key: cord-326983-h6gdck2u
authors: Ferretti, Andrew P.; Kula, Tomasz; Wang, Yifan; Nguyen, Dalena M.V.; Weinheimer, Adam; Dunlap, Garrett S.; Xu, Qikai; Nabilsi, Nancy; Perullo, Candace R.; Cristofaro, Alexander W.; Whitton, Holly J.; Virbasius, Amy; Olivier, Kenneth J.; Buckner, Lyndsey R.; Alistar, Angela T.; Whitman, Eric D.; Bertino, Sarah A.; Chattopadhyay, Shrikanta; MacBeath, Gavin
title: Unbiased screens show CD8+ T cells of COVID-19 patients recognize shared epitopes in SARS-CoV-2, most of which are not located in the Spike protein
date: 2020-10-20
journal: Immunity
DOI: 10.1016/j.immuni.2020.10.006
sha: 
doc_id: 326983
cord_uid: h6gdck2u

Developing effective strategies to prevent or treat COVID-19 requires understanding the natural immune response to SARS-CoV-2. We used an unbiased, genome-wide screening technology to determine the precise peptide sequences in SARS-CoV-2 that are recognized by the memory CD8+ T cells of COVID-19 patients. In total, we identified 3–8 epitopes for each of the six most prevalent human leukocyte antigen (HLA) types. These epitopes were broadly shared across patients and located in regions of the virus that are not subject to mutational variation. Notably, only 3 of the 29 shared epitopes were located in the spike protein, whereas most epitopes were located in ORF1ab or the nucleocapsid protein. We also found that CD8+ T cells generally do not cross-react with epitopes in the four seasonal coronaviruses that cause the common cold. Overall, these findings can inform development of next-generation vaccines that better recapitulate natural CD8+ T cell immunity to SARS-CoV-2.

Identification of CD8 + T cell epitopes in five HLA-A*01:01 patients 

Coronavirus Disease 2019, or COVID-19, is a global pandemic that has claimed over a million lives world-wide and has affected countless more. Developing effective vaccines and therapies requires understanding how the virus and the immune response affect disease pathology and how the adaptive immune system recognizes and ultimately clears the virus. To date, most efforts have focused on the B cell-mediated antibody response to the virus. Notably, the vast majority of current vaccine development efforts are focused on eliciting neutralizing antibodies to the virus, most frequently by immunizing with the spike (S) protein of SARS-CoV-2, or even with just the receptor binding domain (RBD) of the S protein (Vabret et al., 2020) . How cytotoxic CD8 + T cells recognize and clear infected cells is less understood. In individuals who recovered from the closely related coronavirus, SARS-CoV, virus-specific memory CD8 + T cells persist for at least six to eleven years, whereas memory B cells and anti-viral antibodies are largely undetectable at these later timepoints (Peng et al., 2006; Tang et al., 2011) . Similarly, antibody responses to SARS-CoV-2 can be detected in most COVID-19 patients 10 -15 days following symptom onset, but responses decline to baseline in many patients within 3 months (Seow et al., 2020) . These findings suggest that vaccines focused solely on eliciting neutralizing antibodies to the S protein may be insufficient to elicit long-term immunity to coronaviruses. In mice infected with SARS-CoV, virus-specific CD8 + T cells are sufficient to enhance survival and diminish clinical disease (Zhao et al., 2010) and immunization with a single immunodominant CD8 + T cell epitope confers protection from lethal viral infection (Channappanavar et al., 2014) .

These studies highlight the importance of understanding the natural CD8 + T cell response to SARS-CoV-2 as a route to designing more durable vaccines.

SARS-CoV-2-specific CD8 + T cells can be detected both in convalescent patients (Braun et al., 2020; Grifoni et al., 2020; Le Bert et al., 2020; Peng et al., 2006; Sekine et al., 2020; Thieme et al., 2020) and in subjects participating in vaccine trials (Folegatti et al., 2020; Jackson et al., 2020; Sahin et al., 2020) . However, these studies used complex pools of predicted epitopes, and it is therefore not clear what specific epitopes are being recognized and, in the case of vaccine trials, if the epitopes being recognized are the ones driving the natural CD8 + T cell response to viral infection. To circumvent potential bias introduced by epitope prediction algorithms, we built upon an unbiased, genome-wide screening technology (Kula et al., 2019) to simultaneously screen all of the memory CD8 + T cells in convalescent patients against every possible epitope in SARS-CoV-2. We focused on memory cells to identify epitopes that are functionally recognized during the course of SARS-CoV-2 infection and included patients with a J o u r n a l P r e -p r o o f range of symptoms to determine if any obvious associations are observed between CD8 + T cell response and disease severity. CD8 + T cells in these patients responded to a few highly antigenic epitopes in SARS-CoV-2 that were shared among patients with the same HLA type.

These epitopes were largely unique to SARS-CoV-2, were invariant among viral isolates, were frequently targeted by multiple clonotypes within each patient, and did not occur in "common cold" coronaviruses. Only ~10% of the epitopes were found in the S protein, with ~50% located in ORF1ab and the highest density of epitopes located in the nucleocapsid (N) protein. These results provide the necessary tools to better understand the CD8 + T cell response in and provide a path to the design and development of next-generation vaccines.

To determine the global landscape of CD8 + T cell recognition in an unbiased fashion, we built upon a genome-wide screening technology, termed T-Scan (Kula et al., 2019) , that enabled us to simultaneously screen all the memory CD8 + T cells in a patient, one HLA allele at a time, against every possible viral epitope in SARS-CoV-2, as well as the four seasonal coronaviruses that cause the common cold ( Figure 1A ). Briefly, CD8 + T cells were co-cultured with a genomewide library of target cells (modified HEK 293 cells), engineered to express a single HLA allele.

Each target cell in the library also expressed a different 61-amino acid (61-aa) protein fragment.

These fragments were processed naturally by the target cells and the appropriate peptide epitopes were displayed on class I MHC molecules on the cell surface. If a CD8 + T cell encountered its target in the co-culture, it secreted cytotoxic granules into the target cell, inducing apoptosis. Early apoptotic cells were then isolated from the co-culture and the expression cassettes sequenced, revealing the identity of the protein fragment. Because the assay is non-competitive, hundreds to thousands of T cells were screened against tens of thousands of targets simultaneously. To address the bottleneck of extensive sorting needed to isolate rare recognized target cells in high complexity libraries (Kula et al., 2019) , we engineered the target cells to express both a Granzyme B (GzB)-activated fluorescent reporter, as previously described, as well as a GzB-activated version of the scramblase enzyme XKR8, which drives the rapid and efficient transfer of phosphatidylserine to the outer membrane of early apoptotic cells (Supplemental Methods and Figure 1A ). Early apoptotic cells were then enriched by magnetic-activated cell sorting with Annexin V, followed by fluorescence-activated J o u r n a l P r e -p r o o f sorting with the fluorescent reporter. This modification increased the throughput of the T-Scan assay 20-fold, enabling the rapid processing of a large number of patient samples.

To comprehensively map responses to SARS-CoV-2, we generated a library of 61-aa protein fragments that tiled across all 11 open reading frames (ORFs) of SARS-CoV-2 in 20-aa steps ( Figure 1B) . To capture the known genetic diversity of SARS-CoV-2, we included all protein-coding variants from the 104 isolates that had been reported as of March 15, 2020. We also included the complete set of ORFs (ORFeome) of SARS-CoV and the four endemic coronaviruses that cause the common cold (betacoronaviruses HKU1 and OC43, and alphacoronaviruses NL63 and 229E). As positive controls, we included known immunodominant antigens from Cytomegalovirus, Epstein-Barr virus, and Influenza virus (Currier et al., 2002) .

Finally, each protein fragment was represented ten times, each encoded with a unique nucleotide barcode to provide internal replicates in our screens, for a final library size of 43,420

clones.

To identify the epitopes functionally recognized during the course of SARS-CoV-2 infection, we collected peripheral blood mononuclear cells from 78 adult patients who had tested positive by viral PCR (swab test), had recovered from their disease, and had been out of quarantine according to Centers for Disease Control and Prevention guidelines for at least two weeks (Supplemental Methods). Patients were recruited at either of two centers: Atlantic Heath System in Morristown, NJ and Ochsner Medical Center in New Orleans, LA. All patients were HLAtyped, and a summary of their characteristics and HLA types are provided in tables S1 and S2.

As HLA A*02:01 is the most common MHC allele world-wide (Gonzalez-Galarza et al., 2020; Maiers et al., 2007) , we started by selecting nine HLA-A*02:01 patients with a broad range of clinical presentations: six had mild symptoms and were not hospitalized, two required supplemental oxygen, and one required invasive ventilation. In each case, we purified bulk memory CD8 + T cells (CD8+, CD45RO+, CD45RA-, CD57-) by negative selection, expanded the cells with antigen-independent stimulation (anti-CD3), and screened them against the SARS-CoV-2 library. Target cells expressing only HLA-A*02:01 were used to provide unambiguous MHC restriction of discovered antigens. The SARS-CoV-2 screening results for one representative patient and one COVID-19-negative healthy control are shown in Figure 1C .

We found reactivity to at least eight regions of SARS-CoV-2 proteins in the convalescent patient and none in the control. Importantly, we observed reproducible performance of four technical screen replicates, internal nucleic acid barcodes, and overlapping protein fragments, collectively J o u r n a l P r e -p r o o f suggesting robust screen performance. Additionally, we detected reactivity to the control CMV epitope (NLVPMVATV) in the healthy control, who was known to be CMV-positive, and reactivity to two EBV epitopes in both the COVID-19 patient and the healthy control ( Figure 1C ).

Next, we examined the screen results for the full set of nine HLA-A*02:01 patients and detected reactivity to specific segments of SARS-CoV-2 ORFs in 8/9 patients ( Figure 2A ). In keeping with what has been observed for other viruses (Yewdell, 2006) , we found that specific fragments of SARS-CoV-2 were recurrently recognized by the T cells of multiple patients (i.e., are immunodominant). For example, ORF1ab aa 3881-3900 and S aa 261-280 were each recognized by 7/9 patients ( Figure 2A ). Overall, we identified six regions that were targeted by CD8 + T cells from at least three different patients. In addition to being shared across patients, these regions were among the strongest responses observed in each patient.

We next sought to identify the precise peptide epitopes underlying the shared T cell reactivities detected in our screens. The overlapping design of our antigen library allowed us to map the T cell reactivities to specific 20-aa segments. We then used NetMHC4.0 (Andreatta and Nielsen, 2016; Nielsen et al., 2003) to identify specific, high-affinity HLA-A*02:01 peptides in each pre-identified 20-aa stretch. An example of a predicted epitope and the corresponding screen data is shown in Figure 2B . Notably, the fragments scoring in our screens were enriched for high-affinity HLA-binding peptides compared to the library as a whole ( Figure S1 ). To visualize the results across all nine patients, we collapsed the screening data into a single value (mean of screen replicates and redundant tiles), revealing a set of six predicted epitopes that were recurrently recognized by three or more patients ( Figure 2C , Table 1 ). We then synthesized peptides corresponding to each epitope to validate our findings. All six epitopes induced peptide-dependent T-cell activation as determined by interferon-gamma (IFNγ) secretion ( Figure 2D ) and CD137 upregulation ( Figure S2 ). Both IFNγ secretion and CD137 upregulation correlated with the fold enrichment in the T-Scan screen ( Figures S2 and S3 ). As further validation, we constructed MHC tetramers with the six peptides and used them to stain the memory CD8 + T cells of all nine A*02:01 patients, as well as an additional test-set of 18 A*02:01 patients that had not been screened. Positive tetramer staining was observed in a subset of patients for all six peptides, including patients in the independent test-set ( Figure 2E ).

Additionally, the magnitude of enrichment in the screens correlated well with the frequency of cognate T cells in the patient samples (r = 0.73, p < 0.0001; Figure 2F ), allowing us to determine that our screens detected the targets of T cells that were present at ≥0.1% frequency J o u r n a l P r e -p r o o f in the memory CD8 + T cell pool. Notably, the three most commonly recognized epitopes we discovered -KLW, YLQ, and LLY -were each recognized by 67% of the patients we screened, and all nine patients had a detectable response to at least one of the top three epitopes ( Figure   2G ). A similar analysis of the tetramer staining data in all 27 A*02:01 patients showed recognition of at least one of these epitopes in 23/27 patients (85%; Figure 2H ). Taken together, these analyses revealed the limited set of A*02:01-restricted shared epitopes recognized by patient T cells.

As CD8 + T cell responses are profoundly shaped by host MHC alleles, we next mapped memory CD8 + T cell reactivities for five additional MHC alleles: HLA-A*01:01, HLA-A*03:01, HLA-A*11:01, HLA-A*24:02, and HLA-B*07:02. Along with HLA-A*02:01, these alleles provide a broad perspective on the nature of anti-SARS-CoV-2 CD8 + T cell immunity, as ~90% of the U.S.

population and ~85% of the world population is positive for at least one of the six alleles we examined (Gonzalez-Galarza et al., 2020; Maiers et al., 2007) . For each allele, we selected five HLA + convalescent COVID-19 patients and screened their memory CD8 + T cells against the SARS-CoV-2 library in target cells expressing only the single HLA of interest. As some patients were positive for more than one allele, their T cells were used in more than one HLA-specific screen. A total of 25 distinct patients were needed for the 34 HLA-specific screens. As with A*02:01 patients, we found robust T cell recognition of multiple regions in the SARS-CoV-2

ORFeome for patients with each HLA allele ( Figure S4 ) and confirmed that the scoring fragments were enriched for predicted high-affinity MHC binders for each respective allele ( Figure S1 ). We again observed recurrent recognition of specific protein fragments by most or all patients for each allele ( Figure 3A ), indicating a narrow set of shared responses. As before, we used NetMHC4.0 to identify the precise epitopes underlying the top hits from our screens, and validated these peptides using IFNγ secretion ( Figure 3B ) and CD137 upregulation ( Figure   S2 ). We identified three or more recurrently recognized epitopes for each screened MHC allele and found that 92% of patients recognized at least one of the top three allele-specific epitopes for these five additional HLA types ( Figure 3C ). Collectively, we mapped and validated 29 CD8 + T cell epitopes that were shared among COVID-19 patients with the same HLA type (Table 1) .

These epitopes represent the global landscape of MHC class I immunodominance in SARS-CoV-2 across the six most prevalent HLA types.

The unbiased antigen mapping we performed enabled us to interrogate various features of CD8 + T cell immunity to SARS-CoV-2. First, we examined the scope of recognized viral J o u r n a l P r e -p r o o f proteins. We observed broad reactivity to many SARS-CoV-2 proteins, including ORF1ab, S, N, M, and ORF3a ( Figure 4A ). Notably, only three of the 29 epitopes were located in the S protein, with most (15 of 29) located in ORF1ab and the highest density of epitopes located in the N protein. When taken in aggregate, our results are consistent with previous ORF-level analyses using peptide pools (Altmann and Boyton, 2020; Braun et al., 2020; Grifoni et al., 2020; Le Bert et al., 2020; Thieme et al., 2020) . However, our approach provided an increased level of granularity that enabled identification of specific epitope sequences and highlighted HLA allelespecific differences. For example, we observed shared epitopes in the S protein for HLA-A*02:01, HLA-A*03:01, and HLA-A*24:02, but not for HLA-A*01:01, HLA-A*11:01, or HLA-B*07:02. Notably, we detected only one recurrent response in the receptor-binding domain (RBD) of the S protein (KCY on HLA-A*03:01).

Next, we asked how the CD8 + T cell response to SARS-CoV-2 intersects with the emerging genetic diversity of the virus. Recent analyses, which examined the genome sequences of over 10,000 isolates of SARS-CoV-2 sampled from 68 different countries, identified a set of 28 non-synonymous coding mutations detected in at least 1% of strains (Koyama et al., 2020) . Only one of these mutations (M protein T175M, detected in 2% of strains) was found in the shared epitopes we identified (HLA-A*01:01 ATS and HLA-A*11:01 ATS). This suggests that the recognition of the epitopes we identified is unlikely to be significantly influenced by the SARS-CoV-2 genetic diversity observed thus far.

Identifying specific SARS-CoV-2 epitopes allowed us to examine the features of the T cell receptors (TCRs) recognizing these shared epitopes. We used tetramers loaded with three HLA-A*02:01 epitopes (KLW, YLQ, and LLY) to stain and sort antigen-specific memory CD8 + T cells from the initial nine HLA-A*02:01-positive convalescent COVID-19 patients. For each of the other five HLA alleles, we used tetramer or CD137 staining to sort CD8 + T cells reactive to the 3-4 most frequently shared epitopes in two patients each. We then used 10x Genomics single-cell sequencing to identify the paired TCR α and TCR β chains expressed by these T cells. Collectively, we found TCRs recognizing 17 shared epitopes for a total of 421 SARS-CoV-2-reactive TCRs. Next, we examined the TCR sequences themselves, focusing on the three HLA-A*02:01 antigens that were explored across a larger set of patients. We identified paired clonotypes reactive to each antigen in 5/9 (KLW, ALW) or 6/9 (YLQ) patients. For a majority of responses (9/16), we detected oligoclonal recognition by five or more distinct clonotypes.

Striking similarity was observed among the TCRs recognizing each antigen in terms of Vα gene J o u r n a l P r e -p r o o f segment usage and, to a lesser extent, Vβ usage ( Figure 4B ). Specifically, 26/61 KLW-reactive clonotypes used TRAV38-2/DV8, 24/31 YLQ-reactive clonotypes used TRAV12-1, and 14/29 LLY-reactive clonotypes used TRAV8-1. Notably, these dominant Vα genes were used across all of the patients for whom we identified reactive clonotypes. Taken together, these data suggest that certain TCR Vα regions provide the structural features necessary for high-affinity binding to peptide-MHC, and that these features may explain the recurrent recognition of these epitopes among patients with the same HLA type.

Although our study was not designed to test specific clinical hypotheses, we looked for potential associations between virus-specific T cell responses and clinical characteristics. We focused on the 27 A*02:01 patients for which tetramer staining data were available, as this represents the most uniform set of T cell data in our study. No obvious association was observed between T cell response and sex ( Figure S5 ), but a significant negative correlation was observed with time from diagnosis to blood draw (p=0.0012; Figure 4C and Figure S5 ). This is expected, as antiviral T cells, including effector memory cells, naturally contract following an acute infection (Badovinac et al., 2002; Wherry and Ahmed, 2004) . This observation is important, however, as future epidemiological studies should control for this variable. We also observed a trend in which patients with severe disease exhibited fewer virus-specific T cells than those with mild disease (p=0.041; Figure 4D and Figure S5 ). Additionally, older patients had lower T cell responses than younger patients ( Figure S5 ). These observations should be treated with caution as the number of patients in these studies is small, particularly those requiring invasive ventilation. Appropriately powered studies to address this question are warranted, as they could shed light on whether these shared epitopes are potentially protective against severe disease.

Another key question is how pre-existing immunity to other coronaviruses shapes the CD8 + T cell response to SARS-CoV-2. There are four commonly circulating coronaviruses, OC43, HKU1, NL63, and 229E, and cross-reactive responses to these viruses have been theorized as a potential protective factor during SARS-CoV-2 infection (Sette and Crotty, 2020). Moreover, understanding the extent of cross-reactivity has implications for accurately monitoring T cell responses to SARS-CoV-2 and for optimizing vaccine design. If the immune response to SARS-CoV-2 is shaped by pre-existing CD8 + T cells that recognize other coronaviruses, we reasoned that COVID-19 patients should have reactivity to the regions of the other coronaviruses that J o u r n a l P r e -p r o o f correspond to the SARS-CoV-2 epitopes we identified. We therefore examined T-cell reactivity to SARS-CoV-2, SARS-CoV, and all four endemic coronaviruses in the 34 genome-wide screens that we conducted across all patients and MHC alleles ( Figure 5A ). We observed broad reactivity to the corresponding epitopes in SARS-CoV in over half of cases, consistent with a recent study reporting the existence of long-lasting memory T cells cross-reactive to SARS-CoV-2 in patients that had been infected in SARS-CoV during the 2002/2003 SARS outbreak (Le Bert et al., 2020) . In contrast, we detected almost no reactivity to OC43 and HKU1 (2/29 dominant epitopes) and none to NL63 and 229E. Beyond the 29 epitopes, we observed no reproducible cross-reactivity to any other regions of the four endemic coronaviruses, again suggesting that prior exposure to these viruses is unlikely to provide CD8 + T cell-based protection from SARS-CoV-2.

Mapping the specific shared epitopes in SARS-CoV-2 enabled us to determine the molecular basis for this lack of cross-reactivity. In some cases, the corresponding region is poorly conserved in the other coronaviruses and high-affinity binding to MHC is lost (see, for example, the corresponding regions of the KLW epitope in NL63 and 229E; Figure 5B ). In other cases, the corresponding epitopes are still predicted to bind with high affinity to MHC, but SARS-CoV-2-reactive T cells did not recognize them (see, for example, the corresponding regions of the KLW epitope in OC43 and HKU1; Figure 5B ). In one notable case, we did identify a strong cross-reactive response. The HLA B*07:02 epitope SPR, which lies in the N protein, is highly conserved across betacoronaviruses and all four of the patients that demonstrated reactivity to SPR also exhibited reactivity to the corresponding epitopes in OC43 and HKU1

( Figure 5C ). Overall, however, we conclude that the CD8 + T cell response to SARS-CoV-2 is not significantly shaped by pre-existing immunity to endemic coronaviruses. In contrast to reports that unexposed individuals have T cells that cross-react with SARS-CoV-2 Le Bert et al., 2020; Mateus et al., 2020; Weiskopf et al., 2020) , our data showed that the most expanded memory T cells in convalescent patients did not cross-react with endemic coronaviruses. These discordant results may be explained by our focus on CD8 + T cell responses, whereas most cross-reactivity is detected in CD4 + T cells.

Additionally, it is possible that weakly cross-reactive T cells exist in unexposed individuals, but these T cells are overtaken by de novo responses during SARS-CoV-2 infection. If pre-existing memory responses to other coronaviruses efficiently recognize SARS-CoV-2, the reacting T cells should expand, and their targets would likely have been detected in our screens. As a result, the paucity of identified cross-reactive responses argues against substantial protection against SARS-CoV-2 stemming from CD8 + T cell immunity to the four coronaviruses that cause the common cold. We did identify two epitopes that were shared with OC43 and HKU1, however, which could be of interest in the design of vaccines intended to boost pre-existing T cell immunity.

Our findings have broader implications for SARS-CoV-2 vaccine design. The vast majority of shared epitopes we uncovered (26/29) were located in ORF1ab, N, M, and ORF3a;

only three were in S and only one was in the receptor-binding domain of S. These findings J o u r n a l P r e -p r o o f provide high-resolution insight into peptide pool studies observing responses outside the S protein, and are consistent with the detectable but modest CD8 + T cell responses generated by vaccines targeting the S protein Le Bert et al., 2020; Mulligan et al., 2020) .

Importantly, the protective or pathogenic role of CD8 + T cell responses to specific proteins, individual shared epitopes, or epitopes that are only recognized following vaccination remains to be determined. The epitopes we identified can serve as the basis of experimental and correlational studies to address this critical question. Moreover, our findings enable the design and evaluation of next-generation vaccines that more fully recapitulate the scope of natural CD8 + T cell responses to SARS-CoV-2 infection.

While our screening approach assayed all patient memory CD8 + T cells as a pool, it is best suited for the discovery of targets recognized by the most abundant T cell specificities (≥0.1% based on our estimates). Additional specificities recognized by less frequent T cell clonotypes may have been missed. In addition, sample limitations necessitated polyclonal expansion of the memory CD8 + T cells ex vivo that may have altered the relative abundance of some clonotypes.

Finally, our study was underpowered to evaluate the clinical impact of CD8 + T cells recognizing specific epitopes. Additional studies are needed to determine whether CD8 + T cell responses to individual proteins or epitopes are associated with protection from the virus or specific clinical outcomes.

All donors provided written consent. The study was conducted in accordance with the Declaration of Helsinki (1996) We thank the patients and their families who participated in these studies. We also that Stephen

Elledge and Kai Wucherpfennig for helpful discussions. This work was supported by TScan Therapeutics, a privately-owned biotechnology company. 

Peptides generated in this study are available for research purposes upon signing a materials transfer agreement. Peptides and peptide sequences for commercial purposes (e.g., diagnostics or vaccine development) are available through license agreement.

T-Scan screening data are available upon request. Amino acid sequences for the coronavirus peptidome library (43,420 sequences) are available upon request. T cell receptor sequences for the purposes of therapeutic development are available through license agreement.

All donors provided written consent. The study was conducted in accordance with the Declaration of Helsinki (1996) , approved by the Atlantic Health System Institutional Review Board (IRB) and the Ochsner Clinic Foundation IRB and registered at clinicaltrials.gov (# NCT04397900).

Patients who had recovered from COVID-19 were eligible for this study. They were required to be >18 years of age and have laboratory-confirmed diagnosis of COVID-19 using CDC or state health labs or at hospitals using an FDA Emergency Use Authorized molecular assay. Time since discontinuation of isolation was required to be >14 days and discontinuation of isolation followed CDC guidelines (accessed on March 19, 2020) using either test-based or non-test-J o u r n a l P r e -p r o o f based criteria for patients either in home isolation or in isolation at hospitals. Patients were also required to have no anti-pyretic use for >17 days and be able to sign informed consent for blood draws for 4 tubes of whole blood with approximately 7.5 mL of blood per tube. Eligible patients were identified by the participating sites through advertising and direct contact. Case report forms did not contain identifying information. Samples were de-identified at the participating sites with an anonymous code assigned to each sample. Anonymized blood samples were sent to TScan laboratories with limited demographic and clinical data. Demographics included age, sex, and ethnicity. Clinical data included date of diagnosis, specifics of diagnostic testing, duration of symptoms, and whether the patient required hospitalization, supplemental oxygen, or ICU care/ ventilator support. Comorbidities and current medications were also recorded.

Convalescents who met eligibility criteria and consented to described procedures were enrolled and sampled from two sites: Atlantic Health (New Jersey, 51 samples) and Ochsner (New Orleans, 27 samples). These sites played a critical role in treating patients from early epicenters of SARS-CoV-2 outbreaks. Recruitment materials clearly requested patients that had recovered from COVID-19 with the goal of designing effective vaccines and treatments for this indication.

Average self-reported duration of symptoms was 18 days (1-80 days range) in females and 21 days (0-76 days range) in males. Hospitalizations made up ~32% of total convalescent samples received, with 31% requiring oxygen and 5% placed on a ventilator.

Blood samples were collected in four 10-mL K2 EDTA vacutainer tubes (Becton Dickinson) and

processed within 24-30 h to PBMCs or memory CD8+ T cells. A 1-mL sample was removed and centrifuged at 500xg for 10 min to obtain plasma. To isolate PBMCs, blood samples were diluted with an equal volume of MACS separation buffer (phosphate buffered saline, 0.5% bovine serum albumin, 2 mM EDTA), then layered onto lymphocyte separation media (Corning) and centrifuged at 1200xg for 20 min. The interface was removed and washed once with MACS buffer before further processing or cryopreservation. Memory CD8+ T cells were isolated from PBMCs using MACS microbead kits according to the manufacturer's instructions (Miltenyi).

Following separation, purity was confirmed using antibodies to CD3, CD8, CD45RA, CD45RO

and CD57 (Biolegend). Immediately following isolation, memory CD8+ T cells were expanded J o u r n a l P r e -p r o o f by co-culturing with 2x10 7 mitomycin C treated (50 µg/mL, 30 min) allogenic PBMCs in the presence of 0.1 µg/mL anti-CD3 (OKT3, eBioscience), 50 U/mL recombinant IL-2 (Peprotech), 5 ng/mL IL-7 and 5 ng/mL IL-15 (R&D Systems). After 10 days of expansion, the cells were collected and cryopreserved.

Coding sequences of all deposited SARS-CoV-2 strains were downloaded from NCBI on March and incubating for an additional 15 min at room temperature. The stained cells were pelleted and washed three times before resuspending in a 5 µg/mL DAPI solution and analyzed by flow cytometry (Cytoflex S, Beckman Coulter). The limit of detection was defined as the mean + 2 SD of the frequency of three MHC-mismatched controls.

Single-cell TCR-seq (scTCR-seq) libraries were prepared following the 10x Genomics Single

Cell V(D)J Reagent Kit (v1) 

All statistical analyses were performed using GraphPad prism, Excel, or Python. The details of the statistical tests are displayed in the figure legends. 

Immune Functional Assays, From Custom to Standardized Tests for Precision Medicine

SARS-CoV-2 T cell immunity: Specificity, function, durability, and role in protection

Gapped sequence alignment using artificial neural networks: application to the MHC class I system

Programmed contraction of CD8(+) T cells after infection

SARS-CoV-2-reactive T cells in healthy donors and patients with COVID-19

Virus-specific memory CD8 T cells provide substantial protection from lethal severe acute respiratory syndrome coronavirus infection

A panel of MHC class I restricted viral peptides for use as a quality control for vaccine trial ELISPOT assays

Safety and immunogenicity of the ChAdOx1 nCoV-19 vaccine against SARS-CoV-2: a preliminary report of a phase 1/2, singleblind, randomised controlled trial

Allele frequency net database (AFND) 2020 update: gold-standard data classification, open access genotype data and new query tools

Targets of T Cell Responses to SARS-CoV-2 Coronavirus in Humans with COVID-19 Disease and Unexposed Individuals

An mRNA vaccine against SARS-CoV-2-preliminary report

Variant analysis of SARS-CoV-2 genomes

T-Scan: A Genome-wide Method for the Systematic Discovery of T Cell Epitopes

SARS-CoV-2-specific T cell immunity in cases of COVID-19 and SARS, and uninfected controls

Clinical and immunological assessment of asymptomatic SARS-CoV-2 infections

High-resolution HLA alleles and haplotypes in the United States population

Selective and cross-reactive SARS-CoV-2 T cell epitopes in unexposed humans

Phase 1/2 study to describe the safety and immunogenicity of a COVID-19 RNA vaccine candidate (BNT162b1) in adults 18 to 55 years of age: interim report

Memory T cell responses targeting the SARS coronavirus persist up to 11 years post-infection

Reliable prediction of T-cell epitopes using neural networks with novel sequence representations

Long-lived memory T lymphocyte responses against SARS coronavirus nucleocapsid protein in SARS-recovered patients

Concurrent human antibody and TH1 type T-cell responses elicited by a COVID-19 RNA vaccine

Robust T Cell Immunity in Convalescent Individuals with Asymptomatic or Mild COVID-19

Longitudinal evaluation and decline of antibody responses in SARS-CoV-2 infection

recognize 3-8 shared epitopes for each HLA type studied • ~90% of shared epitopes are not located in the Spike protein • CD8 + T cells show almost no cross-reactivity with epitopes in seasonal coronaviruses eTOC Blurb Ferretti et al. reveal specific SARS-CoV-2 epitopes that are broadly shared by CD8+ T cells of COVID-19 patients but exhibit limited cross-reactivity with seasonal coronaviruses

J o u r n a l P r e -p r o o f

J o u r n a l P r e -p r o o f