key: cord-0980200-9sd0zzbt
authors: Dalai, Sudeb C.; Dines, Jennifer N.; Snyder, Thomas M.; Gittelman, Rachel M.; Eerkes, Tera; Vaney, Pashmi; Howard, Sally; Akers, Kipp; Skewis, Lynell; Monteforte, Anthony; Witte, Pamela R.; Wolf, Cristina; Nesse, Hans; Herndon, Megan; Qadeer, Jia; Duffy, Sarah; Svejnoha, Emily; Taromino, Caroline; Kaplan, Ian M.; Alsobrook, John; Manley, Thomas; Baldo, Lance
title: Clinical Validation of a Novel T-cell Receptor Sequencing Assay for Identification of Recent or Prior SARS-CoV-2 Infection
date: 2022-05-06
journal: Clin Infect Dis
DOI: 10.1093/cid/ciac353
sha: fcce1733fac5ea33b16f11b5c7bfe4ecb7a48ae1
doc_id: 980200
cord_uid: 9sd0zzbt

BACKGROUND: While diagnostic, therapeutic, and vaccine development in the COVID-19 pandemic has proceeded at unprecedented speed, critical gaps in our understanding of the immune response to SARS-CoV-2 remain unaddressed by current diagnostic strategies. METHODS: A statistical classifier for identifying prior SARS-CoV-2 infection was trained using >4000 SARS-CoV-2–associated TCRβ sequences identified by comparing 784 cases and 2447 controls from 5 independent cohorts. The T-Detect™ COVID assay applies this classifier to TCR repertoires sequenced from blood samples to yield a binary assessment of past infection. Assay performance was assessed in 2 retrospective (n = 346; n = 69) and 1 prospective cohort (n = 87) to determine positive percent agreement (PPA) and negative percent agreement (NPA). PPA was compared to 2 commercial serology assays, and pathogen cross-reactivity was evaluated. RESULTS: T-Detect COVID demonstrated high PPA in individuals with prior RT-PCR–confirmed SARS-CoV-2 infection (97.1% 15 + days from diagnosis; 94.5% 15 + days from symptom onset), high NPA (∼100%) in presumed or confirmed SARS-CoV-2 negative cases, equivalent or higher PPA than 2 commercial serology tests, and no evidence of pathogen cross-reactivity. CONCLUSION: T-Detect COVID is a novel T-cell immunosequencing assay demonstrating high clinical performance for identification of recent or prior SARS-CoV-2 infection from blood samples, with implications for clinical management, risk stratification, surveillance, and understanding protective immunity and long-term sequelae.

Knowledge gaps in our understanding of immunity to SARS-CoV-2 infection translate into 2 critical areas of unmet need in diagnosis and management of COVID-19 and epidemiologic 3 monitoring of the pandemic. Serologic testing of IgM, IgG, and/or IgA isotypes has been the 4 primary modality for identifying prior SARS-CoV-2 infection, estimating disease prevalence, 5 and evaluating immunity [1, 2] . Although antibody testing has been shown to capture a larger 6 percentage of exposures than polymerase chain reaction (PCR) testing [3] , it is limited by 7 interassay variability [4]; low or absent antibody titers in individuals with asymptomatic or mild 8 infection [5,6]; declining antibody levels over time [7] ; and false-positive results [1] . It also 9 remains unclear whether serology results correlate with long-term protective immunity or 14 Humoral responses vary among vaccinated or exposed individuals, and 5%-20% of individuals 15 recovered from SARS-CoV-2 infection may have no detectable antibodies, depending on isotype 16 and disease severity [10] [11] [12] . Multiple lines of evidence support a central role of the cellular 17 response in SARS-CoV-2 immunity [13, 14] . The majority of patients diagnosed with COVID-18 19, including convalescent patients presenting across a wide spectrum of disease severity, 19 generate CD8 + and CD4 + T-cell responses [9, 15, 16] , which have been associated with milder 20 disease and protection from infection [17, 18] . T cells also play a critical role in activating the 21 humoral response and can precede antibodies as the first detectable immune response to CoV-2 infection, particularly in asymptomatic or mild illness [15] . SARS-CoV-2-specific T [13, [19] [20] [21] . Additionally, the observation that some emerging variants of 2 concern (VOCs) evade antibody responses while largely preserving the T-cell response [16, 22] 3 underscores the critical importance of understanding the resulting effects on infectivity and 4 vaccine-induced immunity. 5 Features inherent to the T-cell response make it a desirable target for identifying and tracking 6 disease exposure. The cellular immune response is sensitive, antigen-specific, and is amplified 7 through expansion of clones that circulate in the blood and are maintained in long-term memory. 8 Here we describe the implementation and clinical validation of T-Detect™ COVID, a novel 9 high-throughput assay that has received Emergency Use Authorization (EUA) for determining 10 recent or prior SARS-CoV-2 infection based on T-cell receptor gene sequencing and subsequent 11 repertoire profiling from whole blood samples [23] . We demonstrate high positive percent 12 agreement (PPA) and negative percent agreement (NPA) in PCR-confirmed SARS-CoV-2 cases 13 across several cohorts and longitudinal timepoints. This assay has equivalent or better 14 performance than commercially available EUA antibody tests and lacks cross-reactivity to 15 several respiratory pathogens.

All samples were collected pursuant to an Institutional Review Board-approved clinical study 

For the PPA and NPA clinical validation studies, samples from both arms were analyzed using 14 the T-Detect COVID assay (Tables 1 and 2) . A detailed description of the allocation plan and 15 study cohorts is included in Supplemental Methods. When available, paired serum samples from 16 cohorts used for secondary PPA analyses (Table 1 ; n=77) were tested using 2 different EUA 17 antibody assays: 1) Elecsys ® Anti-SARS-CoV-2; Roche (all isotypes); and 2) SARS-CoV-2 18 Antibody, IgG; LabCorp (see Supplemental Methods for details). sequences in the sample. A 5-fold cross-validation of the training set was used to identify the 10 final P-value cutoff in the 1-tailed Fisher's exact test that yielded the optimal area under the 11 receiver operating characteristic (AUROC) and was also used to evaluate assay performance 12 with the training data. To further refine the method described in our previous study [19] , the 13 diagnostic model threshold was set to 99.8% specificity against an independent holdout set 14 of 1657 negative controls and 100 positive controls. 16 Process Overview 17 T-Detect COVID consists of a core assay designed to sequence and quantify rearranged 18 TCRβsequences from genomic DNA (gDNA) extracted from peripheral blood and diagnostic 19 software that applies the COVID-specific algorithm to the TCRβ sequence repertoire data to 20 determine a result. An overview of the assay is shown in Figure 1 and described below.

Peripheral whole blood is collected in a 10 -mL ethylenediaminetetraacetic acid (EDTA) 2 vacutainer tube and shipped overnight at ambient temperature to the Adaptive Biotechnologies' 3 clinical laboratory. Upon receipt, the sample is accessioned and stored at 4°C for processing that 4 same day via automated gDNA extraction or stored at -80°C for later extraction. Detailed methods for sample preparation, immunosequencing, and pipeline analysis have been 7 described [19, 26] . Briefly, a target gDNA sample input of 18 µg is isolated from 2 mL fresh or 8 frozen peripheral whole blood (6 mL requested). This target gDNA input ensures that samples 9 meet the minimum unique productive rearrangements input quality control (QC) specification. A 10 multiplex PCR strategy with control synthetic TCRβ molecules added to each reaction is used to 11 amplify rearranged TCR sequences from gDNA. PCR libraries are loaded together on a single 12 sequencing run, and sequencing is performed using the Illumina NextSeq 500/550 System.

Sequence data are extracted, and reads are attributed to data derived from biological versus 14 synthetic templates to calculate template estimates for each identified receptor sequence, as well 15 as input cell counts. 16 

The SARS-CoV-2-specific algorithm (classifier) is applied to the core assay output for each 18 sample to make the COVID-positive/negative call based on the resulting score. As described 19 above, the classifier identifies and quantifies any SARS-CoV-2-associated TCRs from a 20 predetermined list of several thousand SARS-CoV-2-associated TCRs, and it also quantifies all 21 unique non-SARS-CoV-2 TCRs identified. These two variables were used in the machine

learning classifier to produce the final score for each sample. The total number of unique TCR 1 sequences must fall within a threshold for the algorithm to produce a valid result. The pre-2 specified threshold is then applied to classify the patient sample as positive or negative for an 3 immune response to SARS-CoV-2. The classifier was locked to create the COVID-specific 4 algorithm for T-Detect COVID prior to clinical validation. We began with the previously described classifier [19] , which was trained on 784 SARS-CoV-2-9 positive cases (from 5 independent cohorts, detailed in Supplemental Table 1 ) and 2448 controls 10 (from 4 independent cohorts, detailed in Supplemental Table 1 ). Adjustment of the training set to of the training data and using the independent holdout set described above. Performance of the 16 classifier was generally robust to age and sex, although age was weakly associated with COVID While developing a classifier for Lyme disease, we found that incorporating additional 3 sequences that did not meet strict enrichment thresholds but showed other evidence of disease 4 association, such as shared sequence similarity with a Lyme-associated sequence and/or elevated (Table 1 ). In the secondary PPA study, all 77 independent samples tested (from the 18 ImmuneRACE and ImmuneSense COVID-19 cohorts) were from unique individuals, passed QC 19 and threshold requirements, and were included for analysis (Table 1) . Samples were collected a 20 maximum of 106 days from symptom onset. PPA for the T-Detect COVID assay was highest 21 (97.1%) 15 days since diagnosis in the primary analysis and 15 days since symptom onset 22 (94.5%) in the secondary analysis (Table 3) .

Two NPA studies were undertaken to evaluate T-Detect COVID assay performance: a primary 2 NPA analysis of retrospectively sourced whole blood samples from pre-pandemic timepoints 3 (July 2017-Nov 2019; independent samples belonging to the DLS cohort) presumed to be 4 SARS-CoV-2 negative and a secondary NPA analysis of samples prospectively collected from 5 symptomatic individuals in the ImmuneSense COVID-19 cohort who tested negative for SARS- CoV-2 (Table 2 ). In the primary NPA study, 87 of 124 samples were from unique individuals, 7 passed all standard QC and assay threshold requirements, and were included for analysis, 8 yielding an NPA of 100% (Table 4 ). The majority of excluded samples failed to meet assay QC 9 criteria or assay-specific thresholds, which may have been due to the variable collection (Table 4) . 16 17 Additional analyses compared the PPA of the T-Detect COVID assay with that of antibody 18 testing in paired SARS-CoV-2-positive samples from 77 individuals (used in secondary PPA 19 analyses). Results of these analyses showed that the PPA for T-Detect COVID was as high or 20 higher than that of serology, particularly in the early phases of infection (Table 5) . (Table 6) . In this study, we describe a TCR sequence-based assay for identifying recent or prior SARS- 20 CoV-2 infection in whole blood samples that demonstrates high PPA in confirmed SARS-CoV- reaction; SARS-CoV-2, severe acute respiratory syndrome coronavirus 2. Hopkins Lyme, FHCRC cancer, Moffitt pancreatic cancer, and independent DLS samples).

Supplemental Table 1 provides additional cohort details. 

CoV-2 infection and implications for immunity

Immunological memory to SARS-CoV-2 assessed for up 6 to 8 months after infection

Immune evasion of SARS-CoV-2 8 emerging variants: What have we learnt so far? Viruses

Robust T cell immunity in 10 convalescent individuals with asymptomatic or mild COVID-19

Impact of SARS-CoV-2 variants on the total CD4 + and 13

CD8 + T cell reactivity in infected or vaccinated individuals

Antigen-specific adaptive 16 immunity to SARS-CoV-2 in acute COVID-19 and associations with age and disease 17 severity

SAR-CoV-2 responsive T cell numbers are 19 associated with protection from COVID-19:A prospective cohort study in keyworkers

MedRxiv 20222778