Human virus and microbial indicator occurrence in public-supply groundwater systems: meta-analysis of 12 international studies

Groundwater is an important source of drinking water in both developed and developing countries. It constitutes about 95% of the world’s accessible freshwater (Chilton and Seiler 2006; Howard et al. 2006; McKay 2011) and is often used with little or no treatment (Pedley et al. 2006). Contamination of groundwater with human enteric viruses is a global issue (Blaschke et al. 2016; Gotkowitz et al. 2016; Hynds et al. 2014; USEPA 2006a, b), as consumption of contaminated water can result in elevated rates of endemic illness and waterborne disease outbreaks in affected communities (Beer et al. 2015; Borchardt et al. 2011; Cho et al. 2014; Guzman-Herrador et al. 2015; Hilborn et al. 2013; Jack et al. 2013; Wallender et al. 2014; Zhou et al. 2012). Enteric viruses implicated in waterborne outbreaks include enteroviruses, hepatitis A, rotavirus, and norovirus (Craun et al. 2010; Hejkal et al. 1982), but others such as adenoviruses, Aichi virus 1, hepatitis E, and reoviruses, potentially are capable of groundwater-borne transmission. Waterborne viruses cause a wide range of illnesses, including gastroenteritis, paralysis, aseptic meningitis, conjunctivitis, diabetes, fevers, herpangina, rash, myocarditis, and respiratory illness, (Kitajima and Gerba 2015; WHO 2011).

Several nonpathogenic bacteria that are normal flora of the human intestine and other warm-blooded animals can be detected using simple and inexpensive techniques. These include total coliform bacteria, fecal (thermotolerant) coliform bacteria, E. coli, enterococci, and bacterial endospores (Ashbolt et al. 2001; Locas et al. 2007; Tallon et al. 2005). Because pathogens, and especially viral pathogens, occur sporadically and are difficult to detect, these bacteria have been used as indicators of fecal pollution. Total coliform bacteria and E. coli are the most commonly employed water quality indicators, but of the two, E. coli is a more definitive indicator of fecal pollution (Edberg et al. 2000; Tallon et al. 2005).

Coliphages (bacteriophages that infect coliform bacteria) are present in wastewater and have been suggested to be a useful conservative indicator of fecal and viral pollution of groundwater (Deborde et al. 1998a, b; Lucena et al. 2006). They are grouped into two major categories—F-specific and somatic coliphages. F-specific coliphages are similar in size (about 26 nm in diameter) and structure (icosahedral) to human enteric viruses (about 28–38 nm in diameter for enteroviruses and noroviruses), while somatic coliphages have more variability. Unlike enteric viruses, coliphages are detectable by simple, inexpensive and rapid techniques (Lucena et al. 2006). Phages of Bacteroides have also been evaluated as useful indicators (Jofre et al. 1986; Johnson et al. 2011; McMinn et al. 2014).

Although bacterial indicators are used for general water quality monitoring, they are not considered to be adequate indicators of viral contamination because the structure, composition, morphology, size (about 1,000 nm in diameter × 2,000 nm in length for E. coli), and survival characteristics of viruses differ fundamentally from those of bacteria (Ashbolt 2015; Payment and Locas 2011; WHO 2011). Viruses survive longer than vegetative bacteria in the environment (Nasser et al. 1993) and have different transport properties. The usefulness of anaerobic bacterial endospores is limited by very long survival times, the ubiquity of Clostridium species in soil, and by transport properties (Meschke 2001). Aerobic endospores also have major differences in transport and survival properties (Headd and Bradford 2016; Pang et al. 2005).

Viral and other pathogens and microbial indicators enter aquifers through multiple sources and pathways, including leachates from sanitary landfills, on-site septic waste treatment discharges, broken sewage lines, runoff from urban, agricultural and natural areas, and water reuse by direct injection of inadequately treated wastewater into aquifers (Borchardt et al. 2012; Costan-Longares et al. 2008; Gotkowitz et al. 2016). Fecal contamination from the surface may also get into groundwater through improperly constructed, protected, or maintained wells (Hynds et al. 2014).

Despite the potential public health impact from drinking untreated groundwater, the regulations guiding the quality of groundwater supplies used for drinking water are highly varied among countries. Many use the World Health Organization Guidelines for Drinking Water Quality (WHO 2011), which covers drinking water from all sources and suggests that drinking water contain no thermotolerant coliform bacteria or E. coli per 100 ml. Canada’s guidelines state that drinking water should have no detectable total coliforms or E. coli per 100 ml (Health Canada 2017). Australia’s guidelines add intestinal enterococci and coliphages in addition to thermotolerant bacteria and E. coli as agents that should not be detected in 100 ml and suggest that sanitary surveys be conducted for groundwater systems (National Health and Medical Research Council 2016). Sanitary surveys are used to examine a system’s deficiencies that could cause vulnerability to microbial contamination. The surveys are intended to identify deficiencies caused by poor source water, inadequate well construction or maintenance, and improper system operation. Monitoring of water quality is required by regulation in some countries—for example, Japan requires monitoring for E. coli (Wakayama 2016). Korea does periodic monitoring for enteric viruses and their Groundwater Act requires monitoring of groundwater for E. coli (Lee and Kwon 2016; Lee H et al. 2011; Lee S et al. 2011). The European Union requires regulatory authorities to monitor all public drinking-water systems for coliform bacteria and E. coli at a frequency that depends on water volume pumped and/or population served. The United States regulatory instrument is the Ground Water Rule (GWR; USEPA 2006a, b). The key components of the GWR are: sanitary surveys, triggered source water monitoring, corrective actions, and compliance monitoring. Source monitoring may be required rather than triggered if the source water quality is uncertain. The triggered water monitoring provision requires that untreated groundwater systems must conduct triggered source water monitoring for the presence of at least one of the three fecal indicators: E. coli, enterococci, or coliphage following a Revised Total Coliform Rule (RTCR)-positive sample. The RTCR requires regular monitoring of all public systems for total coliforms using 100-ml samples at a frequency that depends on the population served. Any regular monitoring sample that contains any amount of total coliforms must be followed by additional tests for both total coliforms and E. coli. The compliance monitoring provision of the GWR requires that systems that provide 4-log treatment of viruses must conduct compliance monitoring to demonstrate continual treatment effectiveness. The GWR uses a risk-targeted approach to address the likelihood of viral contamination of wells—i.e., well susceptibility to contamination (note: susceptibility herein is defined by the entry of fecal contamination into an aquifer as measured by the demonstrated presence of virus in any sample from an associated well). It suggests that the agencies that implement the rule consider groundwater from aquifers in limestone, igneous and metamorphic rock, and gravel as potentially susceptible. However, among aquifers in gravel settings, gravel/cobble aquifers are more susceptible than those consisting of gravel/sand (Berger 2008). Aquifers in gravel settings have rock grain sizes of 4 mm or larger, whereas those in gravel/cobble settings have grain sizes up to about 256 mm (Wentworth 1922).

While a good indicator would be expected to be positively correlated to virus presence (i.e., always present when viruses are present and absent when viruses are absent), these types of correlations are not always observed in groundwater (Payment and Locas 2011). This might be due to indicator and virus differing in die-off properties, transport characteristics, waste treatment practices before release to the environment, source concentrations—e.g., indicators are constantly present and at higher concentrations in wastewater than are viruses (Berg et al. 1978)—and virus infection dynamics and shedding rates in the host population. In addition, in an analysis of studies correlating indicators and pathogens in water types other than groundwater, Wu et al. (2011) showed that the strength of correlations was related to study sample size and the number of pathogen-positive detections. Mindful of these limitations, this study examined the relationship between indicators and viruses by combining data from 12 studies of public drinking-water groundwater systems. In addition to conventional tests of correlation, the study evaluated the association between indicator and virus occurrence by logistic regression and calculated standard performance measures for diagnostic tests, namely, sensitivity, specificity, positive predictive value, and negative predictive value. These measures were further evaluated for three subsets of wells considered susceptible to fecal contamination based on hydrogeological setting and US regulations.

Table 6 shows the percent of samples and wells that were positive for indicators and virus from the 12 groundwater studies. Total coliforms were detected in 31% of samples and 36% of wells. Overall, 37% of samples and 45% of wells were positive for any indicator, while 15% of samples and 27% of wells were positive for any enteric viruses. The average titer of positive samples containing culturable- or PCR-virus from the 12 studies was 0.4 infectious units and 16 genomic copies per liter, respectively (data not shown).

Spearman Rank Order tests showed that among the indicators there were positive and moderately strong correlations (rho ≥0.5, P < 0.001, n > 1,300) for combinations of total coliforms, E. coli, and enterococci, and between enterococci and somatic phage (data not shown). Between the indicators and culturable virus concentrations on a per-sample basis the trends were positive but weak (rho ≤0.3, P < 0.001, n > 1,200; Table 7). Correlations on a per-well basis were also weak with the highest between somatic coliphage and culturable virus (rho = 0.46, P < 0.001, n = 355). Correlations among indicators and PCR-virus (Table 7) were always weaker than to culturable virus.

The analysis was repeated by restricting the data to virus-positive samples (or wells) to minimize the effect of non-detects on the correlations. With this restricted data set, the highest correlations to culturable virus were E. coli (rho = 0.62, P < 0.001, n = 144) and somatic coliphage (rho = 0.54, P < 0.001, n = 141) on a per-sample basis and somatic coliphage (rho = 0.48, P < 0.001, n = 76) on a per-well basis; however, restricting the data set did not improve the correlations among the indicators and virus detected by molecular assays (data not shown).

Subsequently, the association between indicator and virus occurrence was evaluated (detect or non-detect) using logistic regression models and the data from the 12 studies. Culturable viruses were associated at statistically significant levels with the indicators except for total coliforms measured at the sample level and spores measured at the well level (Table 8). In contrast, PCR-viruses and indicators at the sample level were never statistically associated, and at the level of well only three of the six indicators (total coliforms, E. coli, and somatic phage), were significantly associated with PCR-viruses. The two most commonly used indicators in the US—total coliforms and E. coli—were associated at the well level with both culturable viruses and PCR-viruses, albeit the E. coli and culturable virus association was marginally not significant (P = 0.087; Table 8).

The risk ratios reported in Table 8 give the relative increase in the probability of detecting a virus-positive sample (or well) when an indicator is detected compared to when an indicator is not detected—for example, a positive E. coli sample is associated with a seven-times greater chance of detecting culturable viruses in a corresponding sample compared to a negative E. coli result. For PCR-viruses, the chance of detection in a sample is similar whether a corresponding E. coli sample is positive or negative (risk ratio = 0.9, P = 0.79). However, at the level of well, an E. coli positive well is associated with a 60% greater chance the well, at some time, will be positive for PCR-viruses (risk ratio = 1.6, P = 0.008).

Table 8 reports the associations and risk ratios adjusted for study (i.e., accounting for underlying differences among the 12 studies) and the sample-level analyses additionally include an adjustment for wells. Analyses were also conducted without these adjustments (data not shown). Generally, the adjustments resulted in lower estimated risk ratios and similar conclusions regarding statistical significance of the indicator-virus association, suggesting the probability of detecting a virus when an indicator is present differed across the individual studies and the wells within each study. The effect of the adjustments on the risk ratios could reflect differences in hydrogeological settings, virus contamination sources, laboratory methods, or several unknown factors at the level of study or well that are related to virus and indicator occurrence. Nonetheless, for this evaluation of indicators and viruses in groundwater, it is evident it was important to account for the effects of both study and well.

Indicator test performance measures were examined next. Sensitivities of the indicators for signifying whether a virus was detected in a sample or well were low (2–30%), with estimates being relatively higher when virus positivity was determined by culture (11–30%, Table 9) versus PCR (2–12%, Table 10); note that all test performance measures reported here are adjusted for study and well (sample-level analyses) or study only (well-level analyses). By culture, between 11 and 30% of virus-positive samples and between 37 and 73% of virus-positive wells could be correctly identified as virus-positive by an indicator. In other words, many samples and wells negative for an indicator were, in fact, positive for virus. The corresponding numbers for PCR were ≤12% for samples and ≤39% for wells.

Positive predictive values were also low for the indicators in predicting the detection of both culturable virus and PCR-virus. Many samples and wells that were positive for an indicator were, in fact, negative for virus (i.e., the indicators had a high rate of false positives). Positive predictive values at the well level were higher (but still mostly <50%) than at the sample level. Approximately 30 to 40% of the wells that were positive for an indicator were contaminated with virus at some point during the sampling period.

In contrast, specificities of the indicators were high, with values often near 90% or greater for culturable virus and PCR-virus. For example, E. coli had 95 and 97% specificity at the sample level for culturable virus and PCR-virus, respectively. In other words, a negative E. coli result correctly identified 95–97% of the samples that were negative for viruses. Specificity values at the well level were consistently lower than at the sample level for all six indicators, but the magnitude of the differences were usually not large.

Negative predictive values for the indicators were generally greater than 90%, with the exception of PCR-viruses at the well level, where estimates ranged from 50 to 76%. An indicator with a high negative predictive value means that when a sample or well is indicator-negative, it is likely to be also virus-negative (i.e. the indicator has a low rate of false negatives).

The sample and well data were reanalyzed according to the TCR, Hydrogeology, and GWR susceptibility subsets described in Table 5. Table 11 shows the ratio of the percent of virus-positive samples or wells in each subset to the percent of all virus positive samples or wells. Culturable viruses were twice as likely to be present in samples obtained from wells in each subset as in samples from all wells. Viruses measured this way were only 1.3–1.5 times more likely in wells from the TCR and hydrogeology subsets, but 4 times more likely in the GWR subset. PCR virus was poorly associated with the different subsets (ratios of 0.9–1.6).

Table 12 shows the indicator test performance measures for those combinations of indicators and susceptibility categories that had statistically significant indicator-virus associations. The chance of detecting virus was elevated when total coliform or enterococci bacteria or F-specific bacteriophage were detected in a hydrogeological susceptibility setting and when enterococci or somatic bacteriophage were detected for wells considered susceptible by having a TCR violation. No single indicator improved the chance of detecting virus for those wells considered susceptible by the GWR. Compared to the analyses that included all samples or wells, sensitivities and positive predictive values generally were higher for indicators measured in wells in all three susceptibility categories for both culturable virus and PCR-virus.

This study addresses the relationship between microbial indicators and human enteric virus in groundwater. The focus of the study is on public systems in the US, Canada, Europe, and Asia for which raw data from 746 wells and 2273 samples were available. One or more indicator was found in 44% of the wells, while culturable virus was detected in 8% and PCR virus in 30% of the wells.

Many studies have examined correlations between concentrations of indicators and viruses and often times the correlations are weak or non-existent (see references in Table 1). When 12 studies were combined, the correlations between indicator and virus concentrations were statistically significant, but the low rho values showed the correlations were weak. Contributing factors to the weak relationships are the use of different methodologies, the amalgamation of wells from various hydrogeological settings, and the large number of samples or wells in the dataset that were negative for indicators or viruses. Excluding most negative values from the Spearman rank order test did increase the rho values pertaining to culturable virus, but as negative values cannot be excluded a priori such a relationship has no practical value for well testing.

Hynds et al. (2014) pooled data from 39 studies of private groundwater wells that included both microbial indicator and pathogen data and compared the percentage of indicator positive wells to the percentage of pathogen positive wells. The correlations tended to be weak and some were opposite of what would be expected, for example, viruses in groundwater were negatively correlated with enterococci. These authors recommend direct testing for virus. From 540 sets of indicator-pathogen correlations in studies of surface water, Wu et al. (2011) modeled the probability of finding a significant correlation and concluded that many studies do not find a correlation because of small sample size. Payment and Locas (2011) compared virus concentrations to concentrations of six indicators in 242 samples from 25 groundwater sites and found no correlations.

Logistic regression was used as an alternative approach to examine the association between dichotomous representations of indicator and virus detections. This approach allowed the calculation of risk ratios (i.e. the ratio of virus detection probabilities in the presence versus the absence of an indicator) while adjusting the association for underlying differences among the 12 studies and among wells within a study. Risk ratios and tests for association evaluate the overall correspondence between indicator and virus occurrence. A more complete interpretation of an indicator’s value with respect to viruses is provided by calculating standard test performance measures (e.g., sensitivity, specificity, etc.) that convey the likelihood of false positives and false negatives (Borchardt et al. 2003).

The risk ratios show that, in general, culturable viruses are more likely to be detected when there is a positive indicator result, especially for the indicators E. coli, enterococci and somatic phage at the sample level and F-specific and somatic phage at the well level. For example, the data indicate that it is nine times more likely to find culturable virus in a sample or well that also has somatic coliphage. In contrast, no significant risk ratios were found among indicators and PCR-virus at the sample level and the strongest risk ratios at the well level were less than two. Similarly, in examining the factors associated with statistically significant correlations between indicators and pathogens in surface waters, Wu et al. (2011) showed that studies that use molecular methods for pathogen detection were less likely to find significant correlations.

All six indicators that were evaluated tended to have low sensitivity and low positive predictive values, but high specificity and high negative predictive values. From a practical standpoint, this means that if a well has an unknown virus contamination problem, it is unlikely to be identified by a positive indicator (low sensitivity). Additionally, a positive indicator result does not necessarily mean the well is virus-contaminated; it could likely be a false positive (low positive predictive value). If the well does not have a virus contamination problem, there is a reasonable chance the indicator results will be negative, confirming there is no problem (high specificity). And a well that is indicator negative is unlikely to be virus contaminated, especially for culturable viruses (high negative predictive value). In summary, the downside of the indicators is that many virus positive wells can be missed and there could be many false positives. On the other hand, wells that are indicator-negative are unlikely to have virus contamination problems. Moreover, sensitivities and positive predictive values at the well level were higher for all six indicators compared to the sample level, for the most part without sacrificing a large drop in specificity or negative predictive value. This suggests that multiple samples from a well improves indicator performance in assessing a well’s susceptibility to virus contamination. Indicator performance might also be improved by considering not just the detection of the indicator, but also its concentration. Payment and Locas (2011) showed the probability of human virus detection in the Saint-Lawrence River in Quebec, Canada, increased with increasing concentrations of the indicators thermotolerant coliforms or Clostridia perfringens. An indicator’s sensitivity performance measure may depend on the indicator concentration with higher sensitivities at higher concentrations.

The indicator-virus associations were evaluated for three well susceptibility categories to examine if indicators were more informative based on prior knowledge of well-specific data (Table 5). The TCR category was chosen because total coliforms are the most common indicator used to evaluate well susceptibility to pathogens. The hydrogeology category was selected due to the known susceptibility of karst, fractured bedrock, and gravel/cobble settings, and the GWR category was selected because the criteria for well susceptibility includes two indicators that typically had not been tested in groundwater, enterococci and coliphage. Among the wells in the three susceptibility categories, culturable virus was detected twice as frequently in samples and 1.3–3.9 times more in wells compared to all samples or wells (Table 11). In contrast, PCR-viruses were detected only 0.7–1.6 times more in samples or wells in the susceptibility categories. This difference may be a function of virus fate in the subsurface. It may be easier to detect viruses by PCR in less susceptible settings as the viruses can still be detected long after infectivity is lost and at greater distances from sources of contamination (Ogorzaly et al. 2010).

Sensitivities, positive predictive values, and risk ratios tended to be higher for total coliforms, enterococci, F-specific phage, and somatic phage when they were measured in samples or wells in the TCR or hydrogeology susceptibility categories. In other words, for wells with two or more TCR violations or wells located in susceptible hydrogeological settings, like karst, these indicators performed better at identifying samples and wells that were true virus-positives and with a lower false positive rate. However, for wells that would have met the criteria under the GWR for additional monitoring, none of the single indicators performed any better than when they were measured in all wells regardless of susceptibility status.

The finding of a virus by either culturable or molecular assays is not always easily translatable to the public health risk presented by a particular well. Virus concentration in groundwater can vary rapidly over time (Bradbury et al. 2013). A single data point can fall anywhere within the possible range of virus concentrations during transient contamination, from having no virus, and thus underestimating risk, to by chance collecting the sample during peak concentration and overestimating risk. In addition, as virus recovery from water by various concentration procedures rarely achieves 100% (Cashdollar and Wymer 2013; Gibbons et al. 2010; Ikner et al. 2012, 2011; Karim et al. 2009; Sobsey and Glass 1984; Wu et al. 2013) virus concentration, and therefore risk, can be underestimated. Different virus groups and even different members of the same species can have different recovery efficiencies, while recovery efficiency of a single virus type will vary over time with changes in turbidity and other water quality factors.

To pose a health risk, virus concentrations in the well water must be high enough that individuals drinking the water ingest more virus than the minimum infectious dose necessary to initiate infection. Apart from egregious virus contamination of groundwater that results in an outbreak (e.g. Borchardt et al. 2011), most surveys of groundwater in developed countries show culturable virus concentrations that typically are very low (e.g., combined data from this study), suggesting the health risk is low. On the other hand, culturable virus assays can underestimate virus concentrations and occurrence. Virus in water can be aggregated with each aggregate registering as a single infectious unit in culturable virus assays (Gassilloud and Gantzer 2005); however, aggregates may have a sufficient number of virus particles to cause an infection, if consumed. Culturable assays also underestimate risk in that they do not detect many important waterborne enteric viruses, and cell culture does not necessarily replicate the same favorable conditions for virus replication as when a virus infects a host.

Because PCR detects virus genomes and not necessarily infectious virions, PCR measurements of viruses in groundwater are often thought to overestimate the health risk. On the other hand, virus present can maintain infectivity for months at temperature of 12 °C or less (Charles et al. 2009; Nasser and Oman 1999). PCR also may underestimate risk because enteric viruses, and especially RNA viruses, evolve constantly, leading to genetic variability that prevents their detection by PCR (i.e., false negative results). Nonetheless, the presence of viral genomes in groundwater shows that an aquifer is susceptible to fecal contamination. Importantly, Borchardt et al. (2012) compared the fraction of acute gastrointestinal illness (AGI) associated with drinking untreated groundwater in 14 communities with PCR-virus from community wells. The concentration of enteroviruses in the tap water of households in the communities was associated with AGI in adults, and tap water concentrations of norovirus genogroup I were associated with AGI in all ages. Using quantitative microbial risk assessment, the authors estimate that between 6 and 22% of the overall AGI in the communities and up to 63% of AGI in children less than 5 years of age was associated with drinking untreated tap water. The AGI incidence increased by 63% when the mean norovirus concentration in the communities’ tap water increased from 0 to about 3 genomic copies per liter. Thus, without results to the contrary, the possibility of adverse health risks based upon PCR positive findings should be considered when there are septic tanks, sewer lines, or other sources of contamination near aquifers that reside in similar hydrogeologic settings to those in Wisconsin.