Memory cytotoxic SARS-CoV-2 spike protein-speci � c CD 4 + T cells associate with viral control


 Little is known of the role of cytotoxic CD4+ T-cells in the control of viral replication. Here, we investigate CD4+ T-cell responses to three dominant SARS-CoV-2 epitopes and evaluate antiviral activity, including cytotoxicity and antiviral cytokine production. Diverse T cell receptor (TCR) usage including public TCRs were identified; surprisingly, cytotoxic CD4+ T-cells were found to have signalling and cytotoxic pathways distinct from classical CD8+ T-cells, with increased expression of chemokines and tissue homing receptors promoting migration. We show the presence of cytolytic CD4+ T-cells during primary infection associates with COVID-19 disease severity. Robust immune memory 6-9 months post-infection or vaccination provides CD4+ T-cells with potent antiviral activity. Our data support a model where CD4+ killer cells drive immunopathogenesis during primary infection and CD4+ memory responses are protective during secondary infection. Our study highlights the unique features of cytotoxic CD4+ T-cells that use distinct functional pathways, providing preventative and therapeutic opportunities.


Introduction
Understanding the immune phenotype of SARS-CoV-2-speci c T cells that are protective or pathogenic is pivotal to de ne therapeutic and prophylactic strategies to manage the COVID-19 pandemic 1 . Early T cell responses against diverse SARS-CoV-2 proteins associate with clinical protection 2,3,4,5 . In contrast, patients with severe disease have reduced frequencies of circulating T cells and increased activation markers 6,7,8,9,10,11,12 . There is an urgent need to for in-depth functional characterization of virus-speci c T cells during natural infection and following vaccination.
In addition to their "helper" role, mainly through interaction with antigen presenting cells and soluble mediators, the direct effector and antiviral function of virus-speci c CD4 + T cells have been studied in the setting of several human virus infection settings, such as in uenza, cytomegalovirus (CMV) and Epstein-Barr virus (EBV) 13 . CD4 + cytotoxic T cell (CTL) responses were recently recognized to play an important role in the immune response to viral infections. However, most mechanistic studies investigating cytotoxic CD4 + T cells were conducted in murine models; therefore, it is unclear if insights gained from such studies are relevant to human CTL responses 13 . Recent SARS-CoV-2 CD4 + T cell studies have been focused on the regulatory and helper function of these cells, such as T follicular helper cells (Tfh) and their role in assisting antibody production 14,15 . Very little attention has been paid to the direct effector function and antiviral activity of SARS-CoV-2 speci c CD4 + memory T cells, possibly due to the limited experimental tools available.
Current advances in technologies that combine population speci c single cell transcriptomic pro ling with TCR sequencing analysis and in vitro T cell functional assays have enabled us to study the quality of T cell responses and their ability to control virus replication 5 . Multiple immunodominant SARS-CoV-2 spike protein-speci c T cell epitopes have been identi ed and are frequently detected in individuals who have recovered from SARS-CoV-2 infection and following vaccination 16,17,18 .
In this study we focus on three most dominant CD4 + spike-speci c epitopes identi ed in a cohort of individuals who had recovered from SARS-CoV-2 18 . We explore correlations with the quality of these spike-speci c T cell responses and study their TCR repertoires, public TCR usage and their association with antiviral activity. We examine the potential mechanisms of antiviral activity in a speci c cytotoxic subset of CD4 + T cells using single cell transcriptome analyses of expanded T cell clones bearing the same TCRs as identi ed in single cell analysis.
To further characterize these three dominant T cell responses, we generated 50 S 166−180 -speci c T cell clones, 54 S 751−765 -speci c T cell clones and 49 S 866−880 -speci c T cell clones from convalescent samples and evaluated their functionality. T cell receptors (TCRs) of each clone were also sequenced (Supplementary Table 2). All clones expressed cytokines including TNF-α, IFN-γ and IL-2 upon antigen activation (Fig. 1e). S 866−880 -speci c T cell clones displayed the highest antigen-sensitivity, with the lowest EC 50 calculated from TNF-α, IFN-γ and IL-2 production. Interestingly, in addition to cytokine expression, we observed a substantial proportion of T cell clones (4 from S 166−180 , 8 from S 751−765 and 60% for S 866−880 (Fig. 1f). This highlights the existence of SARS-CoV-2-speci c CD4 + CTLs following SARS-CoV-2 infection. We de ned CD4 + killer cells as clones with a killing capacity of > 10%. Among these spike-speci c T cell clones, S 866−880 -speci c T cells showed the strongest cytotoxic and killing capacity (Fig. 1f, 61% P < 0.001) and proportion of killer cells (Fig. 1g, 60% P < 0.001) whereas S 166−180 T cells showed the least cytotoxic killing potential. The highest effector function and antiviral e cacy was seen for S 866−880 , suggesting signi cant effector function of cytotoxic CD4 + T cell acting on virus infected cells and controlling virus replication.
2. Spike-speci c CD4 + T cell antiviral activity is associated with cytotoxic activity, cytokine production and antigen load Taking advantage of our in vitro SARS-CoV-2 virus infection system 5 , we assessed the antiviral activity of these spike-speci c CD4 + T cells. In brief, Epstein Barr virus (EBV)-transformed B cell lines (BCLs) ectopically expressing ACE2 were infected with SARS-CoV-2 ( Fig. 2a) then cocultured with spike-speci c CD4 + T cells. T cell recognition of virus-infected cells was examined by intracellular cytokine staining (ICS), and the suppression of SARS-CoV-2 replication by T cells was assessed by quantifying the number of viral copies in the infected cells after 48hrs of co-culturing. Our data showed that CD4 + T cells targeting these three dominant spike epitopes can recognise virus-infected cells and produce cytokines after activation (Fig. 2b), with S 866−880− speci c-T cells having the highest proportion of TNF-α (P < 0.001), IFN-γ (P < 0.001) and IL-2 (P < 0.001) producing CD4 + T cells after encountering virus-infected BCLs (Fig.   2c). More importantly, CD4 + T cell clones targeting each of the three epitopes exhibited, to varying extents, direct effector function against the virus, capable of suppressing virus replication (Fig. 2d). In particular, T cells targeting S 866−880 showed signi cantly better antiviral e cacy compared to T cells targeting epitope S 751−765 restricted by the same HLA-DRB1*15:01 (P = 0.002). We sought to investigate whether this strong antiviral activity was a result of high effector function or exposure to high antigen loads by examining single cell gene expression from tetramer-sorted short term cultured T cell lines. First, we compared single cell gene expression pro les of T cells targeting S 751−765 (n = 1629) and S 866−880 cells (n = 2233) (Fig. 2e). We observed signi cant upregulation of genes encoding effector molecules, such as cytotoxic molecules KLRK1 (P = 1.37 x 10 −44 ), GZMB (P = 6.13 x 10 −4 ) and GZMK (P = 5.13 x 10 −10 ), and cytokines CCL3 (P = 2.79 x 10 −6 ), CCL4 (P = 4.4 x 10 −8 ), TNF (P = 9.08 x 10 −12 ) in S 866−800speci c T cells compared to S 751−765 -speci c T cells (Fig. 2f). This further con rms the cytotoxic potential of S 866−880 -speci c T cells.
To estimate the antigen load of each epitope on virus-infected cells, we cultured T cells with the same number of target cells either infected with SARS-CoV-2 or loaded with variable amounts of peptide, then assessed T cell responses by ICS (Fig. 2g). The antigen-load in virus-infected cells was equivalent to the peptide concentration that elicited a similar level of response to virus-infected cells. Surprisingly, much lower concentrations (equivalent to 0.06µM) of S 866−880 peptide were presented on virus-infected cells, when compared to S 166−180 (about 0.11µM) and S 751−765 (equivalent to 2.57µM) (Fig. 2h). Our data suggests that the higher antiviral activity of S 866−880 -speci c T-cells is likely to result from cytotoxicity and high antigen sensitivity even when antigen load on the surface of infected cells was relatively low.
Next, we compared the cytotoxic activity of S 751−765 and S 866−880 tetramer-sorted short term cultured single cells isolated from those who had recovered from mild or severe acute COVID-19 (n = 2 mild, 2728 cells; n = 2 severe, 1823 cells; Fig. 2i). We observed that T cells isolated from patients who had recovered from severe disease were more activated and expressed higher levels of T cell effector function genes CD69 (P = 2.75 x 10 −5 ), CCL5 (P = 1.61 x 10 −22 ), IL2RG (P = 3.27 x 10 −35 ), MX1 (P = 3.76 x 10 −13 ), in particular cytotoxic molecules such as GZMB (P = 6 x 10 −17 ) and GZMM (P = 3.87 x 10 −3 ), compared to cells from mild cases (Fig. 2j). Collectively, these data suggest that cytotoxic CD4 + T cells may play a role in the immunopathogenesis of the severe disease.  (Fig. 3a). This highlighted the importance of the α chain in these spike epitope TCRs; hence we decided to focus on the αV chain when investigating TCR clonotypes further. Single cells from epitope speci c T cells sampled at different timepoints (S 166−180 ex vivo acute, ex vivo 1-3 months, short term culture 9 months convalescence from total 5 patients; S 751−765 and S 866−880 ex vivo acute, ex vivo 1-3 months, short term culture 6 months and 9 months convalescence from total 6 patients) were analysed together to identify public clonotypes (CDR3 amino acid and V gene usage), which are unique clonotypes shared among more than one unrelated patient. We found that public α clonotypes (CDR3α and TRAV) were shared by many patients whereas β clonotypes (CDR3β and TRBV) were shared by a smaller number of patients. For example, the maximum number of patients with one particular α clonotype (CAGTGNNRKLIW, TRAV 25) from S 866−880 -speci c T cells could be 6/6 whereas the highest number of patients sharing any β clonotype from the same epitope was 3/6 ( Table 1 and Supplementary Table 3). Examination of paired αβ public clonotypes revealed that the β public clonotypes were more diverse than α public clonotypes, with no clear dominant Vβ gene usage for any epitope. By focusing on α clonotypes, we reasoned that we would be better able to study the dominant α V genes for each epitope, which should also capture the diversity of β clonotypes.
Next, we sought to compare the proportion of public and private clonotypes from each epitope. S 166−180speci c T cells have higher proportions of TCRs matching public clonotypes compared to the other two epitopes (n = 21 S 166−180 public α clonotypes, n = 16 S 751−765 public, n = 19 S 866−880 public; Fig. 3b). We also investigated if there were any differences between public and private clonotype expansion for each epitope at different timepoints (acute, 1-3 months and 6-9 months convalescence) and found that T cells with public clonotypes were present at higher frequencies compared to cells with private clonotypes, with the exception of S 866−880 at 6-9 months convalescence (acute: S 166−180 P = 1.  Fig. 3). In summary, these results highlight the potential different functional activities between epitope-speci c T cells. 4. Cytotoxicity and function of spike-speci c CD4 + T cells are regulated by more than TCR usage alone.
We next investigated whether the effector function of spike-speci c CD4 + T cells was due to their broad TCR usage by comparing antigen sensitivity and killing capacity of T cell clones bearing the same TCR. We found that T cells with the same TCRs had very different antigen sensitivities, re ected in a wide range of EC 50 values for IFN-γ, TNF-α and IL-2 production (Fig. 4a). These T cell clones also produced different cytokine pro les upon antigen stimulation (Extended Data Fig. 4). Moreover, S 866−880 -speci c T cell clones with shared TCRs had distinct killing capacities (Fig. 4b), for example, clone 2 had 40% killing capacity while clone 3 showed minimal killing despite sharing the same TCR (TRAV12-1/TRBV5-1). This suggested that the cytolytic activity of these spike-speci c CD4 + T cells was due to factors beyond TCR usage. This observation was further con rmed by variable expression levels of cytotoxic molecule genes, such as PRF1, GZMA and GZMB, in tetramer-sorted S 866−880 single cells (n = 399 cells) with the same TCR usage (Fig. 4c). Indeed, we noted a positive association between killing ability with degranulation activity, as measured by CD107a expression, upon antigen stimulation (R = 0.413, P = 0.001, Fig. 4d).
This suggests that high expression of CD107a can act as a potential marker for CD4 + CTLs. 5. Antiviral activity of spike-speci c CD4 + T cells strongly correlates with their killing capacity and IL-2 production As previously highlighted, a number of spike-speci c CD4 + T cell clones were capable of suppressing virus replication (Fig. 2d). We further examined whether this antiviral effector function was mediated via direct killing of virus-infected cells or by the expression of soluble inhibitory factors. Firstly, our data demonstrated that the antiviral activity of the CD4 + T cell clones strongly correlated to the proportion of cells producing IL-2 upon stimulation with virus-infected cells (R = 0.226, P = 0.030, Fig. 5a), but did not correlate with the proportion of cells producing IFN-γ or TNF-α (Extended Data Fig. 5). Secondly, we found a signi cant association between the killing capacity of T cells and their potential to control virus replication. Indeed, we observed that killer cells (more than 10% killing capacity) could suppress virus replication more e ciently than non-killers (less than 10% of killing) ( Fig. 5b suggesting the importance of direct killing of virus-infected cells in viral control by CD4 + CTLs. Strikingly, some non-killer cells were capable of inhibiting viral replication as e ciently as killer cells, indicating the potential contribution of soluble factors to viral control (Fig. 5c). Subsequently we compared cytokine and chemokine production of these viral suppressing non-killer CD4 + T cells with other non-killer cells ( Fig. 5d). We noticed that non-killer CD4 + T cells capable of suppressing virus replication (suppressor clone) produced signi cantly higher concentrations of IL-2 (P = 0.040) and IFN-γ (P = 0.040) than nonkiller CD4 + T cells incapable of viral suppression (non-suppressor clone), highlighting a role for IL-2 and possibly IFN-γ in CD4 + T cell control of SARS-CoV-2 infection.
6. Cytotoxic spike-speci c CD4 + CTLs utilise distinct cytolytic pathways with increased migration potential Activated CD8 + CTLs carry out their killing function primarily by releasing cytotoxic granules such as perforin and granzymes, which subsequently induce apoptosis of target cells. To determine whether the killing of target cells by spike-speci c CD4 + CTLs was also mediated through the perforin-dependent pathway, S 866−880− speci c CD4 + T cell clones were treated with concanamycin A (CMA), an inhibitor of perforin, prior to adding to target cells loaded with peptide. The cytolytic activity mediated by CD4 + T cell clones was completely blocked by CMA, resulting in decreased or ablated killing capacity (Fig. 6a), and reduced or no viral suppression (Fig. 6b). To identify other factors involved with this perforin-mediated effective viral control, we grouped tetramer-sorted S 866−880− speci c single cells from short term cultured lines into perforin-high (n = 693 cells) and perforin-low (n = 724 cells) subsets according to their perforin module scores (Supplementary Table 4) and compared their gene expression pro les (Fig. 6c) and inhibitory receptors such as TIGIT (P = 2.82 x 10 −4 ) and KLRG1 (P = 1.94 x 10 −2 ) (Fig. 6d). This suggests there is increased migration potential of activated cytotoxic CD4 + T cells to infected tissue.
To examine whether memory T cells established following natural infection could provide su cient protection against secondary viral infection, we collected peripheral blood mononuclear cells (PBMCs) from the same patients 6-9 months after infection. Although T cell responses to these three spike epitopes (S 166−180 , S 751−765 and S 866−880 ) signi cantly declined six months after infection (Fig. 7a), we were able to sort, sequence, and expand these spike-speci c CD4 + T cells after in vitro antigen stimulation. We discovered that the diversity of the TCR repertoire of these CD4 + T cells were maintained 6-9 months after infection (Fig. 7b). We then assessed the antiviral e cacy of these bulk spike-speci c T cell lines using our in vitro SARS-CoV-2 infection assays. The bulk lines targeting all three epitopes elicited strong responses against BCLs infected with SARS-CoV-2 (Victoria strain) and variants of concerns (VOCs; Delta and Omicron). The CD4 + T cell lines produced profoundly elevated level of IFN-γ, TNF-α and IL-2 ( Fig. 7c and Extended Fig. 6) and showed signi cant cytotoxic potential with upregulated CD107a expression (Fig. 7d). In addition, we found that these antigen-speci c CD4 + bulk cell lines are capable of suppressing SARS-CoV-2 replication and showed strong inhibition against VOCs (Fig. 7e). Our data highlight the protective role of these dominant spike-speci c CD4 + T cells in secondary infection against different SARS-CoV-2 variants.

Discussion
In this study, we focus on the three most dominant spike-speci c CD4 + T cell responses identi ed in our cohort restricted by common HLA Class feature dominant public TCR usage, in particular TRAV35, that is biased to TH1 cell subset 20 (i.e. high ID2 expression) and suppressed TH2 transcriptomic pro les 21 (i.e. high TH2 suppressor IRF4 and NR4A2 expression). TCR repertoire analysis surprisingly reveals that over 30% of TCRs are public in all three epitope speci c responses at all time points studied, including public clonotypes with shared αV and βV sequences. There are also different dominant TRAV usages across these three epitope speci c T cells, CD4 + cytotoxic T cell responses observed in virus infection and cancer may contributed to disease pathogenesis 22,23 or protection 24, 25, 26 . We observe potent cytotoxic capacity and antiviral e cacy of S 866−880 CD4+ T cells. Looking further into this response, we rst nd that cytotoxicity is likely primarily mediated by perforin, which is in agreement with previous reports 27, 28, 29 . When we compare cells at a single cell level, we nd a signi cant association of CD4 + cytotoxic T cells in individuals recovered from severe in comparison with mild disease; however, the number of patients studied are small, therefore further large-scale studies are needed. Using single cells with high and low perforin scores as a marker for CD4 + cytotoxic T cells, we nd that perforin-high cells have the potential to migrate to affected tissue e ciently by expressing high levels of chemokines (CCL3, CCL4 and CCL5) and tissue homing receptors (ITGB1, ITGA4 and ITGAL), in line with the observation that CD4 + CTLs are expanded in the lung of patients with severe COVID-19 30 . Our data highlights the potential pathogenic role of these potent cytotoxic T cells during primary virus infection. These cells may accumulate in the infected organ/tissue, causing excessive in ammation and bystander killing of cells with elevated MHC Class II expression, including professional antigen presenting cells. This hypothesis merits further investigation.
We also nd that cytotoxic CD4 + T cells may use different signalling and cytotoxic molecules compared to classical CD8 + T cells; CD4 + CTLs express GZMK and GZMM signi cantly higher whereas CD8 + CTLs express much higher GZMB and GZMH. In addition, NK activation receptor NKG2D (KLRK1) and NKG7, known to be important for cytotoxic degranulation 31 , are highly expressed on CD4 + CTLs but not classical CD8 + CTLs.. Importantly, it is known that NKG2D ligands are expressed in the lung affected by COVID-19, therefore the high NKG2D expression on cytotoxic CD4 + T cells could potentially contribute to potent cytotoxicity in infected lungs and provide additional protection in mild COVID-19 cases or excessive in ammation in severe cases.
Surprisingly, we nd that the functionality of these CD4 + T cells, in particular cytotoxicity, appears to involve mechanisms beyond TCR usage alone. S 866−880 -speci c T cell clones sharing the same TCR exhibit a broad range of killing capacity, which correlate with their ability to control virus replication in infected cells. However, T cells with no killing activity can still suppress virus replication and express high levels of IL-2 and IFN-γ. In addition, we observe an overall signi cant association between IL-2 and IFN-γ and the antiviral e cacy of all three epitope responses, suggesting direct antiviral effector function of CD4 + spike-speci c T cells independent of cytotoxicity.
Finally, the immune memory responses from 6 -9 months post-infection demonstrate potent antiviral e cacy to the original SARS-CoV-2, Delta and Omicron Variants, suggesting memory responses to spike protein, induced by vaccine or natural infection, may contribute to protection against secondary infections to all VOCs, including Omicron, by direct killing of virus infected cells and/or antiviral cytokine and chemokine production.

In summary, our study provides evidence of cytotoxic CD4 + T cells in SARS-CoV-2 virus infection and new
insights on the potential mechanisms related to this important group of CD4 + cells at a single cell level.
Induction of potent CD4 + killer cells by vaccination could be an attractive approach for novel vaccine designs to support early viral control.

Declarations
This work uses data provided by patients and collected by the NHS as part of their care and support #DataSavesLives. 19/SC/0296). Clinical de nitions were de ned as previously described 1 .
Generation of ACE2-transduced EBV-transformed BCLs. EBV-transformed BCLs 32 and ACE2-transduced BCLs 5 were established as described previously. In brief, the cDNA for the human ACE2 gene (ENSG00000130234) was cloned into a lentiviral vector backbone (Addgene plasmid ID 17488), then cotransfected with packaging plasmids pMD2.G and psPAX2 into HEK293-TLA using PEIpro (Polyplus) to produce lentivirus. EBV-transformed BCLs were infected with ACE2-coding lentivirus followed by cell sorting via ow cytometry to enrich ACE2-expressing B cells. B cells with stable expression of ACE2 were maintained with 0.5 μg ml −1 of puromycin (Thermo Fisher Scienti c). Mycoplasma testing was carried out every 4 weeks with all cell lines using MycoAlert detection kit (Lonza).
Generation of T cell lines and clones. Short-term SARS-CoV-2-speci c T cell lines were generated as described previously 33 . Brie y, 2 x 10 6 PBMCs were pulsed with 10 μM peptides at 37°C for 1hr and cultured in R10 (RPMI 1640 medium with 10% human serum, 2 mM glutamine and 100 mg ml −1 of penicillin-streptomycin) at 2 x 10 6 cells per well in a 24-well plate (Costar). IL-2 was added to a nal concentration of 100 IU ml -1 on day3. S 751-765 -and S 866-880 -speci c T cell clones were established by sorting tetramer + CD4 + T cells from thawed PBMCs or short-term T cell lines on day 10-14. S 166-180speci c T cell clones were generated by cell sorting with TNF-α, IFN-γ, and IL-2 secretion assay (Miltenyi).
T cell clones were then expanded with irradiated allogeneic PBMCs every 2-3 weeks as described previously 34 .
IFNγ ELISpot assay. Ex vivo assays were carried out using either freshly isolated or cryopreserved PBMCs as described previously 18 . Peptides were added to 2 x 10 5 PBMCs at a nal concentration of 2 μM for 16-18hrs. For in vitro ELISpot assays, autologous and allogeneic EBV-transformed BCLs were loaded with peptides, and subsequently cocultured with T cell clones or bulks at an effector:target (E:T) ratio of 1:50 for at least 6hrs. To quantify antigen-speci c responses, mean spots of the control wells were subtracted from the sample wells, and the results expressed as spot forming units (SFU) per 10 6 PBMCs. Responses were considered positive if results were at least 3x the mean of the quadruplicate negative control wells and >25 SFU/10 6 PBMCs. If negative control wells had >30 SFU/10 6 PBMCs or positive control wells (PHA stimulation) were negative, the results were excluded from further analysis.
10x scRNA-seq data processing. BCL les were converted to FASTQ les using cellranger mkfastq (Cellranger v.5.0.0). Counts matrices and sample demultiplexing was carried out using cellranger count (CellRanger) using with Feature Barcode options. For additional donor deconvolution, cellSNP v.0.3.2 was used to generate a list of SNPs from Cellranger output (BAM les), followed by Vireo v.0.5.6 to demultiplex sequencing data into individual patients from pooled sequenced libraries. The resulting counts matrix was analyzed in R using Seurat.
Single cell RNA sequencing analysis. Cells were ltered using the following criteria: minimum number of cells expressing speci c gene = 3, minimum number of genes expressed by cell = 200 and maximum number of genes expressed by cell = 4000. Cells were excluded if they expressed more than 5 -10% mitochondrial genes. Patient-speci c cells were integrated using Harmony v.1.0 to remove batch effects.
The AddModuleScore function (Seurat) was used to look at the expression of speci c gene sets (Supplementary Table 4). The average expression of a gene set was calculated, and the average expression levels of control gene sets were subtracted to generate a score for each cell relating to that particular gene set. Higher scores indicate that that speci c signature is more highly expressed in a particular cell compared with the rest of the population. Cells with a module score ≥ 1 were de ned as perforin-high; cells with a score ≤ 0.25 were de ned as perforin-low. The FindMarkers function (Seurat) was used to evaluate differentially expressed genes (DEGs) between two conditions using MAST (Modelbased Analysis of Single Cell Transcriptomics) statistical test, with different sequencing batches as latent variables. DEGs were visualized on volcano plots using EnhancedVolcano v1.6.0 and VlnPlot (Seurat).
SmartSeq2 and 10x TCR processing. TCR sequences were reconstructed from SmartSeq2 scRNA-seq FASTQ les using MiXCR v.3.0.13 to produce separate TRA and TRB output les for analysis. The output les were parsed into R using tcR v.2.3.2. TCR sequences were extracted from 10x VDJ sequencing using cellranger vdj (Cellranger). The resulting ltered_contig_annotations.csv le was analysed in R.