High-Throughput Wastewater SARS-CoV-2 Detection Enables Forecasting of Community Infection Dynamics in San Diego County

1Wastewater-based epidemiology (WBE) can facilitate detailed mapping of the extent and spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in a community and has seen a rapid rise in recent months owing to its cost effectiveness as well as its ability to foreshadow trends ahead of diagnostic testing. With over 46 million cases reported globally, 9.3 million of which are from the United States, there is an imminent need for rapid, community-level surveillance in order to identify potential outbreak clusters ahead of diagnostic data. Previous studies have reported high levels of correlation between viral concentration in sewage to clinically reported cases in a community with trends appearing 2 to 8 days ahead in wastewater. A major bottleneck in large-scale wastewater surveillance is the lack of robust, high-throughput viral concentration methodology. Conventional techniques for viral concentration from wastewater typically employ laborious or time-consuming processes, namely, polyethylene glycol (PEG)-based precipitation, direct filtration, or ultrafiltration methods that severely limit throughput. In the present study, we employed an affinity-capture magnetic hydrogel particle (Nanotrap)-based viral concentration method which was incorporated on the KingFisher Flex liquid-handling robot platform robots (Thermo Fisher Scientific, USA), using a 24-plex head to process 24 samples at once in a 40-min run. RNA is then extracted on the same KingFisher system for rapid sample processing. Reverse transcription-quantitative PCR (RT-qPCR) targeting N1, N2, and E gene targets was used for detection and quantification of SARS-CoV-2 RNA. Detection of all three genes in a sample and its replicate (at quantification cycle [Cq] values of <39) was considered positive. All steps from concentration to RT-qPCR plating were performed hands-free and conducted by liquid-handling robots to minimize human error. Using the above pipeline, 96 raw sewage samples were processed in a period of 4.5 h (concentration to RT-qPCR detection/quantification), effectively reducing the processing time by at least 20-fold. The sensitivity of detection of SARS-CoV-2 viral RNA in wastewater by the high-throughput pipeline (i.e., to determine whether individual buildings yield sufficient wastewater signal for high-resolution spatial studies) was established by routinely monitoring the SARS-CoV-2 signatures in the wastewater of a large San Diego hospital building housing active coronavirus disease 2019 (COVID-19) patients. This site was used as a positive control to test correlations with caseload on a daily basis. Sewage samples were collected daily for a period of 12 weeks during which time the hospital’s caseload (specific to COVID-19 patients) varied between 2 and 26. SARS-CoV-2 viral gene copies correlated with the daily hospital caseload (r = 0.75; see Fig. S1 in the supplemental material), suggesting that the wastewater data could at least be used to identify the peaks. SARS-CoV-2 viral RNA was detected in wastewater on all days sampled. Furthermore, the high-throughput protocol has been used as a part of an on-campus wastewater surveillance for the last few months where data from 70 autosamplers covering individual buildings are sampled and analyzed on a daily basis. The method enabled detection of cases as low as 1 asymptomatic individual in buildings with over 400 occupants (unpublished data [data available on request]).

2The efficiency of high-throughput concentration method was compared to the two other commonly implemented concentration methods (electronegative membrane filtration and polyethylene glycol [PEG]-based) using ninefold serial dilutions of heat-inactivated SARS-CoV-2 viral particles spiked into 10-ml volumes of raw sewage which was previously verified to be from a location with no SARS-CoV-2 prevalence and verified by RT-qPCR (see Text S1 in the supplemental material). PMMoV (pepper mild mottle virus) concentration was used to normalize the obtained SARS-CoV-2 viral concentrations (to account for load changes, if any). The high-throughput protocol compared favorably to the conventionally used protocols demonstrating its potential implications for large-scale sample processing (>100 samples/day) (see Fig. S2A, Fig. S3, and Table S1 in the supplemental material). The average viral recovery efficiencies were 27% (standard deviation [SD], 8%), 15% (SD, 9%) and 13% (SD, 8%), respectively, for the high-throughput, PEG, and HA filtration protocol (Text S1). The improved recovery could also be attributed in part due to the specificity of the magnetic beads in recovering viral particles coupled with a magnetic-bead-based nucleic acid extraction protocol as well. The protocol was optimized on the KingFisher for optimal bead binding through homogenization (https://doi.org/10.17504/protocols.io.bptemnje). Furthermore, the automated protocol was hands-free and less prone to user discrepancies than the other two methods which were carried out manually.

3Twenty-four-hour flow weighted composites were collected each day for 3 months between 20 July 2020 and 21 October 2020 from the influent stream of Point Loma wastewater treatment plant. The plant processes over 175 MGD (million gallons per day) of raw sewage and is the primary treatment center for the greater San Diego area serving over 2.3 million residents. SARS-CoV-2 viral RNA was detected in all of the samples processed at an average concentration of 2,010,104 gene copies/liter in the influent. Over the course of the study, the clinically reported cases in the county increased by 29,375 (Fig. 1). Peaks in the wastewater data were frequently followed by peaks in the clinically confirmed cases at a later date. This suggests a correlation between wastewater and the number of new cases with the caveat of a time delay, where the wastewater data predicts future trends in the new number of cases. Although informative, this time-lagged correlation alone is not enough for robust predictions. This served as the main motivation to build a predictive model for forecasting the number of new cases per day in San Diego County. In order to model the number of new reported cases, which correlates not only to wastewater but also to other complex prevalence and spreading dynamics, we embed useful predictors as separate time series to take advantage of any side information that may improve our forecasting results. Here, the day of the week was tracked for each data point in a third time series to capture any weekly trends for a total of three time series with 88 data points each. We used a data-driven approach to train a prediction model that utilizes prior reported cases, wastewater data, and temporal correlations (embedded in the day of the week) in order to forecast the number of new positive cases in San Diego County. The multivariate autoregressive integrated moving average (ARIMA) model was applied to build a prediction model for the number of new positive cases. The (predicted) number of new cases consist of lagged past values from all three series (number of new cases, wastewater data, day of the week), and each term can be thought of as the influence of that lagged time series on the number of new cases. A total of 65 data points corresponding to data collected from 7 July 2020 to 28 September 2020 were used for developing the predictive model and forecasting estimates. The remaining 24 data points, corresponding to data collected from 29 September 2020 to 25 October 2020, were reserved for validating our forecast estimates. Figure 2B shows the results of the model (prediction and forecast) compared to the observed data. The Pearson correlation coefficients (r) and root mean squared error (RMSE) between the observed data and the predicted model were r = 0.84, 0.79, 0.69, and 0.47 and RMSE = 57, 50, 59, and 70 for the trained model, 1, 2, and 3 week advance forecast values, respectively (Table S2). Our data-driven approach obtains forecasts successfully capturing general trends on the number of new cases (shown here up to 3 weeks in advance), ultimately reinforcing that wastewater analysis can be expository of previously undetected SARS-CoV-2 infections in the population which could provide useful insight for county officials for the purpose of early public health interventions. As with most environmental samples, there are biases and caveats associated with interpreting data from wastewater quantitatively, there are trade-offs while attempting to scale-up the process to enable the screening of hundreds of samples on a daily basis with a quick turnaround time (such as the requirement to purchase liquid-handling robots). However, our data show that we can still capture low prevalence cases enabling studies at high spatial resolution. Our study demonstrates that high-throughput wastewater-based surveillance can be successfully leveraged to enable creation of a rapid, large-scale early alert system for counties/districts and could be particularly useful in community surveillance in the more vulnerable populations.