Sensitive detection and quantification of SARS-CoV-2 in saliva

Saliva has significant advantages as a test medium for detection of SARS-CoV-2 infection in patients, such as ease of collection, minimal requirement of supplies and trained personnel, and safety. Comprehensive validation in a large cohort of prospectively collected specimens with unknown SARS-CoV-2 status should be performed to evaluate the potential and limitations of saliva-based testing. We developed a saliva-based testing pipeline for detection of SARS-CoV-2 nucleic acids using real-time reverse transcription PCR (RT-PCR) and droplet digital PCR (ddPCR) readouts, and measured samples from 137 outpatients tested at a curbside testing facility and 29 inpatients hospitalized for COVID-19. These measurements were compared to the nasal swab results for each patient performed by a certified microbiology laboratory. We found that our saliva testing positively detects 100% (RT-PCR) and 93.75% (ddPCR) of curbside patients that were identified as SARS-CoV-2 positive by the Emergency Use Authorization (EUA) certified nasal swab testing assay. Quantification of viral loads by ddPCR revealed an extremely wide range, with 1 million-fold difference between individual patients. Our results demonstrate for both community screening and hospital settings that saliva testing reliability is on par with that of the nasal swabs in detecting infected cases, and has potential for higher sensitivity when combined with ddPCR in detecting low-abundance viral loads that evade traditional testing methods.


Results
Pre-implementation studies: optimization of SARS-CoV-2 assays and RNA extraction. Validation of SARS-CoV-2 PCR assays with nucleic acid standards. We first optimized RT-PCR and ddPCR assays targeting nucleocapsid (N1) and nucleocapsid (N2) genes of SARS-CoV-2 using commercially available RNA standards. For both RT-PCR and ddPCR assays, we used one-step reverse transcription kits to simplify the PCR assays and to minimize errors. Our RT-PCR measurements showed excellent efficiency (> 99.25% for N1 and N2) and limit of detection (LOD; we were able to detect 1 copy of viral mRNA in a reaction). The RT-PCR cycle threshold (Ct) values corresponding to 1 copy number (CN) of N1 and N2 genes were calculated to be 38.2 and 38.6, respectively (Supplementary Fig. 1). Similarly, the ddPCR assay targeting the N1 and N2 genes of SARS-CoV-2 showed excellent linearity and revealed an LOD of 0.06 and 0.21 CN/µL for N1 and N2, respectively ( Supplementary Fig. 2).

Characterization of RT-PCR and digital-PCR sensitivity for detection of SARS-CoV-2.
To characterize the sensitivity of both PCR assays in patient samples, nasal swab specimens collected from 35 COVID-19 negative individuals (as confirmed by the FDA EUA TaqPath COVID-19 kit and EUA Cepheid Xpert Xpress diagnostic systems, as well as research RT-PCR and ddPCR assays) were pooled together, mixed and aliquoted into individual vials. Nasal swab specimen collected from a clinically confirmed COVID- 19  www.nature.com/scientificreports/ then added to the pooled negative aliquots at different dilution ratios, ranging from 1/10 2 to 1/10 7 (Supplemental Table 1). Four identical dilution series were prepared in duplicates. One dilution set was tested with the Cepheid Xpert Xpress SARS-CoV-2 system. For the remaining three dilution sets, RNA was extracted using an automated KingFisher Flex instrument and analyzed using the FDA EUA Applied Biosystems TaqPath COVID-19 kit as well as RT-PCR and ddPCR assays optimized in our research laboratory settings. These assays have different positivity cut-offs. The diagnostic Cepheid Xpert Xpress system used by CML calls a NP or NS sample positive if either N2 or envelope (E) gene shows a Ct value lower than 45 23 . The TaqPath COVID-19 kit calls sample positive if any 2 of the 3 target genes [N gene (N1 and N2 regions), spike glycoprotein (S) gene and ORF1ab gene] show a Ct value lower than 37. The CDC guidelines however recommend calling an RT-PCR test positive when Ct values are lower than 40, and this threshold was used for our research laboratory RT-PCR assay. For ddPCR measurements in these diluted samples, a specimen was called positive when CN/µL for either N1 or N2 was higher than 0.1 and more than two positive droplets were detected in the reaction (as recommended by the manufacturer, Bio-Rad Inc.). While RT-PCR based tests (EUA Cepheid Xpert Xpress, EUA TaqPath and our own research laboratory RT-PCR assay) could reliably detect the presence of a virus in 1/100,000 diluted samples, ddPCR assay was able to detect N1 target gene at 1/1,000,000 dilution (Supplementary Table 1). Although these results are suggestive of higher sensitivity of ddPCR platform, a direct comparison between assays using different RNA extraction methods, gene targets, cycling conditions, and cut-offs should be addressed with caution. We next assessed the sensitivity of the RT-PCR and ddPCR assays (both use the same primers and probes combination) in a cohort of 8 positive and 18 negative NP samples (based on CML results) from patients undergoing clinical testing for suspected COVID-19. For the positive samples we saw 100% concordance of RT-PCR and ddPCR results with CML data (i.e. all positive samples by CML were also positives in our measurements, Supplementary Fig. 3), but out of the 18 samples called negative by the CML, 11 samples were identified as positive by RT-PCR and 13 by ddPCR. These additional positive cases were mostly samples with low viral load, suggesting a higher detection sensitivity of our optimized assays. Both RT-PCR and ddPCR assays have also demonstrated a high agreement in detection of an internal control RnaseP gene ( Supplementary Fig. 3), indicating that detection of additional positive cases by ddPCR is not biased by the input material. These assays were then used for further measurements on additional patient samples (N = 137) and compared to NS testing, as described below.
TRIzol-based RNA extraction for SARS-CoV-2 testing. During the height of the COVID-19 pandemic, many laboratory supplies such as RNA extraction kits were either not available or backordered. We therefore tested if a TRIzol-based RNA extraction (TRE) method, a widely used and simple approach for RNA extraction in research laboratories, would be feasible for COVID-19 screening tests. RNA extracted from the same saliva sample using either TRE or two CDC recommended commercial viral extraction kits 24 (QIAmp MinElute Virus DNA/RNA Spin Kit [QVDRK] and QIAmp Viral RNA [QVK] Mini Kit) was analyzed with RT-PCR and ddPCR to quantify the RnaseP gene, an internal control for extraction procedure. The relative efficiency of TRE method compared to QVK was found to be 17% ( Supplementary Fig. 4). Although this value is relatively low, the high sensitivity of our PCR assays compensates for the reduced extraction efficiency, and the availability of the reagents was considered a major advantage. We therefore proceeded to use this simple and available RNA extraction method for our studies.

Comparative analysis using matched clinical specimens. Curbside sample collection pipeline at The
University of Chicago Medical Center. We established a sample collection pipeline from the clinical curbside testing site at the University of Chicago Medical Center. The pipeline was divided into 3 stages: samples collection at the curbside testing clinic, RNA extraction at a nearby BSL3 facility, and analysis with RT-PCR and ddPCR in a BSL1 research laboratory (Fig. 1). We simultaneously collected three samples (2 NS and 1 saliva sample) from 137 volunteers (Table 1) who were scheduled at the curbside testing site to receive a standardof-care clinical COVID-19 test. 92 of these patients were being tested for COVID-19 related symptoms, and 45 asymptomatic subjects were being tested prior to a planned medical procedure. Upon collection, one NS sample was sent to CML for clinical diagnostic analysis. The other NS and saliva samples were sent to a BSL3 facility for viral RNA extraction (see "Methods"). The extracted viral RNA samples were then transported to a BSL1 lab for concurrent RT-PCR and ddPCR analysis ( Fig. 1).
High concordance of saliva testing with clinical nasal swab assay. We measured the curbside collected patient specimens using our RT-PCR and ddPCR assays. As an assay control, a serial dilution of SARS-CoV-2 RNA standards was always analyzed along each RT-PCR or ddPCR plate (Supplementary Fig. 5 and 6). The distributions of both RT-PCR and ddPCR results demonstrated an excellent concordance between N1 and N2 detection (Fig. 2a,b). All 16 patient samples detected as SARS-CoV-2 positive by the CML were also called positive by our saliva RT-PCR test (Fig. 2c,d), resulting in 100% concordance, and demonstrating the applicability of using saliva for sensitive SARS-CoV-2 diagnostic testing. Furthermore, our saliva testing detected an additional 7 positive cases that were called SARS-CoV-2 negative by the CML. Eighteen paired NS specimens collected from the 23 patients with positive saliva samples were also tested positive using the research RT-PCR assay (Fig. 3).
Empirical ddPCR threshold calculation. To evaluate the concordance of our ddPCR saliva assays with NS results from a clinical laboratory, we first sought to determine the baseline levels of SARS-CoV-2 nucleic acid signals in COVID-19 negative patients. Since no FDA recommended guidelines are available for ddPCR positive calling criteria, and because manufacturers recommendations (BioRad) were provided for nasal swab testing only, we have developed the thresholds for both N1 and N2 genes de novo using specimens collected from www.nature.com/scientificreports/ healthy individuals in our pre-operation screening cohort (pre-op). This cohort contained 45 asymptomatic individuals who participated in the curbside testing as part of their pre-operative screening process. All of these patients were SARS-CoV-2 negative by CML nasal swab test. Using a standard formula for threshold detection (mean + 2*SD, SD: standard deviation), the new ddPCR saliva thresholds were calculated and listed in Supplementary Table 2. Data distribution of the healthy cohort is shown in Supplementary Fig. 7.
Empirical threshold application on clinical specimens. We have applied the empirical ddPCR thresholds to determine the COVID-19 status of the remaining 92 subjects collected at the curbside (Fig. 4). A patient's sample was called positive when either N1 or N2 target genes showed both CN/µL (concentration) and positive droplet number above the respective thresholds (see "Methods"). We found that ddPCR results generated from saliva specimens in our research lab were highly concordant with the diagnostic data obtained from the nasal swab (NS) test performed at the certified CML. Our saliva ddPCR testing identified 15 out of 16 individuals determined as positive by the NS-based CML assay (Fig. 2e,f), achieving a concordance of 93.75% in detecting positive cases. In matched NS specimens, ddPCR assay has identified 17 positive cases, including 14 cases called positive by the CML (Fig. 2e). We next analyzed an independent specimen cohort collected from 29 inpatients admitted to the hospital with COVID-19. For these subjects, NS and saliva were collected simultaneously. NS samples were tested by the CML, whereas saliva specimens were tested by our ddPCR and RT-PCR assays in the research laboratory. Important Table 1. Clinicopathological characteristics of the SARS-CoV-2 saliva testing participants. IQR interquartile range. *Comorbidities were defined as heart failure, coronary artery disease, hyperlipidemia, history of cancer, systemic hypertension, pulmonary hypertension, home oxygen use, chronic lung disease, immunodeficiency, end stage renal disease, obesity, and diabetes. # The risk factors were defined as being older than 60 years old, male, race (black, non-white Hispanic or Latino, American Indian, Alaskan native), presence of underlying comorbidity, and being immunocompromised.  Potential for increased sensitivity in detecting SARS-CoV-2 with saliva digital-PCR readout. Overall, by analyzing saliva samples we were able to identify most curbside subjects that were determined as positives by the CML (100% for RT-PCR, 93.75% for ddPCR assays), (Fig. 2c-f), demonstrating the applicability of using saliva for sensitive SARS-CoV-2 diagnostic testing. While NS-based CML assay has detected 16 of the 92 cases as positives for SARS-CoV-2, ddPCR assay identified 17 and 31 positives from NS and saliva samples respectively (Fig. 2). Due to a higher sensitivity and resolution provided by the ddPCR assay, we were able to catch more positive cases than RT-PCR (8 more) and CML (15 more), especially in cases where the viral copy number is very low (Figs. 2a,b, 3 and 4). This is especially noticeable for saliva samples assessed with ddPCR (black circles in Fig. 2b). The ability to detect low viral load in patients may be useful for identifying potential viral carriers with no or mild symptoms.
To confirm that additional samples detected by ddPCR are true positives and verify whether the readings remain consistent and reproducible over time, we independently re-analyzed remnant RNA specimens extracted from 46 saliva samples (19 positive and 27 negative, Supplementary Table 3) with ddPCR. The samples selected prioritized individuals that were caught positive only by ddPCR but missed by CML or our RT-PCR tests. The results of two independent analyses showed excellent consistency ( Supplementary Fig. 8). Out of the 46 retested samples, 19 were indicated as positives during the first run, of which 15 were confirmed as positives in the second run. The 4 samples with disagreement between the two runs had PCR readings close to our empirical thresholds. While some specimens showed lower readings upon re-analysis compared to first ddPCR run, this may be attributed to the RNA degradation after storage for nearly 5 months. All 27 samples that were negative during the first run were confirmed to be negative upon re-analysis.

Discussion
An important step in containing the spread of COVID-19 is early detection of viral carriers in public and monitoring viral load changes in patients. Conventional sample collection methods such as nasal or nasopharyngeal swabs are invasive, time consuming, labor intensive, and rely on availability of nasal swabs which can be in short supply. In addition, these methods increase the infection risk to health care workers by provoking patients to sneeze and cough 25 . Nasal swab (NS) testing is uncomfortable for patients 26 , possibly leading to low adherence. Saliva testing offers a non-invasive, easy to collect alternative method that could reach a wider range of the population with reduced cost and risk to healthcare workers. Despite the potential of saliva testing, however, previous studies revealed conflicting results on its sensitivity for detection of SARS-CoV-2 in patient specimens and for its concordance with, FDA EUA approved nasal swab testing by clinical laboratories.
To this end, we have compared two different sampling approaches (saliva and NS) by using readouts from RT-PCR and ddPCR methods as well as an EUA approved diagnostic platform, demonstrating the benefits of saliva as a test medium for detection of SARS-CoV-2 infection in patients. We first demonstrate that our optimized    www.nature.com/scientificreports/ This high concordance in detecting the positive patients with clinical diagnostic testing further supports the suitability of saliva testing as a noninvasive alternative method for the detection of SARS-CoV-2 in patients. ddPCR measurements on saliva specimens revealed a higher number of positive patients compared to nasal swabs performed by the approved diagnostic laboratory. Out of 92 curbside collected patient specimens, saliva testing detected 31 positive samples with ddPCR readouts, a substantially higher number than that detected by the clinical diagnostic platform using NS specimens or our own NS ddPCR measurements. These additional positive calls were close to but above the threshold of our assay and were seen in patients with symptoms. Notably, independent re-testing of 46 saliva specimens with ddPCR confirmed presence of SARS-CoV-2 with 91.3% concordance, supporting a high reproducibility of this screening approach. In line with these observations, a comparable higher sensitivity was also obtained by Whyllie et al., using a RT-PCR saliva based readout 7 . Similarly, Suo et al. reported that 26 patients deemed negative by RT-PCR (out of 77 patients) were detected as positive by ddPCR 27 . In another study, multiplex ddPCR was used to test saliva and NP samples from 130 individuals with COVID-19 symptoms, and ddPCR was found to be superior to RT-PCR in determining positive cases 28 . Results published by several recent studies also support the use of ddPCR as a sensitive method for SARS-CoV-2 detection and quantification [29][30][31] .
Our analysis also revealed a very wide range in viral load measurements detected among COVID-19 positive patients, with up to 10 6 fold range in viral load for both curbside patients and inpatients (Supplementary Table 4). Interestingly, some patients showed extremely high viral nucleic acid concentrations, over 10 6 CN/ml. Quantification of viral load enabled us to classify these patients as supercarriers (viral loads > 10,000 CN/ml sample), high www.nature.com/scientificreports/ viral level carriers (between 1000 and 10,000 CN/ml sample), moderate viral level carriers (between 100 and 1000 CN/ml sample), and low viral level carriers (< 100 CN/ml sample). We found that saliva measurements show a more uniform distribution of viral loads among patients compared to nasal swabs. While most of the nasal swab samples showed either very high or very low viral loads, saliva testing revealed a more gradual distribution of viral loads in the same patients, indicating that saliva may be a more representative test medium for quantification of the overall viral load in a patient. The viral load detected by NS may also be influenced by how aggressively the swab is done, whereas there may be less operator-variability in collecting an equivalent volume of saliva. The precise quantification of viral load in saliva is crucial for identifying patients with high potential for transmission, as saliva is a main source of larger droplets and aerosols that are believed to spread the virus 32 . Recent evidence suggests a close relationship between patients with high viral load (supercarriers) and superspreaders 33 , which makes the saliva among the most relevant human specimens to determine superspreaders [29][30][31][32] . In summary, our work demonstrates the utility and sensitivity of saliva-based testing for detection of SARS-CoV-2 in a large group of patients from different settings (healthy, curbside, inpatients), and provides a comprehensive characterization of this method by comparing it to nasal swab testing performed by a clinical microbiology laboratory. We showed that measurement of saliva specimens provides high concordance with nasal swab testing, with possibility to provide additional sensitivity in detecting patients with low viral load. The inherent advantages, sensitivity, scalability and safety of saliva testing, makes it an important alternative to nasal or nasopharyngeal swab-based assays. While we believe that saliva testing is already a feasible alternative to traditional methods and should be immediately utilized, we also believe that improvements in pooled testing methods, automation, and parallelization will enable a much wider, rapid and cost-effective application of this important testing method in the management of the COVID-19 pandemic.

Methods
All methods were carried out in accordance with relevant guidelines and regulations.

Sample collection. This study was approved by the Institutional Review Board (IRB) of the University of
Chicago (IRB 20-0520), and we obtained informed written consent from all patients prior to sample collection. 137 outpatients who were scheduled for clinical curbside testing either due to the presence of influenza-like symptoms (n = 92) or as part of routine pre-operative or pre-procedural screening (n = 45) agreed to participate. Enrolled patients received a clinical COVID-19 test administered by the clinical team, followed by an additional nasal swab and saliva specimen collection administered by the research team. Paired NS and saliva samples were also collected from 29 inpatients admitted with COVID-19. The saliva sample was collected by instructing patients to spit until they had filled a sterile tube to the 2 mL line and then 2 mL of Zymo DNA/RNA shield solution (ZymoR1100-50) was immediately added. NS samples were collected into a tube containing 1 mL of Amies transport solution (Thermo Scientific).

RNA extraction.
For QIAmp MinElute Virus DNA/RNA Spin Kit (QVDRK) (Qiagen) and QIAmp Viral RNA Mini Kit (QVK)) (Qiagen) RNA extraction was performed following the manufacturer's instructions. For Trizol LS Reagent (Thermo Fisher Scientific, #10296010) the manufacturer's protocol was modified as follows: 1 ml of sample (i.e., saliva or nasal swab), 3 ml of Trizol LS Reagent and 0.6 ml of chloroform were mixed and centrifuged for 15 min at 12,000 × g at 4 °C. 0.8 ml of the colorless upper aqueous phase was carefully transferred to a new tube containing 1 ml isopropanol and 1 µl glycogen. The mixture was kept at − 20 °C overnight or was kept at room temperature for 10 min. After that, the mixture was span for 10 min at 12,000 × g, 4 °C, and the supernatant was discarded. The pellet was resuspended in 2 ml 75% ethanol, vortexed mixing and centrifuged for 5 min at 7500 g, 4 °C. The supernatant was discarded, and the pellet was incubated at room temperature for 5-10 min. After ethanol is evaporated, the pellet was resuspended in 50 µl RNase-free water, the quantity and quality of extraction was assessed using Qubit RNA HS kit, and stored at −80 °C. ddPCR assay. The 20 μl ddPCR reaction containing 5 µl One-step ddPCR Supermix, 2 µl reverse transcriptase, 1 µl 300 mM DTT (Bio-Rad), 2 µL N1/N2/RnaseP probe/primers sets (IDT, #10006770), and 6.4 µl of sample were loaded into either manual or automated droplet generator (Bio-Rad). The emulsion was transferred into a 96 well plate, sealed, and cycled using a C-1000 thermal cycler (Bio-Rad) under the following conditions: 25 °C for 2 min, 50 °C for 60 min, 95 °C for 10 min, 45 cycles of 95 °C for 30 s, 55 °C for 1 min, then 98 °C for 10 min, and 4 °C for infinite hold (ramping speed was 2.5 °C/s). After amplification, the plate was transferred to a QX200 droplet reader (Bio-Rad) from which raw fluorescence amplitude data is extracted to the Quantasoft software for downstream analysis. NS samples were called positive when CN/µL for either N1 or N2 was higher than 0.1 and more than two positive droplets were detected in the reaction, according to the manufacturer recommendation. As no Bio-Rad recommended threshold was available for saliva specimens, an empirical threshold was set by collecting and analyzing samples from 45 asymptomatic participants. All these individuals were COVID-19 negative by CML. A standard formula: mean + 2*SD (standard deviation), was utilized for the determination of a threshold. The thresholds are reported in Supplementary www.nature.com/scientificreports/ as negative control. IDT nuclease free water was used as a non-template control. CFX384 Touch Real-time PCR detection system (Bio-Rad,) was programmed as follows based on CDC's recommendations 24  TaqPath COVID-19 assay. RNA was extracted with MagMAX viral/pathogen (ThermoFisher) nucleic acid isolation kit using KingFisher Flex instrument for automated sample purification. Extracted RNA was reverse transcribed to cDNA and amplified/detected using the EUA TaqPath COVID-19 Combo Kit (ThermoFisher) on an Applied Biosystems 7500 Fast RT-PCR instrument. The assay contains probes to detect and amplify three SARS-CoV-2 specific target genes: ORF1ab, N gene, and S gene. If three targets are detected or if any 2 of the 3 target genes are detected, the test reports a positive result.
Cepheid Xpert Xpress SARS-CoV-2 test. The GeneXpert Dx system (Cepheid) is an integrated diagnostic device that performs automated specimen processing and real-time RT-PCR analysis. The Xpert Xpress test consists of two main components: (i) the plastic cartridge, which contains liquid sample-processing and PCR buffers and lyophilized real-time RT-PCR reagents, and (ii) the GeneXpert Dx system, which controls intracartridge fluidics and performs real-time RT-PCR analysis. Statistical analysis. R software was used for statistical analysis and data processing. All data figures were plotted with MATLAB. Relative concentrations for N1, N2, and RnaseP targets assessed by the RT-PCR assay were estimated using linear regression analysis of dilution series of the RNA standard (SARS-CoV-2 RNA Standard, Exact Diagnostics).

Ethics approval. This study was approved by the Institutional Review Board (IRB) of the University of
Chicago (IRB 20-0520), and informed written consent was obtained from all patients before sample collection.