Functional analyses and single cell immunoprofiling uncover sex-specific differences in SARS-CoV2 immune memory development

SARS-CoV-2 infection leads to a broad range of outcomes and immune responses, with the development of neutralizing antibodies generally correlated with protection against reinfection. Here, we have characterized both neutralizing activity and T cell responses in a cluster of subjects with mild disease linked to a single spreading event. Surprisingly, we observed sex-specific associations between spike- and particularly nucleoprotein-specific T cell responses and neutralization, with pro-inflammatory cytokines being linked to higher titers only in males. Using single cell immunoprofiling, which provided matched transcriptome and T-cell receptor (TCR) profiles in restimulated CD4 + and CD8 + cells from these subjects, we identified differences in type I IFN signaling that may underlie this difference in antibody generation. Finally, we also identified several TCRs associated with cytokine producing T cells. Altogether, our work maps the breadth of immunological outcomes of SARS-CoV2 infections and highlight the potential role of sex-specific feedback loops during the generation of neutralizing antibodies.


Introduction
The coronavirus disease 2019 (COVID- 19) pandemic has led to more than 400 million cases and 5 million deaths con rmed as of early 2022 (www.who.int/covid -19). Over the past year and a half, many studies have helped characterize the biology of SARS-CoV-2 and the role of the immune system in both pathogenesis and protection, and have notably led to the development of multiple highly effective vaccines to SARS-CoV-2 as well as some therapeutic options including recombinant neutralizing antibodies (reviewed in 1 ).
In addition to these therapeutic advances, the large-scale immunopro ling of COVID-19 subjects has offered important insights into the basic immunobiology of responses to acute infections. COVID -19 severity has been associated with distinct cytokine pro les and cellular compositions [2][3][4][5] , generally tting with the notion of a "cytokine storm" that contributes to pathology and have demonstrated an expected network across immune cell states. Beyond these studies of acute cases, several groups have pro led the memory responses in convalescent individuals, with an initial focus on identifying neutralizing antibodies developed during infection 6-8 and on the SARS-CoV-2 epitopes recognized by memory T cells [9][10][11] . Of particular relevance here, the application of single cell sequencing technologies has allowed the joint mapping of T and B cell receptor (TCR/BCR) sequences and transcriptional pro les 8, 12 , while larger scale studies have mapped TCR repertoires more broadly 13,14 . In spite of these studies, the link between epitopes, TCRs and T cell function, as well as its impact on the development of antibody responses, remains only partially understood.
Here, we leveraged an existing cohort of convalescent subjects, where we had previously pro led antibody responses 8 , to combine functional analyses of spike and nucleoprotein re-stimulated T cells with single cell sequencing technologies. We observed a diverse array of T-cell cytokines produced in response to SARS CoV-2 antigen -both spike and nucleoprotein -including those associated with Th1, Th2 and Th17 cells. T cell responses to nucleoprotein were a better predictor of previous infection compared to spike protein. Surprisingly, we identi ed a sex-speci c association between T cell cytokine production and the development of neutralizing antibody responses. Upon pro ling of the T-cell responses using scRNA-seq, we observed a number of genes upregulated in a sex speci c manner in SARS CoV-2 responding CD4 + T-cells, including those involved in the type 1 interferon response. In addition, scRNAseq analyses of re-stimulated cells highlighted a broad functional diversity of SARS CoV-2 responding T-cells that is not observed in baseline PBMCs, and links TCR sequence to speci c functional pro les.

Cohort of previously SARS-CoV-2-infected individuals
To investigate and characterize the development of immune memory after COVID-19, we obtained samples from a cohort of 85 individuals that were either infected or close contacts of SARS-CoV-2 infected subjects during the early months of the pandemic in the U.S Northeast. All symptomatic individuals included in the cohort had a mild disease course (1-2 on W.H.O. severity scale). Of these subjects, 38 were men and 47 were women, with an average age of 46 (Table 1). Samples were collected in the 5-6 weeks that followed the onset of symptoms or a potential infectious contact to evaluate the development of antibody responses as well as the phenotype of memory T cells generated during this infection. To this end, we collected both serum and whole blood, and PBMCs were isolated from whole blood within 4 hours of blood draw and cryopreserved. In addition, participants lled out a survey reporting clinical information and including the duration and nature of their symptoms (Table 1; Fig. 1A).
Out of 85 individuals, 34 were con rmed positive by nasopharyngeal swab and PCR. However, given the low availability of testing during the rst wave of infections, it seemed likely that some individuals were infected but not tested, or tested after their SARS-CoV-2 viral load became undetectable. We therefore performed serology testing for both IgM and IgG against the receptor binding domain (RBD) of the spike protein or against the nucleoprotein of SARS-CoV2 on all serum samples from this cohort. As expected, all PCR-con rmed subjects were con rmed positive by serology, with the exception of two individuals that did not appear to mount detectable antibody responses. In addition, we identi ed an additional 17 individuals who tested positive by serology for responses against RBD (red circles in Fig. 1B), and all but one had demonstrable neutralizing responses (vide infra). Ultimately, we treated all subjects with either a positive PCR test or positive serology as "previously infected", and all other individuals as uninfected controls, and we will use these designations hereafter ( Table 1).

Development of SARS-CoV2 neutralizing antibodies
Having established a serological assay, we focused on measuring neutralizing antibodies in serum through two parallel approaches; one using a virus pseudotyped with the SARS-CoV-2 spike protein (reported as ID50_pseudo and ID80_pseudo, the serum titers that neutralize 50 or 80% of the infection) and another using an authentic SARS-CoV-2 infection model (reported as ID50). In both assays, titers obtained from the two approaches were highly correlated (Pearson = 0.675, p < 0.001 for ID50_pseudo and ID50) (Fig. 1C), and the presence of signi cant neutralization activity was only detected in infected subjects (Fig. 1D). We noted that neutralization assays with authentic SARS-CoV-2 virus provided a much wider dynamic range (with ID50s ranging from 20 to ~ 3000, versus 20 to 1000 for pseudotyped virus).
Based on the high correlation between the two assays, this discrepancy in dynamic range is more likely due to technical considerations rather than the biological impact of non-spike targeting neutralizing antibodies, although this would need to be examined in more detail for further con rmation. Nevertheless, we chose to focus on the neutralization titers generated through authentic SARS-CoV-2 assays through the rest of this study, and focused on understanding the correlates of protection in this cohort.
Next, we examined whether any of the clinical parameters collected were associated with the development of neutralizing activity. In spite of the association between age, sex and disease severity as well as symptom duration, we did not observe any correlation between these parameters and neutralizing antibody titers ( Fig. 1E and Supp Fig. 1A-B). As all COVID-19 cases followed here were mild disease by W.H.O criteria, we focused on symptom duration as a surrogate marker of disease course, but again we did not observe any correlation between this parameter and the development of neutralizing antibodies (Supp Fig. 1A). Altogether, these results show that there is a marked diversity in the development of neutralizing antibodies, that does not appear to be correlated with disease course but is probably the result of the inherent heterogeneity of the immune response across subjects.

SARS CoV-2 speci c T cells produce a range of cytokines upon restimulation
To understand how the development of T cell memory and the generation of neutralizing antibodies are correlated after SARS-CoV2 infection, we activated PBMCs from COVID-19 recovered individuals and uninfected controls with a tiled mix of 15mer peptides in a pool, derived from either the spike (S) or nucleoprotein (N) of SARS-CoV-2 (commercially available; see methods). In total, we measured the T-cell cytokine responses across 78 individuals (48 infected and 30 uninfected controls). After restimulation, supernatants were collected and the levels of 10 cytokines capable of being produced by T cells (IL-2, IL-5, IL-6, IL-9, IL-10, IL-17A, IL-17F, IL-22, IFN-γ and TNF-α) were measured with a bead array. First, we examined the presence of SARS-CoV-2 speci c responses across the entire cohort, by comparing the levels of cytokines in stimulated and unstimulated cells. We observed a signi cant induction (FC > 2; p < 0.05, Kruskal-Wallis One-Way ANOVA test) of IL-2, IL-5, IL-17A, IL-17F, IL-22, IFN-γ and TNFα in SARS-CoV-2 infected individuals compared to the uninfected controls (Supp Fig. 2A-B). Of note was the breadth of the cytokine response across SARS CoV-2 infected individuals, and activation of cytokines independent of the canonical antiviral cytokines (e.g. IL-5, IL-17A, IL-22). Among all individuals tested, there was generally a higher amplitude of responses induced by the S-derived peptide pool than by N-derived peptides. This response after restimulation is likely resulting from SARS CoV-2 speci c T-cells given the lack of responses in IL-2, TNF-α and IFN-γ we observe in spike restimulated, pre-pandemic PBMCs collected before 2020 (Supp. Figure 2C). Of note, the nucleoprotein peptide mix did include more total 15mer peptides compared to the spike mix (see methods). Conversely, IL-10 was found to be less speci c to infected individuals and additionally, found to be upregulated even in PBMCs from uninfected individuals (collected before 2020), suggesting that our approach does restimulate cross-reactive cells described previously 15 but that their response can be functionally separated from the responses associated with recent SARS-CoV2 infection (Supp Fig. 2D). We did not observe the induction of IL-4, IL-13 or IL-9 above background levels, and excluded them from further analyses.
To understand the development of SARS-CoV-2 speci c responses after infection while considering other factors that can in uence cytokine levels, we used a multivariate linear regression model which estimates the possible effects of age, sex, infection status and processing (fresh or cryopreserved cells) over cytokine induction (i.e. log fold change over unstimulated cells) ( Fig. 2A-B, Supp Fig. 2E, Methods). As expected, we observed a signi cant increase in IFNγ production in response to S-peptides and infection status (99% increase, FDR < 0.1, Methods). Perhaps surprisingly given the overall lower responses to Npeptides, we observed a greater number of associations between the cytokine inductions after restimulation with the N-peptide pool and the infection status of the donors. IL-2, IL-5, IL-17F, IFNγ and TNFα inductions were positively associated with the infection status in the N-peptides restimulated cells (308%, 65%, 39%, 304% and 40% increase when infected, respectively, FDR < 0.1, Fig. 2A-B, Supp Fig. 2E, Methods). This result suggest that a broader variety of cytokine responses, potentially corresponding to distinct T helper lineages, are induced by nucleoprotein derived antigens. Notably, in spite of their welldocumented in uence on disease severity and of the reported effect of age on the nature of immune responses in acute patients 16 , we did not observe any signi cant in uence of age and sex on the amplitude of cytokine responses, with the exception of a modest association between age and IL-5 production in response to spike antigens (2% increase, FDR < 0.1, Fig. 2A, Methods). From a technical perspective, the processing of the sample as fresh vs. cryopreserved (i.e. frozen) did impact cytokine production and frozen samples generally lowered responses, although the magnitude of this effect depends on the cytokine readout. For example, IL-2 and IL-6 levels were strongly affected (70% and 63% decrease for IL-2 in spike and nucleoprotein respectively, 73% decrease for IL-6 in spike, FDR < 0.1, Fig. 2A, Methods) while IFNγ, and IL-17A were not statistically different between freshly stimulated and previously frozen cells ( Fig. 2A, Methods), highlighting the importance of regressing this parameter out of downstream analyses.

Sex-speci c correlations between neutralizing antibody responses and T cell memory
Having characterized the scope of memory T cell responses against SARS-CoV-2, we next focused on the association between levels of neutralizing antibodies and the nature of anti-spike and anti-nucleoprotein T cell responses. Across all infected individuals, we noticed modest correlations between ID50 and cytokine levels for the large majority of the cytokines associated with infection status, as would be expected since ID50 values and multiple cytokine levels are both directly linked to the infection status of the subject (data now shown). To avoid this confounding factor and directly reveal whether speci c T cell responses are associated with the production of neutralizing antibodies during infection, we therefore focused only on infected subjects, excluding one individual who tested positive by PCR but showed no antibody or T cell responses above background. Strikingly, when this dataset was broken down across sex, we observed positive correlations between neutralizing titers and in ammatory cytokine production after nucleoprotein restimulation in male subjects (TNFα, R = 0.51, p = 0.022, and IFNγ, R = 0.33, p = 0.15), but these associations were not observed in female subjects, with only IL-2 showing a trend towards association (R = 0.35, p = 0.08) and in ammatory cytokines showing no association at all (TNFα, R = -0.098 and IFNγ, R = 0.048) (Fig. 2C). Similar trends were observed in an independent restimulation to a spike peptide pool, albeit to a lesser degree (Fig. 2D). As noted above, this was not caused by differences in overall neutralizing titers or cytokine production across men and women, since these were similar ( Fig. 1E and Fig. 2A), suggesting that the coupling of neutralizing antibody production and memory T cell generation occurs differently across males and females: in males speci cally, higher neutralizing antibodies appear to be associated with more type 1 responses, a phenomenon that is not observed in females. Interestingly, when we looked at associations between T cell cytokine responses and symptom duration, as a surrogate marker of disease severity in these clinically mild cases, the only signi cant association was detected in female subjects, between the levels of TNF elicited after spike stimulation and symptom duration (Supp. Figure 2F-G). Altogether, this suggests that differences in the orchestration of the immune response between male and female subjects can in uence the development of neutralizing antibodies, independently of the general strength of the T cell response or of disease course.

Enrichment and single cell analysis of SARS-CoV-2 responsive T cells
To follow up on this observation and try to de ne the molecular underpinnings of this difference in immune response orchestration, we decided to analyze the phenotype of SARS-CoV-2 speci c T cells by single cell RNA sequencing (scRNAseq). To be able to broadly analyze T cells across all HLA types to multiple SARS CoV-2 epitopes, we designed a strategy based on the isolation of activated T cells after restimulation with a pool of peptides from the spike protein and nucleoprotein. As a rst validation, we measured the upregulation of activation markers CD154 and OX-40 on CD4 + and CD8 + T-cells across 8 previously infected individuals after restimulation with spike and nucleoprotein pools, using ow cytometry (Fig. 3A). In previously infected donors, a signi cant fraction of cells expressed the activation markers CD154 and OX40, suggesting that these markers can be used to enrich SARS-CoV2-speci c Tcells (Fig. 3B). From these data, we also observed a higher percentage of CD4 + and CD8 + T-cells were activated upon re-stimulation with the spike protein compared to the nucleoprotein (Fig. 3B).
Using this sorting strategy, we activated and sorted CD8 + and CD4 + T cells activated with either spike or nucleoprotein peptide pools and performed 5' scRNAseq pairing transcriptome analysis and TCR sequencing. After quality control, we obtained 18,436 cell pro les from 17 subjects across all conditions, with 15,025 of these cells also having a paired fully identi ed TCR sequence (Supp Fig. 3A). To complement this dataset, we also extracted CD8 + and CD4 + T cells sequenced directly from PBMCs of the same subjects, representing an additional 10,141 cells with complete TCR sequences.
As a rst step, we performed Leiden clustering of the resulting dataset to identify the phenotypic pro les of the spike and nucleoprotein restimulated and the unstimulated PBMC cells (Methods). We identi ed 25 transcriptional clusters (TC) across all CD4 + and CD8 + T cells (Fig. 3C, Supp Fig. 3B, Methods). As expected, clusters derived from CD4 + and CD8 + T (Fig. 3D, Supp Fig. 3C, D) cells were generally well separated, although some clusters did contain cells from both populations (e.g. TCs 4 and 6, that likely represent a signature of cell stress characterized by an enrichment in ribosomal proteins and NEAT respectively, Supp Fig. 3B). Overall embedding structure of the CD4 + cells displayed the cells on the axis of variation representing the naïve to activated states differentiation (Supp Fig. 3E,F, Methods). Likewise, CD8 + cells were positioned based on their naïve to cytotoxic cell states (Supp Fig. 3G,H, Methods). Importantly, cells derived from unstimulated PBMCs clustered independently in a small number of subsets (Fig. 3E, Supp Fig. 3I,J), representing naïve CD4 + and CD8 + cells (Fig. 3C,E, TCs 0 and 5 respectively), T helper cells (TC 1), and two subsets annotated as cytotoxic T cells (TC 2) and a small cluster of CXCR6 + cells (TC 17). By contrast, cells derived from restimulated samples (Fig. 3E, spike in red, nucleoprotein in yellow, Supp Fig. 3I, J) showed a much higher phenotypic diversity that was divided across the remaining 20 clusters. The majority of clusters represented shared states observed across multiple subjects, but we also noted the presence of smaller donor-speci c clusters among restimulated cells (in particular TCs 11 and 20 derived from subject 257, Supp Fig. 3K, L). This outlier may re ect speci c states that arise during activation, and it is interesting to note that this population was recovered from both CD4 + and CD8 + T cells in this subject, but it may also be linked to the fact this subject was also an outlier in terms of age.
In order to functionally map these clusters, we extracted biological factors with non-negative matrix factorization (Methods), thus de ning 12 latent factors that generate the main axes of biological variation across these T cells (Supp Fig. 4A-W). These 12 factors mostly did not correlate with each other, possibly because they stand for independent processes that generate the overall transcriptome variation (Fig. 3F, max Pearson's R = 0.3, Methods). We then relied on the top genes in each factor to associate these factors with relevant facets of T cell biology (Supp Fig. 4W). For example, factor 10 was highly enriched in non-restimulated cells, and included SELL, IL7R, LEF1 and TCF7, indicative of a naïve or central memory T cell pro le (Supp Fig. 4R, S). More interestingly, factor 8 highlighted a population of cytokine producing cells, with top genes including IL2, IL21, CCL20 and IFNG as well as a large number of costimulatory molecules including TNFRSF18, 4 and 9 (GITR, OX40 and 4-1BB respectively) (Fig. 3G). A direct calculation of a general cytokine expression score highlighted the same populations (Fig. 3H, Supp  Fig. 4X, Methods). This factor largely mapped to CD4 + T cells restimulated with nucleoprotein peptide pools and was observed across multiple patients. Of note, cytokine genes are generally hard to capture in the context of scRNAseq of PBMCs because of their low expression in non-stimulated cells across multiple methods 17 , but their detection here is likely a re ection of our restimulation approach, since little to no non-stimulated PBMCs scored for this cytokine module.
Across CD8 + cells, factor 1 and 5 captured the bulk of the phenotypic variation observed, with factor 1, also enriched in non-restimulated cytotoxic T cells (Supp Fig. 4A,B), associated with an effector pro le (CD8A, CD8B, GNLY, NKG7, GZMK, GZMA) while factor 5 also included CD8A, CD8B and GNLY but did not include any granzymes (Supp Fig. 4I,J). It is also interesting to note that factor 5 encompassed PTPN22 and EOMES, which have both been associated with exhaustion phenotypes that may become more apparent during our restimulation assay (Supp Fig. 4Y). Altogether, our results highlight a broad diversity of T cell phenotypes that could only be uncovered after restimulation but are not directly apparent in PBMCs.
In particular, given the importance of cytokine responses in driving neutralizing antibody responses in a sex-speci c manner, we focused on the clusters associated with high cytokine production (TCs 16, 18 and 22) and examined sex-speci c differences in gene expression in these cells, as a majority of these cytokine producing cells overlapped with cells enriched in factor 8. As a control, we performed a similar analysis in resting PBMCs from the same subjects, overall capturing a broad diversity of male and female donors in our resting (TCs 0, 1) and activated (TCs 16, 18, 22) clusters (Fig. 3I, Supp Fig. 5A). In the activated group of cells (representing 9 female and 7 male subjects, Fig. 3I, Supp Fig. 5A), we observed a strong enrichment of type I IFN-associated pathways (Supp Fig. 5B) and genes (Fig. 3J) in cells derived from male donors, with IFITM3, ISG15, MX1, IFITM1, OAS1 and ISG20 all upregulated in these cells (FDR < 0.1, log2 fold change > 0.4). We also observed an enrichment of other immune associated genes, including GZMA, CD52, LY6E, CXCR6 and ICAM3, altogether suggesting that cytokine producing CD4 + T cells have a distinct gene signature in male subjects. Interestingly, when we performed a similar male/female comparison across resting cells, the number of differentially expressed genes in this comparison was more limited and did not include clear biological pathways. For example, when looking at type I IFN associated genes, we only observed 2 DE genes, with IFIT1 was upregulated in naïve cells from female subjects and IFITM3 upregulated in naïve cells from males, thus demonstrating that the enrichment of a type I IFN signature in males is only evident in activated but not naïve T cells.

Repertoire analysis of SARS-CoV-2 responsive T cells
Another key advantage of analyzing resting and activated T cells simultaneously is to enrich for spike or nucleoprotein TCRs that are likely to be present at a low frequency across all peripheral T cells but enriched by the sorting of spike or nucleoprotein-activated cells. De ning a clonotype based on a common beta chain (including identical VDJ segment usage and CDR3 sequence), we identi ed several thousand clonotypes across CD4 + and CD8 + T cell populations (Fig. 4A, Methods). CD8 + T cells demonstrated a strikingly stronger clonal expansion than CD4 + T cells, with 25-50% of clones being expanded depending on the condition, versus < 10% for CD4 + T cells. To measure diversity among T cell populations, we then compared the Shannon entropy of PBMCs and restimulated cells and, as expected, observed a signi cantly lower Shannon entropy for restimulated CD8 + T cells, con rming the signi cant enrichment of a more restricted pool of clones after restimulation and sorting (Fig. 4B). While there was a trend towards a lower Shannon entropy in restimulated CD4 + T cells as well, this did not reach signi cance, most likely because of the dominance of single-representative clones in our CD4 + T cell data.
Upon closer inspection of the TCR repertoire, we did not observe a signi cant difference between the restimulated samples and the PBMC samples for beta chain V and J gene usage (Sup Fig. 6A,B,C,D) and alpha chain J gene usage (Sup Fig. 6E, F). However, we detected a signi cant increase in TRAV3 and TRAV8 alpha chain V gene usage in both spike and nucleoprotein stimulated cells compared to PBMCs, exclusively in CD8 + T cells (Sup Fig. 6G, H). Given the abundance of expanded CD8 + T cell clones, we decided to focus on the repertoire composition of this population.
The transcriptomic landscape of the CD8 + cells belonging to the expanded clones (size > 20) associated with a more activated and cytotoxic CD8 + pro le in all spike, nucleoprotein and PBMC samples, even though the cells formed subclusters based on their sample type (Fig. 4C). We next looked more speci cally at the distribution of clones across the PBMCs, spike and nucleoprotein-restimulated samples, with the assumption that the selection of cells activated by a given peptide pool should enrich for clones speci c for SARS-CoV2 antigens, and that these clones might also be present at a low frequency in the total PBMCs of convalescent subjects. Indeed, the majority of cells that belonged to expanded clones were associated with the nucleoprotein restimulation condition, with a small degree of overlap with other samples (Fig. 4D). This suggests that our approach has the potential to reveal antigen speci c clones by enriching them compared to total PBMC T cells, including clones that are directly associated with cytokine-producing clusters (Fig. 4E).

Discussion
The analysis of immune responses of individuals naturally exposed to SARS-CoV2 has provided key insights into the development of protective immunity against this virus, and about responses to novel pathogens more generally. Here, we focused on a cohort of individuals that were infected in a similar timeframe through a single spreading event and developed mild disease. In this well-de ned group, we focused on the establishment of neutralizing antibody responses and T cell memory. Our results demonstrate a large range of neutralization activity that is not correlated with age and gender. We therefore examined T cell responses as a potential correlate of antibody development, and similarly observed a large diversity of both the magnitude and nature of the response, with IFN-γ generally dominating the response to spike peptides but IL-2, IL-5, IL-17F and TNFa all elicited at signi cant levels in response to nucleoprotein peptides. Surprisingly, we observed a robust positive association between TNF-responses and neutralizing activity only in male subjects, and no signi cant association in female subjects.
Sex differences in immune responses have been investigated in many contexts, with female subjects displaying generally higher innate and adaptive immune responses but a higher prevalence of autoimmune disease 18 . Infectious diseases display a more mixed sex-based susceptibility depending on pathogen, but in the case of COVID-19 speci cally, the higher susceptibility of male subjects has been well documented 19 . The analysis of immunological parameters in hospitalized subjects has revealed a number of sex-speci c differences in in ammatory cytokines and T cell responses 20 , although this was not seen in other immunopro ling studies 4 and whether these alterations are directly related to sex or a simply a consequence of differing disease courses has been subject to debate 21 . Of note, multiple non-exclusive mechanisms including differences in mucosa-associated invariant T (MAIT) cells 22 and kynurenic acid levels 23 have been proposed to explain these differences in susceptibility during acute infection. How these differences translate to the development of memory has remained poorly explored. In our relatively homogeneous cohort (recovered subjects, 6-10 weeks post-mild disease), and after regressing out age, it is interesting to note that we did not observe any signi cant differences in the magnitude of B and T cell responses, but rather in the orchestration of B-T collaboration during the development of neutralizing antibodies. Long term studies of vaccinated subjects have shown similar antibody titers but lower neutralization activity in males, both early on and 6 months after vaccination 24 . It is interesting to consider that this may be linked to a distinct contribution of Tfh-dependent and independent B cells, with the latter potentially yielding more diverse but less matured and persistent antibodies in a Tfh-de cient mouse model 25 .
Single-cell RNA sequencing can identify a large array of T cell receptors and has been employed in the case of SARS-CoV2 to study broad changes in the repertoire 26 , but provides limited insight into T cell speci cities. By contrast, approaches that rely on the enrichment of reactive cells can more directly assess speci c responses, without requiring epitope-speci c knowledge or tetramers of a matching HLA, as has been performed in acute COVID-19 cohorts 27 . Here, we leveraged this approach to de ne both the pro le and repertoire of spike and nucleoprotein reactive cells. We were particularly interested in a cluster enriched for cytokine and chemokine production, and identi ed both sex-and activation status-speci c differences within these T cells, suggesting that differences in type I IFN signaling may underlie the sexspeci c impact of T cells on neutralizing antibodies. This approach also allowed us to de ne speci c TCRs associated with this population of cytokine producing cells. It is important to note that these studies were performed on subjects infected with the original Wuhan strain, but multiple studies suggest that novel variants, including Omicron, are generally well-recognized by T cells elicited by a previous strain 28- 30 , even in the context of antibody escape.
Altogether, the combination of approaches used sheds light on sex-speci c differences in the immune responses to natural SARS-CoV2 infection and uncovers potential mechanistic factors involved. While we focused here on COVID-19, we expect that some of these differences in the orchestration of immune responses may apply to other infectious diseases. A deeper understanding of these processes might both inform vaccine development and provide new clues to the impact of sex on immunity.

Declarations Code and Data Availability
Raw sequencing les will be made available through the DUOS portal upon publication.

Human Samples
Whole blood (collected in BD Vacutainer K2EDTA tubes) and serum (collected in BD Vacutainer serum tubes) samples from consented subjects recovered from COVID were collected in EDTA tubes at the Brigham and Women's Hospital, Boston, MA, USA. Results from PCR testing when available, disease and demographic information (Table 1) were collected after blood draws through a RedCap-administered survey. All studies were performed under the IRB protocol number 2020P000849 "Biorepository for Samples from those at increased risk for or infected with SARS-CoV-2." approved at the Brigham and Women's Hospital, Boston, MA, USA.

Serology and neutralization assays
Serum samples were aliquoted from serum tubes and aliquoted for storage at -80ºC. Serology was performed at the Broad Institute as described previously 31 . Brie y, MaxiSorp 384-well plates were precoated with 50uL/well of 2.5ug/mL SARS-CoV2 RBD then incubated with 50μL of 1:100 diluted serum samples for 30 min at 37ºC. After washing, HRP-anti human IgG and IgM (1:25000 dilution, Bethyl Laboratory) was added to each well for 30 min at RT, before washing and incubation with 40 μl/well of Pierce TMB peroxidase substrate (ThermoFisher). The reaction was stopped by adding 40 μl/well of 0.5 M H2SO4 and the OD was read at 450 and 570 nm on a BioTek Synergy HT.
For neutralization assays with pseudotyped viruses 8,32 , HEK293 cells were cotransfected with psPAX2 (AIDS Resource and Reagent Program), pLenti-CMV Puro-Luc (Addgene), and spike protein expressing pcDNA3.1-SARS CoV-2 SΔCT sequences for the Wuhan strain. Supernatant was collected 48 h posttransfection and mixed with 3-fold serial dilutions of heat-inactivated serum samples. After a 1-hour incubation at 37ºC, mixes were used to infect HEK293T-hACE2 cells seeded in 96-well plates at a density of 1.75 × 10 4 cells per well the previous night. After 48h, cells were lysed in Steady-Glo luciferase assay (Promega) and neutralization titers were de ned as the sample dilution at which a 50% or 80% reduction in relative light unit was observed relative to the average of the virus control wells.
For neutralization assays with live SARS-CoV2 virus 31 , serially diluted patient sera samples were mixed with diluted SARS-CoV-2 live virus (D614) and incubated at 37°C for 1 h. Mixes were used to infect Vero E6-TMPRSS2 cells seeded the day prior at 10,000 cells per well in CellCarrier-384-well microplates. After 48h of culture, cells were xed with 4% PFA for 2h, washed and incubated with a mouse anti-SARS-CoV2 NP antibody (Sino Biological) for 1.5h at RT, followed by Alexa488-conjugated goat-anti-mouse antibody (Jackson ImmunoResearch Labs) for 45 min at RT and nuclear staining with Hoechst 33342 (Thermo Fisher Scienti c). Fluorescence imaging was performed using the Opera Phenix™ High Content Screening System (Perkin Elmer) and half-maximal inhibitory dilutions (ID50) were determined using a fourparameter, nonlinear curve tting algorithm, with a total range of 20 to 10,240 (for samples where the minimal dilution did not achieve 50% neutralization or where the maximal dilution exceeded 50% neutralization, respectively).

PBMC isolation and T-cell restimulation
PBMCs were isolated within 3 hours of blood draw from subjects by density gradient separation using Single cell RNA sequencing Cells were separated into droplet emulsions using the Chromium Next GEM Single-cell 5′ Solution (v1.1) and the 10x Chromium Controller. 10,000 cells were loaded per channel of the Chromium Next GEM  cytokines containing IL2, IL5, IL6, IL9, IL10, IL17A, IL17F, IL22, IFNg, TNF, let Y k_spike , Y k_nucleo , and Y k_control stand for the measurement levels (in log2 scale) of k in spike stimulated, nucleoprotein stimulated and unstimulated control conditions. We assessed the associations between subject age, sex (male or female), infection status (yes or no), and sample processing (fresh or Page 16/28 frozen) with fold changes in the level of k upon spike (Y k_spike -Y k_control ) and nucleo (Y k_nucleo -Y k_control ) stimulations with two separate multivariate linear regression models: Y k_spike -Y k_control = 0 + 1 Age + 2 Sex + 3 Processing + 4 Infected Y k_nucleo -Y k_control = 0 + 1 Age + 2 Sex + 3 Processing + 4 Infected

scRNA-seq analysis
For the mRNA data integration, count normalization, dimensionality reduction, clustering, cell scoring, and cluster marker genes detection Seurat R package 35,36 was employed. Batch effects, de ned as the batch of the sequencing run, were regressed out from the normalized count values with the ComBat algorithm 37 implemented in SVA R Package version 3.38.0.
Preprocessing Cells which either do not have 10x standard high-quality heavy and light chain V(D)J sequences, or have more than 10% of their transcriptome reads coming from mitochondrial genes were ltered out before the downstream transcriptome analysis. Sets of TRAV, TRAJ, TRBV, TRBJ genes were discarded in all of the downstream analyses.
For the UMI count normalization step, gene expression counts for each cell were divided by the total counts for that cell and multiplied by 10 6 , which was then log-transformed using log1p.
Dimensionality reduction, graph clustering, and UMAP visualization Dimensionality reduction was done with PCA identifying the rst 50 principal components. For clustering of the cells into expression clusters, a k-nearest neighbor (kNN) graph of the cells was constructed (k=20) using the 50 principal components. Then the Leiden algorithm 38 was used to nd the clusters of the cells based on the generated kNN graph, with a resolution of 0.1. Expression levels of immunoglobulin genes were discarded during the clustering step. Uniform manifold approximation and projection (UMAP) 39 algorithm was run on the rst 50 principal factors to obtain the 2D projections of the cells.

Scoring gene set signatures
Cell speci c expression scores of the cell cycle genes and the cytokines were calculated as previously de ned in 40 , where for each gene in the gene-set, 100 genes were randomly selected as control genes.

Non-negative matrix factorization
The integrated single cell count matrix was non-negative matrix factorized using the Python package Consensus Non-negative Matrix Factorization (cNMF) 41 . The number of high variance genes used to run the factorization was set to 3000, and the NMF loss function was frobenius. The optimal number of latent factors was chosen based on the identi ability of the independent pathways as determined by gene set enrichment analysis. The genes with signi cantly high loadings per factor had loadings greater than three interquartile ranges above the 75th percentile. The clusterPro ler R package was used to perform gene set enrichment of the selected genes 42 .

Pseudotime Analysis
Pseudotime analyses of the CD4+ and CD8+ T cell populations were performed using the R implementation of Monocle3 43 . Root cells were selected based on the transcriptome analyses of the Leiden clusters.

TCR repertoire analysis
While generating the assembled V(D)J sequences, the 10x Genomics V(D)J contig assembly algorithm (https://support.10xgenomics.com/single-cell-vdj/software/pipelines/latest/algorithms/assembly) accounts for many types of noise speci c to scRNA-seq data. Nonetheless, only cells with high-quality V(D)J contig sequences in the beta and alpha chains were chosen. The downstream repertoire analysis excluded cells with more than one high-quality alpha or beta chain sequence (i.e., double expressors).
Cells having the same beta chain V and J genes as well as CDR3 nucleotide sequences were assumed to be clonally related. A small fraction of clones contained representatives in both CD4+ and CD8+ populations which might have represented contamination events, in those cases the clone was assigned to the dominant population for further analyses. Tables   Table 1 is in the supplementary le. Figure 1 Heterogenous development of antibody responses in a mild COVID-19 cohort A) Experimental approach. 85 subjects were recruited from a single spreading event, and PBMCs and serum samples were collected for analyses of antibody responses, T cell activation assays and single-cell RNAseq. B) ID50 levels in an authentic SARS-CoV2 neutralization assay for each subject. Infection status was evaluated based on a PCR positive test (plus signs) or a negative test / absence of test (circles), as well as serology (red indicates a positive result for IgG spike, blue indicates negative). C) Scatterplots displaying the correlations between different neutralization assays, using a pseudotyped virus (ID50_pseudo and ID80_pseudo for titers neutralizing 50% or 80% of infection) or authentic SARS-CoV2 (ID50). Diagonal plots display the density of the values. D) Boxplots displaying the distribution of the ID50_pseudo, ID80_pseudo and ID50 neutralization titer values of the subjects strati ed by their sex and infection status. The boxes represent -2 standard deviation (lower portion), mean (black line), and + 2 standard deviation (upper portion). The values above each violin plot represent the median values of the distribution. Brackets indicate statistical signi cance using a one-sided t-test with **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001 and ns=non signi cant. E) ID50 neutralizing titers against authentic SARS-CoV2 across male (purple) and female (green) subjects. There is no signi cant difference between the means of the two groups based on a two-sided t-test.   TCR repertoire analysis across steady state and spike or nucleoprotein restimulated T cells A) Pie charts displaying the expansion of differently sized CD4+ and CD8+ T cell clones for samples grouped based on their sample type (spike stimulated, nucleoprotein stimulated and PBMC). Numbers in the center of the pie charts stand for the total number of cells of that particular group. A slice in the pie chart with a speci c color displays the percentage of cells belonging to clones of size corresponding to this color.