SARS-CoV-2 infection and vaccination elicit distinct pharyngeal mucosal B cell responses in children Authors: Qin Xu, Lihong Shi, Liya Wang, Liya Wang, Foo Cheung, Aparna Kotekar, Galina Koroleva, Elizabeth Rice, Richard Apps, Justin Lack, Craig Martens, Iyadh Douagi, Can Liu, Juraj Kabat, Hengameh Behzadpour, Lela Kardava, Tovah E. Markowitz, Margery Smelkinson, Kenneth B. Hoehn, Clarisa M. Buckner, Dominic P. Golec, Lorenza Bellusci, Gabrielle Grubbs, Sara Pourhashemi, Juanjie Tang, Asya Khleborodova, Martha Kirby, Rachel Sparks, Andrew J. Martins, John S. Tsang, Susan Moir, Surender Khurana, Pamela Mudd, Pamela L. Schwartzberg, Kalpana Manthiram Published: 2026 ABSTRACT Mucosal immunity is an important correlate of protection against respiratory infections such as SARS-CoV-2. Comparing B cell responses to vaccines and infection at relevant mucosal sites may provide unique and important insights into tissue immunity. Here, we characterize antigen-specific B cells in the tonsils, adenoids, and peripheral blood of children who had been infected with SARS-CoV-2 or vaccinated with SARS-CoV-2 mRNA vaccines. SARS-CoV-2-specific switched memory B cells (BSM) are found in the pharyngeal lymphoid tissues and blood after vaccination or infection. However, infection generates a higher proportion of IgA+ BSM and CXCR3+CD21+ BSM. CXCR3+CD21+ BSM show distinct spatial localization, greater clonal expansion and increased propensity for plasma cell differentiation compared to their CXCR3- counterparts, accompanied by persistent activation of innate and T follicular helper cells in the tissues. Our data provide evidence for tissue-specific B cell memory after either SARS-CoV-2 vaccination or infection, but with distinct characteristics that can influence the quality, durability, and localization of immunity. Introduction The high mortality and far-reaching effects of the COVID-19 pandemic triggered the rapid development of several vaccine platforms, including two mRNA-based vaccines, BNT162b2 (Pfizer) and mRNA-1273 (Moderna), which were shown to have high efficacy in preventing severe COVID-191,2. Although these vaccines generate high-titers of neutralizing serum antibodies and virus-specific memory B and T cells in the peripheral blood and lymphoid tissues3–5, immunity wanes with time, and population-based studies suggest that the durability of immune protection following vaccination may be lower than that from prior infection with SARS-CoV-26–9. Moreover, these vaccines do not elicit significant levels of mucosal IgA in the respiratory tract, which is a correlate of protection against COVID-1910–13. COVID-19 mRNA vaccines have been shown to generate memory B cells and GC B cells (GCB) in the draining axillary lymph nodes and various distal lymph nodes in humans14–18. However, the characteristics of SARS-CoV-2-specific memory B cells in the upper respiratory tract lymphoid tissues, where the host first encounters the virus, and how these memory B cells differ after vaccination versus infection remain incompletely resolved. The tonsils and adenoids are secondary lymphoid structures at the mucosal surface of the upper respiratory tract that contain lymphoid cells not found in the peripheral blood; viral-specific GCBs and tissue-resident memory T and B cells have been described in these mucosal tissues following respiratory infections19–21. The availability of these tissues provides the opportunity to deeply characterize SARS-CoV-2-specific B cell populations as well as evaluate the spatial locations of cells of interest within the tissue architecture22,23. Assessing SARS-CoV-2-specific-B cells in these tissues following intramuscular vaccination may, therefore, offer insights into the characteristics, magnitude, and durability of tissue immunity following intramuscular vaccination. Here, we characterize SARS-CoV-2-specific B cells in the tonsils, adenoids, and peripheral blood of children immunized with a SARS-CoV-2 mRNA vaccine alone compared with those of children previously infected with SARS-CoV-2, taking advantage of the unique period during the COVID-19 pandemic to study immunity to an unencountered airborne pathogen and new vaccines. We find SARS-CoV-2-specific B cells, which are primarily switched memory B cells (BSM) but also some GCBs, in the tonsils and adenoids of children both post-infection and post-vaccination, indicating that immune memory is found and maintained in the upper respiratory tract after either intramuscular vaccination or infection. Nonetheless, vaccination and infection generate BSM with different characteristics, including increased induction of a population defined by CXCR3 and CD21 expression post-infection. This CXCR3+CD21+ BSM population shows a greater propensity for plasma cell differentiation and mucosal homing, has a unique spatial distribution, and correlates with persistent adaptive and innate immune cell activation in mucosal lymphoid tissues. Our findings provide a framework for understanding responses to vaccination and infection, including novel parameters for assessing vaccine-induced mucosal immunity. Results SARS-CoV-2-specific B cells are found in the pharyngeal tissues and blood post-vaccination To assess SARS-CoV-2-specific immunity in vaccinated individuals, we collected serum, peripheral blood mononuclear cells (PBMC), and tonsil and adenoid tissues from 21 children undergoing tonsillectomy and/or adenoidectomy from December 2021 to September 2022 at Children’s National Hospital in Washington, DC, USA, who had received at least one dose of the monovalent mRNA vaccines, BNT162b2 or mRNA-1273 (Fig. 1a, Supplemental Data 1 and 2). These vaccines encode the ancestral Wuhan strain-like spike protein. Of these vaccinated subjects, 10 had no evidence of prior SARS-CoV-2 infection by history of positive PCR or antigen test, nor by positive serum titers against nucleocapsid (NC) and/or open reading frame 8 (ORF8), which have been used for serodiagnosis of SARS-CoV-2 infection24,25 (Supplemental Data 3). These 10 subjects comprised our vaccinated only cohort (VAC).Fig. 1SARS-CoV-2-specific B cells are found in the pharyngeal tissues and blood post-vaccination.a Study design and enrollment of CON (control), INF (post-infection), VAC (vaccinated) participants. Some elements of figure were created in BioRender. Manthiram, K. (2026) https://BioRender.com/ad6q0l3. S = spike, NC = nucleocapsid, ORF8 = open reading frame 8. b Time from vaccination(s) or infection to surgery. c Upper panel: serum neutralizing antibody titers against WA-1 and B1.1.159 (omicron, BA.1). N indicates the number of individual subjects per group (CON N = 12, INF N = 23, VAC N = 10). Lower panel: Proportion of subjects with indicated neutralizing antibodies in INF and VAC groups. d Correlation between serum neutralizing antibody titers to WA-1 or omicron BA.1 and days from last exposure to surgery (INF N = 10, VAC N = 10) using a two-sided Spearman’s rank correlation test. The solid line represents the linear regression fit with 95% confidence intervals shown as shaded area. e Representative flow cytometry plots of S1+RBD+ cells among BSM. Some elements of the figure were created in BioRender. Manthiram, K. (2026) https://BioRender.com/ad6q0l3. f Percentage of S1+RBD+ cells among CD19+ B cells (PBMC: CON N = 12, INF N = 15, VAC N = 10; adenoid: CON N = 10, INF N = 14, VAC N = 9; tonsil: CON N = 12, INF N = 22, VAC N = 7). (g) Percentage of S1+RBD+ cells among CD19+ B cells in matched PBMCs, adenoids, and tonsils from INF (N = 5) and VAC (N = 6) participants compared with paired two-sided Wilcoxon signed-rank test. h Correlations among serum neutralizing titers to WA-1 and percentage of S1+RBD+ B cells (PBMC INF N = 15, VAC N = 10; tonsil INF N = 22, VAC N = 7; adenoid INF = 14, VAC = 9). Correlations assessed with two-sided Spearman’s rank correlation test. Correlation coefficients (r values) are indicated by colors shown in the bar. Size of the circle indicates absolute r value. * p < 0.05, ** p < 0.01, *** p < 0.001. P values were obtained from linear model in panels c and f correcting for participant age (in blue) or from two-sided Mann-Whitney U test (in black). Box plots (c and f) show the median (center line) and interquartile range (25th–75th percentiles; box bounds), with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. P < 0.05 was considered significant. SARS-CoV-2-specific B cells are found in the pharyngeal tissues and blood post-vaccination. a Study design and enrollment of CON (control), INF (post-infection), VAC (vaccinated) participants. Some elements of figure were created in BioRender. Manthiram, K. (2026) https://BioRender.com/ad6q0l3. S = spike, NC = nucleocapsid, ORF8 = open reading frame 8. b Time from vaccination(s) or infection to surgery. c Upper panel: serum neutralizing antibody titers against WA-1 and B1.1.159 (omicron, BA.1). N indicates the number of individual subjects per group (CON N = 12, INF N = 23, VAC N = 10). Lower panel: Proportion of subjects with indicated neutralizing antibodies in INF and VAC groups. d Correlation between serum neutralizing antibody titers to WA-1 or omicron BA.1 and days from last exposure to surgery (INF N = 10, VAC N = 10) using a two-sided Spearman’s rank correlation test. The solid line represents the linear regression fit with 95% confidence intervals shown as shaded area. e Representative flow cytometry plots of S1+RBD+ cells among BSM. Some elements of the figure were created in BioRender. Manthiram, K. (2026) https://BioRender.com/ad6q0l3. f Percentage of S1+RBD+ cells among CD19+ B cells (PBMC: CON N = 12, INF N = 15, VAC N = 10; adenoid: CON N = 10, INF N = 14, VAC N = 9; tonsil: CON N = 12, INF N = 22, VAC N = 7). (g) Percentage of S1+RBD+ cells among CD19+ B cells in matched PBMCs, adenoids, and tonsils from INF (N = 5) and VAC (N = 6) participants compared with paired two-sided Wilcoxon signed-rank test. h Correlations among serum neutralizing titers to WA-1 and percentage of S1+RBD+ B cells (PBMC INF N = 15, VAC N = 10; tonsil INF N = 22, VAC N = 7; adenoid INF = 14, VAC = 9). Correlations assessed with two-sided Spearman’s rank correlation test. Correlation coefficients (r values) are indicated by colors shown in the bar. Size of the circle indicates absolute r value. * p < 0.05, ** p < 0.01, *** p < 0.001. P values were obtained from linear model in panels c and f correcting for participant age (in blue) or from two-sided Mann-Whitney U test (in black). Box plots (c and f) show the median (center line) and interquartile range (25th–75th percentiles; box bounds), with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. P < 0.05 was considered significant. We compared these vaccinated children to subjects we previously recruited from late 2020 to early 2021 who had prior SARS-CoV-2 infection (INF), before the availability of vaccination for children (Fig. 1a, Supplemental Data 1 and 2)21. All had mild or asymptomatic infection. Based on the timing of sample collection and known infection dates, most of these subjects were likely infected with the D614G or alpha strains, which bear sequence similarity to the spike mRNA used in the BNT162b2 or mRNA-1273 vaccines administered to the VAC cohort. The interval from infection to surgery among INF participants was comparable to the interval from the last vaccination to surgery in the VAC group (Fig. 1b, Supplementary Fig. 1a). We also included a group of unvaccinated pediatric controls (CON) with no serologic or cellular evidence of prior COVID-19, who were recruited during our initial study (Fig. 1a, Supplemental Data 2 and 3). To compare VAC and INF groups, we used both direct statistical comparisons (Mann-Whitney U) and linear models correcting for age. All VAC and INF participants had serum neutralizing antibodies to the WA-1 strain (Fig. 1c, Supplemental Data 3) with no significant differences noted between the two groups. Similar to our INF cohort, VAC subjects showed a trend towards lower neutralizing titers to WA-1 with greater time from infection (Fig. 1d), consistent with previous reports26. We also evaluated neutralizing titers to omicron, a variant that emerged in late 2021 and rapidly became the dominant strain, causing numerous breakthrough infections in those who were previously infected or vaccinated27,28. Although neutralizing titers were lower to omicron than to WA-1 in both groups, they were on average higher in VAC than INF, with a higher proportion of VAC subjects having positive titers (Fig. 1c, Supplemental Data 3). We then used fluorescently-labeled probes for the receptor-binding domain (RBD) and S1 portion of the spike protein from the original Wuhan strain to identify SARS-CoV-2-recognizing B cells (S1+RBD+) by flow cytometry (Fig. 1e). As previously reported, S1+RBD+ CD19+ B cells were found in both tissues and blood of most INF participants (Supplemental Data 3)21. Notably, almost all VAC subjects also had S1+RBD+ CD19+ B cells in both their pharyngeal tissues and blood (Fig. 1f, Supplemental Data 3). A higher frequency of S1+RBD+ cells was noted among B cells in the peripheral blood of VAC compared to INF individuals but these trended towards a higher frequency in the adenoids of INF subjects (Fig. 1f). These findings were confirmed by comparing S1+RBD+ B cells in the blood versus tissues within individual subjects by group (Fig. 1g). Among INF subjects, the percentage of S1+RBD+ B cells in the PBMCs, adenoids, and tonsils were all significantly correlated, as previously reported (Fig. 1h)21. In contrast, among VAC subjects, the percentage of S1+RBD+ B cells in the PBMCs correlated significantly with serum neutralizing titers, but not significantly with the percentages of S1+RBD+ B cells in the adenoid and tonsil (Fig. 1h). We also identified omicron-recognizing B cells (RBD-Omi+) using two fluorescently labeled RBD probes from the omicron strain (Supplementary Fig. 1b, c). VAC participants had a higher proportion of S1+RBD+ B cells that also recognized omicron in the blood, adenoid and tonsil compared to INF (Supplementary Fig. 1d), suggesting vaccination provides broader coverage of variants than infection in both the blood and tissues4,16,29. Thus, although vaccinated children had greater B cell responses in the peripheral blood compared to the mucosal tissues, we could detect SARS-CoV-2-specific B cells in the secondary lymphoid tissue of the upper respiratory tract, distal from the site of intramuscular immunization in nearly all vaccinated children. Infection induces a higher proportion of IgA+ SARS-CoV-2-specific BSM than vaccination To provide insight into SARS-CoV-2-specific B cells, we used a high-dimensional flow cytometry panel consisting of 29 markers, including fluorescently-labeled SARS-CoV-2 probes as well as surface markers, to describe B cell subsets, isotype, activation, tissue residence, and homing. In both INF and VAC, the majority of S1+RBD+ B cells in PBMCs, tonsils, and adenoids were BSM (Fig. 2a, b, defined as IgD- CD19+ B cells after exclusion of CD27hiCD38hi plasma cells in PBMCs, and IgD-CD38- CD19+ B cells in tissues, Supplementary Figs. 2 and 3). Unlike neutralizing titers which declined with greater time from vaccination/infection, the percentage of S1+RBD+ BSM in the tissues were stable (VAC) or increased (INF) with time (Fig. 2c).Fig. 2SARS-CoV-2-specific BSM cells have different characteristics post-vaccination and post-infection.a Gating strategy for switched memory B cells (BSM) in PBMCs and tonsils/adenoids (Supplementary Figs. 2, 3). b Composition (mean frequencies) of S1+RBD+ B cells and total CD19+ B cells in INF and VAC samples by flow cytometry. Samples with <25 S1+RBD+ B cells were excluded. PBMC: INF N = 11, VAC N = 9; adenoid: INF N = 14, VAC N = 8; tonsil: INF N = 22, VAC N = 6. USM = unswitched memory B cells, GCB = germinal center B cells, PC/PB = plasma cells/plasmablasts. c Correlation between percentage of S1+RBD+ cells among BSM in adenoid and tonsil and days from last exposure to surgery in INF and VAC groups (adenoid: INF N = 5, VAC N = 9; tonsil: INF N = 10, VAC N = 7) using a two-sided Spearman’s rank correlation test. The solid line represents the linear regression fit with 95% confidence intervals shown as the shaded area. d Immunoglobulin isotype composition among S1+RBD+ BSM cells and total BSM in PBMCs, adenoids, and tonsils of INF and VAC. Samples with <10 S1+RBD+ BSM were excluded. PBMC: INF N = 13, VAC N = 9; adenoid: INF N = 14, VAC N = 8; tonsil: INF N = 22, VAC N = 6. Mean frequencies are shown (Supplemental Data 4). e Percentages of IgA+, IgG+ and IgM+ cells among S1+RBD+ BSM (same samples as d). f Composition of conventional MBCs (cMBC, CD27+CD21+) and other subsets based on CD27 and CD21 expression among BSM. Upper panel: representative flow plot showing gating with CD21 and CD27. Samples with <10 S1+RBD+ BSM were excluded from this analysis (same samples as d). Mean percentages are shown (Supplemental Data 4). P values in e were obtained from linear model correcting for participant age (in blue) or from two-sided Mann-Whitney U test (in black). Box plot e shows the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. P < 0.05 was considered significant. SARS-CoV-2-specific BSM cells have different characteristics post-vaccination and post-infection. a Gating strategy for switched memory B cells (BSM) in PBMCs and tonsils/adenoids (Supplementary Figs. 2, 3). b Composition (mean frequencies) of S1+RBD+ B cells and total CD19+ B cells in INF and VAC samples by flow cytometry. Samples with <25 S1+RBD+ B cells were excluded. PBMC: INF N = 11, VAC N = 9; adenoid: INF N = 14, VAC N = 8; tonsil: INF N = 22, VAC N = 6. USM = unswitched memory B cells, GCB = germinal center B cells, PC/PB = plasma cells/plasmablasts. c Correlation between percentage of S1+RBD+ cells among BSM in adenoid and tonsil and days from last exposure to surgery in INF and VAC groups (adenoid: INF N = 5, VAC N = 9; tonsil: INF N = 10, VAC N = 7) using a two-sided Spearman’s rank correlation test. The solid line represents the linear regression fit with 95% confidence intervals shown as the shaded area. d Immunoglobulin isotype composition among S1+RBD+ BSM cells and total BSM in PBMCs, adenoids, and tonsils of INF and VAC. Samples with <10 S1+RBD+ BSM were excluded. PBMC: INF N = 13, VAC N = 9; adenoid: INF N = 14, VAC N = 8; tonsil: INF N = 22, VAC N = 6. Mean frequencies are shown (Supplemental Data 4). e Percentages of IgA+, IgG+ and IgM+ cells among S1+RBD+ BSM (same samples as d). f Composition of conventional MBCs (cMBC, CD27+CD21+) and other subsets based on CD27 and CD21 expression among BSM. Upper panel: representative flow plot showing gating with CD21 and CD27. Samples with <10 S1+RBD+ BSM were excluded from this analysis (same samples as d). Mean percentages are shown (Supplemental Data 4). P values in e were obtained from linear model correcting for participant age (in blue) or from two-sided Mann-Whitney U test (in black). Box plot e shows the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. P < 0.05 was considered significant. Mucosal IgA is protective against SARS-CoV-2 infection, but mRNA vaccination has been shown to engender less mucosal IgA in the respiratory tract compared to infection9,10,12,13. SARS-CoV-2-specific BSM were predominantly IgG+ post-infection and post-vaccination (Fig. 2d, Supplementary Figs. 2 and 3). However, INF individuals had a greater proportion of IgA+ cells (and less IgG+ cells) among S1+RBD+ BSM in PBMC and tissues compared to VAC subjects (Fig. 2d, e, Supplemental Data 4). Class-switched IgD- BSM in humans are often classified based on expression of CD21 and CD2730,31. CD27+CD21+ BSM are quiescent, highly affinity-matured, GC-derived cells, referred to as “conventional” or “classical” memory B cells32, which we denote cMBC. In both VAC and INF tissues and blood, most of the SARS-CoV-2-specific BSM were CD27+CD21+ cMBCs (Fig. 2f, Supplemental Data 4). SARS-CoV-2-specific BSM in the blood are phenotypically different post-infection and post-vaccination To further compare the phenotypes of S1+RBD+ B cells between the two groups in an unbiased manner, we concatenated the S1+RBD+ B cells from all VAC and INF participants and performed unsupervised clustering based on expression of 19 B cell surface markers. Cells from the peripheral blood and tissues were assessed separately due to differences in cell populations. Unsupervised clustering of S1+RBD+ B cells from PBMCs generated 12 phenotypically distinct clusters, 11 of which were BSM subsets (Fig. 3a–c, Supplemental Data 5, Supplementary Fig. 2). VAC and INF samples largely segregated on principal component analysis (PCA) on PC2; a few VAC subjects were distinct on PC1 (Fig. 3d). We compared the proportions of each cluster in the two groups by Mann-Whitney U or a linear model correcting for age (Fig. 3c, e; Supplementary Fig. 4a). We found that INF subjects had higher proportions of S1+RBD+ B cells in several BSM clusters that expressed CXCR3 and CD21. CXCR3 is a chemokine receptor induced by IFN-γ that directs BSM to areas of inflammation33. Cluster 10, the most significant cluster, represented IgA+CXCR3+CD21+ BSM. Two other clusters (clusters 2 and 6, representing IgG+ cells) showed a trend towards increased percentages in INF compared to VAC individuals (Fig. 3b, e–h).Fig. 3Unsupervised analyses of S1+RBD+ B cells in the peripheral blood.a Uniform manifold approximation and projection (UMAP) of unsupervised clustering based on surface markers from flow cytometric analysis of 4348 S1+RBD+ CD19+ B cells of PBMC from both INF (1275 cells) and VAC (3073 cells) groups. Clusters enriched in VAC group are indicated in yellow and INF group in green in the legend. b Heatmap of marker/antibody expression in each cluster. c Mean proportion of each cluster across subjects among S1+RBD+ B cells in each group. d Principal component analysis (PCA) of each subject based on cluster frequencies. Each point represents an individual and is colored by group. e Comparison of frequencies of each cluster among S1+RBD+ B cells by group. Differentially distributed clusters are shown; data from all clusters are shown in Supplementary Fig. 4a. Box plots show the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. P values obtained from linear model correcting for participant ages (in blue) and from two-sided Mann-Whitney U test (in black) are shown. P < 0.05 was considered significant. f Distribution of S1+RBD+ B cells among clusters by group on UMAP. g Densities of S1+RBD+ B cells in each cluster by group, with density shading indicating the concentration of cells in different areas of UMAP. Color represents cell density. h Heatmaps of selected individual marker/antibody expression overlayed on UMAP. N denotes the number of individual subjects per group: INF N = 15; VAC N = 10(Supplemental Data 5). Unsupervised analyses of S1+RBD+ B cells in the peripheral blood. a Uniform manifold approximation and projection (UMAP) of unsupervised clustering based on surface markers from flow cytometric analysis of 4348 S1+RBD+ CD19+ B cells of PBMC from both INF (1275 cells) and VAC (3073 cells) groups. Clusters enriched in VAC group are indicated in yellow and INF group in green in the legend. b Heatmap of marker/antibody expression in each cluster. c Mean proportion of each cluster across subjects among S1+RBD+ B cells in each group. d Principal component analysis (PCA) of each subject based on cluster frequencies. Each point represents an individual and is colored by group. e Comparison of frequencies of each cluster among S1+RBD+ B cells by group. Differentially distributed clusters are shown; data from all clusters are shown in Supplementary Fig. 4a. Box plots show the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. P values obtained from linear model correcting for participant ages (in blue) and from two-sided Mann-Whitney U test (in black) are shown. P < 0.05 was considered significant. f Distribution of S1+RBD+ B cells among clusters by group on UMAP. g Densities of S1+RBD+ B cells in each cluster by group, with density shading indicating the concentration of cells in different areas of UMAP. Color represents cell density. h Heatmaps of selected individual marker/antibody expression overlayed on UMAP. N denotes the number of individual subjects per group: INF N = 15; VAC N = 10(Supplemental Data 5). Conversely, we found that VAC subjects had a higher portion of S1+RBD+ B cells in several CD21lo clusters, including clusters 1, 3, 4 and 7 (CD21loIgG+ BSM), and cluster 11 (a CD21loIgA+ BSM cluster found in 2 subjects) (Fig. 3a-c, e-h, Supplemental Data 5). Clusters 4, 7 and 11 expressed CD11c, CD85J, FCRL3/5 and CD95, markers associated with a CD27-IgD- double negative (DN) population that is also CD21- and CD11c+ and refered to as DN2 cells. DN2 cells (also called “atypical” memory B cells) are expanded in chronic infections and autoimmune conditions but are also noted after acute infection and vaccination including after COVID-19 vaccination34–44. DN2 cells have been shown to express the transcription factor Tbet, which can promote expression of CXCR345,46. However, these CD21- clusters did not express CXCR3 (Fig. 3b). Cluster 3 was an IgG+ CD27+CD21− cluster that expressed variable CXCR3, CD11c, CD71, and FCRL3/5. Antigen-specific CD27+CD21lo/- BSM populations expressing FCRL5, CD11c, CD71 and/or Tbet have been noted after vaccination and, in one study, found to contribute to long-term recall responses47–50. In contrast, the three CXCR3+ clusters enriched in subjects post-infection did not express CD11c, CD95, FCRL4, or FCRL3/5 and were largely CD21+ (Fig. 3a, b, h)5,39. Thus, unbiased analyses revealed phenotypic differences in SARS-CoV-2-specific B cells in the peripheral blood post-infection and post-vaccination, including enrichment of CXCR3+CD21+ BSM post-infection. Distinct SARS-CoV-2-specifc GCBs and BSM are in the pharyngeal tissues of INF and VAC subjects We next examined the characteristics of SARS-CoV-2-specific B cells in the pharyngeal tissues, generating 14 clusters representing naïve and unswitched memory B cells (USM), GC and pre-GC B cells (CD38+CD10+CD95+CD71+), and BSM (Fig. 4a–c, Supplementary Fig. 3, Supplementary Fig. 4b, c, Supplemental Data 6). Like PBMCs, PCA analyses showed separation between VAC and INF subjects (Fig. 4d).Fig. 4Unsupervised analyses of S1+RBD+ B cells in the adenoid and tonsil.a UMAP of unsupervised clustering based on surface markers from flow cytometric analysis of 20,801 S1+RBD+ CD19+ B cells from adenoid and tonsil of both INF (14,627 cells) and VAC (6174 cells) subjects. Clusters enriched in VAC group are indicated in yellow and INF group in green in legend. b Heatmap of marker/antibody expression in each cluster. c Mean proportion of each cluster across subjects among S1+RBD+ B cells in INF and VAC tissues. d PCA of each subject based on cluster frequency. Each point represents an individual and is colored by group. e Densities of S1+RBD+ B cells in each cluster by group, with density shading indicating the concentration of cells in different parts of the UMAP. Color represents cell densities. f Heatmaps of selected individual marker/antibody expression overlayed on UMAP. g Comparisons of cluster frequencies among S1+RBD+ B cells in INF and VAC groups according to tissue type. Differentially distributed clusters are shown; data from all clusters are shown in Supplementary Fig. 4d. h Percentage of CXCR3+IgA+ or CXCR3+IgG+ cells among S1+RBD+ BSM in PBMCs, adenoid, and tonsils of INF and VAC participants (PBMC INF N = 15, VAC N = 10; adenoid INF N = 14, VAC = 9; tonsil INF N = 22, VAC N = 7). i Frequency of subjects with ≥10 S1+RBD+ GCB cells in at least one tissue in INF and VAC participants (see Supplemental Data 7 for details). For h and g, p values were obtained from linear model correcting for participant ages (in blue) and from two-sided Mann-Whitney U test (in black). P < 0.05 was considered significant. Box plots (h, g) show the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. N denotes the number of individual subjects per group. For all panels except h, adenoid: INF N = 14, VAC N = 9; tonsil: INF N = 22, VAC N = 7 (Supplemental Data 6). Unsupervised analyses of S1+RBD+ B cells in the adenoid and tonsil. a UMAP of unsupervised clustering based on surface markers from flow cytometric analysis of 20,801 S1+RBD+ CD19+ B cells from adenoid and tonsil of both INF (14,627 cells) and VAC (6174 cells) subjects. Clusters enriched in VAC group are indicated in yellow and INF group in green in legend. b Heatmap of marker/antibody expression in each cluster. c Mean proportion of each cluster across subjects among S1+RBD+ B cells in INF and VAC tissues. d PCA of each subject based on cluster frequency. Each point represents an individual and is colored by group. e Densities of S1+RBD+ B cells in each cluster by group, with density shading indicating the concentration of cells in different parts of the UMAP. Color represents cell densities. f Heatmaps of selected individual marker/antibody expression overlayed on UMAP. g Comparisons of cluster frequencies among S1+RBD+ B cells in INF and VAC groups according to tissue type. Differentially distributed clusters are shown; data from all clusters are shown in Supplementary Fig. 4d. h Percentage of CXCR3+IgA+ or CXCR3+IgG+ cells among S1+RBD+ BSM in PBMCs, adenoid, and tonsils of INF and VAC participants (PBMC INF N = 15, VAC N = 10; adenoid INF N = 14, VAC = 9; tonsil INF N = 22, VAC N = 7). i Frequency of subjects with ≥10 S1+RBD+ GCB cells in at least one tissue in INF and VAC participants (see Supplemental Data 7 for details). For h and g, p values were obtained from linear model correcting for participant ages (in blue) and from two-sided Mann-Whitney U test (in black). P < 0.05 was considered significant. Box plots (h, g) show the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. N denotes the number of individual subjects per group. For all panels except h, adenoid: INF N = 14, VAC N = 9; tonsil: INF N = 22, VAC N = 7 (Supplemental Data 6). Similar to PBMCs, VAC tissues had a higher frequency of S1+RBD+ B cells in certain clusters containing CD27+IgG+ BSM with lower or intermediate CD21 expression and low or mixed CXCR3 expression. These included a greater proportion in cluster 2 in VAC tonsils and in cluster 7 (CD62L+) in VAC adenoids and tonsils compared to INF (Fig. 4a–c, e–g; Supplementary Fig. 4c, d). However, these cells did not express high levels of CD11c, FCRL3/5, or FCRL4. In contrast, tonsils and adenoids of INF subjects contained a higher fraction of SARS-COV-2-specific B cells in several CXCR3+CD21+ containing clusters compared to VAC individuals, including IgA+ (clusters 4 and 12) and IgG+ (clusters 5 and 10 in the adenoids and cluster 3 in tonsils) BSM populations that were largely CD27+ (Fig. 4a–c, e–g, Supplementary Fig. 4c, d). Increased percentages of CXCR3+ IgA+ and IgG+ cells in INF subjects were confirmed by manual gating (Fig. 4h). CXCR3+ has been shown to be important for establishment of tissue-residence among memory B cells in murine lungs and production of mucosal IgA in mice, in addition to attracting cells to the mucosa51–53. Although we also observed CXCR3+ BSM in the blood, we saw a trend towards higher proportions of CXCR3+ cells, particularly CXCR3+IgA+ cells, among S1+RBD+ BSM in the adenoid and tonsil compared to blood, supporting the idea that CXCR3+ BSM may preferentially home to mucosal tissues (Supplementary Fig. 4e). Finally, as we previously reported, S1+ cells were observed in GCB clusters in the pharyngeal tissues of most (15/24, 62.5%) INF individuals. However, two out of 10 (20%) VAC subjects also had at least 10 S1+RBD+ B cells within GCB clusters in either the tonsil or adenoid, a threshold based on a recent study18 (Fig. 4i; Supplemental Data 7). These two VAC subjects had their most recent (second) vaccine doses 48 days and 232 days prior to surgery, suggesting this was not temporally associated with recent vaccination (Supplemental Data 2). However, whether these individuals had a transient or abortive SARS-CoV-2 infection without mounting a systemic serum antibody response or had waning NC/ORF8 antibody responses is unknown. Nonetheless, INF tonsils had a higher proportion of S1+RBD+ B cells in the IgG+ GCB cluster (cluster 6) and IgA+ GCB cluster (cluster 11, by Mann-Whitney U) than VAC (Fig. 4a–c, e–g; Supplementary Fig. 4d), supporting the role of natural infection in generating and maintaining localized GC responses in pharyngeal mucosal tissues. Thus, infection and vaccination both gave rise to memory B cell populations in the oro-pharyngeal lymphoid tissues, but were associated with phenotypically distinct memory B cell populations, with higher proportions of CXCR3+CD21+ BSM, including CXCR3+IgA+ cells and of SARS-CoV-2-specific GCB cells post-infection. Evaluation of BSM based on CXCR3 and CD21 expression Given the differential expression of CXCR3 and CD21 on S1+RBD+ BSM from INF and VAC individuals, we used these two markers to define four populations among BSM: P1 (CD21+CXCR3-); P2 (CD21+CXCR3+); P3 (CD21-CXCR3+); and P4 (CD21-CXCR3-) (Fig. 5a). The proportion of P2 (CD21+CXCR3+) cells among S1+RBD+ BSM was significantly higher in INF PBMCs and tissues compared to VAC (Fig. 5a, b). In contrast, VAC individuals had a higher proportion of CXCR3- populations, P4 in blood and P1 and P4 in tissues, highlighting differences in CXCR3 expression between antigen-specific B cells in INF and VAC subjects.Fig. 5Evaluation of BSM based on CXCR3 and CD21 expression.a Left: Representative gating of BSM by CXCR3 and CD21 expression yielding 4 populations (P1-P4). Right: Proportions of P1 (CXCR3-CD21+), P2 (CXCR3+CD21+), P3 (CXCR3+CD21-), and P4 (CXCR3-CD21-) among S1+RBD+ BSM and total BSM in VAC and INF subjects. Samples with <10 S1+RBD+ BSM were excluded. N denotes the number of individual subjects per group (PBMC: INF N = 13, VAC N = 9; adenoid: INF N = 14, VAC N = 8; tonsil: INF N = 22, VAC N = 6). Mean frequencies are plotted (Supplemental Data 4). b Comparison of percentage of P1-P4 among S1+RBD+ BSM in PBMCs, adenoids, and tonsils of VAC versus INF subjects. Same samples as a. Histograms of selected marker expression on P1-P4 subsets among BSM in PBMC (c) and tonsil (d) sample. Representative sample among PBMC (N = 3) and tonsil (N = 3) are shown. Samples listed in Supplemental Data 2. e Correlation between percentages of P1-P4 among S1+RBD+ BSM in PBMCs, adenoids and tonsils and serum neutralizing titers to WA.1. Correlations were assessed with a two-sided Spearman’s rank correlation test. Correlation coefficients (r values) are indicated by colors shown in the bar and size of the circle indicates absolute r value. INF and VAC groups combined for analysis. Samples in panel a with available neutralizing titers were analyzed: PBMC N = 22, adenoid N = 21, tonsil N = 27. * p < 0.05, ** p < 0.01. f Percentage of CD69+ cells among S1+RBD+ BSM and total BSM in PBMCs, adenoids, and tonsils of VAC versus INF subjects. Same samples as a. In b and f, p values were obtained from linear model correcting for participant ages (in blue) and from two-sided Mann-Whitney U test (in black). P < 0.05 were considered significant. Box plots (b and f) show the median (center line), interquartile range (25th–75th percentiles; box bounds), with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. Evaluation of BSM based on CXCR3 and CD21 expression. a Left: Representative gating of BSM by CXCR3 and CD21 expression yielding 4 populations (P1-P4). Right: Proportions of P1 (CXCR3-CD21+), P2 (CXCR3+CD21+), P3 (CXCR3+CD21-), and P4 (CXCR3-CD21-) among S1+RBD+ BSM and total BSM in VAC and INF subjects. Samples with <10 S1+RBD+ BSM were excluded. N denotes the number of individual subjects per group (PBMC: INF N = 13, VAC N = 9; adenoid: INF N = 14, VAC N = 8; tonsil: INF N = 22, VAC N = 6). Mean frequencies are plotted (Supplemental Data 4). b Comparison of percentage of P1-P4 among S1+RBD+ BSM in PBMCs, adenoids, and tonsils of VAC versus INF subjects. Same samples as a. Histograms of selected marker expression on P1-P4 subsets among BSM in PBMC (c) and tonsil (d) sample. Representative sample among PBMC (N = 3) and tonsil (N = 3) are shown. Samples listed in Supplemental Data 2. e Correlation between percentages of P1-P4 among S1+RBD+ BSM in PBMCs, adenoids and tonsils and serum neutralizing titers to WA.1. Correlations were assessed with a two-sided Spearman’s rank correlation test. Correlation coefficients (r values) are indicated by colors shown in the bar and size of the circle indicates absolute r value. INF and VAC groups combined for analysis. Samples in panel a with available neutralizing titers were analyzed: PBMC N = 22, adenoid N = 21, tonsil N = 27. * p < 0.05, ** p < 0.01. f Percentage of CD69+ cells among S1+RBD+ BSM and total BSM in PBMCs, adenoids, and tonsils of VAC versus INF subjects. Same samples as a. In b and f, p values were obtained from linear model correcting for participant ages (in blue) and from two-sided Mann-Whitney U test (in black). P < 0.05 were considered significant. Box plots (b and f) show the median (center line), interquartile range (25th–75th percentiles; box bounds), with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. We then evaluated the expression of additional markers including transcription factors in each population in tonsil and blood. We found that markers associated with “atypical”/DN2 memory B cells (defined as IgD-CD27-CD21-CD11c+ B cells), including T-bet, CD11c, CD95, FCRL4, and FCRL3/5, were expressed primarily in the CD21 negative populations P3 and P4, with the highest expression in P3 (Fig. 5c-d). Nevertheless, backgating revealed that the P1-P4 populations did not fall into traditionally described subsets of memory B cells based on CD27 and CD21 (Supplementary Fig. 5a). Of note, the frequency of P3 among S1+RBD+ BSM correlated with neutralizing titers to WA.1 (Fig. 5e). In contrast, P1 and P2, the CD21+ populations, were largely CD27+ cMBC; they also expressed high levels of CXCR5, a marker associated with follicular localization (Fig. 5c-d; Supplementary Fig. 5a). However, most cells in P2 expressed a low level of T-bet, suggestive of previous exposure to IFN-γ that may have driven and maintained expression of CXCR353,54. CD69 is a marker of tissue resident memory B cells (BRM) that are poised to provide rapid local protection in response to infection or immune challenges55. In tonsil and adenoid BSM, we found that P2 had the highest frequency of CD69+ cells, suggesting that this population is enriched for BRM (Fig. 5d, Supplementary Fig. 5b). Although CD69+ S1+RBD+ BSM were noted post-infection and post-vaccination, the percentage of CD69+ cells among S1+RBD+ BSM was significantly higher in INF tonsils compared to VAC (Fig. 5f, Supplemental Data 4). Thus, CXCR3+CD21+ SARS-CoV-2-specific BSM with BRM features were enriched in upper respiratory tract lymphoid tissue after infection compared to vaccination. Tissue CXCR3+CD21+ S1+ BSM have distinct transcriptomic features To further characterize these populations, we sorted S1+ and S1- B cells from tonsils and/or adenoids of a subset of 8 INF, 4 VAC, and 2 CON subjects (Supplemental Data 8), including the two VAC subjects in whom S1+ GCBs were identified by flow cytometry, and performed Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq) using a panel of 20 antibodies including anti-CXCR3 and CD21 (Supplemental Data 11). In total, 9370 S1+ and 172919 S1- B cells were captured and assessed for surface protein expression, gene expression and BCR sequencing at the single cell level. Unsupervised clustering based on surface protein expression revealed 6 clusters representing naïve/USM B cells, BSM, GCB, and plasma cells/plasmablasts (PC/PB) (Fig. 6a-b). S1+ B cells were predominantly in a BSM cluster (cluster 1) in both INF and VAC tissues (Fig. 6c-d). However, a portion of S1+ B cells in both INF and VAC subjects were in cluster 2, which had a gene expression signature consistent with GCBs (Fig. 6d-e; Supplementary Fig. 6a).Fig. 6Transcriptomic features of tissue S1+ BSM in INF and VAC individuals.a UMAP of unsupervised clustering based on surface protein expression from CITE-seq of sorted S1+ and S1- B cells from tonsils/adenoids. CD38 and IgD surface expression (b) and distribution of S1+ B cells (c) are shown. d Proportion of clusters among INF and VAC subjects. e GCB signature gene module scores in S1+ BSM and S1+ GCB from INF versus VAC (two-sided Mann-Whitney U test). f CXCR3 and CD21 expression on concatenated S1+ and S1- BSM. g Proportion of B cell subsets (gated by surface markers) among S1+ and S1- B cells. h Trajectory analysis of S1+ B cells from INF and VAC subjects using Slingshot. i Heatmap showing gene enrichment scores for selected pathways in S1- B cell subsets. Color represents z score. MBC = memory B cell; atMBC = atypical memory B cell. j Differentially expressed genes (DEGs) identified by comparing pooled pseudobulk libraries of P1 versus P2 S1+ BSM with a linear model (paired). Top 15 genes with adjusted p value < 0.05 are labeled (Supplemental Data 9). Voom–limma with empirical Bayes–moderated linear models was used with FDR-adjusted two-sided moderated t-tests. k Select pathways from GSEA of P2 versus P1 S1+ BSM. Genes were ranked by limma empirical Bayes–moderated t-statistics. Enrichment significance was assessed at both ends of the ranked gene list by permutation testing (fgsea). NES and Benjamini–Hochberg-adjusted p values reported. Dot size denotes NES and color denotes −log10 (adjusted p value). l Differentially accessible regions (DARs) between P1 BSM and P2 BSM with FDR < 0.1. Top 20 peaks are labeled with corresponding gene names (Supplemental Data 10). m Proportion of IRF4-binding complex motifs (AICE, ISRE, and EICE) among DARs differentially enriched in either P1 or P2 (two-tailed Fisher’s exact test). For a–d, g, j, N denotes the number of subjects per group (adenoid: INF N = 6, VAC N = 3, CON N = 1; tonsil: INF N = 6, VAC N = 2, CON N = 2; Supplemental Data 8). For e, h and I, only CITE-seq set 306-4 was used. In d and g, mean proportions among each group of subjects are plotted. Transcriptomic features of tissue S1+ BSM in INF and VAC individuals. a UMAP of unsupervised clustering based on surface protein expression from CITE-seq of sorted S1+ and S1- B cells from tonsils/adenoids. CD38 and IgD surface expression (b) and distribution of S1+ B cells (c) are shown. d Proportion of clusters among INF and VAC subjects. e GCB signature gene module scores in S1+ BSM and S1+ GCB from INF versus VAC (two-sided Mann-Whitney U test). f CXCR3 and CD21 expression on concatenated S1+ and S1- BSM. g Proportion of B cell subsets (gated by surface markers) among S1+ and S1- B cells. h Trajectory analysis of S1+ B cells from INF and VAC subjects using Slingshot. i Heatmap showing gene enrichment scores for selected pathways in S1- B cell subsets. Color represents z score. MBC = memory B cell; atMBC = atypical memory B cell. j Differentially expressed genes (DEGs) identified by comparing pooled pseudobulk libraries of P1 versus P2 S1+ BSM with a linear model (paired). Top 15 genes with adjusted p value < 0.05 are labeled (Supplemental Data 9). Voom–limma with empirical Bayes–moderated linear models was used with FDR-adjusted two-sided moderated t-tests. k Select pathways from GSEA of P2 versus P1 S1+ BSM. Genes were ranked by limma empirical Bayes–moderated t-statistics. Enrichment significance was assessed at both ends of the ranked gene list by permutation testing (fgsea). NES and Benjamini–Hochberg-adjusted p values reported. Dot size denotes NES and color denotes −log10 (adjusted p value). l Differentially accessible regions (DARs) between P1 BSM and P2 BSM with FDR < 0.1. Top 20 peaks are labeled with corresponding gene names (Supplemental Data 10). m Proportion of IRF4-binding complex motifs (AICE, ISRE, and EICE) among DARs differentially enriched in either P1 or P2 (two-tailed Fisher’s exact test). For a–d, g, j, N denotes the number of subjects per group (adenoid: INF N = 6, VAC N = 3, CON N = 1; tonsil: INF N = 6, VAC N = 2, CON N = 2; Supplemental Data 8). For e, h and I, only CITE-seq set 306-4 was used. In d and g, mean proportions among each group of subjects are plotted. Comparison of surface protein to mRNA levels revealed that most CXCR3+ B cells had little or no detectable CXCR3 transcripts (Supplementary Fig. 6b). We, therefore, defined BSM populations using CXCR3 and CD21 surface markers (P1-P4) (Fig. 6f). As expected, we saw a higher proportion of P2 among S1+ B cells post-infection and higher proportions of P1 and P4 (both CXCR3-) post-vaccination (Fig. 6g, Supplementary Fig. 6c). Trajectory analyses using Slingshot suggested distinct patterns of S1+ BSM development in INF and VAC conditions (Fig. 6h). In INF tissues, two branches originating from naive B cells both landed at the P2 state, while in VAC tissues, two diverging trajectories from naïve B cells emerged, including one leading to the P4 population (Fig. 6h). By PCA, the CD21+ BSM populations (P1 and P2) segregated distinctly from the CD21- populations (P3 and P4) on PC1 (Supplementary Fig. 6d). Both CD21- populations (P3 and P4) had high expression of “atypical”/DN2 B cell signature genes including TBX21, FCRL4, NKG7, ITGAX, and IL2RB44,56 as well as HCK and FGR, two SRC family kinases linked to integrin signaling (Supplementary Fig. 6e)57. P3 and P4 also had high expression of gene sets related to “atypical”/DN2 cells, as well as antigen presentation and processing and type I and type II interferon responses, which have been previously reported to be characteristic of these populations (Fig. 6i)34,58,59. In contrast, both CD21+ populations (P1 and P2) in tissues exhibited features of cMBCs, including expression of CR2 and FCER2 and expressed high levels of the tissue-residence marker CD69 (Fig. 6i, Supplementary Fig. 6e)58. We then focused on differences between the CXCR3+ and CXCR3- CD21+ BSM (P2 and P1, respectively), which constituted the majority of the S1+ cells. Differentially expressed genes (DEG) in P1 versus P2 S1+ BSM revealed that cells in P2 expressed higher levels of TCF7, a transcription factor associated with stem-like central memory fate and found in cMBCs34,48 (Fig. 6j, Supplemental Data 9). Upon secondary challenge, BSM in the lymphoid tissue can preferentially differentiate to PB/PC or re-enter GCs based on their transcriptomic and epigenetic profile60. DEG in P1 versus P2 S1⁺ BSM suggested distinct cell fates upon rechallenge. P2 S1+ BSM expressed higher FOS, an AP-1 transcription factor family member expressed highly in pre-plasmablasts61, and expressed lower levels of BACH2, a transcription factor that represses BLIMP1 expression and plasma cell differentiation and is found in memory B cells that preferentially differentiate to GCBs60 (Fig. 6j, Supplemental Data 9). S1+ P2 BSM also expressed lower IL21R, which is critical for GCB differentiation, maintenance, and proliferation62–64 and the GCB marker MEF2B (Fig. 6j, Supplemental Data 9)65. Moreover, pathway analyses revealed enrichment of genes involved PC differentiation, translation, interferon response, antigen presentation and oxidative metabolism pathways, with lower cell cycle-associated genes in P2 compared to P1 S1+ BSM (Fig. 6k). Together, these differences suggest that P2 BSM are better poised to differentiate to plasma cells compared to P1. We further compared the epigenetic profiles of sorted bulk P1, P2, P3 and P4 BSM from three tonsils using ATAC-seq. Again, CD21+ (P1 and P2) and CD21- (P3 and P4) populations clustered distinctly on PCA (Supplementary Fig. 6f), with P3 and P4 enriched in motifs related to AP-1 transcription factors (FOS, BATF, JUN) that are associated with “atypical”/DN2 BSM and PC/PB differentiation65 (Supplementary Fig. 6g). Direct comparison of the CD21+ groups, P1 and P2, revealed that P1 had greater overall open chromatin regions compared to P2 with a few notable exceptions including the TCF7 locus, which was more accessible in P2, consistent with the higher TCF7 expression noted in the transcriptome (Fig. 6l, Supplemental Data 10). The enriched signaling pathways associated with differentially accessible regions in P1 versus P2 mirrored those identified in the transcriptomic analysis, including those downstream of IFN-γ, BCR, and antigen receptors (Fig. 6k, Supplementary Fig. 6h). IRF4 and PRDM1 encode transcription factors essential for PC/PB differentiation66; chromatin accessibility at these loci in P1 and P2 BSM were similar and distinct from plasma cells and P3/P4 (Supplementary Fig. 6i). However, differentially accessible regions in P2 compared with P1 had a greater proportion of AP-1-IRF composite (AICE) and interferon sequence response element (ISRE) binding motifs, which are IRF4-binding complex motifs reported to favor plasma cell differentiation. In contrast, the proportion of the Ets-IRF composite (EICE) motif, which is also an IRF4-binding complex motif but not involved in plasma cell differentiation, was comparable between P2 and P1 enriched peaks66,67 (Fig. 6m). This enrichment pattern suggests the P2 BSM population is primed for IRF4-mediated transcriptional changes driving plasma cell differentiation. Our findings suggest that among the CD21+ BSM populations, P2 are transcriptionally and epigenetically predisposed to antibody-secreting cell differentiation upon antigenic re-exposure compared to P1. Infection induces greater clonal expansion among S1+ B cells in pharyngeal tissues To further understand differences between VAC and INF responses, we assessed isotype, somatic hypermutation, clonal expansion, and clonal overlap by BCR sequencing of the sorted S1+ and S1- B cells characterized by CITE-seq. Single-cell BCR sequencing confirmed that the majority of S1+ B cells were IgG1-class switched cells in both VAC and INF tissues, with a greater proportion of IgA1-expressing cells across all BSM subsets (P1-P4) post-infection (Supplementary Fig. 7a). A greater proportion of S1+ B cells were part of expanded clones in INF compared to VAC tissues (Fig. 7a). S1+ B cells from INF tissues displayed lower clonal diversity compared to S1- B cells and also to S1+ B cells in VAC tissues, consistent with greater antigen-driven expansion in tissues post-infection compared to post-vaccination (Fig. 7b). Furthermore, in INF tissues, P2 S1+ BSM had a trend towards a higher level of expansion than P1 S1+ BSM; this expansion was not seen in P2 S1+ BSM post-vaccination (Fig. 7c).Fig. 7Single-cell BCR sequencing of S1+ B cells in INF and VAC groups.a Proportion of expanded (clone size > 1 cell) clones among S1+ CD19+ B cells from INF and VAC participants. b Simpson’s diversity of S1+ and S1− B cells from adenoids and tonsils from 8 INF and 4 VAC donors and S1− B cells from 2 CON. Adenoid and tonsil samples were assessed independently. Lower Simpson’s diversity indicates a greater frequency of large clones. c Clonal expansion index of S1+ B cells in each B cell subset in INF and VAC tissues. d Somatic hypermutation (SHM) frequency of S1+ or S1- B cell subsets. Each dot represents one cell. e Pairwise transition index (pTrans) showing degree of BCR clone overlap between P2 S1+ BSM and S1+ B cell subsets from adenoids and tonsils. f Heatmap showing clonal overlap between S1+ or S1- B cell subsets from one representative donor who had the largest number of sorted S1+ B cells (CNMC 70, 2413 S1+ B cells). Off-diagonal elements are colored by the Jaccard index indicating degree of clonal overlap and are labeled by the raw number of overlapping clones. Diagonal elements are labeled by the total number of clones within a particular subset. g, h Clonal lineage trees selected from the two largest GCB-bearing S1+ trees from INF (h) and VAC (i) donors. Triangles indicate S1+ cells and tip color indicates B cell subset. Isotype and tissue origin of each clone are listed next to the symbol. Branch lengths represent SHM frequency/codon in VDJ sequence according to the scale bar. For a–e, N denotes the number of subjects per group (adenoid: INF N = 6, VAC N = 3, CON N = 1; tonsil: INF N = 6, VAC N = 2, CON N = 2, listed in Supplemental Data 8). Significance for panels b-d calculated with two-sided Mann-Whitney U test. P < 0.05 was considered significant. Box plots (b and d) show the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. Single-cell BCR sequencing of S1+ B cells in INF and VAC groups. a Proportion of expanded (clone size > 1 cell) clones among S1+ CD19+ B cells from INF and VAC participants. b Simpson’s diversity of S1+ and S1− B cells from adenoids and tonsils from 8 INF and 4 VAC donors and S1− B cells from 2 CON. Adenoid and tonsil samples were assessed independently. Lower Simpson’s diversity indicates a greater frequency of large clones. c Clonal expansion index of S1+ B cells in each B cell subset in INF and VAC tissues. d Somatic hypermutation (SHM) frequency of S1+ or S1- B cell subsets. Each dot represents one cell. e Pairwise transition index (pTrans) showing degree of BCR clone overlap between P2 S1+ BSM and S1+ B cell subsets from adenoids and tonsils. f Heatmap showing clonal overlap between S1+ or S1- B cell subsets from one representative donor who had the largest number of sorted S1+ B cells (CNMC 70, 2413 S1+ B cells). Off-diagonal elements are colored by the Jaccard index indicating degree of clonal overlap and are labeled by the raw number of overlapping clones. Diagonal elements are labeled by the total number of clones within a particular subset. g, h Clonal lineage trees selected from the two largest GCB-bearing S1+ trees from INF (h) and VAC (i) donors. Triangles indicate S1+ cells and tip color indicates B cell subset. Isotype and tissue origin of each clone are listed next to the symbol. Branch lengths represent SHM frequency/codon in VDJ sequence according to the scale bar. For a–e, N denotes the number of subjects per group (adenoid: INF N = 6, VAC N = 3, CON N = 1; tonsil: INF N = 6, VAC N = 2, CON N = 2, listed in Supplemental Data 8). Significance for panels b-d calculated with two-sided Mann-Whitney U test. P < 0.05 was considered significant. Box plots (b and d) show the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. In tonsil, “atypical”/DN2 BSM have been noted to have slightly lower levels of SHM58; we also found that SHM frequency was lower among bulk S1- P3 BSM in the tissues. However, SHM frequency was comparable across all four P1-P4 subsets of S1+ BSM in both INF and VAC (Fig. 7d), revealing distinctions between SARS-CoV-2-specific cells compared to the bulk. Indeed, SHM frequency and heavy chain CDR3 amino acid length and distribution among S1+ IgG+ BSM and GCBs were comparable between INF versus VAC tissues (Supplementary Fig. 7b–d). We then assessed BCR clonal overlap to evaluate BSM development. Using STARTRAC (single T cell analysis by RNA sequencing and TCR tracking) pairwise transition index (pTrans)68,69, which assesses developmental links in lymphocyte populations using clonal sharing, we found that antigen-specific S1+ P2 BSM exhibited high pTrans scores with S1+ P1 BSM and S1+ GCB (Fig. 7e); bulk S1- P2 BSM also had a strong association with S1- GCBs (Supplementary Fig. 7e). We then looked further at BCR sequences from the tissues of one INF subject from whom we sorted a large number of S1+ B cells (CNMC 70). By Jaccard index scores, we observed clonal overlap among S1+ P1, P2, P3, and P4 BSM populations, with particularly strong overlap between P1 and P2. S1+ P1 and P2 BSM also had strong clonal sharing with GCBs, whereas S1+ P3 and P4 BSM had GC overlap but to a lower extent than S1+ P1 and P2 BSM (Fig. 7f). Nonetheless, evaluation of clonal trees showed S1+ BSM bearing various phenotypes (P1-P4, GCB) and isotypes emerging from a common ancestor in both INF (Fig. 7g) and VAC (Fig. 7h) tissues, implying broad differentiation potential and overlap among these populations. Several clonotypes were also shared among adenoid and tonsil S1+ B cells, particularly in P1 and P2 BSM and GCB populations, supporting an immunologic connection between these two oropharyngeal lymphoid tissues, as we previously reported21 (Fig. 7g-h, Supplementary Fig. 7f). Thus, although similar levels of SHM were noted post-infection and vaccination, infection induced greater antigen-specific B cell expansion in the tissues, particularly among P2 BSM, which likely have a GC origin. CXCR3+CD21+ BSM are more responsive to BCR stimulation and poised to differentiate to plasmablasts To determine whether these differences reflect functional differences among P1-P4 BSM populations, we evaluated responsiveness to BCR stimulation, proliferation capacity, and ability to differentiate to antibody-secreting cells. Tonsil cells from a separate group of pediatric subjects who were either post-infection or had hybrid immunity to SARS-CoV-2 (both infected and vaccinated, Supplemental Data 2) were stimulated with soluble anti-human IgA, IgG, and IgM for 2 min and phosphorylation of the tyrosine kinase Syk and phospholipase Cγ2 (PLC-γ2), two downstream signaling proteins rapidly phosphorylated upon BCR stimulation, was measured70. As noted previously, the CD21lo BSM populations (P3 and P4) were poorly responsive to BCR stimulation with soluble immunoglobulin71,72 (Fig. 8a). In contrast, the P2 BSM population had the strongest induction of p-Syk and/or p-PLCγ2 among total BSM and S1+ BSM (Fig. 8a, b, Supplementary Fig. 8a, b). This phenotype was not limited to BSM, as the CXCR3+CD21+ (P2) subpopulation also had the strongest response to BCR stimulation among GCBs (Supplementary Fig. 8c).Fig. 8CXCR3+CD21+ (P2) BSM are predisposed to plasma cell differentiation.a Phosphorylation of Syk and PLCγ2 in S1⁺ BSM subsets, concatenated by group, with and without 2-min stimulation with soluble anti-IgG, -IgA, and -IgM (N = 9 tonsils). b Percentage of cells expressing either or both p-Syk and/or p-PLCγ2 among P1 versus P2 S1+ BSM post-BCR stimulation (percentages of p-Syk and/or p-PLCγ2 from untreated conditions were subtracted from stimulated conditions) (N = 9 tonsils). Geometric mean fluorescence intensity (gMFI) of surface IgG or IgA on isotype-specific B cell subsets from PBMC (c) and tonsil/adenoid (d) (PBMC N = 37, adenoid N = 33, tonsil N = 41). e gMFI of IgG on P1 versus P2 S1+RBD+ IgG+ BSM in INF and VAC groups. Samples with at least 10 cells in both P1 and P2 S1+RBD+ IgG+ BSM populations were analyzed (PBMC: INF N = 6, VAC N = 8; adenoid: INF N = 12, VAC N = 7; tonsil: INF N = 17, VAC N = 6). f Percentages of plasmablasts (CFSE+CD38+CD20-) after 4 days of in vitro culture of sorted P1 and P2 BSM from tonsils with B cell-depleted PBMCs from an unrelated donor, R848, and IL-2. Left: representative plot; right: fold change comparing P1 and P2 calculated from paired P1 versus P2 BSM from three tonsils. g Division index of CFSE-labeled P1 and P2 BSM following in vitro cultures in Fig. 7f. P values for b–g were calculated with two-sided Wilcoxon signed ranks test (paired). Samples used in this figure are listed in Supplemental Data 2. In f and g, mean ± SD are shown in the box plots. Individual data points are shown. P < 0.05 was considered significant. CXCR3+CD21+ (P2) BSM are predisposed to plasma cell differentiation. a Phosphorylation of Syk and PLCγ2 in S1⁺ BSM subsets, concatenated by group, with and without 2-min stimulation with soluble anti-IgG, -IgA, and -IgM (N = 9 tonsils). b Percentage of cells expressing either or both p-Syk and/or p-PLCγ2 among P1 versus P2 S1+ BSM post-BCR stimulation (percentages of p-Syk and/or p-PLCγ2 from untreated conditions were subtracted from stimulated conditions) (N = 9 tonsils). Geometric mean fluorescence intensity (gMFI) of surface IgG or IgA on isotype-specific B cell subsets from PBMC (c) and tonsil/adenoid (d) (PBMC N = 37, adenoid N = 33, tonsil N = 41). e gMFI of IgG on P1 versus P2 S1+RBD+ IgG+ BSM in INF and VAC groups. Samples with at least 10 cells in both P1 and P2 S1+RBD+ IgG+ BSM populations were analyzed (PBMC: INF N = 6, VAC N = 8; adenoid: INF N = 12, VAC N = 7; tonsil: INF N = 17, VAC N = 6). f Percentages of plasmablasts (CFSE+CD38+CD20-) after 4 days of in vitro culture of sorted P1 and P2 BSM from tonsils with B cell-depleted PBMCs from an unrelated donor, R848, and IL-2. Left: representative plot; right: fold change comparing P1 and P2 calculated from paired P1 versus P2 BSM from three tonsils. g Division index of CFSE-labeled P1 and P2 BSM following in vitro cultures in Fig. 7f. P values for b–g were calculated with two-sided Wilcoxon signed ranks test (paired). Samples used in this figure are listed in Supplemental Data 2. In f and g, mean ± SD are shown in the box plots. Individual data points are shown. P < 0.05 was considered significant. In general, as B cells transition to PC/PB, membrane-associated IgG declines due to altered splicing, generating secreted IgG48,73. We analyzed surface IgG and IgA expression (geometric mean fluorescence intensity or gMFI of anti-IgG and anti-IgA antibody staining) among BSM and PC/PB in our flow cytometry data. As expected, CD27hiCD38hi PC expressed markedly less surface IgG compared to BSM in both the peripheral blood and tissues (Fig. 8c, d). Surface IgA was also reduced on plasma cells in peripheral blood, but to a lesser extent (Fig. 8c) and was not reduced on tissue PC compared to GCBs and tissue BSM (Fig. 8d). These observations are consistent with findings of functional surface IgA on gut and bone marrow PC in humans74 and suggest that surface IgA may also play a functional role on PC in the oropharyngeal lymphoid tissue. We then compared surface IgG expression on the two major S1+RBD+ BSM populations, P1 and P2, as a surrogate for PC differentiation48. In INF tissues and blood, S1+RBD+ IgG+ P2 BSM had lower surface immunoglobulin expression than their P1 counterparts, suggesting the SARS-CoV-2-specific P2 BSM post-infection are more prone to differentiate to PC (Fig. 8e). Significant differences were not noted for P1 and P2 in VAC tissues, although a trend was seen in the tonsils (Fig. 8e). Moreover, in the tonsil, S1+RBD+ IgG+ BSM from INF subjects had lower surface immunoglobulin expression compared to VAC subjects (Supplementary Fig. 8d); a similar trend was noted in the adenoid, suggestive of a propensity of tissue BSM induced by infection to differentiate to PC/PB. We then directly assessed the ability of P1 and P2 BSM from tissues to differentiate to plasmablasts and proliferate in vitro. We sorted bulk P1 and P2 BSM cells from tonsils of three subjects with hybrid immunity and labeled them with carboxyfluorescein succinimidyl ester (CFSE). We cultured the labeled P1 and P2 BSM populations with B cell-depleted PBMCs of an unrelated donor, R848 (TLR7/8 agonist), and IL-2 and we assessed proliferation index and percentage of plasmablasts in culture using flow cytometry after 4 days. Due to low cell numbers and poor viability after sorting, we were unable to reproducibly culture and assess P3 and P4 BSM. Compared to P1, a greater proportion of P2 BSM differentiated to CD38+CD20lo plasmablasts (Fig. 8f). P2 BSM also exhibited a higher proliferation index in culture (Fig. 8g). Thus, P2 BSM, which were enriched among INF samples, were better primed for plasma cell differentiation compared to P1 BSM, which may result in greater protective antibody production in the upper respiratory tract tissues upon re-exposure to SARS-CoV-2 in infected individuals compared to vaccinated individuals. Infection, but not vaccination, is associated with persistent immunologic activation in pharyngeal tissues We previously found that SARS-CoV-2 infection early in the pandemic induced persistent expansion of CD4+ and CD8+ T cell populations associated with antiviral, GC and IFN-γ responses in the tonsils and adenoids compared to uninfected controls, with the most significant effects in adenoid21. However, whether vaccination also induces long-lasting immunologic changes in the upper respiratory tract lymphoid tissues is unclear. Consistent with our previous findings, a higher proportion of CXCR3+CCR6-CD57+PD-1hi CD4+ T cells and CXCR3+CCR6- pre-Tfh cells (gated after the exclusion of CD25+CD4+ cells) were noted in INF compared to CON adenoids (Fig. 9a; Supplementary Fig. 9a, Supplementary Fig. 10). In contrast, VAC adenoids did not exhibit changes in these T cell populations compared to CON (Fig. 9a).Fig. 9Persistent immunologic changes are noted in tonsils and adenoids post-infection.a Percentages of CXCR3+CCR6-CD57+PD-1hi and CXCR3+CCR6- pre-Tfh (gated on CD25- cells) among CD4+ T cells in CON (N = 10), INF (N = 14) and VAC (N = 9) adenoids. Samples listed in Supplemental Data 2. b gMFI of HLA-DR expression on CD14+ monocytes, conventional dendritic cells (cDC) and plasmacytoid dendritic cells (pDC) in CON, INF and VAC adenoids. Same samples as in a. c Correlations between neutralizing titers to omicron BA.1 and WA-1 strains, percentages of P1-P4 among S1+RBD+ BSM, CXCR3+CCR6-CD57+PD-1hi among CD4+ T cells, and CXCR3+CCR6- preTfh (gated on CD25-) among CD4+ T cells and HLA-DR gMFI of pDC, cDC, and CD14+ monocytes. Data from tonsils and adenoids and INF and VAC were pooled (N = 51). Correlations were assessed with a two-sided Spearman’s rank correlation test. Correlation coefficients (r values) are indicated by color shown in the bar. Size of the circle indicates absolute r values. Significant p values are indicated as * p < 0.05, ** p < 0.01, *** p < 0.001. d, e Expression of IFN-γ among subsets (based on CXCR3 and CCR6 expression) of pre-Tfh (gated on total CD4+ T cells) and CD57+PD-1hi CD4+ T cells from tonsils and adenoids after PMA/ionomycin stimulation (N = 52, including 26 tonsils and 26 adenoids). For a and b, p values obtained from linear model correcting for participant ages (in blue) and from two-sided Mann-Whitney U (in black) are shown. P values for d, e were calculated with two-sided Wilcoxon signed ranks test (paired) without multiple-comparison adjustment. Box plots (a, b, d and e) show the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. P < 0.05 was considered significant. Persistent immunologic changes are noted in tonsils and adenoids post-infection. a Percentages of CXCR3+CCR6-CD57+PD-1hi and CXCR3+CCR6- pre-Tfh (gated on CD25- cells) among CD4+ T cells in CON (N = 10), INF (N = 14) and VAC (N = 9) adenoids. Samples listed in Supplemental Data 2. b gMFI of HLA-DR expression on CD14+ monocytes, conventional dendritic cells (cDC) and plasmacytoid dendritic cells (pDC) in CON, INF and VAC adenoids. Same samples as in a. c Correlations between neutralizing titers to omicron BA.1 and WA-1 strains, percentages of P1-P4 among S1+RBD+ BSM, CXCR3+CCR6-CD57+PD-1hi among CD4+ T cells, and CXCR3+CCR6- preTfh (gated on CD25-) among CD4+ T cells and HLA-DR gMFI of pDC, cDC, and CD14+ monocytes. Data from tonsils and adenoids and INF and VAC were pooled (N = 51). Correlations were assessed with a two-sided Spearman’s rank correlation test. Correlation coefficients (r values) are indicated by color shown in the bar. Size of the circle indicates absolute r values. Significant p values are indicated as * p < 0.05, ** p < 0.01, *** p < 0.001. d, e Expression of IFN-γ among subsets (based on CXCR3 and CCR6 expression) of pre-Tfh (gated on total CD4+ T cells) and CD57+PD-1hi CD4+ T cells from tonsils and adenoids after PMA/ionomycin stimulation (N = 52, including 26 tonsils and 26 adenoids). For a and b, p values obtained from linear model correcting for participant ages (in blue) and from two-sided Mann-Whitney U (in black) are shown. P values for d, e were calculated with two-sided Wilcoxon signed ranks test (paired) without multiple-comparison adjustment. Box plots (a, b, d and e) show the median (center line) and interquartile range (25th–75th percentiles; box bounds) with whiskers extending to the most extreme values within 1.5× the interquartile range. Individual data points are shown. P < 0.05 was considered significant. COVID-19 induces persistent activation and epigenetic remodeling of innate immune cells, including peripheral blood monocytes and dendritic cells well into convalescence75,76. The frequency of CD14+ monocytes/macrophages, conventional dendritic cells (cDC), and plasmacytoid dendritic cells (pDC) did not differ among the VAC, INF, and CON tissues (Supplementary Fig. 9b, Supplementary Fig. 11). However, these innate immune populations expressed higher HLA-DR in INF adenoids compared to VAC (and trended toward higher expression compared to CON), revealing persistent myeloid cell activation in adenoids post-infection but not post-vaccination (Fig. 9b). Unlike the tissues, peripheral blood CD14+ monocytes/macrophages and cDC in INF subjects did not express higher HLA-DR post-infection, highlighting localized differences in the impact of infection in upper respiratory tract tissues compared to blood (Supplementary Fig. 9c). However, the antiviral type I interferon-producing and CXCR3-expressing pDC population was slightly more activated in both INF blood and tissues (Fig. 9b, Supplementary Fig. 9c)77. Thus, convalescence was associated with local activation of innate cells in tissues, which may reflect differences in the primary lymphoid site of antigen presentation and GC response and/or antigen persistence in infection versus vaccination. We then assessed correlations between populations of interest. The proportion of P2 cells among S1+RBD+ BSM was positively associated with the proportion of CXCR3+CCR6-CD25- pre-Tfh cells and of CXCR3+CCR6- CD57+PD-1hi Tfh cells (Fig. 9c). Furthermore, both of these CXCR3+ Tfh populations expressed high levels of IFN-γ with in vitro stimulation with PMA and ionomycin (Fig. 9d-e, pre-Tfh were gated from total CD4+ T cells). Moreover, the proportion of P2 BSM and CXCR3+ pre-Tfh cells were also positively associated with the activation levels of monocytes/macrophages, cDC, and pDC, pointing to the potential relevance of innate cell activation in development or accumulation of CXCR3+ lymphocytes (Fig. 9c). Thus, infection is associated with prolonged activation of innate or expansion of adaptive immune populations in tonsils/adenoids, which may contribute to the generation of CXCR3+ BSM. CXCR3+ lymphocytes and CXCL9-expressing myeloid cells are in close proximity in the interfollicular region Using multicolor immunofluorescence imaging, we compared the location of CXCR3+ CD20+ B and CD4+ T cells and to their CXCR3- counterparts in 4 tissue samples, derived from tonsils and adenoids of one INF and one VAC subject (Fig. 10a, Supplemental Data 2). CXCR3+BCL6- CD20+ B cells and CXCR3+BCL6- CD4+ T cells were primarily located in the interfollicular region (Fig. 10b–d). Compared to their CXCR3- counterparts, a smaller portion of CXCR3+BCL6- B cells were in follicles, which includes the GC and surrounding mantle (Fig. 10c). A similar pattern of distribution was seen among CXCR3+ CD4+ T cells (Fig. 10d). CXCR3+BCL6- CD4+ T cells and CXCR3+BCL6- CD20+ B cells appeared co-localized or aggregated near one another compared to CXCR3-BCL6- CD4+ T and CXCR3-BCL6- CD20+ B cells, which were more broadly distributed in the interfollicular area (Fig. 10a, b).Fig. 10CXCR3+ cells and CXCL9-expressing cells co-localize in the tissue.a Immunofluorescence images of tonsil from one INF donor, representative of paired tonsil and adenoid from one INF and one VAC donor, with major histologic regions delineated (top left). Insets show greater magnifications. Bottom panels show CD20 (blue), CD4 (purple), BCL6 (orange), and CD138 (plasma cells and epithelial cells, white) with and without CXCR3 channel (green). Arrows indicate CXCR3+CD20+ B cells (arrow 1) and CXCR3+CD4+ T cells (arrow 2). Scale bars: 500 μm (top left), 80 μm (top right), and 20 μm (bottom). b Localization of CXCR3⁺BCL6- CD4+ T and CD20+ B cells (left) and CXCR3⁻BCL6- CD4+ T and CD20+ B cells (right) in the representative tonsil from a. Scale bar 500 μm. Percentages of CXCR3+BCL6- CD20+ or CXCR3-BCL6- CD20+ B cell populations (c) and CXCR3+BCL6- CD4+ or CXCR3-BCL6- CD4+ T cells populations (d) in histological regions of paired tonsils and adenoids of one INF and one VAC subject (N = 4 tissues). P values calculated with two-sided paired t test. P < 0.05 was significant. e Spatial distribution of 8 cell subsets defined by CD4, CD20, BCL6, and CXCR3 expression, shown in the representative tonsil from (d) using SPACE at 10 μm scale. Peaks of local representation of each population along the latent path indicate their spatial proximity to each other. BCL6+ and BCL6- cells were gated with a gap, which is reflected as a gap in the latent path. f Annotated cell types based on transcriptome are shown in nearby section of the representative tonsil from panel a run on Xenium In Situ spatial transcriptomic platform using human 5000 gene panel. g Representative scaled expression of CXCL9, CXCL10, and CXCL11 in each cell type from a representative adenoid. Spatial map showing the location of monocytes/macrophages (Mono/Macro) (h), activated dendritic cells (aDC) (i), pDC (j) and cycling dark zone GCB (DZ GCB) to indicate germinal centers from the tonsil from f. k Spatial map showing the location and expression levels of CXCL9, both overall and specifically within mono/macro and aDC cells in the representative tonsil from f. CXCR3+ cells and CXCL9-expressing cells co-localize in the tissue. a Immunofluorescence images of tonsil from one INF donor, representative of paired tonsil and adenoid from one INF and one VAC donor, with major histologic regions delineated (top left). Insets show greater magnifications. Bottom panels show CD20 (blue), CD4 (purple), BCL6 (orange), and CD138 (plasma cells and epithelial cells, white) with and without CXCR3 channel (green). Arrows indicate CXCR3+CD20+ B cells (arrow 1) and CXCR3+CD4+ T cells (arrow 2). Scale bars: 500 μm (top left), 80 μm (top right), and 20 μm (bottom). b Localization of CXCR3⁺BCL6- CD4+ T and CD20+ B cells (left) and CXCR3⁻BCL6- CD4+ T and CD20+ B cells (right) in the representative tonsil from a. Scale bar 500 μm. Percentages of CXCR3+BCL6- CD20+ or CXCR3-BCL6- CD20+ B cell populations (c) and CXCR3+BCL6- CD4+ or CXCR3-BCL6- CD4+ T cells populations (d) in histological regions of paired tonsils and adenoids of one INF and one VAC subject (N = 4 tissues). P values calculated with two-sided paired t test. P < 0.05 was significant. e Spatial distribution of 8 cell subsets defined by CD4, CD20, BCL6, and CXCR3 expression, shown in the representative tonsil from (d) using SPACE at 10 μm scale. Peaks of local representation of each population along the latent path indicate their spatial proximity to each other. BCL6+ and BCL6- cells were gated with a gap, which is reflected as a gap in the latent path. f Annotated cell types based on transcriptome are shown in nearby section of the representative tonsil from panel a run on Xenium In Situ spatial transcriptomic platform using human 5000 gene panel. g Representative scaled expression of CXCL9, CXCL10, and CXCL11 in each cell type from a representative adenoid. Spatial map showing the location of monocytes/macrophages (Mono/Macro) (h), activated dendritic cells (aDC) (i), pDC (j) and cycling dark zone GCB (DZ GCB) to indicate germinal centers from the tonsil from f. k Spatial map showing the location and expression levels of CXCL9, both overall and specifically within mono/macro and aDC cells in the representative tonsil from f. We then assessed spatial proximity of CXCR3+ and CXCR3- B and CD4+ T cells using spatial patterning analysis of cellular ensembles (SPACE), which assesses spatial relationships among multiple cell populations78. Peak areas of abundance of BCL6+ and BCL6– populations were distinct, clearly delineating GCs (Fig. 10e). Within BCL6- regions, the CXCR3+BCL6- CD20+ B cell and CXCR3+BCL6- CD4+ T cell peaks were in close proximity to each other; similarly, in BCL6+ regions, which represent GCs, the CXCR3+BCL6+ CD20+ B cell and CXCR3+BCL6+ CD4+ T cell were closely aligned, suggesting that CXCR3+ B and T cells were adjacent both within and outside GCs (Fig. 10e). Moreover, compared to their CXCR3- counterparts, CXCR3+BCL6- B and T cells were closer to BCL6+ populations, and in particular, closer to CXCR3+BCL6+ B and T cells, further implying proximity and possible interactions among CXCR3+ Tfh, GCB, and BSM populations in the GC and T-B border regions (Fig. 10e). We performed spatial transcriptomics analysis with the Xenium 5000 human gene panel of the same 4 samples from an adjacent section of tissue. With unsupervised clustering and annotation (see Methods), we identified 25 cell types, which localized to particular regions (Fig. 10f). CXCL9, -10, and -11 are ligands of CXCR3 and enable migration and retention of CXCR3-expressing lymphocytes in tissue79. In tonsils and adenoids, we found that macrophages/monocytes and dendritic cells were the primary expressors of CXCL9, CXCL10, and CXCL11, with CXCL9 being the most abundantly expressed (Fig. 10g, Supplementary Fig. 9d). Moreover, these innate immune cell populations were located in the interfollicular area, where we detected expression of CXCL9 and where we found CXCR3+ B and T cells by immunofluorescence (Fig. 10h-k). We did not analyze CXCR3 gene expression with Xenium due to the lack of correlation between CXCR3 mRNA and CXCR3 protein levels (Supplementary Fig. 6b). Nonetheless, together these spatial analyses suggest that the CXCR3 and CXCL9/10/11 axis facilitates interaction of innate and adaptive immune cells in lymphoid tissue, which may shape both the position and characteristics of memory B cells during viral infection. Discussion Maintenance of virus-specific adaptive immune memory in the respiratory mucosa is important for immune protection against respiratory viral infections, including COVID-19. By assessing tonsils and adenoids of children undergoing tonsillectomy/adenoidectomy, we found SARS-CoV-2-specific B cells, including BSM, BRM and a few GCBs, in these upper respiratory tract lymphoid tissues following vaccination. Distribution of BSM across tissues, including mucosal tissues, after intramuscular vaccination has been noted18,80,81. However, whether and how these cells differ from memory cells found post-infection has not been clear. Using high-dimensional analyses, we show that immune memory is maintained in mucosal lymphoid tissues after intramuscular COVID-19 immunization, but reveal distinct features compared to those generated post-infection. In particular, we identified a subset of memory B cells, defined by expression of CXCR3 and CD21, that was enriched in infected individuals, with increased propensity for plasma cell differentiation and distinct localization within mucosal lymphoid tissues. This population may be responsible for some of the qualitative and clinical differences in humoral mucosal immunity observed with infection and vaccination. Prior studies have found SARS-CoV-2-specfic GC and Tfh cells in the draining axillary lymph nodes following mRNA vaccination14–17,82, but no SARS-CoV-2-specific GC B cells were noted in a small number of contralateral lymph nodes14. However, spike-reactive GCBs have been noted in lung-associated lymph nodes of a few (2/18) vaccinated (and uninfected) organ donors18 and tonsils of a few vaccinated adults, although whether these latter subjects were truly SARS-CoV-2 infection-naïve is unclear since some had B cells reactive to NC59. To address this concern, we measured both anti-NC antibodies and anti-ORF8 antibodies, which are more sensitive than antibodies to NC alone, but may still wane with time24. We too noted the presence of spike-specific GCBs in some vaccinated participants in distal mucosal lymphoid sites, although fewer than among infected subjects. Whether these result directly in response to immunization, from cross-reactivity to common cold coronaviruses, or exposure to SARS-CoV-2 that did not result in a productive infection, perhaps due to the presence of high-affinity antibodies generated by vaccination83,84, remains an interesting question. Although COVID-19 mRNA vaccines have high vaccine efficacy and prevent hospitalization and death from COVID-19, epidemiologic studies suggest that infection provides longer-lasting immunity than vaccination6–8. This may result from differences in the route of viral antigen exposure to the host’s immune system (intramuscular vs. inhaled), breadth of antigenic exposure (spike only vs. all viral antigens), nature of antigens presented (lipid nanoparticles vs. virions) or duration of antigen exposure. These factors, in turn, affect host immune responses, and, indeed, we found that infection and vaccination drive the generation of phenotypically different SARS-CoV-2-specific B cells, suggesting unique BSM developmental pathways depending on the type of exposure. The clonal expansion and persistence of CXCR3+ BSM post-infection suggest that IFN-γ, induced by viral infection and likely produced by activated and expanded innate and Tfh cells in the tissue, may be a driving force for these developmental differences in B cell memory. The CXCR3+ P2 BSM population expressed low levels of Tbet, which may reflect prior IFN-γ exposure during infection, and were distinct from “atypical”/DN2 and FCRL5+ B cell populations which expressed high levels of Tbet. Our results suggest that CXCR3 expression affords two key characteristics in memory B cells that may contribute to enhanced mucosal protection post-infection. First, CXCR3 promotes homing to and retention of BSM cells, particularly IgA+ cells, in mucosal lymphoid tissues, where CXCR3+ cells bear BRM markers, as supported by murine studies as well51,53. Second, we found that among CD21+ BSM, CXCR3+ cells have a predilection for plasma cell differentiation upon secondary challenge compared to CXCR3- cells, perhaps enabling a more robust antibody response upon secondary challenge. Although CXCR3+ BSM were enriched in the tissues, the difference in CXCR3 expression between INF and VAC was also clear in PBMCs, suggesting CXCR3 may be a useful marker for evaluating potential for trafficking even in blood; this is supported by our recent findings in blood of vaccinated adults with hybrid immunity85, which also indicate the applicability of our findings across age groups. Moreover, although our study focused on mRNA COVID-19 vaccines, another recent study found that Ad26.COV2.S elicited more CXCR3+ BSM (that also lacked “atypical”/DN2 features) compared to mRNA vaccines5, suggesting similar cells may be generated by viral vaccine platforms. Nonetheless, the mRNA component of mRNA-lipid nanoparticle vaccines can stimulate IFN-γ production by Tfh cells86, potentially explaining the presence of some CXCR3+ BSM in the blood and tissues of vaccinated subjects in our study. Of note, the frequency of the P2 population among BSM did not correlate with serum neutralizing titers, suggesting this population reflects other aspects of immunity induced by infection and that neutralizing titers cannot be used as a proxy for this population. Conversely, we found that mRNA vaccination induced a greater proportion of virus-specific CD21lo and CXCR3- BSM populations than infection in both blood and tissues29,36,59. Although some studies suggest that CD21lo memory B cells have lower degrees of somatic hypermutation than CD21+CD27+ “conventional” memory B cells and are of extrafollicular origin, particularly in the context of chronic viral or autoimmune conditions34,42, we found that SARS-CoV-2-specific CD21lo populations in tonsils/adenoids had comparable levels of SHM as CD21+ BSM, highlighting the importance of assessing antigen-specific versus bulk cell populations and evaluating cells in patients without conditions characterized by prominent extrafollicular reactions. The frequency of CD21-/lo BSM in the blood declines with time after influenza or COVID-19 vaccination35,47. Nonetheless, FCRL5+Tbet+ CD21loCD27+ memory B cells have been found 1 year after influenza vaccination and contributed significantly to serum antibody levels upon secondary challenge; similarly, a FCRL4lo population was present at least 6 months post-influenza vaccination, suggesting the picture is more complex48,50. We also note that vaccination engendered broader variant coverage, as determined by cross-reactivity to omicron, highlighting an important feature induced by mRNA vaccination. Thus, the plasticity and function of different memory populations following vaccination versus infection remain important questions. As humans are confronted with new pathogens, the development of vaccinations for respiratory pathogens that elicit strong mucosal immunity, which may reduce mucosal shedding and transmission, is a priority. Although intramuscular SARS-CoV-2 mRNA vaccines generate rapid and broad immunity, they elicit low neutralizing antibody levels in the respiratory tract12,13. Trials of intranasal vaccines in humans are ongoing and have shown mixed results87,88. Nonetheless, pre-clinical studies of inhaled vaccine boosters following intramuscular mRNA vaccines in murine models showed enhanced mucosal protection due to trafficking of antigen-experienced BSM to the respiratory tract through the CXCL9/10-CXCR3 axis and enhanced differentiation to IgA-producing plasma cells33. Our study underscores the significance of this axis in homing and shaping immunity in the upper respiratory tract of humans. In summary, our study provides evidence for the generation and maintenance of mucosal B cell memory after mRNA vaccination and further provides a framework for evaluating immunity in the upper respiratory tract and the blood following immunization that may be applicable for multiple respiratory pathogens. Limitations Although study participants were heterogeneous in terms of their age, tonsil condition, time since vaccination/infection, and doses of immunization, we tried to control for age as a covariate using a linear model. Although INF and VAC subjects were recruited at different times during the pandemic, the INF cohort, which was collected in 2020–2021, had similar times from their last known immunologic exposure to surgery as the VAC cohort and were exposed to early circulating strains with spike proteins similar to the strain in the first-generation mRNA vaccines, helping control for these factors. However, we only know the time of infection for approximately half of the infected subjects. As our study is a cross-sectional study, we are unable to assess temporal changes in immune responses over time. We also note that our INF cohort only included patients with mild or asymptomatic infection; severe infection may alter or compromise the humoral immunity generated by infection89. Lastly, due to the increasing percentage of infected individuals, our sample size of vaccinated only individuals was limited, which could affect our ability to fully assess correlations between clinical characteristics and immune profiles. Methods Participant recruitment This study was approved by the Institutional Review Board (IRB) at Children’s National Hospital (IRB protocol number 00009806). Written informed consent was obtained from parent/guardians of all enrolled participants, and assent was obtained from minor participants over 7 years of age. We recruited 21 children who underwent tonsillectomy and/or adenoidectomy at Children’s National Hospital (CNH) in Washington, DC, USA and also received at least one dose of a first-generation COVID-19 mRNA vaccine. The first 17 participants were recruited from December 2021 to April 2022, and the remaining were recruited from August 2022 to September 2022. Because not all tissues or blood were available from each subject, we collected a total of 21 blood samples, 19 adenoids, and 18 tonsils from these 21 participants. No statistical methods were used to predetermine sample size. All participants had negative RT-PCR testing from a nasopharyngeal swab for SARS-CoV-2 within 72 h of surgery. Demographic information and clinical data were collected through parental questionnaires and chart review and inputted and managed in REDCap (https://project-redcap.org/), and biologic samples were acquired in the operating room by the clinical team at CNH. Participants with no evidence of prior infection (negative for anti-nucleocapsid (NC) and anti-open reading frame 8 (ORF8) serum antibodies and/or no clinical history of PCR/antigen confirmed SARS-CoV-2 infection) were classified as the VAC group (vaccinated only, 10 subjects). Samples from the remaining 11 subjects with hybrid immunity (both post-vaccination and infection, 11 subjects) and additional post-infection subjects without vaccination who were recruited in 2022/2023 (12 subjects) were used in some assays. We also used samples obtained from our prior study of 24 children infected with SARS-CoV-2 from 2020 to 2021 (INF), prior to the availability of vaccines for children. Collection of these samples was previously described21. Prior infection in these individuals was confirmed by serologic testing for anti-spike antibodies and/or the presence of SARS-CoV-2-specifc B cells in the blood, tonsil, or adenoid by flow cytometry. From these 24 children, 15 PBMCs, 14 adenoids, and 22 tonsils were analyzed based on availability of cells (Supplementary Data 2). Blood and tissue collection Blood samples were obtained just prior to the surgical procedure in the operating room in serum separator tubes (BD) for serum collection and sodium heparin tubes (BD) for extraction of peripheral blood mononuclear cells (PBMCs) from an intravenous line placed for anesthesia. Once received in the laboratory on the day of collection, serum separator tubes were spun at 1200 g for 10 min, and serum was aliquoted and stored at −80 °C. PBMCs were isolated the day after collection by density gradient centrifugation (Lymphocyte Separation Medium, MP Biomedicals) at 1500 rpm for 30 min at room temperature (RT) with no brake and washed with PBS. If red blood cell contamination was present, cells were lysed with ammonium–chloride–potassium (ACK) buffer (Gibco). Tonsil and adenoid tissues were stored in RPMI-1640 (RPMI) media with 5% heat-inactivated fetal bovine serum (FBS, VWR), gentamicin 50 mg/mL (Gibco), and 1× antibiotic/antimycotic solution (Gibco) on ice immediately after collection. Tissues were processed the day after collection. The tissue was mechanically disrupted and filtered through a 100 μm cell strainer to create a single-cell suspension, lysed with ACK buffer, and washed with PBS three times. Cells were then stored in liquid nitrogen in the presence of FBS with 10% DMSO and thawed according to our published protocol90 for flow cytometry, CITE-seq, and functional assays below. Seroreactivity of samples to SARS-CoV-2 spike, nucleocapsid, and ORF8 by ELISA 96-well Immulon plates were coated with 20 ng/100 µL of recombinant nucleocapsid, spike RBD protein, or Open Reading Frame 8 (ORF8) from WA1/2020 in PBS overnight at 4 °C. Starting at a 1:20 dilution, serum samples were serially diluted 5-fold and applied to the coated well for 1 h at ambient temperature. Serum samples were assayed in duplicate, as described before11,91. After three washes with PBS/0.05% Tween 20, bound human IgG antibodies were detected with 1:5000 dilution of HRP-conjugated anti-human IgG Fc-specific antibody (Jackson Immuno Research). After 1 h, plates were washed with PBST followed by PBS, and o-Phenylenediamine dihydrochloride was added for 10 min. Absorbance was measured at 492 nm. End-point titer was determined as 3-fold above the average of the absorbance values of the binding of serum samples to blank control wells. The end-point titer is reported as the serum dilution that was above this cutoff and was calculated using Prism 9 (GraphPad Software). SARS-CoV-2 serum neutralization assay Samples were evaluated in a qualified SARS-CoV-2 pseudovirion neutralization assay (PsVNA) using SARS-CoV-2 WA1/2020 strain and Omicron BA.1 subvariant. SARS-CoV-2 neutralizing activity measured by PsVNA correlates with PRNT (plaque reduction neutralization test with authentic SARS-CoV-2 virus) in previous studies92–94. Neutralization assays were performed as previously described11,91,93–95. Briefly, 50 µL of SARS-CoV-2 S pseudovirions (counting ~200,000 relative light units) were pre-incubated with an equal volume of medium containing serial dilutions (starting at 1:10) of all samples at RT for 1 h. Then, 50 µL of virus-antibody mixtures were added to 293T-ACE2-TMPRSS2 cells (104 cells/50 μL) in a 96-well plate. The input virus across all SARS-CoV-2 strains was the same (2 × 105 relative light units/50 µL/well). After a 3 h incubation, fresh medium was added to the wells. Cells were lysed 24 h later, and luciferase activity was measured using One-Glo luciferase assay system (Promega). The assay of each sample was performed in duplicate, and the 50% neutralization titer was calculated using Prism 9 (GraphPad Software). The limit of detection for the neutralization assay is 1:20. Two independent biological replicate experiments were performed for each sample and variation in PsVNA50 titers was <10% between replicates. High-dimensional flow cytometry SARS-CoV-2 specific B cell characterization with spectral flow cytometry 5 million cells per sample of PBMC, adenoid, or tonsil were resuspended in PBS with 2% FBS and 2 mM EDTA (FACS buffer). Biotinylated probes to SARS-CoV-2 were crosslinked with fluorochrome-conjugated streptavidin in a molar ratio of 4:1. Fluorochrome-conjugated streptavidin was split into 5 aliquots and conjugated to biotinylated probes by mixing for 20 min/aliquot at 4 °C. Four fluorochrome-conjugated probes were used in this assay, including spike S1 from the Wuhan strain (BioLegend) conjugated to APC, RBD from the Wuhan strain (BioLegend) conjugated to BV421, RBD from the Omicron variant (Acro) conjugated to BUV615, and RBD from the Omicron variant (Acro) conjugated to PE. Cells were first stained with the viability dye, Zombie NIR (1:800 dilution, BioLegend), for 15 min at RT, washed twice and then incubated with the 4 fluorochrome-conjugated probes plus d-biotin (Avidity) and Brilliant Stain Buffer Plus (BD Biosciences) at 4 °C for 1 h. Then, cells were washed twice and resuspended in True-Stain Monocyte Blocker (1:10 in 50 μL) (BioLegend) for 5 min. Anti-CXCR3 antibody and an antibody cocktail containing the rest of the surface antibodies (Supplementary Data 11) and Brilliant Stain Buffer Plus were sequentially added directly to the cells and incubated for 5 min and 30 min at RT, respectively (200 μL total staining volume). Cells were washed three times and fixed in 1% paraformaldehyde for 20 min at RT before washing again and collecting on a spectral flow cytometer (Aurora, Cytek). Antibodies are listed in Supplementary Data 11. Gating strategies are shown in Supplementary Figs. 2, 3. A separate panel, including intracellular transcription factor staining, was also used to stain PBMC and tonsil cells90. Antibodies are listed in Supplementary Data 11. Immunophenotyping with a 37-color spectral flow cytometry panel Two million PBMCs per sample and 5 million cells per adenoid or tonsil were resuspended in FACS buffer after thawing. Cells were stained and acquired as described in our prior study21. Antibodies are listed in Supplementary Data 11. Manual gating for both panels was conducted with FlowJo Software v.10.9.0 (BD Biosciences) as in Supplementary Figs. 10, 11. Unsupervised analysis of high-dimensional flow cytometry data S1+RBD+ B cells from the SARS-CoV-2-antigen-specific B cell panel with 29 parameters were analyzed with unsupervised clustering of surface antibody staining. S1+RBD+ live CD45+CD3-CD14-CD19+ B cells from INF and VAC groups were analyzed. Cells from tonsil and adenoid samples were merged and processed together, while PBMC samples were processed separately due to differences in cell populations. B cell analysis was based on surface expression of IgA, IgD, IgM, IgG, CD27, CD38, CD21, CD95, CD11c, FCRL3/FCRL5, FCRL4, CD10, CD86, CD83, CD69, CD71, CXCR3, CD85J, and CD62L. Channel values (scaled output with compensated parameters) of S1+RBD+ B cells from each sample were exported from FlowJo and then processed in R (v. 4.4.2) via Rstudio (2023.12.1 + 402). Each cell was given an index and labeled with its origin subject’s identification and tissue type. Data were further processed using Seurat96,97 (v. 5.1.0). Cell clustering was performed by applying the FindNeighbors() function96 on a distance matrix generated from the dist function by “euclidean” method, followed by Leiden clustering on the resulting SNN graph using Seurat’s FindClusters() algorithm, with a resolution parameter of 1.0 in PBMC and 0.6 in tissues. Expression of selected markers was visualized with their mean expression in each cluster by pheatmap (v. 1.0.12), and the downstream analysis and results were processed using ggplot2 (v. 3.5.1)98, reshape2 (v. 1.4.4)99, ggpubr (v. 0.6.0)100, ggthemes (v. 5.1.0)101 and tidyverse (v. 2.0.0). Statistical comparison with linear model The majority of participants underwent tonsillectomy for obstructive sleep disordered breathing due to tonsillar hypertrophy (which includes those with obstructive sleep apnea diagnosed with polysomnography); a smaller portion had recurrent or chronic tonsillitis as their primary indication. We and others have noted that age and indication for tonsillectomy influence the immune cell populations in these tissues55,102; however, none of the INF subjects had recurrent tonsillitis. This limitation prevented us from adjusting for this potential confounding factor. Sex had little effect in our cohort (evaluated with PCA analysis, data not shown). Cluster proportion, cell percentages from manual gating, or neutralizing titers in each sample were modeled linearly as a function of (1) age (“Age”) and (2) history of SARS-CoV-2 infection or vaccine exposure (“Group” includes two or three categories: VAC, INF, and CON depending on the comparison. Each comparison involves two categories at a time.). The following formula was used to estimate separate coefficients for each category of Group (adjusted for age): lm (Frequency/Percentage ~ Group + Age). To illustrate the extent of correction achieved by the linear model, both p-values from the two-sided Wilcoxon signed ranks test (in black) and p-values from linear model (in blue) were presented in plots built with ggplot2 (v. 3.5.1)98. Single-cell RNA sequencing Sorting of S1+ and S1− B cells for CITE-seq We performed a combined analysis by merging data from three sets of cellular indexing of transcriptomes and epitopes (CITE-seq) experiments (B-2021, B-306-3, and B-306-4). B-2021 was performed and processed as reported in our prior study21. Briefly, in B-2021, paired PBMC, adenoid, and tonsil samples from three donors (total of nine donor-tissue samples) were assessed and “hashtag” antibodies (BioLegend) were used to uniquely label each of the nine donor-tissue samples103. Two additional sets of CITE-seq experiments were conducted separately to include VAC samples (B-306-3 and B-306-4). For B-306-3, 2 INF and 1 CON tonsil were processed. For B-306-4, 3 VAC and 4 INF adenoid and 2 VAC, 2 INF, and 1 CON tonsil samples were processed as three sets (B-306-4 A, B, C) on 3 separate days over the course of one week to ensure manageable sorting times on each processing day (See Supplementary Data 8 for details on which samples were processed with each experiment). Frozen tonsil and adenoid cells were thawed from liquid nitrogen as described previously21,90. For each donor, 100,000–500,000 thawed tonsillar cells were reserved for bulk RNA-seq. During data analysis, individual cells were demultiplexed using donor-specific single-nucleotide polymorphism (SNP) information obtained from the bulk RNA-seq data (see Methods: CITE-seq processing and demultiplexing). The remaining cells of the same tissue type from each donor were pooled together (i.e. adenoids and tonsils were pooled separately). The number of cells from each donor to pool was estimated using flow cytometry data with the aim of pooling a similar number of S1+ B cells from each sample. Pooled cells were incubated with Fc blocker at 4 °C for 10 min followed by CITE-seq and sorting antibody cocktails in the following order at 4 °C: TotalSeq anti-CXCR3, anti-CCR6, and anti-CXCR5 antibodies for 10 min, and the remaining 21 CITE-seq antibodies and fluorescence-labeled sorting antibodies and viability dye (Aqua) for an additional 30 min (Antibodies are listed in Supplementary Data 11). Cells were then washed with PBS with 0.04% BSA and sorted on a BD FACS Aria™ III sorter (BD Biosciences, San Jose, CA). S1+ B cells from each tissue pool were sorted into a 200 μL low binding PCR tube (Eppendorf) with 50 μL culture medium (10% heat-inactivated FBS (VWR), 2 nM glutamine, 0.055 mM 2-mercaptoethanol, 1% penicillin/streptomycin, 1 mM sodium pyruvate, 10 mM HEPES, 1% non-essential amino acids in RPMI with glutamine), while S1- B cells were sorted into 1.5 mL low binding tubes (Eppendorf) with 300 μl RPMI culture medium with 20% heat-inactivated FBS. All collecting low-binding tubes were precoated with RPMI media with 20% FBS overnight at 4 °C prior to use. Sorting strategy is shown in Supplementary Fig. 12a. Antibodies, including barcoded and fluorescence-conjugated antibodies, are listed in Supplementary Data 11. S1⁺ cells were centrifuged and resuspended for single-cell partitioning without counting. An appropriate number of S1⁻ B cells were counted, aliquoted, centrifuged and resuspended for partitioning. Antibody concentrations used for CITE-seq were inferred based on the concentrations of antibodies of the same clones titrated for use in our flow cytometry staining of tonsil/adenoid cells. CITE-seq processing and demultiplexing Library construction and sequencing for B-2021 were described previously21. For B-306-3 and B-306-4, sorted S1+ and S1- B cells were mixed with the reverse transcription mix and partitioned into single-cell gel-bead in emulsion (GEM) using 10× Chromium Next GEM Single Cell 5’ Reagent Kit v2 for B-306-3 and 10× Chromium Next GEM Single Cell 5’ HT v2 for B-306-4 (10x Genomics). The reverse transcription step was performed in an Applied Biosystems Veriti 96-well thermal cycler (Applied Biosystems). 5’ single cell gene expression (GEX), cell surface protein (CSP), and B cell receptor (BCR) libraries were prepared as instructed by 10x Genomics user guides (https://www.10xgenomics.com/resources/user-guides/; CG000330 Rev F and CG000424 Rev C for the B-306-3 and B-306-4 run, respectively). Library quality and quantities were measured using a TapeStation system (Agilent) and a Qubit fluorometer (ThermoFisher). Libraries were pooled at a concentration of 10 nM and sequenced on the Illumina platform (NovaSeqX for B-306-3 and B-306-4, Illumina) using the following read lengths: Read 1:26 base pairs, Index 1:10 base pairs, Index 2:10 base pairs, Read 2:90 base pairs. Tonsil cells saved for bulk RNA-seq from each participant were resuspended in QIAzol and RNA was extracted using RNeasy micro kit (Qiagen) and standard RNA sequencing libraries were generated using Universal Plus mRNAseq kit (TECAN Genomics). These libraries were used to generate SNP calls for each donor. Sequencing results were demultiplexed and converted to FASTQ format using Illumina bcl2fastq software. The sequencing reads were adaptor and quality trimmed and then aligned to the human genome using the splice-aware STAR aligner and SNP calls were made using BCFtools104. The pooled single-cell RNA sequencing data were demultiplexed with a combination of demuxalot (v. 0.4.1)105 and demuxlet106 to match cells to each donor and identify doublets. For the pool with fraternal twin subjects, CNMC 124 and CNMC 125, cells were assigned by demuxalot, then reads with associated barcodes were extracted and reassigned via demuxlet to further refine the cell identities for twin subjects. CITE-seq data analyses Data from B-2021 set were processed as described21. For B-306-3 and B-306-4, CellRanger (v. 7.1.0)107 was used to map cDNA libraries to the GRCh38 genome reference (GRCh39-2020-A) and to count antibody tag features. Down sampling was performed using the cellranger aggr pipeline (10x Genomics) to normalize sequencing depth across B cell lanes. Data were further processed using Seurat (v. 5.1.0) in R 4.4.2. Cells were demultiplexed by SNP as described above (Methods: CITE-seq processing and demultiplexing). Surface protein library counts were transformed by using dsb (v. 1.0.4)108. For quality control, cells with fewer than 100 detected genes, greater than 30% mitochondrial reads, or gene counts greater than 25,000 were removed. To exclude cells with extremely high surface antibody counts, the top 0.5% of cells in the surface antibody total count distribution were removed. Cell clustering was performed by applying the FindNeighbors() function from Seurat on a distance matrix generated from the dsb-transformed surface protein data, followed by Leiden clustering on the resulting SNN graph using Seurat’s FindClusters() algorithm, with a resolution parameter of 0.3. Expression of selected genes was visualized using the ComplexHeatmap package (v. 2.22.0)109, and the percentage of cells per population for the S1+ and S1− cells was plotted using ggplot2 (v. 3.5.2)98. For CXCR3 protein and mRNA expression correlation analysis (Supplementary Fig. 6b), an equal number of subject- and tissue-matched cells were sampled from each cell type (S1− or S1+) from both INF and VAC subjects. S1− and S1+ cells were then down sampled to the maximum equal number across groups. To further assess the transcriptome of BSM subsets P1 to P4, we manually annotated B cell populations using protein expression data from CITE-seq for three reasons: (1) to align our analysis with flow cytometry data; (2) CXCR3 protein and mRNA expression levels were poorly correlated (Supplementary Fig. 6b); and (3) to minimize batch effects across different experimental sets. We used the removeBatchEffect function from the limma (v. 3.62.1)110 package to correct batch effects in CSP expression within the B-306-4 set. CSP expression from B-2021, B-306-3, and B-306-4 was exported as.csv files and imported into FlowJo (v. 10.9.0) for manual gating. The gated FlowJo files were then processed using CytoML (v. 2.18.0)111 and flowWorkspace (v. 4.18.0)112, and the resulting gates were merged with the CITE-seq metadata by matching identical cell barcodes. To identify differentially expressed genes among BSM P1 to P4 (Supplementary Fig. 6e), we used only the GEX from B-306-4 to minimize batch effects. Differential expression analysis was performed using the FindAllMarkers function in Seurat with the MAST algorithm, incorporating ‘Batch’ (defined as samples processed on different days within the B-306-4, set A-C), ‘Subject’, and ‘Tissue’ as latent variables (min.pct = 0.1, min.cells.feature = 3, min.cells.group = 3, logfc.threshold = 0.1). The PseudobulkExpression function was used to normalize count data for each BSM population, yielding representative expression values for differentially expressed genes within each subset. The top 20 genes with adjusted p-values < 0.05 and log2FC > 0, as identified by FindAllMarkers results, were visualized using pheatmap (v. 1.0.12)113. Pseudo-bulk and differential gene expression analysis Pseudo-bulk gene differential expression analysis and gene set enrichment analysis (GSEA) were performed as described previously114,115. In brief, all sorted cells in a given cell type (S1+ or S1-) and tissue (adenoid or tonsil) within a donor were computationally ‘pooled’ according to their B cell subset assignment by summing all reads for a given gene. Pseudo-bulk libraries composed of only a few cells (less than 5) and with fewer than 30,000 unique molecular identifier counts after pooling per library were removed from the analysis, as they are likely not modeled properly by bulk differential expression methods. Only cell type- and tissue-specific B cell subsets with more than 3 psuedobulk libraries were included for differential comparison. Genes expressed at low levels were removed for each cell type individually using the filterByExpr function from edgeR116. Differentially expressed genes were identified using the limma voom117 workflow, which models the log of the counts per million (CPM) of each gene. Scaling factors for library size normalization were calculated with the calcNormFactors function with method = ‘RLE’. Genes were ranked using the moderated T statistics for the relevant coefficient from the limma voom model. Differentially expressed genes between S1+ BSM P1 and S1+ BSM P2 cells were identified with a model accommodating paired analysis (formula: ~0 + CD21_CXCR3+Subject_Tissue). The S1+ P1 and P2 BSM subset (“CD21_CXCR3”) and tissue-specific subject (“Subject_Tissue”) were modeled as factor variables representing the BSM population (including “BSM P1” and “BSM P2”), tissue type from a specific subject (including “tonsil” and “adenoid”), respectively. The contrasts.fit function was then used to compare the estimated means between S1+ P1 BSM and S1+ P2 BSM. Enriched gene sets were identified using the pre-ranked gene-set enrichment analysis (GSEA) algorithm implemented in the fgsea (v. 1.32.0) R package. Gene set lists used for enrichment assessment (including Gene Ontology Biological Process (GO BP), GO Cellular Component (GO CC), GO Molecular Function (GO MF), Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome, the Molecular Signatures Database’s Hallmark collection, Blood Transcriptomic Modules and a few published datasets58,118,119) were collected and pooled. P values were adjusted using the Benjamini–Hochberg method for the whole gene set list. The pathways shown in Fig. 6k were manually curated from gene sets relevant to immunology. Signature scores of B cell subsets Gene sets for atypical MBC (atMBC) were collected from studies on various chronic diseases in humans and mice including malaria120, HIV infection37, systemic lupus erythematosus121, Sjögren’s syndrome122, rheumatoid arthritis, common variable immunodeficiency38, and rodent malaria Plasmodium chabaudi58, as well as human tonsil58. The gene set for GCB signature was derived from a human tonsil B cell study123. The cMBC signature was used from a recent study on human tonsil and murine malaria58. Other gene signatures used to characterize B cells were obtained from MSigDB124 (msigdbr R package, v. 7.5.1). The AddModuleScore function of Seurat was applied with default parameters to score each signature in each B cell. To represent the levels of each signature set in the populations of interest, the median score for each tissue- and subject-specific sample within the S1- B cell subset from the B-306-4 set was used as an estimate. The median scores of per-subject median scores for each B cell population were then displayed in the heatmap of Fig. 6i. Trajectory analysis To compare the lineage differentiation of S1+ B cells in INF and VAC tonsils and adenoids, we applied Slingshot (v. 2.14.0)125 to infer lineage trajectories. S1+ cells from B-306-4 set were used in this analysis. B cell subsets were assigned by manual gating as described in Fig. 6f. “PC” and “PreGC” cells were removed due to low number of cells for an accurate lineage estimation. First, uniform manifold approximation and projection (UMAP) embedding was performed on the normalized single-cell surface protein expression profile from the CITE-seq dataset to obtain a low-dimensional representation. P1-P4 BSM populations were provided as cluster input and “Naïve/USM B cell” population was appointed as the start cluster to Slingshot for trajectory reconstruction. Cells were ordered through inferred pseudotime based on gene expression to indicate their differentiation progress. Trajectories for INF and VAC groups were estimated separately. BCR sequence analysis and clonal clustering BCR repertoire sequence data were analyzed using the Immcantation (www.immcantation.org, v. 4.5.0) framework. Starting with filtered CellRanger output, V(D)J genes for each sequence were aligned to the IMGT GENE-DB reference database obtained 2/09/24 using IgBLAST (v. 1.22.0)126 and Change-O (v. 1.3.0)127. Nonproductive sequences, cells without associated constant region calls, cells identified as arising from doublets or negative droplets, and cells with multiple heavy chains were all removed. Samples within each subject were pooled and sequences were grouped into clonal clusters, which contain B cells related to each other by somatic hypermutation (SHM) from a common V(D)J ancestor. Using the hierarchicalClones function of scoper (v. 1.3.0)128, sequences within these groups differing by a participant-specific normalized Hamming distance threshold within the CDR3 region were defined as clones using single-linkage clustering (one subject required a slightly lower threshold than others)129. This threshold was determined by fitting a gamma/Gaussian mixture model to the distance to nearest sequence neighbor distribution using SHazaM(v. 1.2.0)130. These heavy chain-defined clonal clusters were further split if their constituent cells contained light chains that differed by V and J genes. Within each clone, germline sequences were reconstructed with D segment and N/P regions masked (replaced with “N” nucleotides) using the createGermlines function within dowser (v. 2.3)131. SHM was calculated as the frequency of non-ambiguous mismatches from each cell to the V gene (IMGT positions 1–312) of its reconstructed germline sequence. Paired scBCR-seq data were integrated with CITE-seq data based on matched cell barcodes. To quantify B cell clonal diversity, we calculated Simpson’s diversity for each tissue-specific sample using the alphaDiversity function of alakazam (v. 1.3.0)127. Lower values of Simpson’s diversity indicate a greater probability of two random sequences belonging to the same clone, consistent with larger clones. To account for differences in sequence depth, samples within each comparison were down-sampled to the same number of sequences, and the mean of 1000 such re-sampling repetitions was reported. Subject/tissue/cell type samples or populations with <70 B cells were excluded, which led to the exclusion of all S1+ cells from CNMC 10 and CNMC 99 (both CON with no history or evidence of SARS-CoV-2 infection or vaccination). Clonal overlap among tissues or B cell subsets can be used as a measure of immunological connectivity. Clonal overlap was calculated using the Jaccard index, which for each pair of B cell subsets is the number of unique clones found in both subsets (intersect) divided by the total number of unique clones among the two subsets (union). Clones were relabeled as “S1-” clone when the ratio of S1⁺ to S1⁻ sorted B cells within the clone was less than 0.1. After clonal clustering, only heavy chain sequences were used for subsequent analysis. Clone expansion and B cell subset transition were estimated with indices (STARTRAC-expa and STARTRAC-tran) from the STARTRAC package (v. 0.1.0)68,69. In STARTRAC-tran indices analysis, S1+ or S1- subject-specific populations with less than 10 cells were removed. Clones from INF and VAC groups were analyzed. To infer lineage trees, we estimated tree topologies, branch lengths, and subject-wide substitution model parameters using maximum likelihood under the GY94 model132,133. Using fixed tree topologies estimated from the GY94 model, we then estimated branch lengths and donor-wide parameter values under the HLP19 model in IgPhyML (v. 2.0.0)132. Trees were visualized using dowser (v. 2.3)131 and ggtree (v. 3.14.0)134. ATAC-seq ATAC-seq data processing and analysis Frozen tonsil cells were thawed and stained with antibodies listed in Supplementary Data 11 for 30 min at RT. Cells were washed with PBS twice and resuspended in RPMI with 10% FBS for sorting. 10,000 viable P1 (CXCR3-CD21+), P2 (CXCR3+CD21+), P3 (CXCR3+CD21-) and P4 (CXCR3-CD21-) BSM were sorted into 500 mL of FACS buffer using an FACSAria Fusion Sorter (BD) (cell sorting strategy is shown in Supplementary Fig. 12b). Cells were pelleted and resuspended in 50 μL of transposase mixture including 25 mL 2xTD buffer (Illumina), 2.5 mL TDE1 (Illumina), 0.5 mL 1% digitonin (Promega) and 22 μL water. Tagmentation was performed by incubation at 37 °C for 30 min at 300 rpm. Following incubation, DNA was purified using a Qiagen MinElute kit, eluting samples in 10 μL. Purified tagmented DNA was PCR-amplified using previously described primers135, with 12 cycles of amplification. Amplified libraries were purified using a Qiagen PCR cleanup kit and sequenced for 50 cycles (paired-end reads) on a NovaSeq 6000 (Illumina). ATAC-seq was done in three biological replicates per BSM subset (samples shown in Supplementary Data 2). ATAC-seq primers are listed in Supplementary Data 11. ATAC-seq data was processed using the chrom-seek pipeline (v. 1.0.0)136 with --assay ATAC (https://github.com/OpenOmics/chrom-seek). Reads were trimmed with Cutadapt (v. 4.4)137. All reads aligning to the Encode hg38 v1 blacklist regions138 were identified by alignment with BWA (v. 0.7.17)139 and removed with Picard SamToFastq. Remaining reads were aligned to an hg38 reference genome using BWA. Reads with a mapQ score less than 6 were removed with SAMtools (v. 1.17)104 and PCR duplicates were removed with Picard MarkDuplicates. Data was converted into bigwigs for viewing and normalized by reads per genomic content (RPGC) using deepTools140 (v. 3.5.1) using the following parameters: --binSize 25 --smoothLength 75 --effectiveGenomeSize 2805636331 --centerReads --normalizeUsing RPGC. Averaged bigwigs were created using the bigwigAverage function of deepTools (v. 3.5.4)140. Peaks were called using macsNarrow141 (macs v. 2.2.7.1) with the following parameters: -q 0.01 --keep-dup = “all” -f “BAMPE”. Differential peaks were called using DiffBind (v. 2.15.2)142 and its Deseq2 differential caller with default parameters. Peaks were considered significant with an FDR value less than 0.1. Motif analysis was completed using the MEME suite (v. 5.5.5)143. Known motif enrichment analysis was accomplished using AME on a combined jolma 2013, jaspar 2018 core vertebrate non-rendundant, and HOCOMOCO (v.11)144 full human mono database. Downstream analyses and results visualization were performed with R (v.4.4.2) and visualized with ggplot2 (v. 3.5.1)98. Differentially accessible regions (DAR) and pathway enrichment analysis Differential peaks from comparing P1 BSM and P2 BSM (adjusted p-value < 0.10) were selected for downstream analysis. To explore the functional significance of these DARs, pathway enrichment analysis was conducted using GREAT (Genomic Regions Enrichment of Annotations Tool, v. 4.0.4)145,146. The genomic coordinates of DARs were converted to BED files and uploaded to the GREAT web server. The analysis was performed mapping to GRCh38 (UCSC hg38, Dec. 2013) and using the default association rules, which map genomic regions to nearby genes based on a basal plus extension model (5 kb upstream, 1 kb downstream, and up to 1 Mb extension from the transcription start site). Enriched terms from GO BP and GO MF, MSigDB, Reactome, and other curated databases were extracted. Pathways and terms with an FDR (adjusted p-value) <0.05 were considered significant. Top20 GO BP pathways ranked by FDR (adjusted p-value) values were visualized with bar plot. IRF4 complex motif search with FIMO To further evaluate the enrichment of IRF4 complex motifs in P1 BSM and P2 BSM, a search with stringent EICE (GGAANNGAAA), ISRE (A/GNGAAANNGAAACT) and two AICE (0 bp/4 bp: 0 bp – GAAATGA(G/C)TCA; 4 bp – TTTCNNNNTGA(G/C)TCA)147 motifs using Find Individual Motif Occurrences (FIMO) on the DAR sequences obtained from P1 BSM to P2 BSM comparison was performed66. The enrichment statistics were calculated as above using a two-tailed version of Fisher’s exact test. For Supplementary Fig. 6i, data for memory B cells (MBC) and plasma cells (PC) were retrieved from a published dataset (https://zenodo.org/records/8373756)97 and loaded into Signac (v1.11.0)148. Plots were created with the Signac CoveragePlot function. BCR signaling assessment by phosphorylation staining Tonsil cells were thawed, rinsed with RPMI supplemented with 0.1% FBS, and rested for 80 min at 37 °C in the same medium. After resting, the cells were incubated with a live/dead stain in PBS for 15 min at RT and then washed once with PBS. This was followed by staining the cells with antibodies against CXCR3 and CXCR5 for 5 min at RT. Subsequently, the remaining surface antibody mix was added (Supplementary Data 11), and the cells were resuspended in FACS buffer for 20 min at RT90. Afterward, the cells were washed twice with RPMI supplemented with 0.1% heat inactivated FBS and then resuspended in pre-warmed RPMI with 10% FBS and stimulated with anti-BCR antibodies, as previously described70,149,150. Stimulation was carried out at 37 °C for 2 min using 10 mg/mL goat F(ab’)2 anti-human IgA/G/M (Jackson ImmunoResearch Laboratories). For detecting phosphorylated signaling intermediates, the cells were fixed and permeabilized using BD Cytofix and Phosflow Perm/Wash buffers (BD Biosciences), then stained with PE-phosphorylated Syk (p-Y348) and Alexa Fluor 488-phosphorylated PLCγ2 (p-Y759) antibodies (BD Biosciences). Samples were acquired on an Aurora cytometer (Cytek), and analysis was performed using FlowJo (v. 10.9.0). Gating strategies are shown in Supplementary Fig. 13a. In vitro plasmablast differentiation and proliferation assay Tonsil cells were thawed and stained with viability dye and surface antibody mix (Supplementary Data 11) at RT for 30 min. Then cells were washed with PBS twice and resuspended in RPMI with 10% FBS at concentration of 5 million cells/mL for sorting. P1, P2, P3 and P4 BSM were sorted on Aria sorter (BD) into 0.3 mL RPMI with 20% FBS in 1.5 mL tubes (Supplementary Fig. 12b). Sorted cells were centrifuged at RT for 10 min and then were labeled with 0.5 μM CFSE (CellTrace CFSE cell proliferation kit, ThermoFisher) in 1100 μL PBS at RT in the dark for 10 min. Then, cells were washed with pre-warmed RPMI + 10% FBS as described149. Allogenic B cell-depleted PBMCs were prepared using a B-cell depletion kit (Dynabeads CD19 Pan B, ThermoFisher) from PBMCs of an unrelated healthy donor. The sorted memory B cells and allogeneic B cell-depleted PBMC were co-cultured at a 1:9 to 2:8 ratio with 2.5 mg/mL R848 and 1000 U/mL recombinant human IL-2 for 4 days at 1 × 106 cells per well in a 96-well flatbottom plate. RPMI with 10% FBS was used for culture. The cells were collected and stained with antibodies against CD19, CD20, CD3, CD27, CD21, IgD, CD38, and CXCR3, fixed (Lysing Solution, BD Biosciences), permeabilized (Permeabilizing Solution 2; BD Biosciences) and stained with antibodies against IgG, IgA, IgM. The cells were acquired on an Aurora cytometer (Cytek) and analyzed using FlowJo (v. 10.9.0). Antibodies are listed in Supplementary Data 11. Gating strategies are shown in Supplementary Fig. 13b. The division index, a measure of the overall proliferative response, is the average number of divisions undergone per cell in the total population, including cells that have not undergone division. T cell functional assays—intracellular cytokine staining We reanalyzed T cell functional assay data from our prior study21, which included 13 INF and 13 CON subjects with paired adenoid and tonsil samples. Briefly, cells were thawed and stimulated with PMA (50 ng/ml, Sigma) and ionomycin (1000 ng/ml, Sigma) for 2.5 h in the presence of anti-CD107a (BioLegend), GolgiSTOP (monensin, BD), and GolgiPlug (BFA, BD), followed by surface staining, fixation, permeabilization, and intracellular staining as described previously21. Tissue processing and staining for immunofluorescence assay Formalin-fixed, paraffin-embedded (FFPE) adenoid and tonsil tissue blocks were cut into 5 μm sections and mounted onto charged slides. Two paired tonsil and adenoid samples (from one INF donor and one VAC donor) were loaded onto the same slide. For staining, slides were deparaffinized and tissue rehydrated in deionized water. Antigen retrieval was performed by incubating slides in antigen retrieval (AR) buffer (Cepham Life Sciences) for 45 min in a steamer (preheated, approximately 95 °C). After 45 min, slides were taken out from the steamer and allowed to cool to RT. Sections were permeabilized, blocked for 1 h in PBS containing 0.3% Triton X-100 (Sigma-Aldrich), 1% bovine serum albumin (Jackson Immune Research). Sections were stained with titrated amounts of non-conjugated primary antibodies, followed by overnight incubation at 4 °C. Slides were then washed with PBS (3 times, 10 min each) and stained with the appropriate secondary antibodies for 2 h at RT. Slides were washed and blocked again for 1 h at RT with a 1:10 dilution of normal mouse and rabbit or goat serum. Then, slides were stained with titrated amounts of directly conjugated antibodies for 2 h at RT. After three final washing steps and staining with the nuclear marker TOPRO (ThermoFisher), slides were mounted with prolonged gold anti-fade mounting media (ThermoFisher) and sealed with a glass coverslip. Antibodies are listed in Supplementary Data 11. Tissue sections were imaged (using confocal system, Leica Stellaris RTB WLL FLIM) as three-dimensional (3D) tile scans and subsequently mosaic-merged to generate a continuous representation. To minimize imaging artifacts, corrections were applied for motion-induced distortions, 3D alignment inconsistencies, and thermal drift across sequential z-sections. Additionally, crosstalk and color calibration adjustments were performed using Huygens Pro (version 24.04.0p3 64-bit, Scientific Volume Imaging BV). Image deconvolution was conducted within the same software to enhance signal resolution. The reconstructed images were further processed using Imaris (v. 10.2.0, Oxford Instruments). A combination of colocalization analysis, ChannelArithmetics, Xtension, Imaris installed machine learning-based classification, and masking techniques were employed to delineate distinct anatomical regions as additional computational channels, including follicles, germinal centers, interfollicular regions, epithelium, and crypt structures. The Surface module of Imaris was utilized to generate cell objects, integrating nuclear signals and perinuclear regions derived from the preceding image processing steps. Quantitative data were extracted from processed files via a custom parser script, which reformatted surface object statistics into “.csv” files optimized for direct import into FlowJo (v. 10.9.0) for gating and further analysis. Statistical analysis and visualization were further processed in GraphPad Prism (v. 10.2.0392). SPACE analysis of immunofluorescence imaging data Spatial Patterning Analysis of Cellular Ensembles (SPACE) is an R package designed to identify complex spatial patterns at the cell and tissue levels78. Cell objects from immunofluorescence images were annotated with 8 populations as in Fig. 10e by manual gating with FlowJo (v. 10.9.0) and labeled by an R-based customized script. The 10 µm radius captures close cellular associations and was chosen for SPACE “census_image” function. The number of neighborhoods was chosen to achieve 5× tissue coverage. Covariation plots were created using the SPACE “learn_pattern” function. Spatial transcriptomic profiling with Xenium in Situ platform Slides were prepared following the manufacturer’s instructions and workflow for FFPE tissue samples (CG000578 Rev A; 10x Genomics). A 5-µm section from the tissue block containing the same paired tonsil and adenoid samples (one from INF donor and one from VAC donor) used for immunofluorescence was carefully attached to the sample area on a Xenium slide (Histoserv, MD). Samples are listed in Supplementary Data 2. Xenium slides were deparaffined and rehydrated and were then assembled into the Xenium Cassette. Deparaffinized slides in the Xenium cassette were decrosslinked (CG000580 Rev A) and immediately underwent probe hybridization, ligation, and amplification (CG000582 Rev D) using the Xenium 5000 human gene panel (Prime 5K Human Pan Tissue & Pathways Panel). With autofluorescence quenching and nuclei staining, the tissue images were captured and analyzed by the Xenium Analyzer (PN-1000569, instrument software v2.0.1.0). Regions of interest were manually selected from the scanned images. Post-run data for each slide was obtained using default parameters for downstream analysis. Data were processed by Xenium analysis software (v 2.0.0.10). The raw count matrix was pre-processed using the Seurat package (v. 5.1.0)96,97 in R (v. 4.4.2). For quality control, cells with at least nCount > 40, 15 nFeature > 10, cell_area > 10 & <200 were retained151. Raw count data were normalized using the SCTransform function with method “glmGamPoi”. Dimension reduction was performed using the runPCA function and the optimal number of principal components was selected using the ElbowPlot function. Cell clusters were determined using the FindClusters function. We annotated cell populations with human tonsil reference version 2 with Seurat (v. 5.0.2)96 and Azimuth (v. 0.5.0)97,152. Cell populations with less than 150 cells per slide were removed (granulocytes, mast, preB/T and PC/doublet cells). Some cell types were merged (B naïve includes B naive and B activated; B memory includes B memory & FCRL4/5+ B memory; CD4 Non-TFH includes CD4 Non-TFH & CD4 TCM; CD4 TFH includes CD4 TFH & CD4 TFH Mem; CD8 non-naïve includes CD8 T & CD8 TCM; gdT_MAIT includes MAIT/TRDV2 + gdT & non-TRDV2 + gdT; Mono/Macro includes Mono/Macro & Cycling myeloid; PB/PC includes PB & PC). Additional statistical analyses Correlations were analyzed using Spearman’s rank correlation test using base R and corrplot (v 0.95). Paired comparisons were performed with Wilcoxon signed-ranks test with ggpubr (v. 0.6.0)153 and visualized with ggplot2 (v. 3.5.1)98 in R (v. 4.4.2). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1-10 Supplementary Data 11 Reporting Summary Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Data 1-10 Supplementary Data 11 Reporting Summary Transparent Peer Review file Source data Source Data Source Data