SARS-CoV-2-specific CD8⁺ T cells from people with long COVID establish and maintain effector phenotype and key TCR signatures over 2 years

We longitudinally analyzed 31 people with long COVID who had been HLA-typed, for up to 24 mo post-acute SARS-CoV-2 infection (Fig. 1A and SI Appendix, Table S1↗). Individuals were designated as long COVID in accordance with the WHO definition of symptoms >3 mo post SARS-CoV-2 infection, last for at least 2 mo and cannot be explained by an alternative diagnosis (37). Participants were part of two prospective cohort, one recruited in Melbourne (Australia) in hospitals and the other in Hong Kong quarantine, both from February to October 2020. At the same time, nineteen individuals without ongoing symptoms were recruited and included as non-long COVID (non-LC) controls. Longitudinal samples were grouped across acute infection (days 0 to 23, median 9 d long COVID/7 d non-LC), ~3 mo (days 24 to 129, median 68/68 d), ~6 mo (days 130 to 269, median 209/180 d), ~12 mo (days 270 to 459, median 367/365 d), ~18 mo (days 460 to 629, median 593/591 d) and ~24 mo (days 630 to 889, median 743/767 d) (Fig. 1A). Our study included individuals across a range of acute COVID-19 severities, with 33% and 25% severe/critical cases, 47% and 33% moderate, and 20% and 42% asymptomatic/mild acute SARS-CoV-2 infections in long COVID and non-LC groups, respectively (Fig. 1B). Of individuals sampled 18 to 24 mo postinfection, all but two people with long COVID received at least one SARS-CoV-2 vaccine dose (average time of first vaccine postinfection was 432 d (range 226 to 635) for long COVID and 461 d (282 to 594) for non-LC) (Fig. 1A). One person with long COVID and one non-LC control reported known SARS-CoV-2 reinfection following vaccination prior to their final study visit (24 mo). The most common reported long COVID clinical syndrome was neurological (90%), including fatigue, headache, memory issues, lack of concentration ("mental fog"), mental confusion, dizziness, tremor, paresthesia, loss of taste or smell, problems speaking, and balance issues (Fig. 1C). Of individuals with long COVID, more than one clinical syndrome was reported by 77% (24/31) (Fig. 1C). Meanwhile 52% (16/31) were affected by ≥3 clinical syndromes, and of those, 38% recovered by their final study visit. Overall, 40% of people with long COVID self-reported being fully recovered by study endpoint (Fig. 1C). For epitope-specific T cell analyses, we determined frequencies of individuals expressing at least one HLA allele with known SARS-CoV-2 T cell epitopes within our personalized tetramer panel, with high prevalence of class I HLA-A*02:01 (32%), HLA-A*24:02 (28%), HLA-B*40:01 (30%), and class II DPB1*04:01 (40%), DRB1*15:01 (28%) observed (Fig. 1D).

To define longitudinal circulating SARS-CoV-2 epitope-specific T cells in people with long COVID, we used SARS-CoV-2 peptide–HLA tetramers to assess T cell responses directly ex vivo against 11 CD8⁺ (A1/ORF1a₁₆₃₇, A1/S₈₆₅, A2/S_269, A3/N₃₆₁, A3/S_378, A24/S₁₂₀₈, B7/N₁₀₅, B15/S₉₁₉, B27/N₉, B35/S₃₂₁, and B40/N₃₂₂) and two CD4⁺ (DPB4/S₁₆₇ and DR15/S₇₅₁) SARS-CoV-2 epitopes (Fig. 1E and SI Appendix, Fig. S1A↗) (38 –43). These epitopes encompass peptides highly conserved across variants of concern (VOC) strains, while B15/S₉₁₉ is cross-reactive with common cold coronaviruses (HKU1/OC43) (40, 41, 44).

Prototypical and robust populations of circulating SARS-CoV-2 epitope-specific CD8⁺ and CD4⁺ T cells were observed at all timepoints, starting from acute disease up to 24 mo post SARS-CoV-2 infection in people with long COVID and non-LC controls (Fig. 1F and SI Appendix, Fig. S2 A and B↗). The mean frequency of SARS-CoV-2 epitope-specific CD8⁺ T cells was stable across time. When data were stratified according to spike (A1/S₈₆₅, A2/S_269, A3/S_378, A24/S₁₂₀₈, B15/S₉₁₉, and B35/S₃₂₁) and nonspike (A1/ORF1a₁₆₃₇, A3/N₃₆₁, B7/N₁₀₅, B27/N₉, and B40/N₃₂₂) CD8⁺ T cell specificities, frequency of spike-specific CD8⁺ T cells in people with long COVID increased from 1.98 × 10⁻⁵ at 3-mo to 7.47 × 10⁻⁴ at 18 mo post SARS-CoV-2 infection (3.77 fold increase, P = 0.0126), while ORF1a- and N-specific CD8⁺ T cell responses remained unchanged (Fig. 1G). The increase in spike-specific CD8⁺ T cells at 18 mo was maintained when repeated measures per epitope (multiple samples per donor within a time grouping) were averaged (P = 0.036) (SI Appendix, Fig. S2C↗). Meanwhile, spike-specific CD4⁺ T cell responses remained stable over time in people with and without long COVID (SI Appendix, Fig. S2A↗).

To ensure the increase in spike-specific CD8⁺ T cells was the result of vaccination and not function of time postinfection, frequency of epitope-specific CD8⁺ T cells was compared between acute infection, during vaccine-naïve convalescence (3 to 24 mo postinfection) and postvaccination convalescence (12 to 24 mo). Spike-specific CD8⁺ T cell frequencies in people with long COVID showed a dip at vaccine-naïve convalescence, followed by a significant increase postvaccination (P = 0.0012) (Fig. 1H), which trended toward significance when repeated measures were averaged (SI Appendix, Fig. S2D↗). Paired analysis of final prevaccination and first postvaccination samples also demonstrated increased spike-specific CD8⁺ T cells postvaccination (P = 0.0073) (Fig. 1I, i). Conversely, frequency of ORF1a/N-specific CD8⁺ T cells (Fig. 1 I, ii and SI Appendix, Fig. S2E↗) and spike-specific CD4⁺ T cells did not change with vaccination (Fig. 1J and SI Appendix, Fig. S2F↗). Influenza tetramers (A2/M1_58, A24/PB1₄₉₈, and B35/NP₄₁₈) were included to assess bystander CD8⁺ T cell activation in people with long COVID. Frequencies of these CD8⁺ T cells were stable across two years post SARS-CoV-2 infection, with no difference between people with long COVID and non-LC controls (SI Appendix, Fig. S2G↗).

Overall, both people with long COVID and non-LC controls had prototypical robust T cell populations of SARS-CoV-2 epitope-specific CD8⁺ and CD4⁺ T cells following acute infection, convalescence, and vaccination across a two-year time period.

Individual's SARS-CoV-2-specific T cell frequencies were next assessed as total epitope-specific T cell responses within each individual over time. Based on HLA-I and HLA-II profiles, some individuals had two to four T cell specificities measured with combinations of up to two spike or nonspike CD8⁺ T cell specificities and two spike CD4⁺ T cell specificities (N = 20 long COVID, 8 non-LC), while some individuals only had one T cell specificity assessed (N = 10 long COVID, 5 non-LC). In those with multiple T cell specificities, we assessed immunodominance hierarchical patterns at acute, convalescent, and postvaccination timepoints to determine whether immunodominance hierarchies and/or magnitude changed over time or were impacted by vaccination or reinfection (Fig. 2 A and B). In people with long COVID, the nonspike CD8⁺ T cell response was typically dominant over spike-specific CD8⁺ T cells at acute and up to three months after infection (epitopes restricted to different HLA allotypes). However postvaccination, nonspike-specific CD8⁺ T cells were relatively stable in frequency, while spike-specific CD8⁺ T cell responses either increased compared to their nonspike counterparts or were further boosted, in agreeance with our earlier grouped analyses of tetramer⁺ frequencies (Fig. 1).

When we analyzed people with long COVID (N = 8) pre- and postvaccination with paired spike and nonspike CD8⁺ T cell responses, albeit restricted to different HLA allotypes, nonspike CD8⁺ T cell responses were higher than spike responses prevaccination for 4/8 individuals. However, postvaccination, spike-specific CD8⁺ T cell responses increased or remained stable in 6/8 individuals with long COVID (P = 0.0391), while their paired nonspike CD8⁺ T cell responses remained relatively stable in 5/8 individuals (Fig. 2C). However, for non-LC controls, only 1/4 individuals showed an increase in spike-specific CD8⁺ T cells, whereas the other three donors had relatively stable frequencies of spike and nonspike T cells.

CD8⁺ T cells from four people with long COVID were assessed longitudinally for HLA-A*01:01-restricted specificities recognizing either ORF1a₁₆₃₇ or S₈₆₅ (Fig. 2A and SI Appendix, Fig. S2H, i↗). Interestingly, A1/ORF1a₁₆₃₇-specific T cells were dominant over the 18-mo time period post SARS-CoV-2 infection in both vaccinated (N = 3) and nonvaccinated (N = 1) individuals. Unfortunately, there was no HLA-A*01:01 expression in non-LC individuals to assess this pattern. Similarly, across different HLA-restrictions, B40/N₃₂₂⁺CD8⁺ T cells, in combination with A2/S₂₆₉, A24/S₁₂₀₈ or B27/N₉ specificities, were generally immunodominant and often even higher in frequency than vaccine-induced spike-specific CD8⁺ T cells (N = 4/5) (SI Appendix, Fig. S2H, ii↗). However, A2/S₂₆₉⁺CD8⁺ T cell responses in the two non-LC individuals were immunodominant at convalescence 6 to 12 mo over B40/N₃₂₂, although no acute samples were available to determine response kinetics.

One individual with long COVID (donor#5) was vaccinated (BNT162b2) before 12 mo and reinfected by 24 mo post primary SARS-CoV-2 infection (Fig. 2A and SI Appendix, Fig. S2H, iii↗). At 8 d post primary infection, CD8⁺ T cells were predominantly directed at the A1/ORF1a₁₆₃₇ epitope, with an absence of B35/S₃₂₁⁺CD8⁺ and DR15/S₇₅₁⁺CD4⁺ T responses. A1/ORF1a₁₆₃₇ was still dominant and stable at 3 (day 109) and 6 mo (day 209), with the emergence of subdominant B35/S₃₂₁⁺CD8⁺ and DR15/S₇₅₁⁺CD4⁺ T cell responses. Post vaccination (day 334-dose 1, 417-dose 2), at 12 and 18 mo sampling (407 and 603 d), T cells directed at both B35/S₃₂₁ and DR15/S₇₅₁ epitopes were higher than those toward A1/ORF1a₁₆₃₇ (2.6 to 11.6-fold higher), however postinfection (day 643) by 24 mo (743 d), spike-specific CD8⁺ and CD4⁺ T cells remained at high levels, while A1/ORF1a₁₆₃₇-specific CD8⁺ T cells increased to become immunodominant (Fig. 2A).

Of 30 individuals with long COVID assessed longitudinally using our personalized tetramer panel, four individuals had very high frequencies of SARS-CoV-2-specific CD8⁺ T cells, over 1000 tetramer⁺ cells per million CD8⁺ T cells (>10⁻³ frequency), early post infection (acute and 3 mo), which was generally attributed to one dominant epitope (A1/ORF1a₁₆₃₇N = 2, B40/N₃₂₂N = 2) rather than the sum of multiple T cell specificities (Fig. 2D). One individual had very high DPB4/S₁₆₇⁺ CD4⁺ T cells early post infection. Post vaccination, five additional donors reached a similarly >10⁻³ high frequency, and the reinfected individual based on their boosted spike-specific CD8⁺ T cell responses. Conversely, only one non-LC individual (out of 13) reached the threshold frequency of >10⁻³ at acute timepoints (B7/N₁₀₅ and A3/S₃₇₈), which declined at 3 mo postinfection, but re-emerged at high levels following vaccination at 12 to 18 mo due to a vaccine-induced A3/S₃₇₈ response.

Collectively, we show that people with long COVID can establish and maintain large expansions of SARS-CoV-2-specific T cell populations, which can be further boosted by COVID-19 vaccination and reinfection.

Previous studies showed phenotypic changes in bulk T cell populations in long COVID and during peptide stimulation (26 –30). We investigated how ex vivo phenotype and activation of SARS-CoV-2 tetramer-specific T cells changed over time in our long COVID cohort, in comparison to non-LC controls, and how COVID-19 vaccination affected memory and activation phenotypes. Total CD8⁺ and CD4⁺ T cell populations of Tnaive, Tcm, Tem, Temra, and Tscm subsets did not significantly change from acute timepoint to 3-, 6-, 12-, 18-, and 24-mo timepoints for long COVID and non-LC groups (SI Appendix, Fig. S3A↗). However, there was a skewing toward terminally differentiated CD8⁺ Temra cells in long COVID people, compared to non-LC controls when all timepoints were pooled together (P = 0.0161) (Fig. 3A). For spike and nonspike CD8⁺ T cell specificities, the acute timepoint was predominantly of Tcm phenotype (>65%) before decreasing to 30 to 50% at convalescent timepoints in people with long COVID (spike P = 0.0111, nonspike P < 0.0001) (Fig. 3B). These convalescent spike and nonspike CD8⁺ T cell phenotypes in both long COVID and non-LC remained predominantly unchanged from 3 mo to 24 mo (SI Appendix, Fig. S3B↗) or following vaccination (Fig. 3B). CD4⁺ T cells directed at spike specificities (DPB4/S₁₆₇ and DR15/S₇₅₁) were both of Tem and Tcm phenotypes at acute timepoints for people with long COVID (SI Appendix, Fig. S3C↗). However, spike-specific CD4⁺ T cells were mainly of the Tcm phenotype, stable at 3 mo postinfection up to 24 mo or postvaccination (SI Appendix, Fig. S3C↗), which was the case for both long COVID and non-LC, and influenza-specific CD8⁺ T cells (SI Appendix, Fig. S3D↗). Phenotypes could not be assessed for non-LC CD4⁺ T cell specificities or influenza-specific CD8⁺ T cell specificities at acute timepoints given the low tetramer counts. In cases when epitopes were shared between long COVID and non-LC groups, phenotype distributions across different SARS-CoV-2 CD8⁺ and CD4⁺ T cell specificities were comparable between long COVID and non-LC groups (SI Appendix, Fig. S4A↗).

People with long COVID with mild acute disease can have increased expression of T cell inhibitory markers PD-1 and TIM-3 at 3 and 8 mo, with resolution by 24 mo post primary COVID-19 (30, 32). High and prolonged coexpression of CD38 and HLA-DR on T cells has been linked to fatal H7N9 influenza virus infection (45). It is also highly coexpressed on innate and adaptive T cells in severe COVID-19 patients requiring intensive care compared to milder to moderate hospital ward COVID-19 patients (46). To determine activation status of total T cells and SARS-CoV-2 tetramer-specific T cells, we assessed activation markers CD38, HLA-DR, and proliferative marker CD71, as well as TIM-3 and PD-1, linked to T cell exhaustion in chronic diseases and cancers (47). We observed high expression of all 5 markers in SARS-CoV-2-specific CD8⁺ and CD4⁺ T cells and moderate expression in total CD8⁺ and CD4⁺ T cells during acute SARS-CoV-2 infection, in both long COVID and non-LC groups (Fig. 3C and SI Appendix, Fig. S4B↗). PD-1 remained highly expressed in SARS-CoV-2-specific T cells at 3 mo, but was reduced by 6 mo (Fig. 3C), with significantly lower PD-1 expression at convalescence compared to acute disease in patients with long COVID (CD8⁺ spike P = 0.0295, CD8⁺ ORF1a/N P = 0.0348, CD4⁺ spike P = 0.0157) (Fig. 3D and SI Appendix, Fig S4C↗). Although there was a trend toward increased PD-1 and TIM-3 expression levels following vaccination, this was not significant on spike-specific CD8⁺ or CD4⁺ T cells (Fig. 3D and SI Appendix, Fig. S4C↗). Conversely, expression of CD38, HLA-DR, and CD71 was minimal at all convalescent time-points, pre- and postvaccination, in long COVID or non-LC groups. Thus, we did not observe prolonged activation of SARS-CoV-2 CD8⁺ or CD4⁺ T cells (or total T cells) in people with long COVID up to 2 years postinfection.

The frequency of bulk CD19⁺ B cells, spike-/N-specific CD19⁺IgD⁻ memory B cells was reported unaltered in long COVID stemming from mild acute disease (32), thus we assessed antigen-specific B cells using spike and N fluorescent probes in our cohort of people with long COVID and non-LC controls (SI Appendix, Fig. S1B↗). Circulating spike- and N-specific CD19⁺IgD⁻ B cells during acute infection (day 0 to 22) were low in both long COVID and non-LC individuals. Spike-specific B cells significantly increased by convalescence in people with long COVID (P = 0.0333), with a similar trend observed in N-specific B cells and non-LC controls (Fig. 3E and SI Appendix, Fig. S5A↗). Following resolution of acute disease, N-specific B cell frequencies remained stable over the 24 mo (SI Appendix, Fig. S5B↗), while spike-specific B cells increased significantly at 18 and 24 mo postinfection as a result of vaccination (SI Appendix, Fig. S5C↗). Frequencies of spike-specific B cells increased in both people with long COVID and non-LC between vaccine-naive convalescence and postvaccination (LC P = 0.0199, non-LC P = 0.0104) (Fig. 3E and SI Appendix, Fig. S5A↗). People with long COVID had increased frequency of IgG⁺ spike-specific B cells postvaccination (P = 0.0211), with a similar trend observed in non-LC, (Fig. 3 F, Left). Additionally, an increase in CD21⁻CD27⁻ B cells (P = 0.0326) and decrease in CD21⁺CD27⁺ resting B cells (P = 0.0326) was found in people with long COVID (Fig. 3 F, Right). Meanwhile, N-specific B cell isotype and phenotype were not affected by vaccination (SI Appendix, Fig. S5 D and E↗). The changes in spike-specific memory B cell isotype (SI Appendix, Fig. S5F↗) and phenotype (SI Appendix, Fig. S5G↗) were also observed when data were graphed against time, with all but two of the people with long COVID vaccinated by their final study visit. Finally, the frequency of spike-specific CD4⁺ T cells weakly correlated with the spike-specific B cells when long COVID and non-LC were combined (r = 0.2677, P = 0.0461) (Fig. 3G).

Overall, we show that SARS-CoV-2 antigen-specific T cells and B cells in people with long COVID have prototypical memory and activation phenotypes following resolution of acute infection and post-COVID-19 vaccination.

Antigen-specific T cell responses are shaped by their TCRαβ repertoires, encoded by the V(D)J gene segment usage and CDR regions. Here, we defined TCR signatures of SARS-CoV-2-specific CD8⁺ and CD4⁺ T cells in long COVID and fully recovered individuals over time. We assessed TCR sequences across six SARS-CoV-2 specificities: A2/S_269, A24/S₁₂₀₈, A3/S_378, B7/N₁₀₅, B35/S₃₂₁, and DPB4/S₁₆₇. No TCR data currently exist for A3/S₃₇₈ and B35/S₃₂₁ epitopes.

To assess TCRαβ repertoires, we single-cell sorted and sequenced 424 SARS-CoV-2-specific tetramer⁺ T cells from 16 people with long COVID and four non-LC controls at acute, convalescent, and postvaccination timepoints. Distribution of TCR sequences per epitope was spread between long COVID (38 to 80%, N = 1 to 6 per epitope) and non-LC individuals (N = 1 to 2 per epitope) with two exceptions (Fig. 4A); B35/S₃₂₁ TCR sequences all came from people with long COVID (N = 4; no non-LC in this cohort expressed HLA-B*35:01) and DPB4/S₁₆₇ TCR sequences were 91% from people with long COVID. kPCA analyses generated by TCRdist revealed that TCR clonotypes from people with long COVID were evenly interspersed with non-LC TCR clonotypes (Fig. 4B). Both long COVID and non-LC groups showed two separate clusters, mainly based on the separation of CD8⁺ and CD4⁺ T cell epitopes (Fig. 4C) rather than timing of sampling (Fig. 4D). In both groups, spike-specific CD8⁺ TCR clonotypes mainly projected upward in the kPCA plot while B7/N₁₀₅ TCR clonotypes and their clonal expansions projected downward away from the spike CD8⁺ TCR clonotypes.

The A2/S₂₆₉ TCR repertoire is widely studied with known prominent CDR3α motifs (TRAV12-2 _ TRAJ30 "CAVNXXDKIIF" and TRAV12-1 _ TRAJ43 "CVVNXXDDMRF," where "X" denotes any amino acid) and CDR3β motifs (TRBV7-9 "CASSPDIEQYF" and TRBJ2-2 "CAS-NTGELFF") (39, 40, 48). Similar A2/S₂₆₉ motifs were present in long COVID and non-LC individuals (SI Appendix, Table S2↗). TRAV12-2 _ TRAJ30 CAVNXDDKIIF was found in 3 people with long COVID from 3 mo up to 24 mo, 2 at convalescence and 1 at postvaccination timepoints. These TCRα chains were either paired with TRBV7-9 (CASSPDIVQFF, CASSSDIXAFF, CASSPDIKQYF) or TRBV20-1 _ TRBJ2-2 clonotypes (CSARDHQAQNTGELFF). The other prominent TRAV12-1 _ TRAJ43 CVVNXXXDMRF motif was also present in 3 long COVID and 1 non-LC individuals at convalescent timepoints with different TCRβ chains (SI Appendix, Table S2↗).

TCR repertoires against the A24/S₁₂₀₈ and B7/N₁₀₅ epitopes showed expanded clonotypes within long COVID and non-LC individuals (>1 clonotype/timepoint). However, most of these clonotypes were private, not shared between donors in either group, with the exception of one B7/N₁₀₅ clonotype in long COVID donor#6 (2 sequences) and non-LC #15 (1 sequence) (Fig. 4E and SI Appendix, Table S2↗). A3/S₃₇₈ TCR repertoire was less expanded but highly diverse in one individual with long COVID and one non-LC individual. B35/S₃₂₁ TCR repertoire was driven by large clonal expansions of 1 to 2 private clonotypes in individuals with long COVID at convalescent timepoints (N = 4) (SI Appendix, Table S2↗). Although clonally expanded within donors, the high diversity between donors was previously observed in A24/S₁₂₀₈ and B7/N₁₀₅ TCR repertoires from COVID-19 individuals (39, 40, 49).

DPB4/S₁₆₇ TCR repertoire included the highly prominent TRAV35_TRAJ42 CAXXNYGGSQGNLIF TCRα motif representing 73% of total DPB4/S₁₆₇ clonotypes (68/93 alpha sequences), observed in all six sequenced donors, including five people with long COVID and one non-LC (SI Appendix, Table S2↗). These TCRα chains were paired with diverse TCRβ chains and were clonally expanded within donors at acute, convalescent, and postvaccination timepoints; however, the TCRβ chains were not shared between donors. The heavily biased TRAV35_TRAJ42 CAXXNYGGSQGNLIF TCRα motif being paired with different TCRβ chains was observed in SARS-CoV-2 infection and vaccination cohorts (40 –42).

Given the overlap in SARS-CoV-2-specific TCRαβ repertoires between long COVID and non-LC groups, we compiled all individuals to generate TCRαβ, TCRα and TCRβ alluvial plots per epitope to assess whether the same clonotype or TCRα or TCRβ sequence was observed within individuals at multiple timepoints up to 24 mo (Fig. 4 E and F and SI Appendix, Fig. S6↗). Indeed, we observed a high level of sharing over time within individuals for the TCRα chain (N = 14) and the TCRβ chain (N = 16) across different epitopes, even following vaccination, which consisted of prominent TCRα/β motifs, and sequences from private and expanded clonotypes (Fig. 4F). Finally, 13 CD8⁺ and 7 CD4⁺ TCRαβ clones were identified across timepoints within individuals, providing strong evidence of clonal persistence over time in individuals with and without long COVID (Fig. 4E and SI Appendix, Table S2↗).

Thus, SARS-CoV-2-specific CD4⁺ and CD8⁺ T cells from long COVID and non-LC individuals are similar in nature. Both groups display key TCR signatures that can be detected over 2 y following disease onset.

Hospitalized COVID-19 patients were recruited via Sentinel Travelers Research Preparedness Platform for Emerging Infectious Diseases (SETREP-ID) Study at sites in Melbourne, Australia. Patients were enrolled during 2020 COVID-19 pandemic wave (January–September 2020), followed longitudinally with sample and data collection to 6 mo. A subset (N = 19) reconsented to additional follow-up to 2 y, with study visits at ~6-monthly intervals. RT-PCR-positive COVID-19 patients were also recruited from Prince of Wales Hospital during quarantine in Hong Kong between February and October 2020, longitudinally sampled at ~6-mo intervals. Long COVID specific-symptom data were collected during follow-up visits.

SETREP-ID study was approved by Melbourne Health (HREC/17/MH/53) and University of Melbourne (#1749349, #13344) human research ethics committees (HREC). The Hong Kong COVID-19 patient cohort was approved by institutional review board HKU/HA Hong Kong West Cluster (UW 20-273, UW20-169, UW 20-132), Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee (CREC 2020.229) and University of Melbourne HREC (#25684). Cohort summary demographics are described in SI Appendix, Table S1↗. The study was conducted in compliance with conditions of the ethics committee approval, NHMRC National Statement on Ethical Conduct in Human Research 2007 (updated 2018) and Note for Guidance on Good Clinical Practice (CPMP/ICH-135/95).

Informed consent was obtained from all participants. Cohort clinical details, ex vivo tetramer enrichment, B cell probe staining, and TCR analysis are described in SI Appendix↗.