key: cord-1016504-ek8n6n18
authors: Fitzgerald, S. F.; Rossi, G.; Low, A. S.; MacAteer, S. P.; O'Keefe, B.; Findlay, D.; Cameron, G. J.; Pollard, P.; Singleton, P. T. R.; Ponton, G.; Singer, A. C.; Farkas, K.; Jones, D. L.; Graham, D. W.; Quintela-Baluja, M.; Tait-Burkard, C.; Gally, D. L.; Kao, R. R.; Corbishley, A.
title: COVID-19 mass testing: harnessing the power of wastewater epidemiology
date: 2021-05-26
journal: nan
DOI: 10.1101/2021.05.24.21257703
sha: 6c084dc6c0878172191d7573cbb5b6a00faef049
doc_id: 1016504
cord_uid: ek8n6n18

Background COVID-19 patients shed SARS-CoV-2 RNA in their faeces. We hypothesised that detection of SARS-CoV-2 RNA in wastewater treatment plant (WWTP) influent could be a valuable tool to assist in public health decision making. We aimed to rapidly develop and validate a scalable methodology for the detection of SARS-CoV-2 RNA in wastewater that could be implemented at a national level and to determine the relationship between the wastewater signal and COVID-19 cases in the community. Methods We developed a filtration-based methodology for the concentration of SARS-CoV-2 from WWTP influent and subsequent detection and quantification by RT-qPCR. This methodology was used to monitor 28 WWTPs across Scotland, serving 50% of the population. For each WWTP catchment area, we collected data describing COVID-19 cases and deaths. We quantified spatial and temporal relationships between SARS-CoV-2 RNA in wastewater and COVID-19 cases. Findings Daily WWTP SARS-CoV-2 influent viral RNA load, calculated using daily influent flow rates, had the strongest correlation ({rho}>0.9) with COVID-19 cases within a catchment. As the incidence of COVID-19 cases within a community increased, a linear relationship emerged between cases and influent viral RNA load. There were significant differences between WWTPs in their capacity to predict case numbers based on influent viral RNA load, with the limit of detection ranging from twenty-five cases for larger plants to a single case in smaller plants. Interpretation The levels of SARS-CoV-2 RNA in WWTP influent provide a cost-effective and unbiased measure of COVID-19 incidence within a community, indicating that national scale wastewater-based epidemiology can play a role in COVID-19 surveillance. In Scotland, wastewater testing has been expanded to cover 75% of the population, with sub-catchment sampling being used to focus surge testing. SARS-CoV-2 variant detection, assessment of vaccination on community transmission and surveillance for other infectious diseases represent promising future applications.

Background 22 COVID-19 patients shed SARS-CoV-2 RNA in their faeces. We hypothesised that detection of 23 SARS-CoV-2 RNA in wastewater treatment plant (WWTP) influent could be a valuable tool to 24 assist in public health decision making. We aimed to rapidly develop and validate a scalable 25 methodology for the detection of SARS-CoV-2 RNA in wastewater that could be 26 implemented at a national level and to determine the relationship between the wastewater 27 signal and COVID-19 cases in the community. 28

We developed a filtration-based methodology for the concentration of SARS-CoV-2 from 30 WWTP influent and subsequent detection and quantification by RT-qPCR. This methodology 31 was used to monitor 28 WWTPs across Scotland, serving 50% of the population. For each 32 WWTP catchment area, we collected data describing COVID-19 cases and deaths. We 33 quantified spatial and temporal relationships between SARS-CoV-2 RNA in wastewater and 34 COVID-19 cases. 35

Daily WWTP SARS-CoV-2 influent viral RNA load, calculated using daily influent flow rates, 37 had the strongest correlation (ρ>0.9) with COVID-19 cases within a catchment. As the 38 incidence of COVID-19 cases within a community increased, a linear relationship emerged 39 between cases and influent viral RNA load. There were significant differences between 40 WWTPs in their capacity to predict case numbers based on influent viral RNA load, with the 41 limit of detection ranging from twenty-five cases for larger plants to a single case in smaller 42

Interpretation 44

The levels of SARS-CoV-2 RNA in WWTP influent provide a cost-effective and unbiased 45 measure of COVID-19 incidence within a community, indicating that national scale 46

The COVID-19 pandemic has necessitated the rapid implementation of surveillance 63 programmes worldwide to track and control the spread of SARS-CoV-2. Initially, such 64 programmes relied on syndromic surveillance, community testing, contact tracing and the 65 monitoring of morbidity and mortality rates (1-3). Community testing relies on voluntary 66

reporting of clinical signs and is only partially able to capture the pre-symptomatic, 67 asymptomatic and pauci-symptomatic cases of SARS-CoV-2 infection that can contribute 68 significantly to community transmission, and are therefore subject to biases, which can 69 influence estimates of disease burden (1, 2). Syndromic surveillance based on hospital 70 admissions is less biased, but is subject to delays between infection and admission (2), while 71

implementing mass swab-testing on a nationally meaningful scale is not economically 72 feasible for most countries (2). 73

Early studies identified SARS-CoV-2 RNA in the faeces of infected individuals and COVID-19 74 has subsequently been associated with a range of gastrointestinal symptoms (4). SARS-CoV-75 7 high resolution allocation of the number of tests, detected cases (i.e. positive tests), and 134 COVID-19 related deaths for each of the CAs. 135

The objective was to understand the association between the concentration and daily viral 137 To test the association between observed cases (Yi,j) and daily WWTP viral RNA load (Xi,j), we 151 fitted a linear mixed model: 152 ",$ = ( + * ",$ + $ + $ ",$ + " , 153

where b0 and b1 represent the fixed intercept and coefficient of the daily WWTP viral RNA 154 load Xi,j. Parameters uj and bj are the random intercept and coefficient, associated with each 155 group j (catchment). Before the estimation, the dependent and independent variables were 156 square root transformed. 157 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 26, 2021. ; https://doi.org/10.1101/2021.05.24.21257703 doi: medRxiv preprint explained by both fixed and random effects (11) and analysed the resulting coefficients 159 (intercept, b0 + uj, and slope, b1 + bj) to assess the consistency of the signal and the potential 160 causes of the differences between WWTPs. We first fitted a series of univariable linear 161 regression models with the site's slope or intercept as the dependent variable and 162 population, population density, number of wastewater samples, latitude, longitude, 163 deprivation and access indices (10) as dependent variables. We then fitted a multivariable 164 model to each coefficient, selecting as independent variables those returning a p-value 165 below 0·2 in the univariable models. Variables were then further selected through a 166 stepwise selection in order to eliminate the statistically insignificant ones. 167

All data manipulation and analysis was done in R 4.0.1 (12) using packages tidyverse (13), 168 lme4 (14), and MuMIn (15). 169

Reliable quantification of SARS-CoV-2 in wastewater requires consistent viral RNA extraction 173 across a broad range of concentrations. To investigate this, aliquots of a single wastewater 174 sample were spiked with a serial dilution of heat-inactivated SARS-CoV-2. There was no 175 association between viral concentration and the efficiency of RNA recovery across five 176 orders of magnitude of SARS-CoV-2 concentration (Fig S1.1.A) . We next validated PRRSv (a 177 porcine enveloped nidovirus that can be cultured in vitro at Containment Level 2) as a 178 suitable surrogate process control virus for SARS-CoV-2. The extraction efficiencies of heat-179 inactivated SARS-CoV-2, live PRRSv and heat-inactivated PRRSv were comparable when used 180 to spike a single wastewater sample (Fig S1.1.B) . Extraction efficiencies were also 181 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. detection of the N1-gene by RT-qPCR was found to be more sensitive than the E-gene (Fig  212   S1 .2.B) and therefore N1-gene detection was adopted for the national programme. 213

The weekly number of SARS-CoV-2 reported cases, deaths and positivity are shown in Fig  215   1A . As of 29/1/2021, 989 wastewater samples, with three replicates each, have been 216 analysed across 28 WWTPs, with the earliest samples taken from late May 2020 ( Fig 1B) . 217

The number of samples per WWTP ranged from 12 (Stornoway, Outer Hebrides) to 112 218 (Shieldhall, Greater Glasgow). The CAs are distributed across Scotland ( Fig 1C) and despite 219 covering only 1·2% of Scotland's land mass, they cover 50% of the population. Daily WWTP 220 influent flow data was missing for 18% of the samples. 221

As evident in Fig 2, wastewater RNA viral concentration (panels A, C and E) and daily WWTP 222 viral RNA load (panels B, D, and F) mimic the trends of the daily positive test rate (number of 223 positive tests over the total) and the daily incidence curves, respectively. This was 224 independent of the CA population size (Fig S2.1 to S2.5 for remaining WWTPs). 225

Preliminary correlation analyses showed that WWTP daily viral RNA load is highly correlated 226 with the number of COVID-19 cases detected in the CA, while the wastewater viral 227 concentration was highly correlated with the positive test rate (Fig S2.6 and Fig S2.7) . The 228 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 26, 2021.

correlations improve as the number of contributing days for case counts before sampling 229 increases from zero to five, at which point it stabilises. 230

The full mixed model explained 78% of the variance in the number of cases in the CA 231 (conditional R 2 = 0·78), while the daily viral RNA load as a fixed effect explained 45% of the 232 variance (marginal R 2 = 0·45). The null hypothesis that the sites' random slope variance was 233 zero, i.e. each site had the same slope gradient, was rejected with a Chi 2 test (p ~ 0). By 234 varying the length of the time period used to calculate the number of cases, the conditional 235 R 2 ranged from 0·71 to 0·89, with an average of 0·76 across the 29 periods considered WWTPs of different size: median 0·84 [0·63-0·90] for the smaller sites (< 10,000 population), 244 2·25 [1·72-3·78] for the medium-sized sites (10,000 to 100,000 population), and 5·30 [3·2-245 6·95] for the larger sites (> 100,000 population). This translates to a threshold of less than 246 one recorded case from which the relationship between viral RNA load and cases is 247 detectable in small catchments, five recorded cases in the medium-sized catchments and 248 twenty-five cases in the large catchments. Among the latter group, Dalmuir and 249

Meadowhead were outliers, with higher intercept and lower slope compared with similar-250 sized catchments (Fig 4C) . 251 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 26, 2021. ; https://doi.org/10.1101/2021.05.24.21257703 doi: medRxiv preprint

The variables that best explain differences in mixed model slopes across WWTPs were the 252 population size and the number of samples taken, although geographical longitude (not 253 significant) was retained after multivariable model stepwise selection (Table 1) Spain (19). Importantly, we demonstrate how WBE can be adopted across a range of 284 catchments, from densely populated urban areas (Edinburgh and Glasgow), to smaller 285 towns, rural areas and islands. In contrast to previous studies, we demonstrate the value of 286 obtaining flow measurements from WWTPs to calculate daily viral RNA loads, which display 287 a stronger correlation with detected community case numbers, compared with viral 288 concentration data alone (Fig S2.6 ). This strong correlation demonstrates that degradation 289 of viral RNA within the wastewater network is minimal over the considered time scale, as 290 reported by other studies (20). 291

Our typically low limits of detection show that wastewater surveillance can be particularly 292 valuable for areas reaching low prevalence and is therefore suitable as a logistically 293 sustainable and cost-effective early warning system, making a targeted community testing 294 strategy viable. For WWTPs collecting wastewater from cities, it is harder to isolate small 295 clusters of infections. This hurdle can be overcome by sampling a site "upstream" to the 296 WWTP (i.e. within the sewerage network) to improve spatial resolution. This is currently 297 taking place in Scotland, with local health boards using sub-catchment wastewater sampling 298 to direct surge testing. 299 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 26, 2021. ; https://doi.org/10.1101/2021.05.24.21257703 doi: medRxiv preprint community interventions, however a potential issue here is the variability in the signal. 301

Specifically, we observed sudden spikes in the viral RNA load or viral concentration in many 302 small WWTPs (Fig 2, E and F; Fig S2.4; Fig S2.5) . While smaller catchments might be more 303 sensitive to individual variations in shedding, these spikes might also be caused by one or 304 two households being infected in a short period of time. Given the sensitivity of these 305 smaller WWTPs to a small number of cases, this may explain these sudden variations in the 306 SARS-CoV-2 daily viral RNA load. 307

Whilst we have shown that daily viral RNA load has the best correlation with detected cases, 308 daily WWTP flow measurements are not always available. This may be more of a problem in 309 smaller WWTPs, where flow rates regularly exceed the working range of the flow meter or 310 in low resource settings, however our model retained substantial detection power when 311 daily flow was estimated using easily obtained ammonium concentrations, with the 312 conditional R 2 dropping by only 2% (R 2 = 0·76). stabilises once detected cases are included up to and including the five days prior to 322 wastewater sampling. Furthermore, even with declining incidence, when the cumulative 323 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 26, 2021. ; https://doi.org/10.1101/2021.05.24.21257703 doi: medRxiv preprint effect of older infections would be expected to have a greater contribution to the overall 324 signal if shedding duration was long, the conditional R 2 of the mixed models did not 325 deteriorate significantly (0·76 compared to 0·78 when incidence was increasing), and was 326 consistent with a short period of peak viral shedding. Unfortunately, there is currently very 327 CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Multivar. intercept --0·01 0·908 Table 1 . Results of the univariable and multivariable linear models to determine the 394 variables that influence the mixed model slope and intercept for different sites. Deprivation 395 and access indices measure the relative deprivation and the access to healthcare services 396 respectively of a datazone. They were included as potential causes of bias in case detection. 397

The R 2 of the two multivariable linear models was 0.45 for the slope, and 0.50 for the 398 intercept (both p < 0.001). 399 400 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 26, 2021. ; https://doi.org/10.1101/2021.05.24.21257703 doi: medRxiv preprint

Wastewater-Based Epidemiology (WBE) and Viral Detection in Polluted Surface Water: A 403

Valuable Tool for COVID-19 Surveillance-A Brief Review

Making waves: Wastewater-based epidemiology for COVID-19 -approaches and challenges 407 for surveillance and prediction

Implementation of environmental 409 surveillance for SARS-CoV-2 virus to support public health decisions: Opportunities and 410 challenges

SARS-CoV-2 412 in wastewater: State of the knowledge and research needs

Faecal shedding models for SARS-CoV-2 RNA amongst 415 hospitalised patients and implications for wastewater-based epidemiology

Presence of SARS

Coronavirus-2 RNA in Sewage and Correlation with Reported COVID-19 Prevalence in the

Show 422 us the Data: Global COVID-19 Wastewater Monitoring Efforts, Equity, and Gaps. medRxiv

theme/population/population-estimates/2011-based-special-area-population-estimates/nuts-432 population-estimates

Extension of Nakagawa & Schielzeth'sR2GLMMto random slopes 434 models

Welcome to 437 the Tidyverse

Fitting Linear Mixed-Effects Models Using 439 lme4

Multi-Model Inference

Measurement of SARS-CoV-2 RNA in wastewater tracks community infection dynamics

First confirmed 445 detection of SARS-CoV-2 in untreated wastewater in Australia: A proof of concept for the 446 wastewater surveillance of COVID-19 in the community

CoV-2 RNA in wastewater anticipated COVID-19 occurrence in a low prevalence area

Decay of 456 SARS-CoV-2 and surrogate murine hepatitis virus RNA in untreated wastewater to inform 457 22

SARS-CoV-2 titers in 465 wastewater are higher than expected from clinically confirmed cases

Estimating the proportion of SARS-CoV-468 2 infections reported through diagnostic testing

Genome Sequencing of Sewage Detects Regionally Prevalent SARS-CoV-2 Variants

An integrated national scale SARS-CoV-2 genomic 473 surveillance network

Environmental surveillance system to track 475 wild poliovirus transmission

Seasonal 477 and spatial dynamics of enteric viruses in wastewater and in riverine and estuarine receiving 478 waters

Detection 480 of Hepatitis E Virus in Sewage After an Outbreak on a French Island