Enteric pathogen reduction targets for onsite non-potable water systems: A critical evaluation

Abstract

Onsite non-potable water systems (ONWS) collect and treat local source waters for non-potable end uses such as toilet flushing and irrigation. Quantitative microbial risk assessment (QMRA) has been used to set pathogen log₁₀-reduction targets (LRTs) for ONWS to achieve the risk benchmark of 10⁻⁴ infections per person per year (ppy) in a series of two efforts completed in 2017 and 2021. In this work, we compare and synthesize the ONWS LRT efforts to inform the selection of pathogen LRTs. For onsite wastewater, greywater, and stormwater, LRTs for human enteric viruses and parasitic protozoa were within 1.5-log₁₀ units between 2017 and 2021 efforts, despite differences in approaches used to characterize pathogens in these waters. For onsite wastewater and greywater, the 2017 effort used an epidemiology-based model to simulate pathogen concentrations contributed exclusively from onsite waste and selected Norovirus as the viral reference pathogen; the 2021 effort used municipal wastewater pathogen data and cultivable adenoviruses as the reference viral pathogen. Across source waters, the greatest differences occurred for viruses in stormwater, given the newly available municipal wastewater characterizations used for modeling sewage contributions in 2021 and the different selection of reference pathogens (Norovirus vs. adenoviruses). The roof runoff LRTs support the need for protozoa treatment, but these remain difficult to characterize due to the pathogen variability in roof runoff across space and time. The comparison highlights adaptability of the risk-based approach, allowing for updated LRTs as site specific or improved information becomes available. Future research efforts should focus on data collection of onsite water sources.

1. Introduction

Onsite non-potable water systems (ONWS) collect and treat locally available water sources for non-potable end uses such as toilet flushing and irrigation. Typical source waters include greywater, combined wastewater, roof runoff (rainwater), and stormwater, with decentralized collection scales ranging from single buildings to city districts (Sharvelle et al., 2017). Fit-for-purpose onsite treatment and use can provide economic, resource and public health benefits (Xue et al., 2016). However, given their untraditional water sources and operational configurations, care must be taken to ensure that ONWS meet acceptable levels of public health protection. To address risk-based treatment needs, quantitative microbial risk assessment (QMRA) has been used to develop pathogen log₁₀-reduction targets (LRTs) for human-infectious enteric bacteria, enteric viruses, and parasitic protozoa to achieve selected risk benchmarks during fit-for-purpose onsite uses (Sharvelle et al., 2017; Schoen et al., 2017). LRTs proposed by Schoen et al. (2017) and presented in the 2017 LRT Guidance by Sharvelle et al. (2017) (“2017 LRT Guidance” herein) have been adopted in various state and local regulations for permitting ONWS (e.g., CO, San Francisco), and other juris-dictions are currently advancing policies to develop onsite reuse programs following the risk-based approach (e.g., CA, MN, HI, WA, City of Austin, TX) (NBRC, 2019). Since publication of the 2017 LRT Guidance, additional research has been conducted to assess the conditions under which these LRTs are applicable (Schoen et al., 2018, 2020b).

Recently, the California State Water Resources Control Board commissioned a new Independent Advisory Panel to review and update the LRTs for use in its state regulations (Olivieri et al., 2021; Pecson et al., 2022b). This effort, referred to herein as the “2021 LRT Update”, recalculated the LRTs using new pathogen data and alternative approaches for characterizing the different source waters. Regulators and other stakeholders are now presented with two LRT tables to select between as they develop and implement ONWS programs.

The purpose of this article is to provide a scientific evaluation and synthesis of the two fit-for-purpose, onsite QMRA efforts, i.e., the 2017 LRT Guidance and the 2021 LRT Update, to inform appropriate selection of pathogen treatment targets for ONWS. We begin by introducing the risk-based framework for onsite water treatment, followed by an overview of the two modeling approaches and comparison of their assumptions through the QMRA process. The resulting LRTs are compared with an emphasis on their underlying assumptions, key differences, and the implications thereof for LRT interpretation. We then discuss relevant considerations when selecting between the pathogen LRTs, as well as outstanding research and guidance needs.

2. Model framework

2.1. Risk-based treatment approach

Risk-based water management frameworks aim to achieve an acceptable level of public health risk by specifying the pathogen reductions required for source waters to meet a stated risk target during given end uses (i.e., fit-for-purpose) (Petterson and Ashbolt, 2016; Schoen and Garland, 2017). In the United States, a benchmark of 10⁻⁴ infections per person per year (ppy) (Regli et al., 1991; Sinclair et al., 2015) has been used for drinking water (e.g., potable reuse in California; Hultquist, 2016; SWRCB, 2016). Other countries and the World Health Organization (WHO) utilize 10⁻⁶ disability adjusted life years (DALYs) ppy, which additionally consider the community disease burden associated with each infection case (Sinclair et al., 2015; Petterson and Ashbolt, 2016; Schoen and Garland, 2017). For a given risk target and water use configuration, QMRA is used to model pathogen LRTs that meet the benchmark based on the range of anticipated conditions, including source water pathogen characterizations, estimated exposure volumes, and pathogen dose-response relationships. Model inputs can be discrete point estimates or probability distributions reflecting variability and/or uncertainty of input parameters, in which case a chosen percentile (typically 50th or 95th) of the resulting LRT distribution is selected. Sensitivity analysis can then be used to assess the influence of model assumptions and explore alternative scenarios.

2.2. Scope of ONWS applications

Both the 2017 LRT Guidance and 2021 LRT Update focused on human and zoonotic enteric pathogens, developing recommended LRTs for enteric viruses, bacteria, and parasitic protozoa to achieve a risk benchmark of 10⁻⁴ infections ppy (the 2017 LRT Guidance also included 10⁻² as an alternative benchmark for voluntary exposures; see Section 8.3). Reference pathogens were similar (Table 1), although the generalized LRTs for each class of microbes differed in specific pathogens on which they were based. The 2021 LRT Update evaluated new data for bacteria and protozoa in roof runoff, but then used only protozoa LRTs in final recommendations (discussed in Section 7.2). Both considered non-potable end uses of unrestricted irrigation (i.e., not limited to sub-surface or drip applications; does not include food crops) and indoor use for toilet flushing and clothes washing. To protect against potential cross-connections to potable supplies and other accidental ingestions of treated non-potable water, indoor use also includes a rare, non-routine drinking exposure for a fraction (10%) of the user population. The 2021 LRT Update considered additional end uses of fire suppression, car washing, and decorative fountains, whereas risk models supporting the 2017 LRT Guidance were later applied to car washing and decorative fountains in subsequent work (Schoen et al., 2020b). The definition of “onsite” was similar between the two efforts; ONWS are intended to represent large buildings or city districts rather than single residences, generally in the range of 500 to 1000 persons. For the 2017 LRT Guidance, other decentralized scales (5- and 100-persons) were also assessed in underlying work (Schoen et al., 2017) and during later applications of the model (Schoen et al., 2018, 2020a; see Section 7.3).

Table 1.

Reference pathogens and dose-response relationships (revised from Schoen et al., 2017).

Reference pathogen	Model	Parameters	Parameter values	Units	Reference	Note	Susceptible fraction
Norovirus (GI)	hypergeometrica	alpha beta	0.04 0.055	genome copies	(Teunis et al., 2008)	Upper-bound	1
Norovirus (GI & GII.4)	fractional Poisson	P u	0.72 1106	genome copies	(Messner et al., 2014)	Lower-bound	1
Adenovirus Types 4, 7, and 16	hypergeometrica,b	alpha beta	5.11 2.80	TCID50	(Teunis et al., 2016)		1
Rotavirus	beta-Poisson	alpha beta	0.2531 0.4265	FFU	(Haas et al., 1999)		0.06/1c
Campylobacter jejuni	beta-Poisson	alpha beta	0.145 7.589	CFU	(Medema et al., 1996)		1
Salmonella enterica	beta-Poisson	alpha beta	0.3126 2884	CFU	(Haas et al., 1999)		1
Giardia lamblia	exponential	r	0.0199	cysts	(Rose et al., 1991)		1
Cryptosporidium spp.	fractional Poisson	P	0.737	oocysts	(Messner and Berger, 2016)	Upper-bound	1
Cryptosporidium spp.	exponential	r	0.09	oocysts	(U.S. EPA, 2005)	Lower-bound	1

Abbreviations: TCID50, median tissue culture infectious dose; FFU, focus forming units; CFU, colony forming units

^aThe 2021 LRT Update used the approximate beta-Poisson dose-response, which resulted in conservative LRT values that ranged from 0.0 to 0.4 higher than the hypergeometric dose-response model.

^bThe 2017 LRT Guidance included an error when reporting the adenoviruses dose-response parameters of alpha= 5.25, beta= 2.95, which were subsequently adopted in the 2021 LRT Update. However, when corrected in the 2021 LRT updated, there was no change to the resulting LRTs.

^cThe 2017 LRT Guidance assumed that only young children (6% of the population) were susceptible (Pott et al., 2012). The 2021 LRT Update used Enterovirus for occurrence and Rotavirus for the dose-response function and assumed the entire population was susceptible.

2.3. Model structure and implementation

The two ONWS QMRA efforts utilized the same overarching model structure proposed by Schoen et al. (2017). The LRT for a selected reference pathogen was calculated by setting the annual probability of infection (Pinf_annual) resulting from a selected end use to the benchmark risk and solving as follows:

$$\text{P}{\text{inf}}_{\text{annual}}={10}^{-4}=S*\left(1-{\prod }_{i=1}^n[1-DR(V*{10}^{{\text{log}}_{{10}^{(C_i)-\text{LRT}}}})]\right)$$ (Eq. 1)

where S is the fraction of people in the exposed population susceptible to each reference pathogen DR.(…) is a dose-response function for the reference pathogen V is the volume of water ingested per day of use n is the number of days of use over a year C_i is the pathogen concentration in the untreated, freshly collected source water on day i.

The predicted annual probability of infection for an end use in Eq. (1) was calculated assuming independent, daily risks; each daily risk was computed from a daily accumulated pathogen dose from all relevant activities (e.g., clothes washing and toilet flushing for indoor use; Section 5). A Monte Carlo analysis was implemented to capture the natural variation in pathogen concentration across time for each combination of input parameters (i.e., source water, end use, and pathogen dose-response selection) by random sampling from selected pathogen distributions for each day i (see Section 4 and Supplementary Material Table S1). The remaining exposure and dose-response parameters were fixed at point estimates and explored through sensitivity analyses. LRTs with non-routine ingestion were calculated by taking the weighted sum of the annual probabilities of infection for populations with and without non-routine ingestion. The 95th percentile LRT, rounded to one decimal point, was reported; although, LRT recommendations based on these results took additional considerations (Section 6).

3. Reference pathogens and dose-response

Reference pathogens represent classes of pathogens (i.e., viral, bacterial and protozoan) with potential adverse health impacts. Here, given the objective to define treatment recommendations for fecally-contaminated source waters, the specific interest was waterborne human-infectious enteric pathogens; water-based opportunistic pathogens that may grow post-treatment being controlled via appropriate system management practices (Sharvelle et al., 2017; Section 8.4). From the most recent estimate of the burden of infectious waterborne disease in the U.S. in 2014, Norovirus accounted for 21.8 M of a total 33.6 M illnesses, more than all other known pathogens combined, followed by giardiasis (415,000 illnesses) and cryptosporidiosis (322,000 illnesses) (Collier et al., 2021). Accordingly, a number of human-infectious enteric viruses, protozoa, and bacteria (i.e., Norovirus, human adenoviruses [Mastadenovirus], Rotavirus, Giardia lamblia/duodenalis/intestinalis species complex, Cryptosporidium spp., Campylobacter jejuni, Salmonella enterica) from human or animal fecal contamination were considered.

Dose-response models relate a healthy adult’s reference pathogen ingestion dose to a resulting probability of infection. The dose-response selections, adopted from peer-reviewed literature as described in Schoen et al. (2017), are listed in Table 1. Both upper- and lower-bound dose–response models were used for Norovirus and Cryptosporidium spp. to account for uncertainty in the dose-response relationships; for other pathogens, a single model was used. The mean parameter estimates of a “generalized” dose-response model that estimates the risk of gastrointestinal (GI) infection from oral exposures using the available exposure data from various adenoviruses was adopted. This model combines dose-response information from inhalation, oral ingestion, intranasal and intraocular droplet inoculation of adenovirus Types 4, 7, and 16 with health outcomes of both GI and respiratory infection Teunis et al., 2016). A conservative assumption was made that 100% of the population of healthy adults is potentially susceptible to all pathogens (i.e., S = 1 in Eq. ((1)), with the exception of Rotavirus in the 2017 LRT Guidance for which a dose-response model for healthy adults was used with the assumption that only young children were susceptible (Pott et al., 2012). For the purpose of developing an estimated viral LRT, the 2021 LRT Update approach followed the U.S. Environmental Protection Agency (EPA) Surface Water Treatment Rule (SWTR) approach for use of a “synthetic virus” (Regli et al., 1991; Macler and Regli, 1993) which combined properties of several pathogenic waterborne viruses of the enterovirus group (e.g., poliovirus, echovirus, coxsackievirus) for occurrence and Rotavirus for the dose-response function assuming 100% of the population is potentially susceptible.

4. Pathogen characterization

4.1. Data criteria and quality considerations

Selected pathogen characterization distributions are an important distinction between the 2017 LRT Guidance and 2021 LRT Update (Table 2). To characterize pathogen concentration and occurrence in the untreated source waters, Schoen et al. (2017) specified the following criteria: 1) analytical methods used to enumerate the pathogens were comparable to those used in the dose-response studies (i.e., “conventional” (culture-based or direct count) methods for all reference pathogens except for Norovirus that utilized qPCR gene copies); and 2) if a large fraction of the samples were non-detects, the limit of detection was specified. Generally, these criteria were not met, or available data were too limited in scope to enable broad representative applicability; thus, Schoen et al. (2017) modeled the pathogen concentrations in source waters (described for each source water below).

Table 2.

Summary comparison of assumptions and rationale for pathogen characterizations in onsite non-potable water system source waters.

Source Water		2017 LRT Guidance	2021 LRT Update
Onsite combined wastewater (WW) (~500–1000 persons)	Approach & Rationale	Municipal WW data (centralized) not representative of small systems with variable pathogen load and limited dilution from other sources (stormwater, infiltration, industrial and commercial inputs) Lack empirical data for collection scales of interest Thus, develop simulated data based on epidemiological modeling (Jahne et al., 2017)	Assume equivalence between municipal WW and decentralized WW Relies on empirical dataset for centralized WW Alternative analysis using available data from one onsite system (Kothari et al., 2020)
	Basic Assumptions	Intermittent pathogen occurrence among small populations; no dilution in collection system Fecal load based on fecal indicator bacteria in fresh collections Pathogen infections simulated based on estimated GI illness rates (as a proxy for infection) attributed to each pathogen by disease monitoring programs (Scallan et al., 2011). Shedding characteristics from clinical studies	Not sufficient information to confirm that onsite WW has higher peak concentrations and more variability than municipal Use DPR-2 municipal WW (centralized) dataset (n = 120 per pathogen, represents WW from roughly 10 million people in California) (Pecson et al., 2022a) Set non-detect values (NDs) equal to limit of detection/quantification (LOD/LOQ) and model as log-normal with 100% occurrence
	Uncertainty & Limitations	Not based on empirical data Requires multiple epidemiological assumptions Simulated Norovirus concentration validated but not adenoviruses (Jahne et al., 2020)	Insufficient data to demonstrate equivalency between centralized and decentralized collections; does not account for differences in WW dilution between decentralized and municipal systems (e.g., including flows from non-domestic WW sources or stormwater flows/infiltration) Limited to 5 sites in CA Collection period occurred partly during a pandemic (see discussion in Pecson et al. (2022a)) Alternative analysis dataset limited to one building for which higher Norovirus measurements were observed in a different study (Jahne et al., 2020)
Onsite greywater (GW)	Approach & Rationale	Lack empirical data Thus, develop simulated data based on epidemiological modeling (Jahne et al., 2017)	Lack empirical data Assume pathogen characterizations are comparable to diluted WW Use industry-best, empirical dataset for centralized WW; adjust to go from WW to GW Alternative analysis using modeled data (Jahne et al., 2017) and limited empirical data for virus (Jahne et al., 2020)
	Basic Assumptions	Same as WW	GW has 1% pathogen concentration of WW; previous LRTs for locally collected wastewater were roughly two orders of magnitude greater than greywater (Schoen et al., 2017) Use DPR-2 municipal WW (centralized) dataset (n = 120 per pathogen, represents WW from roughly 10 million people in California) (Pecson et al., 2022a) Set NDs equal to LOD/LOQ and model as log-normal with 100% occurrence For alternative analysis, modeled Jahne et al. (2020) as log-normal with NDs set to LOD – assume 100% occurrence
	Uncertainty & Limitations	Not based on empirical data Requires multiple epidemiological assumptions Viruses in GW were not well characterized by the model (overpredicting occurrence and concentration; Jahne et al., 2020)	Not based on empirical data GW is not a physical dilution of WW; characterization approach is mathematically contrived Assumes equivalence between municipal and onsite WW Conservative assumption of 100% pathogen occurrence Neglects potential variability in collection types (e.g., shower vs. sinks) WW data limited to 5 sites in CA, part of which was collected during a pandemic
Stormwater (SW)	Approach & Rationale	Develop empirical LRTs using available measurement data from residential, commercial/light industrial, mixed use, and urban watersheds in the U.S. (McBride et al., 2013) To account for poorly understood variability and potential site-specificity in empirical LRTs, estimate SW as dilutions of WW informed by empirically derived LRTs Used variety of data sources to characterize generalized pathogen concentrations in WW	Limited empirical data from stormwater itself to characterize concentrations Estimate SW as dilutions of WW Use empirical dataset for centralized WW Assume same dilutions as 2017
	Basic Assumptions	Used 0.1% and 10% dilutions, representing range of LRTs determined based on measurements Log-normal distribution for Norovirus (Pouillot et al., 2015) across the world and across seasons; later supported by North American data (Eftim et al., 2017) All other pathogens use log-uniform based on range reported in North America (Soller et al., 2017)	Use 0.1% and 10% dilutions following Schoen et al., 2017 and supported by additional studies (Alja’fari et al., 2022b; Secru et al., 2011; Sauer et al., 2011; McBride et al., 2013) Use DPR-2 municipal WW (centralized) dataset (n = 120 per pathogen, represents WW from roughly 10 million people in California) (Pecson et. al, 2022a)
	Uncertainty & Limitations	Based in part on international data Applies conservative assumption on occurrence	Not based on empirical data Applies conservative assumption on occurrence WW data limited to 5 sites in CA, part of which was collected during a pandemic
Roof Runoff (RR)	Approach & Rationale	Lack of empirical data, particularly in U.S. Estimated pathogen concentrations based on animal model assuming fresh fecal deposits (Schoen et al., 2017)	Direct measurement of four pathogen gene targets in U.S. roof runoff (Alja’fari et al., 2022a)
	Basic Assumptions	Based on fecal contamination from seagulls (Levesque et al., 2000) 100% pathogen occurrence	Relies on limited empirical data Empirical data represents 4 cities in U.S. over multiple seasons (n = 80 samples) with similar results to Australian study ( Ahmed et al., 2010, 2012) Focus on protozoan risks 100% pathogen occurrence for log-normal distributions; log-uniform distributions modeled with occurrence assumptions
	Uncertainty & Limitations	Not based on empirical data Only sufficient data to model one animal source (gulls) Only sufficient data to model bacterial pathogens; no protozoa Conservative occurrence assumption Conservative bacterial concentrations and occurrence relative to later U.S. study (Alja’fari et al., 2022a)	Very few empirical pathogen detections to characterize distributions Requires gene copy to infectious unit translation (conservatively assumed 1:1) Conservative occurrence assumption

In contrast to the 2017 LRT guidance document that relied on modeling approaches to generate pathogen concentrations in source waters, the 2021 LRT update relied primarily on empirical waterborne pathogen data (Pecson et al., 2022a; Alja’fari et al., 2022a). Selection criteria included 1) large sample size; 2) monitoring of multiple locations over time; 3) freshly collected samples; 4) high method sensitivity; 5) enumeration methods compatible with dose-response functions; 6) targets human infectious strains or groups; and 7) reports raw data including recovery and limits of detection (Pecson et al., 2022b). Where necessary, assumptions were made to estimate pathogen densities in source waters of interest relative to the available datasets. Additional alternative datasets were also considered for certain source waters (Kothari et al., 2020; Jahne et al., 2020, 2017).

The specific approaches used by each LRT development effort are further described below and in Table 2 with comparison of estimates presented in Fig. 1. Additional data describing source water characterizations and fitted distributions is provided in Supplementary Material (Table S1).

Fig. 1.

Modeled pathogen concentrations in greywater (left) and stormwater containing a 10⁻³ wastewater contribution (right) from the 2017 LRT Guidance and 2021 LRT Update. Greywater symbols denote pathogen occurrence rates in the epidemiology-based simulation, i.e., values below these are zero. Onsite wastewater and stormwater 10⁻¹ distributions follow the same patterns as greywater and stormwater 10⁻³, respectively. Roof runoff reference pathogens did not align for comparison.

4.2. Onsite wastewater and greywater

Direct pathogen monitoring data for fresh collections of wastewater and greywater from onsite reuse systems have been extremely limited (e. g., system-specific) or, for many pathogens, unavailable (Schoen and Garland, 2017; Jahne et al., 2017; Kusumawardhana et al., 2021). The occurrence of GI infections is sporadic for small population sizes; depending on the collection scale (e.g., 5- vs. 1000-person), there may often be no active pathogen shedders among them. However, for onsite systems with low contributing population members there is also limited dilution from other users, no diluting flow from stormwater or industrial wastewater inputs (as may occur in municipal collections), and short wastewater residence time for attenuation or decay prior to treatment, all contributing to potentially high pathogen loads when infections do occur (Jahne et al., 2017; O’Toole et al., 2014; Barker et al., 2013; Fane et al., 2002). Following this logic, the 2017 LRT Guidance adopted an epidemiology-based approach wherein the fecal contamination levels of source-diverted greywaters and onsite combined wastewater were coupled with modeled infections in relevant population sizes (selecting 1000 persons) and the pathogen shedding dynamics of those infected individuals (Jahne et al., 2017).

The 2021 LRT Update used an empirical dataset for centralized wastewater collected at five facilities in California (“DPR-2”) (Pecson et al., 2021, 2022a), assuming equivalence between decentralized wastewater and raw municipal wastewater (Pecson et al., 2022b). The pathogen monitoring campaign developed and implemented an optimized standard operating protocol to characterize the concentration of human pathogens in raw wastewater. Methods to detect relevant viral and protozoan pathogens in raw wastewater were optimized and implemented during a 14-month monitoring campaign. Over 120 samples were collected from five wastewater treatment plants treating a quarter of California’s population (approximately 10 million people). Samples were analyzed for two parasitic protozoa (Cryptosporidium spp. and Giardia) using microscopy methods, three enteric viruses (Enterovirus, adenoviruses, and Norovirus) using culture and/or molecular methods, and male-specific coliphage using culture methods as described by Pecson et al. (2022a). Bacteria were not monitored in this campaign because bacteria are thought to be effectively controlled when managing viral and protozoan risks (Macler and Regli, 1993; Regli et al., 1991; NRC, 2012; see Section 7.2). This dataset was selected for the 2021 LRT update because it met the above data criteria (Section 4.1). The key assumption underlying the use of this dataset is that concentrations in municipal wastewater have sufficient equivalence with onsite wastewater for larger scale systems of interest (greater than 500 persons).

The 2021 LRT Update also compared the centralized wastewater LRT results against LRTs developed from a limited building-scale pathogen dataset (Kothari et al., 2020). However, the Kothari et al. (2020) study reported a high frequency of Norovirus non-detects (9/15 samples) compared to a similar study in the same building which processed a larger sample volume (10 L vs. 0.5 L; 1/28 samples non-detect) (Jahne et al., 2020). The authors further note that gene copy estimates may be underestimated by 1–2 log₁₀ due to low virus recovery in spiked matrices (Kothari et al., 2020).

For greywater LRTs, the 2021 LRT Update relied on three comparative approaches: 1) estimate based on a moderate dilution (10⁻²) of untreated municipal wastewater assuming 100 percent occurrence; (2) simulated epidemiological dataset of Cryptosporidium spp. and Giardia density and occurrence in the community based on Jahne et al. (2017), and (3) limited measured pathogen data (i.e., Norovirus GII and adenoviruses) provided in Jahne et al. (2020).

4.3. Stormwater

The 2017 LRT Guidance used both observed and modeled pathogen concentrations to estimate LRTs for stormwater (Table 2). Using stormwater pathogen concentration observations, the LRTs for each pathogen varied up to 3.0 log₁₀ across stormwater collections due to variability in pathogen concentrations; yet the possible variation was likely underestimated given the vast variety of stormwaters possible across sites and event conditions (Ahmed et al., 2019; Bichai and Ashbolt, 2017). Instead, the 2017 LRT Guidance reported the LRTs for 10⁻¹ and 10⁻³ dilutions of municipal wastewater (i.e., 1:10 and 1:1000 volumetric contributions of sewage to stormwater), selected to represent high and low levels of sewage contamination in stormwater, which is anticipated to drive human health risks (Ahmed et al., 2019; Bichai and Ashbolt, 2017). The LRTs derived from observations were lower than predicted by the 10⁻¹ dilution of raw wastewater and were more comparable with 10⁻³ or 10⁻⁴ dilutions of raw wastewater. Thus, the 10⁻¹ dilution, while conservative, provides the maximum LRT for stormwater in the absence of site-specific information. The two generalized dilutions provide broadly applicable treatment targets while also aiding decision makers presented with a range of likely contamination impacts to their stormwater (Alja’fari et al., 2022b; Schoen et al., 2017). The 2021 LRT Update adopted the same dilution approach but used the empirical municipal wastewater dataset collected at five facilities in California (Pecson et al., 2022a).

4.4. Roof runoff

Due to a lack of data that met the inclusion criteria for Schoen et al. (2017) and the associated 2017 LRT Guidance, the amount of fecal contamination in roof-collected rainwater was estimated using fecal indicator bacteria (FIB) and coupled with estimated levels of pathogens in animal fecal sources (Schoen and Ashbolt, 2010). Of animals with roof access, required data on pathogen densities in fecal samples were only available for seagulls, and then only for bacterial pathogens. Schoen et al. (2017) acknowledged uncertainty of the conservative assumptions made and that protozoan targets were also warranted but lacking data to develop them. The 2021 LRT Update used direct measurement of four pathogen gene targets in roof runoff (Alja’fari et al., 2022a) described by both a log-normal and log-uniform distribution. For the log-uniform distribution, the 2021 LRT Update corrected for occurrence (Olivieri et al., 2021). These data were collected across multiple sites and seasons using molecular methods (Table 2) yet observed relatively few pathogen detections (see Section 6.3.3).

5. Exposure assumptions

There is a lack of empirical data to inform the exposure assumptions related to ingestion volumes for most of the selected uses and the exposure frequency of less common uses. Exposure assumptions, which were generally similar between the two guidance documents, are presented in Table 3 and further described in the Supplemental Material. For toilet flushing, clothes washing, and unrestricted irrigation and dust suppression, the 2017 LRT Guidance and 2021 LRT Update adopted values from Australian reuse guidelines (NRMMC-EPHC-AHMC, 2006) (with the modification of 365 days of use for clothes washing assumed for 2017 LRT Guidance), which used conservative assumptions (described in Supplemental Material). Similarly, fire suppression volumes and exposure frequency were based on conservative assumptions, while decorative fountains and vehicle washing were informed by limited empirical data, as described in the Supplemental Material. Indoor use included clothes washing and toilet flushing; for the 2021 LRT Update, inclusion of decorative fountains was also considered (Supplemental Material). For residential indoor use, the exposure assumptions also included non-routine ingestion, which dominated the LRTs compared to routine exposures. The volume consumed during non-routine ingestion of treated waters or a cross-connection event of treated to potable waters corresponds to one day of potable consumption, which is consistent with the limited available data for cross-connections between potable and treated reclaimed water (Schoen et al., 2018).

Table 3.

Summary comparison of exposure assumptions.

End Use	LRT Guidance	Daily Exposure Volume (L)	Exposure Frequency (days/ year)	Fraction of Population Exposed	Key Assumptions
Toilet flush water	2017/2021	0.00003	365	1	Adopted from Australian guidelines1; approximately 1 second of hand to mouth exposure2; 3 flushes per day
Clothes washing	2017	0.00001	365	1	Adopted from Australian guidelines1; assumed frequency; only used in indoor use calculation
Clothes washing	2021	0.00001	100	1	Adopted from Australian guidelines1; assumed frequency
Unrestricted irrigation and dust suppression	2017/2021	0.001	50	1	Adopted from Australian guidelines for municipal irrigation1; excludes food crops; frequency selected for dry climates
Cross-connection of treated water with potable water	2017/2021	2	1	0.1	Target group for accidental ingestion may be young children (6.5% of population under age 5)3; only used in indoor use calculation
Potable consumption	2017	2	365	1	Potable consumption1
Fire Suppression	2021	0.002	20	1	Data from Water Services Association of Australia “Health Risk Assessment of Fire Fighting from Recycled Water Mains”4
Car Washing	2021	0.001	12	1	Assumed frequency; assumed similar exposure as garden irrigation1
Decorative Fountains A	2021	N(−1.05,0.81)a,b × 10−3	50	1	Assumed similar exposure as car washing5; two distributions fitted to same data
Decorative Fountains B	2021	U(0.06, 3.79)a × 10−3	50	1	Assumed similar exposure as car washing5; two distributions fitted to same data

References:

¹NRMMC-EPHC-AHMC, 2006.

²de Man et al., 2014.

³U.S. Census Bureau, 2011.

⁴Deere et al., 2004.

⁵Sinclair et al., 2016.

^aValues are mL.

^bValues are log₁₀.

6. Pathogen log-reduction targets

6.1. 2017 LRT Guidance

The 2017 LRT Guidance based its recommended LRTs on the 95th percentile LRTs for 10⁻⁴ and 10⁻² ppy infection risk targets reported by Schoen et al. (2017), rounding up to the nearest 0.5 unit. The two risk benchmarks were included to provide flexibility on whether exposure is considered involuntary or voluntary (respectively); we focus on the more stringent primary targets here. For onsite wastewater and greywater, the 1000-person simulation results were selected; these LRTs were greater than for 100-person collections (Schoen et al., 2017, 2020a; Section 7.3). Two sets of stormwater LRTs were included for different levels of sewage contamination (10⁻¹ and 10⁻³ dilutions of municipal sewage) representing the range of LRTs anticipated based on limited available direct pathogen observations (Schoen et al., 2017; Section 7.3).

The LRTs for Norovirus and Cryptosporidium spp. were sensitive to the selected dose-response function. The LRTs for Norovirus calculated using the upper bound dose-response, the hypergeometric model proposed by Teunis et al. (2008), were approximately three log₁₀ greater than the LRTs calculated using the lower bound, the fractional Poisson model proposed by Messner et al. (2014). The LRTs for Cryptosporidium spp. calculated using the fractional Poisson model proposed by Messner and Berger (2016) were approximately one log₁₀ greater than the LRTs calculated using the exponential model (U.S. EPA, 2005). Norovirus (using the lower-bound dose-response relationship), Campylobacter spp., and Cryptosporidium spp. (using the upper-bound dose-response relationship) LRTs were selected when recommending class-wide LRTs for human enteric viruses, enteric bacteria, and parasitic protozoa, respectively. Each of these organisms had the highest modeled LRTs within its respective pathogen class (Fig. 2; Sharvelle et al., 2017); however, it should be noted that subsequent analyses support the upper-bound dose-response model for Norovirus (Teunis et al., 2020). For blended waters, LRTs for the highest-risk water source should be used regardless of proportional volume (Sharvelle et al., 2017).

Fig. 2.

Pathogen log-reduction targets (LRTs) for indoor use (left) and unrestricted irrigation (right) from the 2017 LRT Guidance (solid bars) and 2021 LRT Update (hatched bars). Symbols denote recommended values for enteric viruses (asterisks) and parasitic protozoa (circles) for each source water. Pathogen dose-response models are noted where multiple options were used (HG, hypergeometric; FP, fractional Poisson; EX, exponential). Roof runoff reference pathogens did not align for comparison.

6.2. 2021 LRT Update

The 2021 LRT Update estimated log reductions needed to achieve the risk goal of 10⁻⁴ infections ppy for healthy adults. The 95th percentile value was used to select an LRT from the distributions. Giardia were used for final protozoa LRT recommendations. Cultivable adenoviruses yielded essentially equivalent LRT results as the cultivable Enterovirus/Rotavirus approach in nearly all cases and slightly higher results in others, and were bounded by the upper and lower Norovirus dose-response results. For this reason, it was selected as the reference viral pathogen for LRTs.

Both the primary and alternate datasets were used to test sensitivity of the LRTs to assumptions about source water pathogen concentrations. Overall, there was reasonable alignment in the LRTs regardless of the pathogen datasets used and LRTs were within one log₁₀ value of each other (Supplemental Material Table S2). Larger differences were observed between the greywater LRTs across the pathogen groups. This deviation was due to the differences in approaches used to estimate concentrations: a modified empirical approach based on municipal wastewater (primary) and an epidemiological model-based approach (alternate). These two approaches did not lead to systematic differences, however, as neither approach had uniformly higher (or lower) LRTs, thus providing a weight of evidence in results.

Given the lack of roof runoff pathogen data, the 2021 LRT Update considered the results of both log-normal and log-uniform distributions fit to the Alja’fari et al. (2022a) pathogen dataset. Generally, the highest LRTs required for indoor use of roof runoff ranged from 1.0 to 1.5 across the three pathogens evaluated (only 1 value out of 24 was greater than 1.5). Similarly, unrestricted irrigation generally required 0.3 to 0.7 LRT across the pathogens. The 2021 LRT Update recommend that the more conservative LRT between the log-normal and log-uniform distributions be used for roof runoff given the lack of data. In addition, the 2021 LRT Update recommended that applying the criteria for Giardia alone will provide sufficient protection against pathogenic bacteria, particularly when using validated technologies that comply with approved protozoa crediting frameworks (Section 7.2). Consequently, a separate bacterial LRT was not recommended.

The 2021 LRT Update considered additional end uses including fire suppression, car washing, and decorative fountains. The 2021 analysis indicated that fire suppression and car washing (when considered alone) did not require LRTs higher than those needed for unrestricted irrigation and indoor use. The 2021 LRT Update also estimated LRTs for combined indoor use that included exposure due to decorative fountains as well as toilet flushing, clothes washing, and non-routine ingestion. For the selected reference pathogens, the indoor use LRTs that included decorative fountains were the same as the indoor use LRTs that did not include decorative fountains except for the virus LRTs for onsite wastewater and stormwater, which were 0.1-log₁₀ higher. This would have resulted in a 0.5-log₁₀ increase in the LRT requirements when rounding up to the nearest 0.5-log₁₀ value. For this reason, the 2021 LRT Update noted that the indoor use LRTs would be protective if decorative fountains were included as an indoor use. This is consistent with findings from Schoen et al. (2020b).

6.3. Comparison of LRT results

For onsite wastewater, greywater, and stormwater, LRTs for human enteric viruses and parasitic protozoa were within 1.5-log₁₀ units between the 2017 LRT Guidance and 2021 LRT Update (Table 4). Comparison of results by source water is discussed below. Given the dependence of 2021 LRTs on the DPR-2 municipal wastewater dataset, additional discussion of its use is provided in Supplemental Material.

Table 4.

Comparison of log-reduction target (LRT) recommendations for onsite non-potable water systems.

Water Use Scenario	Human Enteric Viruses		Parasitic Protozoa		Enteric Bacteria
	2021 LRT Update	2017 LRT Guidance	2021 LRT Update	2017 LRT Guidance	2021 LRT Update	2017 LRT Guidance
Untreated Onsite Wastewater
Unrestricted irrigation	7.5	8.0	5.5	7.0	n.d.c	6.0
Indoor use	8.0	8.5	6.5	7.0	n.d.c	6.0
Greywater
Unrestricted irrigation	5.5	5.5	3.5	4.5	n.d.c	3.5
Indoor use	6.0	6.0	4.5	4.5	n.d.c	3.5
Stormwater (10% wastewater contribution)
Unrestricted irrigation	6.5	5.0	4.5	4.5	n.d.c	4.0
Indoor use	7.0	5.5	5.5	5.5	n.d.c	5.0
Stormwater (0.1% wastewater contribution)
Unrestricted irrigation	4.5	3.0	2.5	2.5	n.d.c	2.0
Indoor use	5.0	3.5	3.5	3.5	n.d.c	3.0
Roof Runoff Water
Unrestricted irrigation	n/aa	n/aa	1.0	n.d.b	1.0/0.0d	3.5
Indoor use	n/aa	n/aa	1.5	n.d.b	1.5/1.0d	3.5

^aHuman enteric viruses were assumed to be not applicable in roof runoff.

^bThe 2017 LRT Guidance did not determine protozoan LRTs for roof runoff due to lack of available data.

^cThe 2021 LRT Update did not determine enteric bacteria LRTs for wastewater, greywater, or stormwater.

^dThe 2021 LRT Update calculated enteric bacteria LRTs for roof runoff but did not recommend them; shown are values for Campylobacter/Salmonella spp.

6.3.1. Onsite wastewater and greywater

Onsite wastewater and greywater differed in both characterization approach (Table 2) and selected reference pathogens, yet the same or slightly lower LRTs were determined in the 2021 LRT Update (0.5–1.5 log₁₀ unit difference). The 2017 LRT Guidance utilized the epidemiology-based simulation of Jahne et al. (2017) for Norovirus in its enteric virus recommendations, which has subsequently been supported at a building of relevant scale (Jahne et al., 2020); Schoen et al. (2017) did not propose an LRT for adenoviruses due to lack of data. In contrast, the 2021 LRT Update did not recognize distinction between onsite and centralized collections, selecting to use the new DPR-2 dataset available for municipal wastewater and cultivable adenoviruses as the reference pathogen. Although Norovirus LRTs were lower than those based on the scale-dependent epidemiological model due to its lower concentrations in the DPR-2 dataset (Pecson et al., 2022a), the 2021 LRT Update LRTs for adenoviruses were similar to the 2017 LRT Guidance Norovirus-based recommendations (0.5 log₁₀ units lower) and fell between the two assumptions for Norovirus dose-response (Fig. 2). Decentralized black-water collections (wastewater that excludes shower and laundry sources) follow onsite wastewater LRTs under both guidances, given limited available data from these systems (Olivieri et al., 2021).

Both studies assumed that greywater was approximately 2 log₁₀ units more dilute than sewage. Jahne et al. (2017) based this assumption on the relative loading of FIB in fresh greywater/wastewater collections and Pecson et al. (2022b) followed results of the previous 2017 LRT guidance (Table 2). Despite their difference in reference pathogens, recommended virus LRTs for greywater were the same. For parasitic protozoa, the epidemiology-based simulation and DPR-2 dataset differed in which pathogens presented the greatest modeled risk (Cryptosporidium spp. vs. Giardia). The 2017 LRT Guidance selected the upper-bound dose-response for Cryptosporidium spp., which is conservative for Giardia based on simulation results; in the 2021 LRT Update, the selected Giardia LRTs were greater than those for Cryptosporidium spp. using either dose-response model (Fig. 2). Therefore, the 0.5–1.5 log₁₀ unit lower protozoan LRTs in the 2021 LRT Update may be considered a result of the lower Cryptosporidium spp. measurements in municipal sewage vs. the scale-dependent simulation.

6.3.2. Stormwater

Across source waters, the greatest differences occurred for viruses in stormwater (1.5 log₁₀ greater in the 2021 LRT Update), given the differences in municipal wastewater characterizations used for modeling sewage contributions (i.e., literature review vs. the DPR-2 dataset) and selection of reference pathogens (Norovirus vs. adenoviruses). The systematic Norovirus literature review by Pouillot et al. (2015) used in the 2017 LRT Guidance was intended to capture broad variability across U. S. and international systems (Table 2). The DPR-2 dataset was collected specifically in California and most accurately reflects wastewaters in that region, with the comparison of pathogen distributions across plants supporting the conclusion that the pathogen concentrations do not vary widely in wastewaters across the state (Pecson et al., 2022a).

DPR-2, part of which was collected during the SARS-CoV-2 pandemic, reported generally lower Norovirus than measured previously, consistent with California incidence data (Wigginton et al., 2021). However, an analysis of the data showed no significant change for adenoviruses, on which 2021 LRT recommendations were based, in the periods before and during the pandemic (Pecson et al., 2022a). In the 2017 LRT Guidance literature review, a comprehensive dataset was available for Norovirus LRT development that has been supported by subsequent studies (Eftim et al., 2017), whereas there was limited data for human adenoviruses (Soller et al., 2017). Stormwater LRTs for Norovirus using different dose-response models bracketed those for adenoviruses in the 2021 LRT Update; in the 2017 LRT Guidance, the LRT for adenoviruses was comparable to the lower-bound Norovirus selected for recommended LRTs (<0.5 log₁₀ unit difference). This alignment is important since concentrations of adenoviruses are based on cultivable measurements whereas Norovirus relies on total estimates (infectious and non-infectious) by molecular methods (discussed in Section 8.1). Recommended stormwater LRTs for parasitic protozoa were the same, although based on different pathogens (Fig. 2).

6.3.3. Roof runoff

Recommended roof runoff LRTs were not directly comparable as only enteric bacteria were presented in the 2017 LRT Guidance and only parasitic protozoa were recommended in the 2021 LRT Update. The earlier guidance recognized that protozoan LRTs were warranted but lacked sufficient data to assess them (Schoen et al., 2017; Sharvelle et al., 2017), whereas the 2021 LRT Update is based on detected G. duodenalis β – giardin gene targets (but not C. parvum) in 4/79 samples (5.1%) in four U.S. cities (Alja’fari et al., 2022a). Despite the large number of non-detects, the 2021 LRT Update conservatively assumed that Giardia were always present and set these concentrations to the detection limit (when fitting a log-normal distribution) or used the measured range and detection rate (for log-uniform). Resulting LRTs for protozoa reinforce that treatment processes appropriate for these pathogens are necessary, although LRTs are based on only 4 positive samples and estimates remain uncertain. In the recent measurement study, detection rates of Salmonella and Campylobacter spp. were lower than the 100% occurrence assumed by Schoen et al. (2017), even when accounting for potential censoring (58% and 39% of modeled concentrations would be detectible based on stated limits of detection), although measured concentrations were within the modeled range (above median and below maximum) (Alja’fari et al., 2022a). While bacterial targets were not ultimately recommended in the 2021 LRT Update (Section 7.2), new LRTs based on these measurements were 2–3 log₁₀ units lower than in the 2017 LRT Guidance. This result indicates that the original enteric bacteria LRTs for roof runoff remain protective in the context of new measurements, although their uncertainty is noted by both Schoen et al. (2017) and Sharvelle et al. (2017). Both LRT studies assumed that human-infectious enteric viruses were unlikely to be found in roof runoff (Sharvelle et al., 2017).

7. Implementation considerations

7.1. Applicable end uses

The 2017 LRT Guidance included indoor use (toilet flushing, clothes washing, and rare non-routine ingestion events) and unrestricted irrigation (ornamental crops and dust suppression) (Sharvelle et al., 2017). Subsequent work demonstrated that the indoor use LRTs, which include the accidental ingestion/cross-connection safety factor, are also protective for use of treated non-potable water for decorative fountains and vehicle washing provided that the estimated daily exposures remain below approximately 10⁻⁵ L (for daily use) or 10⁻⁴ L (for 50 days of use per year) (Schoen et al., 2020b). When non-routine ingestion was removed, the change in LRT was dependent on the routine ingestion volume and pathogen; the LRT decreased by up to 1 log₁₀ for indoor water use. The indoor use LRTs were not protective for use of treated wastewater or greywater for bathing and showering (Schoen et al., 2020b).

The 2021 LRT Update included the same primary uses (indoor use for toilet flushing and clothes washing with rare accidental ingestion; unrestricted irrigation for non-food purposes), but also extended the LRTs to fire suppression and car washing for which both indoor use and irrigation LRTs were conservative. In the 2021 LRT Update, different exposure distributions resulted in up to 0.1 log₁₀ increase in indoor use protozoan LRTs when decorative fountains were included. As a result, decorative fountains were included as a separate end use and an increase in indoor use LRTs was not recommended. Microbial risks from decorative fountains, however, may be dominated by water-based opportunistic pathogens (such as Legionella and non-tuberculous mycobacteria) rather than enteric ones when water temperatures may exceed 20 °C and aerosols are generated (Haupt et al., 2012; Kanamori et al., 2016; Salinas et al., 2021; Section 8.4).

7.2. Inclusion of bacterial LRTs

The 2021 LRT Update recommendation was to not include enteric bacteria LRTs within California regulatory requirements, consistent with current reuse regulations in the state (SWRCB, 2018). While risks from enteric bacteria in untreated source waters exist, bacteria do not drive selection of unit treatment processes for wastewater, greywater, and stormwater and are assumed to be sufficiently removed (i.e., would meet LRTs) when systems are designed to meet human virus LRTs (Macler and Regli, 1993; Regli et al., 1991; NRC, 2012). Unit treatment processes that remove viruses by size exclusion also remove bacteria and disinfection processes (UV, chorine, and ozone) generally require the same or higher doses for treatment of viral pathogens compared to bacterial pathogens (Hijnen et al., 2006; WaterSecure, 2017a, 2017b). Accordingly, existing treatment crediting frameworks for pathogen reduction do not include enteric bacteria, thus creating pragmatic issues for bacterial LRT implementation.

Data collected from roof runoff suggests that potentially human pathogenic bacteria exist in roof runoff while viruses that are human pathogenic are assumed not to be present (Schoen et al., 2017; Pecson et al., 2022b). Based on recent data that included several enteric bacteria and human-infectious protozoa (Alja’fari et al., 2022a), the 2021 LRT Update recommended an LRT based only on the protozoan detected. The recommendation for CA also includes a requirement to maintain a disinfection residual in the distribution system that is verified through monitoring. While primarily targeting certain opportunistic water-based pathogens such as Legionella, this disinfection step provides additional protection against enteric pathogens including bacteria as well.

For those using the 2017 LRT Guidance, or not requiring a disinfection residual for all applications, there is motivation to develop a crediting framework for bacteria, and this consideration is particularly important for roof runoff where a viral LRT is not required. Similarly, if a system aims to meet the protozoa target and is not required to maintain a disinfection residual, there is potential for unit treatment processes such as cartridge filtration to be credited for protozoa that may not achieve sufficient treatment of bacteria. Antimicrobial resistance among bacteria also remains an emerging concern during water reuse (Nappier et al., 2018).

7.3. Application across wastewater collection sizes

As noted above, both the 2017 LRT Guidance and 2021 LRT Update for wastewater and greywater are intended for onsite or decentralized systems on the scale of approximately 500–1000 persons, i.e., large buildings or reuse districts. The 2017 LRT Guidance explicitly considered potential scaling effects on pathogen occurrence and densities in small wastewater and greywater collections (Section 4.2) using the scale-dependent models of Jahne et al. (2017). In that work, several population sizes (5, 100, and 1000 persons) were modeled with the 1000-person simulation ultimately selected for LRT recommendations. Later work (Schoen et al., 2020a) showed that LRTs based on the 100-person model were 0.5 log₁₀ units lower and therefore that the original LRTs are also protective for smaller population sizes. Although they did not calculate LRTs, other studies estimated risks for a larger population size (40,000 persons) using the same epidemiology-based approach, finding <1 log₁₀ unit increase in associated risk estimates compared to the previous 1000-person simulation (Schoen et al., 2018). This implies that, following the assumptions of Jahne et al. (2017), risks (and hence LRTs) would increase marginally for larger population scales. This modeling exercise supports the 2021 LRT Update assumption that LRTs developed to protect wastewater collections from large populations are conservative for smaller collection systems, although municipal wastewater also contains influent from other sources (e.g., inflow and infiltration, industrial inputs) that may dampen pathogen concentrations relative to onsite collections (Kothari et al., 2020). Wastewater and greywater LRTs in the 2021 LRT Update, based on municipal wastewater pathogen characterizations, were the same or 0.5–1.5 log₁₀ units lower than those in the 2017 LRT Guidance, based on the scale-dependent approach.

Of further interest is the application of LRTs for large buildings/districts to smaller ones, e.g., single family homes. While beyond the scope of either the 2017 LRT Guidance or 2021 LRT Update, Schoen et al. (2017) modeled LRTs for a 5-person collection scale applicable to this scenario. Given the rare occurrence of pathogen infections within this small population size, only the most prevalent waterborne enteric pathogen (Norovirus) had stable LRTs for 95th percentile annual risk (i. e., to achieve the risk benchmark in 95% of years) following the epidemiology-based approach. Like the 100-person model, these LRTs were lower (<1 log₁₀ unit) relative to the 1000-person collection used in the 2017 LRT Guidance, indicating that enteric virus LRTs remain conservative at smaller scales. For other pathogen classes, Schoen et al. (2017) note that “the 99th percentile log₁₀ reduction for protozoa and bacteria is greater than zero and approximately equal to the 95th percentile target log₁₀ reductions for the 1000-person system”. Therefore, while the 95th percentile LRT is technically zero for these pathogens due to their occurrence in <5% of years (Jahne et al., 2017), they do present non-zero risks and some level of treatment is warranted. Use of LRTs for 1000-person collections would be conservative for this population scale. Likewise, treatment for enteric viruses using 95th percentile LRTs may effectively manage risks from enteric bacteria and parasitic protozoa as well (Section 7.2). However, it should be noted there are likely more significant routes of enteric pathogen exposure within a household (e.g., person-to-person and via contaminated surfaces) relative to water reuse.

7.4. Stormwater LRT selection

The intent for the range of stormwater LRTs in guidance by Sharvelle et al. (2017) was to provide flexibility in selection of appropriate treatment targets based on expected levels of sewage contamination resulting from deteriorating wastewater collection systems, homeless encampments, etc. Pathogen characterizations were based on sewage dilutions for which virus LRTs aligned with those based on data reported for different land uses (Schoen et al., 2017; McBride et al., 2013). However, the range of LRTs recommended by Sharvelle et al. (2017) is large (Table 4), resulting in a wide range of possible treatment requirements.

The wide range of LRTs combined with lack of understanding from the regulatory community and practitioners on appropriate approaches for estimation of sewage contributions has presented ambiguity for selection of LRTs for stormwater treatment systems that are protective of public health. This has resulted in regulatory bodies selecting conservative stormwater treatment targets for the 10⁻¹ sewage dilution (e.g., San Francisco; SFDPH, 2020), resulting in extensive treatment requirements. Data is lacking to support selection of sewage contribution levels and subsequent LRTs based on infrastructure condition and land use, which has not been found to be an adequate predictor of stormwater microbial quality (McBride et al., 2013). One possible approach is to measure human microbial source tracking markers (MSTs), e.g., HF183, to estimate sewage contributions to stormwater. A limitation for use of such MSTs is that their removal times in environmental water matrices have been found to be shorter than the inactivation times of viral pathogens (Ahmed et al., 2014). More data is required to further guide selection of an appropriate suite of human MSTs that detect both long duration and acute sewage contamination events. Data collection should include measurement of human pathogens and a suite of viral and bacterial MSTs in wastewater collection systems impacting collected stormwater. Additionally, these targets should be regularly monitored in source waters during system operation to detect potential changes in pathogen load that warrant treatment adjustment (Alja’fari et al., 2022b).

8. Limitations and future needs

8.1. Use of molecular data

As detailed in Section 3, the LRTs were calculated based on dose-response relationships, which require as input a pathogen dose expressed in the unit corresponding to the method used in the dose-response study (Table 1). The calculated LRTs adhered to quality criteria (Section 4.1) that the data used to characterize dose should be expressed in the same units as the dose-response, with exception of roof runoff in the 2021 LRT Update for which bacteria and protozoa data were collected using molecular methods (Alja’fari et al., 2022a). Therefore, additional assumptions were made in the 2021 LRT Update for this source water to harmonize the calculated dose with the dose-response model such that all gene copies were assumed to be infectious.

Among reference pathogens, only Norovirus dose-response was expressed in gene copies, due to the lack of a readily available cultivation method; recent advances using human intestinal enteroids are under development to address this need (Ettayebi et al., 2016, 2021). While clinical trials used to develop dose-response relationships also used molecular detection methods, the number of infectious viruses in the challenge study inocula remains unknown, introducing uncharacterized uncertainty to the dose-response. As outlined by Van Abel et al. (2017), the conservative assumption that all virus gene copies are infectious is the current best practice in exposure assessment given the lack of additional data on infectivity; both the 2017 LRT Guidance and 2021 LRT update adopted this approach. For other viruses in wastewater, new data have been collected that offer insights into the potential infectivity of the genome segments measured by molecular methods (Pecson et al., 2022a). In that study, the ratios of cultivable Enterovirus and adenoviruses to the number of genome copies present in freshly collected raw wastewater spanned from 1:1 to 1:10,000 (adenoviruses) or 1:100,000 (Enterovirus). These findings support the claim that an assumption of a 1:1 ratio may overestimate infectious dose in environmental samples. However, these viruses are biologically distinct from Norovirus and recent dose-response analysis indicates high infectivity of genome copies in sewage-impacted oysters (Teunis et al., 2020).

The 2017 LRT Guidance included Norovirus in the final specification of viral LRTs, given that Norovirus remains the most important waterborne enteric virus in the U.S. accounting for an estimated 1.3 million annual cases of waterborne disease (Collier et al., 2021). Norovirus was not selected as the reference pathogen for viral LRTs in the 2021 LRT Update due to the challenges in dose-response interpretation and the uncertainty surrounding the ratio of genome copies to infectious units. Use of cultivable adenoviruses and Enterovirus eliminated the need for translation between genome copies and infectious units, but uncertainties in dose-response models for adenoviruses also remain given the lack of dose-response data for serotypes 40 and 41 which cause most GI infections, with the majority of dose-response data for serotypes 4 and 7 that result in respiratory infections (Teunis et al., 2016). The 2021 LRTs for adenoviruses were nearly identical to those resulting from the EPA’s SWTR approach, i.e., using Enterovirus occurrence with the Rotavirus dose-response function (Section 3), and within the predicted Norovirus range.

8.2. Epidemiology-based wastewater modeling

The epidemiology-based wastewater modeling method developed by Jahne et al. (2017) and used to simulate the onsite greywater and wastewater pathogen concentrations for Schoen et al. (2017) preceded the more recent rise in this area of study due to the COVID-19 pandemic (Wolfe et al., 2021; Soller et al., 2022). There are likely improvements to the models, as identified by SARS-CoV-2 modeling (Li et al., 2021; Soller et al., 2022), that could be possible with improved data for reference hazards of interest, such as characterization of the shedding rate and duration for symptomatic vs. asymptomatic infections. Wigginton et al. (2021) critically assessed the potential to link wastewater concentrations with infections in the community to support the work of both wastewater utilities and public health partners. The authors conclude that it may be feasible to use wastewater data to inform illness prevalence estimates and, conversely, outbreak data to inform wastewater sampling of peak pathogen concentrations, given that improved disease prevalence and pathogen shedding estimates are developed.

8.3. Risk benchmark selection

Risk benchmarks are generally focused on public health measures and, for water systems, generally based on annualized risks (e.g., 10⁻⁴ infections or 1 μDALY ppy) (Schoen and Garland, 2017). While such annualized risk benchmarks are foundational to overall system design and assessment needs, actual risks may be driven by short-duration hazardous events that can be better managed when event duration (e. g., daily) risk targets are considered for risk management (Signor and Ashbolt, 2009). Examples of short-duration, hazardous events in drinking water distribution include intrusions, cross-connections and backflows, inadequate management of reservoirs, improper main pipe repair, maintenance work, and biofilms (Westrell et al., 2003; Tng et al., 2015). For water reuse systems, additional hazardous events include low-frequency failures in unit processes (i.e., pathogen control barriers), such as loss of membrane integrity or disinfection efficacy (Antony et al., 2016; Buysschaert et al., 2018; Pype et al., 2016; Amoueyan et al., 2019), an area lacking large datasets necessary to describe durations of problematic pathogen concentrations. For regulation of direct potable reuse in CA, a daily infection risk target of 10⁻⁴/365=2.7 × 10⁻⁷ has been considered (SWRCB, 2019).

The studies described herein utilized an annualized 10⁻⁴ infection ppy benchmark as the primary risk target (Sharvelle et al., 2017; Olivieri et al., 2021). The 2017 LRT Guidance also included a lower benchmark of 10⁻² infections ppy, aligning closer to the illness risk associated with U.S. recreational water quality criteria (i.e., 32 GI illnesses per 1000 primary contact recreators) (Schoen et al., 2017; U.S. EPA, 2012). This secondary target is provided to account for the potential difference in acceptable risk among involuntary exposures (e.g., drinking water) vs. voluntary ones (e.g., recreational water); depending on system configuration, use of ONWS may be considered within either category. Although infection- and illness-based risk benchmarks are typical in the U.S., the WHO and Australia utilize DALYs in their water reuse guidelines (NRMMC-EPHC-AHMC, 2006; WHO, 2006). This metric additionally considers the health impact of resulting disease and allows for comparison across diverse hazard types (Sinclair et al., 2015; Petterson and Ashbolt, 2016; Schoen and Garland, 2017). The DALY benchmark has been noted as equivalent to an infection risk of approximately 10⁻³–10⁻⁴ ppy for several enteric pathogens of concern (i.e., Rotavirus, Campylobacter spp., and Cryptosporidium spp.; WHO, 2006).

8.4. Other pathogens

The LRTs discussed herein protect users against infection from enteric pathogens. Other relevant microbial hazards include opportunistic pathogens (e.g., Legionella pneumophila) which grow within collection and distribution systems (Hamilton et al., 2019) and may present unique risks in reclaimed vs. potable water due to differences in water chemistry, treatment processes, and use characteristics (Garner et al., 2018). Since these pathogens typically present in distribution and premise plumbing systems (i.e., post-treatment), treatment requirements were not specified for these organisms. Rather, they should be controlled using best management practices for operation and maintenance of the building water system (Sharvelle et al., 2017). Similarly for harvested rainwater in warm climates, the free-living pathogenic Naegleria fowleri is likely present (Waso et al., 2018), and the microbiota of greywater is dominated by skin-associated bacteria, some of which may be pathogenic, e.g., Staphylococcus aureus (Keely et al., 2015). Hence, for practical purposes, maintenance of disinfectant residuals and biofilm control become the dominating risk management factors. Exposure to reclaimed water may also expose users to antimicrobial resistant bacteria or genes, yet it is currently difficult to assess the treatment requirements for resistant agents given the lack of dose-response data, among other limitations (Garner et al., 2021).

8.5. Additional source waters and end uses

The current LRTs were developed to comprise major water sources and non-potable end uses within large buildings or city districts; however, they are not exhaustive. Building heating, ventilation, and air conditioning (HVAC) systems can also serve as both sources and sinks of building water. Particularly in humid climates, condensation-based AC systems generate large quantities of water that can help meet onsite non-potable water demands (Arden et al., 2021). While fecal contamination is not generally anticipated in atmospheric condensate collections, cooling coils and drip pans provide moist environments conducive to the growth of environmental bacteria and fungi, including opportunistic pathogens such as Legionella and non-tuberculosis mycobacteria. These pathogens are best controlled by system maintenance (e.g., cleaning to remove biofilms) and the use of appropriate biocide treatments as needed (Glawe et al., 2016). As described by Sharvelle et al. (2017), systems utilizing these waters should follow best management practices common to other building water systems, such as maintaining a disinfectant residual. Given the high purity of this water source, corrosion control may also be required to prevent leaching from plumbing materials and fixtures (Glawe et al., 2016). Conversely, building cooling towers use significant quantities of water for evaporative cooling. This end use can utilize non-potable water but documented cooling tower-associated Legionella outbreaks highlight the importance of system maintenance to control for growth of these pathogens (Walser et al., 2014). Indeed, AC condensate provides an ideal source for cooling tower makeup water, given its low dissolved solids and the correlation of condensate production with cooling tower demand (Glawe et al., 2016). Other potential extensions include foundation drainage, as permitted in San Francisco using stormwater LRTs (SFDPH, 2020), and bathing/showering use, for which separate LRTs have been developed (Schoen et al., 2020b).

9. Conclusions

The 2017 LRT Guidance and 2021 LRT Update apply the same risk-based approach to estimate pathogen treatment targets for onsite non-potable water systems, yet differ in certain assumptions. A comparison of the two efforts found:

• Despite different pathogen characterizations for onsite greywater, wastewater and stormwater, the LRTs for each pathogen class were similar between efforts.

• The risk-based approach has the benefits of being transparent, allowing comparison of assumptions and input data, as well as being adaptable to updated information. This flexibility is exemplified by new municipal wastewater data informing higher stormwater virus LRTs in the 2021 LRT Update.

• The roof runoff LRTs support the need for protozoa treatment but remain difficult to characterize due to limited available data on pathogen variability in roof runoff across space and time.

• Future research efforts should focus on pathogen data collection of onsite water sources.

Abbreviations:

DALY

disability adjusted life year

EPA

United States environmental protection agency

FIB

fecal indicator bacteria

HVAC

heating, ventilation, and air conditioning

LOD

limit of detection

LOQ

limit of quantification

LRT

log10-reduction target

MST

microbial source tracking marker

ND

non-detect value

ONWS

onsite non-potable water systems

ppy

per person per year

QMRA

quantitative microbial risk assessment

SWTR

Surface Water Treatment Rule

WHO

World Health Organization

Data availability

Referenced data is contained in the original works cited herein.