Microbial Source Tracking Using Quantitative and Digital PCR To Identify Sources of Fecal Contamination in Stormwater, River Water, and Beach Water in a Great Lakes Area of Concern

Fecal contamination of recreational water poses a persistent and ongoing problem, particularly in areas of concern around the Great Lakes. The identification of the source(s) of fecal contamination is essential for safeguarding public health as well as guiding remediation efforts; however, fecal contamination may frequently be present at low levels and remain undetectable by certain methodologies. In this study, we utilized microbial source tracking techniques using both quantitative and digital PCR assays to identify sources of contamination. Our results indicated high levels of human fecal contamination within stormwater outfalls, while lower levels were observed throughout the watershed. Additionally, high levels of gull fecal contamination were detected at Rouge Beach, particularly during drier sampling events. Furthermore, our results indicated an increased sensitivity of the digital PCR assay to detect both human and gull contamination, suggesting it could be a viable tool for future microbial source tracking studies.

ABSTRACT

Areas of concern (AOCs) around the Great Lakes are characterized by historic and ongoing problems with microbial water quality, leading to beneficial use impairments (BUIs) such as beach postings and closures. In this study, we assessed river and beach sites within the Rouge River watershed, associated stormwater outfalls, and at Rouge Beach. The concentrations of Escherichia coli as well as human- and gull-specific qPCR microbial source tracking (MST) markers were assessed at all sites. A preliminary comparison of digital PCR (dPCR) methodologies for both MST markers was conducted regarding sensitivity and specificity. Within the watershed, the outfalls were found to be a prominent source of human fecal contamination, with two outfalls particularly affected by sewage cross-connections. However, the occurrence of human fecal contamination along Rouge Beach and in the lower portions of the watershed was largely dependent on rain events. Gull fecal contamination was the predominant source of contamination at the beach, particularly during dry weather. The multiplex human/gull dPCR methodology used in this study tended to be more sensitive than the individual quantitative PCR (qPCR) assays, with only a slight decrease in specificity. Both dPCR and qPCR methodologies identified the same predominance of human and gull markers in stormwater and beach locations, respectively; however, the dPCR multiplex assay was more sensitive and capable of detecting fecal contamination that was undetected by qPCR assays. These results demonstrate the dPCR assay used in this study could be a viable tool for MST studies to increase the ability to identify low levels of fecal contamination.

IMPORTANCE Fecal contamination of recreational water poses a persistent and ongoing problem, particularly in areas of concern around the Great Lakes. The identification of the source(s) of fecal contamination is essential for safeguarding public health as well as guiding remediation efforts; however, fecal contamination may frequently be present at low levels and remain undetectable by certain methodologies. In this study, we utilized microbial source tracking techniques using both quantitative and digital PCR assays to identify sources of contamination. Our results indicated high levels of human fecal contamination within stormwater outfalls, while lower levels were observed throughout the watershed. Additionally, high levels of gull fecal contamination were detected at Rouge Beach, particularly during drier sampling events. Furthermore, our results indicated an increased sensitivity of the digital PCR assay to detect both human and gull contamination, suggesting it could be a viable tool for future microbial source tracking studies.

INTRODUCTION

For regions around the Great Lakes designated areas of concern (AOCs), understanding the source of fecal contamination is of great importance. AOCs are characterized by historic and ongoing problems with microbial water quality leading to beneficial use impairments (BUIs), such as beach postings and closures, which can lead to recreational and economic losses (1) and increased risks to public health (2, 3). Traditionally, microbial water quality is assessed via the quantification of fecal indicator bacteria (FIB), such as Escherichia coli (4, 5). However, the quantification of FIB alone provides no information relating to the source(s) of fecal contamination, potentially frustrating efforts at remediation. Furthermore, different sources of contamination can pose different levels of risk to human health, with human sewage generally considered to be the most harmful (6).

To combat the deficiencies of solely detecting FIB, microbial source tracking (MST) methods have become widely used in field studies, using host-associated DNA markers to identify multiple different sources of fecal contamination, such as from humans, cows, gulls, and dogs (7–12). Additionally, many MST methods are now available as quantitative PCR (qPCR) assays, enabling the rapid quantification of the source(s) of contamination and providing information considerably more quickly than traditional culture-based methods. However, qPCR methods also suffer from several limitations, primarily, their reliance on standard references to interpolate the quantification of the target (13). This is increasingly problematic, as recent studies have shown that the standard reference material used can provide different results when procured from different vendors or from different batches from the same vendor (14, 15). Additionally, qPCR assays are subject to inhibition from common constituents in environmental matrices, impeding accurate quantification (16).

An emerging technology, digital PCR (dPCR) enables the absolute quantification of nucleic acids independent of a standard reference material. In dPCR, the sample is partitioned into tens of thousands of reactions, with the DNA target present in some reactions but not in others. Poisson statistics are then used to estimate the concentration of target DNA copies (13, 17, 18). Therefore, in addition to providing absolute rather than relative quantification, dPCR assays should be more robust in the face of environmental inhibition and face fewer issues relating to substrate competition, enabling a more effective use of multiplex assays (19, 20).

In this study, we assessed the concentrations of E. coli and sources of contamination, via MST qPCR assays for human- and gull-specific contamination, within the Rouge River watershed and Rouge Beach (Toronto, ON, Canada) (Fig. 1). Additionally, for a subset of the samples, a comparison between qPCR and dPCR human and gull assays was conducted, as well as fecal validation from known fecal sources and wastewater treatment plants (WWTPs) within the Toronto region. We hypothesized that the watershed would exhibit problems similar to those of other watersheds within the region, with high concentrations of human contamination in the stormwater outfalls and high concentrations of gull contamination along the beach. Furthermore, we hypothesized that dPCR analysis would be more sensitive and able to better identify low levels of host-associated contamination within the watershed. The identification of the source(s) of elevated E. coli concentrations at Rouge Beach is needed to guide remediation efforts to reduce beach postings and remedy a BUI within the Toronto AOC.

FIG 1.

Map of the Rouge River watershed with marked sampling sites in the full watershed (A) and the lower end of the watershed (B). This map was created using ArcGIS updated in April 2017.

RESULTS

Escherichia coli enumeration.

An analysis of variance (ANOVA) revealed a main effect of site type (river, beach, or outfall) on E. coli concentrations (F_2,290 = 51.48, P < 0.005), with outfalls having significantly higher concentrations than either river or beach sites (P < 0.001 for both) (Fig. 2). Among the river sites, a significant main effect of sampling site was detected (F_10,125 = 4.48, P < 0.005). River sites AG, MT, and 401 had the highest E. coli concentrations, which were significantly higher than at sites MC and FM (P ≤ 0.01 for all analyses) (Fig. 2). Among the outfall sites, a significant effect of sampling site was also detected (F_2,27 = 21.04, P < 0.005), with sites RR1 and RR6 having significantly higher concentrations than at the outfall site UR (P ≤ 0.001 for both). Additionally, a significant effect of sampling site was observed among the beach sites (F_8,117 = 30.80, P < 0.001). Sand pore sites had significantly higher concentrations than chest- or ankle-depth sites (P < 0.001 for all analyses) (Fig. 2).

FIG 2.

Box plot of E. coli concentrations at each of the sampling sites, showing the median E. coli concentrations between the 25th and 75th data quartiles; whiskers extend to the outermost data points within ±1.5 data points of the interquartile ranges. Open circles depict outlier values, and asterisks depict extreme values.

Validation of MST qPCR and dPCR markers.

A fecal validation of MST qPCR and dPCR assays revealed that although both host-specific markers exhibited good specificity, neither demonstrated 100% specificity (Table 1). Both qPCR and dPCR MST assays revealed that the gull marker is cross-reactive with Canada goose samples and the human marker is cross-reactive with chicken samples, while the dPCR assay revealed that the human marker is also cross-reactive with cat samples (Table 1). With the exception of high human marker concentrations in one chicken sample, the levels of cross-reactivity among both assays were always orders of magnitudes lower in nontarget host fecal samples than in target host fecal samples (data not shown).

TABLE 1.

Percent detection of qPCR and dPCR MST markers in fecal samples and sewage influent and effluent

Source	Sample size (n)	qPCR (%)		dPCR (%)
		HF183	qGull4	HF183	qGull4
Canada goose	5	0	40	0	60
Cat	5	0	0	20	0
Chicken	5	20	0	20	0
Cow	5	0	0	0	0
Dog	5	0	0	0	0
Gull	5	0	100	0	100
Influent	5	100	0	100	0
Effluent	5	100	0	100	0

A Bayesian analysis revealed slight differences in the probabilities that each of the qPCR and dPCR markers was detecting a true positive. When assessed on the basis of the presence/absence of marker detection, where a sample was counted as a nondetect if the qPCR or dPCR quantification was less than one, the probability of the dPCR MST assay correctly detecting a true positive was calculated to be 93.89% for HF183 and 82.26% for qGull4. With the same presence/absence assessment, the probability of the qPCR MST assay correctly detecting a true positive was calculated to be 95.37% for HF183 and 86.45% for qGull4.

Microbial source tracking.

A multiple analysis of variance (MANOVA) detected a main effect of site type for qPCR marker concentrations (F_4.576 = 131.54, P < 0.001). Tukey's post hoc analyses revealed that outfalls had significantly greater concentrations of the human marker than river or beach sites, whereas beach sites had significantly greater concentrations of the gull marker than river or outfall sites (P < 0.001 for all analyses) (Fig. 3). No significant difference was detected for either marker among river sites. However, among outfall sites, a main effect of sampling site was observed (F_4,53 = 6.65, P < 0.001), with outfall RR1 having significantly greater concentrations of the human marker than outfalls UR or RR6 (P < 0.001 for both). No significant difference was detected for either marker among beach sites; however, when sampling depth was considered (ankle-depth, chest-depth, or pore water samples), a significant effect was observed (F_4,244 = 3.44, P = 0.009), with sand pore samples having significantly higher concentrations of the gull qPCR marker than chest-depth samples (P = 0.005).

FIG 3.

Box plot of MST qPCR marker concentrations at each of sampling sites. White boxes represent the human marker, whereas dark boxes represent the gull marker. The box plot shows the median marker concentrations between the 25th and 75th data quartiles; whiskers extend to the outermost data points within ±1.5 data points of the interquartile ranges. Asterisks depict extreme values, usually associated with rain events.

Similar to that for the qPCR MST marker quantification, a significant main effect of sample type was observed using dPCR marker concentrations as a response variable (F_4,116 = 19.75, P < 0.001). A post hoc analysis also revealed that outfalls had significantly higher concentrations of the human marker than river or beach sites, while beach sites had significantly higher concentrations of the gull marker than river or outfall sites (P ≤ 0.003 for all analyses). No significant difference in either of the marker concentrations was observed among river or beach sampling sites. While the MANOVA did not detect any significant effect of sampling site among outfalls, the univariate analysis (ANOVA) did detect significantly higher concentrations of the human marker at outfall RR1 than at the UR outfall (P = 0.04).

Effect of rain events on E. coli and qPCR marker concentrations.

Sampling dates were considered rain events if more than 5 mm of rain fell up to 72 h prior to sampling, as this cutoff can be used by the city of Toronto to consider potential stormwater responses. When all sites were considered together, rain events had no significant effect on concentrations of E. coli or either MST qPCR marker. However, among river sites, t tests revealed significantly higher concentrations of E. coli and both MST qPCR markers (P ≤ 0.006 for all analyses) (Fig. 4). Among the outfalls, rain events significantly increased the concentrations of the gull qPCR marker (P = 0.04). Lastly, among the beach sites, rain events significantly increased the concentrations of the human qPCR marker and significantly decreased the concentrations of the gull qPCR marker (P ≤ 0.002 for both) (Fig. 4).

FIG 4.

Mean concentrations of E. coli (◽), the gull qPCR marker (▲), and the human qPCR marker (●) ± SEM during dry sampling events (A) and rain events (B).

Comparison of qPCR and dPCR MST markers.

Among all the sampling sites, paired t tests revealed that the concentrations of both the human and gull markers by dPCR were significantly greater than those by qPCR (P < 0.001 for both) (Fig. 5). This relationship was also true for both markers among the river and beach sampling sites (P ≤ 0.04) (Fig. 5). However, among the outfalls, while the concentrations of the dPCR gull marker were significantly higher than those of the qPCR gull marker (P = 0.03), the concentrations of the qPCR human marker were significantly greater than the concentrations of the dPCR human marker (P = 0.01). Among all the sampling sites, the dPCR human marker was detected in 63% of samples in which the qPCR human marker was not detected, while the dPCR gull marker was detected in 74% of samples in which the qPCR gull marker was not detected. Significant correlations among the concentrations of E. coli and all MST markers were also observed and are presented in Table 2. Each qPCR marker had a significant positive correlation with its respective dPCR counterpart, and all markers were significantly and positively correlated with E. coli concentrations (Table 2).

FIG 5.

Mean concentrations comparing qPCR (open symbols) and dPCR (filled symbols) concentrations of the human marker (A) and the gull marker (B) ± SEMs for all sampling sites on two rain events and two dry weather sampling dates (n = 62).

TABLE 2.

Spearman correlation coefficients for concentrations of E. coli and all MST markers

Assay	Marker	Correlation (r)
		E. coli	qPCR		dPCR
			Human	Gull	Human
qPCR	Human	0.63a
	Gull	0.44a	0.22
dPCR	Human	0.70a	0.84a	0.20
	Gull	0.43a	0.16	0.74a	0.16

^aP < 0.05.

DISCUSSION

Sources of contamination in the Rouge River watershed and at Rouge Beach.

The results of this study identified different levels of fecal contamination throughout the Rouge River watershed. As anticipated, the highest concentrations of E. coli and human fecal contamination were present within the outfall sites, with outfalls RR1 and RR6 likely exhibiting problems with sewage cross-connections, while the predominant source of contamination at Rouge Beach was observed to be from gull contamination. These results are consistent with what has been found within other watersheds within the Toronto and region AOCs in recent years (7, 21). However, the sources and levels of contamination throughout the river sampling sites varied throughout the watershed and, within the lower watershed and toward Rouge Beach, were more highly impacted by rain events.

River sites toward the extremes of both the upper (sites AG, MT, and LW) and lower watersheds (401, WE, and R0) tended to have the greatest concentrations of E. coli and exceeded the guidelines for recreational water of 100 CFU · 100 ml⁻¹ (22) in greater than half of the sampling events. While these sites had the greatest concentrations of E. coli, the sources of fecal contamination were likely different. The sampling sites on the main branch of the upper Rouge River (AG and MT) experienced some level of human contamination, consistent with more urban land use within this section of the watershed (23). Conversely, despite relatively high E. coli concentrations, site LW had no traces of human contamination when assessed via qPCR and low levels (less than 50 copy numbers [CN] · 100 ml⁻¹) when assessed via dPCR. Site LW is situated in the upstream portion of Little Rouge Creek, which tends to have more rural or agricultural land use (23), making it likely that the contamination at this site was due to wildlife inputs.

The concentrations of E. coli and human contamination tended to decline toward the middle sections of the river, becoming elevated once again at site 401, below the confluence of the main branch of the Rouge River and Little Rouge Creek. Additionally, site 401 is immediately downstream of outfall RR6 and a camp site situated around sampling site GR1. Consequently, the high concentration of the human marker detected within outfall RR6 was likely a contributor to the increased concentrations of E. coli and the human marker detected at site 401. Human contamination from this source could have persisted further downstream, contributing to the relatively elevated concentrations of E. coli and the human marker detected within site WE and at the river mouth (site R0).

Human contamination present within the lower watershed was also highly dependent on rain events. The human qPCR marker was almost exclusively detected, in both the middle Rouge River sites as well as at Rouge Beach ankle- and chest-depth sites, during rain events (Fig. 4). Additionally, rain events increased human marker concentrations, particularly at sites 401 and at the river mouth. This is likely due to contamination from outfall RR1, which had the highest concentrations of E. coli and the human marker throughout the duration of the study. Outfall RR1 was not situated immediately adjacent to the surface waters and generally had minimal flow during dry sampling events, although high flow was observed during rain events. During wet weather events, it is likely that the high levels of human contamination from this outfall were flushed into site WE and further downstream toward the river mouth and ultimately Lake Ontario and the Rouge Beach sites.

The Rouge Beach sites, similar to that for the river sites in the lower watershed, were highly influenced by rain events. During dry sampling events, E. coli concentrations at ankle- and chest-depth sampling sites were almost always under 100 CFU · 100 ml⁻¹ Furthermore, the primary source of contamination during dry weather events was predominantly from gull fecal contamination, consistent with previous studies of gull contamination within the region (24). Nearly all exceedances of regulatory guidelines and human marker detection coincided with rain events, with the highest concentrations of both E. coli and the human marker at transect RB1 and decreasing with greater distance from the river mouth. The fact that the qPCR human marker was never detected within pore samples at Rouge Beach (and only in low concentrations via dPCR assessment) suggests that the Rouge River was the likely source of human fecal contamination and that it was not resident at the beach.

Comparison of qPCR and dPCR methodologies.

Of particular interest in this study was the preliminary comparison between qPCR and dPCR MST markers. While several studies have used droplet digital PCR (ddPCR) to assess concentrations of FIB, particularly enterococci, and human-specific fecal markers and make comparisons with qPCR quantification results, these studies tended to focus on marine waters and stormwater discharges (13, 17, 18, 25). Our study is among the first to use dPCR (as opposed to water-in-droplet ddPCR) as part of a full-scale field study, utilizing a variety of sample types (freshwater river samples, stormwater samples, and beach samples) and employing a multiplexed MST assay.

The overall results obtained via dPCR assessment were in agreement with those obtained via qPCR. Both methods identified outfalls as having the highest concentrations of the human marker, while the beach sites had the highest concentrations of the gull marker. Furthermore, the same sites that tended to have the highest concentrations of the human marker within sample types were comparable by both qPCR and dPCR methods (Fig. 5). On the basis of the general agreement between these methodologies, we conclude from this preliminary assessment that our multiplex human/gull dPCR assay could be a viable tool for future MST field studies. However, it should be noted that our results were based on only 4 sampling events wherein qPCR and dPCR results were compared, and further assessment will be required to understand the nature of the specific differences we found between the two methods.

We observed differences between the qPCR and dPCR methods with regard to quantification, sensitivity, and specificity. While dPCR generally resulted in the detection of greater concentrations of both human and gull markers, this was not the case with all individual samples. As noted with outfall sites, dPCR quantification resulted in significantly lower concentrations of the human marker than the qPCR concentrations. One possible explanation for this could be an overestimation of the qPCR marker concentrations as a consequence of inaccurate reference material used for the standard curve. Furthermore, the upper limit of quantification for dPCR, which requires the dilution of samples with high concentrations of the target marker, is considerably less than that for qPCR. These necessary dilutions for outfall sites may have resulted in an underestimation of the human marker concentrations when assessed via dPCR. The lower dynamic range of dPCR than qPCR may be a limitation in these situations, requiring the consideration of additional dilutions and testing for dPCR.

While qPCR may have resulted in an overestimation or dPCR may have underestimated the absolute marker concentrations when present at high levels, dPCR was more sensitive than qPCR. As noted, 63% and 74% of water samples that resulted in a nondetect for the qPCR human and gull markers, respectively, had positive detections, albeit at low concentrations, when assessed via dPCR. These results suggest that our dPCR assay is able to detect marker concentrations below the limit of detection of the human and gull qPCR assays employed, potentially making it a valuable tool for field sites experiencing low levels of contamination. It is possible that aspects such as the partitioning of water samples in dPCR could reduce concerns about PCR inhibition and contribute toward a better detection of targets at lower concentrations (13).

These results may need to be taken with caution, though, as a Bayesian analysis of both methodologies for both markers revealed a slightly lower probability of our dPCR assay detecting a true positive than the qPCR assays employed. It is possible that the increased sensitivity of a dPCR assay may also more readily detect low concentrations of host-specific markers in nonhost fecal sources. However, in our study, these differences stem mostly from the detection of the gull marker in one more Canadian goose sample and of the human marker in one cat sample that was not detected via qPCR assessment. The authors of a previous multilab study noted that gull markers can cross-react with other bird species, particularly waterfowl (26). Additionally, due to the close association between humans and cats as domestic pets, Bacteroides strains from the human gut could be transiently shared with cats, or strains in the cat gut microbiome may have coevolved to be cross-reactive with human-specific markers. It should be noted that specific Bacteroides strains in the human gut could be shared, resulting in small levels of cross-contamination in domestic pets or even farm animals in close contact with humans, albeit usually in concentrations orders of magnitude lower than those found in humans (27, 28). While our dPCR and qPCR primer and probe sequences were identical, and had similar cross-reactivity patterns consistent with those from previous studies (7), our fecal validation was based only on 5 samples from each source. Further research should be conducted to better assess the specificity of this dPCR assay and any implications from an increased sensitivity. It would be worthwhile to also investigate alternatives to Bacteroides-based MST markers, which may be more specific for humans, such as Faecalbacterium prausnitzii or Eubacterium rectale (29, 30).

The practicality of more sensitive detection via dPCR is also an issue that may need to be addressed from a human health and risk management perspective. In ∼19% of the samples wherein both qPCR and dPCR assays were conducted, both qPCR and dPCR detected human contamination, while only dPCR detected gull contamination at very low levels. Furthermore, in ∼8% of samples, both qPCR and dPCR assays detected gull contamination, whereas only dPCR detected low levels of human contamination. While the detection of low levels of human contamination may be more relevant from a public health perspective, further research needs to be conducted to determine what level of human contamination is necessary to constitute an actual threat to public health and worthy of triggering management or remedial actions.

The results of this study suggest that stormwater outfalls within the Rouge River watershed are a prominent source of human fecal contamination in the river, particularly in the lower portions of the watershed and near Rouge Beach. However, elevated E. coli concentrations and associated human fecal contamination at the beach seemed to be present almost exclusively during rain events in our study year. Gull fecal contamination was usually predominant at beach sites, particularly during dry weather. While dPCR detected significantly higher concentrations of gull markers at several sites, particularly in the upper watershed, these results need to be considered with caution, as mentioned before, as this may be indicative of cross-contamination with other bird species; additionally, the concentrations of the gull marker were also generally low, raising the question of whether remediation efforts for gulls at these sites would be worthwhile. Consequently, remediation efforts will need to address problem stormwater outfalls, notably RR1 and RR6, as well as gull contamination at the beach. A rain rule to prevent swimming would seem to be advisable to reduce health risks from sewage contamination. Additionally, the result of this study demonstrated the efficacy of a human/gull dPCR methodology to aid in the assessment of sources of fecal contamination within urban watersheds. This dPCR multiplex assay was more sensitive and capable of detecting fecal contamination that was often undetected by qPCR assays. Furthermore, the cost per multiplexed dPCR was equivalent to the cost of running single-target qPCRs for two targets, making dPCR a cost-effective alternative to qPCR. However, as there are still relatively few studies which have assessed dPCR assays in field studies, further research needs to be conducted to verify the robustness of the results (i.e., from different vendors and with different primer/probe chemistries). Therefore, continued research to evaluate dPCR methods could be useful to develop new MST tools and benefit efforts to remediate BUIs at impaired beaches and better safeguard public health.

MATERIALS AND METHODS

Study area and sample collection.

This study was conducted within the Rouge River watershed and at Rouge Beach in the Toronto AOC. The Rouge River spans 336 km² and land use consists of rural (40%), urban (35%), forest/wetland/meadow (24%), and open water body (1%) (23). The Rouge River combines with the Little Rouge Creek from the east, ultimately flowing into Lake Ontario, with Rouge Beach located at the mouth of the river. Of the urbanized areas surrounding the Rouge River, 20% have no stormwater controls (31). The sampling sites and relative locations can be found in Table S1 in the supplemental material and in Fig. 1.

Water samples were collected weekly from June to August 2016. The water samples from the Rouge River and associated stormwater outfalls were grab samples collected in 500-ml autoclaved polypropylene bottles. At three transects along Rouge Beach, surface water samples were collected at increasing distances from the shoreline at ankle depth (approximately 10 cm) and chest depth (approximately 1.2 m) within the lake, along with a sand pore water sample. To collect the sand pore water sample, a hole was dug down to the water table in the foreshore sand approximately 1 m inland from the lake, and a 250-ml polypropylene bottle was inserted into the hole to collect the water that accumulated (while minimizing sand collection). All water samples were placed on ice and transported to the laboratory for processing within 6 h of collection. Data on rainfall the day of sampling and 24, 48, and 72 h prior to sampling was obtained from the Pearson International Airport site on Environment and Climate Change Canada's website (http://climate.weather.gc.ca/historical_data/search_historic_data_e.html).

E. coli enumeration and DNA extraction.

Water samples were filtered (0.45-μm pore size, 47-mm diameter) over a range of dilutions according to standard membrane filtration methods (32), and E. coli were enumerated on differential coliform (DC) medium supplemented with cefsulodin. E. coli were incubated at 44.5°C for 22 h, and the results are reported as CFU · 100 ml⁻¹. For DNA extraction, an additional 300 ml (100 ml for pore water samples) was filtered as above. The filters were frozen for no more than 1 week at −80°C until ready for DNA extraction. The filters were then folded and placed into Powerbead tubes and extracted using Powersoil DNA isolation kits (Mo Bio Laboratories, Inc., Carlsbad, CA, USA) according to the manufacturer's instructions. Extraction blanks were included in every batch of DNA extractions.

MST qPCR and dPCR.

Assays for qPCR used previously published primer/probe sets for the human HF183 marker and the gull qGull4 marker (33, 34). Each qPCR mixture consisted of 2 μl of an internal amplification control (IAC), 2.5 μl 2 mg/ml bovine serum albumin (BSA), 3 μl nuclease-free water, 12.5 μl TaqMan universal master mix 2.0 (Thermo Fisher, Waltham, MA, USA), 3 μl of a primer/probe mixture (100 μM for both primers and probe), and 2 μl of extracted DNA. The reactions were carried out in 96-well plates using a Bio-Rad CFX96 cycler (Hercules, CA, USA). All reactions were carried out in duplicates, including no template controls, negative controls consisting of 2 μl salmon testes DNA, and positive controls consisting of 2 μl of DNA extracted from a known fecal source. Thermocycler settings were 50°C for 2 min, 95°C for 10 min, and 40 cycles of 95°C for 15 s and 60°C for 1 min for the human marker. Thermocycler settings for the gull qPCR assay were 95°C for 5 min and 45 cycles of 95°C for 15 s and 60°C for 30 s. All qPCR results are reported as CN · 100 ml⁻¹.

Standard curves for all qPCR assays were constructed using synthesized plasmid DNA (pIDTSMART with ampicillin resistance; Integrated DNA Technologies, Coralville, IA, USA). The DNA used for the standard curve was serially diluted using AE buffer (Qiagen, Valencia, CA, USA) to concentrations ranging from 10² to 10⁵ gene copies/reaction. The DNA used for the IAC was similarly constructed using synthesized plasmid DNA (pIDTSMART with ampicillin resistance; Integrated DNA Technologies, Coralville, IA, USA) with complementary primer sites for each assay and included in every reaction to verify that there was no inhibition from the ambient water matrices. All qPCR runs had an efficiency between 90% and 110% with an R² of >0.95, and the results were normalized to the reaction efficiency.

A subset of samples (62 samples) was also assessed via dPCR. These samples included all river sites except for AG, all outfall sites, and only transect 1 (RB1) of the Rouge Beach samples on two rain event sampling days (13 June and 16 August 2016) and two dry weather sampling days (5 July and 24 August 2016). Digital PCRs were conducted as a multiplex assay for human and gull markers, using the same primer/probe sets as the qPCR assays but with the exception that the gull probe was labeled with a 5′-VIC fluorophore. Each dPCR mixture consisted of 1 μl nuclease-free water, 0.75 μl of each 900 nM forward and reverse primer, 0.75 μl of each 250 nM probe, 7.5 μl of QuantStudio 3D digital PCR master mix v.2 (Thermo Fisher), and 2 μl of extracted DNA. The reaction mixtures were loaded into a QuantStudio 3D digital PCR 20K chip v2 (Thermo Fisher) using a QuantStudio 3D chip loader (Thermo Fisher), and the reactions were carried out using a ProFlex thermocycler PCR system (Thermo Fisher). The thermocycler settings were 96°C for 10 min, followed by 40 cycles of 60°C for 2 min and 98°C for 30 s, and then 60°C for 2 min. The chips were then read using a QuantStudio 3D digital PCR instrument (Thermo Fisher). The results were analyzed using the QuantStudio AnalysisSuite, which determined threshold fluorescence values for the ROX reference dye to identify the number of PCR partitions, as well as thresholds for the 6-carboxyfluorescein (FAM) and VIC signals to identify positive reactions for the human and gull marker, respectively. The software then applied a Poisson Plus modeling technique to determine the concentration of each target within the sample. The software results were visually inspected and are reported as CN · 100 ml⁻¹.

qPCR/dPCR fecal validation.

To compare the levels of sensitivity and specificity between qPCR and dPCR assays, both assays were performed with the extracted DNA from 40 fecal samples from various known sources and sewage samples from across southern Ontario. For sewage samples from Toronto's Ashbridge's Bay wastewater treatment plant, Highland Creek wastewater treatment plant, Humber wastewater treatment plant, and North Toronto wastewater treatment plant, 50 ml of both influent and effluent was filtered and DNA was extracted as described above. For five fecal samples from each animal source (Table 1), ∼0.3 g was weighed into a MoBio Powerbead tube and DNA was extracted as described above.

Statistical analysis.

E. coli concentrations (CFU · 100 ml⁻¹), qPCR quantification of MST markers (CN · 100 ml⁻¹), and dPCR quantification of MST markers (CN · 100 ml⁻¹) were log transformed prior to analysis. ANOVA was used to determine the main effect of sample type (river, outfall, or beach), where response variables were concentrations of E. coli. Similarly, ANOVA was also conducted to determine whether there was a main effect of sampling site within each site type. ANOVA was also used to assess differences in E. coli concentrations in sublocations (ankle-depth, chest-depth, and pore water site samples) among beach sites. A MANOVA was used to assess the main effect of sample type, where response variables were either qPCR human and gull marker concentrations or dPCR human and gull marker concentrations. As for the analysis with E. coli, MANOVA was also performed to assess differences among sampling sites within each sample type for MST qPCR or dPCR markers. Tukey's post hoc test was performed if a significant effect was detected for all the analyses described above.

Independent t tests were used to assess the differences in E. coli and MST marker concentrations between rain and dry sampling events. Paired t tests were used to assess the differences between qPCR and dPCR marker concentrations. The results from the 62 dPCR samples were only compared to those from equivalent qPCR samples. Spearman correlations were used to determine the relationships between E. coli and MST (qPCR and dPCR) marker concentrations and rainfall. For fecal validation of the qPCR and dPCR markers, Bayes' theorem was applied to calculate the conditional probability that any detection of host-specific qPCR or dPCR markers was the result of a true positive (35). All analyses were performed in Statistica v.12 (Dell, Tulsa, OK) and considered significant at the α level of 0.05.