Main

Upon encounter with cognate antigens, lymphocytes are endowed with the capacity to form memory cells1,2. Memory lymphocytes are usually long-lived and provide faster and more vigorous immune responses upon secondary contact with their specific antigen2. Some memory cells circulate between blood, secondary lymphoid organs and bone marrow, while others migrate to peripheral tissues and mucosal sites where they can become tissue resident3.

Whereas subdivision of labor in terms of tissue homing and effector functions has been well characterized for memory T cells, functionally different subsets also exist for memory B (Bm) cells. Antigen-stimulated B cells receiving instructive signals from their interaction with helper CD4+ T cells can further differentiate in the germinal centers (GCs) of secondary lymphoid organs or using an extrafollicular pathway. B cells that differentiate in the GC undergo affinity maturation through somatic hypermutation (SHM) of the B cell receptor (BCR) following which B cells can become long-lived plasma cells or Bm cells4,5,6. Long-lived plasma cells can continuously secrete high-affinity antibodies that are protective against a homologous pathogen7, whereas Bm cells encode a broader repertoire which allows protection against variants of the initial pathogen after restimulation8. Upon antigen reencounter, Bm cells differentiate into antibody-secreting plasma cells or reenter GCs where they undergo additional SHM9.

Bm cells can be subdivided into phenotypically and functionally distinct subsets10. In humans, resting Bm cells are typically CD21hi, and express the tumor necrosis factor (TNF) receptor superfamily member CD27. Additionally, CD21CD27+ activated Bm cells11 might represent a GC-derived population prone to plasma cell differentiation12, and CD21CD27 Bm cells have been reported in chronic infection, immunodeficiency and autoimmune diseases and are thought to be of extrafollicular origin13,14,15,16,17,18. However, the differentiation path of CD21CD27+ Bm cells and CD21CD27 Bm cells remains ill-defined. Antigen-specific CD21CD27+ and CD21CD27 Bm cells have been transiently detected after vaccines12,19,20,21,22 and during infection with certain pathogens21,23,24, including severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) (refs. 25,26,27,28,29). CD21CD27 Bm cells depend on the transcription factor T-bet for their development30, are CD11chi and express inhibitory coreceptors, such as Fc receptor-like protein 5 (FcRL5) (refs. 31,32).

In this article, we studied the kinetics, distribution and interrelatedness of antigen-specific Bm cell subsets during acute infection and months 6 and 12 post-infection with SARS-CoV-2 in individuals with mild and severe coronavirus disease 2019 (COVID-19) that have also received SARS-CoV-2 messenger RNA vaccination post-infection, and healthy volunteers before and after SARS-CoV-2-specific vaccination. We found that SARS-CoV-2-specific CD21CD27+ activated Bm cells and CD21CD27 Bm cells were the predominant subsets in circulation during acute infection and upon vaccination. CD21+ resting Bm cells became prevalent at 6–12 months post-infection. Single-cell RNA sequencing (scRNA-seq) indicated that single Bm cell clones adopted different fates upon antigen reexposure.

Results

SARS-CoV-2 infection forms a durable Bm cell response

We longitudinally studied antigen-specific Bm cells in a cohort of 65 patients with COVID-19, 33 females and 32 males, including 42 with mild and 23 with severe disease course, during their acute SARS-CoV-2 infection and at months 6 and 12 post-infection. Of these individuals, 35 received one or two doses of SARS-CoV-2 mRNA vaccination between month 6 and month 12, and three subjects were vaccinated between acute infection and month 6 (Supplementary Table 1 and Extended Data Fig. 1a).

First, we focused on samples from nonvaccinated individuals at acute infection (n = 59, day 14 on average after symptom onset), month 6 (n = 61, day 202 after symptom onset) and month 12 (n = 17, day 374) (Fig. 1a and Supplementary Table 1). SARS-CoV-2-specific Bm cells were identified using probes of biotinylated SARS-CoV-2 spike (S) and receptor-binding domain (RBD) protein multimerized with fluorophore-labeled streptavidin (SAV) and characterized using a 28-color spectral flow cytometry panel (Fig. 1b and Extended Data Fig. 2a). We observed a strong increase in the frequency of S+ and RBD+ Bm cells in SARS-CoV-2-infected individuals at months 6 (median 0.14% and 0.033%, respectively) and 12 post-infection (median 0.068% and 0.02%) compared with acute infection (median 0.016% and 0.0023%) (Fig. 1c and Extended Data Fig. 2b,c). Frequencies of S+ Bm cells were comparable in patients with mild and severe COVID-19 (Fig. 1d). During acute infection S+ Bm cells were mainly immunoglobulin (Ig)M+ and IgG+, whereas IgG+ Bm cells predominated (85–90%) at months 6 and 12 post-infection (Fig. 1e). IgG1 represented the most common subtype (around 65% of S+ Bm cells at months 6 and 12 post-infection), and between 5% and 10% of S+ Bm cells were IgA+ (Fig. 1e,f).

Fig. 1: Longitudinal analysis of SARS-CoV-2-specific Bm cells post-infection.
figure 1

a, SARS-CoV-2-infected patients were analyzed by spectral flow cytometry and scRNA-seq at acute infection and months 6 and 12 post-infection. b, Representative flow cytometry plots show percentages of decoy-negative SARS-CoV-2 S+ Bm cells (gated as in Extended Data Fig. 2a) of patient CoV-P1 pre-exposure to SARS-CoV-2, at days 33 and 152 post-symptom onset and at day 12 post-first dose of SARS-CoV-2 mRNA vaccination (that is, day 166 post-symptom onset). c, Frequency of S+ Bm cells in total B cells was measured by flow cytometry at acute infection (n = 59) and months 6 (n = 61) and 12 post-infection (n = 17). Lines connect samples of same individual. Red line represents fitted second-order polynomial function (R2 = 0.1932). d, Frequency of S+ Bm cells was measured by flow cytometry and separated by mild (acute, n = 40; month 6, n = 39; month 12, n = 11) and severe COVID-19 (acute, n = 19; month 6, n = 22; month 12, n = 6). e, Shown are gating strategy (left) and stacked bar plots (mean + standard deviation; right) of IgG+, IgM+ and IgA+ S+ Bm cells at indicated timepoints (acute, n = 23; month 6, n = 52; month 12, n = 16). f, Violin plots show percentages of IgG1+ (left) and IgG3+ (right) S+ Bm cells at indicated timepoints (acute, n = 23; month 6, n = 52; month 12, n = 16). g, Heat map represents V heavy (VH) gene usage, in RBD+ and RBD Bm cells in scRNA-seq dataset from months 6 and 12. Shown are 30 most frequently used VH segments, sorted by hierarchical clustering, with colors indicating frequencies. Samples in cf were compared using Kruskal–Wallis test with Dunn’s multiple comparison, showing adjusted P values. Frequencies in g were compared using two-proportions z-test with Bonferroni’s multiple testing correction. P values in e and g are shown if significant. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Source data

Next, we performed droplet-based scRNA-seq combined with feature barcoding and BCR sequencing (BCR-seq) on sorted S+ and S Bm cells isolated from the blood of nine patients with COVID-19 at months 6 and 12 post-infection; three patients were nonvaccinated, and six received SARS-CoV-2 mRNA vaccination between month 6 and month 12 (Extended Data Fig. 2d and Supplementary Table 2). Bm cells specific for RBD, wild-type spike (SWT) or spike variants B.1.351 (Sbeta) and B.1.617.2 (Sdelta) were identified by SAV multimers carrying specific oligonucleotide barcodes. The majority of Sbeta+, Sdelta+ and RBD+ Bm cells also recognized SWT (Extended Data Fig. 3a,b). This scRNA-seq approach detected frequencies of about 30% of RBD+ Bm cells within S+ Bm cells that were comparable to flow cytometry (Extended Data Figs. 2a and 3c). Analysis of V heavy and light chain frequencies identified several chains enriched in RBD+ Bm cells compared with RBD Bm cells described to encode RBD-binding antibodies, including IGHV3-30, IGHV3-53, IGHV3-66, IGKV1-9 and IGKV1-33 (refs. 33,34) (Fig. 1g and Extended Data Fig. 3d). Collectively, these data identify a durable, IgG1-dominated S+ Bm cell response forming upon SARS-CoV-2 infection.

Different Bm cell subsets form after SARS-CoV-2 infection

By using uniform manifold approximation and projection (UMAP) we visualized S+ Bm cells from the flow cytometry dataset obtained in nonvaccinated post-infection samples and performed a PhenoGraph clustering (Extended Data Fig. 4a–c). UMAP and clustering grouped Bm cells by IgG (clusters 1–5), IgM (clusters 6 and 7) and IgA (clusters 8 and 9) expression and revealed a phenotypical shift from acute infection to months 6 and 12 post-infection characterized by increased expression of CD21 on S+ Bm cells, whereas expression of Blimp-1, Ki-67, CD11c, CD71 and FcRL5 diminished (Extended Data Fig. 4a,c). PhenoGraph clustering identified an IgG+CD21CD27 cluster (cluster 2), which was TbethiCD11c+FcRL5+, and CD21CD27+ clusters characterized by high expression of CD71, Blimp-1 and Ki-67 (clusters 1, 7 and 8) (Extended Data Fig. 4a–c).

The expression changes in CD21 and CD27 on S+ Bm cells between acute infection and months 6 and 12 post-infection could also be reproduced by manual gating (Fig. 2a). During acute infection S+ CD21CD27+ Bm cells and CD21CD27 Bm cells represented on average 48.1% and 16.4% of total S+ Bm cells, respectively, and they strongly declined at month 6 (6.3% and 5.3%) and month 12 (3.7% and 6.6%) post-infection (Fig. 2b). Conversely, CD21+CD27+ and CD21+CD27 Bm cells were prominent at months 6 and 12, amounting to 60.5% and 29.1% of S+ Bm cells at month 12, respectively (Fig. 2b). These dynamics were comparable in patients with mild and severe COVID-19 (Extended Data Fig. 4d). Expression of Blimp-1, T-bet, FcRL5 and CD71 were increased on S+ Bm cells during acute infection compared with months 6 and 12 post-infection (Fig. 2c), and S+ Bm cells underwent strong proliferation during the acute phase (Fig. 2d). S+ Bm cells continued to show lower but still significantly increased proliferation at month 6, and only returned to background levels at month 12 post-infection (Fig. 2d).

Fig. 2: Phenotypic and functional characterization of circulating SARS-CoV-2-specific Bm cells post-infection.
figure 2

a, CD21 and CD27 expression on S+ Bm cells during acute infection (top) and month 6 post-infection (bottom) of patient CoV-P2 was determined by flow cytometry. b, Violin plots of frequencies of CD21CD27+, CD21CD27, CD21+CD27+ and CD21+CD27 cells within S+ Bm cells are shown at acute infection (n = 23) and months 6 (n = 52) and 12 post-infection (n = 16). Lines connect samples of same individual. Included were only pre-vaccination samples. c, Violin plots represent geometric mean fluorescence intensities (gMFI) or percentages of indicated markers in S+ Bm cells at acute infection (n = 23), and months 6 (n = 52) and 12 post-infection (n = 16), compared with S Bm cells at acute infection (n = 23). d, Shown are representative histograms of Ki-67 in patient CoV-P2 (left) and violin plots of percentages of Ki-67+ S+ Bm cells compared with S Bm cells (right) at indicated timepoints. e, Presented are SHM counts in S+ Bm cells binding SWT, variant S (Sbeta and Sdelta) or RBD at month 6 (n = 634 cells) and month 12 post-infection (n = 197 cells; nonvaccinated); SHM counts in naïve B cells (n = 1,462) are shown as reference. Box plots show medians, box limits and interquartile ranges (IQRs), with whiskers representing 1.5× IQR and outliers (also applies to subsequent figures). f, Waffle plots represent SWT+ Bm cells binding Sbeta and Sdelta in nonvaccinated individuals (n = 9 at month 6 and n = 3 at month 12 post-infection). Samples in bd were compared using Kruskal–Wallis test with Dunn’s multiple comparison correction, showing adjusted P values if significant. In e, two-sided Wilcoxon test was used with Holm multiple comparison correction. Samples in f were compared using two-proportions z-test.

Source data

The scRNA-seq dataset identified a significantly increased SHM count in S+ Bm cells at month 12 compared with month 6 post-infection (Fig. 2e), which correlated with an improved binding breadth, as measured by variant-binding ability of SWT+ Bm cells (Fig. 2f). At the transcriptional level, S+ Bm cells at month 6 post-infection upregulated genes associated with B cell activation and recent GC emigration35, such as NKFBIA, JUND, MAP3K8, CXCR4 and CD83, compared with S+ Bm cells at month 12 (Extended Data Fig. 4e). These data showed that SARS-CoV-2 infection induced a stable CD21+ Bm cell population in the circulation, which continuously matured for more than 6 months.

Tonsillar S+ Bm cells undergo tissue adaptation

To extend our analyses to SARS-CoV-2-specific Bm cells in the peripheral lymphoid organs, we analyzed paired tonsil and blood samples from a cohort of 16 patients (9 females and 7 males) undergoing tonsillectomy who were exposed to SARS-CoV-2 by infection, vaccination or both. Eight patients were vaccinated against SARS-CoV-2 (analyzed on average at day 144 after last vaccination), whereas the other eight patients were considered SARS-CoV-2-recovered based on a history of SARS-CoV-2 infection or positive anti-nucleocapsid (N) serum antibody measurement, with six of them additionally vaccinated against SARS-CoV-2 (assessed on average at day 118 post-last vaccination) (Extended Data Fig. 1b and Supplementary Table 3).

Flow cytometry using the multimer probe approach (Extended Data Fig. 5a,b) identified S+ Bm cells in the blood and tonsils of both vaccinated and recovered individuals, whereas N+ Bm cells were enriched only in recovered individuals (Fig. 3a,b). In tonsils, the S+ Bm cells were less IgG+ (77.4% versus 82.1%) and IgM+ (2.4% versus 5.5%), but more IgA+ (9.1% versus 6%) compared with the circulation (Fig. 3c). Analysis of SARS-CoV-2-specific GC Bcl-6+Ki-67+ B cells detected a trend towards elevated frequencies of S+ and N+ GC cells in recovered compared with vaccinated subjects (Extended Data Fig. 5c).

Fig. 3: Phenotypic and transcriptional makeup of circulating and tonsillar SARS-CoV-2-specific Bm cells post-infection and post-vaccination.
figure 3

a, Flow cytometry plots show decoy S+ (top) and nucleocapsid (N)+ Bm cells (bottom) in paired tonsil and blood samples of a SARS-CoV-2-vaccinated (CoV-T1; left) and SARS-CoV-2-recovered patient (CoV-T2; right). b, N+ (left) and S+ (right) Bm cell frequencies were determined in paired blood and tonsils of SARS-CoV-2-vaccinated (n = 8) and SARS-CoV-2-recovered individuals (n = 8). Lines connect samples of same individual. c, Stacked bar plots (mean + standard deviation) represent isotypes in blood and tonsillar S+ Bm cells from both SARS-CoV-2-vaccinated and SARS-CoV-2-recovered individuals (n = 16; also applies to d and e). d, Contour plots show CD21 and CD27 expression on blood and tonsillar S+ Bm cells of patient CoV-T2 (left) and frequencies of indicated Bm cell subsets (right). Lines connect samples of same individual. e, Representative CD69 histograms in S+ Bm cells of patient CoV-T2 (left) and percentages of CD69+ S+ Bm cells (right) in blood and tonsils. f,g, WNN UMAP of Bm cells was derived from scRNA-seq analysis of blood and tonsillar B cells (n = 4). Bm cells are colored by cluster (f, left), tissue origin (f, right) or SWT binding (g). h, Expression of selected genes (left) and surface protein markers (right) are shown in Bm cell clusters. i, SHM counts are provided for naïve B cells (n = 1,607), blood (n = 170) and tonsillar SWT+ Bm cells (n = 1,128). j, WNNUMAP was derived as in f and colored by tissue origin. Lines connect shared clones. k, Venn diagram shows clonal overlap of SWT+ and SWT– Bm cells in tonsils and blood from scRNA-seq dataset. Samples in b were compared using a Kruskal–Wallis test with Dunn’s multiple comparison correction, in ce with a two-tailed Wilcoxon matched-pairs signed-rank test and in i with a two-sided Wilcoxon test with Holm multiple comparison correction. *P < 0.05, **P < 0.01.

Source data

Among the S+ Bm cell subsets, CD21CD27+ Bm cells and CD21CD27 Bm cells were more frequent in blood, whereas CD21+CD27 Bm cells were more frequent in tonsils (Fig. 3d). Compared with their circulating counterparts, tonsillar S+ and N+ Bm cells expressed, on average, more CD69, less Ki-67, reduced T-bet and several chemokine receptors differently (Fig. 3e and Extended Data Fig. 5d,e). Very few S+ tonsillar Bm cells expressed FcRL4 in both vaccinated and recovered individuals (Extended Data Fig. 5f,g).

We performed scRNA-seq combined with feature barcoding, which allowed us to assess surface phenotype and to perform BCR-seq in sorted S+ Bm cells and S B cells from paired blood and tonsil samples of four patients (two SARS-CoV-2-recovered and two SARS-CoV-2-vaccinated). Weighted-nearest neighbor (WNN) clustering identified naïve B cells (IgMhiIgDhiFCER2hi), naïve/activated B cells (IgMhiIgDhiFCER2hiFCRL5hi), GC B cells (CD27hiCD38hiAICDAhi) and Bm cells (IgMloIgDloCD27int) (Extended Data Fig. 6a–c). Subsequent reclustering of Bm cells resolved six clusters (Fig. 3f–h and Extended Data Fig. 6d,e). The SWT+ Bm cells in the IgG+CD27hiCD45RBhi cluster (cluster 5) were mainly from blood, in the IgG+CD21hi cluster (cluster 2) predominantly tonsillar, while the IgG+CD27lo cluster (cluster 4) contained SWT+ Bm cells from both compartments. The FCRL4hiENTPD1hiTNFRSF13Bhi cluster (cluster 6) probably represented the FcRL4+ B cell subset, and contained very few SWT+ Bm cells (Fig. 3g,h and Extended Data Fig. 6d–g). Differential gene expression identified higher expression of CR2, CD44, CCR6 and CD69 in tonsillar SWT+ Bm cells compared with blood SWT+ Bm cells, whereas the activation-related genes FGR and CD52 were higher in blood SWT+ Bm cells compared with their tonsillar counterparts (Extended Data Fig. 6h). BCR-seq showed similar SHM counts in SWT+ Bm cells in blood and tonsils (Fig. 3i). We identified 16 shared SWT+ Bm cell clones between these compartments (Fig. 3j,k). Taken together, resting antigen-specific Bm cells were found in the tonsils after SARS-CoV-2 exposure, and they carried signs of tissue adaptation and clonal connection to their circulating counterparts.

Bm cell subsets reshift following SARS-CoV-2 vaccination

We probed the Bm cell response to antigen reexposure in 35 of the 65 patients with COVID-19 who had received mRNA vaccination between month 6 and month 12 post-infection (Extended Data Fig. 1a and Supplementary Table 1). The frequency of blood S+ Bm cells was approximately fivefold increased post-vaccination at month 12 compared with pre-vaccination at month 6 post-infection (Fig. 4a,b). Time-resolved analysis identified a peak in the frequency of S+ Bm cells in the first days post-vaccination, reaching 3% of total B cells on average, followed by a slow decrease in frequency over day 150 post-vaccination (Fig. 4c). The scRNA-seq dataset identified a trend towards increased clonality of S+ Bm cells in the six patients vaccinated between month 6 and month 12 post-infection when comparing pre-vaccination with post-vaccination (Fig. 4d). Counts of SHM in S+ Bm cells remained high at month 12 (post-vaccination) compared with month 6 post-infection (pre-vaccination) (Fig. 4e).

Fig. 4: Changes in antigen-specific Bm cell subsets following vaccination-induced antigen reexposure.
figure 4

a, Representative flow cytometry plots of decoy S+ Bm cells are displayed at pre-vaccination (preVac; left; month 6) and day 78 post-vaccination (postVac; right; month 12 post-infection) in patient CoV-P3. b, Paired comparison of S+ Bm cell frequencies within B cells (n = 34) was performed at preVac and postVac. c, S+ Bm cell frequencies within B cells (n = 41) are plotted against time post-last vaccination. Lines connect paired samples. Semilog line was fitted to data (R2 = 0.2695). d, Clonality of S+ Bm cells was analyzed preVac and postVac in scRNA-seq dataset. Each dot represents an individual (n = 6). e, SHM counts of S+ Bm cells were derived at preVac (n = 634 cells), month 12 nonvaccinated (nonVac; n = 197 cells), and early (less than 24 days; n = 838 cell) and late (more than 84 days; n = 1,116 cells) postVac. Naïve B cell (n = 1462 cells), served as reference and are the same as in Fig. 2e, as are preVac and nonVac SHM counts. f, Representative contour plots of CD21 and CD27 expression on S+ Bm cells are shown at preVac and day 9 and day 78 postVac. g, Frequencies (n = 29 pairs; left) and pie charts (right) of indicated S+ Bm cell subsets are provided at indicated timepoints. h, Percentages of S+ Bm cell subsets are plotted against time post-last vaccination. Lines connect paired samples. Linear regressions are fitted to data. We used a two-tailed Wilcoxon matched-pairs signed-rank test in b, d and g, and two-sided Wilcoxon test in e. The Holm–Bonferroni method was used for P value adjustment of multiple comparisons.

Source data

Whereas S+ Bm cells were predominantly resting CD21+ Bm cells at month 6, vaccination strongly induced the appearance of S+ CD21CD27+ and CD21CD27 Bm cells in blood (Fig. 4f,g). S+ CD21CD27+ activated Bm cells peaked in the first days post-vaccination, followed by a rapid decline over the subsequent 100 days (Fig. 4h). Conversely, the frequency of S+ CD21CD27 Bm cells rose quickly and remained stable over 150 days post-vaccination, accounting for about 20% of S+ Bm cells (Fig. 4h).

Subsequently, we analyzed S+ Bm cells in the blood of SARS-CoV-2-naïve individuals (all seronegative for S-specific antibodies) by flow cytometry (n = 11, five females and six males) and scRNA-seq (n = 3) sampled before their SARS-CoV-2 mRNA vaccination, at days 8–13 (week 2) post-second dose, 6 months after the second dose and days 11–14 post-third dose (Extended Data Fig. 1c and Supplementary Table 4). This revealed a potent induction of S+ IgG+ Bm cells at week 2 post-second dose, which stably persisted to month 6 post-second dose, and the frequency further increased early post-third dose compared with month 6 post-second dose (Extended Data Fig. 7a–c). Antigen-specific Bm cells were dominated by CD21CD27+ Bm cells (around 55% of S+ Bm cells) and, to a lesser extent, by CD21CD27 Bm cells (5–15%) at week 2 post-second dose and post-third dose compared to month 6 post-second dose. Conversely, S+ CD21+ Bm cell subsets became predominant at month 6 post-second dose (Extended Data Fig. 7d).

Flow cytometry analysis of S+ Bm cells showed an upregulation of Blimp-1 at week 2 post-second dose compared with month 6, and increased expression of T-bet, FcRL5, CD71 and Ki-67 at week 2 post-second dose and post-third dose (Extended Data Fig. 7e,f). The scRNA-seq data showed that SHM counts in SWT+ Bm cells strongly increased from week 2 post-second (median 3) to month 6 post-second dose (median 13) and even further at week 2 post-third dose (median 14) (Extended Data Fig. 7g). Altogether, these observations indicated that antigen reexposure by SARS-CoV-2 vaccination of SARS-CoV-2-recovered and SARS-CoV-2-vaccinated individuals stimulated S+ CD21CD27+ and CD21CD27 Bm cells.

S+ Bm cell subsets show distinct transcriptional profiles

We used the scRNA-seq of S+ and S Bm cells sorted from recovered individuals with and without subsequent vaccination to interrogate the pathways guiding development of different Bm cell subsets (Extended Data Fig. 8a). WNN clustering of all sequenced Bm cells identified ten clusters that, on the basis of the expression of cell surface markers and Ig isotype, were merged into five subsets annotated as CD21CD27+CD71+ activated Bm cells, CD21CD27FcRL5+ Bm cells, CD21+CD27 resting Bm cells, CD21+CD27+ resting Bm cells and unswitched CD21+ Bm cells (Fig. 5a and Extended Data Fig. 8b,c). Unswitched CD21+ Bm cells were IgM+, whereas the other Bm cell subsets expressed mainly IgG, with IgG1 being the dominant subclass (Extended Data Fig. 8d,e). The flow cytometry data further showed that S+ CD21CD27 Bm cells were enriched in IgG3+ compared with CD21+CD27+ resting Bm cells (Extended Data Fig. 8e,f). In the scRNA-seq dataset, CD21+CD27+ resting Bm cells were the main S+ Bm cell subset at months 6 and 12 post-infection in nonvaccinated individuals, whereas CD21CD27+CD71+ activated and CD21CD27FcRL5+ Bm cells became predominant post-vaccination at month 12 post-infection (Fig. 5a,b and Extended Data Fig. 8g).

Fig. 5: Transcriptional makeup of SARS-CoV-2-specific Bm cell subsets.
figure 5

a, WNNUMAP was derived from scRNA-seq dataset at months 6 and 12 post-infection (n = 9) and colored by indicated Bm cell subsets (top) and S+ and S separated by month 6 preVac, month 12 nonVac and month 12 postVac (bottom). b, Distribution of S+ Bm cell subsets is provided at month 6 preVac, month 12 nonVac and month 12 postVac. c, Heat map shows selected, significantly differentially expressed genes in indicated S+ Bm cell subsets. Functional groups of genes were ordered by hierarchical clustering. d, Representative histograms (left) and violin plots of indicated markers on S+ Bm cell subsets (right) postVac were derived from the flow cytometry dataset (n = 37). e, Heat map shows enrichment scores of selected gene sets that are significantly different between CD27lo/hiCD21+ resting and CD21CD27FcRL5+ S+ Bm cell subsets in a pseudobulk analysis (n = 5 individuals). f,g, GSEA of CD21CD27FcRL5+ S+ Bm cells versus CD21+ resting S+ Bm cells are shown for indicated gene sets. Red dashed lines indicate minimal and maximal cumulative enrichment values. Samples in d were compared using Kruskal–Wallis test with Dunn’s multiple comparison correction, showing adjusted P values if significant. For f and g, statistical analysis of the gene set enrichment and variation analyses was performed as outlined in Methods, and all adjusted P values are shown. GOPB, Gene Ontology Biological Process.

Source data

Analysis of differentially expressed genes indicated that CD21CD27FcRL5+ B cells were the most distinctive subset and had high expression of TBX21 (encoding T-bet), T-bet-driven genes ZEB2 and ITGAX (encoding CD11c), and TOX (Fig. 5c). They were also enriched in gene transcripts involved in interferon (IFN)-γ and BCR signaling and showed high expression of integrins ITGAX, ITGB2 and ITGB7 (Fig. 5c). Moreover, expression of inhibitory receptors, including FCRL2, FCRL3, FCRL5, SIGLEC6, SIGLEC10, LAIR1, LILRB1 and LILRB2, and proteins involved in antigen presentation and processing, such as HLA-DPA1, HLA-DPB1, HLA-DRB1, HLA-DRB5, CD74 and CD86, was particularly high in CD21CD27FcRL5+ Bm cells (Fig. 5c). Several of these differences, such as T-bet, and CD11c, were confirmed at the protein level (Fig. 5d).

Gene set variation and enrichment analysis revealed a strong enrichment of a previously described B cell signature of IgDCD27CXCR5 ‘atypical’ Bm cells from patients with systemic lupus erythematosus (SLE)36, in our SARS-CoV-2-specific CD21CD27FcRL5+ Bm cell subset (Fig. 5e,f). Gene sets involved in antigen presentation and integrin-mediated signaling, as well as B cell activation, BCR and IFN-γ signaling were enriched in CD21CD27FcRL5+ Bm cells compared with other Bm cell subsets (Fig. 5e,g). In summary, the data showed that S+ CD21CD27FcRL5+ Bm cells carried a very distinct transcriptional profile, similar to certain B cells found in autoimmunity.

BCR-seq reveals clonal branching of S+ Bm cell subsets

Comparison of V heavy and light chain usage within S+ Bm cell subsets in the scRNA-seq data from SARS-CoV-2-recovered individuals (months 6 and 12 post-infection) revealed very similar chain usage in S+ CD21+ resting (CD21+CD27+ and CD21+CD27 combined), CD21CD27+CD71+ activated and CD21CD27FcRL5+ Bm cells (Extended Data Fig. 9a). BCR diversity was slightly reduced in S+ CD21CD27FcRL5+ compared with S+ CD21+ resting Bm cells (Extended Data Fig. 9b). BCR-seq detected shared clones mostly between S+ CD21+CD27+ and CD21CD27+CD71+ activated Bm cells, as well as the CD21CD27FcRL5+ Bm cell subset (Extended Data Fig. 9c), indicating that S+ Bm cell subsets had comparable BCR repertoires, although the depth of our analysis was restricted by low cell numbers.

Longitudinal tracking of S+ Bm cell clones between month 6 and month 12 post-infection identified 30 persistent clones in individuals vaccinated during that period (Fig. 6a and Extended Data Fig. 9d). At month 6 post-infection (pre-vaccination), 80% of those 30 clones had a CD21+ resting Bm cell phenotype (Fig. 6b), whereas at month 12 post-infection (post-vaccination) 32% of persistent Bm clones showed a CD21CD27+CD71+ and 28% a CD21CD27FcRL5+ Bm cell phenotype. The S+ Bm cell subset distribution of newly detected clones (n = 1,357 clones) at month 12 post-infection (post-vaccination) was comparable to the persistent clones (Fig. 6b). Between month 6 and month 12 post-infection, persistent Bm cell clones upregulated genes associated with CD21CD27FcRL5+ Bm cells, including TBX21, ITGAX and FCRL5 (Fig. 6c). In addition, reconstruction of clonal lineage trees and visualizing persistent S+ Bm cell clones in a circos plot indicated that individual Bm cell clones acquired different Bm cell fates; for example, a given clone was of a CD21+CD27 resting phenotype at month 6 and adopted CD21+CD27+ resting, CD21CD27+CD71+ or CD21CD27FcRL5+ Bm cell phenotype at month 12 post-infection (post-vaccination) (Fig. 6d,e).

Fig. 6: Temporal analysis of individual SARS-CoV-2-specific Bm cell subsets.
figure 6

a, Donut plots of BCR sequences of S+ Bm cells in three representative patients preVac and postVac. Numbers inside donut plots represent counts of S+ Bm cells. Gray slices indicate individual clones found at one timepoint only, whereas persistent clones found at both timepoints are labeled by the same color. White areas represent BCR sequences found in single cells only. Slice sizes correspond to clone sizes. Percentages indicate frequencies of clonally expanded cells. b, Distribution of S+ Bm cell subsets in persistent and newly detected clones is shown at indicated timepoints. c, Average expression of indicated genes was derived at preVac and postVac in persistent S+ Bm cell clones that contained at least one CD21CD27FcRL5+ S+ Bm cell (n = 14 clones). d, Exemplary dendrograms (IgPhyML B cell trees) display different persistent Bm cell clones at months 6 (triangles) and 12 (dots) post-infection. Colors represent Bm cell subsets. Germline sequences, inferred by the Immcantation pipeline, are shown in white (squares). Branch lengths represent mutation numbers per site between each node. VH and V light (VL) genes are indicated on top of dendrograms. e, Circos plots of all persistent S+ Bm cell clones (left) and those adopting multiple Bm cell fates (right) are shown, with arrows connecting cells of months 6 with 12 and colored according to Bm cell phenotype at month 12. f, SHM counts were calculated in indicated S+ Bm cell subsets (unswitched, n = 53; CD27lo resting, n = 122; CD27hi resting, n = 535; activated, n = 713; CD21CD27FcRL5+, n = 531). g, UMAPs represent Monocle 3 analysis of all Bm cells (left) and S+ Bm cells (right). Colors indicate Bm cell subsets. Black lines indicate trajectory. Samples were compared using paired t-test (c) or two-sided Wilcoxon test (f). Holm–Bonferroni method was used for P value adjustment of multiple comparisons.

Source data

SHM counts were low in unswitched S+ CD21+ Bm cells, slightly higher in CD21+CD27 resting Bm cells, and high by comparison in CD21+CD27+ resting, CD21CD27+CD71+ activated and CD21CD27 Bm cells (Fig. 6f). Pseudotime-based trajectory analysis using Monocle 3 in our scRNA-seq dataset (Extended Data Fig. 9e–g) and visualization of Bm cells on the Monocle UMAP space identified two branches, which strongly separated CD21CD27+CD71+ activated and CD21CD27FcRL5+ Bm cells, both branching out from CD21+ resting Bm cells (Fig. 6g and Extended Data Fig. 9e). Collectively, these observations indicated that individual S+ Bm cell clones could adopt different Bm fates post-vaccination in SARS-CoV-2-recovered individuals.

Discussion

In this study, we demonstrated that individual clones of SARS-CoV-2-specific Bm cells harbored the capacity to follow phenotypically and functionally different trajectories after antigen reexposure, becoming CD21CD27+, CD21CD27 or CD21+CD27+/– Bm cells.

The transient occurrence of vaccine-specific CD21CD27 Bm cells has been described during responses to the influenza vaccine12,20, with one study reporting this Bm cell subset in de novo rather than recall responses20. CD21CD27 Bm cells have also been identified during acute SARS-CoV-2 infection and post-SARS-CoV-2 vaccination22,25,26,27,28,29. The S+ CD21CD27 Bm cells identified here were transcriptionally very similar to their ‘atypical’ counterparts in SLE. These results suggest that CD21CD27 Bm cells partake in the normal immune response to pathogens37. BCR and IFN-γ signaling appears to be a defining feature of CD21CD27 Bm cells, and probably induces and governs the T-bet-dependent transcriptional program in these cells32. We found indication of increased BCR and IFN-γ signaling in S+ CD21CD27 Bm cells, in accord with the increased expression of T-bet and the T-bet target genes ZEB2 and ITGAX30.

The heterogeneity of Bm cells could be explained by several models38,39. The various Bm cell subsets could comprise entirely separate lineages, with distinct BCR repertoires. Alternatively, single B cell clones could give rise to different Bm cell subsets, with stably imprinted phenotypes or show plasticity. Our longitudinal analysis found that distinct Bm cell subsets were clonally related, suggesting plasticity of Bm cell subsets. Studies in patients with SLE or HIV infection have suggested that CD21CD27 Bm cells differentiate through an extrafollicular pathway16,17. We found that the various S+ Bm cell subsets contained comparable amounts of SHM, suggesting that CD21CD27 Bm cells originated either from the GC or from a GC-derived progenitor Bm cell upon antigen rechallenge. The latter possibility fits well with our clonal data. It is unclear whether the CD21CD27 Bm cells observed post-vaccination can again become resting Bm cells or whether this phenotype is terminally fated. Our data showing expression of ZEB2 in CD21CD27 Bm cells suggest unidirectional plasticity, as ZEB2 acts together with T-bet to commit CD8+ effector T cells to a terminal differentiation state and has been proposed to act similarly in B cells16,40.

Whether CD21CD27 Bm cells contribute to protective immunity during infection in humans remains controversial41. T-bet+ B cells have a protective role in mouse models of acute and chronic viral infections38,42. However, antibody responses to several previously applied vaccines were normal in T-bet-deficient patients30. CD21CD27 Bm cells were reported to be able to secrete antibodies when receiving T cell help and to act as antigen-presenting cells24. We found that S+ CD21CD27 Bm cells showed signs of increased antigen processing and presentation; how much this might translate into truly increased capacity of antigen presentation is unclear43.

We found that SARS-CoV-2 infection and vaccination induced long-lived and stable antigen-specific Bm cells in the circulation that continued to mature up to 1 year post-infection, as evidenced by their elevated proliferation rate at month 6, high SHM counts and improved breadth of SARS-CoV-2 antigen recognition. This is in line with previous reports that SARS-CoV-2 infection and mRNA vaccination led to lasting Bm cell maturation through an ongoing GC reaction26,44,45,46.

These observations in circulating Bm cells were paralleled by the appearance of resting Bm cells in tonsils, where they showed high expression of CD69 and CD21 and comparable SHM counts to circulating Bm cells. CD69 expression is a hallmark of tissue residency in T cells3 and has been proposed to characterize resident Bm cells in lymphoid and nonlymphoid tissues47,48,49. Phenotype, chemokine receptor expression and clonal connections suggested these cells formed from CD21+ resting Bm cells, although we cannot exclude that some might have arisen directly in the tonsils.

One limitation of our study is that we performed the clonal analysis after vaccination recall, because the numbers of S+ Bm cells during acute SARS-CoV-2 infection were too low for our sequencing approach. Moreover, our multimer staining approach might miss low-affinity antigen binders50. The inclusion of patients with severe COVID-19 will have increased the average age of our cohort, whereas the individuals from which the tonsil samples were obtained were younger on average.

On the basis of our data, we suggest a linear–plastic model where the antigen stimulation and GC maturation of SARS-CoV-2-specific B cells resulted in the gradual adoption of a CD21+Ki-67lo resting Bm cell state at months 6–12 post-infection. These circulating resting Bm cells might be able to rapidly respond to antigen rechallenge with the acquisition of different Bm cell fates or they might home to secondary lymphoid and peripheral organs to form a CD69+ tissue-resident Bm cells. Our work also provides insight into the CD21CD27 Bm cells, which made up a sizeable portion of Bm cells following acute viral infection and vaccination in humans.

Methods

Patient cohorts

This study was approved by the Cantonal Ethics Committee of Zurich (BASEC #2016-01440). Patients with COVID-19 and healthy individuals were recruited at one of four hospitals in the Canton of Zurich, Switzerland. All study participants provided written informed consent. Serum and blood was obtained, and peripheral blood mononuclear cells were isolated by density centrifugation, washed and frozen in fetal bovine serum (FBS) with 10% dimethyl sulfoxide and stored in liquid nitrogen until use.

A longitudinal cohort (Extended Data Fig. 1a and Supplementary Table 1) consisted of individuals with reverse-transcriptase polymerase chain reaction-confirmed, symptomatic SARS-CoV-2 infection at acute infection (April to September 2020) and months 6 and 12 after infection, including patients with mild (n = 42) and severe (n = 23) COVID-19. Of these, 35 received SARS-CoV-2 mRNA vaccination between month 6 and month 12, and 3 subjects between acute infection and month 6. We included a total of 65 patients of the full cohort51,52 on the basis of a power calculation from pre-experiments and according to sample availability of at least paired samples from two timepoints. The flow cytometry and scRNA-seq subcohort characteristics are presented in Supplementary Tables 1 and 2, respectively.

Another cohort (Extended Data Fig. 1b and Supplementary Table 3) comprised subjects seen at University Hospital Zurich between November 2021 and April 2022 that underwent tonsillectomy for recurrent and chronic tonsillitis or obstructive sleep apnea and were exposed to SARS-CoV-2 by infection and/or vaccination. If they had a confirmed SARS-CoV-2 infection and/or SARS-CoV-2 nucleocapsid-specific antibodies, they were considered ‘SARS-CoV-2-recovered’. The cohort size was based on sample availability. We obtained paired tonsil and peripheral blood mononuclear cell and serum samples. Tonsils were processed according to established protocols47,53. Briefly, they were cut into small pieces, ground through 70 μm cell strainers, and washed in phosphate-buffered saline (PBS), before performing density gradient centrifugation. Subsequently, the mononuclear cells were frozen in FBS with 10% dimethyl sulfoxide and stored in liquid nitrogen until use.

We recruited 11 healthy controls (Extended Data Fig. 1c and Supplementary Table 4) with no history of SARS-CoV-2 infection and seronegative for SARS-CoV-2 S S1-specific antibodies. They donated blood before vaccination, at days 8–13 (week 2) post-second dose, 6 months after the second dose and days 11–14 post-third dose. All individuals received the Pfizer/BioNTech (BNT162b2) mRNA vaccine.

Spectral flow cytometry

To stain antigen-specific B cells, biotinylated SARS-CoV-2 S, RBD, nucleocapsid (MiltenyiBiotec) and H1N1 (A/California/07/2009, SinoBiological) were incubated individually with fluorescently labeled SAV at 4:1 molar ratio for SARS-CoV-2 proteins and 6:1 for influenza antigen, with SAV added stepwise every 15 min at 4 °C for 1 h (refs. 22,54). The probes were mixed in 1:1 Brilliant Buffer (BD Bioscience) and FACS buffer (PBS with 2% FBS and 2 mM EDTA) with 5 μM of free d-biotin. We stained S, RBD, nucleocapsid (for tonsil samples), hemagglutinin (for tonsil samples) or a decoy probe using separate fluorochrome-conjugated SAVs. Frozen mononuclear cells were stained in 96-well U-bottom plates using ZombieUV Live-Dead staining (BioLegend) and TruStain FcX (1:200, BioLegend) in PBS for 30 min, followed by staining with the above-mentioned antigen-specific staining mix (200 ng S, 50 ng RBD, 100 ng nucleocapsid, 100 ng hemagglutinin and 20 ng SAV-decoy per color per 50 μl) at 4 °C for 1 h. Subsequently, cells were stained for 30 min with surface markers, followed by fixation and permeabilization with transcription factor staining buffer (eBioscience) at room temperature for 1 h and intracellular staining at room temperature for 30 min, before washing and acquisition. The antibodies used are listed in Supplementary Tables 5 and 7. Samples were acquired on a Cytek Aurora cytometer using the SpectroFlo software. The same positive control from a SARS-CoV-2-vaccinated healthy control was included in every experiment to ensure consistent results.

Flow cytometry analysis

Flow cytometry data were analyzed with FlowJo (version 10.8.0), with gating strategies shown in Extended Data Figs. 2 and 5. Subsets and markers of antigen-specific B cells and antigen-specific B cell subsets were evaluated only if more than nine or three specific cells per sample were detected, respectively. Dimensionality reduction and clustering analysis of flow cytometry data were performed in R using the CATALYST workflow (CATALYST package, version 1.18.1) (ref. 55). Markers were scaled with arcsinh transformation (cofactor 6,000), samples were subsetted to maximally 25 S+ Bm cells per sample. For UMAP representations and PhenoGraph clustering (Rphenograph package, version 0.99.1) (ref. 56), with k set to 20, the following B cell markers were used: CD11c, CD19, CD20, CD21, CD24, CD27, CD38, CD71, CD80, CXCR5, BAFF-R, FcRL5, IgA, IgD, IgG, IgM, Blimp1, IRF8, Ki67 and Tbet.

SARS-CoV-2-specific antibody measurement

Anti-SARS-CoV-2 antibodies were measured by a commercially available enzyme-linked immunosorbent assay specific for S1 of SARS-CoV-2 (Euroimmun SARS-CoV-2 IgG and IgA)57 or by a bead-based multiplexed immunoassay58.

Cell sorting for scRNA-seq, scRNA-seq and library preparation

scRNA-seq was performed on samples from nine patients of the SARS-CoV-2 Infection Cohort (Supplementary Table 2), three of the SARS-CoV-2 Vaccination Cohort, and paired blood and tonsil samples of four patients of the SARS-CoV-2 Tonsil Cohort (two recovered and two only vaccinated). Samples were stained as described for spectral flow cytometry using biotinylated SWT, RBD, Sbeta and Sdelta (MiltenyiBiotec) and hemagglutinin (SinoBiological) that were multimerized at 4:1 molar ratios with fluorescently labeled and/or barcoded SAV (TotalSeqC, BioLegend). Following 20 min staining with fixable viability dye eFluor 780 (eBioscience) and TruStain FcX and subsequently 1 h antigen-specific staining mix, cells were incubated at 4 °C for 30 min with a surface staining mix containing fluorescently labeled and barcoded antibodies, and each sample was marked with a hashtag antibody that allowed multiplexing (Supplementary Table 6). Cells were sorted on a FACS Aria III 4L sorter using the FACS Diva software. Antigen-specific cells per sample were sorted with 1,500–2,000 nonspecific B cells, as shown in Extended Data Figs. 2d and 6a. Sorted B cells were analyzed by scRNA-seq using the commercial 5′ Single Cell GEX and VDJ v1.1 platform (10x Genomics). After sorting, cell suspensions were pelleted at 400g for 10 min at 4 °C, resuspended and loaded into the Chromium Chip following the manufacturer’s instructions. Fourteen cycles (in one case 17) of initial cDNA amplification were used for all sample batches, and single-cell sequencing libraries for whole-transcriptome analysis (GEX), BCR profiling (VDJ) and TotalSeq (BioLegend) barcode detection (ADT) were generated. Final libraries were quantified using a Qubit Fluorometer, pooled at ratios of 5:1:1 or 10:1:1 (GEX:VDJ:ADT) and sequenced on a NovaSeq 6000 system.

Single-cell transcriptome analysis

Preprocessing of raw scRNA-seq data was done as described51. Briefly, FASTQ files were aligned to the human GRCh38 genome using Cell Ranger’s ‘cellranger multi’ pipeline (10x Genomics, v6.1.2) with default settings, which allowed one to process together the paired GEX, ADT and VDJ libraries for each sample batch. Downstream analysis was conducted in R version 4.1.0 mainly with the package Seurat (v4.1.1) (ref. 59). In the SARS-CoV-2 Infection Cohort, cells with fewer than 200 or more than 2,500 detected genes and cells with more than 10% detected mitochondrial genes were excluded from the analysis. In the SARS-CoV-2 Tonsil Cohort and SARS-CoV-2 Vaccination Cohort, cells with fewer than 200 or more than 4,000 detected genes were excluded from the analysis. For the SARS-CoV-2 Tonsil Cohort, we used a cutoff of 7.5% detected mitochondrial genes. Gene expression levels were log normalized using Seurat’s NormalizeData() function with default settings. Sample assignment of cells was done using TotalSeq-based cell hashing and Seurat’s HTODemux() function. When comparing dataset quality, we noticed a markedly lower median gene detection and unique molecular identifier count per cell in one of our datasets of the SARS-CoV-2 Infection Cohort. We associated this with an incident during sample preparation in one of our experiments and decided to exclude most cells of this dataset from the analysis.

As an internal reference for SHM counts in naïve B cells, we co-sorted naïve B cells in one experiment of the SARS-CoV-2 Infection Cohort. Naïve B cell clusters were identified on the basis of their surface protein expression of CD27, CD21 and IgD and their transcriptional levels of TCLA1, IL4R, BACH2, IGHD and BTG1. Independent datasets were then integrated using Seurat’s anchoring-based integration method. Gene expression data and TotalSeq surface proteome data were integrated separately. Seurat’s WNN analysis was used to take advantage of our multimodal approach during clustering and visualization59. Clustering was performed using the Louvain algorithm and a resolution of 0.4. For UMAP generation in the SARS-CoV-2 Infection Cohort datasets, the embedding parameters were manually set to a = 1.4 and b = 0.75. Differential gene expression analyses were done using assay ‘RNA’ of the integrated datasets. FindAllMarkers and FindMarkers functions were executed with logfc.thresholds set to 0.25 (0.1 for comparing resting Bm cells at month 6 versus month 12) and a min.pct cutoff at 0.1. Heat maps were generated using the ComplexHeatmap package (v2.13.1) or pheatmap package (v1.0.12) (ref. 60).

Gene set enrichment analysis (GSEA) was done as described51. Briefly, lists of differentially expressed genes were preranked in decreasing order by the negative logarithm of their P value, multiplied for the sign of their average log-fold change (in R, ‘-log(P_val)*sign(avg_log2FC)’). GSEA was performed on this preranked list using the R package fgsea (v.1.2). Gene sets were obtained from the Molecular Signatures Database (v7.5.1, collections H and C5) and loaded in R by the package msigdbr (v.7.5.1). To make the results reproducible, seed value was set (‘set.seed(42)’ in R) before execution. A multiple hypothesis correction procedure was applied to obtain adjusted P values. Results were filtered for gene sets that were significantly enriched with adjusted P < 0.05.

Gene set variation analysis with the package gsva (v1.42.0) was used to estimate gene set enrichments for more than two groups61. Transcriptomes of individual cells were used as inputs for the gsva() function with default parameters. Gene set enrichments for individual cells were summarized to patient pseudobulks by calculating mean enrichment values of cells belonging to the same patient. Pseudobulking was done only for patients with more than 20 cells in each cell subset. Resulting scores were used to compute fold changes and significance levels for enrichment score comparisons between cell subsets in limma (v3.50.3) (ref. 62).

Single-cell trajectories were created with Monocle3 (version 1.2.9) (ref. 63). Raw counts obtained from the cellranger gene expression matrix were used to create cell datasets, which were preprocessed using the Monocle 3 pipeline. Different batches were aligned using Batchelor (v.1.10.0) (ref. 64). The num_dim parameter of Monocle’s preprocess_cds() function was set to 20. Functions reduce_dimension(), order_cells() and graph_test() were executed with default parameters.

BCR analysis

B cell clonality analysis was performed mainly with the changeo-10x pipeline from the Immcantation suite65 using the singularity image provided by Immcantation developers. filtered_contig_annotations.csv files obtained from the cellranger multipipeline were used as input for the changeo-10x pipeline. Unique combinations of bases were appended to cell barcodes per batch before combining the data from different batches of sequencing to prevent cell barcode collisions. The clonality distance threshold was set to 0.20 for the longitudinal analysis of the SARS-CoV-2 Infection Cohort dataset and to 0.05 for the SARS-CoV-2 Tonsil Cohort dataset. Visualization of the clonal trees was done using dowser66. BCR variable gene segment usage was additionally quantified using the R package scRepertoire (v.1.3.5) (ref. 67). Clonal diversity between Bm cell subsets was investigated using the alphaDiversity function of Immcantations package Alakazam (v1.2.0) (ref. 65). Segment usage between Bm cell subsets was compared using edgeR (v3.36). For this, a count matrix was created with HC/LC segments as rows and samples as columns. Standard edgeR workflow was used to create a linear model for the count data and to conduct statistical tests for differential segment usage between Bm cell subsets.

Mapping of BCR sequences to antigen specificity

We used an adaptation of LIBRA-seq68 to identify antigen-specific cells in our sequencing data. Following subtraction of raw counts of baiting-negative control from those of all other antigen-baiting constructs in every cell, cutoffs for background binding levels were manually determined for every construct by inspection of bimodal distributions of count frequencies across all cells, and all binding counts below thresholds were set to zero and classified as nonbinding. Seurat’s centered log ratio transformation was applied across features, followed by a scaling of obtained values, resulting in final LIBRA scores. Cells with LIBRA scores >0 for the respective antigens were defined as antigen-specific, and in the SARS-CoV-2 infection, cohort cells were considered S+ if any of the antigens used for baiting (SWT, Sbeta, Sdelta, RBD) were defined as specific.

Statistical analysis

The number of samples and subjects and the statistical tests used in each experiment are indicated in the corresponding figure legends. All tests were performed two-sided. We did not assume normal distribution for the flow cytometry data and used nonparametric tests such as Kruskal–Wallis to test for differences between continuous variables in more than two groups, and P values were adjusted for multiple testing using Dunn’s method. For scRNA-seq data, distribution was assumed to be normal, but this was not formally tested. Statistical analysis was performed with GraphPad Prism (version 9.4.1, GraphPad Software, USA) and R (version 4.1.0). Statistical significance was established at P < 0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.