Persistent cytolytic CD8+ T cells recognize SARS-CoV-2 and herpesvirus epitopes in long COVID Authors: Tongcui Ma, Heeju Ryu, Kailin Yin, Thomas Dalhuisen, Douglas Robbins, Tyrine T. Bailey, Sean A. Thomas, Emily A. Fehrman, J. Daniel Kelly, Jeffrey N. Martin, Kara L. Lynch, Sulggi A. Lee, Steven G. Deeks, Timothy J. Henrich, Michael J. Peluso, Evan W. Newell, Nadia R. Roan Published: 2026 ABSTRACT SummaryViral-specific CD8+ T cells play roles in protective immunity and immunopathology including during COVID-19. Leveraging combinatorial tetramers, we profiled CD8+ T cells specific for SARS-CoV-2, as well as those for cytomegalovirus (CMV) and Epstein-Barr virus (EBV)—herpesviruses implicated in COVID-19 pathogenesis. During severe COVID-19, EBV-specific CD8+ T cells exhibit a stem-like state, whereas CMV-specific ones are terminally differentiated and cytolytic. Comparing post-acute samples from long COVID (LC) individuals to those recovered revealed that LC-associated CMV-specific CD8+ T cells are T central memory cell (Tcm)- instead of terminally differentiated memory T cell (Temra)-biased. In LC individuals, CD8+ T cells specific for SARS-CoV-2, CMV, and EBV are preferentially terminally differentiated, exhausted, and cytolytic. Cytolytic granzyme-B-expressing CD8+ T cells are particularly prominent among LC women. As granzyme B is upregulated on viral-specific CD8+ T cells during acute viremia, an inability to shut down COVID-19-induced cytolytic effector expression may drive preferential persistence of cytolytic CD8+ T cells in people with LC, which may contribute to LC pathogenesis. Introduction Long COVID (LC) is an infection-driven chronic disease state in which symptoms attributable to COVID-19 are present for at least 3 months following acute SARS-CoV-2 infection.1 The condition can persist for years, adversely affecting quality of life and everyday functioning.2 What drives this chronic disease remains incompletely understood.3 As a result, no approved therapies exist for LC. Although LC is a chronic disease, its foundations may be established during the initial immune response to acute SARS-CoV-2 infection. LC incidence is higher among those who experienced severe COVID-19 that required hospitalization compared to those who had only mild illness.4 One potential explanation for this observation might be that high levels of SARS-CoV-2 replication during severe COVID-19 perturbs homeostasis in a manner that hinders a full recovery even after active viral replication has long subsided. SARS-CoV-2 may not be the only virus whose impact during acute COVID-19 drives subsequent development of LC—reactivation of herpesviruses such as Epstein-Barr virus (EBV) and cytomegalovirus (CMV) has also been suggested to play a role. The association of EBV DNA levels during acute COVID-19 with subsequent development of LC,5 along with higher EBV antibody titers during LC,6,7,8 together suggest reactivation of EBV and its downstream effects as potential LC drivers. CMV antibody titers have also been observed to be elevated during LC,8 although in some studies they have paradoxically been associated with protective effects.7 EBV and CMV are classic examples of viruses that persist for the lifetime of infected individuals. SARS-CoV-2 has not classically been considered a persistent virus, yet has been documented to persist for years in people with LC,9 although whether this persistence drives LC symptoms is less clear.10 Regardless, it is conceivable that the inability to control SARS-CoV-2, EBV, and/or CMV viremia during the acute COVID-19 can cause immune perturbations that fail to resolve, which may contribute to LC pathogenesis. During acute viremia, CD8+ T cells play a key role in controlling viral replication. Upon encountering cognate peptide-major histocompatibility complex (MHC) on virally infected cells, CD8+ T cells can exert a multitude of effector functions ranging from cytolytic destruction to cytokine induction. But COVID-19 may also induce disruptions in normal functions of viral-specific CD8+ T cells persisting into the post-acute phase. Indeed, CD8+ T cells are perturbed during LC. For example, CD8+ T cells expressing the cytolytic effectors granzyme A, granzyme B, and/or perforin are more frequent in LC individuals than healthy controls.11,12 Moreover, cytolytic CD8+ T cells appear to be more prevalent among women with LC,13 which is of interest given the higher prevalence of LC in women as compared to men.14 These observations make CD8+ T cells potential mediators of LC. Less is known, however, about the extent to which CD8+ T cells specific for defined viral antigens are perturbed during LC. Studies examining viral-specific CD8+ T cells during LC have largely relied on ex vivo stimulation with viral peptides to identify and characterize these cells. For instance, stimulation of peripheral blood mononuclear cells (PBMCs) with SARS-CoV-2 or CMV peptides followed by sorting of the activated CD8+ T cell population revealed cells with cytolytic signatures to be associated with gastrointestinal LC.5 Peptide stimulation assays also revealed SARS-CoV-2-specific CD8+ T cells with phenotypic features of T cell exhaustion to be more prevalent in LC as compared to recovered individuals.13 One limitation of ex vivo peptide stimulation, however, is that the phenotypes of cells treated in this manner do not completely reflect the cells’ original phenotypes. By contrast, technology based on MHC class I tetramer/multimer binding (whereby the tetramer/multimer comprised of four or more copies of a specific peptide/MHC class I complex is used as a detection reagent) allows for characterization of viral-specific CD8+ T cells in their original state. Only one study to date has used tetramer/multimer staining in the context of LC, and phenotypic comparisons of SARS-CoV-2-specific T cells between those with vs. without LC only revealed expression of CD45RA to be overrepresented in those with LC.15 The limited ability to find phenotypic differences between groups may have been due to restrictions in the numbers of phenotypic markers examined due to limitations imposed by conventional FACS. By contrast, CyTOF-based combinatorial peptide-MHC tetramers provide a strategy that allows for simultaneous deep phenotyping of a large number of antigen-specific CD8+ T cells.16 Here, we designed a combinatorial CyTOF tetramer panel that can profile up to 56 different viral epitope-specific CD8+ T cells for expression levels of 29 phenotypic markers of relevance for CD8+ T cell biology and LC pathogenesis. This panel was applied on specimens collected from donors during active severe COVID-19 as well as from donors 4 months following COVID-19, half of which had recovered and the other half of which met criteria for LC. The post-acute specimens were obtained from individuals enrolled in the UCSF Long Term Impact of Infection with Novel Coronavirus (LIINC) cohort17 and were a subset of those recently analyzed by a variety of “omics” assays,13 enabling cross-study comparisons. Overall, we found LC-associated perturbations among both total and viral-specific CD8+ T cells and discovered shared perturbations among CD8+ T cells specific for SARS-CoV-2, EBV, and CMV. Results Viral-specific CD8+ T cells display different phenotypic patterns during active and post-acute COVID-19 To capture the widest-possible range of viral-specific CD8+ T cells by CyTOF, we leveraged combinatorial tetramer technology16 whereby eight metal channels were dedicated to the detection of MHC peptide tetramers, and each tetramer was indexed by a unique combination of three out of these eight metals (Figure S1A). Combinatorial coding allowed us to cover 56 immunodominant epitopes from common human viruses—including SARS-CoV-2, CMV, EBV, and influenza—that are presented by common human leukocyte antigens (HLAs) (Table S1). The SARS-CoV-2 epitopes were selected based on their established immunodominance and high prevalence across multiple studies of SARS-CoV-2-specific T cell responses,18,19,20,21 while the other chosen viral epitopes are well-described ones previously recognized to be immunodominant and detailed in our prior studies.16,22,23 Our panel also included 29 phenotyping markers (Table S2), enabling an in-depth assessment of the differentiation status, activation states, and effector function profiles of the viral-specific CD8+ T cells. We first applied the panel to determine the extent to which acute SARS-CoV-2 viremia influences the phenotypic distribution of viral-specific CD8+ T cells. To do this, we compared specimens collected from individuals who were hospitalized with severe COVID-19 (n = 12, 0–15 days since symptom onset) (Table S3) to those collected from individuals 4 months following SARS-CoV-2 infection (n = 20) obtained from the LIINC cohort (Table S4). A sequential gating strategy was first implemented to identify CD8+ T cells (Figure S1B). We then used gating strategies to gate on the three-metal code corresponding to each peptide-MHC class I (Figure S1C). UMAP visualization showed that CD8+ T cells specific for the different viruses occupied different areas of this projection in both patient groups (Figure 1A), indicating phenotypic differences between these cells.Figure 1Distribution of viral-specific CD8+ T cells from active and post-acute COVID-19(A) UMAP of total and viral-specific CD8+ T cells from active (n = 12) and post-acute (n = 20) COVID-19.(B) Frequencies of classic CD8+ T cell subsets among viral-specific CD8+ T cells in active and post-acute COVID-19 participants.During both active and post-acute COVID-19, EBV-specific CD8+ T cells are biased toward Tcm.(C) Frequencies of CD69+, CD103+, and CD69+CD103+ Trm cells among viral-specific CD8+ T cells in active and post-acute COVID-19 participants.(D) Frequencies of HLADR+, CD38+, and HLADR+CD38+ activated T cells among viral-specific CD8+ T cells in active and post-acute COVID-19 participants.(E) Frequencies of cells co-expressing homeostatic proliferation markers CD127 and IL-15R among viral-specific CD8+ T cells.(F and G) Expression levels of CD57 (F) or granzyme B (G) reported as median expression intensity (MSI) or as frequencies of positive cells among viral-specific CD8+ T cells.All boxplots show the interquartile range (IQR), with median values indicated by horizontal lines and dots as biological replicates.For each subset, overall group differences were assessed with the two-sided Kruskal-Wallis test. When significant (p < 0.05), Dunn’s pairwise comparisons were performed, and p values were adjusted using the Benjamini-Hochberg method.∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. See also Figure S2. Distribution of viral-specific CD8+ T cells from active and post-acute COVID-19 (A) UMAP of total and viral-specific CD8+ T cells from active (n = 12) and post-acute (n = 20) COVID-19. (B) Frequencies of classic CD8+ T cell subsets among viral-specific CD8+ T cells in active and post-acute COVID-19 participants. During both active and post-acute COVID-19, EBV-specific CD8+ T cells are biased toward Tcm. (C) Frequencies of CD69+, CD103+, and CD69+CD103+ Trm cells among viral-specific CD8+ T cells in active and post-acute COVID-19 participants. (D) Frequencies of HLADR+, CD38+, and HLADR+CD38+ activated T cells among viral-specific CD8+ T cells in active and post-acute COVID-19 participants. (E) Frequencies of cells co-expressing homeostatic proliferation markers CD127 and IL-15R among viral-specific CD8+ T cells. (F and G) Expression levels of CD57 (F) or granzyme B (G) reported as median expression intensity (MSI) or as frequencies of positive cells among viral-specific CD8+ T cells. All boxplots show the interquartile range (IQR), with median values indicated by horizontal lines and dots as biological replicates. For each subset, overall group differences were assessed with the two-sided Kruskal-Wallis test. When significant (p < 0.05), Dunn’s pairwise comparisons were performed, and p values were adjusted using the Benjamini-Hochberg method. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. See also Figure S2. To further define these differences, we gated for the classic CD8+ T cell subsets including T naive (Tn)/T stem cell memory cells (Tscm), T central memory cells (Tcm), T effector memory cells (Tem), T transitional memory cells (Ttm), terminally differentiated memory T cells (Temra), and T cells expressing the recirculating T resident memory (Trm) marker CD10324,25 and/or the early activation/Trm marker CD69, as well as those expressing the classic chronic activation markers HLADR and/or CD38 (Figure S2). Most significant differences between the viral-specific CD8+ T cells were observed in the post-acute specimens only (Figures 1B and 1C). For instance, SARS-CoV-2-specific CD8+ T cells had the highest proportion of Tn/Tscm cells and influenza-specific CD8+ T cells the highest proportion of CD103+ and CD69+CD103+ Trm cells and lowest proportion of Tn/Tscm in these specimens (Figures 1B and 1C). However, EBV-specific CD8+ T cells were biased toward the Tcm subset during both active and post-acute COVID-19 (Figures 1B and 1C). No significant differences were observed among activated CD8+ T cells as defined by HLADR and/or CD38 expression, although there was high within-group variability (Figure 1D). Tcm is a subset known for being long-lived and having high proliferative capacity. Consistent with the abundance of long-lived EBV-specific CD8+ T cells during active COVID-19 was our observation that EBV-specific CD8+ T cells preferentially co-expressed CD127 (the alpha chain of the interleukin-7 [IL-7] receptor) and IL-15RA (the alpha chain of the IL-15 receptor), two receptors key for driving homeostatic proliferation of CD8+ T cells (Figure 1E). In comparison, CMV-specific CD8+ T cells expressed higher levels of the terminal differentiation marker CD57 than did other viral-specific CD8+ T cells during both the early and post-acute phases, although this was statistically significant only in the post-acute group (Figure 1F). Influenza-specific CD8+ T cells expressed the lowest levels of CD57 (Figure 1F). CMV-specific CD8+ T cells also expressed the highest levels of granzyme B, a cytolytic effector molecule associated with terminally differentiated cells (Figure 1G). Together, these results suggest that even though EBV and CMV are both chronic herpesviruses, CD8+ T cells recognizing them harbor different phenotypic features during both active and post-acute COVID-19; specifically, those specific for EBV harbor a more stem-like state while those specific for CMV harbor a more terminally differentiated and cytolytic state. Viral-specific CD8+ T cell subset distributions differ in people with long COVID vs. those who recovered Among the 20 post-acute specimens, half were from individuals who fully recovered from SARS-CoV-2 infection (R, recovered), while the other half were from individuals diagnosed with LC (Table S4). There were no significant differences between the R and LC groups with regard to age, sex, race, BMI, and co-morbidities (Table S5). Of note, these LC and R specimens were earlier time point specimens from a subset of individuals we previously studied at 8 months post-infection,13 none of whom had been vaccinated or re-infected in the 8 months following their infection (Figure 2A). We note that while SARS-CoV-2-specific CD8+ T cells were characterized in the prior study, their phenotypes corresponded to those of ex vivo stimulated cells given that an intracellular cytokine staining assay was used to identify those cells. By contrast, use of combinatorial tetramers in the current study enabled analysis of SARS-CoV-2-specific T cells in their original unstimulated state and further allowed us to simultaneously examine the original phenotypes of CD8+ T cells specific for EBV, CMV, and influenza.Figure 2Overview and subset distribution of total and viral-specific CD8+ T cells from recovered and long COVID participants(A) Study cohort and sample selection. PBMCs were collected 4 months post-symptom onset from recovered (R, n = 10) and long COVID (LC, n = 10) participants for combinatorial tetramer analyses.The 4-month specimens are a subset of participants analyzed at the 8th month in Yin et al.13(B) UMAP of concatenated total CD8+ T cells from R and LC groups.(C) UMAP of viral-specific CD8+ T cells concatenated from all R or LC participants. The number of donors with contributing cells (N) is indicated for each of the R and LC groups.(D) Distribution of total and viral-specific CD8+ T cells among classical subsets in LC compared to R participants. LC participants harbored lower frequencies of Tn/Tscm total CD8+ T cells and CMV-specific CD8+ Temra cells and higher frequencies of CMV-specific CD8+ Tcm cells.All boxplots show the interquartile range (IQR), with horizontal lines indicating medians and dots as biological replicates.p values were calculated by Welch’s t test. ∗p < 0.05, ∗∗p < 0.01. See also Figures S4 and S5. Overview and subset distribution of total and viral-specific CD8+ T cells from recovered and long COVID participants (A) Study cohort and sample selection. PBMCs were collected 4 months post-symptom onset from recovered (R, n = 10) and long COVID (LC, n = 10) participants for combinatorial tetramer analyses. The 4-month specimens are a subset of participants analyzed at the 8th month in Yin et al.13 (B) UMAP of concatenated total CD8+ T cells from R and LC groups. (C) UMAP of viral-specific CD8+ T cells concatenated from all R or LC participants. The number of donors with contributing cells (N) is indicated for each of the R and LC groups. (D) Distribution of total and viral-specific CD8+ T cells among classical subsets in LC compared to R participants. LC participants harbored lower frequencies of Tn/Tscm total CD8+ T cells and CMV-specific CD8+ Temra cells and higher frequencies of CMV-specific CD8+ Tcm cells. All boxplots show the interquartile range (IQR), with horizontal lines indicating medians and dots as biological replicates. p values were calculated by Welch’s t test. ∗p < 0.05, ∗∗p < 0.01. See also Figures S4 and S5. We first asked whether we could detect differences in the CD8+ T cell profiles of the R and LC individuals. UMAP visualization of total and viral-specific CD8+ T cells revealed qualitative phenotypic differences between the LC and R groups (Figures 2B and 2C). When analyzing classic immune subset distribution, we found the most significant differences among CMV-specific CD8+ T cells: those from participants with LC were preferentially of the Tcm phenotype and dis-enriched for Temra cells, compared to those from R participants (Figure 2D). Among both SARS-CoV-2-specific and total CD8+ T cells, there was general enrichment for Tem cells and a dis-enrichment for Tn/Tscm cells in LC participants (Figure 2D). The enrichment for Tem total CD8+ T cells in those with LC within the same cohort was also observed at the 8-month time point,13 suggesting longitudinal stability of the CD8+ T cell phenotypes. Indeed, the 4- and 8-month frequencies of classic CD8+ T cell subsets (Tcm, Ttm, Tem, Temra, and Tn/Tscm) were all significantly associated with one another (Figure S3). With regard to other subsets, there was a trend for enrichment in Trm marker expressing cells among SARS-CoV-2-specific and EBV-specific CD8+ T cells (Figure S4). With regard to activation markers, there were no significant differences in CD38 expression between the R and LC groups, although overall participants with LC exhibited higher CD38 expression among total and viral-specific CD8+ T cells, except for those recognizing influenza (Figure S4). Surprisingly, HLADR expression exhibited the opposite pattern of being more highly expressed among the R group, with the difference among total CD8+ T cells being highly statistically significant (p < 0.0001) (Figure S4). Overall, there were no significant differences in classic subset distribution between the R and LC groups among influenza-specific CD8+ T cells (Figure 2D). When we compared relative subset distribution of the different CD8+ T specificities, we also found differences between the LC and R groups, e.g., in individuals with LC, EBV-specific CD8+ T cells were significantly enriched for the Tcm and CD69+ Trm phenotypes and dis-enriched for the Temra phenotype, as compared to SARS-CoV-2-specific CD8+ T cells; this was not observed in the R group (Figure S5). Together, these results suggest that CD8+ T cells recognizing not only SARS-CoV-2 but also chronic herpesviruses exhibit phenotypic differences in individuals with LC. A cluster of stem-like CD8+ T cells is preferentially lost in individuals with long COVID To leverage all the phenotypic markers of our CyTOF panel, we performed unsupervised clustering using PhenoGraph, which revealed 24 clusters of CD8+ T cells (Figures S6 and S7). Among total CD8+ T cells, cluster 9 was significantly enriched in R individuals compared to those with LC (Figure 3A). Furthermore, cluster 9 ranked as the cluster most enriched among SARS-CoV-2-specific CD8+ T cells from R relative to LC samples (Figure 3B). Cluster 9 cells were CD45RA+CD45RO−CCR7+ (Figure 3C), a profile that is characteristic of both Tn and Tscm. Cluster 9 cells also expressed high levels of stemness-associated marker CD127 and low levels of the terminal differentiation marker CD57 and cytolytic effector granzyme B (Figure 3D). CD29 has been reported to serve as a useful marker of cytolytic CD8+ T cells,26 and it too, along with the cytolytic marker CD56, was low among cluster 9 cells (Figure 3D). In line with cluster 9 exhibiting stem-like features is their low expression of T cell exhaustion markers (PD1, TIGIT, and CD39) (Figure 3E). Cluster 9 also had distinct activation (high CD38 but low HLADR) and homing (high CXCR4 but low CCR5) profiles (Figures 3F and 3G).Figure 3Cluster 9, enriched in recovered participants, exhibits a non-differentiated, non-exhausted phenotype(A) Cluster 9 harbors a significantly higher proportion of total CD8+ T cells in R than LC participants. ∗p < 0.05 as determined by Welch’s t test.(B) Ratios of cluster frequencies among SARS-CoV-2-specific CD8+ T cells from R relative to LC participants. Cluster 9 is the most enriched cluster.(C) 2D plots of CyTOF gating for classical subsets, demonstrating that cluster 9 cells are primarily Tn/Tscm cells. Shown are data from a representative donor.(D–G) Phenotypic analysis of cluster 9 and total CD8+ T cells from all participants, as measured by MSI of markers of differentiation/effector functions (D), exhaustion (E), activation (F), and homing (G). All boxplots show IQR with medians indicated by horizontal lines and dots as biological replicates.p values were calculated by Wilcoxon signed-rank test and adjusted for multiple comparisons using the Benjamini-Hochberg method; ∗p < 0.05. See also Figure S8. Cluster 9, enriched in recovered participants, exhibits a non-differentiated, non-exhausted phenotype (A) Cluster 9 harbors a significantly higher proportion of total CD8+ T cells in R than LC participants. ∗p < 0.05 as determined by Welch’s t test. (B) Ratios of cluster frequencies among SARS-CoV-2-specific CD8+ T cells from R relative to LC participants. Cluster 9 is the most enriched cluster. (C) 2D plots of CyTOF gating for classical subsets, demonstrating that cluster 9 cells are primarily Tn/Tscm cells. Shown are data from a representative donor. (D–G) Phenotypic analysis of cluster 9 and total CD8+ T cells from all participants, as measured by MSI of markers of differentiation/effector functions (D), exhaustion (E), activation (F), and homing (G). All boxplots show IQR with medians indicated by horizontal lines and dots as biological replicates. p values were calculated by Wilcoxon signed-rank test and adjusted for multiple comparisons using the Benjamini-Hochberg method; ∗p < 0.05. See also Figure S8. Interestingly, cluster 11 was also enriched among the R as compared to LC individuals albeit less significantly (Figure S8A), and clusters 11 and 9 exhibited almost identical phenotypic features including low expression of terminal differentiation and exhaustion markers (Figures S8B–S8F). The one antigen distinguishing cluster 11 from 9 was CD69, which was highly expressed in the former but absent in the latter (Figure S8G). Collectively, these results demonstrate that LC participants are preferentially deficient in stem-like, non-exhausted CD8+ T cells. SARS-CoV-2-specific CD8+ T cells from individuals with LC preferentially reside in cellular clusters defined by terminal differentiation, exhaustion, and cytolytic markers In contrast to cluster 9, which was the most enriched cluster of SARS-CoV-2-specific CD8+ T cells among R individuals (Figure 3B), cluster 13 was the most enriched cluster of SARS-CoV-2-specific CD8+ T cells among LC individuals (Figure 4A). Two additional clusters—clusters 2 and 19—were only detected in LC and not in R participants (Figures 4B and 4C). Although clusters 2 and 13 were primarily Temra cells while Cluster 19 was primarily Tem cells (Figure 4D), all three clusters expressed high levels of terminal differentiation marker CD57 and low levels of stemness-associated marker CD127 and the co-stimulatory molecule CD28 important for T cell proliferation (Figure 4E). Among exhaustion markers, PD1 was highly expressed on cluster 19 cells but not cluster 2 and 13 cells, while TIGIT was highly expressed on cluster 2 and 13 cells and not cluster 19 cells (Figure 4F). Expression of activation marker CD38, but not HLADR, was elevated in all three clusters (Figure 4G), and expression of chemokine receptor CXCR4 was reduced in all three clusters (Figure 4H). Both cytolytic marker CD29 and the cytolytic effector granzyme B were significantly elevated in all three clusters (Figure 4I). Interestingly, when we assessed for associations between the frequencies of granzyme-B-expressing SARS-CoV-2-specific CD8+ T cells with inflammatory-associated genes as assessed from RNA sequencing (RNA-seq) analysis from the same donors,13 we found significant positive associations between the percentages of granzyme B+ SARS-CoV-2-specific CD8+ T cells and the expression levels of the alarmin genes S100A8 and S100A11 in R individuals vs. a negative association in those with LC (Figure 4J). This observation may reflect a unique uncoupling of inflammatory pathways in individuals with LC. Overall, these observations suggest that cellular clusters enriched among SARS-CoV-2-specific CD8+ T cells from individuals with LC exhibit shared features of terminal differentiation, exhaustion, and enhanced cytolytic activity.Figure 4Clusters 2, 13, and 19, preferentially enriched among SARS-CoV-2-specific CD8+ T cells from LC participants, exhibit a terminally differentiated, exhausted, and cytolytic phenotype(A) Ratios of cluster frequencies among SARS-CoV-2-specific CD8+ T cells from LC relative to R. Cluster 13 is the most enriched cluster.(B) SARS-CoV-2-specific CD8+ T cells residing within clusters 2 and 19 were only detectable in LC. Cluster 2 was detected in three of six donors, and cluster 19 in two of six donors. p values were calculated by Mann-Whitney.(C) UMAPs depicting cluster 2 and 19 cells (blue) among total CD8+ T cells (gray) in R and LC.(D) 2D plots of CyTOF gating for classical subsets, demonstrating that cluster 2 and 13 cells are primarily CD45RO−CD45RA+CCR7−, phenotypic features of Temra cells, while cluster 19 cells are primarily CD45RO+CD45RA−CCR7−, phenotypic features of Tem cells.(E–I) Phenotypes of clusters 2, 13, and 19 compared to total CD8+ T cells, as measured by MSI of markers of differentiation/stemness (E), exhaustion (F), activation (G), homing (H), and cytolytic activity (I).All boxplots display IQR with median lines, and p values were calculated by Wilcoxon signed-rank test and multiple comparisons between clusters adjusted using the Holm method, with further adjustment for multiple phenotyping marker comparisons performed using the Benjamini-Hochberg procedure. ∗p < 0.05.(J) Association between proportion of granzyme B+ SARS-CoV-2-specific CD8+ T cells and RNA levels of alarmins S100A11 and S100A8 in total PBMCs. The association is positive in R and negative in LC specimens. RNA-seq signatures were generated from Yin et al.13Associations were calculated using Spearman’s rank correlation (ρ), and p values computed using an exact permutation test and adjusted for multiple comparisons using the Benjamini-Hochberg method. ∗p < 0.05. Clusters 2, 13, and 19, preferentially enriched among SARS-CoV-2-specific CD8+ T cells from LC participants, exhibit a terminally differentiated, exhausted, and cytolytic phenotype (A) Ratios of cluster frequencies among SARS-CoV-2-specific CD8+ T cells from LC relative to R. Cluster 13 is the most enriched cluster. (B) SARS-CoV-2-specific CD8+ T cells residing within clusters 2 and 19 were only detectable in LC. Cluster 2 was detected in three of six donors, and cluster 19 in two of six donors. p values were calculated by Mann-Whitney. (C) UMAPs depicting cluster 2 and 19 cells (blue) among total CD8+ T cells (gray) in R and LC. (D) 2D plots of CyTOF gating for classical subsets, demonstrating that cluster 2 and 13 cells are primarily CD45RO−CD45RA+CCR7−, phenotypic features of Temra cells, while cluster 19 cells are primarily CD45RO+CD45RA−CCR7−, phenotypic features of Tem cells. (E–I) Phenotypes of clusters 2, 13, and 19 compared to total CD8+ T cells, as measured by MSI of markers of differentiation/stemness (E), exhaustion (F), activation (G), homing (H), and cytolytic activity (I). All boxplots display IQR with median lines, and p values were calculated by Wilcoxon signed-rank test and multiple comparisons between clusters adjusted using the Holm method, with further adjustment for multiple phenotyping marker comparisons performed using the Benjamini-Hochberg procedure. ∗p < 0.05. (J) Association between proportion of granzyme B+ SARS-CoV-2-specific CD8+ T cells and RNA levels of alarmins S100A11 and S100A8 in total PBMCs. The association is positive in R and negative in LC specimens. RNA-seq signatures were generated from Yin et al.13 Associations were calculated using Spearman’s rank correlation (ρ), and p values computed using an exact permutation test and adjusted for multiple comparisons using the Benjamini-Hochberg method. ∗p < 0.05. CMV- and EBV-specific CD8+ T cells from individuals with LC preferentially reside in cellular clusters defined by terminal differentiation, exhaustion, and cytolytic markers The exclusivity of cluster 19 to LC samples was observed not only among SARS-CoV-2-specific CD8+ T cells (Figure 4B) but also among CMV-specific CD8+ T cells, albeit only in a single donor in whom >60% of CMV-specific CD8+ T cells belonged to this cluster. This individual also had high numbers of cluster 20 cells, which was not detected in any of the R donors (Figure 5A). Although clusters 19 and 20 belonged to different subsets—cluster 19 was predominantly Tem, whereas cluster 20 was a mix of Tem and Temra (Figure 5B)—both harbored terminally differentiated cells (CD57highCD127lowCD28low) and expressed exhaustion markers PD1 and CD39 but not TIGIT (Figures 5C and 5D). They also both expressed high levels of activation marker CD38, low levels of CXCR4, and high levels of cytotoxic markers CD29 and granzyme B (Figures 5E–5G). Cluster 12 was also enriched among CMV-specific CD8+ T cells from LC individuals, but unlike clusters 19 and 20, it was also detectable among R individuals (Figure S9A). Cluster 12 cells were primarily Tcm and Ttm cells (Figure S9B), and like clusters 19 and 20, they were also terminally differentiated (high CD57), exhausted (high PD1, TIGIT, and CD38), activated (high CD38), and cytolytic (high CD29) (Figures S9C–S9F).Figure 5Clusters 19 and 20, preferentially enriched among CMV-specific CD8+ T cells from LC participants, exhibit a terminally differentiated, exhausted, and cytolytic phenotype(A) UMAP of clusters 19 (blue) and 20 (orange) cells among total CD8+ T cells (gray) from R and LC groups. Clusters 19 and 20 cells are only detectable among CMV-specific CD8+ T cells from LC.(B) 2D plots of CyTOF gating for classical subsets, demonstrating cluster 19 cells are predominantly Tem cells while Cluster 20 cells are a combination of Tem and Temra cells.(C–G) Phenotypes of clusters 19 and 20 compared to total CD8+ T cells, as measured by MSI of markers of differentiation/stemness (C), exhaustion (D), activation (E), homing (F), and cytolytic activity (G).For (C)–(G), boxplots display IQR with median lines, and p values were calculated by Wilcoxon signed-rank test and multiple comparisons between clusters adjusted using the Holm method, with further adjustment for multiple phenotyping marker comparisons performed using the Benjamini-Hochberg procedure. ∗p < 0.05. See also Figure S9. Clusters 19 and 20, preferentially enriched among CMV-specific CD8+ T cells from LC participants, exhibit a terminally differentiated, exhausted, and cytolytic phenotype (A) UMAP of clusters 19 (blue) and 20 (orange) cells among total CD8+ T cells (gray) from R and LC groups. Clusters 19 and 20 cells are only detectable among CMV-specific CD8+ T cells from LC. (B) 2D plots of CyTOF gating for classical subsets, demonstrating cluster 19 cells are predominantly Tem cells while Cluster 20 cells are a combination of Tem and Temra cells. (C–G) Phenotypes of clusters 19 and 20 compared to total CD8+ T cells, as measured by MSI of markers of differentiation/stemness (C), exhaustion (D), activation (E), homing (F), and cytolytic activity (G). For (C)–(G), boxplots display IQR with median lines, and p values were calculated by Wilcoxon signed-rank test and multiple comparisons between clusters adjusted using the Holm method, with further adjustment for multiple phenotyping marker comparisons performed using the Benjamini-Hochberg procedure. ∗p < 0.05. See also Figure S9. Among EBV-specific CD8+ T cells, cluster 5 was the most enriched cluster in LC relative to R samples (Figure 6A); this cluster was not enriched among CMV- or influenza-specific CD8+ T cells (Figure 6B), although it was the second-most enriched cluster among SARS-CoV-2-specific CD8+ T cells (Figure 4A). Unlike for SARS-CoV-2 and CMV, no clusters were only detectable among EBV-specific CD8+ T cells in LC and not R individuals. Cluster 5 was a mix of Tem and Temra cells (Figure 6C). Like the clusters enriched among SARS-CoV-2-specific and CMV-specific CD8+ T cells in individuals with LC (clusters 2, 12, 13, 19, and 20), cluster 5 comprised terminally differentiated cells (CD57highCD127lowCD28low) (Figure 6D), lowly expressed CXCR4 (Figure 6E), and exhibited elevated expression of CD38, CD29, and granzyme B (Figures 6F and 6G). It also expressed higher levels of exhaustion markers than total CD8+ T cells, although this did not meet statistical significance (Figure 6H). When we assessed for associations between the frequencies of exhausted EBV-specific CD8+ T cells with genes involved in immunoregulation as characterized by RNA-seq,13 we found that the frequencies of PD1+TIGIT+ EBV-specific CD8+ T cells significantly positively associated with levels of SIGLEC7, SIGLEC11, and SIGLEC15 transcripts—which encode lectin receptors associated with immunosuppression—while the opposite trend was observed in R individuals (Figure 6I). By contrast, the pro-inflammatory lectin SIGLEC14 showed an opposite trend (negative association in R individuals and positive in LC individuals) (Figure 6I). Hence, LC individuals may uniquely exhibit a dysregulation in immune checkpoint-SIGLEC interactions. These distinct associations of PD1+TIGIT+ EBV-specific CD8+ T cell frequencies among LC individuals prompted us to examine the extent to which EBV-specific CD8+ T cells expressing any combination of our exhaustion markers (PD1, TIGIT, and CD39) were upregulated in those with LC, perhaps reflecting increased EBV reactivation. We found this was not the case, in that EBV-specific CD8+ cells expressing PD1, TIGIT, and/or CD39 did not differ in LC as compared to R individuals (Figure S10A). PD1, TIGIT, and/or CD39 expression was also not elevated among CD8+ T cells specific for CMV, SARS-CoV-2, and influenza (Figures S10B–S10D). For CMV-specific CD8+ T cells, surprisingly, exhausted CMV-specific CD8+ T cells tended to be lower in those with LC, with the TIGIT+ and PD1+TIGIT+ populations being statistically significant (Figure S10C).Figure 6Cluster 5, preferentially enriched among EBV-specific CD8+ T cells from LC participants, exhibits a terminally differentiated, exhausted, and cytolytic phenotype(A) Ratios of cluster frequencies among EBV-specific CD8+ T cells from LC relative to R, demonstrating cluster 5 to be the most enriched cluster.(B) Cluster 5 is not enriched among CMV- and influenza-specific CD8+ T cells among LC.(C) 2D plots of CyTOF gating for classical subsets, demonstrating cluster 5 cells are predominantly combination of Tem and Temra cells.(D–H) Phenotypes of cluster 5 compared to total CD8+ T cells, as measured by MSI of markers of differentiation/stemness (D), homing (E), activation (F), cytolytic activity (G), and exhaustion (H).For (D)–(H), boxplots display IQR with median lines and dots as individual participants. p values were calculated by Wilcoxon signed-rank test and adjusted for multiple comparisons using the Benjamini-Hochberg method. ∗p < 0.05.(I) Association between proportion of PD1+TIGIT+ EBV-specific CD8+ T cells and immunosuppressive (SIGLEC7, SIGLEC11, and SIGLEC15) and inflammatory (SIGLEC14) SIGLECs as measured from RNA-seq analysis of bulk PBMCs.13Associations were calculated using Spearman’s rank correlation (ρ), and p values were computed using an exact permutation test and adjusted for multiple comparisons using the Benjamini-Hochberg method. ∗p < 0.05. See also Figure S10. Cluster 5, preferentially enriched among EBV-specific CD8+ T cells from LC participants, exhibits a terminally differentiated, exhausted, and cytolytic phenotype (A) Ratios of cluster frequencies among EBV-specific CD8+ T cells from LC relative to R, demonstrating cluster 5 to be the most enriched cluster. (B) Cluster 5 is not enriched among CMV- and influenza-specific CD8+ T cells among LC. (C) 2D plots of CyTOF gating for classical subsets, demonstrating cluster 5 cells are predominantly combination of Tem and Temra cells. (D–H) Phenotypes of cluster 5 compared to total CD8+ T cells, as measured by MSI of markers of differentiation/stemness (D), homing (E), activation (F), cytolytic activity (G), and exhaustion (H). For (D)–(H), boxplots display IQR with median lines and dots as individual participants. p values were calculated by Wilcoxon signed-rank test and adjusted for multiple comparisons using the Benjamini-Hochberg method. ∗p < 0.05. (I) Association between proportion of PD1+TIGIT+ EBV-specific CD8+ T cells and immunosuppressive (SIGLEC7, SIGLEC11, and SIGLEC15) and inflammatory (SIGLEC14) SIGLECs as measured from RNA-seq analysis of bulk PBMCs.13 Associations were calculated using Spearman’s rank correlation (ρ), and p values were computed using an exact permutation test and adjusted for multiple comparisons using the Benjamini-Hochberg method. ∗p < 0.05. See also Figure S10. In conclusion, just like clusters enriched among SARS-CoV-2-specific CD8+ T cells, clusters enriched among herpesvirus-specific CD8+ T cells tend to be Tem/Temra and exhibit terminal differentiation, exhaustion, and cytolytic phenotypes. Cytotoxic CD8+ T cells are particularly prominent in women with LC and mirror the active phase of severe COVID-19 The observation that clusters enriched among CD8+ T cells specific for SARS-CoV-2, CMV, and EBV antigens in LC relative to R samples (clusters 2, 5, 12, 13, 19, and 20) exhibited higher expression of terminal differentiation marker CD57, as well as cytolytic markers CD29 and granzyme B, suggests a potential role for viral-specific, terminally differentiated cytolytic CD8+ T cells in LC pathogenesis. We hence assessed the extent to which CD57 and the two cytolytic markers are differentially expressed among CD8+ T cells from people with LC vs. those who recovered, irrespective of clustering. The frequencies of total CD8+ T cells expressing CD29, granzyme B, or both antigens were higher in LC than R individuals, with such differences being prominent primarily among females (Figure 7A). Such a female-biased phenotype was also observed for CD57-expressing CD8+ T cells (Figure 7A), as well as CD8+ T cells expressing exhaustion marker TIGIT where in fact males exhibited the opposite pattern (Figure S11).Figure 7Viral-specific cytolytic CD8+ T cells are characteristic of both LC and active COVID-19(A) Frequencies of CD29+, granzyme B+, CD29+granzyme B+, and CD57+ CD8+ T cells among total CD8+ T cells from participants or those that are female or male.(B) Proportions of viral-specific CD8+ T cells that are positive for CD29 or granzyme B in LC and R. For (A) and (B), p values were calculated by Welch’s t test. ∗p < 0.05, ∗∗p < 0.01.(C) Evolution of granzyme B and CD38 expression in total, SARS-CoV-2-specific, and CMV-specific CD8+ T cells from an R and LC individual, as measured by MSI.(D) Proportion of total or viral-specific CD8+ T cells that express CD29, granzyme B, or both in specimens from participants hospitalized for severe COVID-19 as compared to those from post-acute R or LC individuals. CD38 expression was assessed as an activation marker.Elevated expression of granzyme B alone or in conjunction with CD29, but not CD29 alone, is characteristic of active severe COVID-19.For p value calculation, when one-way ANOVA values were p < 0.05, then Welch’s t tests were conducted for pairwise comparisons and those statistically significant labeled. ∗∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.For (A), (B), and (D), boxplots display IQR with median lines and dots as individual participants. See also Figure S11. Viral-specific cytolytic CD8+ T cells are characteristic of both LC and active COVID-19 (A) Frequencies of CD29+, granzyme B+, CD29+granzyme B+, and CD57+ CD8+ T cells among total CD8+ T cells from participants or those that are female or male. (B) Proportions of viral-specific CD8+ T cells that are positive for CD29 or granzyme B in LC and R. For (A) and (B), p values were calculated by Welch’s t test. ∗p < 0.05, ∗∗p < 0.01. (C) Evolution of granzyme B and CD38 expression in total, SARS-CoV-2-specific, and CMV-specific CD8+ T cells from an R and LC individual, as measured by MSI. (D) Proportion of total or viral-specific CD8+ T cells that express CD29, granzyme B, or both in specimens from participants hospitalized for severe COVID-19 as compared to those from post-acute R or LC individuals. CD38 expression was assessed as an activation marker. Elevated expression of granzyme B alone or in conjunction with CD29, but not CD29 alone, is characteristic of active severe COVID-19. For p value calculation, when one-way ANOVA values were p < 0.05, then Welch’s t tests were conducted for pairwise comparisons and those statistically significant labeled. ∗∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001. For (A), (B), and (D), boxplots display IQR with median lines and dots as individual participants. See also Figure S11. Focusing on the cytolytic markers that were the most consistently elevated among the LC-enriched clusters, we gated for frequencies of cytolytic viral-specific CD8+ T cells among R vs. LC individuals. Significantly higher frequencies of granzyme B+ or CD29+ cells were observed among CD8+ T cells specific to SARS-CoV-2, CMV, or EBV but not among influenza-specific CD8+ T cells (Figure 7B). Hence, cytolytic CD8+ T cells are a general LC-associated feature, among total CD8+ T cells and those recognizing SARS-CoV-2, CMV, and/or EBV antigens. Given that cytolytic T cell responses are features of an immune response to acute viremia, we wondered whether elevated frequencies of viral-specific CD8+ T cells expressing CD29 or granzyme B could reflect an early response that did not resolve properly in those with LC. To answer this question, we performed two analyses. First, we picked two individuals—one R and one with LC—with robust SARS-CoV-2- and CMV-specific CD8+ T cell responses and used combinatorial tetramer analysis to monitor CD29 and granzyme B median signal intensity (MSI) on total and viral-specific CD8+ T cells from the acute phase of infection to >5 months post-infection. CD29 remained constant over time among CD8+ T cells (total and specific to either virus) in both individuals (Figure 7C). By contrast, granzyme B expression decreased over time in viral-specific cells from the R individual, whereas it tended to increase in the LC individual (Figure 7C). Interestingly, the expression of activation marker CD38 exhibited a similar pattern as granzyme B: it decreased over time among SARS-CoV-2-specific and CMV-specific CD8+ T cells in the R but not LC individual (Figure 7C). Next, we compared the expression of these three antigens between the severe hospitalized COVID-19 cohort (Table S3) and the post-acute individuals from LIINC. Among viral-specific CD8+ T cells, CD29 levels did not differ significantly between those with active COVID-19 and either the R or LC individuals. By contrast, granzyme B expression was significantly lower in the R individuals as compared to both LC individuals and those with active COVID-19 among total CD8+ T cells and those specific for CMV. CD8+ T cells dually expressing CD29 and granzyme B were also significantly diminished uniquely among the R individuals, and this was observed among both SARS-CoV-2- and CMV-specific CD8+ T cells as well (Figure 7D). This diminution of cytolytic marker expression in R individuals matched that of activation marker CD38, suggesting that it reflects resolution of acute viremia (Figure 7D). Overall, our results suggest that CD29 and granzyme B, two cytolytic markers preferentially expressed by CD8+ T cells in those with LC, exhibit different kinetics and that the high expression of granzyme B in individuals with LC may reflect an inability to resolve a granzyme-B-mediated cytolytic CD8+ T cell response initiated during initial COVID-19 viremia. Discussion In this study, we implemented combinatorial tetramer technology to uncover potential correlations between viral-specific CD8+ T cells and LC. Overall, our findings revealed that LC individuals exhibited phenotypic perturbations among not only SARS-CoV-2-specific CD8+ T cells but also CD8+ T cells recognizing epitopes from EBV and CMV, viruses implicated as potential drivers of the disease. Our findings support the hypothesis that LC entails a prolonged cytolytic CD8+ T cell response against SARS-CoV-2 and resident herpesviruses. They also shed light on the characteristics of viral-specific CD8+ T cells that distinguish people who recovered from those who developed LC. EBV and CMV are herpesviruses known to elicit permanent residence in their human hosts, mostly in a latent state. Interestingly, EBV-specific and CMV-specific CD8+ T cells were phenotypically distinct from one another, during both the active and post-acute phases of COVID-19. CD8+ T cells recognizing CMV were of a more terminally differentiated state, while those recognizing EBV were more stem-like. This may be due to the ability of CMV, but not EBV, to induce memory T cell inflation, the process by which sporadic CMV reactivation drives expansion of CMV-specific CD8+ T cells. However, in LC individuals, we found that overall CMV-specific CD8+ T cells were preferentially of the Tcm phenotype and dis-enriched for Temra, a terminally differentiated subset typically associated with CMV-specific CD8+ T cells. Hence, the typical profile of CMV-specific CD8+ T cells appears to be different in LC individuals. What causes this altered differentiation profile is unclear, but one possibility is that it is driven by CMV reactivation. CMV reactivation has been associated with protection from LC,7 possibly because CMV can dampen immunity through its viral IL-10 homolog.4 As IL-10 suppresses expansion of Tcm,27 it is possible that lower CMV reactivation in LC individuals allows for CMV-specific Tcm to expand, hence increasing their abundance relative to Temra. The CMV IL10 homolog could also have conceivably led to our observation that exhausted (TIGIT+ and PD1+TIGIT+) CMV-specific CD8+ T cells were lower in LC as compared to R individuals—elevated frequencies of exhausted CMV-specific CD8+ T cells could reflect ongoing active CMV, which through production of the IL-10 homolog may create an overall immune-suppressed state protective against LC. To test this hypothesis, future studies should quantitate CMV viremia in longitudinal samples from acute COVID-19 into convalescence, for those who subsequently did vs. did not develop LC, under conditions where the phenotypes of CMV-specific CD8+ cells are monitored. Individuals with LC also exhibited differences in the subset distribution of their SARS-CoV-2-specific CD8+ T cells, relative to R individuals. For instance, their SARS-CoV-2-specific CD8+ T cells trended toward Tem and Trm phenotypes. Given that Tem are potent effector cells, and Trm are cells that may be migrating to or from tissues, their preferential existence in those with LC suggests the existence of a tissue-based SARS-CoV-2 reservoir in these individuals, as has been reported.28 Clustering analyses, which group CD8+ T cells based on high-parameter marker expression profiles rather than assigning them to classic T cell subsets, revealed two very similar stem-like clusters enriched in R as compared to LC individuals. These clusters—9 and 11—expressed low levels of markers of terminal differentiation (CD57), exhaustion (PD1, TIGIT, and CD39), and cytolytic activity (granzyme B, CD29). By contrast, clusters enriched among LC individuals were essentially of the “opposite” phenotype, i.e., terminally differentiated, cytolytic, and exhausted, and to a lesser extent activated as assessed by CD38 expression. These clusters were enriched in cells specific to SARS-CoV-2 (clusters 2 and 13), CMV (clusters 12 and 20), both SARS-CoV-2 and CMV (cluster 19), or EBV (cluster 5). Compared to total CD8+ T cells, these clusters all express high levels of terminal differentiation marker CD57 and cytolytic markers CD29 and granzyme B, which are all associated with T cell immunosenescence and/or aging.29,30,31 Notably, terminally differentiated CD8+ T cells have been associated poor immune control in other chronic infections, including HIV where antiretroviral therapy-free control of infection has been associated with non-differentiated CD8+ T cells with high proliferative and stemness-associated features.32,33,34 Among the terminally differentiated clusters enriched in LC individuals, the patterns of exhaustion marker expression differ, but overall, they express high levels of at least one of the exhaustion markers examined (PD1, TIGIT, and CD39). Interestingly, in the case of exhausted EBV-specific CD8+ T cells co-expressing PD1 and TIGIT, the frequencies of these cells exhibited a significant positive association with expression of immunosuppressive SIGLECs (SIGLEC7, SIGLEC11, and SIGLEC15) in LC individuals only. By contrast, there were trends for associations in the opposite direction in R individuals. Immunosuppressive SIGLECs have immunoreceptor tyrosine-based inhibitory motif (ITIM) domains and participate in immunoregulation by dampening immunity35 and have been associated with other chronic infections (HIV) as well as cancer.36,37 The significance of the positive association between exhausted EBV-specific CD8+ T cells and immunosuppressive SIGLECs in LC individuals is unclear, but could reflect an overall state of dampened immunity (with exhaustion + immunosuppression), perhaps due to chronic inflammation, which is a feature of LC.4 Given that immunosuppression through ITIM-containing SIGLECs can be mediated in part through IL-10,38,39 a possible future LC therapeutic avenue is the targeting of the SIGLEC-IL10 axis. Of note, SIGLEC15—which mediates its immunosuppressive effects in part through IL-10—has already been proposed as a therapeutic target for cancer immunotherapy.40 The increased prevalence of cytolytic CD8+ T cells recognizing SARS-CoV-2, CMV, and EBV during LC is consistent with prior reports that CD8+ T cells from people with LC express more cytolytic effectors including granzyme B.11,12 In the current study, we also observed this preferential LC-associated cytolytic signature among total CD8+ T cells, particularly among females. Women in general harbor CD8+ T cells with more potent cytolytic and inflammatory activity41 and higher frequencies of activated CD8+ T cells.42 Beyond the CD8+ T cell compartment, the overall baseline inflammatory state is also elevated in women: women harbor higher frequencies of neutrophils,43 and their pDCs produce higher levels of interferon α (IFN-α).44 These sex-dependent effects are attributed to both hormonal and chromosomal differences.42 Given that women are more prone to developing LC than men,45 it is possible that cytolytic CD8+ T cells may contribute to the predominance of LC in women, or alternatively LC could preferentially promote maintenance of such cells. Interestingly, the association between cytolytic signature and LC we observed for the SARS-CoV-2-, CMV-, and EBV-specific CD8+ T cells was not observed for influenza-specific CD8+ T cells. SARS-CoV-2, CMV, and EBV, unlike influenza, have been documented to persist for years in humans—CMV and EBV as classic examples of latent viruses, and, more recently, SARS-CoV-2 in the context of LC.28 Hence, the ability of persistent viruses to drive cytolytic CD8+ T cell responses may play a role in LC pathogenesis. Interestingly, we found that the frequencies of cytolytic (granzyme B+) SARS-CoV-2-specific CD8+ T cells positively associated with pro-inflammatory “alarmins” (S100A8 and S100A11) in R individuals, but that this association was lost (and in fact trends oppositely) in LC individuals. Under normal circumstances, it is expected that alarmin levels parallel cytolytic T cell responses, since alarmins can increase cytolytic CD8+ T cell responses.46 But in the context of LC, it appears that cytolytic CD8+ T cell responses develop independent of alarmins. How alarmins can increase cytolytic responses is not clear but may be due to the ability of these proteins to bind to pattern recognition receptors such as RAGE and TLR4 and initiate innate immune signaling. Future studies should interrogate the extent to which CD8+ T cells from LC individuals exhibit perturbations in innate immune signal transduction. Exactly what drives the preferential cytolytic CD8+ T cell responses against these viruses in those with LC is unclear. One potential mechanism is inability to shut down viral-specific CD8+ T cell response against active replicating SARS-CoV-2 and/or reactivated EBV/CMV during acute COVID-19. Supporting this hypothesis is our observation that granzyme B among CD8+ T cells specific for SARS-CoV-2, CMV, and EBV is higher during acute vs. post-acute phase, in a manner that tracks closely with activation marker CD38. Furthermore, our longitudinal analysis suggested that while activated and granzyme B+ CD8+ T cells against SARS-CoV-2 and CMV decrease over time in context of full recovery, they remained elevated during LC. Another important aspect to establish is the extent to which these cytolytic CD8+ T cells actually contribute to LC pathogenesis or are just a bystander outcome. CD8+ T cells play important roles in controlling acute viral infection including during COVID-1947,48 and may well also have a protective role against LC. Future studies leveraging animal models of LC4,49 should assess the extent to which cytolytic CD8+ T cells associate with LC disease. If so, then follow-up studies leveraging gene knockouts or immune cell depletion could inform on the extent to which these cells can directly mediate LC pathogenesis. While our study focused on viral antigen-specific CD8+ T cell responses, we also observed some interesting differences in total CD8+ T cells between LC vs. R individuals worth mentioning. Some of these differences were reflective of differences observed among all the examined viral-specific CD8+ T cells. For example, our observation of LC individuals harboring higher frequencies of CD8+ Tem and Tn/Tscm was also observed among most of the viral-specific CD8+ T cells. Likewise, low expression of HLADR among total CD8+ T cells from LC individuals was also observed among all the viral-specific ones. Notably, this low expression of HLADR argues against a global, chronic state of CD8+ T cell activation in those with LC. Clustering analysis also revealed two clusters (9 and 11) to be underrepresented in those with LC. These two clusters harbored similar phenotypic signatures, including those characteristic of a non-exhausted, non-differentiated, and non-cytolytic state, with only CD69 distinguishing them. Together, these observations suggest that total CD8+ T cells differ in those with LC, in that they harbor a preferentially effector-like, exhausted phenotype with elevated baseline expression of cytotoxic molecules, but given their low expression of HLADR are unlikely to have recently encountered cognate antigen. Limitations of the study A strength of our study was the in-depth analysis of antigen-specific CD8+ T cell responses, as well as the high level of characterization of the participants and their lack of vaccination or re-infection, major confounders of LC research in the modern era. However, our study has limitations. This study was descriptive in nature and hence statistical differences between frequencies of antigen-specific CD8+ T cell populations between patient groups were limited. We note that the limited availability of appropriate specimens and the technological challenges associated with combinatorial tetramer technology limited the size of the cohort. This study was limited to phenotypic analysis and did not perform functional assays per se, but we note that our panel included markers of established functions (e.g., granzyme B for cytolytic activity and exhaustion markers for T cell exhaustion) and also note that functional assays following stimulation with SARS-CoV-2 peptides were previously performed using specimens from the same R and LC individuals.13 Our focus here was instead on assessing the original phenotypes of viral-specific CD8+ T cells. Overall, our study suggests that individuals with LC preferentially harbor CD8+ T cells specific to SARS-CoV-2, CMV, and EBV sharing phenotypic features defined by terminal differentiation, exhaustion, and cytolytic activity. Future studies should assess the extent to which these cells contribute to the pathogenesis of LC, by directly assessing for preferential expression of cytolytic effectors among these viral-specific CD8+ T cells in larger cohorts, as well as in animal models of LC.49 Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Nadia Roan (nadia.roan@gladstone.ucsf.edu). Materials availability This study did not generate new unique reagents. Data and code availability •Raw CyTOF datasets are available for download as FCS files from the following link in Dryad repository: https://doi.org/10.5061/dryad.g1jwstr51.•This paper does not report original code.•Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. •Raw CyTOF datasets are available for download as FCS files from the following link in Dryad repository: https://doi.org/10.5061/dryad.g1jwstr51. Raw CyTOF datasets are available for download as FCS files from the following link in Dryad repository: https://doi.org/10.5061/dryad.g1jwstr51. •This paper does not report original code. This paper does not report original code. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.