SARS-CoV-2 detection from the built environment and wastewater and its use for hospital surveillance

Surfaces were sampled with the P-208 Environmental Surface Collection Prototype kit from DNA Genotek, consisting of a flocked swab and a semi-lytic nucleic acid stabilization solution for post-collection swab immersion. SARS-CoV-2 was detected by quantitative reverse-transcriptase polymerase chain reaction (qPCR) of RNA extracted from sample stabilization solution using the MagMAX Viral/Pathogen II (MVP II) Nucleic Acid Isolation Kit (Thermo Fisher Scientific, Waltham, MA). Extractions were performed in batches of up to 96 in deep-well plates according to the manufacturer’s protocol, with reagent volumes modified proportionally for a 300 μL sample input volume, and an elution volume of 50 μL. Each batch included positive (heat-inactivated SARS-CoV-2 virus) and negative (stabilization solution only) control samples to monitor extraction efficacy and potential cross-contamination. None of the control samples yielded false positive or false negative detections.

qPCR was performed with primers and a TaqMan probe targeting the N1 region of the SARS-CoV-2 nucleocapsid gene (Centers for Disease Control and Prevention 2020) (Supplementary Material 1, Table S1). Each qPCR reaction consisted of 5 μL of extracted RNA, 300 nM of each primer, 200 nM of probe, and 5 μL of 4× TaqPath 1-Step Multiplex Master Mix (No ROX) (Thermo Fisher Scientific, Waltham, MA) in a final volume of 20 μL. Thermal cycling, fluorescence quantification, and quantitative cycle (C_q) determination were performed on a Bio-Rad CFX Connect with the following cycling conditions: 25 °C for 2 min, 50 °C for 15 min, 95 °C for 2 min, and 45 cycles of denaturation (95 °C for 3 s) and elongation (55 °C for 30 s).

Limits of detection (LOD) and quantification (LOQ) were determined by analyzing 10-fold serial dilutions of SARS-CoV-2 RNA (ATCC VR-1986D; lot 70035624; 4.73 × 10³ genome copies/μL) and heat-inactivated virus (ATCC VR-1986HK; lot 70035039; 3.75 × 10⁵ genome copies/μL) standards obtained from Cedarlane Laboratories (Burlington, ON). Dilutions of the SARS-CoV-2 RNA standard were analyzed by qPCR with 14–20 replicates per dilution. For the virus standard, RNA was extracted from serial dilutions of heat-inactivated virus (7 replicate extractions at each dilution), followed by qPCR analysis.

LODs and LOQs were estimated using a LOD Calculator (Klymus et al. 2020) (R script available at: github.com/cmerkes/qPCR_LOD_Calc). The standard curves were modeled as linear regressions based on the middle two quartiles of C_q values for each standard concentration exhibiting >50% detection. The LOD was defined as the lowest input amount that was detected in >95% of replicates, and the LOQ was defined as the lowest input amount yielding reproducible C_q values with a coefficient of variation less than 0.35. The LOD and LOQ of the RNA standards were estimated by a curve-fitting approach (Klymus et al. 2020), while the LOD of the heat-inactivated virus was determined using a discrete threshold (i.e., the lowest tested input amount with >95% detection).

The swabbing method was validated in laboratory experiments measuring recovery of heat-inactivated SARS-CoV-2 virus from sheets of acrylic, stainless steel, and vinyl upholstery. Each material was spiked in triplicate with 10⁵ virus copies, pipetted as 30–40 small spots within 2.25″ × 2.25″ squares. After a 2.5 h drying period, swabs were prewetted with stabilization solution, surfaces were swabbed for 30 s, and swabs were stored in stabilization solution in collection vials. Control samples consisted of virus spiked directly into collection vials with stabilization solution and unused swabs. Following one day of storage at room temperature, RNA was extracted and SARS-CoV-2 RNA was detected by qPCR. Virus genome copy numbers were estimated using the virus standard curve that relates C_q values to input genome copies. The percent virus recovery was determined by dividing the amount of virus estimated for the surface swabbed samples by the amount of virus estimated in the control samples. Significant differences in recovery efficiencies across surfaces were determined by ANOVA and Tukey’s honestly significant difference test.

We prospectively collected built environment swabs from two large tertiary-care, academic hospital campuses (Campus A and Campus B, totaling over 1100 beds) from Ottawa, Ontario, Canada, over a 10-week period between 28 September and 6 December 2020. We obtained between 40 and 50 samples from each campus weekly, generally maintaining location (ward/unit) and surface type longitudinally; however, we did not attempt to sample the exact same item/spot longitudinally or coordinate with cleaning schedules to improve our study generalizability. For example, a computer keyboard sampled in week 1 on Unit A, may be different from a computer keyboard sampled on Unit A in week 2. At both campuses, we collected samples from two COVID-dedicated noncritical care units, two intensive care (ICU) COVID-designated units, two emergency department (ED) areas, non-COVID medical and surgical units (n = 3 at Campus A, n = 4 at Campus B), one ambulatory unit (dialysis), public/nonpatient-care spaces within the hospital (e.g., main elevators), and exterior hospital grounds (e.g., parking garage). A designated HCW space was also sampled at Campus A. Built environment samples were collected by wetting a P-208 kit flocked swab with the stabilization media and sweeping the swab tip across an approximately 2″ × 2″ area (dependent upon the item/surface size) for 30 s. The swab was then transported in the P-208 stabilization media and stored at ambient temperature to be processed/analyzed within 1 month of collection. A run-in week was performed prior to week 1, to test sample collection, logistics, and sample processing, but is not included in this analysis as a number of specimen sites were initially not included for this week, making inter-week comparisons challenging.

Specific (noncritical care) wards in the hospital were designated for the ongoing inpatient care of COVID-19 patients, and these were designated as “COVID wards”. However, we classified both the COVID Wards as well as the ICU and ED as “COVID risk units”, because they were anticipated to be managing many SARS-CoV-2 infected patients.

The visitor policy at the start of the study period limited visitation to one visitor per patient per day; however, at the start of study week 4, visitors were further restricted to essential caregivers only. Environmental cleaning was in keeping with provincial best practice and occurred once daily in both public and clinical areas sampled, with all contact points disinfected with Vert-2-Go everyday disinfectant (quaternary-ammonium based product) and floors mopped with Vert-2-Go Oxy floor cleaner (hydrogen peroxide based product). The only deviation from this was once weekly cleaning of the Campus A parking garage until sometime between study weeks 5 and 7, then once daily thereafter. Infection Prevention and Control (IPAC) policies were also consistent with provincial best practice (Public Health Ontario 2020; van Doremalen et al. 2020). Throughout the study period, all staff, visitors, and patients were screened for COVID-19 symptoms and risk factors (e.g., exposures, international travel) on entry to the hospital, in accordance with Ontario Ministry of Health (2020a) guidance. Universal masking for all staff and visitors was introduced prior to the study period. Universal eye protection (visor or goggles) for staff working in clinical areas was introduced at the start of study week 1.

Raw sewage grab samples from Campus A were taken with a 5 L bucket on a rope between 11:10 am and 12:30 pm each sampling day, which occurred once every 1 to 2 weeks during (and immediately prior to) the study period. The samples were swirled in the bucket and quickly poured into prelabelled sterile 1 L Nalgene bottles which were immediately placed into a cooler with ice packs for transport back to the lab and processed within 1 hr. Temperature of the samples, measured at the time of sampling with an infrared thermometer, for October–December were between 12 and 22 °C. A 200 mL aliquot was concentrated to 0.25–1.25 mL using the Centricon Plus-70 (MilliporeSigma, Burlington, MA) centrifugal ultrafilter with a cut-off of 10 kDa (Medema et al. 2020). The resulting concentrate was extracted using the RNeasy PowerMicrobiome Kit (QIAGEN, Hilden, Germany) according to the manufacturer’s instructions with an elution of 100μL and SARS-CoV-2 RNA was quantified by qPCR.

Primers and probes, and real-time reverse transcriptase (RT RT-PCR) methods developed by the US Centers for Disease Control and Prevention (2020) were used with the Bio-Rad CFX96 Touch Real-Time PCR Detection System. The RT RT-PCR thermal cycling protocol used to amplify the N1 and N2 target regions within the SARS-CoV-2 nucleocapsid gene, was carried out at 50 °C for 30 min, followed by 95 °C for 15 min and 50 cycles of 95 °C for 15 and 60 °C for 30 s. Reaction mixtures for RT RT-PCR included 5 μL RNA template, 4 μL of QIAGEN OneStep RT-PCR 5 × Buffer, 2 μL dNTP mix (10 mM each), 2 μL Enzyme Mix, Primers and Probes (500 nM each), and 2 μL of 4 mg/mL bovine serum albumen. Each reaction was adjusted to a final volume of 20 μL with RNase-free water. Each sample was run in duplicate. An MS2 bacteriophage matrix spike whole process control was used to qualitatively assess the efficiency of the concentration process and RNA extraction as well as confirm RT RT-PCR protocols. RT RT-PCR inhibition control using MS2 bacteriophage spiked into raw sewage RNA extracts and molecular grade water was used to assess the performance of the RNA extraction, RT-PCR, and to detect the presence of inhibitors. Quantification of the N1 and N2 gene targets in the raw sewage sample concentrates was done using standard curves created with a 2019-nCoV_N_Positive Control plasmid (Integrated DNA Technologies, Coralville, IA) Nontemplate controls were run for each mastermix preparation with molecular grade water as the template. For each week under observation, we averaged the detectable copies per millilitre from both N1 and N2 gene targets.

Daily active patient COVID-19 case counts and locations, including admitted and nonadmitted ED cases, were obtained from IPAC. Cases were those confirmed by RT-PCR. We generated a measure of weekly “active” COVID-19 cases by taking the mean of these daily active cases for each week (typically data only available for weekdays). HCW cases were not included in the analysis, as HCWs were more likely than patients to spend time on multiple units and move throughout the hospital. Also, once a HCW became symptomatic or had a positive COVID-19 test result, they would no longer remain within the hospital. Hospital outbreak data were obtained in real time from IPAC and verified post-study on the publicly available Ottawa Public Health (OPH) COVID-19 dashboard (Ottawa Public Health 2021). An outbreak was defined as per the Ontario Ministry of Health (2020b) as, “Two or more laboratory-confirmed COVID-19 cases (patients and (or) staff) within a specified area (unit/floor/service) within a 14-day period where both cases could have reasonably acquired their infection in the hospital”. New and total active COVID-19 cases in Ottawa were obtained daily from the OPH COVID-19 dashboard, and a weekly “active” caseload was calculated as the mean of the daily active cases across a given week.

To evaluate the characteristics associated with positive swab results, we performed a logistic regression model using specimen characteristics as predictors of built environment sample positivity (total or floor specific specimens). Specimen characteristics were coded as categorical variables, including unit COVID risk type, object/site, material, and study week. The outcome (sample positivity) was coded as a binary variable (1 = detected, 0 = not-detected).

We also evaluated the strength of associations between (i) proportion of hospital built environment swabs positive for SARS-CoV-2 or (ii) proportion of hospital floor swabs positive for SARS-CoV-2 and hospitalized active COVID-19 cases. We used a negative-binomial regression model with active cases as the outcome. We did not quantitatively evaluate the association between wastewater and active cases because of the limited number of wastewater data points.

For the majority of the analyses we considered the two campuses in combination, as they reflect a single health care center with transfer occurring within and between units in both campuses. We also chose not to evaluate case burden by specific unit because, in general, as soon as a case was identified on a given non-COVID risk unit, the patient would be transferred to a specific COVID Risk unit either for cohorting or due to severity of illness (e.g., ICU).

All statistical analyses were performed using R (Version 4.0.3). Values used to build the figures are in the datasets found in Supplementary Material 2. Research ethics board approval was obtained through the Ottawa Health Science Network Research Ethics Board for this study.

SARS-CoV-2 RNA was detected and quantified by qPCR analysis using the N1 primer and probe set targeting the nucleocapsid gene (Centers for Disease Control and Prevention 2020). We analyzed a dilution series of a SARS-CoV-2 RNA standard to produce a standard curve relating the C_q values to genome copy number and determine the LOD and LOQ (Fig. 1a). We estimated that reactions containing at least 3 genome copies were reliably detected under our qPCR conditions (detection rate greater than 95%), whereas precise quantification required at least 150 copies (C_q < 30) (Fig. 1a).

Analysis of the built environment swab samples involved a nucleic acid extraction step prior to qPCR analysis. We evaluated the extraction method by qPCR analysis of RNA extracted from samples of a heat-inactivated SARS-CoV-2 virus standard (Fig. 1b). The LOD for the virus standard (28 copies per reaction) was ∼10-fold higher than that of the RNA standard, which can be attributed to RNA loss during the extraction procedure, presence of PCR inhibitors in the eluant, or inaccuracies in the reported titers of the standards. Nevertheless, the procedure was sensitive and reproducible, and the results yielded a linear standard curve for converting C_q values to viral copies while accounting for inaccuracies introduced by sample processing. We validated the swabbing method by estimating the recovery of virus from surfaces spiked with heat-inactivated SARS-CoV-2 (Supplementary Material 1, Fig. S1). The mean recovery efficiencies ranged from 28% to 42% across three different surface types, which are values similar to those reported in a previous study of SARS-CoV-2 surface swabbing (Parker et al. 2020).

Standard curves generated for both N1 and N2 targets gave consistent results with PCR efficiencies between 90% and 110%, and R² values greater than 0.99. Nontemplate controls were negative for all runs. There was a shift in sensitivity from the N2 target showing lower C_q value than N1 for the first 3 weeks followed by lower C_q values for N1 for the remainder of the study period. As a result of this, the data are presented as an average of N1 and N2 gene copy numbers.

Over a 10-week period from 28 September to 6 December 2020 we systematically performed 963 built environment swabs across two tertiary care hospital sites. SARS-CoV-2 was detected from 6% of swabs (61/963), and it was commonly found from floor samples, with a detection prevalence on floors of 27% (51/188). SARS-CoV-2 was less frequently identified from elevator buttons, benches, items (e.g., hand sanitizer pumps, equipment carts), telephones, and computer keyboards.

COVID risk units, namely COVID treatment wards, ICUs, and EDs, were the most common locations to find SARS-CoV-2 in the hospital environment, with detection prevalence of 18% (17/92), 8% (10/118), and 7% (7/100), respectively (Table 1 and Fig. 2). Non-COVID and public or HCW spaces tended to have a low frequency of SARS-CoV-2 detection from the environment. During this study, there were two outbreaks involving study units, at Campus B in the ED (one patient and two HCWs; study week 2–4) and at Campus A on Ward 3 (five patients and one HCW; study week 4–8). A third outbreak occurred on a non-COVID inpatient unit at Campus A, but this unit was not sampled in the study.

Across the 10-week study period, SARS-CoV-2 detection prevalence paralleled total active cases across the two hospital campuses. Viral prevalence gradually declined over time, but with spikes in hospitalized cases being clearly reflected in built environment burden (Fig. 3a), as well as an increase at week 10 reflecting the start of hospitalizations in the second wave of the pandemic within the city. Built environment burden also paralleled citywide cases, but with less concordance (Fig. 3b). Campus B tended to have SARS-CoV-2 positivity largely restricted to COVID risk units, but Campus A seemed to have more activity in non-COVID risk units/public spaces. A multi-variable logistic regression model identified that increasing study week was associated significantly with reduced detection of SARS-CoV-2 (p < 0.05), and that COVID Risk units, Campus A, and some specific materials and objects were significantly associated with increased SARS-CoV-2 detection (Table 1 and Table 2).

Hospital floors were the most common surface from which SARS-CoV-2 was detected. We looked specifically at prevalence on floors and its relationship with cases within the hospital and citywide (Fig. 4). Restricting analysis to only floors showed they parallel the findings seen with all samples (Fig. 3), with stronger correlation with active hospital cases. Hospital wastewater SARS-CoV-2 viral copies per millilitre (Fig. 5) followed a similar trajectory as built environment sampling, but the decline in wastewater levels appeared to predate the declines in both built environment detection as well as Campus A and citywide case burden.

We also evaluated the strength of association between (i) all built environment swab positivity or (ii) floor swab positivity with hospital case burden (Supplementary Material 1, Table S2). We found that floor swabs best described the hospital case burden, having a measure of association with the greatest significance and a model with the best fit.