Comparison of the Multiple-Sample Means with Composite Sample Results for Fecal Indicator Bacteria by Quantitative PCR and Culture

Abstract

Few studies have addressed the efficacy of composite sampling for measuring indicator bacteria by quantitative PCR (qPCR). We compared results from composited samples with multiple-sample means for culture- and qPCR-based water quality monitoring. Results from composited samples for both methods were similarly correlated to multiple-sample means and predicted criteria exceedances equally.

TEXT

Because concentrations of fecal indicator bacteria (FIB) can be variable within a beach area, several water samples from a beach may be required to sufficiently characterize water quality (18). The development of rapid measurement methods, such as quantitative PCR (qPCR), has resulted in the ability to provide much faster water quality assessments, but these methods are valuable only insofar as a sampling scheme can accurately and efficiently characterize a water body in one or a few samples. Due to the cost of the analysis of multiple samples, analyzing composited samples may be a reasonable and less-expensive alternative (11). Compositing involves mixing equal volumes of spatially discrete samples and analyzing the homogenate. In enumeration of FIB by culture-based methods in water samples, results from composite samples have been shown to be significantly correlated to the arithmetic means of single-sample results and to yield similar beach closure decisions (1, 6, 10). Moreover, analysis of a single composite sample has been estimated to be nearly 40% less expensive than multiple single-sample culture analyses (10). Recent studies have indicated that qPCR-based measurements of FIB are statistically associated with gastrointestinal illness among swimmers (15, 16, 17), and composite sampling could generate significant savings for water quality monitoring programs using qPCR. Because the relationships between culture and qPCR results can vary within a beach on small temporal and spatial scales (3, 13), it is unclear whether composite sampling is as efficacious when using qPCR as when using culture methods. The purpose of this study was to determine whether composite sampling gave results statistically similar to those of multiple-point sampling when qPCR is used to measure bacterial water quality and whether the two sampling strategies are likely to produce the same beach closure decisions based on current proposed U.S. Environmental Protection Agency (EPA) guideline values for recreational water quality (14).

Water samples were collected three times daily (at 0800, 1100, and 1500) from Surfside Beach, Surfside, South Carolina, on weekends and holidays between 6 June and 7 September 2009 (n = 84). Three transects perpendicular to the beach were established to represent areas frequented by beachgoers. Transects were approximately 70 m apart. Duplicate water samples were collected along each transect at shin depth (0.3 m) and waist depth (1 m) in sterile polypropylene bottles. One of the duplicate water samples was used for single-sample analyses, and the other was used to create composites. Composites were made by combining 300-ml subsamples from each sample point and mixing. At each time point, three composites were made: one using samples from both depths, one using samples from shin depth, and one using samples from waist depth. One-hundred-milliliter subsamples from single samples and composites were filtered onto polycarbonate filters and stored at −80°C until qPCR analyses.

Culturable Enterococcus spp. were enumerated by membrane filtration (MF), following EPA method 1600 (12). All culture-based nondetects were assigned a value of 0.5 CFU/100 ml, one-half of the method's lower limit of detection. DNA was recovered from frozen filters by bead beating in a solution of salmon testes DNA, as described by Haugland et al. (5). Salmon testes DNA was included as an exogenous, internal control and reference for DNA loss during bead beating and PCR inhibition (5). The resulting material was stored at 4°C and analyzed within 3 days. Enterococcus species and salmon testes DNA were quantified by qPCR, following the method of Haugland et al. (5). Enterococcus species calibrator cell equivalents (CCE) were quantified by the ΔΔC_T method (5). All qPCR results were above the lower limit of quantification of 10 CCE/100 ml.

Geometric means (Ent_GM) and arithmetic means (Ent_AM) of Enterococcus species concentrations were calculated using results from the single samples that made up each composite. To determine whether concentrations of composited samples (Ent_Comp) tended to underestimate or overestimate means of multiple single samples, ratios of Ent_Comp to both Ent_GM and Ent_AM were calculated using untransformed data. All other calculations and statistical tests were conducted using log₁₀-transformed data to reduce skewness. Statistical analyses were completed in the software programs SAS version 9.2 and SigmaPlot version 11.0, and results and relationships were considered significant when P values were <0.05. Pearson correlations were calculated, and the Fisher z transformation was used to determine whether there were significant differences in strengths of correlation (9). The Mann-Whitney U test was used to identify significant differences in mean values among transects and among composite/mean ratios (8).

No statistically significant differences were observed in Enterococcus species concentrations (as measured by qPCR or MF) among transects (Table 1), affirming that the selected site and transects were appropriate for composite sampling. Where there are predictable significant differences, compositing may mask important spatial information. It is also worth noting that qPCR results were significantly greater than MF results for both single samples and composited samples, often by more than an order of magnitude. This difference is consistent with observations from several other studies (e.g., see references 3, 4, and 5) and is likely because qPCR targets genetic material, not culturable cells.

Table 1.

Arithmetic means, medians, and standard errors of Enterococcus species concentrations at each transect

Transect	Enterococcus species concn (CFU/100 ml or CCE/100 ml) by method
	MF			qPCR
	Median	Mean	SE	Median	Mean	SE
Left	0.54	0.57	0.05	1.81	2.02	0.06
Center	0.48	0.51	0.04	1.77	2.07	0.06
Right	0.48	0.55	0.04	1.66	1.95	0.06

Ent_Comp values were significantly correlated with the means of single samples regardless of the enumeration method (qPCR or culture), depth (both, shin, or waist), and mean calculation (arithmetic or geometric) (Table 2). Correlation strengths for qPCR results were slightly weaker than those for MF results, though differences were not significant when using samples from both depths or shin depth only. At waist depth, correlations for qPCR results were significantly lower than those for MF (P = 0.04). This difference may be due to patchiness of DNA from intact dead or viable but nonculturable cells further from the source of contamination (13). Also, concentrations were often lower at waist depth, as has been observed previously at many beaches, and weaker correlations at greater depth may be due to increased variability in qPCR results relative to culture results, where concentrations are low (19). Nevertheless, the differences in correlation strengths between depths suggest that composite sampling strategies are likely to be beach specific and depend upon many factors, including circulation, proximity to pollution source, and sampling depth.

Table 2.

Significant Pearson correlation coefficients (r) (P < 0.001) between means of single-sample results (independent variable) and results for composited samples (dependent variable)^a

Sampling depth	Mean type	r value (r2) for method
		MF	qPCR
Both	Arithmetic	0.80 (0.63)	0.74 (0.55)
	Geometric	0.80 (0.64)	0.71 (0.56)
Shin	Arithmetic	0.80 (0.64)	0.72 (0.52)
	Geometric	0.79 (0.62)	0.74 (0.54)
Waist	Arithmetic	0.77 (0.59)	0.57* (0.31)
	Geometric	0.79 (0.62)	0.60* (0.34)

^aAsterisks (∗) indicate where correlations for qPCR results are significantly lower (P < 0.05) than those for MF results.

In general, the ratios between Ent_Comp and multiple-sample means were near unity for both qPCR (approximately 0.9) and MF (approximately 1.1) (Fig. 1). Ratios did not differ significantly between methods, except when using samples from both depths combined. In this case, the ratio of Ent_Comp to Ent_GM for MF (1.4) was significantly greater (P < 0.001) than the same ratio using qPCR (0.9). This may be due to differences in homogenization of composite samples, which were mixed during the creation of the composite and as part of EPA method 1600 (12). Increased homogenization may have broken chains normally formed by Enterococcus spp. (7), resulting in greater CFU but not affecting qPCR-detectable CCEs.

Fig 1.

Ratios between concentrations of composited samples and means of multiple-sample concentrations (Ent_Comp/Ent_GM or Ent_Comp/Ent_AM) at the various sampling depths (both in the top row, shin in the middle row, and waist in the bottom row) for both qPCR and MF. Ratios using the arithmetic means are presented in the left column, and ratios using the geometric means are presented in the right column. Dashed lines are a reference for a ratio of 1.

Nevertheless, concentrations of composited samples were occasionally outside the range of the single-sample concentrations for both qPCR and culture results (Fig. 2). For MF, Ent_Comp values were higher than maximum single-sample results in 15% of cases and below single-sample minimums in 2.5% of cases. For qPCR, Ent_Comp values were above single-sample ranges in 19.5% of cases and below single-sample minimums in 2.5% of cases. Potential explanations of these anomalous results include method variation and complexity (especially of the relatively newer qPCR method) and possible natural heterogeneity of Enterococcus spp. in the water (13). Previous work has demonstrated that Enterococcus concentrations in samples collected 1 min apart vary by 60% on average and up to 700% (2). Also, 6% of single-sample MF results were below the limit of detection, meaning that the range of single-sample concentrations extended below the detection limit. In these cases, the composite could not have a concentration below the single sample range.

Fig 2.

Results from composited samples (Ent_Comp, represented by Xs), geometic mean values (Ent_GM, represented by circles), and corresponding ranges of single-sample concentrations (represented by the bars). Dashed lines represent the proposed geometric mean criteria; solid lines represent the proposed single-sample criteria. MF results are summarized by the graphs on the left; qPCR results are summarized by the graphs on the right.

To determine whether the slightly greater instances of Ent_Comp falling outside the single-sample ranges for qPCR may have implications for beach closure decisions based upon EPA guidelines, parameters from logisitic regression models were used to calculate probabilities (p) that decisions based upon Ent_Comp would be equivalent to those based on Ent_GM. To guide these analyses, we have used the geometric mean criteria for culturable and qPCR-enumerated Enterococcus spp. proposed in the EPA's current draft of Recreational Water Quality Criteria (14): 35 CFU/100 ml for culture and 475 CCE/100 ml for qPCR. We compared Ent_Comp with Ent_GM rather than Ent_AM for two primary reasons: first, previous work has demonstrated that health effects are correlated to the geometric mean of FIB concentrations and exposure-response models are based on the geometric means (15, 16, 17), and second, the criteria themselves are based on geometric means (20). For these analyses, we created new binary data sets in which criteria exceedances were represented by 1 and nonexceedances were represented by 0. These binary data were fit to logistic regression models, which model the log odds of an event {ln[p/(1 − p)]} (9), using PROC LOGISTIC in SAS. When Ent_GM values were greater than the EPA-recommended criteria, qPCR-enumerated Ent_Comp values were slightly though not significantly more likely to also exceed the EPA guideline than MF-enumerated Ent_Comp values (p = 0.72 and 0.67, respectively). qPCR-enumerated Ent_Comp values were also more likely to trigger theoretical beach advisories than MF-enumerated Ent_Comp values when Ent_GM values were below the EPA-recommended criteria (p = 0.08 and 0.06, respectively). Because many states base water quality decisions on single-sample criteria, the probabilities that Ent_Comp values would trigger advisories (exceed the geometric mean criteria) when a single sample was above the single-sample criteria were also calculated. Again, we used the single sample criteria proposed in the EPA's draft of Recreational Water Quality Criteria (14): 104 CFU/100 ml and 1,000 CCE/100 ml. In this case, the conditional probabilities that Ent_Comp values would be equal to or above the geometric mean criteria when at least one single-sample concentration was above the single-sample criteria were more similar for MF (0.67) than qPCR (0.64).

In summary, correlations between Ent_Comp and multiple-sample means were similar for qPCR and MF, with the exception of waist-depth samples. Also, beach closure decisions based upon Ent_Comp were equally likely to agree with decisions based upon multiple samples for both qPCR and culture. These results provide evidence that compositing of samples for qPCR can be similarly protective of public health compared to compositing when using culture-based monitoring. However, it is important to reiterate that these data were collected from a single beach during a single swimming season. Though these data are generally supportive of the use of composite sampling for qPCR where compositing for culture is already used, the specifics of compositing strategies, such as depth of sampling and locations of transects, will likely be site specific and depend on proximity to pollution sources, mixing patterns, and sampling depth. Before composite sampling is adopted at other beaches, the data from this study need to be ground truthed in other regions.