Our experimental design sought to answer three primary questions: whether the efficacy of SARS-CoV-2 detection is influenced by the following three variables: 1) does the swab storage solution (95% EtOH vs 91% isopropanol) impact the sensitivity of detection; 2) which sample fraction, swab head or eluent, provides better detection fidelity; and 3) does the swab head material type matter? To do this we designed a series of experiments to compare RNA recovery as measured by RT-qPCR using multiple swab types and storage solutions. We additionally performed environmental sampling in a hospital environment with a subset of swab types for comparison.
Feasibility of 95% EtOH for sample storage and extraction from use of swab head rather than eluent
To evaluate the feasibility of switching from VTM to a more readily-available, viral-inactivating sample collection solution, we compared the extraction efficiency of synthetic-tipped plastic-shafted nasopharyngeal (NP) SYN swab samples stored in VTM versus nasal samples collected using SYN swabs stored in 95% ethanol (EtOH) collected from two separate cohorts (Methods). When mirroring the CDC protocol, which calls for extraction from 200 µL of the eluent from VTM surrounding NP swabs, we had significantly lower recovery of human RNA in 95% EtOH eluent compared to the RNA copy concentrations in the eluent of SYN swabs collected in VTM from a separate cohort of COVID-19 positive patients (Figure 1a; one-way ANOVA with Tukey’s multiple comparison, VTM eluent vs. EtOH eluent p<0.001). However, similar levels of human RNA were recovered when extracting from the EtOH-preserved swab head itself (Figure 1a; one-way ANOVA with Tukey’s multiple comparison, VTM vs. EtOH swab p=0.3). In a subset of seven COVID-19 patient nares samples stored in 95% EtOH, we also detected significantly higher SARS-CoV-2 viral load in RNA extracted from the swab head versus eluent (Figure 1b; one-tailed paired Student’s t-test p=0.03).
To more quantitatively determine the effects of alcohol-based preservation media, we extracted RNA from a pure, commercial sample of human RNA added to water, EtOH, or 91% isopropanol, and found no impact on extraction efficiency (Figure 1c; one-way ANOVA, p>0.05). Next, we examined whether alcohol storage solutions had any protective properties of RNA, specifically a possible inhibitory effect on RNases that might be present in the environment. If alcohol inhibits the RNaseA, one would expect to see similar amounts of RNA as without RNaseA added in control experiments. In the presence of abundant RNaseA added to the solution, 95% EtOH protected both human RNA and SARS-CoV-2 RNA better than 91% isopropanol. Only a moderate decrease in total RNA recovery was observed, at the most extreme concentration of 25 µg per reaction, which is equivalent to the standard amount used for RNA removal during DNA extraction (Figure 1d).
Comparison of alternative swab types against standard CDC approved synthetic swab
Given that the performance of eluent vs. swab-based extractions in each alcohol may depend on the swab tip and body composition, we next tested RNA recovery from both the swab head and the surrounding eluent from a range of medical- and consumer-grade swabs (Methods) (Figure 1e). The RNA yield was highest from swab heads compared to eluent regardless of the swab type and whether stored in 95% EtOH (p<0.0001, U=37, Mann-Whitney) or 91% isopropanol (P<0.0001, U=28, Mann-Whitney) (Figure 1e-f). The storage solution did not impact RNA quality (Supplemental Figure 1b, Mann-Whitney, p>0.05), although swab type had a minor impact (Supplemental Figure 1c, Kruskal-Wallis p=0.03, KW=12.17) . To compare impacts of various alternative swabs, we normalized the recovery of each test to SYN eluent, indicated by ‘1’ (Figure 1e), which is the standard CDC approved method. Thus, any sample with a value greater than 1 would indicate an enhanced recovery of RNA, whereby less than 1 indicates a lower recovery of RNA compared to the standard. The RNA recovery ratio of swab-to-eluent and total yield varied among swab type (p<0.0001, KW=28.37, Kruskal-Wallis for eluent, and p<0.0001, KW=15.43, Kruskal-Wallis for swab heads) (Supplemental Figure 2). This difference in performance may relate to the differences in observed adsorption capacity across swab types (Shapiro-Wilkes p=0.1, w=0.8357; ANOVA p=0.0001, F=7.5, R2=0.56). TMI adsorbed the least (84.5 μL, 20.4; mean, SD) followed by plastic shafts (SYN: 141 μL, 23.1; CGp: 143.3 μL, 29.9) (Supplemental Figure 3). CGp swabs had the highest recovery of RNA from the swab head, while TMI swabs had the highest overall recovery of RNA when combining both eluent and direct swab extractions (Figure 1e, Supplemental Figure 2).
SARS-CoV-2 limit of detection comparison across swab types
We next assessed whether the swab type used would impact the recovery of SARS-CoV-2 and alter the limit of detection when using non-CDC-recommended swabs (CGp or TMI swabs compared to SYN swabs). All negative controls for floor swabs were indeed negative for SARS-CoV-2 using N1 and N2 (Supplemental Table 2, Figure 2) and all ‘no-swab’ controls which only had SARS-CoV-2, were negative for human Rp (Supplemental Table 2). For the ‘no-swab’ and TMI swab, SARS-CoV-2 was detected in all of the three replicates at the lowest input of 362.5 genome equivalents ‘GE’, whereas the lowest dilution for all three replicates to be positive for CGp and SYN swabs was 1450 GE (Supplemental Table 2, Figure 2a). This suggests the limit of detection for neat and TMI swabs is likely between 0 and 362.5 GE per reaction whereas both CGp and SYN swabs were less sensitive with an expected limit between 750 and 1450 GE per reaction. There was a strong correlation between the input or theoretical GE and the measured GE with slopes all greater than 0.95 and the R2>0.96. Despite TMI swabs appearing to have the best overall performance in SARSs-CoV-2 detection followed by SYN swabs and then CGp swabs, the total viral yield did not differ across swab types at the lowest dilution of 362.5 (P>0.05, Kruskal-Wallis test) (Figure 2a). Specifically, multiple post-hoc comparisons showed that variation across swab-type only existed at the highest concentration (116,000 GE) with the TMI swabs having a higher viral recovery compared to SYN swabs (P=0.04, KW=7.21) (Figure 2). Rp yield was also compared across swab types and across viral inputs to characterize the variation in input biomass. For each swab type, human RNase P (Rp) gene was equally detected across the titrations indicating the swab method was sufficiently controlled (Supplemental Figure 4a). Swab type, however, did suggest that the Rp gene was highest in the SYN swab as compared to the CGp and TMI swabs (Kruskal-Wallis: P<0.0001, KW=41.41) (Supplemental Figure 4b). This result suggests that SYN swabs may adsorb more biomass. However, when we compared the variation in Cq values of hospital samples of nares and floor from the same hospital using SYN swabs, we observed Rp gene values that varied over six orders of magnitude (Supplemental Figure S5), much greater than the three orders of magnitude observed across swab types. Specifically, for floor samples, the Rp gene yield (copies per extraction) range across swab types was 149-3368 copies for SYN swabs, 0-3980 for CGp swabs, and 0-207 for TMI swabs.
Hospital proof of concept study
Based on the results from these initial experiments, we conducted a proof-of-concept study in the clinical setting by performing RT-qPCR for the SARS-CoV-2 N1 amplicon and Rp gene on RNA extracted from the swab head of nasal samples collected using TMI and/or CGp swabs alongside the recommended SYN swabs. Of the 20 participants sampled, 16 tested positive for SARS-CoV-2 at admission and were designated as COVID-19(+). The average time from diagnosis to sampling was ca. 4.2 days, with a NP swab test occurring within 72 hours of the time of nasal sampling. Of the 12 nasal samples using the SYN swab preserved in EtOH from COVID-19(+) patients, nine were positive for the presence of SARS-CoV-2 or a false negative rate of 25% (Figure 3a) compared to 14/16 SARS-CoV-2 positive NP swabs for the same group of patients, a false negative rate of 12.5%. For CGp and TMI swabs, 8/12 and 5/10 were positive for nares, respectively (Figure 3a). These rates of false negatives are similar, as compared to the 37.5% false negative rate reported for plastic-shafted synthetic-tipped nasal swabs collected in VTM and extracted from the eluent. As the degree of viral shedding is known to vary over the course of the disease , we compared the performance in the subset of COVID-19(+) patients with an NP-positive swab result within 72 hours of the time of sampling, and observed reduced false negative rates of 18.2% (SYN), 25% (TMI), and 30% (CGp). We next compared success rates across swab samples from the built environment. On the floor samples, the CGp swabs had the highest success rate at 75% in detection of SARSs-CoV-2 from SARSs-CoV-2 positive patient rooms whereas SYN swabs detected SARSs-CoV-2 in 63% of rooms, and TMI in 44% of rooms (Figure 3a). Bedrail samples had the lowest frequency of detection, 5/16 (31%), for each swab type (Figure 3a). For SARSs-CoV-2 negative patients admitted to the same hospital for other reasons, all nares and bedrail samples were negative, whereas one floor sample using the SYN swab detected SARSs-CoV-2 (Figure 3b).
The observed differences in detection among nares and environmental samples, taken in context of our previous experimental results demonstrating that the swab type does not inherently impact SARS-CoV-2 detection, suggests that variation in sample collection from the nares and other environmental samples has an important role in detection sensitivity. When swabbing an environmental surface or body site (i.e., nares), there is inherent variation in the swabbing event which can be attributed both to stochastic differences in biomass (i.e. human cells, dust, etc.) present and collected as well as the downstream processes such as nucleic acid extraction and RT-qPCR. To evaluate if certain sampling locations or swab types were more variable than others, we calculated the intra-assay coefficient of variance (CV) of the Rp gene Cq values. When comparing the variation across sample types (environmental samples: bedrail and floor vs. nares swab head and nares eluent), the CV was significantly higher in floor (Mann Whitney p=0.0056), patient nares eluent (P=0.0017), and patient nares swab (Mann Whitney, P=0.0018) when compared to control (human RNA spike-in) samples. The median difference in nares swabs was greatest with a median difference in variance of 2.5 compared to controls (Supplemental S6a). In this case, the CV of the positive control samples would demonstrate the variance in combined extraction and RTqPCR thus is indicative of the total variance in molecular sample processing. Next, we stratified for swab types. Swab types also demonstrated an effect with the inter-swab variation between values measured with CGp and SYN along with CPg and TMI swabs for controls and patient samples being significant, respectively (Mann Whitney: P=0.0012, P=0.0127) (Supplemental Figures S6b). Overall these findings demonstrate how variability in a given sampling (swabbing) event can influence SARs-CoV-2 detection.
To determine the feasibility of co-opting nucleic acid for microbiome processing, we processed a subset of samples (n=40) spanning a total of three patients, two sample types (floor and nasal) and the three swab types. After processing with Deblur, the total number of reads per sample were compared (Figure 4a). Read counts were highly variable across sample types and for each patient but were consistent within the swab types for each comparison. For floor samples in patient room 18, SYN swabs had the highest number of reads followed by TMI and CGp. For nasal samples however, patient 1 had the higher read counts from TMI while patient 7 and 18 both showed slightly lower read counts for SYN swabs as compared to alternative swabs. The differences were minor however and are primarily differentiated by patient room (Figure 4a). After rarifying to 5000 reads, a PCoA plot was generated from using unweighted UniFrac distances (Figure 4b). Samples which were collected using different swab types clustered together when controlling for patient room and sample type, suggesting that the swab type used does not have impact the microbiome analysis (Figure 4b). When analyzing all samples together, sample_type (floor vs nasal) and patient number (7 vs 18) were both significant drivers of the microbiome community (sample_type: PERMANOVA n=24, group=2, P=0.001, F=6.94; patient_num PERMANOVA n=24, group=2, P=0.001, F=6.92) whereas swab type did not have an effect (P=0.164). Distances between swab types were lower than distances between patients for both floor (Supplemental Figure S7a) and nasal (Supplemental Figure S7b) samples, with patient 7 exhibiting higher variation than patient 18. Floor samples generally had a higher microbial diversity compared to nasal swabs, with Staphylococcus, Corynebacterium, Pseudomonas, Streptococcus, and Enterobacteriaceae being the more dominant taxa. Nasal samples however were mostly enriched by either Staphylococcus or Corynebacterium, with patient 7 having a higher abundance of Lawsonella (Figure 4c).