Immunological memory diversity in the human upper airway

Abstract

The upper airway is an important site of infection, but immune memory in the human upper airway is poorly understood, with implications for COVID-19 and many other human diseases^1–4. Here we demonstrate that nasal and nasopharyngeal swabs can be used to obtain insights into these challenging problems, and define distinct immune cell populations, including antigen-specific memory B cells and T cells, in two adjacent anatomical sites in the upper airway. Upper airway immune cell populations seemed stable over time in healthy adults undergoing monthly swabs for more than 1 year, and prominent tissue resident memory T (T_RM) cell and B (B_RM) cell populations were defined. Unexpectedly, germinal centre cells were identified consistently in many nasopharyngeal swabs. In subjects with SARS-CoV-2 breakthrough infections, local virus-specific B_RM cells, plasma cells and germinal centre B cells were identified, with evidence of local priming and an enrichment of IgA⁺ memory B cells in upper airway compartments compared with blood. Local plasma cell populations were identified with transcriptional profiles of longevity. Local virus-specific memory CD4⁺ T_RM cells and CD8⁺ T_RM cells were identified, with diverse additional virus-specific T cells. Age-dependent upper airway immunological shifts were observed. These findings provide new understanding of immune memory at a principal mucosal barrier tissue in humans.

Although peripheral blood sampling has proven useful for answering myriad clinical and research questions, sampling blood inherently misses important information at sites of disease. The upper airway is the primary site of infection for many human pathogens but has proved to be a challenging site to study immunologic memory, which is central to the protective immunity provided by almost all vaccines and infections. Critical outstanding questions remain about human germinal centre (GC) biology and human upper respiratory tract (URT) B cell and T cell immune memory, including development, abundance, maintenance and residency. This is illustrated by two of the largest and most comprehensive human tissue immunological studies, which had no representation of URT tissues^5,6. Tissue resident memory T (T_RM) cells are distinct memory T cells that do not circulate and are detectable in numerous human tissues^5–11. T cell analysis from organ donors provided evidence for human tissue-specific T_RM cells^9,12. Human tissue resident memory B (B_RM) cells also exist in several tissues¹³, but their biology in the upper airway is unknown. Additionally, while GCs are important for the generation of immune memory, they are classically very difficult to study in humans due to their inaccessibility^14–19. Here we show that nasal and nasopharyngeal (NP) swabs can be used to obtain insights into immune memory of the human upper airway in adults, including the presence and persistence of antigen-specific B cell and T cell memory.

Several swab types and swabbing techniques were initially explored before suitable swabs and protocols were established for collecting upper airway cells. To determine the feasibility of upper airway swab sampling for studying human URT immune cell populations, we enroled a cohort of approximately 30 healthy adults to undergo monthly NP swabs for more than 1 year, with samples analysed by spectral flow cytometry. Monthly NP swab collections were well tolerated with minimal attrition of volunteers from the cohort. Up to 5 million viable cells could be collected with a single NP swab (median of more than 100,000; Fig. 1a), with approximately one-third being CD45⁺ immune cells (median 42,190 and maximum 2.75 million; Extended Data Fig. 1a), including B cells, CD4⁺ T cells and CD8⁺ T cells (Fig. 1b,c and Extended Data Fig. 1b). Epithelial cells were the primary CD45⁻ cell types (Extended Data Fig. 1c–e). Upper airway immune cell populations, including total CD45⁺ cells, B cells, T cells, CD4⁺ T cells and CD8⁺ T cells, seemed stable over time (Fig. 1c). T cells from NP swabs included prominent putative T_RM cell populations (Fig. 1d). CD8⁺ T_RM cells were defined by coexpression of CD69 and CD103 on non-naive CD8⁺ T cells²⁰ (Fig. 1d). URT CD4⁺ T_RM cells were defined as CD69⁺ non-naive cells, with subpopulations coexpressing other putative T_RM cell markers such as CD103 (ref. 20) (Fig. 1d). Putative B_RM cells were observed in the NP swabs as CD69⁺ B cells¹³ (Fig. 1d). On examining the longitudinal donor cohort for these immune memory cell types, CD8⁺ T_RM cells, CD4⁺ T_RM cells and putative B_RM cell populations were each observed to be stable over time (Fig. 1d).

Fig. 1 |. Distinct human upper airway cell populations sampled by NP swabs.

a, Number of viable cells per NP swab; each dot represents one NP swab per donor, black line is the median (n = 170 swabs). b, Representative flow cytometry plots of NP swab total B cells and T cells (top), and CD4⁺ and CD8⁺ T cells (bottom). c, URT CD45⁺, B cell and T cell stability over time, calculated as percentage change from median frequency for each donor; individual donor data in grey (n = 20 donors). d, Representative flow cytometry of URT CD4⁺ and CD8⁺ T_RM cells, and putative B_RM cell populations; stability of these populations over time as percentage change from median; individual donor data in grey (n = 20 donors). e, Combined Seurat UMAP for approximately 70,000 cells from 12 NP swabs (n = 12 donors); doughnut plot shows representative proportions of each cell type by scRNA-seq/CITE-seq. f, T cell subset UMAP generated from the approximately 19,000 T cells in e; T cell subset identities assigned by gene expression profiling and CITE-seq. g, Dotplot of select cluster-defining genes for T cell subsets in f. Protein markers (CITE-Seq) in bold with ‘Prot’. h, GSEA analysis of T cell subsets in f. i,j, Dotplots of select genes from subset analysis of T_FH cell (i) and T_reg cell (j) dominated clusters from f. k,l, SARS2-specific T cells in NP swabs determined experimentally by AIM assay (spike (S) or non-spike SARS2 antigen-specific) and personalized SARS2-specific TCR repertoire datasets for each subject (k), or identified bioinformatically using public datasets (l). m, SARS2-specific CD4⁺ versus CD8⁺ T cells, overlaid on UMAP from f. n, SARS2-specific T cell subtypes from k–m. See Extended Data Figs. 2d–g,i–k and 3 for further details and stratifications. abs(NES), absolute value of the normalized enrichment score; CTL, cytotoxic T lymphocyte; DC, dendritic cell; mono/mac, monocyte/macrophage; mT_FH, mantle T_FH; NK, natural killer; ns, not significant; P_adj, adjusted P value; T_eff, effector T.

To better characterize URT immune cell populations, we performed single-cell RNA sequencing (scRNA-seq) together with cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) using a nine antibody CITE-seq panel on single-cell suspensions of NP swabs from 12 adult donors (Supplementary Table 1). Numerous epithelial cell (Extended Data Fig. 1d,e) and immune cell (Fig. 1e) populations were identified by single-cell transcriptome profiling, and were conserved across donors (Extended Data Fig. 1f). Approximately one-half of all NP cells identified were B cells, 28% were T cells, 15% were epithelial cells and the remainder (less than 10%) other immune cells (Fig. 1e), with Seurat analysis confirming high concordance in cell identities by gene expression analysis and CITE-seq antibody binding (Extended Data Fig. 1g,h). B cells and T cells were further clustered independently, and distinct cell types were defined on the basis of differential gene expression, gene set enrichment analysis (GSEA)^21–23 and B cell receptor (BCR) and T cell receptor (TCR) VDJ sequencing analysis. Several B cell populations were identified, including memory B (B_mem) cells, antibody-secreting cells (ASCs), B_RM cells and others (Extended Data Fig. 2a,b), discussed in more detail later. Numerous distinct CD4⁺ and CD8⁺ T cell populations were identified and well represented, including CD4⁺ T_RM cells, CD8⁺ T_RM cells, regulatory T (T_reg) cells, mucosal-associated invariant T (MAIT) cells and follicular helper T (T_FH) cells (Fig. 1f–h and Extended Data Fig. 2c–g). Two CD8⁺ T_RM cell populations were identified, with similar transcriptional patterns to CD103⁺ and CD103⁻CD8⁺ T_RM cells observed in the human intestine²⁴ (Fig. 1h). Compared with CD8⁺ T_RM cells, CD4⁺ T_RM cells are less well defined in human tissues; however, a distinct CD4⁺ T_RM cell cluster was identified that was enriched for tissue residency markers and chemokine receptors, including CD69, ITGAE (CD103), CCR6 and CXCR6 (Fig. 1h). The CD4⁺ T_RM cell cluster was also enriched for KLRB1 and IL7R (Fig. 1h). Higher resolution analysis of several clusters revealed more populations of interest. Closer examination of the ‘naive CD4⁺ T cell’ cluster (CD4 T_N) showed an overlapping CD4⁺ memory T cell cluster distinguished by FAS expression (FAS⁺CD4 T_mem) (Fig. 1f and Extended Data Fig. 2e). Within the T_FH cell cluster, it was possible to identify GC-T_FH cells, GC-T_FH cells with evidence of type I IFN signalling (GC-T_FH^IFN) and mantle T_FH cells (Fig. 1i and Extended Data Fig. 2f)²⁵. More granular examination of the T_reg cell dominated cluster revealed follicular regu latory T (T_FR) cells, T_FH cells and T helper 17 (T_H17) cells (expressing IL17a and IL17f but not FOXP3) in addition to T_reg cells (Fig. 1j and Extended Data Fig. 2g). Overall, a rich diversity of CD4⁺ and CD8⁺ T cell populations were identified in the human upper airway.

It is important to better understand pathogen-specific immune memory in human tissues that are common sites of infection. Single-cell TCR sequencing (scTCR-seq) was completed on all 12 NP swab samples (Fig. 1e). Among these donors, all reported primary COVID-19 vaccinations and seven had evidence of a SARS-CoV-2 (SARS2) breakthrough infection (Supplementary Table 1). We used a combination of experimental and computational methods to query whether SARS2-specific CD4⁺ and CD8⁺ T cells were present in NP tissue¹ (Fig. 1k–n). SARS2 spike-specific T cell clonotypes were defined experimentally from peripheral blood mononuclear cells (PBMCs) by an activation induced marker (AIM) assay antigen-specific CD4⁺ T cell and CD8⁺ T cell sorting and scRNA-seq pipeline (Extended Data Figs. 2h and 3), generating personalized SARS2-specific TCR repertoire reference datasets for each subject, specific for SARS2 spike and non-spike epitopes. Upper airway SARS2 spike- and non-spike-specific memory T cells were then identified by matching TCRs with 100% CDR3 amino acid sequence identity between NP swabs and the PBMC-derived personalized SARS2-specific TCR repertoire datasets for each subject (AIM⁺; Fig. 1k). Additional SARS2-specific CD4⁺ T cells were identified in swab samples by matches to a TCRα sequence motif known to recognize an immune-dominant HLA–DPB1*04 S^167–180 epitope²⁶ (public TCRA; Fig. 1l) or TCRs with 100% amino acid sequence identity to TCRs in select databases^27,28 and papers^29,30 (public ref; Fig. 1l). Altogether, 250 SARS2-specific CD4⁺ and CD8⁺ T cells were identified in the NP swabs of these 12 subjects (Fig. 1m,n). SARS2-specific CD4⁺ and CD8⁺ T cells were identified in the upper airways of all seven subjects who had recent SARS2 infections (Extended Data Fig. 2k and Supplementary Table 1). These included both SARS2 spike and non-spike epitope specificities. Upper airway SARS2-specific CD8⁺ T_RM cells were found in most of the previously infected individuals and represented a plurality of the SARS2-specific T cells present (Fig. 1n and Extended Data Fig. 2k). SARS2-specific CD4⁺ T_FH cells and T_RM cells were also identified in upper airway tissue of most subjects (Fig. 1n and Extended Data Fig. 2k). Several other types of differentiated SARS2-specific CD4⁺ T cells were identified in upper airway samples from previously infected individuals, including central memory T (T_CM) cells, T_reg cells and T_H17 cells (Fig. 1n and Extended Data Fig. 2g). A small number of spike-specific CD4⁺ or CD8⁺ T cells were identified in subjects who only had evidence of COVID-19 vaccination or had evidence of a remote subclinical SARS2 infection (Extended Data Fig. 2k and Supplementary Table 1). In sum, these experiments demonstrate both virus-specific memory CD4⁺ T_RM cells and CD8⁺ T_RM cells in the human upper airway after an infection. The experiments further demonstrate diverse differentiated T cell types in the upper airway in these same individuals to the same antigens. Last, these experiments demonstrate antigen-specific T cell clones with identical TCRs located in both peripheral blood and as T_RM cells in the human upper airway.

Unexpectedly, B cells with surface phenotypes consistent with GC B (B_GC) cells (CD19⁺CD20⁺CD38⁺) were present in NP swab samples from many individuals according to flow cytometry (Fig. 2a and Extended Data Fig. 4a). GC-T_FH CD4⁺ T cells were identified in the same individuals by flow cytometry (CD4⁺CXCR5⁺PD-1^hi) (Fig. 2a and Extended Data Fig. 4a). B_GC cells and GC-T_FH cells were also validated by intracellular staining for BCL6 (Extended Data Fig. 4b). Collection of B_GC cells and GC-T_FH cells by NP swabs was confirmed by scRNA-seq (Fig. 1i and Extended Data Fig. 2f for GC-T_FH cells; Fig. 2b–d and Extended Data Fig. 2a for B_GC cells), including SARS2-specific GC-T_FH cells (Extended Data Fig. 2f). Among SARS2-specific T_FH cells, several subpopulations were observed (Extended Data Fig. 2f). For subjects with recent SARS2 infection, T_FH cells with spike and non-spike specificities were observed (Extended Data Fig. 2i), and were the most abundant SARS2-specific CD4⁺ T cell subset overall (Fig. 1n and Extended Data Fig. 2k). SARS2-specific T_FH cells were found exclusively in subjects with breakthrough infections (Extended Data Fig. 2i–k and Supplementary Table 1). More granular B_GC cell scRNA-seq analyses identified dark zone (‘DZ’; for example, AICDA, CXCR4, MKI67) and light zone/intermediate (‘LZ/Int’; for example, LMO2, CD83) differentiation states (Fig. 2b,c and Extended Data Figs. 2a and 4c,d; GSEA, Fig. 2d). Finally, cell cycle of the B_GC cells was assessed and demonstrated to match canonical B_GC cell-cycle states (Extended Data Fig. 4c). Reproducible sampling of GC lymphocytes by NP swabs indicated that swabbing may sample immune cells derived from both URT lymphoid tissue and mucosal epithelium.

Fig. 2 |. Upper airway mucosal B_GC cells and GC-T_FH cells consistently sampled by NP swabs.

a, Representative flow cytometry plots of GC populations from NP swabs (B_GC: CD20⁺CD38⁺ of CD19⁺ B cells and GC-T_FH: CXCR5⁺PD-1^hi of CD4⁺ T cells). b, B_GC cell subset UMAP of approximately 5,500 cells from the B_GC cell clusters shown in Extended Data Fig. 2a, demonstrating distinct dark zone (DZ) and light zone (LZ)/intermediate (Int) states. c,d, Characterization of B_GC cells from NP swabs by scRNA-seq transcriptomic profiling (c) and GSEA (d). e, Schematics and representative images of endoscopic-guided MT and NP (adenoid) swab sampling from the same donor. f, Immunophenotyping from paired swabs collected by endoscopy compared with control adenoid tissue (from a different donor). g, Stability of URT GC populations (B_GC cells and GC-T_FH cells) from NP swabs over time by flow cytometry (data shown for NP swabs from original longitudinal cohort donors with B_GC cell counts greater than 100; n = 15 donors). h, Clonal lineage of a B_GC cell from an NP swab, represented as a phylogenic tree. Each tip of the tree represents a unique heavy chain (HC) sequence determined by its transcriptomic profile; coloured by heavy chain constant region gene (isotype). Grey dotted lines indicate intervals of five estimated heavy chain nucleotide mutations. Panel e was created in part using BioRender (https://Biorender.com).

Adenoids are a secondary lymphoid organ located at the posterior NP wall. Unlike encapsulated lymph nodes, adenoids have a single layer of fenestrated epithelium with channel-like invaginations known as crypts. On the basis of the swab results, we considered that adenoid tissue with functional GCs may be present in many adults and that NP swabs may directly sample GC lymphocytes from adenoid tissue through crypts. To test this suggestion, we pursued several experimental approaches. First, using nasal endoscopy to directly visualize the anatomic site being sampled, we separately collected swab samples from the nasal cavity epithelium approximately at the midpoint of the inferior nasal turbinate (MT) and from the adenoids of healthy adult volunteers (Fig. 2e and Supplementary Videos 1–3). Spectral flow cytometry immunophenotyping of samples collected by endoscopic-guidance showed that B_GC cells and GC-T_FH cells were found exclusively in adenoid swabs (Fig. 2f). Cells from an adenoid tissue biopsy were examined as a comparator. Longitudinal NP swab sampling of numerous individuals with consistently detectable adenoids by flow cytometry demonstrated that GC cells were sampled reproducibly, and relatively stable over time (Fig. 2g). Furthermore, substantial somatic hypermutation (SHM) was observed in B_GC cells in clonal lineages collected by swabs (Fig. 2h and Extended Data Fig. 4e).

Paired NP posterior wall and MT swab samples were next collected from 38 adult donors for high parameter spectral cytometry immunophenotyping exploring immune memory cell types. Highly divergent frequencies of immune cell types existed between the two URT anatomic sites (Fig. 3a–q,u and Extended Data Fig. 5a–h). Compared with adenoids, turbinate epithelium possessed a significantly lower CD4⁺:CD8⁺ T cell ratio (P < 0.0001, Fig. 3g), few naive B cells (P < 0.0001; Extended Data Fig. 5b), few naive T cells (P < 0.0001; Extended Data Fig. 5c,d) and a higher frequency of T_reg cells (P < 0.001; Fig. 3u and Extended Data Fig. 5a,e). B_GC cells and GC-T_FH cells were found exclusively in NP tissue (P < 0.0001 for both; Fig. 3h,j). Among memory lymphocytes, turbinate epithelium possessed higher frequencies of CD8⁺ T_RM cells (CD69⁺CD103⁺; 95% versus 78%; P < 0.001; Extended Data Fig. 5f), most of which were also CD49a⁺ (Fig. 3m and Extended Data Fig. 5a). CD4⁺ T_RM cells (Fig. 3l) and putative B_RM cells (Fig. 3k) were detected in turbinate epithelium in addition to NP tissue. Many of these cell types involved in immune memory were not detected in blood (Extended Data Fig. 5i). Notably, B_mem cells in turbinate epithelium (including B_RM cells) were significantly more likely to be IgA⁺ than B_mem cells in NP tissue or blood (P < 0.001 and P < 0.0001, respectively; Fig. 3o–q). Differential class-switch usage was also seen between NP B_GC cells and B_mem cells (Fig. 3r–t). In sum, immune memory populations were substantially different between blood and two URT sites, with samples from upper airway tissue sites separated by centimetres demonstrating distinct local immune memory in healthy adults.

Fig. 3 |. Nasal cavity and nasopharynx mucosa possess distinct adaptive immune cell populations.

a, Parts of a whole graph demonstrating average percentages of principal immune cell populations in MT versus NP swabs (n = 38 donors). Error bars, s.e.m. b–n, Differences in immune cell frequencies, ratios and subsets in MT versus NP swabs by flow cytometry (n = 88 NP and 55 MT swabs; 38 donors); see Extended Data Fig. 4a for gating strategy: percentage of total CD19⁺ B cells and CD3⁺ T cells (of live CD45⁺) (b,c); ratio of B cells to T cells (b:c) (d); percentage of CD4⁺ (e) and CD8⁺ T cells (f); ratio of CD4⁺ to CD8⁺ T cells (e:f) (g); percentage of B_GC cells (h), B_mem cells (CD20⁺CD38⁻IgD⁻) (i) and B_RM cells (CD20⁺CD69⁺) (k) of total B cells; percentage GC-T_FH cells of total T cells (j); percentage CD4⁺ T_RM cells of non-naive CD4⁺ T cells (l); percentage CD49a⁺ of CD69⁺CD103⁺CD8⁺ T_RM cells (m); percentage of antibody-secreting cells (CD20⁻CD38^hi) of total B cells (n). o–q, Differences in B cell isotype profiles in MT and NP swabs and PBMCs (frequency of IgA⁺ (o), IgG⁺ (p) and IgM⁺ (q)), n = 58 NP and 46 MT swab, 56 PBMC samples; 22 donors. r–t, Differences in isotype profiles of NP B_mem cells versus B_GC cells (IgA⁺ (r), IgG⁺ (s) and IgM⁺ (t)), n = 58 NP swabs; 22 donors. Bars in b–t, median (pairwise comparisons by Mann–Whitney test); *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. u, Volcano plot summarizing significant differences in NP versus MT populations. v, Correlation between number of B_GC cells sampled per NP swab and age across healthy adults; Pearson r and P values (n = 117 donors; one swab per donor). w, Correlation between number of B_GC cells and GC-T_FH cells sampled per NP swab (n = 80 donors; one swab per donor); Pearson r and P values.

With the ability to non-invasively sample adenoid lymphoid tissue from healthy adult donors, it was possible to address the question of how long immunologically functional adenoid tissue lasts during adulthood, as this relates to the capacity for new generation of immune memory. NP swabs collected from more than 100 healthy adult subjects aged 18–68 years were used to determine adenoid GC activity across a large age spectrum. Seventy percent of the adult donors had detectable adenoid B_GC cells. B_GC cells per swab decreased with age, with approximately log-linear progression (Pearson r = −0.54, P = 4.9 × 10⁻¹⁰; Fig. 3v). Naive B cells per swab correlated with B_GC cells (r = 0.71, P = 7.4 × 10⁻³⁷; Extended Data Fig. 5k,n,o), indicating that B_GC cell detection was both an indicator of the presence of mucosal lymphoid tissue and the functionality of B cell responses in that tissue. GC-T_FH cell numbers decreased commensurately with age (r = −0.58, 6.1 × 10⁻¹²; Extended Data Fig. 5l–m). GC-T_FH cell frequencies in NP swabs correlated tightly with B_GC cells (r = 0.96, P = 6.8 × 10⁻⁴⁵; Fig. 3w). B_GC cells were observed in individuals as old as 68 years. Together, the data indicate that, whereas adenoid tissue atrophies with ageing, most healthy adults possess functional adenoid tissue for GC responses.

Adults with and without a history of recent SARS2 infection were enroled in a longitudinal study with monthly sample collections to better understand both functional attributes and durability of human URT B cell memory (Supplementary Table 1). Using fluorescently conjugated SARS2 antigen probes plus multiparametric spectral flow cytometry, upper airway SARS2-specific B_mem cells were detected in NP tissue in 100% of subjects and in MT tissue in 31% of subjects following breakthrough infection (Fig. 4a–c and Extended Data Fig. 6a,b). SARS2-specific B_mem cells in NP samples were preferentially IgA⁺ compared with blood (20% versus 5.9% for nucleocapsid, P = 0.0004; 9.1% versus 3.1% for spike, P < 0.0001; Fig. 4a,b). IgA class switching was even more pronounced in the turbinate epithelium, with around 60% of nucleocapsid-specific B_mem cells and around 30% of spike-specific B_mem cells being IgA⁺ in that tissue (Fig. 4a). Notably, SARS2 nucleocapsid-, receptor binding domain (RBD)-, and spike-specific B_mem cells were maintained in the upper airway (NP) for at least 6 months, and included both IgA and IgG isotypes (Fig. 4c and Extended Data Fig. 6c). Thus, both IgA⁺ and IgG⁺ virus-specific B_mem cells can be durable in the URT for more than 6 months.

Fig. 4 |. SARS2-specific upper airway mucosal B cells.

a,b, Frequencies of SARS2 nucleocapsid (N)-, RBD-, and spike-specific IgA⁺ (a) and IgG⁺ B_mem cells (b) in NP and MT swabs and PBMCs. Bars, median. Pairwise comparisons by Mann–Whitney; *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. c, Longitudinal probe staining of SARS2 nucleocapsid-, RBD- and spike-specific NP B_mem cells by flow cytometry (for gating strategy, see Extended Data Fig. 6a; n = 29–47 donors). Grey lines, data for individual donors. Dotted line, limit of detection. d, Seurat UMAP, 9,462 NP B cells from breakthrough donors 4202, 4809 and 4379 (one NP swab per donor) coloured by B cell subsets. e, Expression of genes of interest for cells in d. f,g, SHM (f) and class-switch recombination (CSR) (g) (isotype) analysis of cells in d. h, SARS2 RBD-, spike- and nucleocapsid-specific B cells overlaid on UMAP from d. i, Phenotypes of SARS2-specific B cells. Two cells that were class switched (IgG1, IgA1) with moderate-to-high SHM were reclassified from naive to class switched resting B_mem cells (Resting B^CSR). j, Isotypes (left) and heavy chain (HC) SHM (right) for early and late B_PC. k, SARS2 RBD-, spike-specific B_PC. l, Phylogenic trees representing Omicron-specific B cell clonal lineages from donor 4809. Tip sizes, number of cells with identical features; colours, cell type; labels, validated antibodies and K_D by SPR; *K_D for BA.4/5 spike. m, Summary of ELISA, PSV neutralization and SPR data for ancestral (D614G), Omicron BA.4/5 spike, RBD for antibodies in l. B, binding of ancestral and Omicron; O, Omicron-specific. Grey, undetectable binding/neutralization. n, Comparison of heavy chain mutations for donor 4809 RBD- and nucleocapsid-specific B_GC. Student’s t-test, ****P ≤ 0.0001.

Paired PBMC samples and NP swabs were collected from three further subjects with recent breakthrough infection for scRNA-seq, CITE-seq and single-cell BCR sequencing (scBCR-seq), combined with oligonucleotide-barcoded, fluorescently conjugated SARS2 spike, RBD and nucleocapsid probes (Methods). Data were obtained for more than 9,400 total NP swab B cells. Twelve distinct B cell subtypes or states were identified among the URT B cells, including B_GC cells, ASCs and B_RM cells (Fig. 4d,e). Several URT B_mem cell clusters were delineated from each other by differences in class-switch recombination and SHM, following a continuum from low to high SHM and IgG class-switch recombination (Fig. 4f,g). Putative URT B_RM cells had the highest proportion of IgA isotypes, with IgA1⁺ cells comprising nearly 50% of B_RM cells (Fig. 4g).

Using an experimental and computational pipeline combining (1) SARS2 antigen probe B cell binding data (n = 102 cells), (2) personalized SARS2-specific BCR reference datasets from matched peripheral blood samples for each subject (n = 7 cells), and (3) BCR amino acid sequences identified computationally by matches in public SARS2 antibody datasets (n = 37 cells, with specificity calls revised on the basis of experimentally validated recombinant antibodies) (Extended Data Fig. 6e,f; Methods), we identified 146 SARS2-specific B cells in NP samples (Fig. 4h). RBD, spike non-RBD, and nucleocapsid protein specificities were all identified. Notably, SARS2-specific B cells identified in upper airway swabs included B_mem cell, B_GC cell, B_RM cell and ASC populations (Fig. 4i), demonstrating rich complexity of virus-specific B cells in the adult upper airway.

There is recent evidence in animal models of the importance of B_RM cells in protective immunity at barrier tissues^31–34, and thus analysis of antigen-specific putative human URT B_RM cells was of particular interest, given the limited knowledge of this immune memory cell type. URT B_RM cell SHM equalled the highest SHM amounts of any B_mem cell cluster (Fig. 4f). Among SARS2-specific B_mem cells, both putative B_RM cells and conventional B_mem cells were observed (Fig. 4i). The putative URT B_RM cell cluster had a distinct gene signature, including upregulation of FCRL4, FCRL5, CCR6, CCR1, ZEB2 and SOX5 (Fig. 4e), that is, genes previously reported in tissue-associated B_mem cells in palatine tonsils and atypical B_mem (aB_mem) cells in other tissues^{19,31,35–37}. Expression of TBX21 (T-bet)—a transcription factor seen in many circulating aB_mem cells^38,39—was absent in URT B_RM cells (Extended Data Fig. 6d). In sum, URT B_RM cells have distinctive features and both SARS2-specific IgA⁺ and IgA⁻ B_RM cells can be found in the URT of previously infected individuals.

ASCs comprised a distinct B cell cluster with high expression of XBP1, IRF4, PRDM1, CD27 and SDC1 (CD138) (Fig. 4e). To investigate whether this ASC cluster was composed entirely of short-lived plasmablasts, which would not constitute memory, we analysed the NP ASC cluster by comparison with published transcriptomic data of human bone marrow plasmablasts and plasma cells (B_PC)⁴⁰ (Fig. 4j and Extended Data Fig. 6g). There was minimal evidence of URT plasmablasts, on the basis of expression of MKI67 or other genes (Extended Data Fig. 6g). Instead, the ASCs exhibited gene expression profiles consistent with human bone marrow early and late B_PC (Extended Data Fig. 6g). Compared with early B_PC, the late B_PC cluster had increased class switching and SHM (Fig. 4j). SARS2-specific cells were found as both early and late B_PC (Fig. 4k and Extended Data Fig. 6h). In sum, mature SARS2-specific B_PC were found locally in the URT, with transcriptional profiles of long-lived B_PC.

We identified two extensive spike-specific B cell clonal lineages that included several B_PC that were highly related to B_GC cell clones (Fig. 4l), indicating both active local maturation of SARS2 B_mem cells and a potential reservoir for local long-lived antibody production in the human nasopharynx. These clonal families consisted primarily of B_GC cells and B_PC, with further differentiation into B_RM cells and other B_mem cell subtypes (Fig. 4l and Extended Data Fig. 7a). These spike-specific B cell clonal family members were class switched ubiquitously to IgA1 (Extended Data Fig. 7b). Most cells had heavy chain SHM counts ranging from 15 to 30 mutations (Extended Data Fig. 7b). To further investigate these clonal families from donor 4809, we synthesized antibodies distributed across each phylogeny. Many antibodies did not bind ancestral proteins or neutralize ancestral virus appreciably but bound BA.4/5 proteins and neutralized BA.4/5 virus strongly, indicating an Omicron-primed B cell response (Fig. 4m and Extended Data Fig. 7c). Antibody affinities of clonal lineage 3 to BA.4/5 RBD ranged from 0.11 to 36 nM, with 3.3- to 8.5-fold higher affinity for local B_PC (O_01) and B_mem cell (O_08) clones compared with the nearest neighbour to the inferred germline (O_10) (Fig. 4l,m), indicating local affinity maturation. Antibodies based on inferred germline BCRs for both clonal lineage 2 and 3 (GL2_V01 and GL3_V03) bound weakly to BA.4/5 RBD or spike, providing further evidence of local affinity maturation (Fig. 4l,m). This subject had received a bivalent COVID vaccine (containing both ancestral and BA.4/5 spike) before Omicron breakthrough infection and sample collection (Extended Data Fig. 7d). Analysis of B_GC cell SHM across SARS2 antigen-specificities in donor 4809 revealed a striking dichotomy in the number of heavy chain mutations between nucleocapsid-specific and RBD-specific B_GC cells (median 7.5 versus 20, P < 0.0001; Fig. 4n). The significantly lower number of heavy chain mutations for nucleocapsid-specific B_GC cells provides evidence for local priming of B cells in response to SARS2 breakthrough infection in the adult URT. Together, a parsimonious explanation of the RBD-specific B cell data is priming and differentiation of Omicron RBD-specific B_mem cells after bivalent vaccine administration, followed by recall of B_mem cells to the URT upon breakthrough infection to initiate secondary GCs in the adenoid, whereas the nucleocapsid-specific B cell responses were primed locally in the URT. In summary, complex local B cell responses can occur in the URT in response to viral infection, including local priming, recall, class-switch recombination and affinity maturation, and develop complex populations of local immune memory, including B_PC and B_RM cells.

Discussion

The URT is the primary site of infection for many human pathogens, but there has been limited knowledge available on URT immune memory cellular content, durability, differentiation states or anatomical complexity in adult humans. Here, using more than 315 total swabs from over 100 donors, comprised of several cohorts of adults in a normal physiologic state or following recent acute viral upper airway infections, with sampling of NP posterior wall and nasal turbinate epithelium sites, combining diverse experimental techniques, we provide new insights into human upper airway immune memory and provide rich evidence of complexities of normal upper airway immunology. CD8⁺ T_RM cells, CD4⁺ T_RM cells and putative B_RM cell populations were defined, and total proportions of these memory populations were stable over time for more than 1 year in healthy adults. Active adenoids with GCs were demonstrated in most healthy adults, with age-dependent declines in adenoid immunobiology observed over a 50-year span. Upper airway B_mem cells and B_RM cells were shown to be preferentially IgA⁺, with highest amounts of SHM in B_RM cells. Evidence of B_PC longevity was demonstrated, which may represent a reservoir of long-lived mucosal immunity in the human URT. Complex populations of SARS2-specific T cells were demonstrated⁴¹, with numerous differences in T cell and B cell memory between adjacent lymphoid and mucosal epithelial tissue sites and many differences between the upper airway and blood. Shared CD4⁺ and CD8⁺ T cell clones were shown between the upper airway and blood, also demonstrating interactions between these compartments. Antigen-specific URT CD8⁺ T_RM cells, CD4⁺ T_RM cells and B_RM cells were all demonstrated. SARS2-specific B_mem cells were demonstrated in both NP and MT sites, enriched for IgA⁺ cells, and durability of antigen-specific IgA⁺ and IgG⁺ B_mem cells for more than 6 months was shown. Local SARS2-specific GC responses were also shown. Finally, the evidence indicated local B cell priming, class switching, affinity maturation and differentiation into several immune cell types including B_mem cells, B_RM cells and B_PC with a transcriptional signature of longevity.

In sum, these findings highlight the presence and maintenance of diverse antigen-specific B cell and T cell subtypes that contribute to human upper airway immune memory and may contribute to protection against respiratory pathogens. Finally, these experimental approaches are applicable to diverse studies of human upper airway health and disease.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-024-07748-8.

Methods

Study populations

Healthy adults aged 18 years or older were eligible to participate in the study. Sample sizes were not predetermined by statistical methods. Randomization and blinding were not performed. The La Jolla Institute for Immunology (LJI) institutional review board provided oversight for all human subject research performed in this study, including all sample and clinical data collection. Further institutional review board oversight for tonsillar and adenoid tissue collection was provided by the University of California San Diego and Rady Children’s Hospital. All individuals provided written informed consent and assent for participation. Consent was provided to display and publish de-identified photographic images and videos. Please see Supplementary Table 1 for further details about the participants in the study cohorts.

Sample collection and processing

Swab (NP posterior wall/adenoid and MT) and blood samples were obtained using standardized collection protocols and processed the same day as collection. Participants were asked to sit in a chair designed for phlebotomy, to blow their nose and to close their eyes before swab sample collection to help with relaxation and ease of sampling. A registered nurse, doctor or other qualified and trained provider performed NP swab sampling. A flocked swab (Puritan) was inserted gently into one nostril, along the floor of the nasal cavity (parallel to the hard palate), until the posterior NP was reached (around 8–9 cm maximum insertion distance), followed by approximately five circular rotations of the swab while in the posterior NP. MT swabs were collected in a similar fashion, but by inserting the swab approximately half the depth as for an NP swab (around 3–4 cm maximum insertion distance) into the nasal cavity, then rotating the swab in a circular motion to sample the inferior turbinate mucosa. We performed swab sampling bilaterally for both NP and MT swabs, using the same flocked swab to sample both sides before placing the swab in a collection tube containing RPMI medium (Gibco) supplemented with 10% fetal bovine serum. One NP and/or MT swab was collected per donor per session. If NP and MT swab samples were paired, MT samples were collected first. In the case of endoscopic swab sampling, a licensed and board-certified rhinologist performed the endoscopic sample collection using standard technique to perform an endoscopic examination of the nasal cavity and NP with a rigid nasal endoscope (Stryker), followed by endoscopic swab sample collection and photo/video documentation of the sampling site. After collection, swabs were vortexed briefly in the capped collection tube to release cells from the swab. More medium was then used to rinse adherent cells from the swab by placing the swab in a 40 μm mesh filter over a 50 ml conical tube and applying medium on top of the swab. The filtered cell suspension was then centrifuged at 500g for 7 min at 4 °C. Swab cells were then processed for downstream applications.

PBMCs and plasma were isolated from blood using standard protocols, and cryopreserved until ready for use^42,43. Surgically resected, deidentified human adenoid tissue was collected using standard protocols, and processed the same day as collection to yield single-cell suspensions, as previously described⁴⁴. These adenoid cells were cryopreserved and stored in liquid nitrogen and thawed in the same fashion as cryopreserved PBMCs when ready for use.

Immunophenotyping of URT cell populations by flow cytometry

For flow cytometry, samples were sequentially stained with Fc block (BioLegend), viability dye, SARS2 probes (45–60 min at 4 °C) and/or antibody (30–60 min at 4 °C) before data acquisition with a 5-laser Cytek Aurora spectral analyser. For details about flow cytometry reagents/Ab panels, please see the Supplementary Information.

Detection of antigen-specific B cell subsets by flow cytometry/fluorescence-activated cell sorting

Biotinylated SARS-CoV-2 proteins were multimerized individually using fluorescently labelled streptavidin at 4 °C for 1 h. Full-length SARS-CoV-2 spike (AcroBiosystems, catalogue no. SPN-C82E9–200ug) and RBD (AcroBiosystems, catalogue no. SPD-C82E9–200ug) were used. Biotinylated spike was mixed with streptavidin BUV737 (BD, catalogue no. 612775) and streptavidin Alexa Fluor 647 (ThermoFisher Scientific, catalogue no. S21374) at a 20:1 ratio (around 4:1 molar ratio). Biotinylated RBD was mixed with streptavidin BV605 (BioLegend, catalogue no. 405229) and BV711 (BioLegend, catalogue no. 405241) at a 2.2:1 ratio (around 4:1 molar ratio). Biotinylated SARS-CoV-2 full-length nucleocapsid (Sino Biological, catalogue no. 40588-V27B-B) was multimerized using streptavidin BV421 (BioLegend, catalogue no. 405226) and streptavidin BB515 (BD, catalogue no. 564453) at a 5.5:1 ratio (around 6:1 molar ratio)⁴². Individual tubes of the antigen probes were prepared as described and were then mixed in Brilliant Buffer (BD Bioscience, catalogue no. 566349) containing 5 μM free d-biotin (Avidity, catalogue no. Bir500A). Samples were prestained using LIVE/DEAD Fixable Blue Stain Kit (ThermoFisher Scientific, catalogue no. L34962) and human Fc block (BioLegend catalogue no. 422302) in DPBS buffer at room temperature for 15 min. PBMCs and MT swabs were stained with 50 μl of antigen probe cocktail containing 100 ng of spike per probe (total 200 ng), 27.5 ng of RBD, 40 ng of nucleocapsid per probe (total 80 ng) and 5 μM free d-biotin at 4 °C for 1 h to ensure maximal staining quality before surface staining with antibodies as listed in Supplementary Table 2 in Brilliant Buffer at 4 °C for 30 min. NP swabs were stained with 100 μl of the same probe master mix and surface antibodies. PBMCs were counted and stained in an appropriate volume of probe master mix and surface antibodies for fluorescence-activated cell sorting (FACS) in preparation for scRNA-seq. For PBMC sorting by FACS for scRNA-seq, one of the streptavidins for each biotinylated target (RBD, spike, nucleocapsid) was replaced with a custom streptavidin dual conjugated with a fluorophore (typically BV421; BioLegend) and a unique oligonucleotide hashtag (BioLegend TotalSeq) such that RBD, spike and nucleocapsid were each assigned one unique fluorophore and one unique oligonucleotide hashtag and cells could be sorted as shown in Extended Data Fig. 5a. For flow cytometric analysis, stained samples were acquired on a 5-laser Cytek Aurora (with SpectroFlo software v.2 or higher) and analysed using FlowJo10 (BD Bioscience). For FACS, PBMCs were stained and sorted as described below.

The limit of detection was set at a minimum of three probe double-positive cells for calling a sample positive for SARS2 nucleocapsid-, RBD- or spike-specific B_mem cells and five probe double-positive cells for determining the isotype frequencies of SARS2 nucleocapsid-, RBD- or spike-specific B_mem cells. For Fig. 4a,b and Extended Data Fig. 5b, only data points above the limit of detection are shown. For any given donor’s samples to be included in longitudinal analyses (Fig. 4c and Extended Data Fig. 5c), they had to have detectable SARS2-specific B_mem cells for at least 2 consecutive months. Please see Supplementary Table 1 for further details of sample sizes.

AIM assay

PBMCs were isolated from peripheral blood samples drawn on the same day as swab collection. As previously described^42,43,45, PBMCs were plated in 96-well plates with 1 × 10⁶ PBMCs per well, and cultured for 24 h at 37 °C in an incubator with 5% CO₂ in the presence of negative control (equimolar amount of DMSO vehicle), positive control (Staphylococcal enterotoxin B at 1 μg ml⁻¹) or SARS-CoV-2 peptide megapools (1 μg ml⁻¹ per megapool)^43,46,47 containing spike or non-spike epitopes (CD4-RE or CD8-RE dominant and subdominant megapools). Before stimulation, PBMCs were incubated at 37 °C for 15 min with 0.5 μg ml⁻¹ anti-human CD40 monoclonal blocking antibody (Miltenyi Biotec). After a 24-h incubation, the plates were centrifuged, cells were washed with PBS and then stained with LIVE/DEAD Fixable Blue (Invitrogen) 1:1,000 in PBS with Fc block (5 μl per sample; BioLegend) for 10–15 min at room temperature, washed with FACS buffer (3% FBS in DPBS without calcium or magnesium), then surface stained for 30 min at 4 °C (see Supplementary Information for details of antibody staining panels). Cells were then sorted according to surface marker expression. AIM⁺ gates were drawn on the basis of T cell responses to megapools relative to controls. AIM⁺ memory T cells that were reactive to SARS2 spike and/or non-spike megapools were isolated by FACS and subjected to scRNA-seq/scTCR-seq to generate SARS2-specific TCR repertoire reference libraries. Samples were labelled with TotalSeq-C anti-human hashtag antibody (BioLegend) for each donor and megapool stimulation condition. Samples from each donor and stimulation condition were sorted separately, with CD4⁺ and CD8⁺ T cells sorted into separate tubes. Cells were then washed, centrifuged, pooled (if appropriate) and subjected to downstream 10x processing (10x Genomics). CD4⁺ T cells were never pooled with CD8⁺ T cells.

Cell sorting

Cells were surface stained, as described in other sections, then subjected to FACS isolation using a 4-laser BD FACSAria Fusion (with FACSDiva software v.6 or higher) or 5-laser Cytek Aurora CS cell sorter (with SpectroFlo software v.2 or higher) in the LJI Flow Cytometry Core. For FACS antibody panels and other details, see the respective Methods sections and Supplementary Table 2.

Sample preparation for scRNA-seq

For scRNA-seq experiments using swabs, cell suspensions were subjected to dead cell removal (Miltenyi Dead Cell Removal Kit) before sequential staining with Fc block (10 min at room temperature; BioLegend), and then SARS2 probes and CITE-seq antibody. For CITE-seq, swab-derived cells were stained with the TotalSeq-C Human TBNK Cocktail (BioLegend) according to the manufacturer’s protocol. Most dead epithelial cells, as well as a portion of viable epithelial cells from swabs, were removed by the dead cell removal method, which also significantly enriched overall cell viability of swab samples. PBMCs were stained sequentially with Fc block and viability dye, SARS2 probes and then surface antibody for FACS. SARS2 probes for scRNA-seq were generated as described above for flow cytometry but using a custom streptavidin containing both a fluorophore and unique oligonucleotide hashtag (BioLegend TotalSeq-C) to allow for both FACS sorting of PBMC-derived probe-binding B cells using a combination of fluorophore-conjugated probes and bioinformatic identification of probe-binding B cells according to the unique oligonucleotide ‘hashtag’ present on one of the probes for each antigen. Swab cells were stained using only the hashtag-containing probe as these samples were not sorted and did not require dual probe staining for FACS. Cell suspensions from sorted PBMCs or processed swabs were prepared for Chromium Single Cell Immune Profiling using Chromium Next GEM Single Cell 5′ HT Reagent Kits (v.2), recommended supplies and equipment, according to the manufacturer’s protocol (10x Genomics). Samples were prepared for sequencing and sequenced on a NovaSeq 6000 sequencer (Illumina) in the LJI NGS Core. Please see Supplementary Information for further details.

Multimodal scRNA-seq analysis

Data were prepared for analysis in Cell Ranger (v.7.0.0 or newer; 10x Genomics) to process and align FASTQ reads from the gene expression (GEX), ADT/HTO and V(D)J sequencing libraries to the provided prebuilt GRCh38 human genome references. Analysis was then performed in R studio using R (v.4.0.0 or newer) and Seurat (v.4 for most figures; v.5 for Extended Data Fig. 5d)⁴⁸. The data were filtered using the default parameters min.cells = 3, percent.mt <5, min.features = 200. The data were filtered to exclude cells with more than 4,500 unique feature counts. For Fig. 4, percent.mt was set to less than 10. Before analysis, gene expression was normalized using the ‘NormalizeData’ function in Seurat (default parameters). CITE-seq data were normalized in Seurat using centred log ratio transformation across features (margin 1). The 10,000 most highly variable features were selected and used for data scaling and dimension reduction through principal component analysis (PCA). Individual samples were then integrated using the reciprocal PCA method with default parameters except as follows: the top 50 dimensions were used to specify the neighbour search space; the top 10,000 variable features defined above were used to find anchors; two representative datasets (samples) were used as references. Following integration, data scaling and PCA dimension reduction were again performed on the complete, integrated dataset. Cells were clustered using the ‘FindNeighbours’ and ‘FindClusters’ functions. Further dimension reduction and visualization were performed with uniform manifold approximation and projection (UMAP). The ‘subset’ function was used to examine B cells (including all B cells and then B cells in the B_GC cell and ASC clusters) and T cells (including all T cells, and then T cells in the T_FH cell and T_reg cell clusters) more closely independently. The resolution of each UMAP was adjusted to achieve optimal separation for cell cluster identification using a combination of gene expression and CITE-seq data. The resolution was set to 1.2 for the UMAP in Fig. 1e, 0.6–0.8 in Fig. 1f, 0.4 in Extended Data Fig. 2a, 0.4 in Extended Data Fig. 2e,f, 0.8 in Fig. 4d and 0.4 in Fig. 4k. We analysed differentially expressed genes (DEGs) with the ‘FindAllMarkers’ function in Seurat using a Wilcoxon Rank Sum test with a log₂ fold-change threshold of 0.25, and including only genes expressed in at least 25% of cells. We performed GSEA using the Fast GSEA package v.1.24 with the fgseaMultilevel function and using default arguments⁴⁹. MSigDB human gene sets were imported using the msigdbr package v.7.5.1. Several published, curated T cell^21–23 and B cell^{37,40,50–52} gene sets were used for GSEA analysis (please see Supplementary Information for details). For Fig. 4, when curated gene sets contained more than 100 DEGs, only the top 100 upregulated and downregulated genes according to the adjusted P value and average log₂ fold-change were included for GSEA. Cells mapping to naive B cell clusters that were class switched to IgA or IgG isotypes were reassigned as ‘Resting B^CSR’. Similarly, a single cell in clone 4809_2 (Fig. 4l) mapped to a naive cell cluster and was reassigned as ‘Resting B^CSR’ as it had 20 heavy chain mutations, no isotype call and the remaining 58 out of 59 cells in the clone all had an IgA1 isotype. Please see Supplementary Information for more details.

Single-cell TCR sequencing analysis

Cell Ranger V(D)J sequencing outputs were further processed and analysed using the Platypus R package⁵³. The TCR data were processed initially using the ‘VDJ_GEX_matrix’ function with default parameters, and GEX data were processed separately in Seurat, as described above. We performed clonotyping using the ‘VDJ_clonotype’ function, and cells with 100% nucleotide sequence identity for paired CDR3α and CDR3β sequences were grouped as clones. MAIT were identified by a combination of GEX and scTCR-seq, and included 52 canonical (TRAV1–2, TRAJ33 expressing) MAIT cells.

Assignment of SARS2-specific T cells

A combination of computational and experimental methods was used to identify SARS2-specific T cells. We examined T cells in NP swabs by searching public databases^27,28 and peer-reviewed publications^29,30 for TCR sequences with 100% amino acid sequence identity to our paired TCRα and TCRβ scTCR-seq data (public reference). Putative SARS2-specific T cells were also identified by TCRs with CDR3α sequences matching an immunodominant CDR3α motif (CA[G/A/V] XNYGGSQGNLIF) that recognizes the HLA-DPB1*04 restricted spike (S^167–180) epitope (spike public TCRA²⁶). Separately, we defined SARS2 spike-specific T cells experimentally by matching TCRα and TCRβ CDR3 sequences with 100% amino acid sequence identity between NP swabs and PBMC-derived, sorted AIM⁺ T cells. Approximately 99% concordance between amino acid and nucleotide sequence identity was observed, demonstrating that nearly all the overlap identified between these sites represents true clonal relationships between NP and PBMC-derived TCR sequences rather than false assignment of TCR sequence overlap on the basis of convergent amino acid sequences.

Antibody-secreting cell subtype analysis

The ‘subset’ function was used to examine more closely the phenotypes of ASCs found in the B cell clusters. The resolution of the UMAP was adjusted to achieve optimal separation for cell cluster identification using GEX sequencing data. Published transcriptomic data from human ASC maturation states in bone marrow were used to define the upper airway ASCs into B_PC subclusters⁴⁰. DEGs identified by Duan et al.⁴⁰ in the bone marrow ‘plasmablast’ and ‘early’ plasma cell clusters 1–2 were treated as a single gene set defined as ‘early’ in our analysis, whereas DEGs identified in ‘late’ cell clusters 1–4 were treated as a single gene set defined as ‘late’ in our analysis. DEGs from other cell clusters in ref. 40 were treated as discrete gene sets and retained cluster nomenclature from the original study. GSEA was used to compare our ASC data with the published DEGs, as described above.

Single-cell BCR sequencing analysis

Cell Ranger V(D)J sequencing outputs were further processed and analysed using R and Python packages from the Immcantation framework^54–57. The National Center for Biotechnology Information IgBLAST v.1.21.0 (ref. 58) and the ‘AssignGenes.py’ function from Change-O v.1.3.0 were used to reannotate V(D)J genes by alignment to the ImMuno GeneTics reference database⁵⁹. Cells with non-functional sequences, multiple heavy chain sequences or with unpaired heavy or light chain sequences were removed. For cells containing heavy chain paired with both κ and λ light chains, the κ chains were removed as λ chain rearrangement occurs only when rearrangement of both κ chain alleles is ineffective. We performed V(D)J clonotyping using scoper v.1.2.1 (refs. 56,57). For each donor, nucleotide sequences were first grouped by shared HC V genes, J genes and CDR3 sequence lengths. A normalized Hamming distance threshold for the CDR3 sequence repertoire was determined using the ‘findThreshold’ function in SHazaM v.1.1.2 (ref. 54) with default parameters. Nucleotide sequences differing from one another by this threshold were defined as distinct clones. Clones differing by light chain V gene, J gene or CDR3 sequence length were further subdivided to generate the final clone assignments. For each clone, consensus clonal germline V and J sequences were reconstructed using the ‘createGermlines’ function in dowser v.1.2.0 (refs. 60–62). The ‘observedMutations’ function in SHazaM was used to calculate SHM as the total number and frequency of non-ambiguous nucleotide mismatches from each cell to the V and J genes of its consensus clonal germline sequence. The SHazaM SHM outputs were plotted in R using geom_boxplot in the ggplot2 package (https://ggplot2.tidyverse.org/reference/geom_boxplot.html#summary-statistics).

B cell lineage tree construction

We constructed maximum-likelihood lineage trees for B cell clonal families using the IgPhyML model in the ‘GetTrees’ function of the R Package dowser (v.2.0)^60–62. Branch lengths are indicative of the estimated cumulative number of mutations that have occurred in each heavy chain sequence since its divergence from the most recent common ancestor within the lineage. Tip sizes are indicative of the number of cells with identical features (heavy chain nucleotide sequence, isotype, cell type); tip colours are indicative of the cell type or isotype, as in the corresponding figures and figure legends.

Single-cell probe binding analysis of SARS2-specific B cells

Unnormalized probe unique molecular identifier (UMI) counts were extracted from the Seurat ADT assay for downstream analysis in R using proprietary scripts. UMI count thresholds for NP swab B cells and sorted B_mem cells were set for each probe to achieve an equivalent percentage of probe-positive cells as observed in the flow cytometry data from an equivalent timepoint from the same donors. In a minority of cells, UMI counts were above the probe-positive threshold for more than one antigen probe. In these cases, final antigen specificity was assigned according to the following criteria: (1) RBD-specific cells have RBD probe UMI counts more than the RBD probe-positive threshold and greater than twice the nucleocapsid probe UMI counts for the same cell; (2) spike (non-RBD)-specific cells have spike probe UMI counts greater than the spike probe-positive threshold, RBD probe UMI counts less than the RBD probe-positive threshold, and spike probe UMI counts more than twice the nucleocapsid probe UMI counts for the same cell; (3) nucleocapsid-specific cells have nucleocapsid probe UMI counts greater than nucleocapsid probe-positive threshold and more than twice the spike and RBD probe UMI counts for the same cell; (4) if none of the above criteria were met, the antigen specificity was assigned a ‘not applicable’ value. Antigen specificity was determined experimentally through a combination of direct probe binding to NP swab B cells and personalized references prepared as follows: (1) matched PBMCs from the same donors were stained with SARS2 antigen probes (RBD, spike, nucleocapsid) and sorted to isolate SARS2-specific B_mem cells before being subjected to scRNA-seq and analysis to generate BCR references for each donor; (2) NP swab B cells with BCRs that matched PBMC references on the basis of 70% or greater sequence identity and exact matches for heavy chain and light chain V and J genes as well as heavy chain CDR3 sequence length were considered antigen specific.

Computational identification of SARS2-specific BCRs

To identify putative SARS2-specific B cells from NP swabs, we developed a computational pipeline to match our paired heavy chain and light chain BCR sequencing data to publicly available SARS2-specific BCR and antibody sequences. We first looked for BCRs with exact matches to heavy chain V and J genes, CDRH3 amino acid sequence length and at least 70% CDRH3 amino acid sequence identity to human, SARS2-specific, reference sequences in the Coronavirus Antibody Database (CoV-AbDab)⁶³. BCRs matching these criteria were further filtered by (1) exact matches to paired LC V and J genes in CoV-AbDab⁶³, (2) heavy chain and light chain V gene pairs enriched in RBD-specific antibodies^64,65, or (3) public heavy chain V genes and CDR3 sequence lengths enriched in RBD-specific antibodies^64–66. We synthesized 20 antibodies corresponding to predicted RBD-specific BCRs and used these antibodies to experimentally validate RBD binding by enzyme-linked immunosorbent assays (ELISA; details below). Of the 20 antibodies, 7 were confirmed as RBD binders, and the computational pipeline was refined as follows: (1) method 3 above was removed from the pipeline as it had poor predictive value for determining SARS2-specificity and (2) the threshold for percent CDRH3 sequence identity to CoV-AbDab reference sequences was increased to 80%. Together, these changes resulted in a revised computational pipeline with a sensitivity of 35%, specificity of 87.5%, positive predictive value of 77.8% and negative predictive value of 51.9% (Supplementary Table 3). Predicted and validated RBD-specific BCRs were included in the final SARS2-specificity annotations, including three cells from the breakthrough infection donors shown in Fig. 4.

Validation of expanded SARS2-specific clones

Clone 4809_2 comprised 59 cells, including 1 cell that was validated experimentally as an RBD probe binder and 1 cell that was identified computationally as RBD-specific using the above computational methods. Similarly, clone 4809_3 comprised 36 cells, including 1 cell that was identified computationally as RBD-specific. As 56 out of 59 cells in clone 4809_2 and 35 out of 36 cells in clone 4809_3 were B_GC cells and B_PC, which have decreased BCR surface expression, we considered that the antigen probe assay may not have been sensitive enough to detect most RBD-specific cells within these clones. We thus synthesized recombinant human IgG monoclonal antibody representing BCRs distributed across distinct clades within these clonal lineages and validated them by ELISA, pseudovirus (PSV) neutralization and surface plasmon resonance (SPR) affinity measurements, as described in detail below; 100% of the antibodies tested from these clonal lineages (n = 6 for 4809_2 and n = 6 for 4809_3), including antibodies based on inferred clonal germline sequences, were validated as Omicron RBD-specific by all three methods. As such, all BCRs/B cells belonging to these clones were assigned as RBD-specific.

Antibody validation of SARS2-specific B cells

An initial batch of 20 bioinformatically identified putative SARS2-specific BCRs from the SARS2 breakthrough infection donors whose sequencing data are shown in Figs. 1 and 4 (and related Extended Data figures) were cloned as recombinant human IgG monoclonal antibodies, expressed in CHO cells (GenScript) and purified for validation. These 20 monoclonal antibodies were confirmed to be human IgG by ELISA. The antibodies were then tested for binding and neutralization of ancestral (D614G) and Omicron BA.4/5 SARS2 by ELISA and PSV neutralization assays, respectively. In total, 8 out of 20 antibodies recognized variants of SARS2 RBD as determined by ELISA: 6 bound strongly to both ancestral and BA.4/5 RBD and 2 bound strongly only to BA.4/5 RBD (the index antibodies from clones 4809_2 and 4809_3). Antigen-specificity was confirmed by performing ELISAs and PSV neutralization assays. The validated B cells that corresponded to the three breakthrough infection donors shown in Fig. 4 (three of the eight validated RBD binding antibodies) were added to Fig. 4h–i. A second batch of 25 antibodies, including 4 more antibodies from clone 4809_2 and 5 more antibodies from clone 4809_3 (11 in total, including the index antibodies), was synthesized for further validation. These also included eight antibodies (four per clone) based on inferred clonal germline sequences from clones 4809_2 and 4809_3, which were determined as follows: (1) for each clone, the nucleotide sequences for the germline junction regions predicted by both dowser and IgPhyML R packages were aligned with the junction regions of seven distinct sequenced BCRs with the fewest computed mutations; (2) ambiguous nucleotides were assigned by inferring the most likely nucleotide on the basis of the sequence alignments; (3) the final inferred germline nucleotide sequence for the entire heavy chain was compared with the nearest neighbour in the lineage tree using the National Center for Biotechnology Information BLASTn function to verify that the computed mutation number matched the number of nucleotide differences between the two sequences. Owing to the inherent ambiguity in predicting germline antibody sequences, four inferred germline iterations per clone were synthesized, representing the antibody sequences most likely to match the true germline. Only one iteration per clone showed evidence of binding to BA.4/5 RBD/spike, and these antibodies are indicated as GL2_V01 and GL3_V03 in the main text. Full variable region sequences for all validated SARS2-specific antibodies can be found in Supplementary Table 3. The antibody data are summarized in Fig. 4m.

SARS2 spike and RBD IgG ELISA

ELISAs were performed as described previously⁶⁷. Positive control plasma pooled from convalescent COVID-19 donors was used to standardize each assay and normalize across experiments. Previously published and validated SARS2-specific monoclonal antibody were also used as positive controls⁶⁸, and a previously published HIV envelope broadly neutralizing antibody was used as a negative control⁶⁹. For the assay, 96-well half area plates were coated with 1 μg ml⁻¹ of SARS2 Ancestral (D614G) spike or RBD (Saphire laboratory), Omicron BA.4/5 RBD (AcroBiosystems) or a negative control (EBV Nuclear antigen 1, EBNA1; Abcam) diluted in PBS overnight at 4 °C. Plates were then blocked with 3% milk in PBS containing 0.05% Tween-20 for 1.5 h at room temperature. Monoclonal antibodies were diluted in 1% milk containing 0.05% Tween-20 in PBS and added to the plate at a starting concentration of approximately 16,667 ng ml⁻¹ (1 μg) of monoclonal antibodies, followed by serial threefold dilutions across the plate. Similarly, heat-inactivated, pooled human plasma controls were added to the appropriate wells and serially diluted across the plate. Plates were incubated for 1.5 h at room temperature, then washed. Plates were then incubated with conjugated IgG secondary antibody for 1 h. Plates were then washed and developed with 3,3′,5,5′- tetramethylbenzidine ELISA substrate solution (ThermoFisher). The optical density of the wells was then read using a SpectraMax Plate Reader at 450 nm using the SoftMax Pro software (v.7.1).

SARS2 PSV neutralization

PSV neutralization assays were performed as described previously⁶⁷. In brief, a monolayer of Vero cells (ATCC) was generated by seeding 2.5 × 10⁴ cells in black flat-bottom 96-well plate (Corning) and cells were incubated at 37 °C for 5–6 h. Pretitrated recombinant virus was incubated with serially diluted monoclonal antibody (threefold or sixfold serial dilutions from a starting concentration of 25,000 or 2,000 ng ml⁻¹, respectively, depending on the activity of each specific antibody) for 1 h at 37 °C, added to confluent Vero cell monolayers, and incubated for 16–20 h at 37 °C in 5% CO₂. Previously published and validated SARS2-specific monoclonal antibodies were used as positive controls, and a previously published HIV envelope trimer-binding broadly neutralizing antibody was used as a negative control (as above for ELISAs). Cells were then fixed in 4% paraformaldehyde in PBS with 10 μg ml⁻¹ Hoechst (ThermoFisher) in the dark for 30 min at room temperature. Cells were imaged using a Cell Insight CX5 imager to quantify the total number of cells and infected GFP expressing cells. Neutralization titre (inhibition dose 50; ID₅₀) were calculated from complete dose-response curves using the One-Site Fit Log IC₅₀ model in GraphPad Prism. Recombinant SARS2 spike-pseudotyped VSV-ΔG-GFP were generated as described previously⁷⁰ by transfecting HEK293T cells (ATCC) with phCMV-3-SARS-CoV-2 spike and collecting supernatants containing rVSV-SARS2 spike for virus variants with the specific amino acid mutations, as listed in the GISAID database: ancestral (D614G) or Omicron BA.4/5 (BA.5; T19I, L24S del25/27, del69/70, G142D, V213G, G339D, S371F, S373P, S375F, T376A, D405N, R408S, K417N, N440K, L452R, S477N, T478K, E484A, F486V, Q498R, N501Y, Y505H, D614G, H655Y, H655Y, N679K, P681H, N764K, D796Y, Q954H, N969K). Cells obtained from ATCC were not tested to authenticate cell line identity or for Mycoplasma contamination.

SARS2 spike protein production and purification for affinity testing

Recombinant SARS2 spike proteins were generated in the HexaPro background (containing residues 14–1,208 (GenBank: MN908947) of the ectodomain, six proline substitutions (F817P, A892P, A899P, A942P, K986P, V987P)), as well as the D614G mutation and replacement of cleavage site residues 682–685 (RRAR to GSAS)⁶⁸. Each spike was cloned into a phCMV mammalian expression vector containing an N-terminal Gaussia luciferase signal sequence, a C-terminal HRV-3C cleavage site and a Twin-StrepII-Tag purification. HexaPro constructs also contained a foldon trimerization domain between the C terminus of S2 and the HRV-3C site. Plasmids were transformed into Stellar competent cells and isolated using a Plasmid Plus Midi kit (Qiagen). All clones were fully sequenced to ensure that no further mutations were introduced.

ExpiCHO-S cells (ThermoFisher) were transiently transfected with SARS2 spike constructs. CHO cells (ATCC) were maintained and transfected according to the manufacturer’s protocols. For all ExpiCHO cultures, the manufacturer’s High Titer protocol was used with a 7-day culture incubation to assess relative expression. In brief, plasmid DNA and Expifectamine were mixed in Opti-PRO SFM (Gibco) according to the manufacturer’s instructions and added to the cells. On day 1, cells were fed with manufacturer-supplied feed and enhancer as specified in the manufacturer’s protocol, and cultures were moved to a shaker incubator set to 32 °C, 5% CO₂ and 115 rpm. On day 7, cultures were clarified by centrifugation, followed by addition of BioLock (IBA Life Sciences), passage through a 0.22 μM sterile filter and purification on an ÄKTA Pure system (Cytiva) using a 5 ml Strep-Tactin XT column equilibrated with TBS buffer (25 mM Tris pH 7.6, 200 mM NaCl, 0.02% NaN₃), and eluted in TBS buffer supplemented with 10 mM d-desthiobiotin (Sigma Aldrich). Proteins were then purified by size-exclusion-chromatography on a Superdex 6 Increase 10/300 column (Cytiva) in the same TBS buffer.

High-throughput SPR binding kinetics and data analysis

Binding kinetics measurements were performed on the Carterra LSA platform using a CDMP sensor chip (Carterra). The chip was activated with a freshly prepared solution of 130 mM 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride (catalogue no. PG82079, Pierce) and 33 mM N-hydroxysulfosuccinimide (catalogue no. 24510, ThermoFisher Scientific) in 0.1 M MES pH 5.5 using a single flow cell. A surface capture lawn was prepared with 50 μg ml⁻¹ of goat anti-human IgG Fc secondary antibody (catalogue no. 103255–066, VWR) in 10 mM sodium acetate (pH 4.5)/0.01% Tween. Unreactive esters were quenched with a 7-min injection of 1 M ethanolamine-HCl (pH 8.5). Purified monoclonal antibodies were captured using the 96-print head with 1× HBSTE buffer (10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA and 0.01% Tween-20) as running buffer and antibody diluent. Three independent dilutions, at two concentrations (5 μg ml⁻¹ and 0.5 μg ml⁻¹ for spike analytes and 10 μg ml⁻¹ and 20 μg ml⁻¹ for RBD analytes), were prepared for each monoclonal antibody, with each dilution captured on two independent array spots.

A twofold dilution series of spike or RBD was prepared in 1× HBSTE-BSA buffer (10 mM HEPES pH 7.4,150 mM NaCl, 3 mM EDTA, 0.05% Tween-20, supplemented with 0.5 mg ml⁻¹ bovine serum albumin (BSA)). Protein was injected onto the chip surface (at 25 °C) using the single flow cell, from the lowest to the highest concentration, without regeneration in between. Five buffer injections of the buffer before the lowest non-zero concentration were used for signal stabilization. For each concentration, baseline data were collected for 120 s, association data for 300 s and dissociation data for 900 s. After the titration of each analyte, the huFc capture lawn surface was regenerated with two pulses (17 s per pulse) of 10 mM glycine, pH 2.0 and monoclonal antibodies recaptured in between each analyte. The running buffer for all kinetic steps was 1× HBSTE-BSA. Titration data were processed with the Kinetics software package (Carterra v.1.9.1.4215 or higher), including reference subtraction, buffer subtraction and data smoothing. Analyte binding timecourses for each antibody were fitted to a 1:1 Langmuir model to derive k_a, k_d and K_D values.

Statistics

Statistical analyses were performed in R (v.4.0.0 or newer), GraphPad Prism 10 (GraphPad Software) and FlowJo (v.9 or higher, BD). All flow cytometry and FACS data were analysed using FlowJo (v.9 or higher). For all analyses, a two-sided 5% type I error rate was used unless otherwise noted. Further details can be found in relevant Methods sections, the results, figures and corresponding figure legends.

Reporting summary

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

Extended Data

Extended Data Fig. 1 |. NP swab sampling reveals diverse URT cell populations.

a. Total viable CD45⁺ immune cells collected per NP swab (n = 315 swabs). b. Example flow gating of CD45⁺ immune cell populations upstream of the T_RM and putative B_RM population gates shown in Fig. 1d c. Left: total viable CD45⁻EpCAM⁺ (CD326) epithelial cells collected per NP swab (n = 164 swabs); Right: example CD45 × EpCAM gating with CD45⁻EpCAM⁺ epithelial cell populations in the lower right quadrant. d. Epithelial cell UMAP from scRNA-seq of the approximately 8,000 epithelial cells derived from the same 12 NP swabs/donors as Fig. 1e. e. Dot plot demonstrating expression patterns of epithelial cell cluster defining genes. f. UMAP from Fig. 1e split by donor to show the contribution of NP swabs from each donor. g. Feature plots of gene (top panels) vs surface protein (bottom panels) expression for CD45, CD19, CD3E RNA and their corresponding proteins for the combined immune cell UMAP shown in Fig. 1e. h. Feature plots of gene (top panels) vs surface protein (bottom panels) expression for CD4 and CD8A RNA and their corresponding proteins for the T cell subset UMAP shown in Fig. 1f. Related to Fig. 1.

Extended Data Fig. 2 |. Characterization of adaptive immune populations of the human URT.

a. B cell subset UMAP generated from the approximately 35,000 B cells from Fig. 1e, with subset identities assigned by gene expression. b-c. (b) B cell and (c) T cell profiles (percentage of each lymphocyte subset for each donor) for each of the 12 swab donors whose B and T cell data are shown in Extended Data Fig. 2a and Fig. 1f, respectively. d. GSEA of the T cell subsets in Fig. 1f for the MAIT gene signature: NES = normalized enrichment score; -log(p_adj) = negative log₁₀ Benjamini-Hochberg adjusted p-values (see Methods for details). MAIT were identified by a combination of gene expression and scTCRseq analysis (see Methods for details). e. Feature plot of FAS expression with the CD4⁺ T_Mem cluster from Fig. 1f circled. f-g. The clusters labelled T_FH and T_reg in Fig. 1f were named based on their most prominent cell type. Subset UMAPs generated from each of the (f) ~3,700 cell T_FH and (g) ~1,500 T_reg cluster cells from Fig. 1f showing additional cell types identified by subset analysis. Within the T_FH cluster: mantle T_FH (mT_FH), GC-T_FH, and GC-T_FH with a type 1 IFN signaling signature (GC-T_FH^IFN). Within the T_reg cluster: T_reg, T_H17, T_FH, and T_FR. See Fig. 1j for gene expression profiles. Virus-specific T cells were identified in these clusters and are indicated by colored halos based on the key shown in panels f,g. h. Example gating for T cell sorting for scRNA-seq/scTCR-seq based on AIM marker expression. i. SARS2 spike- and non-spike—specific T_FH counts identified by AIM assay for each donor. j. SARS2-specific T_FH counts (number per breakthrough infection donor swab) plotted by days post-COVID symptom onset. k. SARS2-specific T cell subsets identified in NP swabs split by donor. Related to Fig. 1.

Extended Data Fig. 3 |. Experimental workflow for defining personalized SARS2-specific T cell TCR references to identify upper airway antigen-specific T cells.

a. Workflow diagram for the 12 participants who underwent paired sample collection of peripheral blood (to isolate PBMCs) and NP swabs for CITE-seq plus scRNA-seq/scTCR-seq. SARS2-specific NP T cells were identified based on TCR overlap between NP swabs and blood. b-d. Representative flow plots for stimulated PBMC-derived SARS2-specific (b) CD4 T cells as OX40⁺CD40L⁺, (c) CD4 T cells as OX40⁺41BB⁺, and (d) CD8 T cells as CD69⁺41BB⁺ by AIM assay from a convalescent COVID-19 control donor with robust CD4 and CD8 S and non-S responses. e-g. Frequency of SARS2-specific (e) OX40⁺41BB⁺ CD4 T cells, (f) OX40⁺CD40L⁺ CD4 T cells, and (g) CD69⁺41BB⁺ CD8 T cells by AIM assay. For e-g the dotted line = limit of quantitation. h-j Stimulation index (SI) = fold-change above negative control for SARS2-specific CD4 (h) OX40⁺41BB⁺ and (i) OX40⁺CD40L⁺ and CD8 (j) CD69⁺41BB⁺ T cell responses by AIM assay. For h-j the dotted line = SI of 2. For e-j, data is shown from the 12 donors in panel a; positive responses are greater than the dotted line and negative responses are set to baseline (0.5 × LOQ or SI = 1); bar = geomean; dots are color-coded based on COVID-19 status as per the legend. Related to Fig. 1. Panel a was created using BioRender (https://Biorender.com).

Extended Data Fig. 4 |. Adult adenoids are URT secondary lymphoid organs with functional GCs.

a. Example gating for NP swab (adenoid) B_GC and GC-T_FH. ASC = antibody secreting cells. b. Top panels: flow plots showing intracellular Ki67 and BCL6 staining for B_GC compared with non-B_GC; Bottom panels: histograms showing intracellular BCL6 expression in GC-T_FH, non-T_FH CD4⁺ T cells, and B_GC. c. UMAP showing the cell cycle phases of the B_GC in Fig. 2b. d. Dot plot showing expression of additional B_GC genes (related to Fig. 2c). e. A B cell clonal lineage from the NP swab of donor 4379, represented as linear phylogenic trees by B cell type (left) and by isotype (right). Related to Fig. 2.

Extended Data Fig. 5 |. Adenoids and nasal turbinate mucosa have distinct immune cell profiles.

a. Example gating for B and T cell subsets and B cell isotypes from NP and MT swabs shown in Fig. 3 and in b-h. b-h. Additional differences between NP (adenoid) and MT swab populations: (b) % naive B of total B cells, (c) % naive CD4⁺ of total T cells, (d) % naive CD8⁺ of total CD8⁺ T cells, (e) %T_reg of CD4⁺ T cells, (f) %CD8⁺ T_RM (CD69⁺CD103⁺) of CD8⁺ T cells (g) % non-naive CD4⁺ of total CD4⁺ T cells, and (h) % non-naive CD8⁺ of total CD8⁺ T cells. For b-h: Pairwise testing by Mann-Whitney; bars = median, two-tailed p-values: *** = p < 0.001, **** = p < 0.0001 (n = 38 donors; 88 NP and 55 MT swabs). i. Example flow plots for putative B_RM (CD69⁺ B cells), CD4⁺ T_RM (CD69⁺CD103^+/−), CD8⁺ T_RM (CD69⁺CD103⁺), B_GC (CD19⁺CD20⁺CD38⁺), and GC-T_FH (CXCR5⁺PD-1^hi) in blood (PBMC). j-o. Additional correlations related to adenoid atrophy with aging: (j) frequency of B_GC (% of total B cells) per NP swab vs age (n = 138 donors; 1 swab/donor), (k) B_GC per NP swab in log₁₀ vs % naive B (of total B cells) (n = 233 swabs), (l) frequency of GC-T_FH (% of CD4⁺ T cells) per NP swab vs age (n = 138 donors; 1 swab/donor), (m) GC-T_FH in log₁₀ per NP swab vs age (n = 119 donors; 1 swab/donor), (n) frequency of naive B (% of total B cells) per NP swab vs age (n = 138 donors; 1 swab/donor), (o) naive B per NP swab in log₁₀ vs age (n = 138 donors; 1 swab/donor); For j, k, n, and o: Spearman r and two-tailed p-values. For l-m: Pearson r and two-tailed p-values. Related to Fig. 3.

Extended Data Fig. 6 |. Multimodal validation of SARS2-specific URT mucosal B cells.

a. Example flow gating related to SARS2 probe staining for SARS2 S-, RBD-, and N-specific B cells. b. Proportion of IgG⁺ SARS2-specific B_mem in NP (n = 58 subjects) and MT swabs (n = 11 subjects) versus PBMC (n = 24 subjects). Bars = median. Pairwise comparisons by Mann-Whitney; two-tailed p-values: ns = not significant, * = p ≤ 0.05, **** = p ≤ 0.0001. Please see Methods for additional details. c. % SARS2 N-, RBD-, and spike (S)-specific IgA⁺ or IgG⁺ B_mem of total B_mem over time post-breakthrough infection by flow cytometry with fluorescent probe staining (n = 32–41 subjects). Dotted line = limit of detection. d. Select differences in gene expression patterns between NP swab putative B_RM (orange) and PBMC-derived atypical B_mem (teal). PBMC-derived atypical B_mem often express TBX21. e. Gating strategy for sorting SARS2 S-, RBD-, and N-specific B_mem in PBMC for scRNA-seq/scBCR-seq. f. SARS2-specific B cells from the 3 breakthrough infection donors in Fig. 4h color-coded by method used for assigning Ag-specificity. g. Heat map of human bone marrow PB and early vs late PC gene expression by NP ASC cells in Fig. 4k based on published gene signatures. h. UMAP related to Fig. 4k showing B_PC belonging to donor 4809 that are from the SARS2-specific clones 2 and 3 shown in Fig. 4l, and B_PC belonging to a non-SARS2-specific clone (4809_10). Related to Fig. 4.

Extended Data Fig. 7 |. Expansion of SARS2 Omicron BA.4/5 variant-specific B_mem in the upper airway.

a. UMAP plots demonstrating the distribution of Omicron-specific clones 2 and 3 from donor 4809 across the B cell clusters from Fig. 4d. b. Two SARS2 Omicron-specific B cell clonal lineages from donor 4809 represented as linear phylogenic trees as in Fig. 4l. Tip sizes represent the number of cells with identical features. Tip colors represent the isotype. Tip labels represent validated antibodies. c. ELISA IgG dilution curves for ancestral and Omicron BA.4/5 RBD for the antibodies generated for SARS2-specific B cell validation (related to Fig. 4m summary data). Mean and standard deviation (error bars) for three independent experiments are shown for controls and the 23 mAb in Fig. 4m. d. Timeline of vaccination, breakthrough infection, and NP swab collection for single-cell experiments for donor 4809. Related to Fig. 4.

Data availability

The authors confirm that the source data underlying the findings are fully available. Gene expression, BCR and TCR sequencing data generated in the current study are available in the Gene Expression Omnibus database (GEO) under the GEO Series record GSE270518. All data obtained from publicly available sources is cited in the text and the references include the URL for the data source.