Estimating true human and animal host source contribution in quantitative microbial source tracking using the Monte Carlo method

Abstract

Cultivation- and library-independent, quantitative PCR-based methods have become the method of choice in microbial source tracking. However, these qPCR assays are not 100% specific and sensitive for the target sequence in their respective hosts’ genome. The factors that can lead to false positive and false negative information in qPCR results are well defined. It is highly desirable to have a way of removing such false information to estimate the true concentration of host-specific genetic markers and help guide the interpretation of environmental monitoring studies. Here we propose a statistical model based on the Law of Total Probability to predict the true concentration of these markers. The distributions of the probabilities of obtaining false information are estimated from representative fecal samples of known origin. Measurement error is derived from the sample precision error of replicated qPCR reactions. Then, the Monte Carlo method is applied to sample from these distributions of probabilities and measurement error. The set of equations given by the Law of Total Probability allows one to calculate the distribution of true concentrations, from which their expected value, confidence interval and other statistical characteristics can be easily evaluated. The output distributions of predicted true concentrations can then be used as input to watershed-wide total maximum daily load determinations, quantitative microbial risk assessment and other environmental models. This model was validated by both statistical simulations and real world samples. It was able to correct the intrinsic false information associated with qPCR assays and output the distribution of true concentrations of Bacteroidales for each animal host group. Model performance was strongly affected by the precision error. It could perform reliably and precisely when the standard deviation of the precision error was small (≤0.1). Further improvement on the precision of sample processing and qPCR reaction would greatly improve the performance of the model. This methodology, built upon Bacteroidales assays, is readily transferable to any other microbial source indicator where a universal assay for fecal sources of that indicator exists.

Introduction

Fecal contamination is one of the primary concerns for surface water quality in the world. US EPA estimated that approximately 13% of surface waters in the US exceed their designated use criteria based on fecal indicator bacteria (FIB) concentrations (US EPA, 2005). Fecal contamination poses a significant risk to human health (Anderson et al., 2005, Dick et al., 2005a, Dick et al., 2005b; Santo Domingo et al., 2007). It also leads to economic losses from closure of recreational areas and shellfish production. The need to identify the sources of fecal contamination, especially non-point sources, has spawned a rapid development of microbial source tracking (MST) methodology (Scott et al., 2002, Santo Domingo et al., 2007, Stoeckel and Harwood, 2007, Witty et al., 2009). MST methods have been used to develop total maximum daily load targets and to evaluate best management practices in the United States (Houck, 2002, Reckhow, 2003, Cooter, 2004, Bambic et al., 2007).

Cultivation- and library-independent MST methods that detect host-specific markers directly after PCR are now widely used. Of the bacteria that demonstrate some kind of host-specificity, strictly anaerobic members of the order Bacteroidales are currently the most commonly used source identifiers in water. Universal and a panel of host-specific Bacteroidales assays have been designed, including assays specific for human, dog, ruminant, bovine (cow), swine (pig), equine (horse), elk, and gull fecal pollution (Bernhard and Field, 2000, Dick et al., 2005a, Dick et al., 2005b, Layton et al., 2006, Reischer et al., 2006, Kildare et al., 2007, Okabe et al., 2007, Shanks et al., 2008, Shanks et al., 2009, Jeter et al., 2009, Lamendella et al., 2009, Silkie and Nelson, 2009; Mieszkin et al., 2009). The existence of a wide spectrum of host assays is advantageous when there is more than one dominant fecal source in environmental samples. It allows the investigator to detect multiple sources and compare their relative contributions to the total fecal contamination. Most of the original species-specific Bacteroidales assays were intended for conventional PCR applications, which provide only information about presence or absence of a target sequence. Later, real time quantitative PCR (qPCR) assays were designed to quantify the host-specific genetic markers (Layton et al., 2006, Reischer et al., 2006, Kildare et al., 2007, Okabe et al., 2007, Shanks et al., 2008, Shanks et al., 2009, Jeter et al., 2009, Lamendella et al., 2009, Silkie and Nelson, 2009, Mieszkin et al., 2009).

However, no qPCR assay is 100% specific and sensitive for its intended target. Even if the qPCR reaction itself is performed adequately and after the efficiency loss and inhibition factor are corrected for (Rajal et al., 2007, Silkie and Nelson, 2009), there remains intrinsic false information associated with the assay itself. The factors that may lead to false positive and false negative information in qPCR results are known and well defined. False information arises from the following occasions: (1) the presence and abundance of target genetic markers varies from individual to individual for a specific animal host; (2) a primer set for a specific host might amplify Bacteroidales DNA from other hosts (false positive information for that host-specific assay); (3) Bacteroidales DNA from a specific host might not be amplified by a primer set designed for that host (false negative information for that host-specific assay); (4) measurement errors may be generated in the sample preparation and the amplification steps during the qPCR process. It is, therefore, highly desirable to have a way of removing such false information to estimate the true concentration of host-specific genetic markers and help guide the interpretation of environmental monitoring studies.

The objective of this study was to use a statistical model based on the Law of Total Probability to assess the true concentration of indicator bacteria in a water sample that originated from various sources; such a model can be used to quantify fecal contribution from different land uses. The model’s performance was validated by real world samples and its accuracy and precision were evaluated by simulation studies.