mRNA vaccine-induced T cells respond identically to SARS-CoV-2 variants of concern but differ in longevity and homing properties depending on prior infection status

To characterize the phenotypic features of mRNA vaccination-elicited SARS-CoV-2-specific T cells, we procured 33 longitudinal blood samples from the COVID-19 Host Immune Response and Pathogenesis (CHIRP) cohort. Four of the participants had received the Moderna (mRNA-1273) vaccine, while the remaining seven had received the Pfizer/BioNTech (BNT162b2) one. For all participants, longitudinal specimens were obtained at three timepoints: prior to vaccination, ~ 2 weeks (range 13–18 days) after the first vaccine dose, and ~2 weeks (range 6–38 days) after the second dose. Five of the participants were never infected with SARS-CoV-2, while the remaining six had completely recovered from mild (non-hospitalized) COVID-19 disease (Table 1). These prior infections had all occurred in the San Francisco Bay Area between March and July of 2020, when the dominant local strain was the original ancestral strain. Each specimen was phenotyped using a 39-parameter T cell-centric CyTOF panel (see Materials and methods and Table 2) at baseline (to establish the overall phenotypes of total CD4+ and CD8+ T cells), and following 6 hr of stimulation with overlapping 15-mer peptides spanning the entire original (ancestral) SARS-CoV-2-spike, B.1.1.7 spike, B.1.351 spike, or the original SARS-CoV-2 nucleocapsid (the latter as a control for a SARS-CoV-2-specific response not boosted by vaccination). Including all the baseline and stimulation conditions, a total of 165 specimens from the 11 participants were analyzed by CyTOF.

We first confirmed our ability to identify SARS-CoV-2-specific T cells by stimulating PBMCs from vaccinated individuals with spike peptides. In line with our prior studies implementing a 6 hr peptide stimulation Neidleman et al., 2021; Neidleman et al., 2020b, spike-specific CD4+ T cells could be specifically identified through intracellular cytokine staining for IFNγ, and a more robust response was observed among CD4+ than CD8+ T cells (Figure 1 ). No specific induction of IL4 or IL17 by CD4+ T cells were observed in response to peptide stimulation (Figure 1—figure supplement 1). In addition, activation-induced markers (AIM) such as Ox40, 4-1BB, and CD69 could also be identified in T cells after spike peptide stimulation, but with a higher background in the baseline (no peptide stimulation) specimens relative to the intracellular cytokine staining approach (Figure 1—figure supplement 1). For these reasons, in this study we exclusively used IFNγ positivity in the peptide-stimulated samples as a marker of antigen-specific T cells.

In the infection-naïve participants, the first vaccination dose primed a spike-specific CD4+ T cell response, which was further boosted with the second dose (Figure 1B, top left). For each participant and time point, similar numbers of cells were stimulated by exposure to the ancestral or variant spikes. This finding suggests that vaccine-elicited spike-specific CD4+ T cells recognize ancestral and variant spike equally well, and is consistent with their recently reported ability to recognize variant strains Tarke et al., 2021. The response of vaccine-elicited CD8+ T cells to spike peptides was weaker, and mostly apparent only after the second dose (Figure 1B, top right). As expected, vaccination did not elicit T cells able to respond to nucleocapsid peptides (Figure 1C, top panels).

In contrast to the infection-naïve individuals where spike-specific CD4+ T cells were clearly elicited and then boosted upon the second dose, spike-specific CD4+ T cell responses in convalescent individuals did not show a consistent upward trend. Convalescent donor PID4112 had a large frequency of pre-vaccination SARS-CoV-2-specific CD4+ T cells that increased to >1% of the total CD4+ T cell frequency after the first dose and then dampened after dose 2 (Figure 1B , bottom left). PID4112 also exhibited an elevated nucleocapsid-specific CD4+ T cell response after the first vaccination dose (Figure 1C, bottom left), which may have been due to bystander effects resulting from the concomitant large spike-specific response. In comparison, PID4112’s spike-specific CD8+ T cell response was low after dose 1, and boosted after dose 2 (Figure 1B, bottom right). In contrast to PID4112, the remaining five convalescent donors exhibited an overall weak spike-specific T cell response. In fact, when comparing these five donors to the five infection-naïve donors, there was a significant decrease in the magnitude of the spike-specific CD4+ T cell response, while the spike-specific CD8+ T cell response was equivalent between the two groups (Figure 1D). These results were unexpected and suggest that, when excluding outlier PID4112, the magnitude of the vaccine-elicited spike-specific CD4+ T cell response (after full vaccination) was lower in convalescent individuals than in infection-naïve individuals.

These assessments of the magnitude of the spike-specific T cell response together suggest that (1) in infection-naïve individuals the CD4+ T cell response is boosted by the second vaccination dose, (2) convalescent individuals exhibit a more disparate response, with most donors mounting a weaker response than infection-naïve individuals, and (3) the response is more robust among CD4+ than CD8+ T cells. As a higher number of SARS-CoV-2-specific CD4+ T cells were available for analysis, we focused on this subset for our subsequent analyses.

Leveraging our ability to not only assess the magnitude but also the detailed (39-parameter) phenotypic features of SARS-CoV-2-specific CD4+ T cells, we first determined whether the ancestral and variant spike epitopes stimulated different subsets of vaccine-elicited spike-specific CD4+ T cells. Such differences could theoretically result from the fact that ~5–10% of the spike epitopes differ between variants and ancestral strains Tarke et al., 2021, and may therefore act as APLs steering responding cells towards different fates. We isolated the datasets corresponding to both post-vaccination timepoints for all eleven donors, and then exported the data corresponding to spike-specific CD4+ T cells (as defined by IFNγ production, Figure 1 ). After reducing the multidimensional single-cell data for each individual specimen to a two-dimensional datapoint through multidimensional scaling (MDS) Ritchie et al., 2015, we observed the ancestral spike-stimulated samples to be interspersed among the B.1.1.7- and B.1.351-responding ones (Figure 2A). We then visualized the spike-specific CD4+ T cells at the single-cell level. When visualized alongside total (baseline) CD4+ T cells, spike-specific cells occupied a distinct ‘island’ defined by high expression of IFNγ (Figure 2B), suggesting unique phenotypic features of these cells. To better analyze these spike-responding CD4+ T cells, we visualized them in isolation within a new tSNE which clearly demonstrated complete mixing of the cells stimulated by the ancestral, B.1.1.7, and B.1.351 spike proteins (Figure 2C). Almost all the responding cells expressed high levels of CD45RO and low levels of CD45RA (Figure 2D), suggesting them to be mostly memory cells. These memory CD4+ T cells included central memory T cells (Tcm), T follicular helper cells (Tfh), and those expressing multiple activation markers (CD38, HLADR, CD69, CD25) and receptors known to direct cells to tissues including the respiratory tract (CXCR4, CCR5, CCR6, CD49d) (Figure 2E). The expression levels of these and all other antigens quantitated by CyTOF were not statistically different between CD4+ T cells responding to the three spike proteins (Figure 2—figure supplement 1). To confirm the identical phenotypes of the three groups of spike-responding cells, we implemented unbiased clustering by FlowSOM. Spike-stimulated cells were clustered into eight subsets, and no subset was preferentially enriched in any one of the three groups (Figure 2F). Together, these data suggest that vaccine-elicited spike-specific CD4+ T cells respond in the same manner to spike epitopes from the ancestral or variant strains, and would probably mount similar responses in vivo to infection by all three virus types.

We next took advantage of our longitudinal study design to assess for any changes in the differentiation of spike-specific T cell responses over the course of the 2-dose vaccination. As the data presented above suggested no phenotypic differences between CD4+ T cells responding to the ancestral, B.1.1.7, and B.1.351 spike proteins, our subsequent analyses combined these datasets. We first assessed whether, among infection-naïve individuals, the phenotypes of spike-specific CD4+ T cells were different after the first and second doses. While MDS and tSNE visualizations of the data revealed that the cells from the two timepoints were somewhat interspersed (Figure 3A-B), FlowSOM clustering suggested some differences in cluster distribution (Figure 3C-D). Direct comparison of the cluster frequencies revealed a cluster (B8) significantly enriched after the first dose, and a different cluster (B5) significantly enriched after the second dose (Figure 3E). As these two clusters differentially expressed the Tcm markers CD27 and CCR7 (Figure 3F), we then assessed whether Tcm cells were differentially represented among spike-specific CD4+ T cells after each of the vaccination doses. Indeed, Tcm cells were significantly higher after the first dose (Figure 3G), consistent with Cluster 8 (enriched after the first dose) expressing high levels of these two receptors. Assessment of other canonical CD4+ T cell subsets – in particular naïve (Tn), stem cell memory (Tscm), effector memory RA (Temra), effector memory (Tem), T transitional memory (Ttm), Tfh, and regulatory T cells (Treg) – revealed Tn cells, like the Tcm subset, to be decreased after the second dose. By contrast, Ttm cells were found to be higher after the second dose, while the remaining subsets were not altered (Figure 3G-H). Overall, Tcm and Tfh were the most abundant subsets among the spike-specific CD4+ T cells (Figure 3G-H). These data together suggest that after receiving the second dose, infection-naïve individuals’ spike-specific CD4+ T cells increase in quantity (Figure 1B), and alter their phenotypes as reflected by a decrease Tcm cells and an increase in Ttm cells.

We then conducted a similar analysis in the convalescent individuals. As the pre-vaccination timepoint included spike-specific CD4+ T cells primed by prior SARS-CoV-2 infection, we included all three timepoints in this analysis. When the data were visualized by MDS, it was apparent that most of the pre-vaccination specimens localized away from the post-vaccination specimens, which were interspersed with each other (Figure 4A). Similar distinctions between pre-and post-vaccination specimens were visualized at the single-cell level by tSNE, which was particularly apparent when visualized as contour heatmaps (Figure 4B and C). Clustering of the cells by FlowSOM revealed that the cluster distribution was markedly skewed among the pre-vaccination cells (Figure 4D and E), with one cluster being under-represented (C2) and one over-represented (C5) as compared to both post-vaccination timepoints (Figure 4F). Cluster C3 was the only cluster that was significantly different after 1 vs 2 doses (Figure 4F) but as this cluster comprised only <5 % of the cells it was not analyzed further. To assess what may drive the differences between the phenotypes of the pre- vs. post-vaccination spike-specific CD4+ T cells, we assessed for markers differentially expressed between clusters C2 and C5. Cluster C2 cells preferentially expressed the Tcm markers CD27 and CCR7, the Tfh markers PD1 and CXCR5, and the co-stimulatory receptors ICOS and Ox40, while among these only CD27 was preferentially expressed in Cluster C5 (Figure 4—figure supplement 1). Manual gating confirmed Tcm, Tfh, and ICOS+ Ox40+ cells to be preferentially enriched in the post-vaccination specimens (Figure 4G-I). None of the canonical subsets were differentially abundant after the first vs. second vaccination dose. Together, these results suggest that, in contrast to the infection-naïve individuals, convalescents’ spike-specific CD4+ T cells were similar after the first vs. second vaccination dose; however, in these individuals vaccination drastically altered the phenotypes of the pre-existing spike-specific CD4+ T cells (presumably elicited from the original infection).

We next determined whether there were any phenotypic differences between the vaccine-induced spike-specific CD4+ T cells from the infection-naïve vs. convalescent individuals. Removal of convalescent outlier PID4112 revealed the magnitude of the spike-specific CD4+ T cell response to be lower in the convalescents than in infection-naïve participants after full vaccination (Figure 1D). But when all donors were included there was no statistically significant difference in response magnitude (Figure 5A). However, the spike-specific CD4+ T cells from the convalescent and infection-naïve individuals exhibited clear phenotypic differences when assessed by both MDS (Figure 5B) and tSNE contours (Figure 5C); this was more apparent after the second vaccine dose, but could already be observed after the first. Since the cells after the second dose are more clinically relevant (as they are the ones persisting in vaccinated individuals moving forward), we focused our subsequent analysis on just this timepoint. When visualized as a dot plot, it was apparent that the spike-specific CD4+ T cells from infection-naïve individuals segregated away from those from the convalescents (Figure 5D). Clustering of the data also demonstrated differences between the two patient groups (Figure 5E and F), which was confirmed by demonstration of a significant difference in Cluster A1 abundance (Figure 5G).

To identify these phenotypic differences, we first assessed the relative distributions of the main canonical CD4+ T cell subsets. Interestingly, the vaccinated convalescents harbored significantly more spike-specific Tcm and Tn, and less spike-specific Ttm (Figure 6A). By contrast, Tfh and Treg frequencies were not different between infection-naïve and convalescent vaccinees (Figure 6B). To broaden our analysis, we assessed for unique features of Cluster A1, which was over-represented in the infection-naïve donors, and Cluster A3, an abundant cluster which was over-represented in the convalescent donors albeit insignificantly (Figure 5G). Interestingly, Cluster A1 expressed low levels of CD127, CXCR4, and CCR7 in contrast to Cluster A3 (Figure 5—figure supplement 1A). As Cluster A1 is enriched among the infection-naïve individuals, these findings suggest that these three receptors may be expressed at lower levels on the cells from these individuals, relative to those from vaccinated convalescents. This was confirmed by our detection of higher expression of CD127, CXCR4, and CCR7 on spike-specific CD4+ T cells from the convalescents, although for CXCR4 the difference did not reach statistical significance (Figure 5—figure supplement 1B).

We then followed up on each of these three differentially expressed markers. CD127, the alpha chain of the IL7 receptor, can drive IL7-mediated homeostatic proliferation of SARS-CoV-2-specific CD4+ T cells Neidleman et al., 2020b, and serves as a marker of long-lived precursor memory cells Kaech et al., 2003. To assess the potential longevity of the spike-specific CD4+ T cells, we determined the percentage of CD127+ cells expressing low levels of the terminal differentiation marker CD57. After the second dose of vaccination, convalescent individuals harbored more long-lived (CD127+ CD57-) spike-specific CD4+ T cells than infection-naïve individuals (Figure 6C). CXCR4, the second preferentially-expressed marker among the convalescents’ spike-specific CD4+ T cells, was recently suggested to direct bystander T cells to the lung during COVID-19, and to be co-expressed with the T resident memory / activation marker CD69 Neidleman et al., 2021. Interestingly, spike-specific CD4+ T cells from convalescent donors harbored a highly significantly elevated proportion of CXCR4+ CD69+ cells (Figure 6D), suggesting a potentially superior ability to migrate into pulmonary tissues. The last differentially expressed antigen, CCR7, is a chemokine receptor that directs immune cells to lymph nodes. As CD62L, a selectin that also mediates lymph node homing, was also on our panel, we assessed whether CCR7+ CD62L+ cells were enriched among the spike-specific CD4+ T cells from the convalescent donors, and found this to be the case (Figure 6E).

Our finding that the convalescent donors’ spike-specific CD4+ T cells were preferentially CXCR4+ CD69+ and CCR7+ CD62L + suggested that they may preferentially migrate out of the blood into lymphoid tissues. Supporting this possibility was our observation that, after the second vaccine dose, the percentages of CCR7+ CD62L + spike-specific cells increased as the percentages of spike-specific CD4+ T cells decreased (Figure 6F). This suggests that the low spike-specific CD4+ T cell response after the second dose of vaccination in some convalescent donors (Figure 1D) may have resulted from these cells preferentially leaving the blood compartment. This was further supported by our finding that the expression levels of CCR7 and CD62L on spike-specific CD4+ T cells inversely correlated with the magnitude of the spike-specific CD4+ T cell response (Figure 6G). To assess whether the CCR7+ CD62L + and CXCR4+ CD69+ CD4+ T cells have the potential to migrate into the nasopharynx, the most common site of SARS-CoV-2 entry, we obtained paired blood and nasal swabs from one of the participants (PID4101) and phenotyped total CD4+ T cells isolated from these specimens. There was a marked enrichment of both CCR7+ CD62L + and CXCR4+ CD69+ CD4+ T cells in the intranasal specimens (Figure 6H), suggesting that CD4+ T cells expressing these markers preferentially exit the blood and enter the nasopharynx. Together, these data suggest that after vaccination, spike-specific CD4+ T cells from convalescent individuals differ from those in infection-naïve individuals in that they appear to be more long-lived, and may more readily migrate out of the blood to mucosal sites, thus explaining their overall lower frequencies measured from the blood.

Finally, we assessed to what extent the main similarities and differences observed with spike-specific CD4+ T cells were also seen for spike-specific CD8+ T cells. Similar to the CD4+ T cells, spike-specific CD8+ T cells stimulated by the three different spike proteins (ancestral, B.1.1.7, B.1.351) did not differ in their phenotypic features (Figure 6—figure supplement 1A-C). Also similar to the CD4+ T cells, spike-specific CD8+ T cells elicited by vaccination differed phenotypically in the infection-naïve vs. convalescent individuals (Figure 6—figure supplement 1D-F). Unlike the CD4+ T cell data, however, these phenotypic differences could not be accounted for by distribution changes among the main canonical subsets Tn, Tscm, Temra, Tcm, Tem, and Ttm (Figure 6—figure supplement 1G). Also unlike the CD4+ T cells, these differences were also not explained by differential abundance of the CD127+ CD57-, CXCR4+ CD69+, or CCR7+ CD62L + subsets (Figure 6—figure supplement 1H). Instead, the differences appear to be due to other phenotypic changes, including elevated frequencies of activated cells in the convalescent donors, in particular those co-expressing the Tcm marker CD27 and activation marker CD38, and the checkpoint inhibitor molecule CTLA4 and activation marker 4-1BB (Figure 6—figure supplement 1I). These results suggest that vaccine-elicited spike-specific CD8+ T cells, like their CD4+ counterparts, respond equivalently to the B.1.1.7 and B.1.351 variants, and exhibit qualitative differences in convalescent individuals but via different phenotypic alterations than their CD4+ counterparts.

As this study was aimed at using in-depth phenotyping as a discovery tool, it focused on deeply interrogating many different conditions (e.g. spike variants, longitudinal sampling) rather than many donors. Therefore, although a total of 165 CyTOF specimens were run, only 11 donors were analyzed. The main findings reported here should be confirmed in larger cohorts using more cost-effective and high-throughput alternatives to CyTOF such as conventional flow cytometry. Such follow-up studies should also examine the functional outcomes of the discoveries made here (e.g. effect of chemokine receptor expression on homing of infection- and vaccine-elicited SARS-CoV-2-specific T cells), including in animal models of SARS-CoV-2 infection. A second limitation of the study was the need to stimulate the specimens in order to identify and characterize the vaccine-elicited T cells, and our limiting our analyses of SARS-CoV-2-specific T cells to those inducing IFNγ, which may have restricted our ability to characterize subsets such as Tfh cells that are relatively poor producers of this cytokine. We note however that we limited peptide exposure to 6 hours to minimize phenotypic changes caused by the stimulation, similar to our prior studies Neidleman et al., 2021; Neidleman et al., 2020b. Our analysis focused on CD4+ T cells because the overall numbers of detectable spike-specific CD8+ T cells were low. Nonetheless, the main findings we made with the CD4+ T cells – that they recognize variants equivalently, and that the phenotypes of the responding cells differ by prior SARS-CoV-2 natural infection status – were recapitulated among CD8+ T cells. Future studies should assess the phenotypes of non-stimulated, vaccine-elicited SARS-CoV-2-specific T cells using peptide-MHC tetramers/multimers. Such studies, however, would be limited to analyzing responses against a small number of epitopes, although use of combinatorial tetramers in conjunction with high-parameter phenotyping Newell et al., 2013 would increase the ability to characterize a larger breadth of the vaccine-elicited T cell response. Such studies however would be limited for the most part to CD8+ T cells as tetramer reagents for CD4+ T cells are less robust. A final limitation is that serological analyses were not performed in this study. As coordination between the humoral and cellular arms of immunity are likely key to effectively controlling viral replication, future studies should assess to what extent the breadth, isotypes, and functional features of spike-specific antibodies elicited by vaccination associate with the herein described phenotypic features of vaccine-elicited SARS-CoV-2-specific T cells.

Eleven participants from the COVID-19 Host Immune Pathogenesis (CHIRP) cohort were recruited for this study. Six were previously infected with SARS-CoV-2 as established by RT-PCR, and had fully recovered from a mild course of disease. Importantly, infections of these six individuals had all occurred in the San Francisco Bay Area between March and July of 2020, when the dominant local strain was the original ancestral (Wuhan) strain. The remaining five participants were not previously infected with the virus. All eleven participants were vaccinated with both doses of either of the Moderna or Pfizer/BioNTech mRNA vaccines (Table 1). Blood was drawn from each of the eleven participants prior to vaccination, ~ 2 weeks after the first vaccine dose, and ~2 weeks after the second vaccine dose (33 specimens total). On the day of each blood draw, PBMCs were isolated from blood using Lymphoprep (StemCell Technologies), and then cryopreserved in 90 % fetal bovine serum (FBS) and 10 % DMSO. For participant PID4101, an additional blood-draw and intranasal swab specimens were obtained for immunophenotyping studies. This study was approved by the University of California, San Francisco, and all participants provided informed consent (IRB # 20–30588).

Cryopreserved PBMCs were revived and cultured overnight to allow for antigen recovery. The cells were then counted, and then 2 million cells per treatment condition were stimulated with the co-stimulatory agents 0.5 μg/ml anti-CD49d clone L25 and 0.5 μg/ml anti-CD28 clone L293 (both from BD Biosciences), in the presence of 0.5 μM of overlapping 15-mer SARS-CoV-2 spike peptides PepMix SARS-CoV-2 peptides from the original SARS-CoV-2 strain, B.1.1.7, or B.1.351, or overlapping 15-mer SARS-CoV-2 nucleocapsid peptides (all from JPT Peptide Technologies). Stimulations were conducted for 6 hours in RP10 media (RPMI 1640 medium (Corning) supplemented with 10 % FBS (VWR), 1 % penicillin (Gibco), and 1 % streptomycin (Gibco)), in the presence of 3 μg/ml Brefeldin A Solution (eBioscience) to enable detection of intracellular cytokines. To establish the phenotypes of total T cells in the absence of stimulation, 2 million cells were cultured in parallel with the stimulated samples, but in the presence of only 3 μg/ml Brefeldin A.

After culture, the cells were treated with cisplatin (Sigma-Aldrich) as a live/dead marker and fixed with paraformaldehyde (PFA) as previously described Neidleman et al., 2020b; Ma et al., 2020. Cisplatin treatment and fixation was performed as follows: first, cells were resuspended in 2 ml PBS (Rockland) with 2 ml EDTA (Corning), followed by addition of 2 ml PBS/EDTA supplemented with 25 μM cisplatin (Sigma-Aldrich) for 60 s. Cisplatin staining was then quenched with 10 ml of CyFACS (metal contaminant-free PBS (Rockland) supplemented with 0.1 % FBS and 0.1 % sodium azide (Sigma-Aldrich)), centrifuged, and resuspended in 2 % PFA in CyFACS. Fixation was allowed to proceed for 10 minutes at room temperature, after which cells were washed twice with CyFACS, and then resuspended in CyFACS containing 10 % DMSO. Fixed cells were stored at –80 °C until analysis by CyTOF. For paired blood/swab specimens from PID4101, cells were immediately cisplatin-treated and fixed, without prior cryopreservation.

CyTOF staining was conducted in a fashion similar to recently described methods Neidleman et al., 2021; Neidleman et al., 2020b; Ma et al., 2020; Cavrois et al., 2017; Neidleman et al., 2020a; Xie et al., 2021. Cisplatin-treated cells were thawed, counted, and each treatment condition was barcoded using the Cell-ID 20-Plex Pd Barcoding Kit (Fluidigm). After the cells were barcoded and washed, the barcoded samples were combined and diluted to 6 × 10⁶ cells / 800 μl CyFACS per well in Nunc 96 DeepWell polystyrene plates (Thermo Fisher). Cells were blocked with mouse (Thermo Fisher), rat (Thermo Fisher), and human AB (Sigma-Aldrich) sera for 15 min at 4 °C, and then washed twice in CyFACS. Surface CyTOF antibody staining (Table 2) was conducted for 45 min at 4 °C, in a volume of 100 μl / sample. Cells were then washed three times with CyFACS and fixed overnight at 4 °C in 100 μl of 2 % PFA in PBS. The next day, samples were washed twice with Intracellular Fixation & Permeabilization Buffer (eBioscience), and incubated for 45 minutes at 4 °C. After two additional washes with Permeabilization Buffer (eBioscience), samples were blocked for 15 min at 4 °C in 100 μl of Permeabilization Buffer containing mouse and rat sera. After one additional wash with Permeabilization Buffer, samples were stained with the intracellular CyTOF antibodies (Table 2) at 4 °C for 45 min in a volume of 100 μl / sample. Cells were then washed once with CyFACS, and stained for 20 min at room temperature with 250 nM of Cell-ID Intercalator-IR (Fluidigm). Immediately prior to sample acquisition, cells were washed twice with CyFACS buffer, once with MaxPar cell staining buffer (Fluidigm), and once with Cell acquisition solution (CAS, Fluidigm). Cells were resuspended in EQ Four Element Calibration Beads (Fluidigm) diluted in CAS immediately prior to acquisition on a Helios-upgraded CyTOF2 instrument (Fluidigm) at the UCSF Parnassus flow core facility.

CyTOF datasets, exported as flow cytometry standard (FCS) files, were de-barcoded and normalized according to manufacturer’s instructions (Fluidigm). FlowJo software (BD Biosciences) was used to identify CD4+ T cells (live, singlet CD3+ CD19 CD4+ CD8-) and CD8+ T cells (live, singlet CD3+ CD19 CD4-CD8+) among all analyzed samples. IFNγ+ in the stimulated samples were considered to be the SARS-CoV-2-responsive cells. For high-dimensional analyses of SARS-CoV-2-specific T cells among the stimulated samples, we excluded samples with an insufficient number of events ( ≤ 3) to limit skewing of the data. Manual gating analysis was initially performed using FlowJo, and then select populations were exported as FCS files and then imported into R software as GatingSet objects. Using the CytoExploreR package, 2D-gates were manually drawn on the imported samples. The 2D dot plots and statistical results were exported for data visualization, bar-graph generation, and statistical comparisons as previously described (https://github.com/DillonHammill/CytoExploreR↗; Hammill, 2021). High-dimensional analyses (MDS, tSNE, and FlowSOM) were performed using R software by implementing a CyTOF workflow recently described Nowicka et al., 2017.

For MDS plot generation, we used the plotMDS function from the limma package with default settings. Euclidean distances between all samples were calculated using the arcsinh-transformed median expression levels with cofactor 5, of the lineage and functional markers listed below.

The first (MDS1) and second (MDS2) MDS dimensions were plotted to show the dissimilarities between samples from the indicated conditions as described Ritchie et al., 2015.

tSNE was performed using the Trsne function from the Rtsne package using arcsinh-transformed expression of lineage markers (no PCA step, iterations = 1000,, perplexity = 30, theta = 0.5). Events corresponding to unstimulated T cells were down-sampled to 1000 cells per sample, and SARS-CoV-2-specific cells (cell numbers ranging from 4 to 229 per sample) were all included in the tSNE analyses without down-sampling. Each cell was displayed in a tSNE plot for dimension reduction visualization and colored with arcsinh-transformed cell marker expression as heatmaps, or pseudo-colored by the appropriate group.

Unsupervised cell subset clustering was performed using FlowSOM Van Gassen et al., 2015 and ConsensusClusterPlus packages using arcsinh-transformed expression levels of the lineage markers indicated above Wilkerson and Hayes, 2010. For clustering of SARS-CoV-2-specific T cells, we set the meta-cluster number to eight and cluster number to 40. The frequency of each cluster within each sample was calculated using the following equation:

(Frequency of cluster in specified sample) = (Cell count of cluster / Total cell count of specified sample).

This was then converted to a percentage by multiplying by 100. The percentages of each cluster from the selected samples were plotted as box plots with jittered points, followed by statistical analysis between the groups. To compare the abundance distribution of clusters between groups, frequencies of clusters in samples from each group were normalized using the equation below:

(Normalized frequency of cluster in specified sample) = (Frequency of cluster in specified sample/ Total number of samples in each group).

This was then converted to a percentage by multiplying by 100, and plotted as stacked bar charts.

The statistical tests used in comparison of groups are indicated within the figure legends. For 2-group comparisons, student’s t-tests were performed and p-values were adjusted for multiple testing using the Holm-Sidak method where applicable. For comparisons of three or more groups, significance between groups was first evaluated by one-way ANOVA, and then the p-values were adjusted for multiple testing using the Holm-Sidak method where applicable. For datasets with significant ANOVA-adjusted p-values ( ≤ 0.05), we performed Tukey’s honestly significant difference (HSD) post-hoc test to determine the p-values between individual groups.

For this study, a total of 120 specimens were analyzed by CyTOF. Each specimen included both CD4+ and CD8+ T cells. For each specimen, we gated separately on events corresponding to CD4+ T cells (live, singlet CD3+ CD19- CD4+ CD8-) and CD8+ T cells (live, singlet CD3+ CD19- CD4 CD8+), and exported the files as 240 individual FCS files. These 240 raw CyTOF datasets are available for download through the public repository Dryad via the following link: https://doi.org/10.7272/Q60R9MMK↗.