Temporal Assessment of the Impact of Exposure to Cow Feces in Two Watersheds by Multiple Host-Specific PCR Assays

Abstract

Exposure to feces in two watersheds with different management histories was assessed by tracking cattle feces bacterial populations using multiple host-specific PCR assays. In addition, environmental factors affecting the occurrence of these markers were identified. Each assay was performed using DNA extracts from water and sediment samples collected from a watershed directly impacted by cattle fecal pollution (WS1) and from a watershed impacted only through runoff (WS2). In WS1, the ruminant-specific Bacteroidales 16S rRNA gene marker CF128F was detected in 65% of the water samples, while the non-16S rRNA gene markers Bac1, Bac2, and Bac5 were found in 32 to 37% of the water samples. In contrast, all source-specific markers were detected in less than 6% of the water samples from WS2. Binary logistic regressions (BLRs) revealed that the occurrence of Bac32F and CF128F was significantly correlated with season as a temporal factor and watershed as a site factor. BLRs also indicated that the dynamics of fecal-source-tracking markers correlated with the density of a traditional fecal indicator (P < 0.001). Overall, our results suggest that a combination of 16S rRNA gene and non-16S rRNA gene markers provides a higher level of confidence for tracking unknown sources of fecal pollution in environmental samples. This study also provided practical insights for implementation of microbial source-tracking practices to determine sources of fecal pollution and the influence of environmental variables on the occurrence of source-specific markers.

The fecal bacterial group is one of the top three contaminants impacting the quality of recreational waters and drinking water sources in the United States (30). Bodies of water with high levels of pollution are believed to be polluted primarily by nonpoint sources, such as wildlife, domesticated animals, and agricultural and urban runoff (30). While inadequately treated sewage and septic system failure can occasionally impact surface waters, animal fecal pollution is unquestionably the major source of pollution in developed countries (23). The annual production of animal feces in the United States is approximately 1 × 10⁹ tons (29), and this number will likely increase due to current trends in the American diet. Poultry, cattle, and swine are responsible for 44, 31, and 24% of the total annual production of animal feces, respectively (14).

Detecting and ranking the sources of fecal pollution are challenging as conventional methods used to monitor pollution cannot discriminate among the various potential nonpoint sources impacting a body of water (30). As a consequence of the development and implementation of total maximum daily loads, several microbial source-tracking (MST) methods have been developed and tested in field applications (3, 5, 19, 26, 27). Among these methods, culture-independent host-specific PCR assays are gaining acceptance because of their potential for high throughput and quick turn-around times (22). In most MST studies using host-specific PCR assays, researchers have employed the one assay-one genetic marker approach to detect a suspected fecal source. Given the complexity of environmental conditions and the poorly understood ecology of targeted microbial populations, we surmise that the latter approach has limited value in field applications. In addition, a variety of parameters could be responsible for introducing spatial and temporal variability into targeted microbial populations, possibly reducing the effectiveness of MST assays. For example, seasonal differences in precipitation, temperature, sunlight, and host diet and spatial differences in the proximity of host animals to a body of water (i.e., direct or indirect contact with water), transport from sediments to the water column, and the hydrogeology of areas upstream and downstream from a suspected pollution source(s) can affect the performance of MST methodologies (4, 8, 10, 12, 13, 17, 32). Despite the fact that spatial and temporal factors directly impact successful implementation of MST approaches, these factors are often ignored.

The objectives of this study were to assess exposure to feces in two watersheds that were impacted by cattle farms with different management practices and to simultaneously elucidate the effects of environmental factors related to spatial and temporal variability on targeted microbial populations by using a combination of 16S rRNA gene (3) and non-16S rRNA gene (25) assays. This is the first time that the recently developed cattle-specific non-16S rRNA gene markers have been evaluated using both water and sediment samples collected along watershed transects. Because both sampling locations that were used are known to be impacted mainly by cattle fecal pollution, this study provides a case scenario for understanding the relationships among MST assays, numbers of fecal indicators, and environmental variables.

MATERIALS AND METHODS

Sampling locations.

The study sites included two watersheds (WS1 and WS2) containing beef cattle farm operations (Fig. 1). Each sampling location was identified with a Trimble GPS Pathfinder Pro XRS receiver with a TSC1 datalogger (Trimble Navigation Ltd., Sunnyvale, CA). Postprocessed code differential correction was applied using GPS Pathfinder Office v. 2.70. WS1, impacted by farm A, is located in Madison County, GA, while WS2 starts in farm B located in U.S. Department of Agriculture-owned land in Watkinsville, GA. In WS1, samples were collected from four sites along a creek and a pond (Fig. 1A). Cattle had direct access to all sampling sites in WS1 except site 1, which was located upstream from the farm outside the property fence (Fig. 1A). On average, 60 cattle were present on this farm. In addition, wildlife, including geese and deer, were spotted in the sampling area during the course of the study and could have contributed to the total fecal load.

FIG. 1.

Aerial views of sampling sites in WS1 (A) and WS2 (B). The arrows indicate the direction of flow. The delimited areas indicate a property fence on farm A and cattle plots on farm B. PA, pond on farm A; PB-1, agricultural pond on farm B; PB-2, community pond outside farm B. (Aerial photographs courtesy of GlobeXplorer.com.)

There were 12 sampling sites in WS2 (Fig. 1B); 7 of these sites were located along a headwater stream and a pond on the farm, while 5 sites were located on the same creek downstream outside the farm. A spring was located at site 1, while sites 5 and 11 were located in agricultural and community ponds, respectively (Fig. 1B). An average of 140 cattle were kept and rotated among 16 fenced pastures on farm B. The distance between the fence and the stream ranged from 19.8 to 48.6 m. Unlike the cattle on farm A, the cattle on farm B had no access to the stream or the pond. Other possible fecal sources affecting the stream and ponds in this area included wildlife, such as deer, geese, and raccoons, and septic systems (the primary human source).

Sample collection.

Water, sediment, and fecal samples were collected monthly between September 2005 and February 2007. Water samples were collected in sterilized 1-liter bottles, transported on ice to the lab, and processed for enterococcal enumeration and nucleic acid extraction within 6 h after collection. Sediment samples were collected from each site except sites 5 and 11 in WS2 in sterile 50-ml tubes from the top 10 cm below the water-sediment interface. Two cow fecal samples per sampling time were collected aseptically from each farm. Sediment and fecal samples were stored at −20°C until they were processed.

Analytical and microbiological methods.

The temperature and pH of water samples were determined onsite using a portable pH meter (Orion 250Aplus; Thermo Orion, Beverly, MA). The daily precipitation data for WS1 and WS2 were obtained from the National Oceanic and Atmospheric Administration (station ID 092517; http://hurricane.ncdc.noaa.gov/dly/DLY) and the Georgia Automated Environmental Monitoring Network (http://www.georgiaweather.net/), respectively. Turbidity was measured in the laboratory using a 2020 turbidimeter (LaMotte Co., Chesterfield, MD) according to the manufacturer's instructions. Enterococcal densities were determined using the membrane filtration technique described in EPA method 1600 (28). The colonies were counted after 24 h of incubation at 41°C.

Molecular methods.

One hundred milliliters of water, 0.5 to 1.0 g of sediment, and 0.2 to 0.25 g of feces were used for DNA extraction with an UltraClean soil DNA kit (MoBio Inc., Carlsbad, CA) used according to the manufacturer's instructions, with some modifications. Briefly, water samples were filtered onto polycarbonate membranes (0.22 μm; Millipore Inc., Bedford, MA), which were then transferred to a 6-ml sterile tube containing a bead solution and solution S1 and vortexed for 10 min. Inhibitor removal solution was added after solution S2, which was followed by the steps described in the manufacturer's instructions. The nucleic acid fraction was eluted in 65 μl of Tris-EDTA buffer and quantified using a NanoDrop ND-1000 UV/visible spectrophotometer (NanoDrop Technologies, Wilmington, DE). The DNA concentration was adjusted to 10 ng/μl.

Host-specific PCR assays targeting 16S rRNA and functional genes were used to monitor cattle fecal pollution (Table 1). Three non-16S rRNA gene markers reported for the first time in this study (Bac4, Bac5, and Bac6) were obtained from cow metagenome enrichments and were presumed to target Bacteroides-like genes (for details concerning the development of non-16S rRNA gene markers, see reference 25). All PCR assays were performed using GoTaq Green master mixture (Promega, Madison, WI). The annealing temperature for each PCR assay was determined using a gradient PCR. The thermal cycling conditions for the 16S rRNA gene-based assays were an initial denaturation step of 2.5 min at 94°C, followed by 30 cycles of denaturation at 94°C for 30 s, annealing at an optimized temperature for 30 s (Table 1), and extension at 72°C for 60 s and then a final extension for 5.5 min at 72°C. The conditions for the non-16S rRNA gene-based assays were initial denaturation for 3 min at 94°C, followed by 35 cycles of 94°C for 60 s, annealing at an optimized temperature for 60 s (Table 1), and 72°C for 60 s and then a final extension for 5 min at 72°C. Amplification products were visualized on a 2% agarose gel stained with 0.2× Sybr Safe DNA gel stain (Invitrogen, Carlsbad, CA). DNA extracts from cow feces freshly obtained at each sampling time were used as positive controls and to test the host distribution of markers used in this study. The practical detection limits for each assay were determined using 10-fold dilutions of two randomly selected cattle DNA extracts. The negative controls included extraction blanks (containing no samples) and extraction controls (DNA extracts from membrane filters through which Nanopure water was filtered) for DNA extraction, as well as PCR blanks with no DNA template.

TABLE 1.

Sequences of primers used in this study

Primer	Sequence (5′ to 3′)	Expected size of PCR product (bp)	Annealing temp (°C)a	Reference
Bac32F	AACGCTAGCTACAGGCTT	676	58	2
Bac708R	CAATCGGAGTTCTTCGTG
CF128Fb	CCAACYTTCCCGWTACTC	580	53	3
HF183Fb	ATCATGAGTTCACATGTCCG	525	59	3
Bac1F	TGCAATGTATCAGCCTCTTC	196	55	24
Bac1R	AGGGCAAACTCACGACAG
Bac2F	GCTTGTTGCGTTCCTTGAGATAAT	274	55	24
Bac2R	ACAAGCCAGGTGATACAGAAAG
Bac3F	CTAATGGAAAATGGATGGTATCT	166	55	24
Bac3R	GCCGCCCAGCTCAAATAG
Bac4F	TGGGAATGGCGGTAATCTCG	187	58	This study
Bac4R	CAACAGCCGGTCGTCTTCCT
Bac5F	ACTCCCTGCGCTCCGAAGATA	150	55	This study
Bac5R	GGCCCAGGCACCATTTACAGT
Bac6F	CTCCGTCTTTCTCCGTCCTGTTCT	430	58	This study
Bac6R	GATCCCCCTCGCCTCCGTCCT

^aEach annealing temperature was adjusted after gradient PCR.

^bUsed with Bac708R.

Statistical analyses.

To detect and avoid severe multicollinearity among the measured environmental variables (pH, temperature, turbidity, 3-day precipitation [sum of the precipitation for the previous 72 h, including the sampling day], and enterococcal density), we examined Pearson correlation coefficients. Binary logistic regressions (BLRs) were used to identify associations between these variables and the occurrence of the molecular markers. A watershed factor (WS1 versus WS2) was included, as well as a “site grouping” factor nested within watersheds. Groups of sampling sites within each watershed were formed based on the effects of physical and chemical stream characteristics on the occurrence of the molecular markers. The six site groups were as follows: WS1 site 1, WS1 sites 2 and 3, WS1 site 4, WS2 sites 1 to 4, WS2 sites 5, 6, 11, and 12, and WS2 sites 7 to 10. A seasonal factor was created by grouping samples for the following 3-month intervals: March to May (spring), June to August (summer), September to November (fall), and December to February (winter). When continuous independent variables are used in logistic regression models, a primary assumption is that there is linearity in the log odds of the response variable over the entire range of the continuous independent variable. The deviance statistic (−2 log likelihood, chi-square distribution) was used to test for model lack of fit, and a nonsignificant deviance was considered a lack of evidence for violation of the assumption of log odds linearity. BLR analyses were performed only for the markers whose frequency of occurrence at all sites was between 0.2 and 0.8. The significance level for variable evaluation was set at alpha = 0.05.

A generalized linear model (GLM) was constructed using the natural logarithm of enterococcal density as the response variable. Three categorical factors (watershed, site group, and season) and four continuous variables (pH, temperature, turbidity, and 3-day precipitation) were examined as potential explanatory variables. A backward selection procedure was performed for final model determination; i.e., the analysis started with the full model (all potential predictors), and then the least significant explanatory variable (based on its P value) was sequentially deleted until only significant variables (P < 0.05) remained.

RESULTS

Detection of molecular markers in fecal and water samples.

The practical detection limits for the PCR assays were 4.17 × 10⁻⁷, 4.17 × 10⁻⁶, and 4.17 × 10⁻⁴ g of wet cattle feces for general Bacteroidales marker Bac32F, ruminant-specific Bacteroidales marker CF128F, and metagenome-based markers, respectively. These values compare favorably with previously reported detection limits (3, 16). No amplification signals were obtained with the negative controls. Except for one occasion, all DNA extracts of cow feces were positive for Bac32F (Table 2). The detection frequencies for Bac1, Bac2, Bac3, and Bac5 were between 0.68 and 1.0 for fecal samples, which were lower than the frequency observed for CF128F (0.97 to 1.0). Bac4 and Bac6 assays were positive for feces collected in WS1 (frequencies, 0.03 and 0.15, respectively), while they were negative for feces collected in WS2 (Table 2).

TABLE 2.

Average detection frequencies for 16S rRNA gene and non-16S rRNA gene markers in bovine fecal samples collected over a 17-month period

Marker	Detection frequencies
	WS1 (n = 34)	WS2 (n = 22)
Bac32F	0.97 (0.17)a	1
CF128F	0.97 (0.17)	1
Bac1	0.74 (0.45)	0.86 (0.35)
Bac2	0.85 (0.36)	1
Bac3	0.76 (0.43)	0.77 (0.43)
Bac4	0.03 (0.17)	0
Bac5	0.68 (0.47)	0.73 (0.46)
Bac6	0.15 (0.36)	0

^aThe numbers in parentheses are standard deviations.

Significant site-to-site variability was observed for water chemistry and the occurrence of molecular markers within each watershed and between watersheds (Tables 3 and 4). The frequency of Bac32F in water samples ranged from 0.76 to 1.0 in WS1 and from 0 to 0.88 in WS2 (Table 4). Bac32F was detected at all sites except site 11 in WS2. The latter site was located in a large community pond. CF128F was detected at frequencies of 0 to 1.0 and 0 to 0.12 in WS1 and WS2, respectively. The highest frequency of CF128F (>0.88) was obtained for sites 2 and 3 in WS1, which coincided with the frequent access of cattle at these sites. In WS2, the highest frequency (0.12) of this marker was obtained for sites 3 and 4, while the frequencies for the other sites were less than 0.07. In the non-16S rRNA gene-based assays, only the Bac1, Bac2, Bac3, and Bac5 markers were detected in WS1. The average frequency of detection for Bac1, Bac2, and Bac5 was 34% for water samples collected in WS1. In agreement with the results for CF128F, the detection frequency for these markers was higher for sites 2 and 3 (0.53 to 0.71). The detection frequency for Bac3 was the lowest among the four detectable markers (0 to 0.29). The detection frequencies for the non-16S rRNA gene markers were much lower (0 to 0.06) in WS2 than in WS1.

TABLE 3.

Results of water quality analysis of stream samples collected monthly in WS1 and WS2 over a 17-month period

Site	n	pH		Temp (°C)		Turbidity			Enterococcal density (CFU/100 ml)a
		Mean	95% confidence interval	Mean	95% confidence interval	Geometric meanb	95% confidence interval		Geometric meanb	95% confidence interval
							Lower boundary	Upper boundary		Lower boundary	Upper boundary
WS1 sites
1	17	6.04	±0.28	12.85	±2.44	44.15	17.97	108.48	74	9	640
2	17	6.48	±0.14	13.64	±2.59	30.73	16.52	57.18	2,600	1,011	6,685
3	17	6.45	±0.16	14.21	±2.79	26.56	14.89	47.40	1,423	795	2,547
4	17	7.32	±0.54	15.75	±4.05	162.15	106.82	246.14	93	42	207
WS2 sites
1	17	5.15	±0.20	16.14	±1.50	0.12	0.03	0.49	1	0	6
2	17	6.41	±0.29	13.59	±2.82	58.40	26.85	127.01	81	10	658
3	17	6.44	±0.30	13.06	±3.07	17.44	9.95	30.56	41	14	118
4	17	6.41	±0.13	14.25	±2.45	6.13	4.24	8.87	225	83	612
5	15	6.60	±0.69	16.66	±3.42	11.20	8.26	15.19	2	0	15
6	15	6.78	±0.39	17.89	±3.85	11.51	8.37	15.82	5	1	36
7	15	6.49	±0.21	17.08	±3.52	8.05	5.90	10.99	89	38	208
8	12	6.72	±0.21	15.99	±3.95	5.82	3.78	8.98	89	7	1,142
9	12	6.70	±0.20	16.11	±3.86	6.63	4.97	8.84	121	18	822
10	15	6.60	±0.23	14.67	±3.32	8.43	5.43	13.10	229	89	588
11	15	7.56	±0.54	17.45	±4.11	5.55	4.11	7.50	4	1	24
12	15	6.71	±0.16	14.03	±2.86	7.49	5.74	9.75	5	1	30

^aEnterococcal densities recorded as too numerous to count were converted to 20,000 for WS1 and to 2,500 for WS2 based on the maximum enterococcal numbers observed in each watershed.

^bValues of zero were converted to 0.01.

TABLE 4.

Detection frequencies for molecular markers in stream water samples collected during a 17-month period

Site	Detection frequencies (avg ± 95% confidence interval)
	Bacteroidales 16S rRNA gene markers		Non-16S rRNA gene markers
	Bac32F	CF128F	Bac1	Bac2	Bac3	Bac5
WS1 sites
1	0.76 ± 0.09	0	0	0	0	0
2	1	1	0.59 ± 0.11	0.71 ± 0.10	0.29 ± 0.10	0.65 ± 0.11
3	1	0.88 ± 0.05	0.65 ± 0.11	0.59 ± 0.11	0.24 ± 0.09	0.53 ± 0.12
4	0.88 ± 0.05	0.71 ± 0.10	0.12 ± 0.05	0.18 ± 0.07	0	0.12 ± 0.05
WS2 sites
1	0.41 ± 0.11	0.06 ± 0.03	0.06 ± 0.03	0.06 ± 0.03	0.06 ± 0.03	0.06 ± 0.03
2	0.76 ± 0.09	0.06 ± 0.03	0	0.06 ± 0.03	0.06 ± 0.03	0.06 ± 0.03
3	0.82 ± 0.07	0.12 ± 0.05	0.06 ± 0.03	0.06 ± 0.03	0.06 ± 0.03	0.06 ± 0.03
4	0.88 ± 0.05	0.12 ± 0.05	0	0.06 ± 0.03	0	0
5	0.33 ± 0.11	0.07 ± 0.03	0	0.07 ± 0.03	0	0
6	0.13 ± 0.06	0.07 ± 0.03	0	0.07 ± 0.03	0	0
7	0.73 ± 0.10	0.07 ± 0.03	0	0.07 ± 0.03	0	0
8	0.67 ± 0.13	0	0	0	0	0
9	0.67 ± 0.13	0	0	0	0	0
10	0.73 ± 0.10	0.07 ± 0.03	0	0.07 ± 0.03	0	0
11	0	0	0	0	0	0
12	0.20 ± 0.08	0	0	0	0	0

Relationships among environmental variables, enterococcal density, and the occurrence of molecular markers.

The Pearson correlation coefficients did not indicate high multicollinearity among the environmental variables, so all of the variables were included in BLR and GLM model development. Based on the overall occurrence of the markers, two BLR models were developed, one for Bac32F and one for CF128F (Table 5). The deviance for both of these models was insignificant, showing no lack of model fit and providing some evidence that the assumption of log odds linearity was not violated. The analyses showed that pH (P < 0.001) negatively affected the presence of Bac32F, while WS1 (P = 0.002) had a positive effect on the occurrence of both Bac32F and CF128F. Interestingly, enterococcal density (P < 0.001) was positively correlated with the occurrence of both Bac32F and CF128F. Season was also identified as a significant explanatory variable for both Bac32F and Bac128F (Table 5). Bac32F occurred more frequently in spring and summer than in winter, while CF128F occurred more often in winter than in fall. BLR analyses for the non-16S rRNA gene markers were performed only for sites 2 and 3 of WS1, where the frequencies of occurrence of the markers (Bac1, Bac2, Bac3, and Bac5) were between 0.2 and 0.8. The pH (P = 0.008) was a significant explanatory variable for Bac3, while no significant explanatory variables were observed for Bac1, Bac2, and Bac5 (data not shown).

TABLE 5.

BLR results for 16S rRNA gene and non-16S rRNA gene markers in WS1 and WS2

Response variable	Explanatory variablea	P	Odds ratio 95% confidence interval		Test of global null hypothesis			Model fit
			Lower boundary	Upper boundary	Chi square	df	P	Deviance	df	Pb	Rescaled R2
Bac32F	pH (−)c	<0.001	0.24	0.64	110.8	6	<0.001	194.2	224	0.925	0.52
	Log (enterococcal density)	<0.001	1.44	2.02
	Seasond
	Spring	<0.001	2.55	17.11
	Summer	0.003	1.81	17.97
	Fall	0.439	0.57	3.67
	Watershed (WS1)e	0.002	1.91	15.86
CF128F	Log (enterococcal density)	<0.001	1.35	2.08	114.8	5	<0.001	107.9	177	1.000
	Season
	Spring (−)	0.339	0.18	1.79
	Summer (−)	0.080	0.08	1.16
	Fall (−)	0.017	0.05	0.74
	Watershed (WS1)	<0.001	8.41	55.35

^aA minus sign in parentheses indicates a negative correlation.

^bA deviance P value of >0.05 indicates no significant lack of fit.

^cpH values were transformed by dividing the values by 10.

^dThe season factor can be interpreted as the difference between marker occurrence in a season and marker occurrence in winter; e.g., “spring” being significant (P < 0.001) means that the marker occurrence in the spring was found to be higher than the marker occurrence in the winter.

^eThe watershed factor can be interpreted as the difference between marker occurrence in WS1 and marker occurrence in WS2 (i.e., WS1 marker occurrence − WS2 marker occurrence).

The GLM analysis (Table 6) using enterococcal density as the response variable indicated that all explanatory variables (watershed, season, site group, pH, temperature, turbidity, and precipitation) were significant. All covariates (pH, temperature, turbidity, and 3-day precipitation) had a positive effect on enterococcal density. The GLM indicated that the enterococcal density was higher in WS1 than in WS2; within the watersheds, the greatest densities were observed at sites 2 and 3 in WS1 and at sites 1 to 4 in WS2. The enterococcal density was highest in fall, decreased in winter and spring, and was lowest in summer.

TABLE 6.

Significant explanatory variables of enterococcal density as determined by the GLM for stream water samples collected in WS1 and WS2^a

Explanatory variable with log (enterococcal density) as the response variable	Coefficient	P
Constant	−2.424	0.122
pH	0.656	0.004
Temp	0.144	0.002
Turbidity	0.004	0.006
Three-day precipitation	1.401	<0.001
Watershed (WS1)b	1.184	<0.001
Seasonc
Spring	−0.389	0.117
Summer	−0.961	0.021
Fall	0.770	0.007
Site groupingd
WS1 site 1	−0.378	0.454
WS1 sites 2 and 3	2.164	<0.001
WS2 sites 1 to 4	0.830	0.002
WS2 sites 5, 6, 11, and 12	−2.069	<0.001

^aThe model fit adjusted R² was 0.497.

^bThe watershed factor can be interpreted as the difference between enterococcal densities in WS1 and WS2 (i.e., WS1 enterococcal density − WS2 enterococcal density).

^cThe season factor can be interpreted as the difference between the enterococcal densities in a season and winter; e.g., “Summer” being significant (P = 0.021) means enterococcal densities in the summer were found to be lower (negative coefficient) than enterococcal densities in the winter.

^dThe site grouping factor can be interpreted as the difference between enterococcal densities at a group of sites and WS1 site 4 (for WS1 samples) or WS2 sites 7 to 10 (for WS2 samples). For example, the positive and significant (P = 0.002) coefficient for “WS2 sites 1 to 4” means that the enterococcal densities at WS2 sites 1 to 4 were found to be greater than the enterococcal densities at WS2 sites 7 to 10.

Detection of 16S rRNA gene and non-16S rRNA gene markers in sediments.

Sediment samples were collected along with water samples to examine the persistence of a group containing the strict anaerobe Bacteroides when it was exposed to the open environment. Overall, molecular markers were detected less frequently in sediment samples than in water samples (Table 7). The host-specific markers, including CF128F, Bac2, Bac3, and Bac5, were detected in sediment samples only from sites 2 and 3 in WS1 with frequencies between 0.06 and 0.59. For the most part the host-specific markers were not detected in WS2; the only exception was CF128F, which was detected once at sites 4 and 10. In contrast, Bac32F was detected at every sampling site, and the frequency of detection ranged from 0.06 to 0.88 in WS1 and from 0.17 to 0.59 in WS2.

TABLE 7.

Average detection frequencies for molecular markers in sediment samples collected in WS1 and WS2

Site	n	Detection frequencies (avg ± 95% confidence interval)
		Bacteroidales 16S rRNA gene markers		Non-16S rRNA gene markers
		Bac32F	CF128F	Bac2	Bac3	Bac5
WS1 sites
1	17	0.18 ± 0.07	0	0	0	0
2	17	0.82 ± 0.07	0.41 ± 0.11	0	0.06 ± 0.03	0.06 ± 0.03
3	17	0.88 ± 0.05	0.59 ± 0.11	0.06 ± 0.03	0.12 ± 0.05	0
4	17	0.06 ± 0.03	0	0	0	0
WS2 sitesa
1	17	0.12 ± 0.05	0	0	0	0
2	17	0.24 ± 0.09	0	0	0	0
3	17	0.18 ± 0.07	0	0	0	0
4	17	0.59 ± 0.11	0.06 ± 0.03	0	0	0
6	15	0.13 ± 0.06	0	0	0	0
7	15	0.27 ± 0.10	0	0	0	0
8	12	0.17 ± 0.08	0	0	0	0
9	12	0.33 ± 0.13	0	0	0	0
10	15	0.47 ± 0.13	0.07 ± 0.03	0	0	0
12	15	0.07 ± 0.03	0	0	0	0

^aWS2 sites 5 and 11 were located in ponds; thus, no sediment samples were collected.

DISCUSSION

Cattle fecal pollution was assessed using assays targeting 16S rRNA gene and metagenomic fragments with samples from two watersheds associated with cattle farms with different management practices. The frequencies of most non-16S rRNA gene markers (all markers except Bac4 and Bac6) were comparable to that of CF128F in the bovine fecal DNA samples (Table 2). However, the ruminant-specific 16S rRNA gene marker, CF128F, was detected more frequently than the non-16S rRNA gene markers in water and sediment samples in the stream directly impacted by cattle (Tables 4 and 7). Differences in the abundance, host distribution, and environmental survival rates of targeted populations may partially explain the different detection rates observed for the molecular markers tested with water samples (1, 22). In addition, differences in copy number between targeted genes and amplification efficiency can also contribute to the differences in sensitivity among MST assays. Of the markers developed and tested for the first time in this study (Bac4, Bac5, and Bac6), Bac5 was found in 70% of fecal samples, while Bac4 and Bac6 were rarely detected. These results are consistent with the fact that Bac5 was the only new marker that was detected in water samples. These findings suggest that besides abundance, host distribution is an important factor for detection by host-specific assays for environmental waters. Both of these factors are critical parameters for the development of quantitative environmental monitoring assays (18).

Except for one time at site 1 in WS1 (April 2006), the human-specific Bacteroidales marker (HF183F) was not detected at any of the sampling sites, suggesting that there was no detectable human fecal contamination in the watersheds. The ruminant-specific marker (CF128F) was either not detected or detected at very low frequencies in WS2. In contrast, the general Bacteroidales marker, Bac32F was detected at every sampling site except site 11 in WS2. The fact that the source-specific markers were not detected at some of the sampling sites where Bac32F was present indicates that the general Bacteroidales marker is more abundant in the environment (16). The higher frequency of this marker could be due to a variety of factors: (i) the presence of the gene in other fecal sources, such as wildlife; (ii) the presence of the targeted gene in nonfecal sources (indigenous microbial populations); or (iii) a higher copy number of the targeted ribosomal gene in the population.

On seven different occasions (data not shown), CF128F was detected in sediment samples but not in the water samples collected from the same sampling sites. This was possible as anaerobic ruminant bacteria may survive for an extended period of time in sediments. Although strict anaerobes are not expected to survive when they are exposed to oxygenic environments (6, 9, 15), it was recently reported that Bacteroides could continue to grow for up to 24 h under aerobic conditions (31). This further suggests that repetitive introduction of fecal material into watersheds may lead to continuous detection of fecal indicators in sedimentary environments; however, without the repetitive introduction of fecal material, the detection of a marker signal may fade over time.

The host-specific markers were detected in WS2 only when cattle were present in the pasture adjacent to the stream in combination with a rain event within 3 days of sampling. For instance, in January 2007 when these two conditions were met, the non-16S rRNA gene markers were detected at sites 1 through 3. In contrast, during the same month, CF128F was detected at sites 1 to site 4 (0.4 mile downstream from site 3), suggesting that there was longer transport of the populations carrying the16S rRNA gene marker in water samples or that the sensitivity of the CF128F assay was greater than that of the non-16S rRNA gene-based assays. While 16S rRNA gene markers are more sensitive than currently available cattle-specific non-16S rRNA gene markers, 16S rRNA gene-based assays have exhibited nonspecific cross-amplification (7, 11). In contrast, some of the non-16S rRNA gene markers used in this study have shown good host specificity (25); therefore, the use of a combination of 16S rRNA gene and non-16S rRNA gene markers provides a higher level of confidence when unknown sources of fecal pollution in environmental samples are tracked.

BLR analyses performed for Bac32F and CF128F showed that season (a temporal variable) was an important factor affecting the occurrence of Bacteroides species. Moreover, the significance of watershed in the BLRs could suggest that the different management techniques at the two farms contributed to the differential fecal loads and subsequent detection of fecal indicators. The BLR algorithm failed to converge when site grouping was included in the models, so within-watershed spatial sampling effects could not be estimated. Interestingly, BLR analyses revealed that enterococcal density was a significant explanatory variable in the occurrence of Bac32F and CF128F, demonstrating that there was a correlation between the occurrence of the molecular markers and the abundance of a traditional indicator. This information supports the conclusion that source-specific molecular indicators could be used as tools for monitoring impaired waters. Because enterococcal density provides information only for the level of impairment of a body of water (20, 21) and does not identify the source of contamination (24), the combination of molecular and traditional methods can provide more accurate and reliable risk assessment for prevention or mitigation of fecal pollution in waters.

In a GLM analysis in which site grouping was not included, watershed and 3-day precipitation were significant explanatory variables of the natural logarithm of enterococcal density (data not shown). However, every explanatory variable became significant and the model's adjusted R² increased from 0.2 to 0.497 when site grouping was added, suggesting that within-watershed spatial factors are crucial and must be considered in MST studies (Table 6). This analysis also showed that the enterococcal density was lowest in summer, suggesting that intense UV radiation, heat, and dryness negatively affect the survival of the bacteria.

Results of this study provide practical insights into the implementation of MST practices for identifying sources of contamination and reducing fecal pollution sources reaching aquatic resources. First, the proximity and/or the access of cattle to the stream in each watershed and within the same watershed may have resulted in the significant differences in the levels of fecal pollution detected, suggesting that limiting access of cattle to streams helps protect stream water quality. Second, the lower densities of enterococci and molecular markers in ponds in both watersheds suggest that placing artificial ponds or other types of reservoirs in areas with impacted streams may increase the retention time and could lead to reductions in the levels of fecal indicators and pathogens. Within a reservoir, the reduction in the level of fecal bacterial indicators could be a function of factors such as UV radiation, protozoan grazing, and sedimentation.

Accurately tracking the sources of fecal contamination depends on several factors intrinsic to the method used, including the host specificity of targeted markers and the assay detection limits. Our results indicate that environmental factors also can be relevant to the success of source-tracking activities. Statistical analyses indicated that there was significant watershed, within-watershed, and seasonal variability in the occurrence of the markers used and the abundance of a traditional fecal indicator. Different management practices, watershed land use, stream topography, and measurement of selected environmental parameters are among the critical factors to consider when source-specific marker performance is evaluated, and these variables should be examined before a source-tracking method is used in an area of interest.