U.S. Environmental Protection Agency
EVALUATION OF MULTIPLE INDICATOR COMBINATIONS
TO DEVELOP QUANTIFIABLE RELATIONSHIPS
U.S. Environmental Protection Agency
Office of Water
December 2010
EPA 822-R-10-004
-------
U.S. Environmental Protection Agency
Table of Contents
Executive Summary 1
1 Introduction 3
1.1 Purpose, Objectives, and Importance 3
1.2 Approaches for Developing Water Quality Measures Corresponding to Equivalent
Risks 4
1.2.1 Risk Link 4
1.2.2 Water Quality Link 6
2 Background 8
2.1 Review and Assessment of Relevant Epidemiology Studies andDatasets 8
2.1.1 Prospective Cohort Studies 8
2.1.2 Randomized Control Trials 9
2.1.3 Studies that have Generated Health Effects Relationships 9
2.1.4 Factors that Preclude Direct Comparison of Epidemiology Studies 10
2.2 Differences in Indicator Performance for Indicators as Measured by Culture-
Based and qPCR-Based Methods 11
2.2.1 General Method Performance 11
2.2.2 Factors Impacting Indicator/Method Performance 16
2.2.3 Uncertainty of Indicator/Method Combinations at Low Density 20
3 Analyses 21
3.1 Risk Link 21
3.1.1 Review of NEEAR Study and Other Health Effects Relationships for qPCR
and Culturable FIB 22
3.1.2 Direct Application of the Risk Link Approach: Linking Bacteroidales
Measured by qPCR to Enterococcus Measured by qPCR 24
3.1.3 Linking E. coli Densities as Measured by MF via Health Effects Curves from
Different Epidemiology Studies 24
3.1.4 Application of the Risk Link Approach for Health Effects Curves Based on
Different Illness Definitions 27
3.2 Water Quality Link Approach 29
3.2.1 Datasets Used in Water Quality Link Demonstrations 29
3.2.2 Linear Models of Log-Transformed Indicator Data 30
3.2.3 Demonstration of the Water Quality Linkage Approach using Linear
Regression Models 35
3.2.4 Broken Stick Models of Log-Transformed Indicator Data 38
4 Discussion 45
5 References 46
-------
U.S. Environmental Protection Agency
List of Tables
Table 1. Epidemiology study differences that limit direct comparison of health effects
curves 6
Table 2. Epidemiology studies that have established relationships between FIB density and
excess GI illness due to swimming 10
Table 3. Summary of factors affecting culture-based and/or qPCR data for fecal indicator
bacteria 11
Table 4. Summary of the target organisms and primer sets used for a selection of qPCR
assays 15
Table 5. Summary of health effects relationships from USEPA NEEAR studies and marine
studies conducted to support development of the current RWQC 23
Table 6. Risk-linked Bacteroidales and Enterococcus densities 24
Table 7. Results from ANCOVA comparing data from the USEPA (1984) and Marion
et al. (2010) epidemiology studies 26
Table 8. Equivalent Criteria in Terms of HCGI and NGI Definitions of Illness 28
Table 9. Examples of equivalent criteria values for marine waters via the statistical
linkages approach 29
Table 10. Results of ANCOVAs for sets of data for Huntington Beach with different
collection times 34
Table 11. Summary of models resulting from retention of data pairs with culture counts
above three thresholds 44
-------
U.S. Environmental Protection Agency
List of Figures
Figure 1. Illustration of the risk link approach 5
Figure 2. Illustration of the Water Quality Linkage approach 7
Figure 3. Paired qPCR and culture data for Enterococcus: Huntington B each 12
Figure 6. Correlation of qPCR Enterococcus densities from replicate samples 14
Figure 7. Trend lines for dilutions of laboratory cultures of E. coli and C.jejuni enumerated
using qPCR and culture methods 17
Figure 8. Comparison of bacterial data from culture-based and multiplex real-time PCR
methods, and statistical sample description 18
Figure 9. Health Effects Data and Trend Lines from the USEPA and Marion Studies 25
Figure 10. Typical plot of paired Enterococcus densities as measured by qPCR and culture-
based methods 31
Figure 11. Paired qPCR and culture Enterococcus data, all NEEAR study Great Lakes
beaches 32
Figure 12. Paired culture and qPCR Enterococcus data, all samples from Huntington Beach ....33
Figure 13. Paired qPCR and culture Enterococcus data, Huntington Beach: 8:00 AM
samples only 33
Figure 14. Paired qPCR and culture Enterococcus data, Huntington Beach: 3:00 PM
samples only 34
Figure 15. Linear model, data from all NEEAR freshwater beaches 36
Figure 16. Linear model, data from Huntington Beach 37
Figure 17. Linear model, 8:00 AM data from Huntington Beach 38
Figure 18. Huntington Beach broken stick model fits for all data and 8:00 AM data 39
Figure 19. West Beach broken stick model fits for all data and 8:00 AM data 40
Figure 20. Silver Beach broken stick model fits for all data and 8:00 AM data 40
Figure 21. Washington Park Beach broken stick model fits for all data and 8:00 AM data 41
Figure 22. Model resulting from retention of all data above the "kink" in the broken stick 42
Figure 23. Model resulting from retention of all data above 80 CFU/100 mL 43
Figure 24. Model resulting from retention of all data above 104 CFU/100 mL 43
IV
-------
U.S. Environmental Protection Agency
Executive Summary
This report was written to meet the EPA Critical Path Science Plan1 element P15. The
objectives of this work were to compare the fecal indicator bacteria and health effect
relationships for multiple indicator/method combinations used in epidemiology studies and to
evaluate multiple indicator/method combinations to develop quantifiable relationships. To meet
those objectives, two approaches were employed to relate the different indicator/method
combinations to gastrointestinal (GI) illness risks, namely the Risk Link and Water Quality Link.
In the Risk Link approach, indicator-method combinations are linked via health effects curves2
generated by epidemiology studies. Three demonstrations of the Risk Link approach are
presented. In the first demonstration, linkages are established between indicator densities for
multiple indicators used in the same epidemiology studies. This demonstration is a
straightforward implementation of the Risk Link approach and entails linkage of Enterococcus
density as measured by qPCR with Bacteroidales density as measured by qPCR. The second
demonstration illustrates an assessment of differences between epidemiology studies conducted
at different places or times, using an analysis of covariance (ANCOVA) to compare data from
(Marion et al.3) and (USEPA4) EPA recreational water epidemiology studies in freshwater. The
analyses indicate that although the two sets of studies were conducted at different times, the
slopes of their health effects curves are statistically similar.
The third Risk Link was performed using health effects relations from epidemiology studies with
Enterococcus and GI relationships, measured by either qPCR or culture-based method. The EPA
recreational water epidemiology study in marine waters5 and the EPA's National
Epidemiological and Environmental Assessment of Recreational (NEEAR) Water Studies
(marine) were linked to determine comparable Enterococcus culture and qPCR indicator
densities at various GI health risk levels. To enable the calculation of comparable Enterococcus
densities for the two studies, results from the 1983 USEPA study were first translated such that
the GI illness definition matched that of the NEEAR GI illness (NGI) definition. Although
numeric qPCR-based water quality criteria have not been established at this time, assumptions
about the way those criteria will likely be developed can be used to demonstrate potential qPCR
and culture-based criteria that are consistent with the same level of risk (table below).
1 USEPA. 2007. Critical Path Science Plan for the Development of New or Revised Recreational Water Quality
Criteria. U.S. Environmental Protection Agency, Offices of Water and Research and Development, Washington,
DC.
2 A health effects curve refers to a mathematical relationship between fecal indicator bacteria (FIB) density and
observed illness in epidemiology studies of recreational waters. All health effects curves referenced in this report
relate the incidence of gastrointestinal (GI) illness to the log-transformed FIB density
3 Marion, J.W., Lee, I, Lemeshow, S., Buckley, TJ. 2010. Association of illness and recreational water exposure
during advisory and non-advisory conditions at an inland U.S. beach. Water Research 44(16): 4796-4804.
4 USEPA 1984. Health Effects Criteria for Fresh Recreational Water. EPA-600/1-84-004. Research Triangle Park,
NC: USEPA.
5 USEPA 1983. Health Effects Criteria for Marine Recreational Waters. EPA-600/1-80-031. Research Triangle
Park, NC: USEPA.
-------
U.S. Environmental Protection Agency
Tolerable
attributable
illness level (as
HCGI* per 1000
swimmers)
8
10
19
Tolerable
attributable
illness level (as
NGI* per 1000
swimmers)
35
43
82
Hypothetical
qPCR
Enterococcus
density
(CCE§/100ml_)
427
610
3460
Hypothetical
geometric mean
membrane filtration
Enterococcus density
(CFU/100ml_)
11
14
35
75th
percentile
value*
(CFU/100
mL)
33
42
104
T The 75th percentile value for Enterococcus density based on the calculated geometric mean
Enterococcus density, assuming Enterococcus densities are log-normally distributed, and
assuming a typical standard deviation of log-transformed Enterococcus density for marine sites
of 0.7.
t Highly Credible Gastrointestinal Illness, as defined in the epidemiology studies conducted in
support of the 1986 water quality criteria
* NEEAR study Gl illness (NGI, per the definition used in the NEEAR epidemiology studies)
§ Calibration cell equivalent; using a calibration sample containing a known concentration of the
target sequence, CCEs are the normalized values of the test sample cell equivalents
The Water Quality Link approach links an indicator-method combination for which there is no
health effects relation to an indicator-method combination with a health effects relation via a
quantifiable relationship between their measured fecal indicator bacteria densities. In short, the
Water Quality Link Approach uses paired water quality data, rather than the linkage of health
effects curves alone, as the basis for an alternative method for establishing culture-based criteria.
Demonstrations of this approach were made using densities of Enterococcus enumerated by
qPCR and by membrane filtration (MF) and linked to the health effects relationship from EPA's
NEEAR freshwater studies. It was found that while the Water Quality Link may be useful on a
site specific basis, in this particular demonstration the relationships between paired Enterococcus
data, measured by culture and qPCR, were not consistent among NEEAR study freshwater
beaches when simple linear and broken stick (segmented) regression models were employed.
Further, the regression fits to this dataset exhibited heteroskedasticity (uneven distribution of
residuals) and both the simple linear regression and broken-stick models resulted in comparable
Enterococcus culture criteria values far in excess of the current criteria.
The exploration of the Risk Link and the Water Quality Link approaches provides a proof of
concept with the currently available data. It is likely that new health effects data or improved
models of the co-occurrence of indicators will allow for additional or improved linkages to be
established.
-------
U.S. Environmental Protection Agency
1 Introduction
1.1 Purpose, Objectives, and Importance
Pathogens within fecal matter arising from human and animal sources that enter coastal and
inland waters pose risks to recreational swimmers. Because it is not feasible to monitor for all
pathogens that may occur in ambient waters, monitoring strategies currently involve the
detection of indicator organisms that are present in fecal material in greater numbers than the
pathogenic organisms (NRC 2004). For decades, EPA has relied on the use of epidemiology
studies to assess the risk of gastrointestinal (GI) illness in swimmers exposed to increasing
densities of fecal indicator bacteria (FIB).6 Previous epidemiology studies used only culture-
based methods for enumerating FIB. More recent epidemiology studies have evaluated
associations between GI illness and a wider range of indicator organisms, using both improved
culture techniques and molecular-based (rapid) methods.
The primary purpose of this report is to help meet one of the elements (Project PI 5) in the U.S.
Environmental Protection Agency (EPA or the Agency) Critical Path Science Plan for
Development of New or Revised Recreational Water Quality Criteria (CPSP or science plan)
(USEPA 2007). The science plan is a key component of EPA's overall process to develop new
or revised Section 304(a) Ambient Water Quality Criteria (RWQC) for recreational waters that
will be used by States, Territories, and Tribes to develop their own water quality standards
(WQS7).
The objectives of CPSP Project PI5 are as follows:
• To compare the GI illness response to exposure relationship for multiple fecal indicator
organism/method combinations; and
• To develop quantifiable relationships between the results from the various
indicator/method combinations.
More generally, PI5 commissions scientific studies to explore correlations and linkages between
available indicator organisms (FIB) and methods. These linkages are established to ensure that
equal protection is provided for any future criteria associated with each indicator-method
combination.
Section 2 provides a background review and assessment of relevant epidemiology studies, as
well as a review of differences in FIB performance for indicators enumerated culture and
quantitative polymerase chain reaction (qPCR). The background supports the analyses described
in Section 3 that analyze available epidemiology datasets using both the "Risk Link" and the
6 "Traditional" FIB include culturable total coliforms, fecal (thermotolerant) conforms, Escherichia coli (an
important member of the coliform group), and Enterococcus (enterococci). Although the presence of FIB indicates
the presence of fecal matter, and the potential presence of (enteric) pathogens, FIB are not pathogenic (NRC 2004;
WHO 2003).
7 Under the Clean Water Act (CWA) States, Territories, and Tribes are required to adopt new or revised WQS for
those pathogens and pathogen indicators for which EPA's new or revised criteria have been developed. Once
approved by EPA, WQS are used for various CWA purposes and programs and are the effective (enforceable)
standards to protect waters for specified designated uses, such as "primary contact recreation."
-------
U.S. Environmental Protection Agency
"Water Quality Link" approaches. Both of these approaches are described in detail in later
sections of this report. Section 4 provides a discussion of the analyses, including limitations and
assumptions.
1.2 Approaches for Developing Water Quality Measures Corresponding to Equivalent
Risks
At the core of RWQC is a quantitative relationship between a measure of fecal pollution in water
and the risk of adverse health outcomes (health effects curves) arising from primary recreational
contact (e.g., swimming) in the polluted water. Although the expense, complexities, and time
associated with conducting epidemiology studies are significant, several studies have been
completed and are ongoing in the U.S. and abroad since the issuance of the current (1986)
RWQC. Many of those studies were conducted in direct support of the development of RWQC,
standards, and guidance. In addition, new technologies, such as rapid molecular-based detection
methods, have been developed for identifying and enumerating FIB and other (alternate)
indicator organisms that are fundamentally different from traditional culture-based methods.
None of these approaches (new methods, alternative indicators, or alternate risk assessment
methods) involves direct measurement of the pathogen(s) suspected of causing illness in
recreational water users. Therefore, linkages are made between the indicators that are being
measured and the adverse health effects observed in relevant and available epidemiology studies.
The health effects curves from some of the epidemiology studies may not be compared without
prior analysis and transformations. Differences that necessitate transformations include, for
example, differences in study design (prospective cohort [PC] vs. randomized control trial
([RCT]) (see more below), differences in definitions of GI illness, and differences in the settings
(marine vs. freshwater environments). Ongoing epidemiology studies may yield health effects
curves for additional indicator/method combinations and these health curves may be suitable for
use in developing linkages between indicator/method combinations.
Further, incorporation of molecular-based methods into RWQC based on culture methods (or
incorporation of culturable methods into criteria based on molecular methods) is complicated by
the differences in their targets. Because different methods measure different targets with
different abundances in environmental samples, it is not a given that health effect relationships
developed for one particular indicator-method combination are valid for another indicator-
method combination. This report places emphasis on the linkages between (1) Enterococcus
measured by qPCR and Bacteroidales measured by qPCR for publicly owned (sewage) treatment
works (POTW)-impacted marine settings; (2) E. coli measured by membrane filtration (MF) for
studies conducted at different times and in different settings; and (3) Enterococcus measured by
qPCR and Enterococcus measured by MF for studies of marine beaches conducted at different
times.
1.2.1 Risk Link
1.2.1.1 Approach Description
The most direct route to meeting the CPSP PI 5 objectives is through use of health effect-FIB
density curves to determine quantifiable relationships for indicator/method combinations that
correspond to the same risk level for swimmers. This approach, termed the Risk Link, is
illustrated in Figure 1. For this approach to be feasible, all indicator-method combinations
4
-------
U.S. Environmental Protection Agency
should have an associated relationship between indicator density and health effects (the solid
lines in both plots in Figure 1). At present, all such health effects curves have been developed
based on epidemiology studies of surface water recreation sites. Future relationships may be
developed using QMRA, watershed modeling approaches, or other methods developed
specifically for that purpose. Assuming the indicator densities and illness rates for the standard
and alternative indicator health curves are comparable (and this assumption is addressed below),
the indicator density corresponding to a selected acceptable level of risk is calculated using the
health effects relationship for each of the indicator/illness combinations.
Illnesses
/1000
Acceptable
risk
Illnesses
/1000
Criterion
Log10 (Indicatordensity)
Figure 1. Illustration of the risk link approach
Log10 (Alternative indicatordensity}
The rationale for using a Risk Link approach is that it maximizes use of direct measurements of
health effects from relevant and available epidemiology studies. Over the recreational season,
the pathogens to which swimmers are exposed and the densities of fecal indicators are variable.
Further, the characteristic pathogens and FIB densities at different beaches likely differ,
depending on the alignment of the beach with fecal pollution sources and other factors such as
climate and rainfall. Given this variability, measurement of the association of adverse health
effects with FIB through epidemiology studies appears to be the most direct and reliable means
for relating water quality to health effects.
1.2.1.2 Harmonizing Data from Disparate Studies
For health effects curves to be comparable, they should relate to risks for the same illness.
Further, the statistical measure used for characterizing water quality (e.g., geometric mean of
multiple samples on a single day vs. the water quality from a single sample taken in a zone and at
a time where swimming occurs) should be taken into account. An extreme example of curves
that would not be comparable is curves that correspond to excess GI illness in swimmers and
excess respiratory infection in non-swimmers. A summary of potential differences between
epidemiology studies that could prevent direct comparison of health effects curves is presented
in Table 1.
-------
U.S. Environmental Protection Agency
Table 1. Epidemiology study differences that limit direct comparison of health effects curves
Cause
Differences in data construct
Different
epidemiology
study design
(PC vs. RCT)
Exposures in prospective cohort (PC) studies are not prescribed and have a much
wider range of durations and ingestions typical of recreation events; exposures in
randomized control trial (RCT) studies are controlled and have more consistent
ingestion volume, but less variability in exposure.
RCT studies control the location of exposure far more than in PC studies; thus, the
impact of this element of study design is site specific.
Water quality associated with illness incidence in PC studies is based on an
average indicator (usually FIB) density for the recreation site and over the entire
study day; water quality associated with illness incidence in RCT studies is based
on the indicator density for a sample taken at the same time and location as
swimmer exposure. Proponents of RCT designs state that this feature reduces
bias, but this claim may not be true depending on the magnitude of short-term
variability.
Different
definition of
human health
outcome
Epidemiology studies differ in their choices in health endpoints and their definitions of
those endpoints, including Gl illness.
Different settings
Indicators may be associated with different risks at sites with different
• fecal pollution sources;
• level of treatment of fecal pollution;
• loading characteristics (continuous vs. event); and
• proportion of FIB resuspended in sediments.
Examples of pairs of studies that may have significant setting-related effects include
• studies conducted in inland and coastal waters,
• studies conducted in the United States and Europe, and
• studies conducted at sites in (sub)tropical and temperate climates.
Regulation
changes and
technological
advances
Regulations and advances in technology have changed wastewater (including POTW)
treatment practices and the resulting indicator and pathogen loads to recreation sites.
The most important of these changes are disinfection of wastewater, development of
improved animal waste treatment technologies, and regulations on the land application
of biosolids (treated wastewater sludge) and animal wastes.
1.2.2 Water Quality Link
Many indicator/method combinations are not, at present, associated with health effects curves or
have not been applied directly in epidemiology studies. In such cases, a less direct approach for
linking indicator/method combinations can be applied. The Water Quality Linkage approach
relates paired data from two indicator/method combinations and links the method/indicator
combinations to each other through their relationship, rather than through a direct linkage to
health effects curves. This approach is illustrated in Figure 2, where in the top graph, a criterion
density for a standard indicator/method combination is established based on a selected level of
tolerable risk and a health effects curve. A quantifiable relationship for an alternative
method/indicator combination is established through a model relating the density of the standard
indicator/method combination to the density of the alternative method/indicator combination.
-------
U.S. Environmental Protection Agency
>
_
3
o
(Indicator
density)
Log10
(Indicator
density)
Figure 2. Illustration of the Water Quality
Linkage approach
As with the Risk Link approach, the Water Quality Link should be employed when paired water
quality data are from comparable studies. Further, a Water Quality Linkage at one site may not
be applicable to another site—even if the sites have the same primary fecal pollution impact. A
critical component in the application of the Water Quality Link is the development of a useful
statistical model that relates Fffi-method combinations. At present, models relating culturable
and qPCR indicator counts are either site-specific (e.g., Byappanahalli et al. 2010; Lavender and
Kinzelman 2009) or have been developed based on pooling of datasets under the assumption that
datasets are similar and may be pooled (Haugland et al. 2005; Whitman et al. 2010). In this
report, linear, and "broken stick" models for relating log-transformed culture and qPCR indicator
densities are explored. It is possible that other statistical models linking these types of analytical
methods will be proposed and evaluated in the future.
-------
U.S. Environmental Protection Agency
2 Background
Ordinarily, FIB themselves do not cause illness, but their densities can provide estimates overall
levels of fecal contamination. As a result, their densities cannot usually be used to directly
estimate health risks through a risk assessment approach. The relationship between indicators
and health risks are best and most commonly established by epidemiology studies (NRC 2004;
WERF 2009). Epidemiology studies (1) establish microbial water quality, typically via FIB
density measurements, with a sufficient number and timing of samples to establish a
characteristic indicator density during swimming; and (2) associate the FIB density with the
adverse health effects observed among the population swimming compared to a non-swimming
control population. This section includes a review and assessment of relevant and available
epidemiology studies, as well as a review of differences in FIB performance for culture and
qPCR indicators.
2.1 Review and Assessment of Relevant Epidemiology Studies and Datasets
Since the 1950s, numerous epidemiology studies have been conducted in the United States and
abroad, most commonly at beaches impacted by sewage/wastewater effluent (e.g., POTWs), to
evaluate the association between recreational water quality and adverse health outcomes. In
these studies, attempts were made to relate a quantitative microbial indicator of water quality to
health effects (usually some form of GI illness) using log-transformed data or a geometric mean
to characterize exposure to water quality indicators. However, eye infections; skin irritations;
ear, nose, and throat infections; and respiratory illness have also been evaluated. Many of these
recreational water epidemiology studies are reviewed in one or more of the meta-
analyses/systematic reviews of Priiss (1998), Wade et al. (2003), and Zmirou et al. (2003). More
recent studies are reviewed in WERF (2009), which notes that all recreational epidemiology
studies identified higher rates of at least some self-reported health end points (usually GI illness)
in relation to water exposure (usually swimmers vs. non-swimmers). That is, recreational water
contact by its nature is associated with increased risk of adverse health effects—even if the
excess risk is not correlated with increases in fecal indicator organisms.
Both PC and RCT (also called prospective randomized exposure studies [Fleisher et al. 2010])
epidemiology study designs have been used to evaluate recreational waters. The primary
difference between RCT and PC design is that in RCT studies the volunteers are randomly
assigned to swim in a selected area where the water quality is measured during the timed
swimming exposure. A brief overview of each design is provided below.
2.1.1 Prospective Cohort Studies
EPA's current (1986) recreational water quality are based on the observed occurrence of GI
illness associated with swimming in fresh or marine recreational waters receiving point sources
of effluent from POTWs as determined through several PC studies conducted in the 1970s
through the early 1980s. Over the past several years, the EPA conducted a PC study called
National Epidemiological and Environmental Assessment of Recreational (NEEAR) Water
Study. This series of epidemiology studies evaluated sites at four POTW-impacted Great Lakes
beaches and five marine recreational beaches (four POTW-impacted and one non-POTW-
impacted). The results of the freshwater and marine studies have been published (e.g., Wade et
-------
U.S. Environmental Protection Agency
al. 2006, 2008, 2010). Further, there have been relatively few studies of inland (non-Great
Lakes) waters or of nonpoint source-impacted recreational sites.
2.1.2 Randomized Control Trials
The other major epidemiology study design used for evaluating health risks associated with
recreational exposures is the RCT. This design has been used extensively in Europe, and more
recently in the United States to determine the association between microbial water quality and
increased risk of adverse health effects in swimmers vs. non-swimmers in both marine and fresh
recreational waters. The WHO Guidelines for Safe Recreational Water Environments (WHO,
2003) apply to both fresh and recreational waters and are largely based on RCT studies
conducted in the late 1980s and early 1990s at POTW-impacted marine recreational beaches in
the United Kingdom (UK) (Fleisher et al. 1996; Kay et al. 1994). In 2006, the European Union
adopted a new directive for the management of bathing water quality that is also based, in large
part, on the same UK marine recreational water studies. It also includes the results of a more
recent RCT study of POTW- and nonpoint source-impacted fresh recreational waters in Germany
(Wiedenmann et al. 2006). Epibathe (2009a, 2009b) describes the results of a series of RCT
studies conducted at marine and fresh recreational waters in Europe in 2006 and 2007. Finally,
the results of an RCT study of a non-POTW-impacted marine beach in Miami, Florida have been
recently published (Fleisher et al. 2010).
2.1.3 Studies that have Generated Health Effects Relationships
A number of epidemiology studies have established a significant relationship between indicator
organism density and increased GI illness (Table 2; see also WERF, 2009). Highlighted rows in
Table 2 indicate relevant and available epidemiological datasets that were obtained and evaluated
in this report (see Section 3). It is beyond the scope of this report to list all epidemiology studies
and datasets of recreational waters that have been conducted.
Among the studies listed in Table 2, three have established relationships for inland waters, and
all of those studies were of waters that are likely predominantly impacted by human fecal
pollution sources. A number of the unique features for the studies that resulted in health effects
relations are the following:
• the USEPA (1983, 1984) studies related illness rates to season-averaged culture indicator
densities;
• the USEPA (1984) study health effects relationship was developed using pooled data
collected for both inland and Great Lakes beaches;
• the USEPA (1983, 1984) studies used different definitions of GI illness than the NEEAR
studies;
• the study by Marion et al. (2010) was conducted on a relatively small inland lake, though
as shown in Section 3.1, the health effects observed in that study were similar to those
observed by USEPA (1984).
In contrast to studies of waters potentially impacted primarily by POTW effluent, epidemiology
studies of marine and freshwater non-POTW impacted-recreational waters tend to yield weak
associations between increased densities of traditional indicators (Enterococcus and E. coli) and
health risk. Calderon (1991) found no association of indicator density with incidence of adverse
-------
U.S. Environmental Protection Agency
health effects at a non-POTW-impacted pond with no known human fecal pollution effects.
Similar results were reported by Colford et al. (2007) for a California coastal beach suspected to
be affected primarily by birds, and for a subtropical coastal marine beach most likely affected by
dog and human nonpoint sources of fecal pollution (Abdelzaher et al. 2010; Fleisher et al. 2010;
Sinigalliano et al. 2010). Results from a study of stormwater impacts on GI illness rates are
difficult to interpret (Haile et al. 1999). It is possible that densities of alternative (non-
traditional) indicators might be associated with risk and that relationships may be established as a
result of ongoing epidemiology studies.
Table 2. Epidemiology studies that have established relationships between FIB density
excess GI illness due to swimming
and
Study(s)
Marion et al.
(2010)
NEEAR
NEEAR (Wade et
al. 2006, 2008)
NEEAR (Wade et
al. 2006, 2008)
Wiedenmann et
al. (2006)
Fleisher et al.
(1 996) and Kay et
al. (1994)
USEPA(1984)
USEPA(1983)
Indicator and primary
detection method
£. co/; culture
Density of Bacteroidales
measured by qPCR
Density of Enterococcus
spp. measured by qPCR
Density of Enterococcus
spp. measured by qPCR
Enterococcus and £. co/;
chromogenic substrate
Enterococcus culture
Enterococcus and £. co/;
culture
Enterococcus culture
Setting and sources
U.S. impounded freshwater beaches,
human and other sources
U.S. marine beaches, POTWs
US freshwater beaches (Great Lakes),
POTWs
U.S. marine beaches, POTW
German, freshwater beaches with one of
more point- (including POTWs) and
nonpoint sources
U.K. marine beaches, POTWs
U.S. Great Lakes and inland freshwater
beaches, POTWs
U.S. marine beaches, POTWs
Study
type
PC
PC
PC
PC
RCT
RCT
PC
PC
2.1.4 Factors that Preclude Direct Comparison of Epidemiology Studies
Although epidemiology studies may be able to identify a general association between a given
fecal pollution source and indicator organism densities by estimating incidence of disease, a
major and ongoing concern is that their results may be limited to describing risk only for beaches
similar to those evaluated in the epidemiology studies (e.g., similar fecal sources). The ability to
conduct valid comparisons between PC and RCT epidemiology studies depends on several key
and potentially interrelated factors. These factors relate to several critical differences in the
details of RCT and PC studies, and include the following:
• the manner in which microbial water quality is associated with illnesses;
• the specific exposure that swimmers experience;
10
-------
U.S. Environmental Protection Agency
• the definition of (GI) illness and the duration of follow-up;
• the age and make up (e.g., tourists vs. locals) of the subject pool;
• the source of fecal indicators and pathogens in the recreational water, their distribution,
and temporal-spatial variability of those distributions; and
• the methods used to enumerate the indicators and the corresponding relationships
between the indicators and the pathogens, as measured by those methods.
2.2 Differences in Indicator Performance for Indicators as Measured by Culture-Based
and qPCR-Based Methods
Cell counts obtained by qPCR are generally greater, often by orders of magnitude, than those
provided by MF analyses of the same samples (e.g., as observed by He and Jiang, 2005). This
section begins with a review of select studies illustrating the differences between qPCR and MF
methods. Common causes of variations in FIB counts via the different methods, as shown in the
literature, are listed below by category and subcategory and are discussed subsequently. Results
from studies on paired comparisons of different indicator/method combinations are also
highlighted throughout the section.
2.2.1 General Method Performance
The performance of selected analytical methods and the fate and transport of cells and genomic
material can be influenced by a number of factors, as summarized in Table 3.
Table 3. Summary of factors affecting culture-based and/or qPCR data for fecal indicator bacteria
Influencing factor
Time of day for
sample collection
Variations in method
performance
Bacterial viability
Variables
Exposure to sunlight
(UV radiation)
• qPCR amplification
efficiency
• Limits of detection
• Inter-laboratory
variability
• Choice of target
gene(s)
• Sample preparation
• Concentration
factor
Enumeration of live,
dead, and/or VNBC
FIB
Outcome
Diurnal variation in FIB
density
Inconsistencies in data
generation and
interpretation
Under-representation of
VNBC FIB using MF; over-
representation of viable
bacteria using qPCR
Reference(s)
Haugland et al. (2005)
Haugland et al. (2005)
Khan et al. (2007)
Kinzelman etal. (2010;
preliminary report)
Mocker and Camper (2006)
Mocker et al. (2006)
Bae and Wuertz (2009)
Varma et al. (2009)
With regards to the sampling approach, the time of day at which samples are collected has been
shown to influence the relationship between culture-based and qPCR data (e.g., Haugland et al.
2005). During EPA's NEEAR studies conducted at freshwater sites in the Great Lakes, water
samples were tested for Enterococcus by qPCR and MF. Samples were collected three times per
day (8:00 AM, 11:00 AM, and 3:00 PM), at multiple depths (waist, shin, and knee), and along
three transects at all four beaches being studied. A model relating enterococci qPCR and culture
data can be developed using results from all the sites and times, or using data that are selected
11
-------
U.S. Environmental Protection Agency
because they appear to have similar underlying trends. All samples collected from one of these
beaches are shown in Figure 3, along with a 45° line showing perfect agreement. Although one
may observe a general trend at high FIB density, at densities below -500 CFU/100 mL, no trend
is apparent and there is no obvious choice for the form of a model relating enterococci density
derived from qPCR methods to culturable enterococci density. The influence of sample
collection time on the relationship between culture and qPCR FIB density is explored in Section
3.2.1.
s> -I
10
100 1000
CFU/100mL
10000
Figure 3. Paired qPCR and culture data for Enterococcus:
Huntington Beach
Variations in the data because of the performance of the analytical method have been
demonstrated in a number of studies. The factors affecting the analytical performance include
the efficiency of the qPCR method, the influence of detection limits, and variations in inter-
laboratory performance. Haugland et al. (2005) compared a qPCR-based enumeration method
for Enterococcus to EPA Method 1600 (MF enumeration). The qPCR method employed in that
NEEAR-related study demonstrated very high amplification efficiency (0.99, or 99% of
amplicons doubling with each cycle) for DNA in dilution water. A modest level of false positive
results from qPCR (19% of 217 negative control samples) was speculated to be the result of
aerosolized DNA contamination that occurred in analytical laboratories. DNA recoveries in
calibration samples (from seeded filters) from the two beaches sampled in the study were 82%
and 51% in beach samples, respectively. Results of qPCR and MF enumerations were
reasonably well correlated (R2 = 0.68) and log-normal distributions described the densities of
Enterococcus for samples collected at each beach and for both qPCR and MF enumerations. The
authors performed linear regression of qPCR results (as cell equivalents) against MF results (as
CFU) for samples collected on multiple days and at multiple locations on two beaches. The
resulting relation:
12
-------
U.S. Environmental Protection Agency
0.531og10(CFE7) [1]
was proposed to describe the variation in qPCR cell counts with those from MF.
Khan et al. (2007) evaluated qPCR enumeration of E. coli in waters of agriculture-dominated
watersheds against MF methods, finding as did Haugland et al. (2005), that (1) qPCR-based
enumerations yielded consistently higher estimates of density than culture methods (attributed to
lack of discrimination between DNA from live and dead cells); and (2) that standard curves (in
this case, based on both dilution water and autoclaved agricultural water) had high coefficients of
variation. Khan and colleagues noted that the detection limit in agricultural waters was 10 cells,
whereas in dilution water it was 1 cell, and that sample preparation had a profound impact on the
viability of the qPCR method. Overall, qPCR enumerations were found to be consistently higher
than those from MF methods—the cell counts for all samples analyzed via MF ranged from 1.0
to 2800 CFU/100 mL, whereas the range for qPCR counts was 15 to 9900 cells/100 mL. The
authors concluded that qPCR results were less variable than MF results; that qPCR could be used
effectively in diverse agricultural watersheds; and that qPCR is more rapid, producing
meaningful results far faster than culture-based indicator methods. Application of qPCR
methods and interpretation of qPCR results may require consideration of the source of indicator
organisms as well as the physiological status of FIB at the time of sampling. For example, the
selection of target gene plays an important role in qPCR enumeration. The number of target
genes per bacteria cell is dependent upon physiological state of the cell and, for example, during
the bacterial logarithmic-growth phase up to 36 rrn genes/cell (range: 12-36 copies/cell) have
been reported for E. coli (Bremer and Dennis 1996).
Inter-laboratory variability was assessed by Kinzelman et al. (2010; preliminary report) through
the analysis of replicate samples in multiple labs to assess the uncertainty and recovery of qPCR.
Archived (MF) filters were used to develop replicate samples and qPCR was used to determine
Enterococcus calibrator cell equivalents (CCEs) in the replicate samples. Multiple laboratories
conducted the study and all reported similar findings. A typical plot of the paired samples is
presented in Figure 4. In all cases there was very little difference between replicate samples at
high indicator density and higher scatter at lower densities. The scatter at low densities appears
to be Poisson distributed (analysis not shown) and attributable to sampling uncertainty. The
relatively tight distribution of points around the 45° line indicates a consistent and relatively high
recovery.
13
-------
U.S. Environmental Protection Agency
1000000 n
100000 -
10000 -
CM
£ 1000
(c
o
"5.
a: 100
10 -
R2= 0.9655
10
100 1000
Replicate 1
10000
100000
1000000
Figure 4. Correlation of qPCR Enterococcus densities from
replicate samples (Source: Kinzelman et al. 2010)
Comparing some of the key papers used to review general method performance, it is important
that the choice of target gene(s) and reaction conditions for the qPCR assays be accounted for
when comparing research approaches. As shown in Table 4, there are a variety of bacteria
targeted in these key papers, some targeting the genus as a whole (e.g., Enterococcus spp.),
others targeting a particular species (e.g., E. coli), and others targeting a particular strain of a
species (e.g., E. coli O157:H7).
Accounting for bacterial viability is an important consideration when comparing culture-based
methods with qPCR assays. Because of its sensitivity, qPCR will amplify DNA regardless of
whether or not it is contained within an intact cell. However, successfully culturing bacteria
requires the organism to be intact. There is an intermediate phase termed viable but non-
culturable (VBNC), where bacteria appear to not be culturable but retain the ability to reproduce
under appropriate conditions. By their nature, culture-based methods will discriminate against
free DNA and VNBC bacteria, whereas qPCR methods will quantify genetic material from cells
in all states of existence. In the case of detection of DNA from dead cells, this ability is a
significant and widely recognized shortcoming of qPCR methods and a potential cause of false
positives (NRC 2004).
Several researchers have explored PCR techniques that enable discrimination between DNA
from live and dead cells. For example, Nocker and Camper (2006) used a chemical reaction of
DNA from dead cells with ethidium monoazide (EMA) to prevent reaction of DNA from dead
cells with PCR reagents. Thus, prior to qPCR determination, EMA was added to sample water.
EMA, which bonds strongly with DNA and is inactivated in water, can enter only bacterial cells
with compromised cell walls. Following this step, DNA was extracted from the live cells using
conventional techniques and qPCR was performed for determination of E. coli O157:H7 and
14
-------
U.S. Environmental Protection Agency
Table 4. Summary of the target organisms and primer sets used for a selection of qPCR assays
Reference
Haugland etal. (2005)
Khan et al. (2007)
Mocker and Camper (2006)
Mocker et al. (2006)
Bae and Wuertz (2009)
Varma et al. (2009)
Target
organism
Enterococcus
E. coli
E. coli
0157:H7
Salmonella
enterica
E. coli
O157:H7
Bacteroidales
Enterococcus
spp.
Bacteroidales
Target gene
23R rRNA
Internal transcribed
spacer (ITS) region
between the 16S-23S
rRNA
Shiga toxin 1
invA gene
Shiga toxin 1
16SrRNA
16SRNA
16SrRNA
Primers
ECST748F
ENC854R
IEC-UP
IEC-DN
stxl -forward
stxl -reverse
invA2-F
invA2-R
stxl -forward
stxl -reverse
Bacllni-520f
BacUni-690r1
BacUni-690r2
BacHum-160f
BacHum-241r
ECST748F
ENC854R
In Siefring et al. (2008)
Salmonella. The authors suggested that use of EMA with DNA-based methods, if refined, could
be a viable alternative to the use of more complicated RNA methods (RNA degrades rapidly
after cell death) in developing cell density estimates that exclude dead cells. In a different study,
Nocker et al. (2006) suggested that propidium monoazide (PMA) may be a better reagent for use
in preventing DNA from dead cells from being amplified in the PCR reaction. In that study,
EMA was found to penetrate live cells of some microbiological species, preventing amplification
of DNA from the penetrated cells and reducing the accuracy of qPCR estimates of cell density.
In contrast, PMA was observed to be selective only for dead cells for the organisms tested in that
study.
Similarly, Bae and Wuertz (2009) used PMA to inhibit amplification of DNA from dead cells.
The use of PMA (rather than EMA) was suggested because, and as noted by Nocker et al.
(2006), EMA is believed to cause degradation of some DNA from viable cells for E. coli
O157:H7, Campy lobacterjejuni, Listeria monocytogenes, and perhaps other pathogenic and
indicator bacteria. Bae and Wuertz (2009) determined PMA reaction conditions (PMA
concentration and light exposure time) that optimized removal of target DNA (host-specific
Bacteroidales genetic markers) from dead cells in wastewater plant influent and effluent. For
samples from the wastewater plant effluent, gene copies from qPCR with PMA were only 30%
15
-------
U.S. Environmental Protection Agency
of those from qPCR without PMA. The difference between qPCR with and without PMA was
greater than two orders of magnitude for samples of wastewater plant effluent.
Varma et al. (2009) investigated the difference between PMA-qPCR and qPCR without PMA for
Enterococcus and Bacteroidales in wastewater. Objectives of the study included assessment of
water matrix effects on the efficiency of the PMA reaction and exploration of use of qPCR for
analysis of wastewater treatment plant operation. Some of the conclusions from that study were
the following:
• high levels of biomass or suspended solids in water samples appeared to interfere with
the ability of the PMA-qPCR method to specifically detect live cells, and
• standard POTW chlorine disinfection practices resulted in substantially greater reductions
in fecal indicator bacteria CPU densities than those observed for PMA-qPCR detectable
target sequences.
When reviewing general method performance, the influence of variability in the data generated
from different methods and their influence on health effect relations has also been described.
Similar to Haugland et al. (2005), albeit for groundwater samples, Lleo et al. (2005) found that
qPCR estimates of E. coli and Enterococcus faecalis density were significantly and consistently
higher than those obtained using MF methods. The authors ascribed the difference to the ability
of qPCR to detect VBNC cells and suggested that the qPCR method provides estimates of FIB
density that are more protective of human health—particularly given the potential for VBNC
cells to resuscitate under appropriate conditions.
In the course of EPA's NEEAR studies conducted at Great Lakes beaches, Wade et al. (2008)
observed that, in contrast to counts from culture methods, qPCR counts of enterococci were
relatively constant during the day. In contrast, counts of enterococci from culture methods show
a distinctive diurnal variation, with densities for early morning samples being as much as two
orders of magnitude higher than for late afternoon samples. In addition to observing less hourly
variation in indicator density measurements using qPCR compared to culturable methods, Wade
and colleagues also noted that increasing qPCR densities (expressed as cell equivalents) were
associated with excess risk of GI illness in swimmers vs. non-swimmers.
2.2.2 Factors Impacting Indicator/Method Performance
2.2.2.1 Choice of Target Organism
Although this report focuses on the methods used to detect and quantify Enterococcus in relation
to health effects, it is important to consider how different FIB compare to one another when
detected in recreational waters by various methods. Differences in the comparison of qPCR and
culture signal exist for different microorganisms, as demonstrated by Rothrock et al. (2009). In
that study, triplicate analyses were performed on samples inoculated with dilutions of laboratory
cultures of E. coli and C.jejuni. Plots of qPCR counts vs. MF counts for the two organisms are
presented in Figure 5, where the dashed line is a 45° line (exact concordance between methods).
In general, the qPCR and culture counts are 1:1, with better agreement at the higher counts
(lower dilutions) than at lower counts. The curve for C. jejuni has a slope of 1:1, but is shifted
above the 45° line. This shift is potentially due to low recoveries typical for this bacterium via
culture methods. As suggested in Figure 5, the impact of difference in recovery between the
methods is manifested as a shift of the trendline above or below the hypothetical line of perfect
16
-------
U.S. Environmental Protection Agency
agreement. Differences in recovery among MF counts may results from damage or shock to
bacteria during sample preparation, clumping of cells, or difficulty in maintaining growth
conditions for some organisms. Differences in recovery among qPCR counts may arise from the
presence of substances inhibiting the PCR reaction and differences in specificity, efficiency, and
sensitivity for PCR assays targeting different microorganisms.
l.OOE+09
l.OOE+08
l.OOE+07
O- l.OOE-06
o
3 l.OOE-05
Ji l.OOE+04
"5
iJ 1.00E*Oi
l.OOE-02
l.OOE+01
l.OOE+00
l.E
^ £ CO// X
H .^
UC.jejuni .~
s
s
m '*
• X
• x
^' *
X *
X
X
X
X
x
-00 l.E+02 l.E+04 l.E+06 l.E+08
E. coli (CFU /100 ml)
Figure 5. Trend lines for dilutions of laboratory cultures of
E. coli and C.jejuni enumerated using qPCR and culture
methods (SOURCE: Rothrock et al. 2009)
Simultaneous detection of Enterococcus and Bacteroidales in recreational marine water was
undertaken by Elmir et al. (2009) using MF and qPCR detection methods. Two pools, one large
and one small, were filled with local offshore marine water and recreational users (adults and
toddlers) were required to be in the water for pre-determined periods. Using three analytical
methods, Enterococcus and Bacteroidales were enumerated from the water samples before and
after users were in contact with the water. Microbial concentrations in the source water were
generally low (MF = 5 [std. dev. ± 7] CFU, qPCR = 29 [± 49] genomic equivalent units).
Bacteroidales was detected only using qPCR human markers (UCD and FIF8), yielding genomic
equivalent units (GEU) of 45 (±183) and 3 (± 10) GEU/100 mL, respectively. The
concentrations of Enterococcus and Bacteroidales in both pools were variable following
recreational use. Enterococcus densities calculated using MF ranged from l.SxlO4 to 2.0xl06,
whereas values calculated using qPCR ranged from 3.8xl05 to 5.5xl06. Bacteroidales values
ranged from below the limit of detection (1.4x 103 GEU) to 1.3 x 106. The authors concluded that
the bathers appeared to release significant amounts of FIB via shedding from their bodies and
into the water column. For this study, an added significance of the study is that culture and
qPCR counts for enterococci from bathers were generally in the same range. By comparison,
effluent from POTWs employing chlorination may have culture and qPCR counts that differ by
many orders of magnitude (e.g., He and Jiang 2005, Bolster et al. 2005) and qPCR and culture
17
-------
U.S. Environmental Protection Agency
counts in POTW-impacted coastal waters may differ by smaller margins than observed in
chlorinate POTW effluent (Byappanahalli et al. 2010).
A suite of FIB from brackish water was analyzed by Ortega et al. (2009). Eighteen sampling
events were conducted over five sampling trips in the St. Lucie River Estuary, South Florida.
Enterococci were enumerated using MF, whereas E. coli were assayed using the most probable
number (MPN) method. The values for enterococci and E. coli were similar (101 to 102 CFU or
MPN/100 mL) and anR2 value of 0.53 was reported when correlating the two fecal indicators.
Finally, Agudel et al. (2010) used MF and multiplex real time PCR to compare the
concentrations of Bacteroides spp., E. coli, and enterococci in a variety of environmental
matrices in Barcelona, Spain. Seventy-four samples from rivers, wells, urban groundwater, and
wastewater were analyzed for FIB counts using standardized MF methods for E. coli, total
coliforms and enterococci, and qPCR assays for total Bacteroides spp. and enterococci. Based
on the authors' statistical analyses, Figure 6 summarizes the enumeration of the fecal indicators.
The authors concluded that bacterial quantification data were more homogeneous when using
PCR than when using conventional culture microbiology.
n=4S
= 55
n=46
n=66
6-
e-
4-
2-
0-
\
1
P 1
i
,x:::
1
Faecal Total
enterococci conforms
I
,
Culture cfu/100 m(
.-•.;•::::
•••. •••:•
<
^
E.coli
\
—
—
._.
•-•
log copyi'lOOin*
? f ? 71 ?
Faecal entaraeoccl Bactorlodoi
I
r
PCR
copy/100 ni
it
1
Figure 6. Comparison of bacterial data from culture-based and multiplex
real-time PCR methods, and statistical sample description (SOURCE:
Agudel et al. 2010)
2.2.2.2 Water Matrix
Inhibition of the qPCR amplification reaction is an important factor to consider when comparing
performance of qPCR with the Enterococcus MF technique. Duprey et al. (1997) hypothesized
possible seawater inhibition of the qPCR assay through secretion of substances by phytoplankton
and zooplankton, while Haugland et al. (2005) also noted frequent inhibition of qPCR in
undiluted Great Lakes water. However, such inhibition is reduced or eliminated by performing
serial dilutions of the sample (Ahmed et al. 2009).
Characteristics of the water matrix also impact quantification assays. Sinton et al. (2002) showed
that salinity increased bacterial cells rate of decay, which would widen the already demonstrated
18
-------
U.S. Environmental Protection Agency
gap between culture-based and qPCR results in marine waters. Duprey et al. (1997) noted in
their seawater study that DNA persistence was lower during the summer for free and dead cell
DNA, and could extend up to 55 days in the winter for dead cell DNA. The persistences of
naked DNA and a small inoculum of dead cell DNA were the lowest at two days. Walters et al.
(2009) reported that in sewage microcosms exposed to sunlight, culturable Enterococcus
concentration fell below detection limit within 5 days while qPCR signal persisted for at least 28
days. These observations are consistent with the general observation of higher qPCR-based
bacterial densities (enumerated in cell equivalents) than culture-based counterparts in both
marine and freshwaters (Byappanahalli et al. 2010; Bower et al. 2005; Haugland et al. 2005;
Lavender and Kinzelman 2009; Morisson et al. 2008; Whitman et al. 2010)—except for Elmir et
al. (2009) who found comparable densities from both techniques in a marine setting and Noble et
al. (2010) who found underestimation of bacterial density via qPCR for marine settings.
The two studies by Noble et al. (2010) were interpreted by the authors as demonstrating an
underestimation of Enterococcus by qPCR with regards to culture-based methods in marine
water. However, on this point there are several concerns in interpretation that need to be
considered. The primer set used was not specifically described in the study. In the discussion it
was indicated that the primers used were designed for high specificity for E. faecalis and E.
faecium (which would detect a narrower population of Enterococcus) in this study than in studies
that used the EPA method. The use of a single species calibrator standard could also result in
disagreement between the techniques, as different species of enterococci may have different
numbers of the target gene. Finally, the authors pointed that the relative quantification method
with a calibrator and a salmon testes DNA control used by other teams expressing results in
CCEs could explain the observed discrepancy between underestimation and overestimation.
They discarded the possibility of error originating from PCR chemistry by showing no
significant difference between the Taqman and Scorpion assays for Enterococcus. However, this
team confirmed the previously published findings of significant correlation between qPCR and
culture-based assays for both Enterococcus and E. coli, with a stronger agreement forE.coli than
for Enterococcus in terms of beach management decisions (88% vs. 94%). With regards to
technique implementation, the authors showed that microbiologists with little qPCR training
produced similar results as their experienced counterparts, while finding DNA extraction
complex and time-consuming. This surprising finding was attributed to a simplified PCR
preparation protocol.
The forgoing findings illustrate the complexities of applying qPCR methods, particularly since
this technique can amplify all DNA present in a sample, regardless of its viability state. This
implies that ambient background DNA (i.e., naked and dead cell DNA; see Lavender and
Kinzelman 2009) needs to be assessed when employing qPCR methods, regardless of the water
matrix. Lavender and Kinzelman (2009) successfully demonstrated the use of a site-specific
corrective factor to correlate culture and qPCR counts and highlighted its importance for the
prediction of beach microbial water quality status.
2.2.2.3 Primary Fecal Pollution Source on Indicator/Method Performance
Fecal pollution sources can include point- and nonpoint sources. Common point sources are
POTW effluents and stormwater outfalls, while nonpoint sources include wildlife, human
shedding and soils and sediments. Although most U.S. POTW effluents are disinfected, some
are not, and in some facilities seasonal disinfection is employed. Disinfection usually yields
19
-------
U.S. Environmental Protection Agency
minimal to null culture-based results for FIB, while qPCR results can still remain high,
accounting for the VBNC and background fractions of enterococci and other FIB. However,
POTW monitoring results typically represent human-derived enterococci, with direct relevance
to public health. Comparatively, indicators in stormwater outfalls can originate from a variety of
sources, in which case qPCR primers are chosen based on an understanding of the origin of the
fecal pollution. Such human/nonhuman speciation would be especially necessary in rural
settings with agriculturally developed watersheds but could be recommended at all recreational
sites due to potential contribution of native Enterococcus to beach advisories (Yamahara et al.
2009). Simultaneous probing with Enterococcus and Bacteriodales primer sets can provide such
speciation (Converse et al. 2009; Elmir et al. 2009; Shanks et al. 2009). The human shedding of
enterococci from bathers was considered minimal in marine waters during initial load or bathing
cycle (about 4%) and proportional to the body's surface area (Elmir et al. 2009). These
researchers noted that additional bathing cycles could increase the predominance of human
shedding of FIB.
2.2.2.4 Treatment and Environmental Factors
Lavender and Kinzelman (2009) and Varma et al. (2009) investigated the effects of wastewater
treatment on culture-based and qPCR results for enterococci. Their similar findings showed that
culture-based numbers are strongly reduced (2 to 5 orders of magnitude), especially by
secondary treatment and disinfection, while qPCR numbers experience smaller reductions or
remain unchanged. As noted previously, this is probably due to the strong impact of disinfection
on culturable cells. Environmental factors (e.g., rain events, UV light) can also significantly
affect qPCR and MF results differently. Lavender and Kinzelman showed that culture-based
Enterococcus densities at stormwater outfalls were significantly different depending on rain
events and location, while no significant differences were observed for qPCR. These researchers
also showed that ambient background DNA appeared insignificant during stormwater events.
Varma et al. (2009) also found that qPCR results were virtually unaffected between normal and
stormflow operations. The contribution of FIB from undeveloped watersheds to reference
beaches studied by Griffith et al. (2010) implies that native sands are at least partially responsible
for some FIB exceedances of water quality thresholds. Furthermore, Walters et al. (2009)
showed that sunlight seems to inhibit the cultivability of cells and accelerate their degradation,
reducing persistence of enterococci by half for culture-based results. Neither dead cell nor naked
DNA (ambient background DNA) appeared to be affected by sunlight. Given the nature of
qPCR and the relative small fraction of viable and culturable FIB, qPCR are not greatly affected
by insolation while culture-based methods experience strong reduction by photoinactivation
during sunlight exposure.
2.2.3 Uncertainty of Indicator/Method Combinations at Low Density
As noted previously, qPCR measurements of FIB exhibit higher variability at low densities
(Haugland et al. 2005). This trend is probably explained by the presence of ambient background
DNA, the abundance of which is site-specific. Lavender and Kinzelman (2009) proposed to
reconcile qPCR counts with culture counts by introducing a site-specific corrective factor in the
calculation of qPCR-derived densities. Bae and Wuertz (2009) proposed a more universal
solution by modifying the traditional qPCR approach with the introduction of PMA, allowing
qPCR to distinguish viable from non-viable cells. This approach was found to be promising and
was used successfully by Varma et al. (2009).
20
-------
U.S. Environmental Protection Agency
3 Analyses
This section describes and demonstrates analyses comparing the illness response to exposure
relationship for multiple indicator/method combinations used in epidemiology studies and
evaluating multiple indicator/method combinations to develop quantifiable relationships. The
approach taken to meet the report's objectives is through the use of available data to provide a
proof of concept for the application of various statistical techniques to develop equivalence
between indicator densities for different indicator-method combinations. These initial analyses
illustrate the most direct approaches for linking indicator-method combinations and the steps that
can be undertaken to effect meaningful comparison of results across disparate datasets.
Two approaches for developing quantifiable relationships for "linking" indicator/method
combinations, the Risk Link approach and Water Quality Link approach, are demonstrated in
this section.
The Risk Link approach is evaluated first using three different statistical linkage analyses. The
first is a straight-forward linkage of Enterococcus density as measured by qPCR to Bacteroidales
density as measured by qPCR. The second demonstration shows analyses that can be conducted
prior to linkage of indicators using curves from different studies. That demonstration links
curves (Marion et al. 2010 and USEPA 1984) based on the same indicator-method combination
(E. coli by MF), but from epidemiological studies conducted in different settings, with different
samples sizes, and at different times. The third demonstration illustrates the linkage of
Enterococcus enumerated by MF to Enterococcus enumerated by qPCR, both from studies
conduted in POTW-impacted marine waters. That demonstration also illustrates the translation
of GI definitions between studies, prior to conducting the Risk Link analysis. Comparisons
beyond the three Risk Links demonstrated here are possible, and additional linkages may be
conducted as alternative statistical analyses or datasets become available.
Secondly, the Water Quality Linkage approach is evaluated using two statistical models, simple
linear regression and broken stick (segmented) regression. Using paired water quality data from
the NEEAR freshwater dataset, the culture-qPCR relationships appear to vary significantly
among beaches. The linkage results in translated culture Enterococcus criteria values generally
significantly higher than the existing culture criteria for those specific beaches. Other site-
specific models, or models that incorporate data beyond paired indicator-method data, may be
developed to overcome the limitations of the evaluated regression models in the future.
3.1 Risk Link
Several Risk Link demonstrations are provided in this section. The simplest and perhaps
strongest application of the Risk Link approach is the linkage of densities of indicators for
multiple indicator-method combinations via health effects curves generated in the same
epidemiology study. In this comparison, illness definitions, sampling strategies, statistical
interpretations and fecal pollution sources are the same, and no conversions or intermediate data
analyses are necessary. However, Risk Linkages can also be made using health effects curves
from different epidemiology studies. In that event, analyses must be conducted prior to the
linkage of indicator densities, resulting in linkages that are less direct than those established
using multiple health effects curves from a single epidemiology study.
21
-------
U.S. Environmental Protection Agency
The Risk Linkages demonstrated in this section illustrate both the direct application of the Risk
Link using health effects relationships from the same study, and the less direct Risk Link, that
requires preliminarily conversions or analyses. The first demonstration illustrates the linkage of
Bacteroidales as measured by qPCR and Enterococcus as measured by qPCR. Health effects
curves for both of those indicator-method combinations were developed in NEEAR study marine
beach epidemiology studies in which both indicator densities were measured concurrently and
associated with the same illness levels. The second demonstration illustrates techniques for
assessing the differences in health effects associations with indicator densities for a common
indicator-method combination (E. coli by MF) for epidemiology studies with similar designs, but
conducted at different times and in different settings. The two studies from which health effects
curves were drawn are a recent study of illness-indicator density associations at a small inland
lake (Marion et al. 2010), and the studies conducted as a prelude to establishment of the current
RWQC (USEPA 1984). The third demonstration illustrates the harmonization of health effects
curves, from two epidemiology studies using different illness definitions, prior to the application
of the Risk Link approach. In that study, qPCR-based Enterococcus health effects curves from
the NEEAR study marine beaches are linked to health effects curves based on Enterococcus as
measured by MF in the USEPA (1983) epidemiology studies. Implicit in the third demonstration
is an assumption that the conditions at the beaches studied in the two epidemiology studies were
sufficiently similar to allow meaningful comparison of their health effects curves.
3.1.1 Review of NEEAR Study and Other Health Effects Relationships for qPCR and
Culturable FIB
The NEEAR studies provide the best current association between water quality measures and
illness rates (i.e., health effects curves), as they are the most recent, largest, and most carefully
designed epidemiology studies conducted by EPA. In this section, health effects relations from
the NEEAR studies are presented. Those relationships may be used in development of new
qPCR-based criteria. Note that the results presented in this report are only a small portion of the
findings of the NEEAR studies. More detailed results and full descriptions of the studies can be
found in Wade et al. (2006, 2008, 2010).
To date, EPA's NEEAR recreational water epidemiology studies have employed both qPCR and
culture-based methods in determining FIB density. Further, the studies have yielded health
effects relationships between Enterococcus density as measured by qPCR, Bacteroidales density
as measured by qPCR, and health effects (GI illness) for both marine and Great Lakes POTW-
impacted sites. The development of associations of illnesses with those indicator-method
combinations is particularly significant because qPCR is a rapid method. Health effects curves
generated from observed indicator densities in the NEEAR study marine beaches and for the
USEPA (1983) study are presented in Table 5. Candidate NEEAR study health effects relations
for marine beaches were provided in a personal communication from T. Wade (2010).
22
-------
U.S. Environmental Protection Agency
Table 5. Summary of health effects relationships from USEPA NEEAR studies and marine studies
conducted to support development of the current RWQC
Setting
Great Lakes
Marine
beaches
Marine
beaches
Marine
beaches
Marine
beaches
Health effects relationship
N/l 000 = 0.0214 log10(C£Arr)- 0.00918
7V/1000 = 0.05171og10(C£Arr)- 0.101
TV/1000 = 0.00406 log10(CBArr)+ 0.0144
N/l 000 = 0.03681og10(CBj4C)- 0.0962
N/l 000 = 0.0219 loglcl(CBArr)- 0.01485
Indicator and method
Enterococcus
measured by qPCR
(CE/100ml_)
Enterococcus
measured by qPCR
(CCE/1 00 ml)
Enterococcus
measured by MF
(CFU/100ml_)
Bacteroidales
measured by qPCR
(CCE/1 00 ml)
Enterococcus
measured by MF
(CFU/100ml_)
Reference
Wade et al.
(2006)
T. Wade,
personal
communication,
unpublished
data, 2010
T. Wade,
personal
communication,
unpublished
data, 2010
T. Wade,
personal
communication,
unpublished
data, 2010
USEPA (1983)
In the EPA study of marine beaches impacted by POTW effluent, Wade et al. (T. Wade, personal
communication, unpublished data, 2010) found that swimmers experienced more GI illness than
non-swimmers on days when Enterococcus density (as measured by MF) exceeded the current
geometric mean guidelines. However, the association among swimmers was not statistically
significant. The authors contrast this lack of association with those for Enterococcus and
Bacteroidales enumerated by qPCR, both of which were associated with a consistent trend of
excess GI illness among swimmers. Note that a weak association of Enterococcus density as
measured by MF with health effects (GI illness) was also observed in EPA's Great Lakes
epidemiology studies (Wade et al. 2006, 2008).
For a rate of swimming-associated GI illness of 35 per 1000 swimmers (risk levels similar to
those underlying the USEPA1986 criteria) the Great Lakes health effects curve indicates an
indicator density of 116 qPCR Enterococcus CCE/100 mL. For a swimming-associated GI
illness rate of 35 per 1000 swimmers, the marine health effects curve based on qPCR
enumeration of enterococci yields a corresponding indicator density of 427 qPCR Enterococcus
CCE/100 mL. The culture-based NEEAR study health effects curve for marine waters produces
a high FIB density corresponding to a swimming-associated illness rate of 35 per 1000
swimmers.
23
-------
U.S. Environmental Protection Agency
3.1.2 Direct Application of the Risk Link Approach: Linking Bacteroidales Measured by
qPCR to Enterococcus Measured by qPCR
Application of the Risk Link approach to determine densities of Enterococcus measured by
qPCR and Bacteroidales measured by qPCR that are consistent with comparable risks is
straightforward. The Enterococcus and Bacteroidales health effects curves presented in Table 5
were generated using water quality and attributable illness rate data collected concurrently at the
same beaches and from the same cohorts. Thus the health effects curves (Table 5) may be used
without prior analyses to establish risk-linked indicator densities. Risk-linked Enterococcus and
Bacteroidales densities for marine waters at three risk levels of interest are presented in Table 6
Table 6. Risk-linked Bacteroidales and Enterococcus densities
Tolerable attributable
illness level (as NGIf
per 1000 swimmers)
35
43
82
Enterococcus density,
qPCR enumeration
(CCE/100ml_)
427
610
3460
Bacteroidales density,
qPCR enumeration
(CCE/100ml_)
3680
6060
69,600
t
NEEAR study definition of Gl illness
3.1.3 Linking E. coli Densities as Measured by MF via Health Effects Curves from Different
Epidemiology Studies
In the second application of the Risk Link approach, health effects curves based on densities of
E. coli measured by MF are used to demonstrate the viability of comparing results from studies
conducted at different times and in different settings. The analyses performed in this
demonstration are intended to illustrate techniques for assessing whether epidemiology study
findings differ between studies and provide insights regarding the combinability of data from
multiple studies. The demonstration relates to the PI5 objectives in that it illustrates methods for
comparing the GI illness response to exposure relationships, though in this case the relationships
are for the same indicator-method combination. Note that the analyses presented are not the only
ones that could be performed. Other meta-analyses that have explored the combination of health
effects data from disparate epidemiology studies are summarized in a text box concluding this
section.
A recent epidemiology study conducted by Marion et al. (2010) resulted in a health effects curve
relating E. coli density via MF to both HCGI and GI illness. Limitations of this study's use in
developing general, quantifiable relations are that the study was relatively small (minimum
number of swimmers on a study day was 11 and the maximum was 88); the study was conducted
for an inland waterbody (impounded reservoir); the dominant fecal pollution source was not
clearly POTW discharge; and E. coli by MF was the sole indicator/method investigated. Despite
these limitations, the analyses seek to evaluate whether the health effects curves generated in the
Marion study are comparable to those generated in the epidemiology studies conducted by
USEPA (1984) on freshwaters in support of the 1986 criteria. Although not conclusive, the
comparison indicates that for the beaches studied by USEPA (1984) and Marion et al. (2010),
FIB densities are associated with similar risks.
24
-------
U.S. Environmental Protection Agency
The Marion et al. (2010) and USEPA (1984) studies differ in the following respects:
• the decade in which the study was conducted,
• the manner in which water quality and health effects are characterized, and
• the settings at which data were collected.
Plots of health effects vs. indicator densities for the two studies are presented in Figure 7. Data
from the USEPA studies are blue diamonds and the regression line through the data is blue. The
data from Marion and colleagues are presented in two different ways. In that study, illness rates
are provided for ranges of FIB densities. Those data are plotted as purple crosses. Data for each
study day were provided by the author in a personal correspondence (Marion, personal
communication, 2010) and those data and associated regression line are shown as red circles and
a red line. Individual study day data from the Marion et al. (2010) study show wide scatter,
which is not surprising given the overall low rates of attributable illness and the relatively small
number of swimmers on each study day. Overall, the ranges of attributable illness rates for the
two sets of studies are similar (based on the binned data from the Marion study) and the slopes of
the health effects curves also appear similar.
o
o
o
T— CD
-
0)
• USEPA 1984 study data
x Averaged Marion Data
* Marion study data, individual study days
1e-01 1e+00 1e+01 1e+02
E. coli(CFU/100mL)
1e+03
Figure 7. Health Effects Data and Trend Lines from the USEPA and
Marion Studies
To test whether the data from the two studies indicated the same or different trends, an analysis
of covariance (ANCOVA) for the individual study day and USEPA datasets was performed. The
ANCOVA tests the hypothesis that the slope and intercept estimates for regression lines are the
25
-------
U.S. Environmental Protection Agency
same, accounting for covariation of the slopes and intercepts. The results of the ANCOVA are
presented in Table 7. In ANCOVA analyses, the p-value is the probability that the parameter is
the same for both models and a p-value above a value of 0.05 indicates that the null hypothesis is
accepted. For the two epidemiological datasets the p-values for both the slopes and intercept are
significantly above 0.05. Thus, it is concluded that the health effects curves are not different
from each other and that the data may be pooled.
Table 7. Results from ANCOVA comparing data from the
USEPA (1984) and Marion et al. (2010) epidemiology studies
Parameter
Slope
Intercept
P-value for comparison between models
0.85
0.10
The finding that the health effects models for the USEPA and Marion epidemiology studies are
not significantly different has two ramifications. First, the study by Marion et al. (2010) was
conducted at an inland lake (impounded reservoir) with mixed fecal pollution sources, one of
which was discharge from small POTWs into tributaries to the lake. The studies conducted by
USEPA (1984) took place on Lake Erie and at a freshwater inland lake (Keystone Lake,
Oklahoma). While not conclusive, the ability to pool health effects data from these sites
indicates that health effects at POTW-impacted inland waters are similar to those for POTW-
impacted coastal waters. Second, the ANCOVA supports health effects relations for the two
studies being similar despite the long time period between the studies. This finding cannot,
however, be used to assert that water quality at all sites has not changed in the intervening years
between the studies. However, the similarity in the health effects curve for the Marion et al.
(2010) and USEPA (1984) studies may indicate that the earlier relationship is indicative of
indicator-health effects relationships that may be observed for some types of current recreational
sites.
Again, it is noted that the comparison of the USEPA (1984) and Marion et al. (2010) studies is
conducted to demonstrate techniques for establishing Risk Links. In this case, the analyses are
conducted to establish that indicator densities are associated with similar illness levels in
different studies. Analyses such as these may be required to establish linkages in the event
straightforward linkages with data from a single study or set of studies are not possible.
Finally, we note that meta-analyses of epidemiology studies have evaluated either implicitly or
explicitly the similarities and differences in observed health effects and in health effects
associations with indicator densities for different beaches and times. The objectives of those
meta-analyses are in some respects similar to the analyses conducted to establish that the USEPA
(1983) health effects curves are generally applicable today. Brief descriptions of other meta-
analyses that have been performed are provided in the text box below.
26
-------
U.S. Environmental Protection Agency
Prtiss (1998) conducted a systematic review following discussions between the WHO Regional Office
for Europe and WHO Headquarters to initiate development of new guidelines for recreational use of
the water environment. The comprehensive review of 22 published studies on sewage pollution of
recreational water and health outcomes concluded that there was a causal association between Gl
illness symptoms and increased bacterial indicator density (i.e., enterococci for marine, enterococci
and E. coli for fresh) in recreational waters.
A meta-analysis of 18 published studies (Zmirou et al. 2003) provided a scientific basis for
establishing new standards for the microbial quality of marine and fresh recreational waters to replace
the 30 year-old European Union bathing water quality guidelines. The researchers provided four
major results: (1) increased concentrations of fecal conforms or E. coli and enterococci in both fresh
and marine recreational waters are associated with increased risks of acute Gl illness, with
enterococci eliciting four to eight times greater excess risks than fecal conforms or E. coli at the same
concentrations; (2) Gl illness risks associated with enterococci occur at lower concentration in marine
versus fresh recreational waters; (3) increased concentrations of total conforms have little or no
association with Gl illness risk; and (4) no evidence exists of a threshold of indicator density below
which there would be no Gl illness risk to bathers.
Wade et al. (2003) conducted a systematic review and meta-analysis of 27 published studies to
evaluate the evidence linking specific microbial indicators of recreational water quality to specific
health outcomes under non-outbreak (endemic) conditions. Secondary goals included identifying and
describing critical study design issues and evaluating the potential for health effects at or below the
current regulatory criteria (USEPA, 1986). The researchers concluded that (1) enterococci and to a
lesser extent £. coli are adequate indicators (predictors) of Gl illness in marine recreational waters,
but fecal conforms are not; (2) the risk of Gl illness is considerably lower in studies with enterococci
and £. coli densities below those established by EPA (1986), thus providing support for their
regulatory use; (3) £. coli is a more reliable and consistent predictor of Gl illness than enterococci or
other indicators in fresh recreational waters; and (4) based on heterogeneity analyses, studies that
used a non-swimming control group and that focused on children found elevated Gl illness risks.
3.1.4 Application of the Risk Link Approach for Health Effects Curves Based on Different
Illness Definitions
Two steps are required to demonstrate the Risk Linkage between Enterococcus measured by MF
in studies of POTW-impacted marine sites conducted by USEPA (1983) and Enterococcus
measured by qPCR in NEEAR marine water studies:
• illness rates from the USEPA (1983) studies are translated to equivalent rates reflecting
different illness definition used in NEEAR studies, and
• demonstration that the association of indicator densities and health effects observed in the
two studies are similar.
Direct comparison of health risks associated with water quality published in 1986 and the health,
water quality relationships developed by Wade (2009 to 2010) cannot be done because the case
definitions underlying the relationships are not the same. Wymer (L. Wymer, personal
communication, 2010) suggested a translation algorithm for converting the health criteria
developed in 1986, based on a HCGI case definition, to an equivalent health criteria that uses
NEEAR study data based on a NGI case definition. The translation would be independent of
water quality data and uses health data from the 1972 to 1981 and 2002 to 2009 EPA studies.
Conversion of the 1986 criteria to potential new criteria requires only the illness rates (using
their respective case definitions) from the non-swimming populations from both study periods
27
-------
U.S. Environmental Protection Agency
and the 1986 criteria value. The algorithm uses the HCGI illness rate (IK) in non-swimmers and
the acceptable HCGI illness rate from the criterion, and calculates a relative risk (RR) value as:
^1986 -
Criteria acceptable IR19S6 + Non - swimmer IR
T986
Non - swimmer IR
[2]
1986
An equivalent criterion value (ECV) can be calculated by multiplying the relative risk from the
1986 data by the 2002-2009 non-swimmer illness rate (NGI) and subtracting the 2002-2009 non-
swimmer illness rate (NGI) from that value.
ECV = (RR19S6 x Non - swimmer IR2009)- Non - swimmer IR2
:009
[3]
Non-swimmer illness rates observed in the epidemiology studies conducted to support the!986
criteria and in the NEEAR studies were 13.6/1000 swimmers and 59/1000 swimmers,
respectively (USEPA 1986). Based on the translation and observed non-swimmer illness rates,
equivalent criteria values for three risk levels of interest are provided in Table 8.
Table 8. Equivalent Criteria in Terms of HCGI and NGI
Definitions of Illness
HCGI criterion (per 1000
swimmers)
8
10
19
ECV (NGI equivalent)
(per 1000 swimmers)
35
43
82
At present, EPA has not selected new RWQC based on qPCR or MF measurement of
Enterococcus. Therefore, this section concludes with illustration of the risk linkage approach for
a range of risk levels and approaches for selecting criteria. Reasonable levels of risk for EPA to
evaluate when selecting new criteria are the tolerable illness rates from the 1986 RWQC for
recreation in freshwaters (8 HCGI attributable illnesses per 1000 swimmers), the tolerable illness
rate for recreation in marine waters (19 HCGI attributable illnesses per 1000 swimmers), or some
intermediate risk level (e.g., 10 HCGI attributable illnesses per 1000 swimmers).
The FIB densities used in the USEPA (1983) health effects relationship are geometric mean
values for indicator densities taken over entire recreation seasons. Recognizing that water
quality varies over the recreation season, current criteria for single sample maximums for
designated beaches are based on the 75th percentile of a log-normal distribution whose geometric
mean corresponds to the indicator density from the USEPA health effects relationship and whose
log-transformed standard deviation is 0.7. New or revised criteria for Enterococcus using MF
methods can be based on the 75th percentile value or other values selected based on improved
characterization of the temporal variability of Enterococcus densities in typical marine waters.
Considering all of these and other factors, example criteria values for qPCR and culture-based
datacorresponding to equivalent risk levels are presented in Table 9. The qPCR example criteria
were calculated using the NEEAR study health effects relationship for all marine POTW-
28
-------
U.S. Environmental Protection Agency
impacted beaches combined. Note that all alternatives to the current culture-based criteria value
for designated beaches (104 CFU/100 mL) are significantly lower than the current RWQC.
Table 9. Examples of equivalent criteria values for marine waters via the statistical
linkages approach
Tolerable
attributable
illness level (as
HCGI per 1000
swimmers)
8
10
19
Tolerable
attributable
illness Level (as
NGI per 1000
swimmers)
35
43
82
Enterococcus
density
measured by
qPCR(CCE/100
mL)
427
610
3460
Geometric mean
Enterococcus
density measured
by MF (CFU/100
mL)
11
14
35
75th percentile
value*
(CFU/100mL)
33
42
104
T The 75th percentile value for Enterococcus density based on the calculated geometric mean
density and assuming a typical standard deviation of log-transformed density for marine sites
of 0.7.
3.2 Water Quality Link Approach
The Water Quality Link approach requires two steps: (1) development of a model relating
indicator density as measured by one indicator-method combination to indicator density as
measured by another and (2) establishment of equivalent values of indicator densities based on
two indicator-method combinations using the resulting model and a health effects curve for one
of the indicator-method combinations. As noted in Section 2, approaches for establishing
models relating indicator-method pairs are in development. Recently published studies have
included models based on simple linear regression (e.g., Noble et al. 2010; Whitman et al. 2010)
and suggested correction factors that account for site- or time-specific conditions influencing the
ratio of indicator counts by different methods (e.g., Lavender and Kinzelman 2009).
This section demonstrates both the modeling component and the linkage component of the Water
Quality Link approach. First, general features of the paired data used in the Water Quality
Linkage are presented. Next, development of the statistical models (functional form of the
relationship between the densities by the different methods and data used to determine the model
parameters) for linking indicator-method density pairs is demonstrated. The demonstration
includes techniques for determining which data to use in generating the model and alternative
model forms for NEEAR study freshwater beach data pairs for Enterococcus density as
measured by qPCR and Enterococcus density as measured by MF. ANCOVA (described in
Section 3.1.3) is suggested as a potential means for screening data prior to model development.
Simple linear regression and broken stick (segmented) regression models are fit to the data and
compared. The broken stick regression models were investigated because simple linear
regression models of the NEEAR study data exhibited distributions of residuals that indicate the
simple linear models may not adequately describe the relationship between the indicator method
combinations.
3.2.1 Datasets Used in Water Quality Link Demonstrations
The NEEAR study freshwater data are used in demonstrating the Water Quality Link approach.
Briefly, the NEEAR study water quality data include paired data for Enterococcus as measured
29
-------
U.S. Environmental Protection Agency
by MF, and Enterococcus as measured by qPCR for samples collected at multiple transects at
four beaches and at three sample collection times for each sample day. The four beaches
included Huntington Beach (Lake Erie), Washington Park Beach (Lake Michigan), Silver Beach
(Lake Michigan) and West Beach (Lake Michigan). Samples were collected along multiple
transects on each sample day at 8:00 AM, 11:00 AM, and 3:00 PM and at shin and waist depths.
Other water quality and environmental data were collected during the NEEAR studies, but those
data were not used in the current demonstration.
3.2.2 Linear Models of Log-Transformed Indicator Data
The most critical component of the Water Quality Linkage approach is establishing a model
relating the densities of paired water quality samples for different indicator/method
combinations. Several researchers have proposed linear models for relating log-transformed
Enterococcus qPCR densities and culture counts (e.g., Haugland et al. 2005; Noble et al. 2010;
Whitman et al. 2010;). This section describes a process by which a linear model of log-
transformed data may be developed. Model development includes regression of data to find
model parameters (slope and intercept). Additionally, the data undergo a selection process by
which similar data are included, and dissimilar data are excluded, in the analysis.
To introduce the Water Quality Link approach and modeling of paired indicator density data, a
plot of paired data (Enterococcus density as measured by MF and Enterococcus density as
measured by qPCR, Figure 8) is presented and described. The data and fits shown in this
example provide background for development of the linear regression model and orient the
reader to general features of the relationship between densities of Enterococcus measured by MF
and Enterococcus measured by qPCR. The data shown in Figure 8 were collected and analyzed
by Ferretti et al. (2008) in a study of multiple beaches along the New Jersey shore. They show a
general trend toward a 1:1 correspondence between cell equivalents (CE) and CPUs at high
indicator densities, with significantly greater scatter, and possibly a different trend, at low
indicator densities. These trends are both expected and, as described in Section 2, are likely due
to the following:
• the association of each live, culturable cell with genetic material measured via qPCR;
• high uncertainty in qPCR at low densities due to analysis of very small samples volumes
(in terms of original sample volume);
• divergence in the culture and qPCR signals as organisms age; and
• the difference in the culture to qPCR ratio arising from uneven loading of culture and
qPCR targets to recreational sites.
In Figure 8, paired Enterococcus densities for two New Jersey beaches, in Monmouth County
(red circles) and Ocean County (blue circles), are illustrated. A dashed green regression line
shows the best fit of a linear model to the raw data (not log-transformed) while a dashed red
regression line shows a linear regression fit to the log-transformed data. The black 45° line is a
hypothetical line showing perfect correspondence between Enterococcus densities when
measured via the two methods. The paired data for these two beaches are similar to those
reported for other beaches—high variability and uncertainty in the qPCR counts at low density,
with a correlation approaching a 1:1 ratio at higher indicator density. In this example, it is not
obvious whether the data from the two beaches are sufficiently similar to be pooled and used to
30
-------
U.S. Environmental Protection Agency
develop a single regression model for linking the indicator/method combinations. It is also not
clear which water quality model is the best for fitting the data, though both have the potential for
application. These two observations (i.e., uncertainty regarding data pooling and uncertainty
regarding model form) indicate that statistical tests should be used to evaluate the similarity
between datasets prior to their inclusion in the model, and that alternative functional forms
should be evaluated for the model relating Enterococcus density as measured by culture (MF)
methods to Enterococcus density as measured by qPCR.
ID
UD
o
o
O
O
OJ
to
o
o
tU
tO
OJ
to
o o
o
o
o
Monmouth County
Ocean County
Linear regression
Log-transformed data
I
10
I
20
50 100
CPU 100 ml
200
500
Figure 8. Typical plot of paired Enterococcus densities as measured by
qPCR and culture-based methods
ANCOVA, a statistical method for assessing differences in model slopes and intercepts for data
corresponding to different factors (e.g., different beaches or different times of day), can be used
to assess whether trends observed at different locations or times are similar, and to help ensure
that only related data are used in model development. In the event that paired water quality data
are determined to be similar, the data may be pooled and model parameters may be estimated
using the pooled dataset. If models using datasets that are determined to be dissimilar, it is
assumed that there are differences in the systems that generated the data and those differences
are not accounted for in the model relating the two datasets. The use of ANCOVA for
developing datasets is illustrated with the following example.
31
-------
U.S. Environmental Protection Agency
As part of EPA's NEEAR studies conducted in the Great Lakes, Enterococcus densities as
measured by MF and by qPCR were collected (Wade et al. 2006, 2008). Samples were collected
three times per day (8:00 AM, 11:00 AM, and 3:00 PM), at multiple depths (waist, shin, and
knee), and along three transects at each of four beaches studied. A model relating the paired
Enterococcus densities as measured by both qPCR and MF can be developed from all the sites
and times, or from data that are selected because they appear to have similar underlying trends.
All the paired water quality data collected are shown in Figure 9, along with a 45° line showing
perfect agreement. Although one may observe a general trend at high indicator densities, no
trend is apparent at densities below roughly 500 CFU/100 mL. Additionally, there is no obvious
choice for the form (equation relating Enterococcus density as measured by qPCR and as
measured by membrane filtration) of a model relating the data from the two analytical methods.
Figure 10 shows paired water quality data from Huntington Beach only, one of the four Great
Lakes beaches studied. Again, the FIB densities appear linearly related at high densities and
scattered at low densities, though the scatter is considerably less than that observed when data for
all the beaches are plotted together.
a
o
10
100 1000
CFU/100mL
10000
Figure 9. Paired qPCR and culture Enterococcus data, all
NEEAR study Great Lakes beaches
32
-------
U.S. Environmental Protection Agency
ffi
o
9 -I
10
100 1000
CFU/100mL
1 ,;„;„;„;,
Figure 10. Paired culture and qPCR Enterococcus data, all
samples from Huntington Beach
Figure 11 and Figure 12 show data for only 8:00 AM and 3:00 PM, respectively, along with 45
lines and regression fits developed via linear regression of the log-transformed densities. Note
that the trend lines for the 8:00 AM and 3:00 PM samples are markedly different and that
segregation of data by time of the day markedly reduces the scatter in the plots.
LU
s
-------
U.S. Environmental Protection Agency
LO
O
CO
o
o
o
LU
O
o data
— regression fit
45degline
10
100
CPU /100mL
1000
10000
Figure 12. Paired qPCR and culture Enterococcus data, Huntington
Beach: 3:00 PM samples only
ANCOVAs can be used to assess whether the differences in the regression lines observed for the
8:00 AM and 3:00 PM data are statistically significant. ANCOVAs were run to compare the
regression lines for sets of data corresponding to each of the collection times at Huntington
Beach. As shown in Table 10, the ANCOVA indicates that there are significant differences in
the models for the 8:00 AM and 11:00 AM dataset, and for the 8:00 AM and 3:00 PM datasets,
but not for the 11:00 AM and 3:00 PM datasets. This finding is not surprising given that the 8:00
AM and 3:00 PM samples do not correspond to the same model, and because solar radiation is
known to reduce culture counts for afternoon samples while qPCR counts remain relatively
constant throughout the day.
Table 10. Results of ANCOVAs for sets of data for
Huntington Beach with different collection times
Collection times
8:00 AM and
11:00 AM
8:00 AM and 3:00
PM
11:00 AM and
3:00 PM
P-value
Intercepts
<0.0001
<0.0001
0.52
Slopes
0.002
0.01
0.60
Additional ANCOVAs were performed to compare models for all of the NEEAR freshwater
study datasets and to determine whether sample water depth, time of day, and beach location
altered the model relating qPCR and culture-based datasets. Findings from the ANCOVAs are
as follows:
• Samples corresponding to different sample collection depths can be pooled.
34
-------
U.S. Environmental Protection Agency
• Models for different times of day differ for two beaches (Huntington and West Beach)
but do not differ for the other two beaches (Silver Beach and Washington Park Beach).
Agreement between models for different times of day for Silver Beach and Washington
Park Beach might be the result of extreme scatter in the data for those two beaches.
Based on these findings and knowledge of the variation in culture and qPCR signals
during the day, 8:00 AM Enterococcus measured by MF are fit by different models than
samples from other times of day..
• For these data, models for different beaches are dissimilar and pooling of the data may be
inconsistent with simple linear models.
3.2.3 Demonstration of the Water Quality Linkage Approach using Linear Regression Models
In this section, the Water Quality Link approach is demonstrated using NEEAR study Great
Lakes Enterococcus density data (as measured by both MF and qPCR), the NEEAR study
freshwater health effects curve (presented in equation form in Table 5), and linear models
relating log-transformed data of Enterococcus density as measured by qPCR and MF. As noted
previously, these linear models may not adequately characterize the correlation between qPCR
and culture Enterococcus density for some datasets. Therefore, the illustration presented below
may be used for developing an understanding of the Water Quality Linkage approach but not for
developing specific equivalences between densities of Enterococcus as measured by MF and as
measured by qPCR. The Water Quality Link demonstration entails
1. selection of an Enterococcus density based on the NEEAR study freshwater beaches
health effects curve;
2. use of linear regression to develop several candidate models relating density of
Enterococcus as measured by qPCR to Enterococcus density as measured by MF; and
3. use of the linear relationships developed in step 2 to establish candidate Enterococcus
densities as measured by MF to the Enterococcus density selected in step 1.
After the Water Quality Linkage approach is demonstrated the results are critically evaluated and
used to suggest alternative Water Quality Linkage approaches.
The NEEAR study health effects curve for Great Lakes beaches (see Table 5) indicates that an
Enterococcus density of about 125 qPCR CE/100 mL relates to a risk of about 45 NGI illnesses
in 1000 swimmers. The rate 45 NGI illnesses in 1000 is approximately equivalent to a risk of 8
HCGI illnesses in 1000 swimmers—the level at which current (1986) freshwater criteria are
based (see Section 3.1.4). Thus for this demonstration, the qPCR level of interest is selected to
be 125 qPCR CE. This selection does not imply that new or revised criteria will be based on the
Great Lakes health effects curve or on a risk level of 45 NGI illnesses in 1000 swimmers. It is
selected for demonstration purposes only and because it is based on risk in a range consistent
with current (USEPA 1986) criteria.
The choice of a model relating the Enterococcus density data from qPCR to density data from
MF has a profound influence on the outcome of a Water Quality Linkage analysis. This is
illustrated using the NEEAR study data. When data from all the NEEAR study freshwater
-------
U.S. Environmental Protection Agency
beaches are used to develop a model relating Enterococcus qPCR and culture densities (the
purple line in Figure 13), the regression line has a slope <1 and falls significantly below the 45°
line, for the culture-based density above about 100 CFU/100 mL. The model mean value for
CPU equivalent to CE for 45 illnesses/1000 swimmers is 1212 CFU/100 mL and the confidence
interval for CFU equivalent to CE for 45 illnesses/1000 swimmers is <519 CFU/100 mL, 4628
CFU/100mL>. Thus, these Enterococcus densities, measured by MF, are clearly very high and
inconsistent with the selected level of tolerable risk.
o
o
LU
o
0 Data
Regression line
Cl
45degline
Health risk level
i i i i i
1 10 100 1000 10000
CFU/100mL
Figure 13. Linear model, data from all NEEAR freshwater beaches
When only Huntington Beach data are used, the results appear to be more reasonable. The
model line (the purple line in Figure 14) still has a slope <1, which is inconsistent with expected
trends at high indicator density, but the slope is steeper than that of the model for data from all
beaches. The model mean value for CFU equivalent to CE for 45 illnesses/1000 swimmers is
10.8 CFU/100 mL, and the confidence interval around that value is <7.1 CFU/lOOmL, 15.4
CFU/100mL>. These equivalent values appear unrealistically low, though more reasonable than
the equivalent values derived using the model for data from all beaches.
36
-------
U.S. Environmental Protection Agency
O <">
o
LU
O
9 4
° Data
Regression line
Cl
45 deg line
Health risk level
1
10
100 1000 10000
CFU/100mL
Figure 14. Linear model, data from Huntington Beach
The most reasonable equivalent culture densities result from a model developed using only 8:00
AM data from Huntington Beach. The model, equivalent value, and confidence interval are all
shown in Figure 15. The model mean value for CPU equivalent to CE for 45 illnesses/1000
swimmers is 29.4 CFU/100 mL and the confidence interval around that value is <17.8
CFU/lOOmL, 45.9 CFU/100mL>.
37
-------
U.S. Environmental Protection Agency
o
o
O
O 0
o Data
Regression line
Cl
45degline
Health risk level
10
100 1000
CFUflOOmL
10000
Figure 15. Linear model, 8:00 AM data from Huntington Beach
Based on the analysis of paired water quality NEEAR study Enterococcus density data
measured by qPCR and MF and using simple linear model of log-transformed variables, several
observations may be made, though the extent to which these may apply to datasets other than the
ones studied is unknown. First, model selection (inclusive of data selection) profoundly impacts
the Water Quality Linkage approach. For this dataset, when all data are pooled and a water
quality model is generated, the resulting densities of Enterococcus as measured by MF that are
equivalent to the density of Enterococcus as measured by qPCR appear very high in comparison
to the current criteria. When only data from morning samples at a single beach are used to
develop the linear model, the resulting equivalent Enterococcus densities for MF are closer to the
current criteria. The non-robust performance of linear models of log-transformed indicator
densities in this demonstration suggests exploration of alternative models that may be capable of
modeling more complex phenomena and providing an adequate fit to a larger set of data.
3.2.4 Broken Stick Models of Log-Transformed Indicator Data
The relationship between paired Enterococcus densities as measured by MF and Enterococcus
densities as measured by qPCR data might not be fit best by a simple linear regression model of
log-transformed data. An alternative model evaluated in the development of the Water Quality
Linkage approach is the broken stick regression model, also known as the segmented model
(Draper and Smith 1998). In broken stick regression models, the data are divided in two parts,
each of which is fit with a different linear model. The two linear models meet at a "kink" in the
regression line. Broken stick regression models have four parameters (two slopes and the
coordinates of the point where the segments of the broken stick meet) that may be determined via
38
-------
U.S. Environmental Protection Agency
optimization, whereas simple linear regression models have two parameters (slope and
intercept). The broken stick model appears a good choice for analyzing paired densities of
Enterococcus as measured by MF and Enterococcus as measured by qPCR because the data
appear to behave differently at low and high indicator densities.
Broken stick regression models were developed for each of the four NEEAR Great Lakes study
beaches separately. Two models were developed for each beach—one for all data for that beach
pooled and one for only data from samples collected at 8:00 AM. The resulting curves are
shown in Figure 16 to Figure 19.
E
o
o
5
O
E
o
o
5
o
100 1000
CFUI100 ml
10000
10
100 1000
CFU/100ml_
10000
Figure 16. Huntington Beach broken stick model fits for all data and 8:00 AM data
39
-------
U.S. Environmental Protection Agency
E
o
o ^_
^r o
8 I
i-
Break CPU =
100
I-
E
o T-
2 V -
^-i &
UJ •-
o
I-
10
100
CFU/100ml_
1000
10000
10
100
CFUMOO mL
1000
10000
Figure 17. West Beach broken stick model fits for all data and 8:00 AM data
o
South Beach, all samples
E ?
a
o
South Beach, 8 AM samples
Break CPU -
74
10 100 1000
CFU/100 mL
10 100 1000
CFU/100mL
Figure 18. Silver Beach broken stick model fits for all data and 8:00 AM data
10000
40
-------
U.S. Environmental Protection Agency
E 8
8 I
5
o
Washington Park Beach
All samples
I
S
o
Washington Park Beach
8 AM samples
10 20 50
CFU/100 ml
100 200
500
10 20 50
CFUHOO mL
100 200
500
Figure 19. Washington Park Beach broken stick model fits for all data and 8:00 AM data
Broken stick regression models of Huntington and West Beach data both exhibit slopes closer to
1.0 in the high FIB density portion of the broken stick, compared to the low density portion. For
the Washington Park and Silver Beach datasets, the presence of water quality data pairs with
relatively high Enterococcus culture density and very low Enterococcus qPCR density appear to
exert an undue influence on the slope portion of the broken stick in the high culture density data
range. For the four beaches, the location of the "kink" in the stick varies within the range 12-
100 CFU/100 mL for models of data including all sample times, and in the range 28-62
CFU/100 mL for the 8:00 AM samples of the Huntington and West Beach data.
Given the empirical and theoretical support for different correlation of qPCR and culture counts
at high and low densities, the broken stick model offers advantages over models of simple linear
regression of log-transformed densities, including
• better fits to data,
• reduction in heteroskedasticity of regression fits, and
• consistency with observed relationships between qPCR and culture densities.
Regression analysis of correlations between the qPCR and culture densities, as well as the
literature, support different relationships between the two water quality measures at low and high
indicator counts. At low indicator densities, the relationship of the culture indicator with the
fecal pollution source becomes tenuous and the qPCR counts become highly uncertain. At
higher densities, the abundance in targets for the two methods is correlated. These findings both
support and suggest performing regression analysis only for those data pairs for which culture
densities exceed a threshold value.
Three regression analyses were performed to evaluate whether using only data above such
threshold values yielded robust models for estimating culture densities that are equivalent to
qPCR densities. In all of the analyses, only FIB density data collected in the morning (8:00 AM
41
-------
U.S. Environmental Protection Agency
sample time) were included, due to potential discrepancies between enumeration methods at later
times. In the first analysis, only points above the "kink" in the stick were retained for analysis
for each of the four Great Lakes beaches at which EPA's NEEAR epidemiology studies were
conducted. The data retained for each beach were determined by the "kink" location for each of
the four beaches. In the second and third analyses, a threshold common to all beaches was
chosen as the basis for retaining data. Two thresholds were explored—80 CFU/lOOmL and 104
CFU/100 mL. The value 104 CFU/100 mL was selected based on the current single sample
maximum for designated beaches; 80 CFU/100 mL was selected because it lies between the
current criteria and the geometric mean on which it is based (35 CFU/100 mL).
Plots showing the resulting three regression models and the culture equivalent to a qPCR density
of 126 CE/100 mL are shown in Figure 20 (data above the kink in the stick), Figure 21 (data
above 80 CFU/100 mL), and Figure 22 (data above 104 CFU/100 mL). The qPCR density
chosen for this illustration (126 CE/100 mL) is a plausible value for qPCR-based criteria (Table
9). In all cases, the new regression models and qPCR value of interest result in equivalent
culture criteria exceeding the current single sample maximum value for infrequently-used fresh
waters of 151 CFU enterococci /100 mL. Regression model lines and culture equivalents to a
potential qPCR criterion of 126 CE/100 mL are provided in Table 11.
o
o
_
O
qPCRvalue of interest (CCE/100ml_) 126
Equivalent cufture value (CFU/100mL) 178
5 10 50 100 500
CFU/100mL
SlJIJM
50000
Figure 20. Model resulting from retention of all data above
the "kink" in the broken stick
42
-------
U.S. Environmental Protection Agency
qPCR value of interest (CCE/1 OOmL) 126
Equivalent culture value (CFU/1 OOmL) 192
10
100 1000 10000
CFU/100 ml
Figure 21. Model resulting from retention of all data
above 80 CFU/1 OOmL
qPCR value of interest (CCE/1 OOmL) 126
Equivalent culture value (CFU/1 OOmL) 246
GJ
O
10
100 1000 10000
CFU/100 ml
Figure 22. Model resulting from retention of all data above
104 CFU/1 OOmL
43
-------
U.S. Environmental Protection Agency
Table 11. Summary of models resulting from retention of data pairs with culture counts above
three thresholds
Treatment
Retain data for which CPU count
is above the break in the stick
Retain data for which CPU count
is above 80 CFU/1 00 ml
Retain data for which CPU count
is above 126 CFU/1 00 ml
Regression line
loglo(CqpCR)= 0.393 + 0.759 log10(CCFU)
log10(CqPCR)= -0.056 + 0.944 loglo(CCFU)
loglo(CqpCR)=-1.289 + 1.3581oglo(CCFU)
Equivalent culture
density of interest
(CFU/1 00 mL)
178
192
246
Several observations can be made regarding development of a regression model based on data
only above thresholds. First, the model results appear quite sensitive to the selection of the break
point above which data are retained. In this regard, there does not appear to be a widely accepted
methodology for selecting the appropriate break point. Both Lavender and Kinzelman (2009)
and Byappanahalli et al. (2010) determined that the qPCR-culture relationship varies with
environmental conditions which, in turn, vary differently at different sites. Second, in all cases,
data pairs with much lower qPCR densities than culture densities appear to have a pronounced
effect on the regression models. At present, we have no explanation for the occurrence of those
data pairs. Because the pairs occur above the cutoff value for culture densities, it is unlikely that
sampling error (i.e., small qPCR sample size in terms of original sample volume) is the cause. In
summary, regression models based on the broken stick or threshold approach were dependent on
a cut-off value, the choice of which could be viewed as arbitrary. Further, culture densities
equivalent to potential qPCR densities were higher than the current single sample maximum for
infrequently-used freshwater beaches when the Water Quality Link approach is applied to the
NEEAR study freshwater indicator dataset, using a broken stick regression model to relate
densities of Enterococcus as measured by MF and qPCR.
44
-------
U.S. Environmental Protection Agency
4 Discussion
Two approaches for linking indicator-method combinations to a common risk level were
demonstrated—the Risk Link approach and the Water Quality Link approach. Using currently
available data and techniques, the exploration of these methods meets both the objective of
developing quantifiable relationships between indicator-method combinations and provides
practical experience in the complexities underlying the two approaches.
Applying the Risk Link approach can be relatively straightforward when multiple indicator
health relationships data are available from a single epidemiology study, as demonstrated in the
Bacteroidales (measured by qPCR) and Enterococcus (measured by qPCR) Risk Link
comparison. However, when directly comparable data are not available, additional steps are
added to the Risk Link approach. These steps may include, but are not limited to, the
harmonization of GI definitions used in two epidemiology studies or the generality of health
effects relations established for sites with specific sources of fecal pollution.
Regarding the Water Quality Link approach, generating a relationship between indicator
densities, as measured by qPCR and culturable methods, is more complex than simple linear or
broken stick regression models. In our demonstration, the relationship, between paired water
quality data corresponding to different indicator-method combinations, appears to vary from
beach to beach. The Water Quality Link may be a useful tool for the development of site-
specific standards by states; but as formulated in this report, is not necessarily useful for the
development of National criteria.
It is possible that additional analyses or datasets might allow for alternative quantifiable
relationships to be explored using both Link approaches. For example, if statistical techniques
are developed that allow for the direct comparison of results between RCT and PC epidemiology
study designs, additional health effects relationships, such as those from the Epibathe studies,
could be evaluated.
45
-------
U.S. Environmental Protection Agency
5 References
Abdelzaher, A., M. Wright, C. Ortega, H. Solo-Gabriele, G. Miller, S. Elmir, X. Newman, P.
Shih, T.D. Bonilla, CJ. Palmer, T. Scott, J. Lukasik, V.J. Harwood, S. McQuaig, C.
Sinigalliano, M. Gidley, L. Piano, X. Zhu, J.D. Wang, and L. Fleming. 2010. Presence of
pathogens and indicator microbes at a non-point source subtropical recreational marine
beach. Applied and Environmental Microbiology 76(3): 724-732.
Agudel, R.M., Codony, F., Adrado, B., Fittipaldi M., Pefiuela, G., and J. Morato. 2010
Monitoring bacterial faecal contamination in waters using multiplex real-time PCR assay
for Bacteroides spp. and faecal enterococci. Water SA, 36(1): 127-132.
Ahmed, W., S. Sawant, F. Huygens, A. Goonetilleke, and T. Gardner. 2009. Prevalence and
occurrence of zoonotic bacterial pathogens in surface waters determined by quantitative
PCR. Water Research 43: 4918-4928.
Bae, S., and S. Wuertz 2009. Discrimination of viable and dead fecal bacteroidales bacteria by
quantitative PCR with propidium monoazide. Applied and Environmental Microbiology
75(9): 2940-2944.
Bolster, C.H., J.M. Bromley, and S.H. Jones. 2005. Recovery of chlorine-exposed Escherichia
coli in estuarine microcosms. Environmental Science and Technology 39(9):3083-3089
Bower, P.A., C.O. Scopel, E.T. Jensen, M.M. Depas, and S.L. McLellan. 2005. Detection of
genetic markers of fecal indicator bacteria in Lake Michigan and determination of their
relationship to Escherichia coli densities using standard microbiological methods.
Applied and Environmental Microbiology 71(12): 8305-8313.
Bremer, H. and Dennis, P. P. 1996. Modulation of chemical composition and other parameters of
the cell growth rate. In Escherichia coli and Salmonella: Cellular and Molecular Biology,
2nd ed., pp. 1553-1568. Edited by F. C. Neidhardt and others. Washington DC:
American Society for Microbiology.
Byappanahalli, M.N., R.L. Whitman, D.A. Shively, and M.B. Nevers. 2010. Linking non-
culturable (qPCR) and culturable enterococci densities with hydrometeorological
conditions. Science of the Total Environment 408: 3096-3101.
Calderon, R.L., E.W. Mood, and A.P. Dufour . 1991. Health effects of swimmers and nonpoint
sources of contaminated water. International of Envionmental Health Research 1: 21-31.
Colford, J.M.J., T.J. Wade, K.C. Schiff, C.C. Wright, J.F. Griffith, S.K. Sandhu, S. Burns, M.
Sobsey, G. Lovelace, and S.B. Weisberg. 2007. Water quality indicators and the risk of
illness at beaches with nonpoint sources of fecal contamination. Epidemiology 18(1):
27-35.
Converse, R., A. Blackwood, M. Kirs, J. Griffith, and R. Noble. 2009. Rapid QPCR-based assay
for fecal Bacteroides spp. as a tool for assessing fecal contamination in recreational
waters. Water Research 43(19): 4828-4837.
Draper, N.R., and Smith, H. 1998. Applied Regression Analysis, 3rd ed. John Wiley and Sons,
NY.
Duprey, E., M.P. Caprais, A. Derrien, and P. Fach, 1997. Salmonella DNA persistence in natural
seawaters using PCR analysis. Journal of Applied Microbiology 82: 507-510.
European Commission (Epibathe consortium). 2009a. Accessible Report to Commission Staff
Responsible for Reporting to the Parliament and to the Council Under Article 14 of the
Bathing Water Directive of 15th February 2006. Assessment of Human Health Effects
46
-------
U.S. Environmental Protection Agency
Caused by Bathing Waters (EpiBathe project). European Union Sixth Framework
Programme, Scientific Support to Policies, FP6 Project Reference 022618.
European Commission (Epibathe consortium). 2009b. Appendix 2 Section 1. Risk of
Gastroenteritis from Recreational Bathing Water: Analysis of Epibathe data from
Hungary and Spain.
Elmir S.M., T. Shibata, H.M. Solo Gabriele, C.D. Sinigalliano, M.L. Gidley, G. Miller, L.R.W.
Piano, J. Kish, K. Withum, and L.E. Fleming. 2009. Quantitative evaluation of
enterococci and Bacteroidales released by adults and toddlers in marine water. Water
Research 43(18): 4610-4616.
Ferretti, J.A., Tran, H.V., Loftin, V.K., Cosgrove, B., and Protonentis, J. 2008. Comparison of
Enterococcus Measurements in Marine Beach and Bay Samples by Quantitative (Real-
Time) Polymerase Chain Reaction, Membrane Filtration, and Enterolert. Edison, NJ: US
EPA Region 2.
Fleisher, J.M., D. Kay, R.L. Salmon, F. Jones, M.D. Wyer, and A.F. Godfree. 1996. Marine
waters contaminated with domestic sewage: nonenteric illnesses associated with bather
exposure in the United Kingdom. American Journal of Public Health 86(9): 1228-1234.
Fleisher, J.M., L.E. Fleming, H.M. Solo-Gabriele, J.K. Kish, C.D. Sinigalliano, L. Piano, S.M.
Elmir, J.D. Wang, K. Withum, T. Shibata, M.L. Gidley, A. Abdelzaher, G. He, C. Ortega,
X. Zhu, M. Wright, J. Hollenbeck, and L.C. Backer. 2010. The BEACHES Study: health
effects and exposures from non-point source microbial contaminants in subtropical
recreational marine waters. International Journal of Epidemiology 39(5): 1291-1298.
Griffith, J., K.C. Schiff, G.S. Lyon, and J.A. Fuhrman. 2010. Microbiological water quality at
non-human influenced reference beaches in southern California during wet weather.
Marine Pollution Bulletin 60(4): 500-508.
Haile, R.W., J.S. Witte, M. Gold, R. Cressey, C. McGee, R.C. Millikan, A. Glasser, N. Harawa,
C. Ervin, P. Harmon, J. Harper, J. Dermand, J. Alamillo, K. Barrett, M. Nides, and G.
Wang. 1999. The health effects of swimming in ocean water contaminated by storm drain
runoff. Epidemiology 10(4): 355-363.
Haugland, R.A., S.C. Siefring, L.J. Wymer, K.P. Brenner, and A.P. Dufour. 2005. Comparison
of Enterococcus measurements in freshwater at two recreational beaches by quantitative
polymerase chain reaction and membrane filter culture analysis. Water Research 39(4):
559-568.
He, J.-W., and S. Jiang. 2005. Quantification of enterococci and human adenoviruses in
environmental samples by real-time PCR. Applied and Environmental Microbiology
71(5): 2250-2255.
Khan, I.U.H., V. Gannon, R. Kent, W. Koning, D.R. Lapen, J. Miller, N. Neumann, R. Phillips,
W. Robertson, E. Topp, E. van Bochove, and T.A. Edge. 2007. Development of a rapid
quantitative PCR assay for direct detection and quantification of culturable and non-
culturable Escherichia coli from agriculture watersheds. Journal of Microbiological
Methods 69(3): 480-488.
Kinzelman, J.L., R.N. Bushon, R.T. Noble, and S.E. Dorevitch. 2010. WERF - PHASE I
Progress Report. Comparative Evaluation of Molecular and Culture Methods for Fecal
Indicator Bacteria for use in Inland Recreational Waters. In: WERF.
Lavender, J.S., and J.L. Kinzelman. 2009. A cross comparison of QPCRto agar-based or defined
substrate methods for determination of Escherichia coli and enterococci in municipal
water quality monitoring programs. Water Research 43: 4967-4979.
47
-------
U.S. Environmental Protection Agency
Lleo, M.M., B. Bonatol, M.C. Tafil, C. Signorettol, C. Pruzzo2 and P. Canepari. 2005.
Molecular vs culture methods for the detection of bacterial faecal indicators in
groundwater for human use Letters in Applied Microbiology 40:289-294
Marion, J.W., J. Lee, S. Lemeshow, and TJ. Buckley. 2010. Association of illness and
recreational water exposure during advisory and non-advisory conditions at an inland
U.S. beach. Water Research 44(16): 4796-4804.
Morrison, C.R., D.S. Bachoon, andK.W. Gates 2008. Quantification of enterococci and
bifidobacteria in Georgia estuaries using conventional and molecular methods. Water
Research 42(14): 4001-4009.
Nocker, A., and A.K. Camper. 2006. Selective removal of DNA from dead cells of mixed
bacterial communities by use of ethidium monoazide. Applied and Environmental
Microbiology 72(3): 1997-2004.
Nocker, A., C.Y. Cheung, and A.K. Camper. 2006. Comparison of propidium monoazide with
ethidium monoazide for differentiation of live vs. dead bacteria by selective removal of
DNA from dead cells. Journal of Microbiological Methods 67: 310-320.
NRC (National Research Council). 2004. Indicators for Waterborne Pathogens. Washington
DC: National Academies Press.
Ortega, C., M. Helena, S. Gabriele, A. Abdelzaher, M. Wright, Y. Deng, and L.M. Stark 2009.
Correlations between microbial indicators, pathogens, and environmental factors in a
subtropical estuary. Marine Pollution Bulletin 58(9): 1374-1381.
Petterson, S.R., R.S. Signer, and NJ. Ashbolt. 2007. Incorporating method recovery
uncertainties in stochastic estimates of raw water protozoan concentrations for QMRA.
Journal of Water and Health 5(Supplement 1): 51-65.
Priiss, A. 1998. Review of epidemiological studies on health effects from exposure to
recreational water. International Journal of Epidemiology 27(1): 1-9.
Rothrock, M. J. J., K. L. Cook and K. H. Bolster 2009. Comparative quantification of
Campylobacter jejuni from environmental samples using traditional and molecular
techniques. Canadian Journal of Microbiology 55(6):633-641.
Siefring, S.C., M. Varma, E. Atikovic, L.J. Wymer, and R.A. Haugland. 2008. Improved real-
time PCR assays for the detection of fecal indicator bacteria in surface waters with
different instrument and reagent systems. Journal of Water and Health 6(2): 225-237.
Shanks, O., C.A. Kelty, M. Sivaganesan, M. Varma, and R. Haugland. 2009. Quantitative PCR
for genetic markers of human fecal pollution. Applied and Environmental Microbiology
75(17): 5507-5513.
Sinigalliano, C.D., J.M. Fleisher, M.L. Gidley, H.M. Solo-Gabriele, T. Shibata, L. Piano, S.M.
Elmir, D. Wanless, J. Bartkowiak, R. Boiteau, K. Withum, A.M. Abdelzaher, G. He, C.
Ortega, X. Zhu, M.E. Wright, J. Kish, J. Hollenbeck, T. Scott, L.C. Backer, and L.E.
Fleming. 2010. Traditional and molecular analyses for fecal indicator bacteria in
nonpoint source subtropical recreational marine waters. Water Research 44: 3763-3772.
Sinton, L.W., C.H. Hall, P.A. Lynch, and R.J. Davies-Colley. 2002. Sunlight inactivation of
fecal indicator bacteria and bacteriophages from waste stabilization pond effluent in fresh
and saline waters. Applied and Environmental Microbiology 68(3): 1122-1131.
USEPA (U.S. Environmental Protection Agency). 1983. Health Effects Criteria for Marine
Recreational Waters. EPA-600/1-80-031. Research Triangle Park, NC: USEPA.
USEPA (U.S. Environmental Protection Agency). 1984. Health Effects Criteria for Fresh
Recreational Water. EPA-600/1-84-004. Research Triangle Park, NC: USEPA.
48
-------
U.S. Environmental Protection Agency
USEPA (U.S. Environmental Protection Agency). 1986. Ambient water quality for bacteria -
1986. EPA440/5-84-002.
USEPA (U.S. Environmental Protection Agency). 2000. Improved Enumeration Methods for the
Recreational Water Quality Indicators: Enterococci and Escherichia coli.
USEPA (U.S. Environmental Protection Agency). 2007. Critical Path Science Plan for the
Development of New or Revised Recreation Water Criteria. Washington DC: Office of
Water and Office of Research and Development, 823-R-08-002
Varma, M., R. Field, M. Stinson, B. Rukovets, L. Wymer, and R. Haughland. 2009. Quantitative
real-time PCR analysis of total and propidium monozide-resistant fecal indicator bacteria
in wastewater. Water Research (early publication online ahead of print).
Wade, T.J., N. Pai, J.N. Eisenberg, and J.M. Colford. 2003. Do U.S. Environmental Protection
Agency water quality guidelines for recreational waters prevent gastrointestinal illness?
A systematic review and meta-analysis. Environmental Health Perspectives 111(8): 1102-
1109.
Wade, T.J., R.L. Calderon, E. Sams, M. Beach, K.P. Brenner, A.H. Williams, and A.P. Dufour.
2006. Rapidly measured indicators of recreational water quality are predictive of
swimming-associated gastrointestinal illness. Environmental Health Perspectives 114(1):
24-28.
Wade, T.J., R.L. Calderon, K.P. Brenner, E. Sams, M. Beach, R. Haugland, L. Wymer, and A.P.
Dufour. 2008. High sensitivity of children to swimming-associated gastrointestinal illness
- results using a rapid assay of recreational water quality. Epidemiology 19(3): 375-383.
Wade, T.J., E. Sams, K.P Brenner, R. Haugland, E. Chern, M. Beach, L. Wymer, C.C. Rankin,
D. Love, Q. Li, R. Noble, and A.P. Dufour. 2010. Rapidly measured indicators or
recreational water quality and swimming-associated illness at marine beaches: a
prospective cohort study. Environmental Health 9: 66.
http://www.ehj ournal. net/contents/9/1/66
Walters, S.P., K.M. Yamahara, and A.B. Boehm. 2009. Persistence of nucleic acid markers of
health-relevant organisms in sewater microcosms. Water Research (early publication
online ahead of print).
WERF (Water Environment Research Federation). 2009. Report on the Experts Scientific
Workshop on Critical Research and Science Needs for the Development of Recreational
Water Quality Criteria for Inland Waters. Dallas/Ft. Worth, Texas, February 18-20,
2009. PATH4WO. Alexandria, VA.
Whitman, R.L., Z. Ge, M.B. Nevers, A.B. Boehm, E.C. Chern, R.A. Haugland, A.M. Lukasik,
M. Molina, K. Przybyla-Kelly, D.A. Shively, E.M. White, R.G. Zepp, and M.N.
Byappanahalli. 2010. Relationship and variation of qPCR and culturable enterococci
estimates in ambient surface waters are predictable. Environmental Science and
Technology 44: 5049-5054.
WHO (World Health Organization). 2003. Guidelines for Safe Recreational Water
Environments. Volume 1 Coastal and Fresh Waters. World Health Organization.
Geneva, Switzerland.
Wiedenmann, A., P. Kruger, K. Dietz, J. Lopez-Pila, R. Szewzyk, and K. Botzenhart. 2006. A
randomized controlled trial assessing infectious disease risks from bathing in fresh
recreational waters in relation to the concentration of Escherichia coli, intestinal
enterococci, Clostridiumperfringens, and somatic coliphages. Environmental Health
Perspectives 114:228-236.
49
-------
U.S. Environmental Protection Agency
Wymer, L., K.P. Brenner, J.W. Martinson, W.R. Stutts, S.A. Schaub, and A.P. Dufour 2005.
The EMPACT Beaches Project: Results from a Study on Microbiological Monitoring in
Recreational Waters. EPA 600/R-04/023 USEPA, Office of Research and Development,
National Exposure Research Laboratory.
Yamahara, K.M., S.P. Walters, and A.B. Boehm. 2009. Growth of enterococci in unaltered,
unseeded beach sands subjected to tidal wetting. Applied and Environmental
Microbiology 75: 1517-1524.
Zmirou, D., L. Pena, M. Ledrans, and A. Letertre. 2003. Risks associated with the
microbiological quality of bodies of fresh and marine water used for recreational
purposes: summary estimates based on published epidemiological studies. Archives of
Environmental Health 58(11): 703-771.
------- |