Many people living in rural areas rely on privately owned wells as their primary source of drinking water. These water sources are at risk for fecal contamination of human, wildlife, and livestock origin. While traditional bacteriological testing involves culture-based methods, microbial source tracking (MST) assays present an opportunity to additionally determine the source of fecal contamination. This study investigated the main host sources of contamination in private well water samples with high levels of Escherichia coli (E. coli), using MST with human and multi-species specific markers. Fecal contamination of human origin was detected in approximately 50% of samples, indicating that current contamination prevention strategies require reconsideration. The relationship between cattle density and fecal contamination of bovine origin was investigated using a Bovine Bacteroidales specific MST assay. Regional variations of microbial sources were examined, and may inform local primary prevention strategies. Additionally, in order to assess MST and E. coli quantitative real time polymerase chain reaction (qPCR) assays as indicators of fecal contamination, these were compared to E. coli culture methods. Variation in results was observed across all assay methods investigated, suggesting the most appropriate routine bacteriological testing methodology cannot be determined without comparison to a method that directly detects the presence of fecal contamination.

INTRODUCTION

Globally, millions of people in developing and affluent nations rely on private groundwater wells as their primary drinking water source. It is well established that aquifers supplying these wells can be become contaminated with human and/or animal feces, which may contain pathogens leading to serious human health risks such as gastrointestinal (GI) illness (Charrois 2010). Microbial source tracking (MST) methods can be used as delineators of fecal contamination through attribution to specific host species, potentially identifying the origin of the contamination and thereby allowing for targeted prevention and remediation strategies. Traditionally, MST methods have been used to evaluate fecal pollution of localized surface waters, including: storm, stream and recreational water sources (Lee et al. 2010; Sauer et al. 2011; Haack et al. 2013; Krentz et al. 2013). However, to date, few studies have utilized these techniques in groundwater derived drinking water, particularly from privately owned wells.

At this time, it is not feasible to directly detect pathogens from water samples due to their low levels, sporadic and erratic presence, and the complex methodologies required to isolate them (Cabral 2010). Traditionally, MST methods are not used as routine measures to indicate fecal contamination. Rather, the current standard of practice uses fecal indicator bacteria (FIB) as markers of fecal contamination when testing environmental water supplies for potability, given that normal intestinal inhabitants are typically present alongside certain pathogenic species (Leclerc et al. 2001). A recent systematic review by Cabral (2010) defined six (three major and three minor) ideal criteria for a good bacterial fecal indicator. The first three require that the FIB have a high concentration in feces, are generally non-pathogenic to humans, and are amenable to cost-effective and timely detection techniques. In addition, the indicator organism should not reproduce external to the intestinal tract, be present in higher quantities than associated pathogens in water and possess a similar decay rate as the pathogens. Utilizing these criteria, previous investigations have determined that, of all intestinal flora, Escherichia coli (E. coli) remains the ideal, reliable fecal pollution indicator for environmental waters (Leclerc et al. 2001).

While utilizing E. coli as FIB is the current best practice, weaknesses have been recently identified. For example, novel evidence documents environmental E. coli populations that originated from animal intestines and adapted to proliferate in environmental conditions (Perchec-Merien & Lewis 2012). Moreover, E. coli has been documented to enter into a viable but non-culturable state, which may yield false negative results during water testing (Liu et al. 2008; Lothigus et al. 2009). In general, the use of FIB does not reveal possible sources of fecal contamination and the use of MST methods may provide insight into these potential contributing host species. Additionally, the use of MST assays alongside traditional FIB will allow for the evaluation of MST as a potential indicator of fecal contamination.

In Ontario, three distinct areas with a higher burden of E. coli contamination among private wells have been identified (Krolik et al. 2013). In one of these regions, the possible origins of fecal contamination were investigated using MST methods (Krolik et al. 2014). Half of the samples had a positive signal for human markers, suggesting septic tanks as the potential origin. In this study, two additional areas with higher E. coli contamination rates were investigated, as well as a region with high cattle density. Traditional culture methods, MST assays using quantitative real time polymerase chain reaction (qPCR) and E. coli qPCR were performed. The specific aims were to: (1a) determine the predominant host-specific origins of contamination among private well water samples from the remaining two regions where burden of E. coli contamination is greater, (1b) determine whether a higher cattle density region exhibits higher levels of bovine contamination in private well water samples, (1c) look for general regional variations in predominant primary host contamination; and (2) evaluate, through comparison, MST and E. coli qPCR assays alongside E. coli culture methods, as indicators of fecal contamination.

MATERIALS AND METHODS

Study area and sample sets

Multiple convenience samples of private water supplies (predominantly private wells), collected over multiple years and regions, were utilized in this investigation. The first group was comprised of waters submitted for bacteriological analysis to the Public Health Ontario Laboratory in 2012 from Region A as described in Figure 1, and which tested positive for E. coli. The 2013 group consisted of water samples submitted during the summer period from Region A, and which tested negative for E. coli. The final sample set, from 2014, included submissions originating from the other two regions with higher E. coli contamination (Figure 1); namely Regions B and C (Region A1 was investigated previously). Additionally, samples from an area with high cattle density were included (Region D) (Agriculture Survey Stats Canada 2011).
Figure 1

Map of sample collection sites: A. Submission catchment for 2012 and 2013 samples; A1. Subset of A that was an area of high E. coli contamination. B and C. Areas of high E. coli contamination. D. High cattle density region. Map of southern Ontario provided in bottom right demonstrates original regions of elevated risk of E. coli contamination in private water submissions (Krolik et al. 2013), which were used to define collection sites A1, B and C.

Figure 1

Map of sample collection sites: A. Submission catchment for 2012 and 2013 samples; A1. Subset of A that was an area of high E. coli contamination. B and C. Areas of high E. coli contamination. D. High cattle density region. Map of southern Ontario provided in bottom right demonstrates original regions of elevated risk of E. coli contamination in private water submissions (Krolik et al. 2013), which were used to define collection sites A1, B and C.

Sample processing and nucleic acid extraction

For bacteriological analysis, 100 mL of sample water was filtered using a 0.45 μm pore size mixed cellulose esters filter (Millipore Billerica, MA) using a partial vacuum. E. coli colony forming units (CFUs) were enumerated from each filter, after 24 h ± 2 h at 35 °C incubation on differential coliform (DC) media (Oxoid, ON) with 5-bromo-4-chloro-3-indolyl-D-glucuronide (BCIG). Colonies are colored blue when BCIG is released into the medium via glucuronidase activity, as it is insoluble, accumulating within the cell; these cells are considered positive for E. coli (Ogden & Watt 1991). Additionally, total coliforms (TC), which include environmental and gut microbiome bacteria, appear pink on DC media (Oxoid, ON). E. coli and TC culture results are counted up to 80 CFU/100 mL, after which they are designated overgrown. Additionally, below or equal to 15 CFU/100 mL the method is considered not quantifiable (ISO 2000).

A second 100 mL of water was filtered through a mixed cellulose esters filter (Millipore Billerica, MA), rolled and placed in 2 mL of Lysis Buffer (bioMérieux, QC) for 1 h at 37 °C on rotation. The DNA was extracted using a magnetic silica bead system (bioMérieux, QC) and eluted to either a 60 μL of a 100 μL volume. The difference in final volumes was accounted for when results were expressed in cells per 100 mL. Several collection sites performed membrane filtration, bacteriological analysis and DNA extraction occurred within 24 hours of collection. DNA samples were frozen at −80 °C and shipped on dry ice to a single site for all qPCR analysis.

qPCR assays – Bacteroidales and E. coli

The MST assay used in this study was previously described by Lee et al. (2010) for surface water, and subsequently applied to private well water by Krolik et al. (2014). Briefly, the MST assay is comprised of three singleplex assays: a human specific Bacteroidales assay, a bovine specific Bacteroidales assay, and a general Bacteroidales assay, all of which target the 16S rRNA gene of Bacteroidales and are, henceforth, referred to as BacHuman, BacBovine and BacGeneral, respectively. The BacGeneral assay is designed to detect this gene in a broad array of hosts (human, cow, pig, deer, horse, dog, cat, gull, goose and raccoon). Assay primer and Taqman probes, as well as reaction conditions, have been previously described (Lee et al. 2010; Krolik et al. 2014). It has been shown that an adjusted limit of detection (LOD) of 10 gene copies per reaction is appropriate (Krolik et al. 2014). The EC23S857 E. coli assay was previously described by Chern et al. (2011) and targets the 23S rRNA gene region. It is a Taqman assay with primer and probe concentrations of 1 μmol/L and 80 nmol/L, respectively, and 5 μL of DNA, for a total volume of 25 μL. An initial incubation at 50 °C for 2 min followed by 95 °C for 10 min was followed by 40 PCR cycles of 95 °C for 15 s and 60 °C for 60 s. All assays were performed using Environmental Mix 2.0 Master Mix on the ViiA7 platform (Thermofischer Scientific, ON). The master mix was chosen to accommodate potential presence of inhibitors and is designed to analyze environmental samples (Thermofischer Scientific, ON).

Standard curves for MST assays were created as previously described (Krolik et al. 2014) using plasmids with concentrations ranging from 6.5 million to 6.5 gene copies per reaction. The E. coli standards were prepared using genomic DNA from ATCC (American Type Culture Collection) strain 25922 and ranged from 1 million to 1 cell per reaction. Subsequently E. coli standards, to a limit of 1 gene copy per reaction, were run independently.

Data manipulation and statistical analyses

The seven datasets (spanning the three sample collection years) that were used to investigate assay relationships are shown in Table 1.

Table 1

Dataset descriptors

Dataset CFU/100 mL result n Region 
A_2012 E. coli positive 716 (MST assay) 706 (E. coli assay) Any sample submitted to Region A in 2012 
A1_2012 E. coli positive 102 Any sample submitted to Region A1 in June–August 2012 (subset of A_2012) 
A_2013_NEG Negative 93 Any sample submitted to Region A in 2013 
A_2013_TC TC positive 73 Any sample submitted to Region A in 2013 
B_2014 E. coli positive 66 Any sample submitted to Region B in 2014 
C_2014 E. coli positive 129 Any sample submitted to Region C in 2014 
D_2014 E. coli positive 59 Any sample submitted to Region D in 2014 
Dataset CFU/100 mL result n Region 
A_2012 E. coli positive 716 (MST assay) 706 (E. coli assay) Any sample submitted to Region A in 2012 
A1_2012 E. coli positive 102 Any sample submitted to Region A1 in June–August 2012 (subset of A_2012) 
A_2013_NEG Negative 93 Any sample submitted to Region A in 2013 
A_2013_TC TC positive 73 Any sample submitted to Region A in 2013 
B_2014 E. coli positive 66 Any sample submitted to Region B in 2014 
C_2014 E. coli positive 129 Any sample submitted to Region C in 2014 
D_2014 E. coli positive 59 Any sample submitted to Region D in 2014 

MST and E. coli qPCR analyses were performed on all samples with the exception of A_2012. A_2012 represents submissions from a full year; the rest of the datasets were captured for June–August period.

Based on which species specific markers were detected (for BacGeneral, BacBovine and BacHuman MST assays), each sample was assigned a status – positive for general markers only, positive for human markers only, positive for bovine markers only, positive for both human and bovine markers, or negative for all assays. These statuses were used to tabulate contributing host-distributions for each region. To further illuminate the predominant fecal source, Spearman correlations between host specific markers and general markers were performed for each of the four regions as well as for the A_2012 dataset.

Previous findings (Krolik et al. 2014) determined that human sourced contamination was the predominant source for Region A. The MST host abundance results were divided into two categories: human sourced and non-human sourced. In order to investigate geographic differences in fecal origins, a Chi-square test was used to detect significant differences across the four regions. Subsequently, the Benjamin, Hochbery and Yekutieli (BHY) post-hoc comparison was used to determine where the differences occurred.

A possible relationship between the regional rates of human sourced fecal contamination and the density of septic systems, a previously hypothesized source of said human fecal contamination, was explored. Unfortunately, septic density is not available as records are not maintained provincially, and thus, a proxy measure – domestic well density – was utilized. The assumption is that owners of a private well would also require private wastewater disposal, and therefore a 1:1 ratio is reasonable. Using the boundaries developed for sample collection, each of the four regions was assigned a set of Statistics Canada-developed dissemination areas (DAs) (small, relatively stable geographic census units) (Statistics Canada 2011). Well densities were determined using well records from the Well Water Information System (WWIS 2013) and expressed as number of domestic-use wells divided by DA area (km2) using ArcGIS 10.2 (ESRI Inc., Redlands, USA). For each region, a median across all the DAs of wells/km2 was calculated, transformed logarithmically and graphically visualized.

In order to compare the three fecal indicator methods (E. coli culture, E. coli and MST qPCR), two different statistical analyses were used. The comparison between CFU and PCR tests required categorical analysis given the majority of samples were within the non-quantifiable CFU range. As such, Chi-square tests with odds ratios were utilized, where the break between categorical groups was determined by the quantifiable limits. Quantifiable ranges existed for both qPCR assays; Spearman correlations were used to investigate possible associations as the data did not follow a normal distribution. All statistical analyses were performed in R 3.2.0 (R Core Team, Vienna).

RESULTS

The MST assays have a natural cut-off of 10 gene copies per reaction as previously published (Krolik et al. 2014); this was confirmed with pooled 2014 data (data not shown). The E. coli assay has a LOD of 1 cell (7 gene copies) per reaction. Both qPCR assays, for all tested samples, had a PCR efficiency between 90 and 110% and an R2 value above 0.995.

Host-specific contamination

In this analysis, the previously published 2012 MST dataset was limited to samples from the region of elevated risk for E. coli contamination and the summer months, in order to most closely match samples from the other regions (Region A1, n = 103 for June–August). Approximately 50% of these samples contained human fecal contamination. The host (human, bovine and general) distributions for this data alongside the other three areas B, C and D are shown in Figure 2.
Figure 2

Distribution of microbial sources stratified by region. Datasets: A1_2012, C_2014, B_2014, D_2014.

Figure 2

Distribution of microbial sources stratified by region. Datasets: A1_2012, C_2014, B_2014, D_2014.

Table 2 shows the number of human and non-human sourced samples (of those that tested positive for BacGeneral) by region. A statistically significant difference between groups was found by the Chi-square test (Chi-square = 12.48, df = 3, p< 0.01). Given the initial significant result, a post-hoc comparison was performed with an adjusted p-value alpha of 0.05 (Table 3). Statistically significant differences between Region A1 compared to Region B and D, as well as C compared to B and D were found. As such, Regions A1 and C have statistically significantly higher rates of human sourced fecal contamination.

Table 2

Number of samples positive and negative for BacHuman out of all samples positive for BacGeneral for each of the four regions

MST positives A1 
Human 57 (70%) 69 (67%) 23 (48%) 17 (44%) 
Non-human 25 (30%) 34 (33%) 25 (52%) 22 (56%) 
MST positives A1 
Human 57 (70%) 69 (67%) 23 (48%) 17 (44%) 
Non-human 25 (30%) 34 (33%) 25 (52%) 22 (56%) 

Datasets: A1_2012, C_2014, B_2014, D_2014.

Table 3

Raw and adjusted p-values from the BHY post-hoc comparisons of regional rates of human sourced fecal contamination

Comparison Raw p-value Adjusted p-value 
A1 vs C 0.7526 0.8290 
A1 vs B 0.0164 0.0328* 
A1 vs D 0.0092 0.0328* 
B vs C 0.0318 0.0477* 
B vs D 0.8290 0.8290 
C vs D 0.0130 0.0328* 
Comparison Raw p-value Adjusted p-value 
A1 vs C 0.7526 0.8290 
A1 vs B 0.0164 0.0328* 
A1 vs D 0.0092 0.0328* 
B vs C 0.0318 0.0477* 
B vs D 0.8290 0.8290 
C vs D 0.0130 0.0328* 

*Signifies statistically significant for adjusted p-values. Datasets: A1_2012, C_2014, B_2014, D_2014.

The frequency distributions for each region of domestic well density by DA on a logarithmic scale are shown in Figure 3. It demonstrates the data is non-normally distributed and therefore medians per region are appropriate for comparison purposes (Table 4). Both visually, by the frequency distributions, and by comparing medians, it is evident that Regions A1 and C have higher domestic well densities.
Table 4

Domestic-use well densities per DA area (km2) expressed as medians for each region

  Median domestic-use wells/DA area (km2
A1 3,905 
1,347 
2,955 
2,100 
  Median domestic-use wells/DA area (km2
A1 3,905 
1,347 
2,955 
2,100 
Figure 3

Regional frequency distribution of domestic-use well densities by DA expressed as well count per km2 and logarithmically transformed.

Figure 3

Regional frequency distribution of domestic-use well densities by DA expressed as well count per km2 and logarithmically transformed.

Spearman correlations between BacHuman and BacGeneral MST assays, used to further illuminate the predominant source, were found for each of the four areas. The results revealed that there is always a strong relationship between BacHuman and BacGeneral assay results (Table 5). Additionally for the 2012 dataset (where n was substantially larger), BacBovine and BacGeneral had a moderate correlation.

Table 5

MST assay correlations

Dataset Test n Spearman's ρ p-value 
A_2012 BacHuman vs BacGeneral 349 0.790 <0.001 
A_2012 BacBovine vs BacGeneral 88 0.509 <0.001 
A1_2012 BacHuman vs BacGeneral 57 0.823 <0.001 
B_2014 BacHuman vs BacGeneral 23 0.704 <0.001 
C_2014 BacHuman vs BacGeneral 69 0.859 <0.001 
D_2014 BacHuman vs BacGeneral 17 0.914 <0.001 
Dataset Test n Spearman's ρ p-value 
A_2012 BacHuman vs BacGeneral 349 0.790 <0.001 
A_2012 BacBovine vs BacGeneral 88 0.509 <0.001 
A1_2012 BacHuman vs BacGeneral 57 0.823 <0.001 
B_2014 BacHuman vs BacGeneral 23 0.704 <0.001 
C_2014 BacHuman vs BacGeneral 69 0.859 <0.001 
D_2014 BacHuman vs BacGeneral 17 0.914 <0.001 

Comparing CFU, E. coli and MST qPCR

The MST assay gene copy maxima per 100 mL of water filtered were: 20,367,074, 179,916, and 99,885,456 for BacHuman, BacBovine and BacGeneral assays respectively. The E. coli qPCR assay maximum cell number was 18,757 expressed per 100 mL.

A comparison of E. coli qPCR positives between quantifiable and non-quantifiable CFU/100 mL ranges revealed that the probability of obtaining an E. coli positive qPCR result are 2.98 (95% CI 1.98–4.58) times greater in the quantifiable range (Table 6) (Chi-square = 26.94, df= 1, p < 0.0001).

Table 6

E. coli PCR results stratified by quantifiable/non-quantifiable cultured counts of E. coli per 100 mL of filtered water

CFU/100 mL 1–15 > 15 
E. coli PCR result n (%) n (%) 
Positive 280 (49.1) 101 (74.3) 
Negative 290 (50.9) 35 (25.7) 
CFU/100 mL 1–15 > 15 
E. coli PCR result n (%) n (%) 
Positive 280 (49.1) 101 (74.3) 
Negative 290 (50.9) 35 (25.7) 

Dataset: A_2012.

Initial analysis showed that the percentages of positives for the BacGeneral assay were 76.2% and 75.3% respectively for E. coli and TC CFU positives (Table 7). In addition, E. coli and TC positives can be combined to represent contaminated samples, and compared against culture negatives (the non-contaminated samples) using a Chi-square test. The test revealed that there was a 2.62 (95% CI 1.68–4.08) times greater probability of observing a BacGeneral positive result for fecally contaminated samples compared to non-fecally contaminated samples (Chi-square = 18.40, df = 1, p < 0.0001).

Table 7

MST BacGeneral results stratified by membrane filtration contamination results of water samples, where negative represents no growth on the filter

Culture result E. coli positive TC positive Pooled E. coli TC data Negative 
MST result n (%) n (%) n (%) n (%) 
Positive 538 (76.2) 55 (75.3) 593 (76.1) 51 (54.8) 
Negative 168 (23.8) 18 (24.7) 186 (23.9) 42 (45.2) 
Culture result E. coli positive TC positive Pooled E. coli TC data Negative 
MST result n (%) n (%) n (%) n (%) 
Positive 538 (76.2) 55 (75.3) 593 (76.1) 51 (54.8) 
Negative 168 (23.8) 18 (24.7) 186 (23.9) 42 (45.2) 

Datasets: A_2012, A_2013_NEG, A_2013_TC.

A comparison between positive and negative results for E. coli and MST qPCR assays revealed that the probability of obtaining an E. coli positive is 3.71 (95% CI 2.57–5.43) times greater for samples that also tested positive for the MST BacGeneral assay compared to samples that tested negative for the BacGeneral marker (Table 8) (Chi-square = 49.57, df = 1, p < 0.0001).

Table 8

Number of samples positive for E. coli qPCR assay with respect to MST BacGeneral assay results

MST Positive Negative 
E. coli PCR result n (%) n (%) 
Positive 331 (61.3) 50 (29.9) 
Negative 208 (38.7) 117 (70.1) 
MST Positive Negative 
E. coli PCR result n (%) n (%) 
Positive 331 (61.3) 50 (29.9) 
Negative 208 (38.7) 117 (70.1) 

Dataset: A_2012.

Overall, a very weak to weak correlation was found between both BacHuman and BacGeneral positive results and E. coli positive results. However, when investigated further, it was observed that correlation strength varied with levels of contamination as determined by traditional E. coli culture methods. The relationship strengthened to a moderate relationship for both MST assays in the overgrown group (Table 9).

Table 9

Correlations between E. coli and MST BacHuman/BacGeneral qPCR assays stratified by bacteriological analysis results

Data Culture group (CFU/100 mL) n Spearman's ρ p-value 
E. coli vs BacHuman 1–15 152 0.051 0.531 
16–80 42 0.194 0.217 
>80 32 0.469 0.007* 
All 227 0.164 0.013* 
E. coli vs BacGeneral 1–15 240 0.233 <0.001* 
16–80 52 0.242 0.083 
>80 39 0.443 0.005* 
All 331 0.295 <0.001* 
Data Culture group (CFU/100 mL) n Spearman's ρ p-value 
E. coli vs BacHuman 1–15 152 0.051 0.531 
16–80 42 0.194 0.217 
>80 32 0.469 0.007* 
All 227 0.164 0.013* 
E. coli vs BacGeneral 1–15 240 0.233 <0.001* 
16–80 52 0.242 0.083 
>80 39 0.443 0.005* 
All 331 0.295 <0.001* 

Dataset: A_2012.

DISCUSSION

Regional MST results

The predominant host contributing to fecal contamination of private wells varies by region; notably, A1 and C had statistically higher rates of human fecal contamination. It should be noted that, while the Chi-square test revealed a difference in human contamination abundance among the regions, the post-hoc test applied (BHY) to identify where the differences were is a less conservative test. The Bonferroni post-hoc test was considered, but deemed inappropriate given the test's inherent over-conservative nature and the number of comparisons required. As such, the noted differences in rates of human fecal contamination between A1 and C compared to B and D (the high cattle region) should be interpreted with some caution.

Septic systems are the most likely source of the observed human fecal contamination in private wells given that most rural residents do not have access to municipal wastewater services. The exploratory analysis employed in this study revealed that greater well density may be related to greater abundance of human fecal contamination in private wells at a regional level. As domestic well density was chosen as a proxy for septic tank density, this provides initial evidence that directly implicates septic systems as the regional source of fecal contamination. Previous research in other regions supports this finding. A directly proportional relationship between fecal coliform concentration and septic tank distance was demonstrated by Arnade (1999). Additionally, the likelihood of pumping septic leachate has been found to increase with increasing septic system spatial density (Bremer & Harter 2012). Furthermore, where properties are smaller (and therefore densities are higher), domestic wells are more likely to be at a smaller distance from septic systems on the same and neighboring properties. Therefore, owners of private wells may have their wells contaminated from septic tanks owned by different individuals. As such, these owners cannot directly protect their well, but instead are dependent on local septic regulations. Consumption of human sourced fecal contamination in water has many negative health implications as various potential pathogens have been found in human waste (Cabral 2010; Li et al. 2015). As such, a review of septic system regulations should be conducted to ensure that private well owners are not being placed at risk.

The majority of fecal contamination is non-human sourced in both Regions B and D (the high cattle region). This suggests that the fecal animal sources could be either bovine or originating from any of the other eight species (pig, deer, horse, dog, cat, gull, goose and raccoon) targeted by the BacGeneral MST assay (Lee et al. 2010). Furthermore, the initial hypothesis stated that a higher rate of bovine contamination would be observed in the high cattle region. However, low rates were observed across all regions and, as such, it was not possible to test this hypothesis statistically. A possible reason for the low bovine rates observed is that private wells, as a groundwater source, require modes of entry for the bovine feces (as well other animal-sourced feces) to enter the aquifer. Typically this occurs due to groundwater under the direct influence (GUDI) of surface water which contains bovine fecal contamination via agricultural runoff (Health Canada 2013). These results suggest that most private wells studied were not from GUDI. In general, this study's observation of regional variation in primary fecal source suggests that primary prevention for groundwater contamination has to be regionally informed. Recent work by Åström et al. (2015) suggests that MST marker prevalence in combination with regional experts’ judgements may more accurately predict the origin of fecal contamination.

The strong correlations between human and general markers in all regions indicate that where samples are contaminated with human feces, human feces are the predominant source of fecal contamination with little interference from the other nine species targeted by the BacGeneral assay. Conversely, given the weaker relationship between BacBovine and BacGeneral, samples contaminated with bovine feces may also be contaminated by other fecal sources (including human). Lee et al. (2014) investigated host-specific fecal contamination in a river watershed and found significant correlations between human-specific and general markers where the primary contamination source was a wastewater treatment plant. In addition, they found no statistically significant relationship between bovine and general markers in the same samples. Reverse results were found in a study of surface water samples, where the primary contamination source was agricultural run-off (Lee et al. 2014). It is noteworthy that the water samples in this study originate from various locations, as previous MST studies are traditionally performed using water originating from a point location (Blanch et al. 2011), and yet still exhibit strong marker-to-marker correlation.

Indicators of fecal contamination

Minor, but significant, correlations were observed between E. coli PCR and BacHuman and BacGeneral results (using A_2012 data) with a stronger relationship observed between the latter. When stratified by levels of contamination (defined by culture methods), variation in correlation was observed. For BacHuman, the correlation was only statistically significant for the overgrown group (despite having the lowest statistical power). For BacGeneral, the strength of the correlation varied with level of contamination. The relationships observed for both E. coli and BacHuman and BacGeneral assays on overgrown samples were moderate in strength. These findings indicate that a more stable relationship exists between indicators when E. coli contamination is greater according to traditional culture methods. Numerous previous studies have attempted to correlate MST and E. coli as indicators of fecal contamination with varying results (and also varying power to detect) across a variety of watershed types. For instance, Layton et al. (2006) found a significant correlation (r2 = 0.85) between results for an assay detecting human, cattle and horse feces and E. coli concentrations for creek water samples. Lee et al. (2014) also reported a significant correlation (r = 0.73) between BacGeneral gene copies and culturable E. coli results when pooled for multiple sampling sites of a river watershed. When assessed individually, correlation results varied depending on sampling site. In addition, Lee et al. (2010) found no significant correlation between general Bacteroidales markers and E. coli cell numbers for creek and river water samples. Conversely, Sauer et al. (2011) found no significant correlation between human Bacteroides qPCR markers and culturable E. coli and a weak correlation (r2 = 0.34) between a wide-ranging Bacteroides marker with culturable E. coli.

The previously noted studies investigated point-source surface waters and were able to quantitatively correlate CFU/100 mL and MST qPCR results, given a much larger range of values for culturable E. coli. Given the lower range of CFU counts per 100 mL in private well waters, the E. coli qPCR assay was used in this study to compare across species. Sauer et al. (2011) found significant correlations (r2 = 0.16 and r2 = 0.46) between E. coli qPCR and human and wide-ranging Bacteroides markers, which were within the same range as the r2 values found by this study.

Previous studies have only investigated samples that were culturable for E. coli. This study investigated the presence/absence of MST markers in conjunction with bacteriological results for private well water samples, which included findings of TC only (no evidence of E. coli) when cultured, or absolute negative results. While, overall, culturing for E. coli (or other coliforms) led to greater probability of testing positive using MST, 54.8% of E. coli culture negative samples tested positive for the BacGeneral assay. The same phenomenon occurred with the E. coli qPCR assay, where a qPCR E. coli positive sample had greater probability of having a BacGeneral positive result, but 38.7% were E. coli qPCR negative and BacGeneral positive. These findings raise questions about the mechanisms involved in obtaining a BacGeneral positive and E. coli negative result. Overall in this study, 90.2% of samples that were culture negative, but BacGeneral positive samples (46 samples), tested positive using the E. coli qPCR assay. Disinfection of water at the well may explain these discordances. Potential water treatment would render any fecal-sourced E. coli cells non-culturable, but bacterial DNA would still be present and detectable within all qPCR assays. However, the remaining 9.8% of samples (n = 5) tested negative for E. coli in both culture and qPCR assay, but positive for BacGeneral; hence, water treatment alone cannot explain the disagreement between indicators. Additionally, 55% of samples positive for TC only, by culture, tested positive for the BacGeneral assay. These specimens are a further indication of the limits of E. coli as an indicator, as the viability of TC prevents treatment as an explanation. The relationship between TC and recent fecal contamination was demonstrated by Savichtcheva et al. (2007) who found a strong correlation (r2 = 0.8) between TC/100 mL and Bacteroides gene copies/100 mL.

Furthermore, 23.8% of samples culture-positive for E. coli and 29.9% of E. coli qPCR samples were negative for the BacGeneral assay. This finding is similar to results from a previous study by this research group (Krolik et al. 2014), where a quarter of the samples tested negative for the BacGeneral assay yet positive for E. coli by culture. The possibility of false positives by culture cannot be eliminated, although the high prevalence of BacGeneral negatives rules this out as the sole explanation. In instances where the CFU/100 mL count for E. coli is low, it is possible that the MST assay was not sensitive enough to detect trace amounts of host-specific fecal contamination. Additionally, the general MST markers include only 10 species whereas the source of fecal contamination may be attributed to other hosts. Finally, several studies have documented the existence of naturalized E. coli populations, capable of surviving for extensive periods in environmental matrices and exhibiting genetic differences from animal GI tract populations (Byappanahalli et al. 2006; Ishii et al. 2006; Ishii & Sadowsky 2008; VanderZaag et al. 2010). These E. coli subpopulations are not of recent fecal origin and therefore not representative of direct fecal contamination in private well water samples.

The probability of having an E. coli PCR positive result were higher if the sample had a higher level of culturable E. coli. As per the MST assay, it was not possible to further quantitatively relate the two tests, though the literature suggests a moderate correlation (Ahmed et al. 2012). However, there was a high level of disagreement between E. coli CFU/100 mL and E. coli qPCR assay (46% of samples that tested positive by culture were negative by qPCR). Additionally, when stratified to only include the quantifiable range (>15 CFU/100 mL), 65% of samples were positive by both E. coli culture and qPCR methods. This increase in agreement between the two methods once culturable E. coli reaches the quantifiable range suggests that the qPCR assay is unable to detect a signal in samples with low E. coli CFU counts. The assay was performed using a single cell (7 gene copy) cut-off, which, when tested against samples that were culture positive for E. coli, TC positive only (no E. coli), and negative for both E. coli and TC, yielded the following results: 55%, 25% and 15% positive, respectively. Additionally, a large portion of samples had a signal below the LOD and a supplemental analysis was conducted to further investigate this phenomenon. It was determined that 2.5 gene copies was a stable LOD, and at this level all datasets were at least 70% positive for E. coli, potentially indicating limits with the bacteriological testing. However, the 30–45% of E. coli culture positives (depending on the qPCR assay cut-off) that did not test positive by the PCR assay remain to be of concern. The high prevalence of signal below the LOD, alongside the correlation results, may also indicate that the E. coli qPCR assay used in this study is not sufficiently sensitive for small-volume water samples (i.e., low E. coli counts are difficult to distinguish from background signal). One of the reasons for the low sensitivity may be the volume of eluted sample used for qPCR, however these volumes are standard practice for qPCR methodology.

Traditionally, when evaluating laboratory diagnostic methodologies, methodologies are compared to definitive tests that are used to designate true positives and true negatives. This principle could not be applied in this study, because all tests are indicators of fecal contamination, not direct evidence of fecal contamination as it relates to the presence of pathogens and the risk of illness, the true aim of testing for fecal indicators. The analyses employed by this study compared each potential indicator to one another and noted their agreement of positives and negatives (where appropriate quantities allowed), but made no direct conclusions regarding the cases where the indicators disagreed given the uncertainties found in the tests themselves. Given that disagreement was found between all three methods, it is necessary to find and rigorously test techniques that are capable of designating ‘true positives and negatives’ for fecal contamination. Newer molecular techniques offer a promising new avenue and should be further investigated. Although not expected to be feasible for routine application in laboratories, use of these sophisticated technologies could enable discernment regarding which indicator assay is best.

Moreover, before MST methods can become integrated into routine laboratory testing as indicators of fecal contamination, characteristic ranges must be established. Currently, a general lack of MST method standardization creates a barrier for cross-comparison studies. Thus, guidelines need to be established regarding not only acceptable, standardized methodology, but also the limits or concentrations of various molecular markers that are required in order to deem water sufficiently safe to drink.

CONCLUSIONS

This study was among the first to track fecal contamination sources using Bacteroidales in groundwater non-point sources across multiple regions. While, in general, human Bacteroidales markers were the most common individual marker, statistically significant variation across regions was found. This indicates that primary prevention against groundwater contamination must be informed by investigations at a regional level. Additionally, this study utilized the breadth of samples collected for MST to investigate the effectiveness of fecal indicator tests (the traditional E. coli culture methods versus E. coli and MST qPCR assays). While there were overall trends toward agreement between methodologies, there was still considerable disagreement. Better assessment tools for determining the presence of fecal contamination in water samples are required.

ACKNOWLEDGEMENTS

The authors wish to thank the employees of Public Health Ontario for their kind support, in particular, staff at the Kingston, London and Hamilton laboratory locations. A special thank you also to Dai Young Lee and Susan Weir who shared their molecular expertise and to Kayla Poirier and Sophie Felletier. The funding for this study was provided in part by the Physicians’ Services Incorporated Foundation.

REFERENCES

REFERENCES
Åström
J.
Pettersson
T. J.
Reischer
G. H.
Norberg
T.
Hermansson
M.
2015
Incorporating expert judgments in utility evaluation of bacteroidales qPCR assays for microbial source tracking in a drinking water source
.
Environmental Science and Technology
49
(
3
),
1311
1318
.
Blanch
A. R.
Hagedorn
C.
Harwood
V. J.
2011
Microbial Source Tracking: Methods, Applications, and Case Studies
.
Springer
,
New York
.
Bremer
J. E.
Harter
T.
2012
Domestic wells have high probability of pumping septic tank leachate
.
Hydrology and Earth Systems Science
16
(
8
),
2453
2467
.
Cabral
J. P.
2010
Water microbiology. Bacterial pathogens and water
.
International Journal of Environmental Research and Public Health
7
(
10
),
3657
3703
.
Census of Agriculture, Farm and Farm Operator Data-95-640-XWE
2011
(accessed 1 April 2013)
.
Charrois
J. W.
2010
Private drinking water supplies: challenges for public health
.
Canadian Medical Association Journal
182
,
1061
1064
.
Chern
E. C.
Siefring
S.
Paar
J.
Doolittle
M.
Haugland
R. A.
2011
Comparison of quantitative PCR assays for Escherichia coli targeting ribosomal RNA and single copy genes
.
Letters in Applied Microbiology
52
(
3
),
298
306
.
Haack
S. K.
Fogarty
L. R.
Stelzer
E. A.
Fuller
L. M.
Brennan
A. K.
Isaacs
N. M.
Johnson
H. E.
2013
Geographic setting influences Great Lakes beach microbiological water quality
.
Environmental Science and Technology
47
(
21
),
12054
12063
.
Health Canada
2013
Guidance on the use of the microbiological drinking water quality guidelines – Catalogue No H144-12/2013E
.
Water and Air Quality Bureau, Healthy Environments and Consumer Safety Branch
.
Ottawa
,
Ontario
.
International Organization for Standardization (ISO)
2000
Water quality – Guidance on validation of microbiological methods – ISO/TR 13843
.
Switzerland
. .
Ishii
S.
Ksoll
W. B.
Hicks
R. E.
Sadowsky
M. J.
2006
Presence and growth of naturalized Escherichia coli in temperate soils from Lake Superior watersheds
.
Applied and Environmental Microbiology
72
,
612
621
.
Krentz
C. A.
Prystajecky
N.
Isaac-Renton
J.
2013
Identification of fecal contamination sources in water using host-associated markers
.
Canadian Journal of Microbiology
59
,
210
220
.
Krolik
J.
Maier
A.
Evans
G.
Belanger
P.
Hall
G.
Joyce
A.
Majury
A.
2013
A spatial analysis of private well water Escherichia coli contamination in southern Ontario
.
Geospatial Health
8
(
1
),
65
75
.
Krolik
J.
Evans
G.
Belanger
P.
Maier
A.
Hall
G.
Joyce
A.
Guimont
S.
Pelot
A.
Majury
A.
2014
Microbial source tracking and spatial analysis of E. coli contaminated private well waters in southeastern Ontario
.
Journal of Water and Health
12
(
2
),
348
357
.
Layton
A.
McKay
L.
Williams
D.
Garrett
V.
Gentry
R.
Sayler
G.
2006
Development of Bacteroides 16S rRNA Gene TaqMan-Based Real-Time PCR assays for estimation of total, human and bovine fecal pollution in water
.
Applied and Environmental Microbiology
72
(
6
),
4214
4224
.
Leclerc
H.
Mossel
D. A.
Edberg
S. C.
Struijk
C. B.
2001
Advances in the bacteriology of the coliform group: their suitability as markers of microbial water safety
.
Annual Review of Microbiology
55
,
201
234
.
Li
X.
Harwood
V. J.
Nayak
B.
Staley
C.
Sadowsky
M. J.
Weidhaas
J.
2015
A novel microbial source tracking microarray for pathogen detection and fecal source identification in environmental systems
.
Environmental Science and Technology
49
(
12
),
7319
7329
.
Liu
Y.
Gilchrist
A.
Zhang
J.
Li
X. F.
2008
Detection of viable but nonculturable Escherichia coli o157:H7 bacteria in drinking water and river water
.
Applied and Environmental Microbiology
74
(
5
),
1502
1507
.
Perchec-Merien
A. M.
Lewis
G. D.
2012
Naturalized Escherichia coli from New Zealand wetland and stream environments
.
FEMS Microbiology Ecology
83
,
494
503
.
Statistics Canada
2011
Dissemination area (DA) – Census Dictionary. Available from: https://www12.statcan.gc.ca/census-recensement/2011/ref/dict/geo021-eng.cfm (accessed 1 June 2015).
VanderZaag
A. C.
Campbell
K. J.
Jamieson
R. C.
Sinclair
A. C.
Hynes
L. G.
2010
Survival of Escherichia coli in agricultural soil and presence in tile drainage and shallow groundwater
.
Canadian Journal of Soil Science
90
,
495
505
.
Well Water Information System (WWIS)
2013
Ontario Ministry of the Environment, Canada. http://www.ontario.ca/environment-and-energy/well-record-data
(accessed 1 February 2013)
.