key: cord-0925963-fx9c4uax
authors: Deming, Meagan E; Dong, Tracy Q; Agrawal, Vaidehi; Mills, Margaret G; Huang, Meei-Li W; Greninger, Alexander L; Jerome, Keith R; Wener, Mark H; Paasche-Orlow, Michael K; Kissinger, Patricia; Luk, Alfred; Hoffman, Risa M; Stewart, Jenell; Kottkamp, Angelica C; Bershteyn, Anna; Chu, Helen Y; Stankiewicz Karita, Helen C; Johnston, Christine M; Wald, Anna; Barnabas, Ruanne; Brown, Elizabeth; Neuzil, Kathleen
title: Detection and kinetics of subgenomic SARS-CoV-2 RNA viral load in longitudinal diagnostic RNA positive samples
date: 2022-02-12
journal: J Infect Dis
DOI: 10.1093/infdis/jiac048
sha: 53c269a26b54bb4b622a8bcecc07e07f19685b8b
doc_id: 925963
cord_uid: fx9c4uax

While detection of SARS-CoV-2 by diagnostic RT-PCR is highly sensitive for viral RNA, the nucleic acid amplification of subgenomic RNAs (sgRNA) that are the product of viral replication may more accurately identify replication. We characterized the diagnostic RT-PCR and sgRNA detection from nasal swabs collected daily by participants in post exposure prophylaxis or treatment studies for SARS-CoV-2. Among 1932 RT-PCR-positive swabs with sgRNA tests, 40% (767) had detectable sgRNA. Above a diagnostic PCR viral load threshold of 5.1 log10 copies/mL, 96% of samples had detectable sgRNA with viral loads that followed a linear trend. The trajectories of diagnostic and sgRNA viral loads differed, with 80% peaking on the same day but duration of sgRNA detection being shorter (8 versus 14 days). With a large sample of daily swabs we provide comparative sgRNA kinetics and a diagnostic PCR threshold that correlates with replicating virus independent of symptoms or duration of illness.

Coronaviruses such as SARS-CoV-2 are large, positive sense, single-stranded RNA viruses that generate structural and accessory proteins through a process of discontinuous transcription, with the resulting subgenomic RNAs (sgRNA) encoding the leader transcription regulatory sequence in close proximity to the target gene. [1, 2] These sgRNA transcripts indicate active viral replication and may be used to discriminate recently transcribed viral RNA from residual genomic material. [3, 4] Prior studies have indicated that persistently detectable SARS-CoV-2 by diagnostic reverse transcriptase polymerase chain reaction (RT-PCR) beyond 10 days may reflect residual genomic material rather than replicating virus, suggesting that findings of prolonged, positive RT-PCRs in treatment and transmission studies may not accurately reflect replicating virus. [5, 6] This time course of infectious viral shedding is influenced by multiple factors including disease severity, degree of immunocompromise, and emerging immune responses. [5, 7, 8] Prolonged diagnostic RT-PCR positivity after an individual is no longer infectious poses public health challenges. For example, during the Omicron variant wave of COVID-19 in the United States, such a large fraction of the populated tested positive that staff shortages were experienced in critical sectors such as healthcare, leading health authorities to shorten recommended self-isolation periods and recommend return-to-work even for individuals who continue to test positive by diagnostic RT-PC. Discrimination between replicating virus and residual genomic material could help to optimize the period of self-isolation to reduce onward transmission while also reducing unnecessary loss of productivity. It could also help to identify persons with SARS-CoV-2 infection who would benefit clinically from antivirals and inform prevention strategies among their contacts. Viral culture and quantification by plaque assays provide a gold standard of assessing infectivity but are not readily available at large scale for a diagnostic assay. Detection of sgRNA in clinical samples has been suggested as an additional diagnostic tool to track infectious virus. [6, 9] We leveraged a large sample A c c e p t e d M a n u s c r i p t set from two outpatient studies that longitudinally collected nasal swabs after exposure to or infection with SARS-CoV-2, capturing pre-and asymptomatic infections in the Post-Exposure Prophylaxis (PEP) study and early infections (<72 hours) in high risk outpatients in the Treatment study. [10, 11] Our goal was to investigate the relationship between sgRNA and diagnostic PCR viral loads, identify the diagnostic PCR viral load that correlates with replicating virus independent of symptom onset or time from first positive test, and characterize the kinetics of sgRNA expression over time.

The study population was derived from two double blinded randomized controlled trials (NCT04328961 and NCT04354428) evaluating PEP and treatment for SARS-CoV-2. [10, 11] The study enrolled participants from around the United States in a remote trial between March and August 2020. For PEP, we evaluated hydroxychloroquine compared to an ascorbic acid control for prevention of SARS-CoV-2 infection beginning at a median of 2 days after exposure to a SARS-CoV-2 positive contact. [10] PEP participants self-collected daily nasal swabs for 14 consecutive days, followed by a swab on day 28. In the parallel Treatment study, infected participants diagnosed within the preceding 72 hours self-collected nasal swabs on days 1-14, 21, and 28 and were treated with hydroxychloroquine with or without azithromycin. [11] The PEP participants were recruited based on early exposure (within 96 hours) to SARS-CoV-2 whereas Treatment participants represented recently infected individuals. Study procedures were conducted with IRB approval and informed consent was obtained from all study participants. Neither prophylactic nor treatment medication was found to be effective at reducing SARS-CoV-2 infections or shedding as determined by RT-PCR, and thus we utilized intervention and control samples for all analyses. [10, 11] A c c e p t e d M a n u s c r i p t

Nasal swabs were tested for SARS-CoV-2 RNA by real time RT-PCR using the CDC N1 and N2 primer sets targeting the nucleocapsid gene as previously described. [12] In brief, RNA was extracted using MagNA Pure LC or MagNA Pure 96, with 200ul input and 50ul elution volumes. [13] As an internal control, EXO RNA was added at the start of RNA extraction to ensure that any negative results were not the result of nonspecific inhibition of the assay. [14] RNase P amplification was performed on 1,933 (14%) of the 13,839 swabs collected across both studies as a sample integrity control; more than 99% (1,915) had RNase P detected. PCR for both N and sgE were performed on ABI 7500 thermocyclers using AgPath-ID One-Step RT-PCR kits, with cycle threshold (C T ) cutoffs of 40.

Diagnostic assays for N were initially performed as two separate assays (N1 and EXO, and N2 and EXO) with 5µL of RNA per reaction, then with a single triplex reaction (N1, N2, and EXO) with 10µL of RNA per reaction after validation showed no difference in the limit of detection between the assays.

Samples with detectable SARS-CoV-2 RNA were tested for sgRNA targeting sgE with a paired leader and E primer set (5'-CCAACCAACTTTCGATCTCTTGT-3´, 5´-CGTACCTCTCTCTTCCGAAACG-3´), and a probe (5´-FAM-TCTCTAAACGAACTTATGTACTC-3MGBEC-3´) set on the leader/E junction. [13] Quantification standards were run with both assays to provide a viral load as copies/mL during assay validation (Supplemental Material). Sample stability on swabs over 7 days at room temperature was assessed prior to trial initiation, demonstrating a 2.1 (N1) and 0.6 (N2) C T loss for diagnostic RT-PCR and no appreciable loss over 7 days for sgRNA. [13] Statistical analysis Participant age and sex at enrollment were summarized overall and by study. RT-PCR test results were summarized on a participant-level as well as swab level, including counts of positive results and concordant results. The time (in days) between the day of peak observed diagnostic and sgRNA viral load was summarized with a histogram.

A c c e p t e d M a n u s c r i p t

The Pearson correlation coefficients were calculated between the diagnostic and sgRNA viral loads among samples that were tested positive for both. We presented a scatterplot of the diagnostic and sgRNA viral loads and fit a hinge regression model with the sgRNA viral loads as the outcome variable and the diagnostic RNA viral loads as the explanatory variable. [15] The empirical receiver operating characteristic (ROC) curves for classification of positive sgRNA samples based on diagnostic RNA viral load thresholds were presented graphically for pre-and post-peak diagnostic RNA viral load and overall.

Three cohorts of participants were further defined based on the observed viral load trajectories during the first 14 days of follow-up. First, the uncensored peak cohort consisted of participants who had an uncensored peak diagnostic viral load during the 14-day follow-up (i.e., peak diagnostic RNA viral load was not measured on the first or last swab collected). Second, the sustained shedding cohort was defined as the subset of the uncensored peak cohort who had at least two positive diagnostic RNA tests during the 14-day follow-up. Last, the trajectory modeling cohort consisted of participants who had at least 2 swabs that tested positive for sgRNA during the 14-day follow-up.

Using data from the sustained shedding cohort, we calculated the sgRNA detection rate and the ratio of diagnostic and sgRNA viral loads over time. We also fit a linear mixed-effects model to test whether the difference between the RNA and sgRNA were significantly different before and after the peak diagnostic viral load (Supplemental Material).

A c c e p t e d M a n u s c r i p t Kaplan-Meier curves were generated for diagnostic and sgRNA viral clearance since peak viral load for participants in the uncensored peak cohort. Viral clearance was defined as two consecutive swabs without diagnostic/sgRNA detected after peak viral load. Participants who did not reach the viral clearance endpoint were right-censored. Cumulative incidence and median event time were calculated using the Kaplan-Meier method.

Using a piece-wise linear mixed-effects model (Supplemental Material), we estimated the diagnostic and sgRNA viral load trajectory with data from the trajectory modeling cohort. [16] We compared three viral dynamics characteristics between the diagnostic and sgRNA viral load trajectories: magnitude of peak viral load, time from viral shedding onset to peak, and time from peak to viral clearance. Analyses were performed using R, Table 1 and Supplemental Figure S1 ). Among the 202 participants who were positive for both diagnostic and sgRNA, 162 (80%) had their peak diagnostic RNA and peak sgRNA viral load observed on the same day; 13 (6%) and 27 (13%) had their peak sgRNA viral load observed before and after the peak diagnostic viral load, respectively. A total of 187 (93%) participants had their peak sgRNA viral load observed within 1 day of the peak diagnostic RNA viral load ( Table 1 and Figure 1 ). Among the 75 participants who only had a single positive diagnostic RNA swab and a test for sgRNA, only 2 (2.7%)

were positive for sgRNA ( Table 1 and Supplemental Figure S2 ).

A c c e p t e d M a n u s c r i p t A total of 2,123 swabs were PCR-positive for SARS-CoV-2, of which 1932 (91%) were tested for sgRNA. Of those tested for sgRNA, 767 (40%) had detectable sgRNA ( Table 1) . The overall Pearson correlation coefficient between the diagnostic and sgRNA viral loads was 0.93, which was similar in the PEP (0.93) and Treatment (0.90) samples. Despite fewer available swabs collected before the peak diagnostic RNA viral load, there was an overall Pearson correlation coefficient of 0.93 before and after the peak ( Table 1) . The sgRNA viral load was consistently low in samples with low diagnostic viral load and appeared to follow a positive linear trend with the diagnostic RNA viral load on the log 10 copies/mL scale beyond a certain threshold (Figure 2) . Fitting a hinge regression model, at a diagnostic RNA viral load of 5.1 log 10 copies/mL (95% CI: 5.0, 5.3) there was a marked increase in positivity for sgRNA. Among the 627 samples with diagnostic RNA viral load  5.1 log 10 copies/mL that were tested for sgRNA, 599 (96%) were positive for sgRNA. The Pearson correlation coefficient between diagnostic and sgRNA viral load among those with diagnostic RNA viral load ≥5.1 log10 copies/mLis 0.94, drastically higher than that among the swabs with diagnostic RNA viral load < 5.1 log 10 copies/mL (0.28) (Table 1 and Figure 2) .

To evaluate the accuracy of the diagnostic RNA viral loads in identifying samples positive for sgRNA, we constructed the empirical receiver operating characteristic (ROC) curves using various diagnostic RNA viral load thresholds (Figure 3) . We identify an "optimal" diagnostic viral load cutoff of 4.5 log 10 copies/mL that provided a sensitivity of 90% and specificity of 91%. This cutoff is "optimal" in the sense that it resulted in a point on the ROC that is closest to the true positive rate (TPR, sensitivity) of 100% and false positive rate (FPR, 1-specificity) of 0%. Samples from the pre-peak subset demonstrated greater areas under the ROC, with an "optimal" cutoff of 4.8 log 10 copies/mL that provided a sensitivity of 90% and specificity of 95%.

Within the sustained shedding cohort that consisted of 187 participants, the sgRNA detection rate culminated at 66% on the day of the observed peak diagnostic RNA (Figure 4a) . Setting the sgRNA A c c e p t e d M a n u s c r i p t viral load at 0 log 10 copies/mL for swabs without sgRNA detected, the mean ratio of sgRNA to diagnostic viral load also peaked on the day of the peak diagnostic RNA, mimicking the trend of sgRNA detection rate (Figure 4b) . For those swabs with detectable sgRNA, the mean sgRNA to diagnostic viral load ratio remained relatively constant over time (Figure 4b) . Using a linear mixedeffects model (Supplemental Material), the mean difference between the sgRNA and diagnostic RNA viral load was not statistically significantly different before and after peak (p = 0.33).

Among the 238 participants in the uncensored peak cohort, we conducted a Kaplan-Meier analysis to calculate the median time to diagnostic or sgRNA viral clearance, defined as two consecutive swabs without diagnostic/sg RNA detected ( Figure 5 ). The median time to diagnostic RNA clearance was 8 days (95% confidence interval, CI: 7 -9) among individuals with at least one positive sgRNA test (black) and 1 day (95% CI: 1 -1) for individuals with no positive sgRNA tests (gray). For those with at least one positive sgRNA test (red), time to sgRNA clearance was 3 days (95% CI: 3 -4).

We fit two separate piece-wise linear mixed-effects models for diagnostic and sgRNA viral load trajectories for the 153 participants in the trajectory modeling cohort. The estimated peak diagnostic viral load was 7.2 log 10 copies/mL (95% credible interval, CI: 7.0 -7.3) and occurred on average 1.8 days (95% CI: 1.5 -2.1) after shedding onset with viral clearance reached 12.2 days (95% CI: 11.8 -12.6) after the peak. In contrast, sgRNA viral load was estimated to reach a lower peak of 5.7 log 10 copies/mL (95% CI: 5.5-5.8) 1.2 days (95% CI: 1.1 -1.4) after shedding onset and to reach viral clearance much faster, about 6.7 days (95% CI: 6.3 -7.1) after peak (Figure 6 ). Additional results from the models are presented in Supplemental Material (Figures S3 and S4) . Studies of SARS-CoV-2 transmission and antiviral therapies may benefit from a better distinction between non-infectious viral genomic material and replicating virus, particularly if this distinction A c c e p t e d M a n u s c r i p t can be provided with PCR assays currently in use. The detection of sgRNA may indicate replicating virus without the potential loss of sensitivity or necessary BSL-3 containment of viral culture. [3, 6] In addition to the potential clinical utility of sgRNA assays themselves, a diagnostic PCR with a validated cutoff below which sgRNA is reliably detected may similarly allow for improved identification and monitoring of individuals with actively replicating virus, which could have implications for isolation periods.

The trajectories of the diagnostic and sgRNA viral loads over time differed, with sgRNA peaking slightly sooner (1.2 vs 1.8 days after shedding onset) and with faster clearance at 6.7 days from peak compared to >12 days for diagnostic PCR. Although real time prediction of peak viral load is not feasible, available studies indicate that symptom onset occurs on or soon after peak viral load in community cases. [16] Assuming individuals test soon after the onset of symptoms, the current Centers for Disease Control (CDC) recommendations for a 5 day isolation period approach the duration of detectable sgRNA in our study.

[22] However, the onset of isolation may be variable depending on access to testing. Additionally, this study was conducted when the D614G variant was circulating and prior to the availability of vaccines, and may vary with currently circulating variants or in vaccinated individuals.

One caveat of this study is that detection of sgRNA is not identical to detection of infectious virions, particularly at later timepoints after infection and when neutralizing antibodies are present. [21] Thus, this study may overestimate the potential infectiousness of later samples. In contrast, viral culture, while highly specific for infectious virus, is less sensitive and may underestimate replicationcompetent virus. In an outpatient study evaluating viral culture compared to detection by PCR, a threshold of 6.4 log 10 copies/mL optimized sensitivity and specificity at 0.81 and 0.9, A c c e p t e d M a n u s c r i p t respectively.

[21] In comparison, using detectable sgRNA as an indicator of replicating virus we show reliable detection with diagnostic RNA viral titers of 5.1 log 10 copies/mL and greater.

Another limitation of this study is that we did not target the same genes for sgRNA and diagnostic RNA assays. The lower viral loads resulting from sgRNA PCR compared to paired diagnostic RNA may reflect the lower molar ratios of E transcript (sgE) compared to N transcript (sgN) for replicating virus. [2, 3] Genomic N is a common target for diagnostic SARS-CoV-2 assays. However, the transition from detectable to undetectable sgE transcripts appears to better track with the transition from infectious virus to residual genetic material than the detection of sgN, possibly due to the relative abundance of sgE vs sgN transcripts. [9, 19] Compared to other studies of viral shedding, we had the advantage of capturing the full course of infection from first detectable virus by RT-PCR. These results may not be directly comparable to samples collected late after the onset of symptoms, and instrument variability and differences in sample collection may limit reproducibility of precise viral load cutoffs. Further, the study population was entirely unvaccinated, and the diagnostic and subgenomic viral load correlations may be altered in vaccinated individuals. These data can complement studies of transmission, preand post-exposure prophylaxis, and early therapy by providing a PCR threshold that can correlate with replicating virus independent of symptoms or duration of illness. In the context of frequent testing, these data could inform the threshold viral load at which treatment and prevention interventions have the potential to decrease viral replication with clinical benefits for persons with SARS-CoV-2.

M a n u s c r i p t Figure 1 . The distribution of days from the observed peak diagnostic RNA viral load to the observed peak subgenomic RNA (sgRNA) viral load. Among the 202 participants who had at least one positive diagnostic RNA swab that was also tested positive for sgRNA, 162 (80%) had their peak diagnostic RNA and peak sgRNA viral load observed on the same day, 13 (6%) had their peak sgRNA viral load observed before the peak diagnostic viral load, and 27 (13%) had their peak sgRNA viral load observed after the peak diagnostic viral load. A total of 187 (93%) participants had their peak sgRNA viral load observed within 1 day of the peak diagnostic RNA viral load. positive diagnostic RNA that were tested for sgRNA. An "optimal" cutoff was calculated as the diagnostic RNA viral load threshold (in log 10 copies/mL) that maximized the sum of sensitivity (true positive rate) and specificity (1 -false positive rate). This cutoff is "optimal" in the sense it weighs both sensitivity and specificity equally. Overall, to ensure a sensitivity of 90%, the highest diagnostic RNA viral load threshold above which a sample was identified to be positive for sgRNA was 4.5 log 10 copies/mL. among the 238 participants in the uncensored peak cohort, defined as the subgroup who had an uncensored peak diagnostic viral load during the 14-day follow-up (i.e., peak diagnostic RNA viral load was not measured on the first or last swab collected). Time to diagnostic RNA viral clearance is depicted by Kaplan-Meier curves for individuals with at least one positive sgRNA test (black) or no detectable sgRNA tests (gray). Clearance of detectable sgRNA for individuals with at least one positive sgRNA test is depicted in red. Viral clearance was defined as two consecutive swabs without diagnostic/sg RNA detected.

Nidovirus transcription : how to make sense

Continuous and Discontinuous RNA Synthesis in Coronaviruses

Comparison of Subgenomic and Total RNA in SARS-CoV-2 Challenged Rhesus Macaques

Coronavirus genomic RNA packaging

Duration of infectiousness and correlation with RT-PCR cycle threshold values in cases of COVID-19

SARS-CoV-2 Virus Culture and Subgenomic RNA for Respiratory Specimens from Patients with Mild Coronavirus Disease

Prolonged SARS-CoV-2 cell culture replication in respiratory samples from patients with severe COVID-19

Infectious Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) Virus in Symptomatic Coronavirus Disease 2019 (COVID-19) Outpatients: Host, Disease, and Viral Correlates

SARS-CoV-2 Total and Subgenomic RNA Viral Load in Hospitalized Patients

Hydroxychloroquine as Postexposure Prophylaxis to Prevent Severe Acute Respiratory Syndrome Coronavirus 2 Infection

Hydroxychloroquine with or without azithromycin for treatment of early SARS-CoV-2 infection among high-risk outpatient adults: A randomized clinical trial

Comparison of commercially available and laboratory-developed assays for in vitro detection of SARS-CoV-2 in clinical laboratories

Predicting infectivity: comparing four PCR-based assays to detect culturable SARS-CoV-2 in clinical samples

Mucosal Shedding of Human Herpesvirus 8 in Men

Estimation and inference for upper hinge regression models

Trajectory of Viral RNA Load Among Persons With Incident SARS-CoV-2 G614 Infection (Wuhan Strain) in Association With COVID-19 Symptom Onset and Severity

JAGS: a program for analysis of Bayesian graphical models using Gibbs sampling

A c c e p t e d M a n u s c r i p t A c c e p t e d M a n u s c r i p t A c c e p t e d M a n u s c r i p t A c c e p t e d M a n u s c r i p t Tables and Figures   Table 1 . Individual-level and swab-level summary statistics for the diagnostic and subgenomic RNA (sgRNA) data from the Post-Exposure Prophylaxis (PEP) study, the Treatment (Tx) study, and both studies combined. *The distribution of days from the observed peak diagnostic RNA viral load to the observed peak sgRNA viral load is presented in Figure 1 . No. of participants with peak sgRNA viral load observed before peak diagnostic RNA viral load* 13 5 8 No. of participants with peak sgRNA viral load observed after peak diagnostic RNA viral load* 27 14 13 Participants with a single positive diagnostics RNA swab (n1) that was also positive for sgRNA (n2), presented as n1/n2 (%) Pearson correlation coefficient between the diagnostic and subgenomic RNA viral load (log 10 copies/mL) among swabs tested positive for both