Investigation of Discordant SARS-CoV-2 RT-PCR Results Using Neat Saliva


 Saliva is an attractive sample for coronavirus disease 2019 testing due its ease of collection and amenability to detect viral RNA without processing. Using a direct-to-RT-PCR method with saliva self-collected from confirmed COVID-19 positive volunteers, we observed 32% false negative results. Confirmed negative and healthy volunteer samples spiked with 106 genome copies/mL of heat-inactivated severe acute respiratory syndrome coronavirus 2 showed false negative results of 10% and 13%, respectively. Additional sample heating or dilution of the false negative samples conferred modest improvement. These results highlight the potential to significantly underdiagnose COVID-19 infections when testing directly from heterogeneous neat saliva samples.


Introduction
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has impacted daily life on a global scale since its emergence in December 2019. Scientists rallied quickly to gather and disseminate information to create sensitive molecular detection methods for SARS-CoV-2 to protect the population 1 . Public health authorities have relied heavily on the "gold standard" reverse-transcription polymerase chain reaction (RT-PCR) testing of nasopharyngeal swabs (NPS). However, this type of sample collection requires professional healthcare workers for procurement, which has prompted researchers to evaluate saliva as a simple, self-collectable specimen. Many studies have shown comparable diagnostic sensitivity for viral genomic RNA extracted from saliva specimens versus NPS when using RT-PCR for COVID-19 detection 2 .
To further increase simplicity, several methods, such as the covidSHIELD test developed at the University of Illinois Urbana-Champaign (UIUC) 3 , the SalivaDirect test developed at Yale 4 , and the SalivaSTAT method developed at the Medical College of Georgia 5 , utilize direct analysis from unprocessed saliva to avoid the RNA extraction step.
Our groups are actively involved in the development of a point-of-care antigen test for COVID-19 utilizing neat saliva as the specimen. We chose the covidSHIELD RT-PCR processing method to aid in validation of our test 6 due to its relative ease and reported sensitivity. However, it was observed during test development that the inherent complexity of neat saliva 7 produced a signi cant number of false negative SARS-CoV-2 RT-PCR results, prompting us to investigate further.

Assessment center and healthy volunteer saliva samples
The research and all associated experimental protocols, including those related to sample collection, processing and testing, were approved by the Research Ethics Board of McMaster University and its a liated hospitals (Hamilton Health Sciences and St. Joseph's Healthcare). The research was performed in accordance with the institutional guidelines/regulations and the Declaration of Helsinki. Informed consent was obtained from all participants (≥ 18 years of age) and/or their legal guardians attending a Hamilton Health Sciences a liated COVID-19 assessment center (Hamilton, Ontario) for NPS SARS-CoV-2 testing. Saliva (1 to 5 mL) was self-collected under supervision following the instructions accompanying a provided saliva collection kit immediately after the nasopharyngeal swab. Patients were con rmed as SARS-CoV-2 positive (22 patients: NPS positive) or negative (92 patients: NPS negative) from the NPS sample at the Hamilton Regional Laboratory Medicine Program at St. Joseph's Healthcare (Hamilton) using an established method 8 . Patient samples were stored at 4°C for <72 hours prior to transport to McMaster University for storage at -80°C. Healthy adult volunteers (23) from the McMaster University research community provided saliva samples in sterile 50 mL polypropylene tubes following the same self-collection instructions. Samples were aliquoted then frozen or used on the same day as collection.
RT-PCR RT-PCR was performed using the Power SYBR® Green RNA-to-C T ™ 1-step Kit (Fisher Scienti c, Ottawa, Ontario, Canada) with the CDC N1 and RNaseP forward and reverse primers from Integrated DNA Technologies (IDT: NC, USA). All assessment center and healthy volunteer saliva samples were processed based on the original covidSHIELD method described by Ranoa et al. 3 . Brie y, a 20-50 µL aliquot of the saliva sample was heated at 95°C for 30 minutes followed by the addition of an equal volume of 2x TBE-  Figure S1) and the limit of detection obtained with these reaction conditions (see details in Supplementary), a threshold cycle value (C t value) of >36 was considered negative.
Spiking saliva samples with heat-inactivated SARS-CoV-2; heating and diluting saliva samples Heat-inactivated 2019 Novel Coronavirus (Isolate USA-WA/2020) was purchased from the ATCC through Cedarlane (Burlington, Ontario, Canada). SARS-CoV-2 NPS negative and healthy volunteer saliva samples were spiked with 4.2 x 10 6 genome copies (gc)/mL of heat-inactivated SARS-CoV-2 prior to processing for RT-PCR as described above. Saliva samples that were heated prior to spiking and RT-PCR processing were incubated at 95°C for 30 minutes. Saliva samples that were diluted prior to RT-PCR processing were done so with an equal volume of sterile RNase-free water.

Results
All saliva samples included in this study gave a positive result for the human RNaseP internal control using the processing method of the covidSHIELD RT-PCR laboratory developed test 3 (Supplementary  Table S1).
To test our work ow and reagents with the covidSHIELD RT-PCR processing method, we initially examined SARS-CoV-2 negative saliva samples (NPS negative) spiked with 10 6 gc/mL heat-inactivated SARS-CoV-2. At this "viral load", a C t value of 29 was expected based on our calibration data (Supplementary Figure S1). We observed great RT-PCR negativity (C t value >36) with the spiked samples when the processed saliva was mixed prior to addition to the ampli cation mixture (Figure 1a). Allowing the same samples to settle improved detection markedly, therefore this settling step was included in all subsequent experiments.
All assessment center saliva samples that tested SARS-CoV-2 negative by NPS also tested negative using the RT-PCR method (100% speci city). However, only 15  Saliva is a complex matrix reported to contain unknown RT-PCR inhibitors 9 , and RNAses that could degrade viral RNA upon release from the capsid 10 . To assess if such inhibitors contributed to the observed false negative results with the spiked saliva samples, four SARS-CoV-2 negative and three healthy volunteer saliva samples that did not respond to spiking were heated at 95°C for 30 minutes to denature/degrade salivary interferants prior to spiking and processing as before. Four of the seven samples showed positive C t values (average ΔC t [spiked-heated] of 7.9 ± 1.7; C t range of 28.6 -32.9) (Figure 2a).
To determine if chemical interferants could be su ciently reduced by dilution, which is often used to reduce viscosity 11 and aid with automated sample processing 12 , saliva samples were diluted 50% before the 50% dilution step incorporated in the RT-PCR processing method creating a 25% saliva solution prior to addition to the RT-PCR reaction. The saliva samples chosen for examination were four SARS-CoV-2 positive samples that tested negative using the RT-PCR method and two healthy volunteer samples whose saliva remained negative after spiking with SARS-CoV-2. The SARS-CoV-2 positive samples remained negative after dilution while the two healthy volunteer samples gave positive results, with a reduction in C t value of 7.6 and 6.7 units (Figure 2b).

Discussion
Accurate reporting of COVID-19 infections to monitor populations and control outbreaks has been a primary goal worldwide from the start of the pandemic. Among the various methods currently in use, tests using saliva have the advantage of easy sample acquisition, which could allow collection without a health-care practitioner. However, our study has indicated that neat saliva poses a problem for conventional RT-PCR assays due to complications surrounding its physical attributes (viscosity, heterogeneity) and unidenti ed chemical constituents.
A key observation was that dispersal of saliva particulates via sample mixing prior to the RT-PCR assay increased the number of observed false negative results and increased C t values generally, even in positive samples (Figure 1a). Allowing the prepared sample to settle for at least 5 minutes after processing greatly improved RT-PCR detectability. In our hands, brief centrifugation worked as well as settling (Supplementary Figure S3), although Ranoa et al. 3 did not observe the same bene t. Based on the visual assessment of several hundred assessment center negative saliva samples including and beyond those used for this study, there was no obvious correlation between any physical attribute and an erroneous RT-PCR result. Furthermore, the rate of false negative results did not seem to depend on the sample group (NPS positive, NPS negative, Healthy volunteer) (Figure 1b; Supplementary Figure S2).
Pre-heating of seven saliva samples that consistently returned false negative results after spiking with heat-inactivated virus showed marked improvement-four samples returned a positive result (Figure 2a, black triangles)-suggesting that some inhibitory components in the saliva could be disrupted or degraded. However, the remaining three samples showed resistance to the treatment, and it was unclear why the long heating step already included in the covidSHIELD method was not su cient for inactivation of the interferants. Sample dilution was unable to produce correct results for false negative saliva samples (NPS positive, saliva RT-PCR negative) (Figure 2b, grey circle and square), though this method did remedy false negative results from spiked healthy samples.
While the utility of saliva as a diagnostic specimen cannot be disputed 13 , and even considering the documented reduction in detection sensitivity relative to NPS samples 14 , our observations provide evidence for a potentially signi cant issue with a high false negative rate when using unprocessed saliva samples for COVID-19 RT-PCR testing. Indeed, the original manuscript describing the covidSHIELD method 3 reported 11% false negative results upon initial testing, but 100% sensitivity upon retesting; Kandel et al. 15 reported 38 of 432 samples (8.8%) as invalid upon initial testing, improving to three invalid samples after retesting; and Sahajpal et al. 5 reported 5% invalid results (though no false negatives).
Our study indicated that valid samples showing expected C t values for the control RNaseP sequence could repeatedly result in a false negative SARS-CoV-2 result, suggesting an inherent issue with interferants in some saliva samples. We note that several other factors could lead to variability in saliva samples, such as the timing of sample collection with respect to disease progression, symptoms, and time of day; and the volume of saliva provided. We also recognize that disease prevalence in the population at the time of sample collection and our limited number of samples may have contributed to our observations. However, general agreement in the rate of false negative SARS-CoV-2 RT-PCR results across 137 samples from assessment centre attendees and healthy volunteers presented here, and seen in other studies, points to unknown factors that produce signi cant false negative results with a widely used RT-PCR method. These results indicate that while direct use of unprocessed saliva is desirable to increase the simplicity and reduce the cost for COVID-19 testing, the relatively high rate of false negative results may require additional sample processing, such as RNA extraction, to reduce the possibility of COVID-19 positive citizens unknowingly transmitting the virus within the community. Competing interests. All authors declare no completing interests. Figure 1 Detection of SARS-CoV-2 in unprocessed saliva samples from COVID-19 positive and negative volunteers.

Figures
(a) Improved detection of spiked saliva samples allowed to settle after processing versus being mixed after processing. Solid black triangles represent samples that became detectable (Ct <36) after settling; dark grey open squares represent samples that showed improved Ct values but remained above the cutoff Ct value (Ct > 36, black dashed line); light grey open circles represent samples that were either detectable in both groups or did not show an appreciable change in Ct value. A Ct value of 40 represents a negative result. (b) Prevalence of false negative SARS-CoV-2 RT-PCR outcomes using neat saliva samples. The sample number is indicated above the data columns. The percent of false negatives is shown in brackets beside each group. All NPS positive saliva samples were tested as is. All NPS negative and healthy volunteer saliva samples were spiked with 106 genome copies/mL heat-inactivated SARS-CoV-2 prior to testing. All samples tested at least in triplicate. See Supplementary Figure S2 for individual sample Ct values.

Figure 2
Effect of heat treatment or dilution of spiked saliva samples prior to testing. (a) Saliva samples from NPS negative or healthy volunteers that produced a false negative result after spiking with 106 genome copies/mL heat-inactivated SARS-CoV-2 (Spiked) compared with the same saliva samples heated prior to spiking (Heated). (b) Spiked saliva samples that produced a false negative result (Neat) tested after dilution (Diluted). Solid black triangles represent samples that became detectable (Ct <36) after settling; dark grey open squares represent samples that showed improved Ct values but remained above the cutoff Ct value (Ct > 36, black dashed line); light grey open circles represent samples that were either detectable in both groups or did not show an appreciable change in Ct value (In panel B only, the light grey open circle represents 3 identical, undetectable samples.) A Ct value of 40 represents a negative result; the black dashed line at Ct = 36 represents the cut-off value for our system. Samples tested at least in duplicate.