Primary and secondary treatment of municipal wastewater contributes to virus removal upstream of advanced purification to produce water for potable reuse. In this study, virus occurrence by cultivable and molecular methods was measured over a 24-month period in raw wastewater influents and secondary effluents from two municipal wastewater treatment plants that together provide the recycled water source for an advanced water purification facility. Using a rank-paired, covariance-based statistical approach, virus log removal values were determined for four wastewater treatment processes that operate in parallel at the two facilities (two activated sludge processes, trickling filter process, and trickling filter/solids contactor process). The trickling filter process exhibited the lowest observed removal of cultivable enteric virus with a median removal of 1.0 log10 (or 90% removal) and a 5th percentile log removal of 0.73 (or 82%), compared to the greatest removal observed for one of the activated sludge processes (median log removal of 2.4 or 99.6% and 5th percentile of 2.1 or 99.2%). Median log removal observed for male-specific (MS) and somatic (SOM) coliphage was 1.8 (98.6% removal) and 0.5 (70%), respectively, for trickling filter and 2.9 (99.9%) and 2.0 (99%) for activated sludge. Thus, coliphage removal was fairly similar to removal observed for cultivable enteric virus. The cultivable enteric virus 5th percentile removal (0.7) from the trickling filter treatment process was proposed to the state regulator for credit towards state requirements for virus removal related to groundwater augmentation with purified recycled water. Receiving pathogen removal credits for secondary wastewater treatment would allow for an improved margin (safety factor) of credits beyond the minimum required; and in this case may also increase the number of viable future groundwater recharge sites closer to drinking water production wells by reducing the underground travel time otherwise required to obtain sufficient credits.
Citation: Polanco JA, Safarik J, Dadakis JS, Johnson C, Plumlee MH (2023) Enteric virus removal by municipal wastewater treatment to achieve requirements for potable reuse. PLOS Water 2(9): e0000052. https://doi.org/10.1371/journal.pwat.0000052
Editor: Silvia Monteiro, Universidade de Lisboa Instituto Superior Tecnico, PORTUGAL
Received: July 15, 2022; Accepted: April 30, 2023; Published: September 27, 2023
Copyright: © 2023 Polanco et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All data in supporting information files.
Funding: Grants and agreement numbers that funded this study: 1) Metropolitan Water District (Future Supply Actions Funding Program), Agreement Number: 189638. Authors who received this award: JD, MHP, CJ, JS, JP 2) Water Research Foundation Subscriber Priority Research Program (2018), Project Funding Agreement Number: 5041 (Project 5041). Authors who received this award: JD, MHP, CJ, JS, JP The authors received no funding from commercial companies for this work. The authors did not receive any salary of other funding from commercial companies for this work. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
As arid regions continue to develop, providing a reliable and resilient source of drinking water to match demand is critically important. Planned potable reuse addresses this challenge by implementing intentional reuse of reclaimed water as a drinking water source. “Indirect” potable reuse (IPR) refers to the use of treated reclaimed water to augment an existing drinking water source that is stored in the environment, i.e. groundwater aquifers or surface water reservoirs [1, 2]. The practice represents an engineered, sustainable solution for water security.
In 2014, the California State Water Resources Control Board (SWRCB) Division of Drinking Water (DDW) finalized regulations for IPR via groundwater recharge. These regulations established the requirement of IPR projects to meet high log reduction values (LRVs) of 12, 10, and 10 for virus, Giardia and Cryptosporidium, respectively (the “12/10/10 rule”) . IPR projects in the state such as the Orange County Water District (OCWD) Groundwater Replenishment System (GWRS) must demonstrate that the required removal for virus (12 logs or 99.9999999999%) and protozoa (10 logs) is met between the source raw wastewater (i.e., untreated influent to the wastewater treatment plant [WWTP]) and the drinking water source (i.e., groundwater produced at drinking water wells downgradient of the recycled water aquifer recharge location) [3–5]. Log removal credits can be assigned to different unit treatment processes that occur between these two points and are summed to meet the requirement .
There are a limited number of datasets available that quantitatively describe the extent of virus removal by primary and secondary wastewater treatment. Two recent studies performed at wastewater treatment facilities near San Diego, CA evaluated the removal of enterovirus, Giardia, and Cryptosporidium [7, 8]. The City of San Diego’s Pathogen Monitoring Study  implemented a long-term sampling plan for their North City Water Reclamation Plant to characterize virus removal and was the first study to obtain virus log removal credits in California. Both studies demonstrated virus log removal where Pecson et al.  proposed a 5th percentile 0.7 log removal value and Bartolo et al.  (for a different facility) proposed a median 2.8 log removal value for enteric viruses for activated sludge wastewater treatment. Despite the general expectation that virus removal occurs during wastewater treatment, removal is not recognized (credited) unless an appropriate site-specific study is conducted and approved to establish a conservative virus log removal value . Obtaining site-specific microbial data and discerning the appropriate statistical approach that quantifies virus removal is the focus of the study reported herein.
The objectives of the present study were to measure virus occurrence and removal during conventional wastewater treatment using cultivable and molecular methods, along with other microbial indicators, and to select and execute an appropriate statistical approach to use the virus data to conservatively calculate the total removal of virus (log removal value) toward potential regulatory credit for potable reuse treatment. This was done by monitoring microbial targets over 24 sampling events in raw wastewater influents and secondary effluents from one wastewater collection agency that operates two municipal wastewater treatment plants. One plant currently serves as the source of recycled water for an advanced water purification facility, with the other plant to begin contributing supply in the near future . While the original objective was to complete the 24 events in a one-year period (covering all seasons), sampling was extended over a two-year period due to the SARS-CoV-2 pandemic. Two different statistical approaches for data analysis were compared as recommended by an independent expert committee [10, 11], including a Monte Carlo simulation-based approach and a rank-paired, covariance-based approach. Microbial concentration data were analyzed using each distinct approach to generate probability distributions and calculate virus log removal values. During the sampling period, routine WWTP operational performance data collected by the facilities was analyzed to establish the operational envelope representative of normal performance and conditions.
Potable reuse projects like the one studied here are highly regulated and scrutinized. Demonstrating additional LRV credits for existing treatment at any treatment stage ranging from wastewater treatment, advanced purification, and including soil aquifer treatment (SAT) increases confidence and would allow for additional margin (safety factor) for potable reuse projects to reliably meet stringent regulatory requirements (i.e., 12/10/10 rule). For some projects such as GWRS, it could also reduce the required underground travel time otherwise required to obtain sufficient credits and therefore increase the number of viable recharge sites at locations closer to drinking water projection wells.
OC San and OCWD facilities description
The Orange County Sanitation District (OC San) is a resource recovery agency that collects, treats, disposes, and recycles wastewater that is generated by approximately 2.6 million people in central and northern Orange County, California. OC San operates two treatment plants, the OC San Reclamation Plant No. 1 (P1) in Fountain Valley and Treatment Plant No. 2 (P2) in Huntington Beach. OCWD operates an Advanced Water Purification Facility (AWPF) co-located with OC San P1 and currently receives essentially all OC San P1 effluent as source water to be recycled for reuse either via advanced treatment for potable reuse or via tertiary treatment for non-potable reuse. The P2 secondary effluent is currently not recycled by OCWD, but the majority will be upon completion of an expansion of the AWPF as part of the GWRS Final Expansion (GWRSFE) in 2023.
The GWRS is a joint project of OC San and OCWD and is an IPR project that advance purifies secondary treated wastewater received from OC San to produce 100 million gallons per day (MGD) of highly purified water from wastewater that would otherwise be discharged to the ocean. Upon completion of GWRSFE, it will produce 130 MGD. The major components of the GWRS include the AWPF, pump stations, and the pipelines that convey the purified recycled water to OCWD’s recharge basins and injection wells. The OCWD AWPF treatment train is comprised of microfiltration (MF), or ultrafiltration (UF) followed by reverse osmosis (RO) and an ultraviolet advanced oxidation process (UV/AOP), and finally decarbonation and lime stabilization. Following advanced treatment, a portion of this highly purified water is injected into a nearby seawater intrusion barrier and the remainder pumped along a 13-mile pipeline for delivery to other OCWD injection wells and finally to spreading basins (percolation ponds) for groundwater replenishment. The AWPF and related groundwater replenishment infrastructure is operated by OCWD to recharge and augment the region’s principal drinking water supply—groundwater stored in the Orange County Groundwater Basin.
With respect to virus removal credits, the AWPF currently receives 2-log removal credits from RO treatment and 6-log removal credits for UV/AOP (8-log total credits). No credit is given for virus removal by MF which occurs before RO, due to the small size of virus relative to larger size of MF pores, nor is credit given for chlorination (chlorine addition before MF to form chloramine for membrane biofouling control). After treatment at the AWPF, GWRS is credited one log virus removal for every one month underground based on travel time between the locations where purified water is infiltrated or injected and the drinking water well production sites. DDW grants this credit due to general expectation of removal of viruses during SAT (conservatively based on Yates et al., 1985 [12, 13]), though it should be noted that during normal operations, any such pathogen occurrence in purified finished water is incredibly unlikely . Currently for GWRS, four months of underground retention time is therefore necessary to meet the 12-log enteric virus removal requirement [3, 12]. The rationale for each treatment step’s virus credit including SAT is outside the scope of this summary but is generally awarded in California (and in other states with similar potable reuse regulatory frameworks) based on a combination of site-specific studies, modeling, and/or established guidance from the state.
OC San sampling locations
A total of six (6) sampling locations from OC San P1 and P2 were monitored. Simplified flow diagrams for the OC San P1 and P2 treatment processes can be found in the S1 Fig. Sampling locations for OC San P1 were raw wastewater influent (raw influent), trickling filter (TF) secondary effluent, activated sludge 1 (AS1) secondary effluent, and activated sludge 2 (AS2) secondary effluent. Raw wastewater from P1 was collected after primary bar-screening but before primary clarification and chemical addition. The first P1 treatment train routes the primary effluent through the TF process followed by secondary clarification. Treated effluent from the TF process after secondary clarification was sampled. P1 primary effluent is also sent through two parallel activated sludge (AS) treatment trains, designated separately as AS1 and AS2. Secondary effluent samples were taken from each AS process following secondary clarification. Both AS trains operate in nitrification-partial denitrification (NDN) mode with the major difference between the AS1 and AS2 processes being that AS1 does not receive mixed liquor return, while the newer AS2 facility does receive it.
OC San P2 sampling locations were limited to raw influent and trickling filter/solids contactor (TF/SC) secondary effluent (i.e., following secondary clarification), since only P2 TF/SC secondary effluent from this plant will be supplied to GWRS in 2023 as a result of the GWRSFE and blended with the secondary effluent supplied from P1. OC San P2 treats wastewater from the western and coastal parts of the OC San service area along with the Santa Ana Regional Interceptor (SARI) trunkline (brine line), the centrate flows from thickening and dewatering of biosolids from P1 and P2, and other side stream flows. For the present study, raw influent for P2 was collected after preliminary screening but before chemical addition, primary clarification, and centrate return flows.
Plant performance and operational parameters
Operating range values, or ORVs, were determined and defined in this study as parameter values that describe typical and normal treatment process operations at the WWTPs as observed during this study’s microbial sampling period. The purpose of developing these ORVs is that any future awarded credit for virus removal at the WWTPs toward the 12-log requirement is contingent on the facilities meeting the ORVs, based on a 30-day average performance that is compared to study-proposed lower or upper baseline ORVs.
ORVs were derived for each secondary treatment process supplying influent to OCWD AWPF, i.e., P1 TF, AS1, AS2 and P2 TF/SC for key plant operational parameters (namely, mean cell residence time (MCRT) for AS1, AS2, and TF/SC, and total biological oxygen demand [BOD-T] for TF). An exceedance of a designated ORV represents a deviation from normal treatment performance. In other words, an exceedance would indicate that the WWTP is operating outside the normal operating envelope, and virus log credit would not be granted for that period. This deviation could be related to an unexpected event but also due to planned activities such as operational maintenance of treatment systems or flow adjustments by the OC San operators. For detailed information on ORVs and values calculated for this study, (see S7 Text and S3 Table).
Microbial targets and sample analysis
Raw influent samples and secondary effluent samples were analyzed for enteric viruses, male-specific (MS) coliphage, somatic (SOM) coliphage, and total and fecal coliform. To analyze samples for enteric viruses, samples were split and analyzed using culture-based and molecular detection methods as described in EPA Method 1615 . For MS and SOM coliphage enumeration, two-liter grab samples were collected in sterile polypropylene Nalgene bottles and processed using EPA Method 1602 with double agar layer modification . For total and fecal coliform analysis, 100 mL grab samples were collected in sterile IDEXX bottles and enumerated according to standard methods 9222B and 9222D, respectively . Details on methods used for samples analysis, including minor modifications to EPA 1615 method for molecular detection of viruses is provided in Polanco et al. .
Raw influent samples.
Raw influent grab samples were collected in two, 1-liter sterile-autoclaved high-density polyethylene bottles. The sample tap was flushed for 2–3 minutes before sample collection. Care was taken to ensure no physical contact was made between the sample bottle and sample tap. Samples were placed on ice for shipment and processing. Grab samples were split after PEG concentration for culture and molecular assays .
Secondary effluent samples were collected by filtering 25 liters through a single-use dead-end polysulfone hollow-fiber ultrafilter (Rexeed-25 S ultrafilter kit by Innovaprep). Filtration was performed on-site according to manufacturer’s instructions. All tubing, fittings, and caps were sanitized and autoclaved in a controlled laboratory setting. Using aseptic technique, a pressure gauge and peristaltic pump were installed ahead of the hollow fiber filter to pump and monitor inlet water pressure. Filtration volume (permeate) was monitored with a flow meter installed on permeate tubing. After filtration, ultrafilters were carefully unmounted, re-capped, wiped with a 10% solution of bleach and preserved on ice.
Secondary effluent concentrates were extracted from each ultrafilter using a high-volume elution canister consisting of pressurized 0.075% Tween 20 phosphate buffer solution. Ultrafilter eluents were filtered through sterile 0.22 μm pore filter and divided into a series of subsamples for culture-based enteric virus analysis and for DNA extraction (molecular detection) . Detailed information on the ultrafiltration procedure can be found in the S2 Text.
A total of 24 sampling events were completed over the course of the study. Samples were collected from each of the six sampling locations and spanned between one to three days constituting one sampling event. On average, two sampling events were scheduled per month and spanned a total of 24 months due to pandemic-related postponement of sampling. Duplicate samples were also collected for all microbial targets on a staggered schedule at each site. With exception to the P2 raw influent duplicate sample, which was collected once, a total of two duplicate samples were collected for all sites during the study. Grab samples for coliphage evaluation were collected once a month from all six locations contemporaneously with enteric virus samples, such that a total of 12 coliphage sampling events were reported over the study. All samples were shipped and processed immediately to comply with sample hold time criteria and quality assurance.
No attempt was made to time collection of the effluent versus influent samples according to the average system hydraulic residence time, i.e. Lagrangian sampling, since this is generally difficult and unreliable and because the anticipated data analysis method to determine log removal did not require it. Both of the data analysis approaches utilized for this study calculate removal of virus based on a statistical evaluation of the observed influent and effluent distribution of concentrations. Considering the use of a statistical analysis that utilizes the full sampling distribution may be considered superior to same day influent-effluent pairing .
To determine virus recovery efficiency, matrix spike recovery (MSR) samples were collected contemporaneously with a subset of the enteric virus and coliphage samples. A total of 9 raw influent MSR samples were collected per raw influent sampling location over the 24 total samples. For each secondary effluent sampling location, five MSR samples were collected, which given four secondary effluent sampling locations resulted in 20 total secondary effluent MSR samples over the course of the study. Male-specific (MS) coliphage (ATCC 15597-B1), somatic (SOM) coliphage (ATCC 13706-B1) and poliovirus (ATCC VR1562) of known concentration were spiked into raw influent samples directly and into the ultrafiltration unit prior to elution to determine and evaluate method recovery. For ddPCR recoveries, armored RNA targets for enterovirus and norovirus GII were spiked into raw influent samples and into the ultrafiltration unit at known concentrations. Additional information on how MSR samples were processed can be found in the (see S3 and S4 Texts).
Determination of median and 5th percentile virus removal
Based on precedent from past California studies, DDW recognizes the 5th percentile LRV as the preferred statistically conservative removal value on which to base an awarded pathogen credit for the particular treatment process, as opposed to the 50th percentile or mean [3, 8]. To calculate 5th percentile LRVs, two mathematical approaches were used and compared using enteric virus and coliphage microbial concentration data for each respective secondary treatment process from OC San P1 and P2.
These two approaches consisted of a rank-paired, covariance-based approach and a Monte Carlo simulation-based approach. Although OCWD GWRS receives a blend of secondary effluent from OC San P1 treatment processes (and by 2023 also OC San P2 treatment process), log removal of enteric viruses and coliphage were determined by separately calculating the 5th percentile LRV for each distinct secondary effluent treatment process from OC San P1 and P2 and not the final blended effluent. In other words, after trialing some potential data amalgamation approaches for the influent and effluent distributions, it was decided not to attempt to combine the observed separate plant datasets to determine one “overall” 5th percentile LRV as a means to represent the multiple wastewater treatment processes preceding the AWPF. It was recognized that instead, and more simply, virus LRV credit from wastewater treatment could be conservatively proposed to be based on the lowest observed individual treatment process enteric virus LRV. In other words, the lowest 5th percentile LRV amongst the four wastewater treatment processes studied was proposed as the credit value toward the potable reuse 12-log virus requirement.
For the rank-paired, covariance-based approach as described by , influent and effluent data from the WWTP are assumed to be dependent (correlated). The approach uses an estimate of sample means followed by the calculated standard deviations of sample covariance derived from two correlated random normal distributions, namely, the raw influent and effluent virus concentration distributions. Microbial concentration data were analyzed using Microsoft Excel software. For both influent and effluent distributions, each sample result (i.e., virus concentration) is first log-transformed, and the average LRV (μLRV) is calculated as shown in Eq 1: (1)
Where μinf is the average log-value for the influent distribution and μeff is the average log-value for the effluent distribution.
To determine covariance, the log-transformed influent and effluent data are sorted from smallest to largest. These ranked values from the influent distribution are paired with ranked values from the effluent distribution (i.e., rank paired) in order to calculate rank-paired sample covariance. To determine if two distributions are covariant, the number of entries from each distribution must be the same, that is, the number of influent samples must equal the number of effluent samples. Eq 2 was then used to estimate the LRV probability distribution using the average LRV term calculated in Eq 1, the z-score statistic, and the standard deviation of sample covariance: (2)
Where μLRV is the average LRV calculated in Eq 1, Z represents the z-score or standard score for a given probability such as the 5th percentile, and σvar-Δ is the standard deviation of sample covariance between the influent and effluent distributions.
The standard deviation of sample covariance (σvar-Δ) is calculated using Eq 3, below, which requires solving for the variance of the influent distribution, variance of the effluent distribution, and the covariance between both influent and effluent distributions. These parameters can be calculated using Microsoft Excel’s variance and covariance.s formulas, respectively, for each respective log-transformed distribution.(3)
Where var(Inf) is the variance of the influent distribution, var(Eff) is the variance of the effluent distribution, and cov(Inf,Eff) is the covariance correlation between both the influent and effluent distributions.
Monte Carlo simulation
The second statistical approach used to calculate 5th percentile LRVs was the Monte Carlo simulation approach . The Monte Carlo simulation approach involves performing a statistical analysis of log-normal distributions from both the raw influent and secondary effluent microbial concentration data to generate a probability distribution model of log removal values for a given microbial target. Since the Monte Carlo simulation estimates the statistical parameters from two independent log-normal variables, any result from the Monte Carlo simulation represents an approximation of the analytical solution defined by the Normality of the LRV calculation . For detailed information on this approach, refer to S8 Text. A key modification of the Monte Carlo approach for the present study was the censoring of negative LRVs observed at the very low percentiles of the resulting LRV distribution before determining the 5th percentile LRV that could be proposed for regulatory crediting. This modification was executed based on the recognition that negative removal (human enteric virus formation) is not possible during wastewater treatment. Hence, herein the method is referred to as “modified” Monte Carlo approach. The modification is discussed further in the later discussion, as well as the (see S8 Text) .
Data analysis and probability distributions
Microbial concentration data were plotted using the ggplot2 software in R. Microbial probability distribution plots and LRV plots (i.e., quantile plots) were plotted using the KaleidaGraph Version 4.5.4 Software by Synergy Software using traditional parameters as described in literature [8, 22, 23]. For each sampling site, the distribution of concentration values was modeled with a best-fit line. Each distribution was tested for outliers using the Dixon’s Q test and Grubbs statistical outlier tests. Finally, all distributions were tested for lognormal fitness using Contchart Software. Finally, to obtain percentile rank data, including 5th percentile values, the Microsoft Excel percentile formula was used. Prior to applying the formula, log removal values were converted to percent removal which is necessary for accurate percentile calculation . As noted above, to plot the LRV distribution as calculated by the covariance analysis method, the 5th percentile, median and 95th percentile values were plotted on a probability plot. For the Monte Carlo simulation method, a probability plot for the LRV distribution was plotted using all simulated (n = 10,000) LRVs.
Microbial concentration data
Fig 1 summarizes microbial concentration data for cultivable enteric virus, enterovirus and norovirus GII targets, MS coliphage, SOM coliphage, and total and fecal coliforms (Fig 1). Cultivable enteric virus, MS coliphage, and SOM coliphage concentrations were all above the detection limit and were well within the method virus acceptance criteria. Concentration data for cultivable enteric virus, MS coliphage and SOM coliphage were corrected by virus recovery as described in the S4 Text . In contrast, enterovirus and norovirus GII molecular concentration data were reported as non-detect for a number of secondary effluent samples (see S1 Table) and poor recovery was oftentimes observed for influent and secondary effluent samples. Due to poor virus recoveries from molecular methods (recoveries were outside of acceptance criteria), non-detect values and statistical outliers in MSR experiments, native enterovirus and norovirus GII concentrations were not corrected by recovery. Finally, coliform data were not adjusted for recovery as these data served as standard microbial indicator controls throughout the study. Detailed statistical attributes for all microbial concentrations can be found in the S2 Table.
Cultivable enteric virus data (A) were corrected for virus recovery and obtained via EPA 1615 cultivable infectivity assay and are reported as most probable number per liter (MPN/L). Gene copy detections of enterovirus and norovirus GII (B) obtained via droplet digital Polymerase Chain-Reaction (ddPCR) are also shown. Male-Specific and somatic coliphage data (C) obtained via EPA 1602 were also corrected by virus recovery and are reported as plaque forming units per liter (PFU/L). Fecal and total coliform data (D) were enumerated using Standard Methods 9222B and 9222D, respectively, and are reported as colonies per milliliter. Raw influent concentrations are shown in gray, and secondary effluent concentrations are shown in dark green (P1 TF), dark blue (P1 AS1), light blue (P1 AS2), and light green (P2 TF/SC).
Cultivable enteric virus concentrations observed for P1 raw influent (5.5x101 to 4.5x104 MPN/L) were often greater than P2 (5.4.101 to 1.6x104 MPN/L), as indicated by the cultivable enteric virus geometric mean concentration that is greater for P1 (1.3x103 MPN/L) than P2 (7.5x102 MPN/L), but not significantly different (S2 Table). P1 TF effluent geometric mean concentration for cultivable enteric viruses was greater than the other effluents at 1.3x102 MPN/L, compared to 4.9, 1.3x101, and 3.7x101 MPN/L for P1 AS1, P1 AS2, and P2 TF/SC effluents, respectively (S2 Table). Most importantly, data obtained by the cultivable infectivity assay show that cultured enteric viruses are significantly removed at all monitored treatment facilities. When plotted chronologically by sampling date, enteric virus removal is observed throughout all sampling dates (see S2 Fig), despite some variation in the amount of removal by each treatment process (Fig 1).
Uncorrected molecular concentration data show that the raw influent enterovirus concentration range was similar between P1 (1.8x104 to 1.8x106 GC/L) and P2 (1.0x104 to 2.7x106 GC/L). Similarly, norovirus GII concentrations were relatively similar between P1 (4.7x104 to 3.8x106) and P2 (3.4x104 to 3.8x106). For both P1 and P2, raw influent norovirus GII geometric mean concentrations were higher than enterovirus. With respect to secondary effluent concentrations, both enterovirus and norovirus GII geometric mean concentrations were highest in P1 TF effluent followed by P2 TF/SC effluent, and the P1 AS1 and P1 AS2 effluents (S2 Table). All secondary effluent sites demonstrated removal of virus genetic material, where (as with cultivable enteric virus) P1 TF showed the least removal.
MS and SOM coliphage concentrations for raw influent were similar between P1 and P2, with a wider range of concentrations observed for MS coliphage compared to SOM coliphage (S2 Table). While both MS and SOM coliphage concentrations for secondary effluent varied by process, all processes also showed removal of both MS and SOM coliphage. AS1 and AS2 treatment processes at P1 show the lowest coliphage geometric mean concentrations (highest removal), while P1 TF effluent demonstrated the greatest mean concentrations (S2 Table).
Finally, fecal coliform concentration ranges for raw influent were also similar between P1 and P2, as were the total coliform concentration ranges (S2 Table). Total coliform geometric mean concentration was greater than fecal coliform for both plants’ raw influent. After secondary treatment at P1 and P2, total and fecal coliform demonstrated a reduction in concentration. P1 TF effluent exhibited the greatest geometric mean concentration of all secondary effluent sites for total and fecal coliform (least removal), followed by P2 TF/SC effluent, P1 AS2 effluent, and P1 AS1 effluent.
Virus recovery and data corrections
The average cultivable enteric virus recovery for P1 and P2 raw influents was 99% and 103%, respectively (S1 Table). Average cultivable enteric virus recovery for secondary effluent samples varied by treatment process (49% for TF, 64.5% for AS1, 59.6% for AS2, and 66% for TF/SC), but were largely similar. Native cultivable enteric virus concentrations in raw influent were corrected by the average of the cultivable recovery of MS coliphage, SOM coliphage, and poliovirus measurements for each plant (99% and 103% for P1 and P2, respectively) similar to the approach by Pecson, et al. . Individual method recoveries are tabulated in S1 Table. For secondary effluents, recovery data from the same three targets were pooled to calculate a single average recovery value which was then used to correct native concentrations. In other words, the average recovery of all P1 secondary effluent treatment processes was used to correct all P1 secondary effluent data (58%). The P2 secondary effluent data was separately corrected by the average recovery (66%). Finally, coliphage concentration data were corrected using the average cultivable recovery of MS coliphage and SOM coliphage only. Overall, these secondary effluent virus recovery results were significantly lower than the average raw influent recovery, therefore requiring a larger correction factor to be applied to the dataset. Since raw influent virus recovery was near ~100% compared to secondary effluent near ~60%, correcting for observed recovery had the conservative impact of reducing the calculated removal (LRV).
Notably, recovery values obtained from the molecular assay (ddPCR) were inconsistent for both enterovirus and norovirus GII targets. Norovirus GII recoveries were highly variable, ranging from frequent instances of 0% recovery to up to 180% for raw influent and up to 20% for secondary effluent. A limited number of extremely high outliers (>6,000%) were also observed in P1 raw influent (S1 Table). While there were fewer outlier values for enterovirus recovery, there were also frequent instances of very low recovery values. Reasons for the wide recovery performance are unclear but may be related to potential incompatibility of use of the ddPCR analysis with armored RNA targets which are traditionally used as an inhibition control in EPA 1615 method using qPCR (S5 Text). As a result, native molecular concentration data were not corrected by these measured recoveries. Thus, uncorrected native ddPCR data was used for log removal estimations.
Virus log removal
Raw influent and secondary effluent probability plots for cultivable enteric virus concentration data obtained from P1 and P2 are shown in Fig 2 using a log-scale y-axis. Probability plots for molecular targets and coliphage targets can be found in the supporting information. All distributions are best modeled with an exponential best-fit line suggesting that all data are lognormally distributed. The distribution of raw influent cultivable enteric virus concentration data (plotted as black circles) was consistently higher than secondary effluent concentration data collected at all percentiles for P1 and P2. The difference between the raw influent distribution and each respective secondary effluent distribution represents removal of enteric viruses. LRV distributions obtained by the rank-paired covariance approach and the Monte Carlo simulation approach are shown in Fig 3.
Probability distributions for cultivable enteric virus concentrations obtained from raw influent and secondary effluent samples taken at OC San P1 (left) and P2 (right). Each point represents one sampling event and the solid line represents a best-fit regression. The coefficient of determination (R2 value) is also shown. Raw influent and secondary effluent cultivable enteric virus data obtained from both P1 and P2 are lognormally distributed.
5th, 50th, and 95th percentile LRVs obtained using the rank-paired covariance approach are shown (left). All calculated LRVs obtained using the Modified Monte Carlo approach are shown (right).
Using the covariance approach (Fig 3, left panel), the median LRV for cultivable enteric viruses at P1 TF and P2 TF/SC was 1.0 and 1.3 (or 90 and 95%), respectively. The median LRVs for P1 AS1 and P1 AS2 were greater at 2.4 and 2.0 (or 99.6 and 99.0%), respectively. Using the modified Monte Carlo approach, the median LRV was similar to the covariance approach at 1.2 and 1.4 for cultivable enteric viruses at P1 TF and P2 TF/SC, respectively, and 2.4 and 2.0 for P1 AS1 and P1 AS2, respectively.
While the median LRVs were found to be similar between the two calculation approaches across the different treatment processes, the 5th percentile LRV is greater for the covariance approach, e.g., 0.73 (82%) compared to the modified Monte Carlo approach at 0.18 (33%) for the cultivable enteric virus measured for P1 TF. Using the covariance approach, 5th percentile LRV for cultivable enteric virus at P2 TF/SC, P1 AS1 and P1 AS2 were 1.1 (92%), 2.1 (99.2%), and 1.5 (97%), respectively. When compared to the modified Monte Carlo approach, P2 TF/SC, P1 AS1 and P1 AS2 5th percentile LRVs were 0.3 (45%), 1.0 (91%), and 0.7 (82%). This is consistent with the greater steepness of the LRV distribution for the modified Monte Carlo approach. LRVs calculated for each microorganism target using both calculation methods are also reported in Table 1.
The lower 5th percentile LRVs for modified Monte Carlo is due to the random pairing of relatively high effluent concentration values with unrelated low influent values (see S8 Text). Due to the overlapping concentration ranges observed for this study for some treatment processes and microbial targets over the 24 sampling events (Fig 2), sometimes the modified Monte Carlo simulation randomly pairs raw influent and secondary effluent concentration values that are quite similar, and these cases produce the lowest LRVs in the distribution which drives the low 5th percentile value.
Conversely, the higher 5th percentile LRV observed for the covariance approach is related to how the rank paired influent and effluent concentration values are correlated (dependent) . The covariance method generates its LRV probability distribution by pairing influent and effluent concentration data by simple rank order. To test dataset correlations, each raw influent distribution and each respective secondary effluent distribution were modeled with a linear trendline to calculate Pearson’s correlation (R2) coefficient of determination (see supporting information). This test indicated that the distribution of values observed in raw influent are related to the distributions of values seen in each distinct secondary effluent process (S4 Table).
LRVs for each microbial target were tabulated for all secondary effluent treatment processes (Table 1). For LRVs obtained for cultivable enteric virus and all other microbial targets, P1 TF and P2 TF/SC processes exhibit lower median and 5th percentile LRVs than P1 AS1 and AS2 using the covariance approach. In addition, P1 AS1 and AS2 show similar LRVs for any given microbial target. Overall, the P1 TF process typically had the lowest 5th percentile LRV. This result is also true for LRVs calculated using the modified Monte Carlo method, with exception to MS coliphage for P1 TF/SC, which showed marginally higher removal than P2 TF/SC (Table 1). Overall, P1 TF effluent exhibited higher microbial target concentrations for all targets compared to the other three effluents (Table 1), which is consistent with the lower LRV calculated and the general degree of treatment expected from each process.
LRVs calculated from the molecular dataset (ddPCR) were similar to those LRVs obtained with the cultivable enteric virus dataset for most treatment processes, regardless of the calculation approach and despite the inherent differences in the methodological approach of each assay. This is readily apparent in Fig 4, which shows calculated LRVs along a bar plot comparing the microbial target (and associated method) for each secondary treatment process. The lowest median LRV observed for the molecular dataset was P1 TF (1.6 and 0.9 for enterovirus and norovirus GII, respectively) and P2 TF/SC (1.8 and 1.1 for enterovirus and norovirus GII, respectively). LRVs for enterovirus and norovirus GII were higher for P1 AS1 (2.1 and 2.0) and AS2 (2.2 and 1.7). This trend is similar to those LRVs calculated for the culture-based assay (Fig 4). Regardless of the calculation approach, both median and 5th percentile LRVs for norovirus GII LRV were less than or equal to cultivable enteric virus LRV for all treatment processes except for the 5th percentile LRV for P1 AS2, which was slightly higher (Fig 4). Conversely, both median and 5th percentile LRVs for enterovirus ddPCR were typically slightly higher than the 5th percentile cultivable enteric virus LRV, except for P1 AS1.
Lastly, it should be noted that while there are significant differences in LRVs between microbial targets (Table 1 and Fig 4), LRVs with differences between +/- 0.4 (e.g., 1.0 versus 0.7) are not large when considering the result in terms of percent removal, where 1.0 and 0.7 LRV correspond to 90% and 80%, respectively. This is especially true when comparing larger LRVs, such as 2.0 and 1.5 LRVs, which correspond to a 99% and 97% removal. Thus, apparent differences in removal between different microbial targets or treatment processes in this study are not always significant, which should be considered during interpretation. Thus, apparent differences in removal between different microbial targets or treatment processes in this study may not be significant.
Selection of a conservative virus LRV
To evaluate virus log removal during wastewater treatment, various microbial targets including enteric viruses, MS coliphage and SOM coliphage were monitored over a 2-year period consisting of 24 sampling events. Microbial concentration data measured from each of the four secondary treatment processes were enumerated with both cultivable and molecular methods and virus log removal was calculated using two mathematical approaches including a rank-paired covariance-based approach and a modified Monte Carlo simulation approach. Median LRVs obtained using each approach were very similar for all microbial targets, suggesting that despite the analytical differences between the two methods, virus removal can be evaluated for each of the treatment processes using both approaches (Table 1). However, in contrast to the observed median values, 5th percentile LRVs varied with each approach where Monte Carlo LRV results were typically lower when compared to the covariance approach. This is of interest given the regulatory precedent in California for using the 5th percentile LRV for treatment unit process crediting.
Treatment processes that demonstrated the least microbial removal drove selection of the most conservative virus removal estimate (5th percentile LRV) for choice of potential regulatory credit value. LRVs reported here show that both P1 TF and P2 TF/SC treatment processes consistently removed all virus targets less efficiently than the two AS processes. In fact, this reduced efficiency of removal highlighted a limitation in the Monte Carlo approach when estimating LRVs at low-end percentiles for these two processes. Very low calculated LRVs at the low percentile ranks for these lower performing treatment processes are believed to be a mathematical artifact of the Monte Carlo’s random pairing of influent-effluent values (see Monte Carlo approach in supporting information) and are therefore a potential weakness of the approach. In the case of P1 TF and P2 TF/SC performance, this is illustrated by the overlap in concentration values observed between the raw influent distribution and secondary effluent distributions. As a result, 5th percentile LRV estimations using the Monte Carlo approach were determined to not be representative of the actual low-end virus log removal observed for all four treatment processes (Table 1). In other words, the Monte Carlo simulation approach (with or without censoring to remove excessively low end LRVs such as negative values observed at low percentiles) may be overly conservative and not reflective of the true low-end plant performance for virus removal. More work is required to confirm this observation and to continue to develop a superior statistical and/or sampling method.
In contrast, the covariance approach generally had higher 5th percentile LRVs for microbial targets compared to the modified Monte Carlo and underscores the underlying differences between each calculation approach. The primary difference with the rank-paired covariance analysis approach lies with the correlative relationship of the raw influent and secondary effluent distributions, in contrast to the Monte Carlo approach which assumes each distribution is mutually independent and distinct. Given the observed lack of any negative LRV pairings for the rank-paired covariance method, no censoring of LRV results was necessary (see supporting information). Therefore, for the present study and proposal of potential regulatory credit, the covariance method was deemed more appropriate for quantitating the differences (virus reductions) between said distributions and determining a 5th percentile LRV. Future work may determine a more meaningful method for determining covariance beyond simple rank-pairing used in this study. Based on the covariance approach, a conservative 5th percentile LRV of 0.73 (82% removal) was the basis for proposed virus log removal credit to California DDW for the four wastewater treatment processes monitored in this study; DDW review is underway at the time of this report. Enterovirus (ddPCR), norovirus GII (ddPCR), and MS and SOM coliphage were not utilized as the basis for the proposed credit given prior precedent in California for use of cultivable enteric virus for other wastewater treatment plant assessment.
Microbiological method testing performance
While microbial detection via molecular methods (e.g., qPCR, ddPCR) holds promise, it is still unclear whether removal of gene copy number across treatment is sufficient to assess performance compared to removal of cultivable enteric virus targets given the lack of correlation in concentration between the two methods observed in this study (analysis reported in Polanco et al. 2022) . This may be related to the fact that virus assessment using molecular assays can include both viable and non-viable virus–though in the present study, lack of correlation may also be related to the inconsistent recovery for the ddPCR targets. While correlative analyses can be performed between cultivable and molecular assays, the fundamental differences between these assays (i.e. testing for biological activity vs. detection of genetic material), suggests that these assays cannot be compared using a simple correlational analysis and requires additional investigation.
Other studies have shown greater removal of genetic material over cultivable enteric virus at the 5th percentile and mean LRV [8, 26]. This could be related to the present study’s use of ddPCR instead of qPCR as well as the lack of recovery correction for the ddPCR results (see S3 Text). It is clear that further work is needed to establish the relationship between these assays to understand how targeted molecular analysis informs virus viability and quantitation.
Armored RNA was selected for this study based on its long history of use for quality control purposes in the EPA 1615 method, though not particularly used for determining recovery [18, 27]. Recovery results for norovirus GII exhibited major inconsistencies, which included large value outliers, and necessitated the decision to not correct the native dataset. In addition, molecular recovery data for both norovirus GII and enterovirus were often negative (i.e., spiked sample had less target than the paired native sample) or near zero (see S1 Table). As a result, it was not acceptable to use these recovery data to correct native results given a minimum acceptance criterion of 5% recovery. More research is needed to further understand how detection of armored RNA may be affected by sample-matrix interference to inform virus target recovery.
Treatment performance and sampling approach
Higher LRVs were observed for activated sludge treatment over TF or TF/SC. In fact, regardless of the statistical method used to calculate LRV (rank paired covariance or Monte Carlo), all treatment processes ranked similarly for microbial removal (Table 1), i.e. AS > TF/SC > TF. This is a notable observation that could inform the choice of sampling approach and/or LRV calculation method for future similar studies because it indicates that the conservative calculation approach (i.e., conservative meaning lower LRVs for regulatory credit), namely the Monte Carlo analysis, may be acceptable to evaluate performance of activated sludge processes depending on how much additional credit is being sought.
In this study, the covariance approach 5th percentile LRV of 0.73 (82% removal) observed for the process with the lowest LRV (OC San P1 TF) was ultimately proposed as the credit value for the OCWD GWRS potable reuse treatment facility. This is because P1 TF represented a conservative choice that avoided the need to combine the LRV datasets of the four effluents that blend to serve as influent to GWRS toward a single LRV credit value. However, reliance on strictly the P1 TF process significantly reduces the overall credit value for wastewater treatment given the much higher observed 5th percentile LRV for AS1 and AS2, essentially under-crediting the GWRS wastewater treatment process.
Future studies that seek to evaluate enteric virus log removal for wastewater treatment (i.e., secondary or tertiary treated wastewater effluent) can benefit from the strategies used in this study to further develop and execute an efficient microbial monitoring study for log removal crediting. While the covariance-based approach was ultimately used in the present study to calculate a proposed LRV credit for wastewater treatment (currently under review by regulator), future research should seek to further develop and advance appropriate statistical and/or sampling techniques for determining conservative (e.g., 5th percentile) LRVs, based on the knowledge that true virus removal across wastewater treatment is likely significant (e.g., this study found median removals of 2.0+ LRV or 99%+ for activated sludge processes and 1.0 to 1.3 LRV or 90 to 95% for trickling filter-based processes). Methods must consider that concentration data for a target like cultivable enteric virus will be highly variable, which could be addressed through study sampling design (e.g., residence-time paired sampling; MSRs for every sample grab) or by considering the use of a more stable virus indicator such as MS and SOM coliphage instead of cultivable enteric virus, or by developing a more appropriate statistical method. In this study and others, coliphage concentration data have been shown to be much more stable in both the influent and wastewater effluent samples, suggesting that MS and SOM coliphage can be reliable indicators for virus removal when using culture-based assays.
Addressing sample variability
The observed variability in cultivable enteric virus occurrence in both OC San P1 and P2 influent and effluent is likely real given the generally good performance of duplicate samples, as opposed to being a result of analytical method performance issues. Nevertheless, an improved study design could include duplicates or triplicates at every sampling event, such that the average used to represent that event may dampen any analytical variability and thus provide a more accurate measurement for the calculation of conservative (e.g., 5th percentile) LRV. If duplicates/triplicates are very consistent, they could be tapered off as the study continues.
In contrast to calculating LRV from same-event paired influent/effluent concentration values (which makes the dubious assumption that the same water parcel was sampled), the Monte Carlo method fundamentally does not require residence-time-paired samples because it assumes no relationship between the influent and effluent distribution and randomly chooses sample pairs from along each distribution of all results across the different days sampled. When the target constituent (in this case cultivable enteric virus) is highly variable in the influent for the different days sampled, as seen in this study, the Monte Carlo LRV distribution will therefore be quite steep with very low LRVs on the low end of the LRV distribution (i.e., the low/conservative percentiles) based on random pairings of unrelated high-effluent samples with low-influent samples. With the requirement to take the 5th percentile LRV, the proposed credit value will be driven by these cases and be very low.
The use of the 5th percentile for regulatory crediting purposes assumes that the 5th percentile estimate is the true low-end performance of the treatment process. However, based on this study, this may not always be true for treatment processes that generally remove less virus particles since, in this case, influent and effluent distributions can overlap. When this occurs, the very low 5th percentile LRVs within the distribution are improbable due to non-meaningful influent/effluent pairings in the LRV calculation of the Monte Carlo simulation, thereby not representing the true range of treatment removal (i.e., these values are exceedingly low). Nevertheless, this conservative approach may be acceptable for treatment processes with high virus removal, such as the activated sludge processes studied here, where at least ~1.0 virus log credit at 5th percentile is likely attainable. However, it appears to be problematic for non-NDN processes where the proposed credit value may approach zero, as observed in this study. Hence, the covariance-based approach was favored here since it is a fundamentally different statistical approach that does not randomly pair influent-effluent concentrations.
Hydraulic residence time and composite sampling
Collecting samples that are appropriately staggered in time by the average hydraulic residence time of the process to allow LRVs calculated from same-day residence-time-paired influent/effluent samples may result in more accurate (and potentially higher) log removal estimate. However, this is very difficult to execute correctly in practice due to the long hydraulic residence times of a wastewater treatment plant, which will be difficult to estimate accurately. Furthermore, it is possible that return flows from sidestreams will influence final concentration results, complicating the attempt to sample the same water parcel from influent-to-effluent. Another related complication is that the influent sample likely utilizes a simple grab sample (an instantaneous timepoint) versus the effluent sample likely requires on-site ultrafiltration (a multi-hour composite), depending on the microbial target.
To attempt to address these challenges, future work could consider focusing on individual unit processes of a wastewater treatment plant expected to demonstrate the highest virus removal, in addition to or instead of sampling for removal from raw influent compared to final secondary or tertiary effluent, because an individual unit process will have a shorter residence time potentially reducing some of this uncertainty and perhaps the influent variability. It may be possible to optimize the sampling approach so that influent and effluent are both grabs or are both composites. Further, pre-studies could characterize the variability in virus concentration over timescales relevant to the residence time to support study design (e.g., if the estimated residence time is some number of hours, and there is lack of significant microbial target concentration variability in that timeframe, it is not necessary to stagger influent/effluent sample collection by residence time). An appropriate statistical method to calculate LRV from residence-time-paired influent/effluent samples would need to be developed, such as based on simple influent/effluent pairing (using the lowest or 5th percentile observed LRV as proposed credit value) or a covariance-based approach with pairing based on each collected sample set.
S1 Fig. OC San simplified flow diagrams.
Simplified flow diagrams illustrating secondary treatment trains and the study sampling locations. (A) Two parallel secondary treatment trains at OC San Reclamation Plant No. 1. Plant secondary clarifiers that follow TF, AS1, and AS2 have engineering differences illustrated simplistically in the diagram. (B) Trickling Filter/Solids Contactor (TF/SC) secondary treatment train at OC San Treatment Plant No. 2.
S2 Fig. Concentrations of enteric viruses observed for raw influent and secondary effluent using a modified culture infectivity assay (EPA 1615).
Data shown above have been corrected for virus recovery and include data from OC San P1 (left panel) and OC San P2 (right panel). The gap in sampling corresponds a study interruption due to the global pandemic. Raw wastewater influent data points are shown as black dots while secondary effluent data are shown in dark green (P1 TF), dark blue (P1 AS1), light blue (P1 AS2), and light green (P2 TF/SC).
S3 Fig. Concentrations of enterovirus and norovirus GII (ddPCR).
Gene copy detections of enterovirus (top row) and norovirus GII (bottom row) using droplet digital Polymerase Chain-Reaction (ddPCR) for raw and secondary wastewater from OC San Plant No. 1 (left panels) and OC San Plant No. 2 (right panels). The gap in sampling corresponds to a study interruption due to the SARS-CoV-2 pandemic.
S4 Fig. Probability distributions for enterovirus and norovirus GII concentrations from droplet digital PCR (ddPCR) analysis of raw influent and secondary effluent samples taken at OC San P1 and P2.
Each point represents one sampling event the dashed line represents a best-fit regression. The coefficient of determination (R2 value) is also shown. Raw influent and secondary effluent molecular assay data obtained from both OC San P1 and P2 are lognormally distributed.
S5 Fig. Log raw influent and secondary effluent correlation plots for OC San P1 and P2.
Ranked concentration values (n = 24) from each P1 secondary treatment process (P1 TF, AS1, AS2) were compared to ranked raw influent concentration values for P1 (left panel). Similarly, ranked P2 TF/SC secondary effluent concentration values (n = 24) were compared to ranked raw influent concentration values from P2 (right panel). The linear model represents the correlation between each distribution from which the correlation coefficient of determination (R2) was calculated.
S1 Table. Summary of percent recoveries for all virus targets.
Average recovery percentages shown in bold were used to correct each respective native dataset, e.g., 120% recovery was used to correct all native P1 raw influent samples for SOM coliphage. The average percent recovery used to correct the native cultivable enteric virus dataset was obtained by taking an average of SOM, MS and Poliovirus recoveries, which are shown as the bold mean values for Poliovirus. An average recovery value was not used to correct the native molecular (enterovirus and norovirus GII) datasets.
S2 Table. Microbial target concentrations.
Microbial concentration ranges, medians and geometric means for each microbial target monitored obtained during the study are shown. Enteric viruses (cultured), MS coliphage, and SOM coliphage have been corrected by virus recovery. Enterovirus, norovirus GII values were not corrected by virus recovery (S4 Text). Total coliform and fecal coliform data were not adjusted.
S3 Table. OC San P1 and P2 Operating Range Values (ORVs) and plant performance during sampling period.
AS = activated sludge, TF = Trickling Filter, SC = Solids Contactor, MCRT = mean cell residence time, BOD-T = total biological oxygen demand. MCRT for all secondary treatment processes are calculated daily by OC San using the below equations. OCWD proposes to use the OC San-calculated values of MCRT for assessment of ORVs related to LRV credit value. BOD-T (P1 TF) is measured by OC San as a daily composite. Mean Cell Residence Time (MCRT) Calculations: P1 Activated Sludge Processes MCRT = (Volume of reactor x MLSS) ÷ [(WAS flow x WAS MLSS) + (Effluent Flow x Effluent TSS)]; where MLSS = mixed liquor suspended solids, WAS = waste activated sludge, and TSS = total suspended solids. P2 Trickling Filter Solids Contactor MCRT = [(Volume of reactor x MLVSS) + (Volume of reactor RSS VSS)] ÷ [(Waste Flow x RSS VSS) + (Effluent Flow x Effluent VSS)]; where MLVSS = mixed liquor volatile suspended solids, RSS VSS = raw sewage sludge volatile suspended solids, and VSS = volatile suspended solids.
S4 Table. Probability values (p-values) for OC San P1 and P2 raw influent and secondary effluent correlations.
Correlations for raw influent and secondary effluent concentration distributions for P1 and P2 are illustrated. These correlations demonstrate how ranked log concentration values from the influent and effluent are related. Correlations for each secondary effluent process were modeled with a linear trendline to calculate Pearson’s correlation coefficient of determination (R2). R2 values measure the amount of variation between two distributions and range between -1 and 1, where 0 represents no correlation and 1 or -1 represent a perfect correlation. Each correlation shown in S5 Fig is statistically significant, with probability values (p-values) less than 0.1% (p < 0.001), as shown in Table 1.
S1 Text. OC San sampling location description.
A total of six (6) sampling locations from OC San P1 and P2 were monitored. Sampling locations for OC San P1 were raw wastewater influent, trickling filter (TF) secondary effluent, activated sludge 1 (AS1) secondary effluent, and activated sludge 2 (AS2) secondary effluent, while sampling locations from OC San P2 consisted of raw wastewater influent and trickling filter/solids contactor (TF/SC) secondary effluent. Sampling at P2 was limited to characterizing the TF/SC process and not other parallel treatment processes. Raw wastewater entering OC San P1 is treated through preliminary screening and primary clarification with chemical addition and is then diverted into one of two secondary treatment trains that operate in parallel (trickling filter process or activated sludge process). Raw wastewater from P1 was collected after primary bar-screening but before primary clarification and chemical addition. Secondary effluents generated by three parallel treatment processes at P1 were sampled in this study. The first P1 treatment train routes the primary effluent through a trickling filter (TF) process followed by secondary clarification. Treated effluent from the TF process was sampled. P1 primary effluent is also sent through two parallel trains of the activated sludge (AS) treatment trains, designated separately as AS1 and AS2. Secondary effluent samples taken from each AS process following secondary clarification was sampled. Both AS trains operate in the nitrification-partial denitrification (NDN) mode. The major difference between the AS1 and AS2 processes is that AS1 does not receive mixed liquor return, while the newer AS2 facility does receive it. Microbial concentrations of both the AS1 and AS2 effluent streams from P1 were of interest due to the operational differences between the two processes described above.
S2 Text. Filtration procedure and ultrafilter elution.
For collection of secondary effluent samples using an ultrafilter, all tubing, fittings, and caps were sanitized and autoclaved in a controlled laboratory setting prior to use. Sterilized equipment was unpacked on-site. All ultrafilter modules were prepared at each sampling location using the same procedure as follows: the blue end of the ultrafilter cell, which houses the hollow fibers, was used to mark the feed port (end receiving OC San sampled water), while the red end of the filter cell was set as the permeate/retentate port (drain). Due to limited access for some secondary effluent sites (P1 AS1 and P1 AS2), a 50-liter carboy was used to collect 25 liters prior to filtration. In this case, a carboy spigot served as the secondary effluent tap for filtration immediately after collection. For remaining sites (P1 TF and P2 TF/SC), secondary effluent was sampled directly from a sampling valve that was compatible with the ultrafilter tubing assembly. In this case, the sample tap was purged for 3 minutes prior to sample collection. To monitor filtration on-site, a pressure gauge and peristaltic pump was installed onto the feed tubing assembly preceding the hollow fiber filter to pump and monitor inlet water pressure. To complete the assembly, a downstream flow meter was installed to the permeate tubing to monitor the volume of water sampled. Upon filtration of 25 liters (at a rate of 1 liter per minute), the peristaltic pump was turned off and the inlet tube line was removed from the sample tap. Any remaining water within the tubing assembly was removed by ensuring all pinch-clamps are placed in the open position and inserting a sterile 2 μm cartridge filter on the inlet tube to avoid aerosol contamination of intake. With the peristaltic pump switched on for approximately 1 minute, the remaining void water was purged from the tubing assembly. Prior to disassembly, sample time, date and location were written on the sample label. Ultrafilter cells were then immediately and carefully unmounted, re-capped, wiped and preserved on ice. To extract contents of the filter, each ultrafilter was eluted using a high-volume elution canister consisting of pressurized 0.075% Tween 20 PBS provided by the manufacturer. Using a sterilized canister adapter fitting, the sample is eluted by manual de-pressurization of the canister with the ultrafilter cell retentate port (drain) open, and the filtrate port (permeate port) closed. An approximate volume of 100–120 mL is eluted off the filter and immediately passed through a sterile 0.45 μm filter to remove large debris and bacterial cells. Each sample (including raw influent concentrates) were divided into a series of subsamples by the Water Quality and Environmental Microbiology Laboratory at Michigan State University to protect samples from multiple freeze-thaw cycles. For sample analysis, one subsample is used for culture-based enteric virus analysis, and one subsample is used for molecular-based virus analysis and the last subsample is stored for archival purposes.
S3 Text. Modifications to EPA 1615 method (ddPCR).
For molecular detection, enterovirus and norovirus GII targets were enumerated using methods and materials as described in the EPA 1615 standard method with modifications adapted to the BioRad QX200 Droplet Digital PCR (ddPCR). Primers and probe sequences used were identical to the EPA method 1615, apart from replacing the probe fluorescent quencher 3ʹ TAMRA with black hole quencher 1 (BHQ1) for both probes and replacing the Norovirus GII probe 5ʹFAM with 5ʹHEX to allow for duplex analysis on the ddPCR platform. The One-Step RT-ddPCR Advanced Kit for Probes (BioRad) PCR kit was used for amplification. Thermal cycling conditions for Enterovirus and Norovirus GII targets were optimized and programmed as follows: 42°C for 60 minutes, 95°C for 10 minutes, 40 cycles of 95°C for 30 seconds and 57°C for 1 minute. Amplification cycles were followed by 98°C incubation for 10 minutes before a 4°C hold cycle. For each virus target, fluorescence-positive or fluorescence-negative droplets were counted after PCR to calculate gene copiers per liter (GC/L) volume using Poisson statistics.
S4 Text. Matrix spike recovery procedure.
Cultivable enteric virus, MS coliphage, and SOM coliphage native concentrations were corrected by the measured method recoveries. Recovery measurements were performed by spiking in MS coliphage (F-specific coliphage, ATCC# 15597-B1), somatic coliphage (ATCC# 13706-B1), and Poliovirus (ATCC VR1562) at a known concentration directly into raw influent grab samples and into the ultrafiltration unit prior to elution. This procedure was performed for P1 and P2 wastewater samples specifically collected for the purpose of matrix spike recovery determination for a subset of sampling events over the course of the study. Method recovery was estimated by calculating the percentage of virus recovered from the matrix spike sample: % recovery = 100 x (observed # in spiked sample–observed # in native) ÷ # spiked virus added Where ‘# spike virus added’ represents the initial spiked concentration of a given virus target, and ‘observed #’ corresponds to either ‘plaque forming units per liter (PFU/L)’, or ‘most probable number per liter (MPN/L)’ for enteric viruses or coliphage, respectively. To correct native cultivable enteric virus concentration data, the average cultivable recovery of MS coliphage, SOM coliphage, and poliovirus was used. Recovery data from all three targets were pooled to calculate a single average recovery value which was then used to correct the native results. Coliphage concentration data were corrected using the average cultivable recovery of MS coliphage and SOM coliphage only. Due to matrix differences (raw wastewater and secondary effluent matrices), recovery corrections were calculated separately for P1 raw influent, P2 raw influent, P1 secondary effluent and P2 secondary effluent each using their respective corresponding group of matrix spike samples. For P1 secondary effluent, recovery measurements were pooled and averaged for all three P1 secondary treatment processes (i.e., P1 AS1, AS2, TF) based on similar ranges of recoveries. For the molecular assay (ddPCR), virus recovery was measured from matrix spike samples by spiking armored RNA (which consists of a gene fragment of the virus target at a known concentration), which is used as an inhibition control in EPA 1615 method, to samples as described above for cultivable enteric virus and coliphage. Instances of measured negative recovery in the present study were converted to zero percent recovery. Reasons for the wide recovery performance are unclear but may be related to potential incompatibility of use of the ddPCR analysis with armored RNA targets (which are traditionally used as an inhibition control in EPA 1615 method using qPCR).
S5 Text. Variability of matrix spike recovery using molecular methods.
For recovery measurements using molecular methods, more research is needed to effectively determine sample recovery using armored RNA as the matrix spike target for ddPCR. Armored RNA was selected for this study based on its long history of use for quality control purposes in the EPA 1615 method (though not for determining recovery as was done for this study). In this study, the norovirus ddPCR recovery results exhibited major inconsistencies necessitating disregarding the recovery data, while the calculated enterovirus ddPCR recoveries were often negative (i.e., spiked sample had less target than the paired native sample) or near zero. As a result, it was deemed not acceptable to use these recovery data to correct the study’s native concentration data given a minimum acceptance criteria of 5% recovery. Thus, native concentration data obtained using ddPCR were not corrected by recovery. Should any future studies consider the use of ddPCR for enumeration of multiple virus targets (i.e., enterovirus, norovirus GII, poliovirus as in this study), it is highly recommended that a robust preoptimization study be performed to address any potential issues with sample-matrix interference, detection of armored RNA, and virus recovery.
S6 Text. Plant performance, operational parameters, and Operating Range Values (ORVs).
Operating range values, or ORVs, were defined in this study as parameter values that describe typical and normal treatment process operations, as observed during this study over the microbial monitoring period. ORVs were derived for each secondary treatment process supplying influent to OCWD AWPF, i.e., P1 TF, AS1, AS2 (current) and P2 TF/SC (future). An exceedance of a designated ORV represents a deviation from the normal treatment performance. This deviation could be related to an unexpected event but also due to planned activities such as operational maintenance of treatment systems or flow adjustments by the OC San operators. These ORVs are proposed to be used for contingent LRV virus credits for GWRS, as with prior pathogen crediting schemes from wastewater treatment for potable reuse in California. This approach is preferred by California regulators such that any virus removal observed from a one-time, past approved study that is used as the basis for ongoing WWTP virus removal credit is only awarded if the WWTP day-to-day performance is normal and acceptable (i.e., within the established ORVs). It should be noted that the ORVs are not directly related to any benchmark, performance goal, or effluent limitation stated in the OC San National Pollutant Discharge Elimination System (NPDES) permit. Thus, failure to meet the ORVs does not signify secondary effluent is of poor quality or unsuitable for reclamation, but rather that the effluent does not meet the normal conditions observed during the microbial sampling study associated with the observed virus removal value and therefore the credit may not be applicable. As a result of the present study, OCWD has proposed to DDW the herein described ORVs framework where each of the four OC San effluents serving OCWD GWRS features an ORV, as does the combined (blended) effluents in the form of Microfiltration Feed (MFF) and Microfiltration Effluent (MFE) at GWRS, all of which must meet their respective ORVs in order to receive virus LRV credit. To determine a recommended ORV for each OC San treatment process, performance data from P1 and P2 were collected as part of routine plant monitoring during the LRV study sampling period. Parameters chosen to develop operational envelope ORVs are shown further below in S3 Table. Performance data were reviewed and analyzed by calculating 30-day averages and interquartile ranges. In addition to ORVs from OC San’s treatment processes, GWRS AWPF MFF and MFE monitoring locations were also selected to represent normal GWRS microfiltration influent and effluent quality. ORVs for each selected parameter were calculated using a baseline threshold equation from 30-day average data as follows: (Eq S1) (Eq S2) where: Q1 = 25th percentile of the 30-day running average dataset, Q3 = 75th percentile of the 30-day running average dataset, and IQR = Interquartile range, defined as Q3 –Q1. The baseline threshold approach (i.e. defined by interquartile ranges) was used to define excursions of baseline conditions of a treatment process using a statistical model derived from a large dataset. Lower thresholds are defined by values below the 25th percentile by 1.5 times the IQR, while upper thresholds are above the 75th percentile by 1.5 times the IQR, respectively.
S7 Text. Performance monitoring and ORV results.
Performance data from OC San P1 and P2 were collected to determine a recommended operating range value (ORV) for each OC San treatment process. Total biological oxygen demand (BOD-T) was selected as the ORV for the OC San P1 TF process. An upper baseline threshold for BOD-T is proposed because higher-than-normal BOD-T (30-day running average of > 26 mg/L) in this sampling location would suggest deviations from the typical performance that was documented during the enteric virus sampling. Lower baseline thresholds of Mean cell residence time (MCRT) was selected and proposed as the ORV for OC San P1 AS1, P1 AS2, and P2 TF/SC treatment processes. Lower thresholds were chosen because a lower-than-normal MCRT (<3 days for P1 AS1, and <4 days for P1 AS2, < 1 day for P2 TF/SC) in these processes would suggest deviation from typical performance. In addition to OC San treatment processes, turbidity parameters for both the GWRS AWPF Microfiltration Feed (MFF) and Microfiltration Effluent (MFE) monitoring locations. OCWD proposed turbidity parameters for both MFF and MFE to be included within the ORV framework in part because of an existing limit set for the AWPF MF treatment process for Giardia and Cryptosporidium removal credit. In addition to this, OC San effluent must meet a specified turbidity requirement to provide blended secondary effluent to the AWPF. Using the described ORVs framework for the overall potable reuse project, each of the four OC San effluents serving GWRS features an ORV, as well as the combined (blended) effluents in the form of MFF and MFE, must meet their respective ORVs to receive virus LRV credit.
S8 Text. Monte Carlo simulation.
For the Monte Carlo approach, LRVs for each microbial target were calculated using the MATLAB software (1984–2020 MathWorks, Inc., version R2020a 22.214.171.1249463), equipped with the Statistics and Machine Learning Toolbox. Microbial concentration data were imported into the MATLAB software using a simplified tab-delimited file. Once imported, a statistical model for each influent and effluent dataset for a given microbial target was generated using the maximum likelihood estimates function. Briefly, influent and effluent microbial concentrations were used to generate a statistical distribution. From these modeled influent and effluent distributions, one independent and random value was selected from each distribution and subsequently paired to calculate an LRV as shown in Eq S3: (Eq S3) Where Ceff is the concentration of the microbial target taken from the secondary effluent distribution, and Craw is the concentration of the microbial target taken from the raw wastewater distribution. This calculation was performed 10,000 times to generate a distribution of n = 10,000 LRVs. All LRVs were then sorted from low to high and assigned a rank, i, over the total number of data points, n. A cumulative probability, p, for each value was assigned as shown in Eq S4. (Eq S4) Where p is the cumulative probability, i is the rank assignment, and n represents the total number of calculated data points (10,000). The Monte Carlo simulation approach was also modified for the present study to generate only non-negative LRVs (n = 10,000+ non-negative log removal values). This was done because the standard Monte Carlo simulation approach generated a fairly large number of negative LRVs as a portion of the total 10,000 LRVs, which reduced the resulting 5th percentile LRV. To address the negative LRVs, approximately 1,000 additional LRVs were calculated for a total of 11,000 samples. Negative LRVs within the n = 11,000 dataset were removed such that the remaining number of positive LRVs were at least n = 10,000. This modified (censored) Monte Carlo approach was used to determine process-specific LRVs, imposing a condition of reality on the statistically determined outcome. A modified Monte Carlo simulation was executed to address the negative LRVs calculated for the P1 TF and P2 TF/SC distributions when using the standard Monte Carlo approach. While this only occurs a fraction of the time, these negative values are recorded as a possible LRV at the low end of the percentile distribution. The calculated negative LRVs were believed to be a mathematical artifact of the Monte Carlo’s random pairing of influent-effluent values, specifically attributed to the large overlap in concentration values observed between the raw influent distribution and OC San P1 TF effluent distribution. The modified (censored) Monte Carlo approach was used to determine process-specific LRVs, imposing a condition of reality on the statistically determined outcome. Negative LRVs generated with the modified Monte Carlo simulation were removed such that the remaining number of positive LRVs were at least n = 10,000. It is physically impossible to generate a negative LRV for an enteric virus during wastewater treatment, as virus cannot be created within primary or secondary treatment processes due to the lack of a host organism. Despite the attempt to resolve this issue by censoring data to remove negative LRVs, this estimation is not representative of the actual low-end virus log removal observed for all four treatment processes (Table 1).
S9 Text. Probability plots for enterovirus and norovirus GII targets (molecular targets).
Non-detect measurements were observed for a subset of sampling events for enterovirus including P1 AS1 (n = 8), P1 AS2 (n = 11), and P2 TF/SC (n = 5) secondary effluent. Only one non-detect measurement was observed for norovirus GII for one sample of P1 AS2 effluent. These cases were substituted with the method detection limit as a conservative upper end estimate of the effluent concentration for purposes of completing the LRVs statistical determination; use of a statistical technique to estimate values below the detection limit would result in comparatively lower effluent concentrations assigned to the non-detect results and correspondingly higher LRVs. Covariance and modified Monte Carlo simulation LRV distributions generated for each treatment process are presented in S4 Fig for enterovirus and norovirus GII. Irrespective of the statistical method used, the least efficient secondary treatment process for both enterovirus and norovirus GII is OC San P1 TF, which is consistent with other targets measured in this study (cultivable enteric virus, SOM coliphage and coliform). Of all the treatment processes, the AS secondary treatment processes at OC San P1 demonstrates the highest removal of gene copies. Similar to the cultivable enteric virus, 5th percentile covariance LRVs for both enterovirus and norovirus GII are larger than the Monte Carlo 5th percentile LRVs. For covariance approach for enterovirus, the P1 TF and P2 TF/SC show the same 5th percentile LRV (rounding to 1.3 LRV) while the modified Monte Carlo analysis shows P1 TF as the lowest 5th percentile LRV. It should be noted that since the method detection limit was substituted in place of non-detect for several samples in the case of enterovirus analysis, this may affect the correlation variable in the covariance analysis method, which may explain the converging distributions at the lower percentile (5th percentile) LRVs.
We would like to acknowledge the Metropolitan Water District of Southern California (Future Supply Actions Program) and the Water Research Foundation (WRF, Subscriber Priority Research Program) including the WRF Project Advisory Committee for providing critical review of this study. We gratefully acknowledge our project partners including those in the Orange County Sanitation District operations and laboratory services teams who assisted with sample collection and study review; as well as the Water Quality and Environmental Microbiology Laboratory at Michigan State University (Professor Joan Rose) and Biological Consulting Services of North Florida (George Lukasik, Ph.D.) for processing of samples. Finally, the authors would like to thank our colleagues who participated in panel discussions and provided critical review and feedback for this work, including George Tchobanoglous, Rhodes Trussell, and Joan Rose.
- 1. U.S. Environmental Protection Agency (EPA). Potable Reuse Compendium, EPA/810/R-17/002. Washington D.C. 2017. Available from: https://www.epa.gov/ground-water-and-drinking-water/2017-potable-reuse-compendium
- 2. U.S. Environmental Protection Agency (EPA). Guidelines for Water Reuse, EPA/600/R-12/618. Washington D.C. 2012. Available from: https://www.epa.gov/sites/default/files/2019-08/documents/2012-guidelines-water-reuse.pdf
- 3. Regulations related to recycled water. California State Water Resources Control Board, Department of Drinking Water (DDW) 2018 [cited 2023 Feb]. Available from: https://www.waterboards.ca.gov/drinking_water/certlic/drinkingwater/documents/lawbook/RWregulations_20181001.pdf
- 4. Gerrity D, Pecson B, Trussell RS, Trussell RR. Potable reuse treatment trains throughout the world. J Wat Supply: Res Technol—Aqua. 2013;62:321–38. https://doi.org/10.2166/aqua.2013.041.
- 5. World Health Organization (WHO). Potable Reuse: Guidance for producing safe drinking water. Water S, Hygiene and Health, editor. Geneva, Switzerland.2017. 152 p. Available from: https://apps.who.int/iris/handle/10665/258715
- 6. Polanco J, Safarik J, Plumlee MH. Demonstrating Virus Log Removal Credit for Wastewater Treatment and Reverse Osmosis for Potable Reuse at OCWD. Final Report. Water Research Foundation (WRF) and Metropolitan Water District of Southern California (MWD), 2022. WRF 5041.
- 7. Pecson B, Dale C, Trussell S, Rose J, Ives R, Idica E, et al. Findings from a One-Year Pathogen Monitoring Study to Support Potable Reuse at the City of Oceanside, CA. WEFTEC 2017; Proceedings of the Water Environment Federation: Water Environment Federation; 2017. p. 5130–41.
- 8. Bartolo M, Chen E, Kolakovsky A, Bear S, Kenny J, Pecson B, et al. Pathogen Monitoring Study at the North City Water Reclamation Plant, Final Report. 2017.
- 9. White’s handbook of chlorination and alternative disinfectants. Black & Veatch Corporation. 5th ed. Hoboken, New Jersey: John Wiley & Sons, Inc.; 2010.
- 10. National Water Research Institute (NWRI). Review of the Orange County Water District Groundwater Replenishment System. Meeting 17 Panel Report. 2021 January 6. Publication Number: NWRI-2020-18. Available from: https://www.nwri-usa.org/
- 11. National Water Research Institute (NWRI). Review of the Orange County Water District Groundwater Replenishment System. Meeting 18 Panel Report. 2021 November 10. Available from: https://www.nwri-usa.org/
- 12. Yates MV, Gerba CP, Kelley LM. Virus Persistence in Groundwater. Appl Environ Microbiol. 1985;49(4):778–81. pmid:4004211
- 13. Yates MV, Stetzenbach LD, Gerba CP, Sinclair NA. The effect of indigenous bacteria on virus survival in ground water. Journal of Environmental Science and Health Part A: Environmental Science and Engineering and Toxicology. 1990;25(1):81–100.
- 14. Betancourt WQ, Kitajima M, Wing AD, Regnery J, Drewes JE, Pepper IL, et al. Assessment of virus removal by managed aquifer recharge at three full-scale operations. Journal of Environmental Science and Health, Part A. 2014;49(14):1685–92. Epub October 16, 2014. pmid:25320855
- 15. U.S. Environmental Protection Agency (EPA). Method 1615 Measurement of Enterovirus and Norovirus Occurence in Water by Culture and RT-qPCR, EPA/600/R-10/181. Version 1.3 ed. Cincinnati, OH. 2014. Available from: https://cfpub.epa.gov/si/si_public_file_download.cfm?p_download_id=522923
- 16. U.S. Environmental Protection Agency (EPA). Method 1602: Male-specific (F+) and Somatic Coliphage in Water by Single Agar Layer (SAL) Procedure. EPA 821-R-01-029. Washington, D.C. 2001. Available from: https://www.epa.gov/sites/default/files/2015-12/documents/method_1602_2001.pdf
- 17. Standard Methods Committee of the American Public Health Association (APHA), American Water Works Association, and Water Environment Federation. 9222 membrane filter technique for members of the coliform group. In: Standard Methods For the Examination of Water and Wastewater. Lipps WC, Baxter TE, Braun-Howland E, editor.: APHA Press.; 2006.
- 18. Ji P, Aw TG, Van Bonn W, Rose JB. Evaluation of a portable nanopore-based sequencer for detection of viruses in water. J Virol Methods. 2020;278:113805. Epub 20191228. pmid:31891731.
- 19. Trussell SR, editor Monitoring Pathogen Concentrations through the City of Oceanside’s San Luis Rey Wastewater Treatment Plant. Proceedings of the 11th International Water Association International Conference on Water Reclamation and Reuse; 2017 July 23–27, 2017; Long Beach, CA: International Water Association (IWA).
- 20. Tchobanoglous G, Kenny J, Leverenz H. Rationale for constant flow to optimize wastewater treatment and advanced water treatment performance for potable reuse applications. Water Environ Res. 2021;93(8):1231–42. Epub 20210215. pmid:33547686; PubMed Central PMCID: PMC8451933.
- 21. Schwartzmann A. Pathogen Monitoring Study at the North City Water Reclamation Plant—Statistics Review and Analysis Report. University of California, San Diego, 2018.
- 22. Helsel DR, Hirsch RM, Ryberg KR, Archfield SA, and Gilroy EJStatistical methods in water resources: U.S. Geological Survey Techniques and Methods. U.S. Geological Survey2020. p. 117–43.
- 23. Blom G. Statistical Estimates and Transformed Beta-variables [PhD dissertation]. Stockholm: Almqvist & Wiksell: Stockholm College.; 1958.
- 24. Schmidt PJ, Anderson WB, Emelko MB. Describing water treatment process performance: Why average log-reduction can be a misleading statistic. Water Res. 2020;176:115702. Epub 20200309. pmid:32247998.
- 25. Pecson B, Darby E, Di Giovanni G, Leddy M, Nelson K, Channah R, et al. Pathogen Monitoring in Untreated Wastewater. Final Report. Water Research Foundation (WRF), California State Water Resources Control Board, and Metropolitan Water District of Southern California, 2021 WRF 4989.
- 26. National Water Research Institute (NWRI). Expert Advisory Panel for Pure Water San Diego. Subcommittee Report on Statistics and Pathogen Monitoring.; February 112019. Available from: https://www.nwri-usa.org/
- 27. Stevenson J, Hymas W, Hillyard D. The use of Armored RNA as a multi-purpose internal control for RT-PCR. J Virol Methods. 2008;150(1–2):73–6. Epub 20080418. pmid:18395804; PubMed Central PMCID: PMC7119664.