Improving wastewater-based epidemiology performance through streamlined automation

Abstract

Wastewater-based epidemiology (WBE) has enabled us to describe Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infections in populations. However, implementation of wastewater monitoring of SARS-CoV-2 is limited due to the need for expert staff, expensive equipment, and prolonged processing times. As WBE increases in scope (beyond SARS-CoV-2) and scale (beyond developed regions), there is a need to make WBE processes simpler, cheaper, and faster. We developed an automated workflow based on a simplified method termed exclusion-based sample preparation (ESP). Our automated workflow takes 40 min from raw wastewater to purified RNA, which is several times faster than conventional WBE methods. The total assay cost per sample/replicate is $6.50 which includes consumables and reagents for concentration, extraction, and RT-qPCR quantification. The assay complexity is reduced significantly, as extraction and concentration steps are integrated and automated. The high recovery efficiency of the automated assay (84.5 ± 25.4%) yielded an improved Limit of Detection (LoD_Automated=40 copies/mL) compared to the manual process (LoD_Manual=206 copies/mL), increasing analytical sensitivity. We validated the performance of the automated workflow by comparing it with the manual method using wastewater samples from several locations. The results from the two methods correlated strongly (r = 0.953), while the automated method was shown to be more precise. In 83% of the samples, the automated method showed lower variation between replicates, which is likely due to higher technical errors in the manual process e.g., pipetting. Our automated wastewater workflow can support the expansion of WBE in the fight against Coronavirus Disease of 2019 (COVID-19) and other epidemics.

Graphical Abstract

1. Introduction

Wastewater-based epidemiology (WBE) is a promising tool for the monitoring of disease outbreaks, including the COVID-19 pandemic [1]. Fecal shedding of SARS-CoV-2 virus is present in more than 60% of the patients who tested positive for COVID-19 [2] and has been observed in patients that are asymptomatic or pre-symptomatic [3]. In fact, WBE-guided clinical testing of several university dormitories showed that around 80% of positive cases were asymptomatic [4]. Thus, WBE-driven surveillance may result in a more accurate snapshot of infection within a building/community [5], [6], particularly in situations where individual testing is performed at home or deferred. Furthermore, given that WBE can pinpoint communities/facilities as “hot spots”, clinical testing can be deployed in a more efficient manner, especially when overall prevalence is low [7], which is particularly important in regions or countries where healthcare resources are limited [8]. In addition to efficiently targeting infected individuals, WBE can effectively reveal the trends of infection among populations [9], as concentrations of SARS-CoV-2 RNA in wastewater are shown to correlate with the number of COVID-19 cases in wastewater sewersheds [10]. Indeed, several groups around the world were able to detect and monitor the levels of SARS-CoV-2 in wastewater samples in the early days of COVID-19 [11], [12], [13]. In summary, wastewater surveillance has shown its merit as a sustainable and cost-effective tool that can be fitted into current infrastructure for future monitoring of pandemics [14], [15].

While WBE has provided valuable guidance to public health stakeholders during the COVID-19 pandemic, WBE utilization remains disproportionately biased toward regions that can support the infrastructurally-complex demands of current WBE workflows [8], [16]. Therefore, as implementation of wastewater surveillance for SARS-CoV-2 increases in scope and scale, there is a need to make the process simpler, faster, and more reliable [17]. The process workflow can be divided into sampling, concentration/extraction of the biomarker, and quantification using tools such as RT-qPCR ( Fig. 1). Different sampling techniques such as direct collection from a wastewater stream (termed a “grab sample”) [18], composite collection over time using autosamplers [19], and passive sampling techniques (e.g., Moore swabs) [20], [21] are being used currently. Common concentration methods include ultracentrifugation, ultrafiltration, Poly (ethylene glycol) (PEG) precipitation, adsorption-extraction, and flocculation [22], [23], [24], [25], and these steps are typically followed by RNA extraction using solid phase extraction [26]. In most cases, the concentration and/or extraction steps are process bottlenecks, as conventional methods are labor-intensive and often include multiple complex filtration and centrifugation steps, which are difficult/expensive to automate.

Fig. 1.

The overall process of wastewater surveillance of SARS-CoV-2. Process bottlenecks are depicted in the black box and involve the purification of RNA, which includes concentration and extraction steps. An automated method can help overcome this bottleneck.

Despite the inherent difficulty in automating key WBE processing steps, some promising progress has been made. For example, Yang et al. developed a semi-automated aluminum hydroxide adsorption-precipitation method [27], where $\mathit{Al}{\mathit{Cl}}_3$ solution is added to the sample and mixed mechanically and then filtered using the automated equipment. The precipitates are then collected using a manual scrapper and virus is eluted using EDTA-2Na dissolution. The process takes about two hours for one sample with a recovery rate of $4.8\pm 1.4\%$at a virus concentration of 100 copies/mL. In another study [28], an automated RNA extraction method based on the Maxwell RSC instrument was used after the concentration steps. Although the wash steps for RNA extraction were performed by an automated platform, the concentrated sample required 16000 g centrifugation before extraction, which makes the process dependent on another expensive instrument. Recently, an automated process containing both concentration and extraction steps was reported using magnetic Nanotrap™ particles (Ceres Nanosciences) paired with ThermoFisher’s KingFisher Flex automated liquid handler system [29]. Following magnetic enrichment of SARS-CoV-2 virions, RNA was purified using existing commercial extraction kits (e.g., MagMAX viral/pathogen nucleic acid isolation kit) [30]. In these cases, cost can be a drawback as the commercial RNA extraction kits and the KingFisher system are expensive.

Exclusion-based Sample Preparation (ESP) is a process that has recently been used to simultaneously concentrate and extract viral RNA from SARS-CoV-2 [31]. ESP has been used in other applications to simplify the isolation of paramagnetic particle (PMP)-bound analyte from complex samples [32]. In brief, ESP uses the surface tension of the reagents and the hydrophobicity of the consumables to stabilize a series of air/aqueous interfaces. Pulling PMPs through these interfaces results in a high efficiency purification and concentration of PMP-bound analyte from the background matrix [33], [34], [35]. The relatively simple motion in ESP technology enables straightforward automation while decreasing the required time and complexity of the workflow. ESP-based WBE was successfully implemented by Strike et al. [31] using a completely manual method. While this study yielded promising results, we anticipate that a low-cost automation method may improve performance (e.g., analytical sensitivity, precision) and promote adoption due to a workflow that is both simpler and faster. Here, we report the development and implementation of an ESP-based wastewater concentration/extraction system that is based on the low-cost Gilson PIPETMAX® liquid handling robot. Compared to conventional concentration and extraction methods, our automated protocol does not require complex instruments (e.g., high speed centrifuges, ultrafiltration systems) and is more cost effective (compared to kit-based RNA extractions), because it merges both extraction and concentration processes in one step.

2. Materials and methods

2.1. Wastewater sampling

Wastewater samples from nine locations in central and eastern Kentucky were collected in September and October 2021. Effluent wastewater samples from two nursing homes in Lexington, Kentucky with 150 and 189 residents were collected at the manhole connected to the sewage system of the building. Moreover, ten influent samples were collected from eastern Kentucky wastewater treatment plants (WWTPs) serving 2500 to 22,000 residents. All the samples were untreated. Manual wastewater analysis showed a sharp rise in the SARS-CoV-2 concentrations in these samples during the Delta variant wave in September and October 2021 (data not shown). We collected 24-hour composite wastewater influent samples from the WWTPs and 24-hour composite effluent wastewater samples from nursing homes. The autosampler wastewater reservoir was shaken and 200 mL of sample was transferred on ice to the lab on the same day where they were stored at $4℃$ and processed the same or next day.

2.2. Manual nucleic acid extraction/concentration

We extracted SARS-CoV-2 RNA directly from 250 $\mathrm{\mu }\mathrm{L}$ of wastewater sample as described in Fig. 2A and [31]. In brief, viral lysis was performed by 3:2 vol ratio addition of lysis buffer containing 4 M guanidine thiocyanate (ThermoFisher) and 10 mM 4-morpholineeethanesulfonic acid (MES) sodium salt (Sigma Aldrich) dissolved in 1:1 v/v absolute ethanol:water. Two different types of paramagnetic particles (PMPs) with different diameters (Serasil-Mag™, #29357369 and #29357374, Cytiva) were then added to the lysed sample and incubated for 20 min at 50 ℃ with vortexing every 5 min. In the next step, tubes were tumbled for 20 min and centrifuged to collect the beads. The beads were washed in two wash buffers. The first wash buffer contained 1 M guanidine thiocyanate (Thermo Fisher), 10 mM Tris buffer pH 8 (Thermo Fisher), and 1% Tween 20 solution (Sigma Aldrich) in distilled water. The second wash buffer had 10 mM Tris buffer pH 8 dissolved in absolute ethanol. The wash steps were performed on polypropylene Extractman plates (22100008, Gilson) using Extractman (22100000, Gilson), a commercially available device based on ESP technology. As the Extractman head slides over wells on the extraction plates, the upper magnets immobilize the PMPs on a hydrophobic strip (22100007, Gilson) and when the head slides over to the next well, the lower magnet pulls the PMPs from the strip into the wash well (Fig. 2A). With movement of the lower magnets, the PMPs are magnetically washed several times in wash buffers (15 times in the first wash buffer and 5 times in the second wash buffer). After washes, the PMPs drop into the elution buffer containing 100 µL of nuclease-free water. These beads were collected in 1.5 mL microcentrifuge tubes and incubated at 70 ℃ for 20 min to promote elution. After the elution, the PMPs were immobilized on a magnetic rack and the RNA was collected for downstream quantification via RT-qPCR.

Fig. 2.

A) Schematic of ESP-based RNA extraction method. Samples containing magnetic beads are loaded on the plate. RNA is purified with magnetic beads which are transferred between buffers using upper and lower magnets. B) PIPETMAX® (Gilson, Inc.) liquid handling robot. The ESP technology is fitted in the benchtop instrument.

2.3. Automated nucleic acid extraction/concentration

We developed an automated version of the ESP-based RNA concentration and extraction method using the PIPETMAX® (Gilson, Inc.) liquid handling robot shown in Fig. 2B. The hardware and software of this instrument is customizable, enabling the use of the same consumables and method described for the manual process. The lysis buffer, PMPs, wash buffers and nuclease-free water were preloaded in microcentrifuge tubes (source tubes) located in a tube rack (Gilson 424) inside the instrument. After receiving the sample in the lab, $200\mathrm{\mu }\mathrm{L}$ of untreated wastewater was loaded in clean microcentrifuge tubes and placed inside the instrument. The robot started the extraction by mixing $200\mathrm{\mu }\mathrm{L}$ of sample with $300\mathrm{\mu }\mathrm{L}$ of lysis buffer and $20\mathrm{\mu }\mathrm{L}$ of PMPs on a heat block followed by 4 mixes (aspirate and dispense) every 5 min to ensure viral lysis and RNA capture. The microcentrifuge tube heat block (Thermo Scientific, #88870101) was heated to 70 ℃ using a digital block heater (Thermo Scientific, 88870001) and placed inside the PIPETMAX. Between the mixing steps, the robot filled the Extractman wash plate with first and second wash buffers and nuclease-free water. Upon finishing the lysis step, the robot removed the samples from the heat block and filled the sample wells on the Extractman wash plate. The Extractman strip loaded automatically onto the robot head and during the wash steps, the robot captured PMPs from sample wells, washed them four times in wash buffers using the magnets inside its head, and released them in the elution buffer which is 100 µL of nuclease-free water. The elution buffer with beads was manually collected and incubated at $70℃$ for 20 min. The eluted RNA was collected using a magnetic rack to immobilize the PMPs. A video of this process is shown as Supplemental Video S1 in the online version. Fig. 3 illustrates the steps required for the complete WBE lab analysis process, highlighting those that are automated.

Fig. 3.

Flowchart of the process linking raw wastewater to RT-qPCR analysis. The automated part (bound by the dashed line) is performed by the automated system.

To validate the performance of our automated protocol, confirmed negative wastewater samplers were spiked with Heat-Inactivated SARS-Related Coronavirus 2 (NR52350, Batch No. 7003663, Isolate USA-WA1/2020, BEI Resources). The stock concentration of this standard was quantified by the BEI resources using the droplet digital PCR (ddPCR) and is equal to 3.4E8 genome equivalents per mL. The standard was aliquoted to make 1:10000, 1:1000, 1:100, and 1:10 dilutions in nuclease free water. Then, 29.4 µL of each dilution is spiked in 10 mL of wastewater, resulting in 1E2, 1E3, 1E4, and 1E5 copies per mL of wastewater, respectively. Moreover, 88.2 µL of the 1:10000 dilution is added to 10 mL of wastewater to result in 300 cp/mL of wastewater. Given the wide range of concentrations of SARS-CoV-2 present in wastewater samples, shown by other studies [36], [37], [38], we are aiming to show the capability of our method across this wide range of concentrations. After spiking, the samples were vortexed for 1 min and then kept at 4 ℃ for at least 30 min before starting the manual/automated process. Samples are briefly vortexed before processing to break up bigger solid particles. The negativity of the wastewater was confirmed using manual extraction and RT-qPCR. In these experiments, we ran 8 replicates of each sample to account for the heterogeneity of wastewater samples.

2.4. Detection and quantification using RT-qPCR

We performed one-step RT-qPCR assays using a LightCycler 480 II (Roche Diagnostics). SARS-CoV-2 RNA in extracted samples was quantified using the N1 gene primer and probe sequence shown in Table S1, as recommended by the CDC [39]. The total volume of the reaction mixture was $20\mathrm{\mu }$L, consisting of $10\mathrm{\mu }\mathrm{L}$ of the extracted RNA, $5\mathrm{\mu }\mathrm{L}$ of TaqMan 4X Fast Virus 1-Step Master Mix (Applied Biosysyems), $1\mathrm{\mu }\mathrm{L}$ of a 20X mix of primers and probe, and $4\mathrm{\mu }\mathrm{L}$ of nuclease-free water. The N1 primers and probe were synthesized by ThermoFisher at a final concentration of 60X, where the final 1X concentration in the reaction is equal to 900 nM of each primer and 250 nM of the probe. The TaqMan probe was synthesized with a FAM fluorophore and ThermoFisher proprietary MGB quencher. Thermal cycling conditions included reverse transcription at $50℃$ for 5 min and a hot start of $95℃$ for 20 s, followed by 45 cycles of $60℃$ for 1 min and $95℃$ for 20 s, while measuring the real-time fluorescence signal from FAM. The reaction was validated on each PCR plate with SARS-CoV-2 Genomic RNA (NR-52508, Isolate USA-CA4/2020, BEI Resources) as the positive control. The same nuclease free water used in the process was used in the no-template control (NTC). Analysis was done on the raw RT-qPCR data using the Roche software version 1.5.1.62. The software employs a maximum second derivative methodology to calculate and assign Cq values, after automatically handling the background baseline. After analysis, the results are verified visually since rarely with this software, a background signal spike can create a falsely reported result.

The RT-qPCR standard curve parameters for SARS-CoV-2 N1 gene were calculated by spiking serial 1:10 dilutions of Heat-Inactivated SARS-CoV-2 virus (NR52350, Isolate USA-WA1/2020, BEI Resources) in nuclease free water as described by Bivins et al. [40]. As shown in Fig. S1, the standard curve had a slope between − 3.21 and − 3.38 with a y-intercept between 39.52 and 40.24. The correlation coefficient (r²) was between 0.9977 and 0.9993 and the amplification efficiency was between 97.54% and 104.60%. This standard curve is used to convert the Cq from extracted wastewaters to copies per reaction. The manual process starts with 250 µL of wastewater and the elution volume is 100 µL, from which 10 µL is used in the PCR reaction. Using these volumes, we can convert copies per reaction to copies per mL of wastewater using a conversion rate of 40. The conversion rate for the automated method is 50 since we start with 200 µL of wastewater. The lowest number of copies that the PCR assay can detect 100% of time was measured to be 3 copies in each reaction (Fig. S2). As a result, our PCR assay limit of detection (PCR LoD) is 3 copies/reaction.

To determine the presence of inhibitors in the extracted RNA samples from automated and manual methods, we spiked three SARS-CoV-2 negative wastewater samples with varying levels of turbidity as shown by Fig. S3. Since SARS-CoV-2 RNA is not naturally present in the chosen wastewater samples (proven by RT-qPCR), this target can be used as an external control for the inhibition test. Moreover, since the SARS-CoV-2 N1 assay is used further in the study, analyzing the inhibitory effects of wastewater on the amplification of this gene is our purpose. Each wastewater was spiked with the same volume of heat-inactivated SARS-CoV-2 (NR52350, Isolate USA-WA1/2020, BEI Resources). All samples were processed in triplicates. Serial 1:4 dilutions of the extracted samples were prepared and amplified. In theory, each 1:4 dilution should result in a Cq drop of 2. As shown in Fig. S3, in which the x-axis is in log base 2, the R² values are higher than 0.9. As a result, there was no indication of inhibitors in the extracted RNA samples from both methods. The high number of washes in our methods (2 ×15 wash cycles in the first wash buffer and 2 ×15 wash cycles in the second wash buffer), which is enabled by the ESP technology, can effectively wash out the inhibitors.

2.5. Statistical Analysis

The LoD of both manual and automated methods was calculated based on the linear regression model and using equation (LoD=3 *SD/b), where SD is the standard deviation of y-intercept and b is the slope of linear calibration curve [41]. We compared the efficiency of the automated method versus its manual analog using Mann-Whitney U-test (Wilcoxon rank-sum). Our null hypothesis was that the probability of getting higher RNA concentration from automated method is equal to the probability of getting higher RNA concentration from its manual analog. In this regard, a p-value less than 0.05 rejects the null hypothesis and concludes that the probability of getting higher RNA concentration from one method is higher, resulting in a better efficiency for one method. For further differentiation between the two methods, standardized effect sizes (d) were calculated using Eq. (1). A positive effect size indicates a higher efficiency for the automated method and a negative effect size indicates a higher efficiency for the manual method. The effect sizes were calculated based on Cohen [42] and interpreted using the following metrics: effect sizes of less than 0.4 are considered “small”, those from 0.4 to 0.7 are “moderate”, and those above 0.7 are “large”. We also compared the variability of our methods using the coefficient of variation (CV), which is calculated by dividing the standard deviation between replicates by means of replicates in each method (CV=SD/M). A higher CV means a higher variability in the measurement method. [42].

$$d=\frac{Mea{n}_{Automated}-Mea{n}_{Manual}}{S{D}_{pooled}},S{D}_{pooled}=\sqrt[2]{S{D}_{Automated}^2+S{D}_{Manual}^2}$$ (1)

The recovery efficiency of both manual and automated processes is estimated using Eq. (2). The “spiked concentration” is calculated based on the dilution factors of the standard control which has a known concentration of heat-inactivated SARS-CoV-2. Moreover, the Cq values from the analyzed wastewaters were used to calculate the “recovered concentration” using the standard curve. Then, the arithmetic average and standard deviations (N = 8) of recovery efficiencies at each spiked concentration is calculated to compare the two methods.

$$\mathit{Recovery}\mathit{Efficiency}=\frac{\mathit{Recovered}\mathit{Concentration}}{\mathit{Spiked}\mathit{Concentration}}$$ (2)

We used a Bland-Altman plot to assess the agreement between the automated and manual methods [43]. This plot shows the difference between the methods against their mean. Since we do not know the true concentration of SARS-CoV-2 RNA in the wastewater, the mean of the two methods is our best estimate. The line of mean of differences and its 95% confidence interval (Mean±1.96 *SD) is also drawn, as the limits of agreement advised by Bland et al. [43]. A Bland-Altman plot can reveal possible biases in measurement techniques.

3. Results and discussion

3.1. Benchmarking with spiked samples

The automated and manual extraction processes successfully extracted SARS-CoV-2 RNA from spiked wastewater samples down to 100 copies per mL of wastewater with the automated process outperforming the manual process. This hypothesis was benchmarked using spiked negative wastewater as described in methods sections. At 100,000 and 10,000 copies per mL of wastewater, all 8 replicates had detectable SARS-CoV-2 RNA ( Table 1 and Fig. 4) with the manual and automated processes. At 1000 copies per mL of wastewater, all replicates from the automated process had detectable SARS-CoV-2 RNA while 7 out of 8 replicates from the manual process had detectable SARS-CoV-2 RNA. At 100 copies per mL of wastewater (2.5 copies per PCR reaction), 7 and 6 of 8 replicates had detectable SARS-CoV-2 RNA from the automated and manual method, respectively. At 100 copies per mL of wastewater (2.5 copies per reaction), we are already lower than our PCR LoD (3 copies per reaction), which is one of the reasons as to why not all replicates are positive. To overcome this stochastic variability, we perform a relatively large number of replicates (N = 8) to overcome this effect. The LoD was calculated based on the linear regression model described in the methods and [41]. However, since the samples were analyzed in 8 replicates, concentrations lower than the LoD might be reported since measurable (>LoD) replicates were averaged with undetectable (reported as 0 cp/mL) replicates.

Table 1.

Detection ratio and LoD of heat-inactivated SARS-CoV-2 virus in wastewater using both manual and automated methods.

Method	Number of positive replicates (out of 8) at different concentrations of spiked Heat-Inactivated SARS-CoV-2 in wastewater (copies/mL of wastewater)				LoD (copies/mL of wastewater)
	1E5	1E4	1E3	1E2
Manual	8/8	8/8	7/8	6/8	205.9
Automated	8/8	8/8	8/8	7/8	40.3

Fig. 4.

Recovery efficiency of automated and manual ESP at different spiked concentrations of heat-inactivated SARS-CoV-2 in wastewater. The error bars show standard deviations (N = 8). Effect sizes (d) are calculated based on Cohen, (d>0.7 indicates a large effect size, suggesting that the automation method has higher recovery efficiency). *Statistically significant determined by Mann-Whitney U-test (p-value<0.05).

Fig. 4 shows the recovery efficiency of both methods at different concentrations of SARS-CoV-2, calculated based on Eq. 2. At each concentration, 8 replicates are processed by each method to help us compare the methods. As seen in Fig. 4, both methods exhibited recovery efficiencies higher than 60%. At spiked concentrations of 100,000 and 10,000 copies/mL, the effect sizes were 0.579 (p-value=0.0583) and 1.059 (p-value=0.0104). At these concentrations, the effect sizes were either moderate (0.4 <d<0.7) or large (d>0.7), indicating that the automated method exhibited higher recovery efficiency compared to the manual method. However, a Mann-Whitney U-test comparison of these methods shows statistical significance only at 10,000 cp/mL. At lower concentrations (1000, 300 and 100 cp/mL), the effect sizes were small, and all p-values were greater than 0.05, indicating no statistical difference between the automated and manual methods.

The automated method was more precise at lower concentrations, as illustrated by a lower coefficient of variation (CV). At 1000, 300 and 100 copies per mL, the CV for the automated method was 0.199, 0.343, and 0.704, respectively, while the CV for the manual method was 0.793, 0.595, and 0.805. Measurement variation was likely due to a combination of biological (e.g., heterogeneity across sample replicates) and technical (e.g., pipette error, particularly with the manual method) factors. Thus, it is unsurprising that CV increases substantially for the manual method as the mean RNA concentration decreases toward the assay LoD. It should be noted that both methods have LoDs that allow them to effectively observe the trends in SARS-CoV-2 concentrations in wastewater which could potentially predict the case trends in a community (e.g., our manual method was recently used to successfully identify SARS-CoV-2 infection in university residence halls [31]). Our analysis showed that the average concentration of SARS-CoV-2 RNA during the Delta variant outbreak was 234 copies per mL, while the average concentration in a two-month period before this outbreak was 33 copies per mL. Moreover, the average concentration of SARS-CoV-2 RNA during the recent BA4/5 outbreak was 665 copies per mL of wastewater. The range of observed viral load in this study was 78–3555 copies per mL, which was during the Delta variant outbreak.

Using spiked samples, the automated and manual ESP methods were also compared with the AllPrep DNA/RNA Micro Kit (Qiagen, 80284), which is a widely used kit in WBE. The results in Table S2 showed high recovery efficiency (>60%) for all methods. However, the AllPrep kit has low throughput (2 h for 4 replicates/samples) and include several high-speed centrifugation steps (>13000 g), which makes it difficult to automate.

3.2. Comparison of manual and automated methods on samples with endogenous levels of SARS-CoV-2

After benchmarking the automated method, 12 wastewater samples with endogenous levels of SARS-CoV-2 were processed. The processing steps for the samples from WWTPs and nursing homes are the same. Our technology was able to detect SARS-CoV-2 in wastewater samples from residential buildings (nursing home) and WWTPs. We did not see any significant difference in the concentrations of SARS-CoV-2 among WWTPs and residential buildings wastewater samples. We assessed the correspondence between the measured SARS-CoV-2 RNA concentrations using the manual and automated ESP methods. Fig. 5A shows a scatter plot of the measurements from both methods, where the X-axis represents the automated method, and the Y-axis represents the manual method. The line of equality (y = x) is drawn to better compare the two protocols. The Pearson coefficient (r) is 0.953 (P-value=0.0025) showing a strong positive relation between the two methods. Note that each point on the figure is a wastewater sample and the RNA concentrations are the arithmetic average of eight replicates. The Bland-Altman plot is shown in Fig. 5B to assess the agreement between the two methods. Since the concentrations of SARS-CoV-2 in these samples range from 78–3555 cp/mL, the log-transformed data is shown. All the samples fall between the upper and lower limits of agreement, suggesting a good agreement between the two methods. Note that for 8 out of 12 samples (66.7%) the difference (automated concentration minus manual concentration) is positive, meaning that the automation protocol more often yields higher concentrations of SARS-CoV-2 RNA. Since there is no evidence of a pattern in the Bland-Altman plot, there is no indication of a concentration bias in our methods (e.g., a bias where the automated method outperforms the manual method only at certain concentrations). This analysis shows that the automated method has an equal or higher analytical sensitivity than the manual method.

Fig. 5.

Comparison of measured SARS-CoV-2 RNA concentrations in wastewater samples by automated and manual protocols. Each point shows the averaged RNA concentration across 8 replicates of one wastewater sample. A) Scatter plot of the data with line of equality B) Bland-Altman plot of the data.

To compare the precision of the two methods, the scatter plot of the coefficient of variation (CV) is shown in Fig. 6A. The line of equality helps us compare the within-sample variability (measured SARS-CoV-2 RNA in each replicate) between methods for each sample. In 10 of 12 (83%) samples, the automated method showed a lower CV, indicating that the automated method was more precise. Fig. 6B shows the CV of each method for each sample against the measured concentrations of SARS-CoV-2. The reason behind the high CV is the highly variable recovery efficiency of concentration/extraction methods due to the heterogeneity of wastewater, shown in other studies [44], [45]. Another reason is that at low concentrations of SARS-CoV-2, samples contain numbers of target lower than the PCR LoD, which increases the chance of negative replicates. As a result, we perform eight replicates of our wastewater samples to ensure that we get at least one positive measurement for positive-but-low samples. However, given that these samples are near the assay limit of detection (and represent only a few copies per PCR reaction), stochasticity naturally increases the CV at low concentrations. This effect is shown in Figs. 4 and 6B.

Fig. 6.

Comparing the precision of automated and manual methods. A) Scatter plot of Coefficient of Variation (CV) of methods with line of equality. B) CV against SARS-CoV-2 concentrations for both methods.

4. Conclusion

The results have shown that our automation improved the LoD of SARS-CoV-2 (from 205.9 cp/mL for manual to 40.3 cp/mL for automated), resulting in increased analytical sensitivity. The wastewater SARS-CoV-2 RNA concentrations from the automated process correlated strongly (r = 0.953) with the concentrations reported from the manual process, and the Bland-Altman plot (Fig. 5B) suggests increased analytical sensitivity with the automated approach. Wastewater heterogeneity can result in big variations between replicates of the same sample; however, our automated approach had lower coefficients of variation (Fig. 6) compared to the manual approach, suggesting improved measurement precision with automation. By automating an established manual ESP-based WBE process, we also reduced turnaround time by 43% (40 min for automated compared to 70 min for manual), while maintaining the high recovery efficiency and utilizing existing consumables and reagents. The details for the breakdown of the time are shown in Table S3. Table 2 compares our automated method with manual ESP and other automated methods.

Table 2.

A comparison between automated methods for detection of SARS-CoV-2 in wastewater. Since each study calculates the LoD differently, the lowest detected viral load (RNA copies per mL of wastewater) is reported.

Method	Pre-Treatment	Concentration	Extraction	Time*	Recovery Efficiency* *	Lowest Detected Viral Load (copies/mL of wastewater)
Adsorption –Precipitation[27]	Manual	Automated	Manual	2 h	4.8 ± 1.4%	64
Maxwell RSC RNA Extraction[28]	Manual	Manual	Automated	3 h	30.2 ± 17.7%	36
King Fisher Flex RNA Extraction[30]	None	Manual	Automated	2.5 h	60.6 ± 31.2%	80
King Fisher Flex RNA Extraction + Ceres Nanotrap™[29]	Manual	Automated	Automated	3 h	27.0 ± 8.0%	∼100
Manual ESP[31]	None	Manual	Manual	70 min	75.5 ± 38.1%	206
Automated ESP	None	Automated	Automated	40 min	84.5 ± 25.4%	40

*The time it takes from raw wastewater to extracted RNA ready for PCR reaction.

* * Average recovery efficiency over the reported range of viral concentrations.

As WBE increases in scope (beyond SARS-CoV-2) and scale (into regions with a diversity of preexisting infrastructure), there is a need to make WBE analysis faster, simpler, and more robust. Traditional WBE techniques for virus concentration and extraction are labor-intensive, have low sample processing capacities and require specialized equipment and personnel. With the increasing demand for wastewater surveillance of COVID-19, several automated sample processing methods have been developed in recent years, providing a significant improvement in WBE. However, there is still a need for further optimization of these protocols and reduction of costs to make them more accessible and efficient. While existing WBE technologies are inherently difficult to automate, we have demonstrated that an ESP-based WBE protocol can be automated in an effective and straightforward manner using a low-cost benchtop automated liquid handler. Table 3 compares the cost for some of the conventional and automated methods. The details for the cost analysis of ESP methods are presented in Tables S4, S5 and S6, while details on the cost calculation for the other methods can be found in references [46], [47]. The automated methods have higher startup costs due to the need for the automated instrument. However, by amortizing the instrument startup costs, we can conclude that this cost would be less than a $ 1.00 per replicate/sample. For the automated ESP, the startup cost per sample/replicate would be $ 0.32, by amortization as shown in Table S6. Consequently, the primary expenses are the reagents and consumables. One of the major drawbacks of these methods is the use of commercial extraction kits, which can be costly. We were able to reduce this cost greatly using the ESP technology. When including the RT-qPCR reagents for quantification of SARS-CoV-2, the total assay cost per replicate/sample for our automated method is $6.50. In summary, automation of an ESP-based protocol has streamlined workflow while simultaneously improving performance.

Table 3.

The startup cost included the instruments and equipment prices. The reagent and consumable cost from raw wastewater to extracted RNA ready to be quantified with RT-qPCR or ddPCR.

	Method	Startup Cost ($)	Reagent and Consumable Cost ($ per replicate/sample)
Manual	Electronegative Membrane with Bead Beating[46]	15,368	7.02
	Electronegative Membrane with Elution[46]	11,160	11.32
	PEG[46]	20,288	16.54
	Ultrafiltration[46]	9000	17.62
	Manual ESP	2874	3.81
Automated	Promega Maxwell RSC[47]	N/A	∼ 40.00
	King Fisher Flex[47]	N/A	∼ 40.00
	Automated ESP	25,939	3.81

CRediT authorship contribution statement

Mohammad Dehghan Banadaki: Conceptualization, Methodology, Investigation, Software, Writing – original draft. Soroosh Torabi: Methodology, Investigation, Writing – review & editing. William Dalton Strike: Investigation. Ann Noble: Investigation, Resources. James W. Keck: Writing – review & editing, Funding acquisition. Scott M. Berry: Conceptualization, Methodology, Writing – review & editing, Project administration, Funding acquisition.

Declaration of Competing Interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests. Scott M. Berry reports a relationship with Salus Discovery that includes equity. Scott Berry has an ownership interest in Salus Discovery, LLC, which has licensed the ESP technology described in the text. Dr. Berry has also been granted patents related to the ESP process.

Editor: P. Fernández−Ibáñez

Appendix A. Supplementary material

Supplementary material

.

Supplementary material

.

Data Availability

Data will be made available on request.