distributed over 43 DiversiLab types (DTs). However, most strains (70%) belonged to only 7 DTs. Overlap of DTs for strains from both water and sediment samples indicated the potential interaction of water and sediments, consistent with the 33% overlap of E. coli genotypes between soil and water (48). Populations of Salmonella may be ecoregion specific in that 63% of strains did not cluster with any reference strain from the national database but were genetically more similar to isolates derived from this study or from other aquatic sources in Georgia and Florida. Furthermore, spatial and temporal distribution of Salmonella populations, based on the rep-PCR results, revealed that certain genotypes were identified in multiple months from the same pond, implying environmental persistence over sustained periods. Alternatively, the original source (i.e., wildlife) may be reseeding ponds over time.

Although strains from this study were not serotyped, DiversiLab profiles are a good predictor of serotype (33, 34), and nearly half of the strains from irrigation ponds were assigned to 24 serotypes based on rep-PCR identity. The most prevalent serotypes were Montevideo, Newport, and Hadar, which differed from the other survey of the same ponds that used PFGE and predicted only 9 serotypes, with Newport (57%) and Enteritidis (12%) as the most prevalent (24). Overlapping serotypes from both studies included Newport, Thompson, Muenchen, Saintpaul, Javiana, and Bareilly; however, two common serotypes in our study (Montevideo and Hadar) were not detected in the previous study. Interestingly, strains that were serotyped as Newport by PFGE in the prior study were frequently MDR, while no MDR Newport strains were identified by rep-PCR in our study. Montevideo and Newport have been identified in several multistate outbreaks associated with fresh produce (4, 6, 37, 49), while Hadar was more associated with meat or poultry (50, 51). Strains with genetic similarity to serotypes Saintpaul, Braenderup, Thompson, and Javiana have also been recovered in produce-related outbreaks (6, 42, 52–54), and Javiana is the most commonly reported serotype for salmonellosis in Florida (Carina Blackmore, Florida Department of Health, personal communication). Recently, Newport, Cerro, and Thompson were reported as the most common serovars of isolates from produce-growing regions of New York State, while Saphra, Newport, and S. enterica subsp. diarizonae serovar 50:r:z were most common in south Florida (56). The disparities among studies could represent differences in either sampling methodologies or the discriminatory powers of the different typing methods.

In summary, irrigation ponds within the Suwannee River watershed contain widely distributed, sustainable, and diverse populations of Salmonella at levels that are probably below the threshold for public health concerns. Occasional spikes in the levels of Salmonella were not predicted by microbial indicators and were not consistent with any of the environmental parameters examined. Thus, the best predictor for Salmonella is likely to be an assay for detection of Salmonella itself. Improvements in PCR (57–59) and non-PCR methods, such as the application of multiple agars (25), may offer alternatives to fecal indicators for future agricultural water quality analysis. The virulence potential of Salmonella isolates associated with agricultural water remains to be determined, but the antibiotic resistance patterns and genetic relatedness of pond isolates to serovars commonly associated with disease support the need for continued monitoring and surveillance.