Temporal profiling of human lymphoid tissues reveals coordinated defense against viral challenge

Main

Secondary lymphoid organs orchestrate the spatial and temporal arrangement of different immune cell subsets to enable appropriate cellular interactions and the generation of timely pathogen-specific responses. T and B lymphocytes are largely segregated into distinct areas within lymphoid tissues, except within germinal centers (GCs), highly organized structures with light and dark zones, where GC B cells acquire antigen from follicular dendritic cells (FDCs) and present it to PD⁺CD4⁺ T follicular helper (T_FH) cells¹. This interaction leads to iterative rounds of GC B cell proliferation, during which there is somatic hypermutation and the emergence of both memory B cells and antibody-secreting plasmablasts or cells that defend against current and future infection^{1,2,3,4,5,6,7}. As lymphoid tissues are critical for the generation of adaptive immunity, they need robust defense mechanisms. These processes are well studied in animal models, for example, in mice, subcapsular sinus macrophages contribute to lymph node defense, preventing the spread of lymph-borne viruses and bacteria, the latter via interactions with innate lymphocytes^8,9. In contrast, profiling of human lymphoid tissue responses during infection is limited, resulting partly from the challenges of sampling these structures in humans.

The nasopharynx contains the pharyngeal tonsils (also known as adenoids), which form part of Waldeyer’s ring of lymphoid tissue and are the major site of nasal-associated lymphoid tissue (NALT). Structurally, they have a surface crypt epithelium with antigen-transporting M cells and underlying B cell follicles and T cell-containing interfollicular areas, which together generate protective antibody-secreting and memory cells after a nasal challenge^10,11. Adenoid tissue typically macroscopically involutes in adolescence and the extent to which NALTs contribute to local mucosal adaptive immune responses in adult humans is unclear. CD68⁺ macrophages have been identified in human adenoid NALT¹², but macrophage heterogeneity and contributions to lymphoid tissue defense or repair post-infection in humans are completely unknown. The posterior nasal space can be readily visualized in humans using an endoscope, raising the possibility that endoscopic adenoid NALT biopsies might offer a feasible method to profile adaptive immune responses in secondary lymphoid organs after an upper respiratory tract immune challenge. To test this, we undertook a ‘natural experiment’, sampling adult participants after severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection—a virus that uses nasal epithelium as a point of entry and replication^13,14—to understand how NALTs defend themselves in the face of local infection to ensure that the cells and infrastructure required to support an ongoing adaptive immune response are maintained.

Although SARS-CoV-2 has already been well studied in humans, much of this work has used peripheral blood or epithelial sampling in live participants, which fails to capture lymphoid follicular components, such as GC B cells^15,16,17. Lymphoid tissue sampling in SARS-COV-2 infection has been conducted in children undergoing adenoidectomy, suggesting a robust GC response in pediatric NALT whereas, in contrast, autopsy samples from the spleen and lymph nodes in adults with fatal SARS-CoV-2 infection show minimal or absent GCs^18,19,20.

In the present study, we applied single-cell RNA sequencing (scRNA-seq), multi-parameter flow cytometry and confocal imaging to adult NALT biopsies, to understand the cellular molecular adaptations that support, polarize and defend human secondary lymphoid tissue in adults in the face of a viral immune challenge.

Results

Postnasal space biopsy enables lymphoid tissue profiling

We developed a well-tolerated endoscopic technique for collecting NALT biopsies from live humans using topical local anesthesia (Extended Data Fig. 1a,b). 23 participants underwent sampling (aged 19–91 years), including 10 healthy controls, 8 patients with acute COVID-19 (sampled within 1 week of a first positive SARS-CoV-2 reverse transcriptase–polymerase chain reaction (RT–PCR) test) and 5 convalescent patients with COVID-19 (clinically asymptomatic participants sampled 3–5 weeks after the first positive SARS-CoV-2 RT–PCR; Fig. 1a and Supplementary Table 1). Postnasal space biopsies (with additional nasal brushing or curettage in some cases) were collected, alongside paired blood samples and processed for flow cytometry, scRNA-seq and confocal microscopy (Supplementary Table 1).

**Fig. 1: Experimental overview and cellular landscape of blood and NALT in SARS-CoV-2 infection.**

After quality control, we generated data on 162,738 cells and identified 20 major cell clusters across peripheral blood mononuclear cells (PBMCs) and nasal tissues, including epithelial cells, stromal cells and a variety of immune cell populations (Fig. 1b and Extended Data Fig. 1c–e). There was a significant overrepresentation of GC B cells and T_FH cells in nasal samples (Fig. 1c,d), reflecting the presence of GC-containing lymphoid structures. GC B cells were particularly abundant in convalescent NALT samples (Fig. 1d), consistent with an expansion of virus-specific B cells. NALT samples also differed from adult nasal cavity and bronchial brushings¹⁷, in which T_FH cells and GC B cells were scarce (Extended Data Fig. 1f–h). Confocal imaging of NALT biopsies confirmed the presence of B cell follicles containing CD21-expressing FDCs and spatially distinct T cell-rich areas, with overlying surface EPCAM⁺ epithelial cells (Fig. 1e). Notably, the number of differentially expressed genes (DEGs) in COVID-19 samples (both acute or active (A) and convalescent (C)), relative to controls, was greater in NALT than in blood (Fig. 1f). Similarly, gene-set enrichment analysis (GSEA) of these DEGs showed a greater magnitude, significance and number of enriched gene sets in NALT compared with peripheral blood, including interferon (IFN)α response pathway genes (Fig. 1g).

Overall, these data show that nasopharyngeal biopsies can be used to profile functional lymphoid tissue in adults, with capture of follicular cell types not found in blood or nasal cavity brushings. NALT sampling presents a feasible method to interrogate the cellular molecular processes occurring during the generation of adaptive immune responses to a nasal viral infection in humans.

MNPs contribute to lymphoid tissue defense and repair

Postnasal space sampling 3–5 d after diagnostic testing for SARS-CoV-2 provided a unique opportunity to understand how early innate immune responses act to defend NALT from local viral attack. Among mononuclear phagocytes (MNPs), we identified 11 cell types: conventional dendritic cells (cDCs) (CD141⁺ cDC1 and CD1C⁺ cDC2), plasmacytoid dendritic cells (pDCs), six monocyte clusters (classical, intermediate, nonclassical, SIGLEC1⁺, CD163^hi and C1Q⁺), basophils and macrophages (Fig. 2a and Extended Data Fig. 2a). The proportion of classical monocytes in both nasal tissue and peripheral blood increased in acute COVID-19 compared with healthy controls and convalescent COVID-19 (Fig. 2b,c), consistent with tissue infiltration (confirmed by flow cytometry; Extended Data Fig. 2b). These monocytes showed increased expression of several self- and neutrophil-recruiting chemokines (CCL2 and CXCL2, CXCL3 and CXCL8, respectively) (Fig. 2d). S100A8/9 transcripts were also increased in classical monocytes in active and convalescent COVID-19 samples (Fig. 2d,e and Extended Data Fig. 2c). S100A8/9 dimerizes to form calprotectin, which sequesters metal ions, limiting their availability to bacteria, including upper respiratory tract-dwelling organisms such as staphylococci²¹. SARS-CoV-2 is known to directly infect respiratory tract epithelium and, indeed, we found SARS-CoV-2 spike protein in epithelial cells overlying NALT in acute COVID-19 (Extended Data Fig. 2d) and in some subepithelial T cells (Extended Data Fig. 2e). CD14⁺ monocytes predominantly localized to the subepithelium in acute COVID-19 (Fig. 2f and Extended Data Fig. 3a–c) and surrounded some B cell follicles in convalescent disease (Fig. 2f), forming a shield around them. In addition, neutrophils were also present in the epithelial and subepithelial areas, some with histone citrullination, required for formation of neutrophil extracellular trap (NET) structures that bind and kill invading bacteria²² (Fig. 2g). Therefore, calprotectin-producing monocytes and NET-losing neutrophils may work together to defend the underlying lymphoid tissues after epithelial cell damage by invading virus.

**Fig. 2: Recruited monocytes and inflammatory DCs form a defensive shield in the subepithelium.**

Macrophages were found in NALT but not in blood and were increased in convalescent COVID-19 samples. The number of DEGs in NALT-resident macrophages was also greater in convalescent COVID-19 compared with acute infection (Fig. 1f), including interleukin (IL)-18 (Fig. 2d), reminiscent of studies in murine lymph nodes showing macrophage production of IL-18 in the context of bacterial challenge⁹. It is interesting that a population of circulating C1Q⁺ monocytes was also expanded in convalescent COVID-19 (Fig. 2c and Extended Data Fig. 2a), with trajectory analysis suggesting that these represent tissue macrophage precursors (Extended Data Fig. 3d).

GSEA²³ showed a greater magnitude of enrichment in nasal cell types compared with their circulating counterparts (Fig. 2h and Extended Data Fig. 3e). Considering the majority myeloid populations in NALT, ‘IFNα response’ and ‘INFγ response’ genes were significantly enriched in macrophages, CD1C⁺ dendritic cells (DCs) and pDCs in acute COVID-19 compared with controls (Fig. 2h and Extended Data Fig. 3e). In convalescent COVID-19, this IFN response gene enrichment became less prominent in cDCs and pDCs but was sustained in macrophages (Fig. 2h). Furthermore, ‘epithelial-to-mesenchymal transition’ (EMT) gene-set enrichment increased in macrophages in convalescent disease and, indeed, was exclusively enriched in macrophages at this time point (Fig. 2h and Extended Data Fig. 3e), consistent with a role in the restoration of tissue homeostasis.

To further explore myeloid cell heterogeneity in NALT, we reclustered nasal MNPs in isolation from those in blood (Fig. 3a). This enabled the further identification of so-called ‘inflammatory’ or ‘monocyte-derived’ DCs (moDCs, based on the expression of CD14, S100A8/9 and FCGR1A^24,25 and two distinct clusters of macrophages based on C1Q and FOLR2 expression; Fig. 3a and Extended Data Fig. 4a). The FOLR2⁺ macrophages expressed additional markers associated with tissue-resident macrophages, including TIMD4 (ref. ²⁶) (Extended Data Fig. 4a). Monocytes and moDCs were increased in acute COVID-19 (Fig. 3b). CD14⁺ moDCs localized to the subepithelial region (Fig. 3c) and expressed ficolin (FCN1) (Fig. 3d), a membrane c-type lectin that mediates CXCL8 upregulation on pathogen encounter²⁷. A small proportion of moDCs also expressed CXCL8 (Fig. 3d) and showed reduced enrichment of antigen presentation genes in acute COVID-19, in contrast to NALT cDCs (Fig. 3e). Altogether, this is consistent with a functional switch in moDCs in acute COVID-19 away from antigen presentation and toward neutrophil recruitment.

**Fig. 3: Tissue-resident macrophages take on pro-repair phenotype in convalescence.**

C1Q^hi macrophage representation in NALT remained stable in health, acute and convalescent COVID-19, but FOLR2⁺ macrophages significantly increased in convalescent COVID-19 (Fig. 3b). Spatially, macrophages were predominantly located in the T cell zone, including the FOLRB⁺ subset (Fig. 3f and Extended Data Fig. 4b). These FOLR2⁺ macrophages also showed temporal changes in their transcriptional program between acute and convalescent COVID-19, demonstrating enrichment of IL-10- and IL-4-stimulated macrophage gene signatures²⁸ in convalescent disease (Fig. 3g and Extended Data Fig. 4c). EMT pathway genes highly expressed in FOLR2⁺ macrophages included COL6A (Fig. 3h), an important structural component in many tissues²⁹. These FOLR2⁺ macrophages also expressed the tolerogenic receptor LILRB4, matrix metalloproteinases and CCL18 (Fig. 3h), a profibrotic cytokine associated with the differentiation of ‘M2’-like macrophages³⁰. Flow cytometric analysis confirmed increased expression of CD206, a marker of M2 macrophages, in nasal CD14⁺ cells in convalescent COVID-19 (Extended Data Fig. 4d).

Altogether, our analysis suggests that NALT myeloid cells play differing roles during the course of SARS-CoV-2 infection; infiltrating monocytes and moDCs produce self- and neutrophil-recruiting chemokines to set up a proinflammatory positive feedback loop in acute COVID-19 and localize to the periphery of NALT, forming a defensive shield. Meanwhile, NALT macrophages reside predominantly within the T cell zone and include a FOLR2⁺ subset that is expanded and transcriptionally activated in convalescent COVID-19, adopting a pro-repair, anti-inflammatory transcriptome, promoting tissue integrity and ensuring that the infrastructure for the generation of local adaptive immune responses is maintained.

Clonally expanded NALT CD8 T_RM cells show prolonged activation

We next considered T cell, natural killer (NK) cell and innate-like lymphocytes in isolation and identified 19 cell types (Fig. 4a and Extended Data Fig. 5a). We found distinct clusters of naive and memory CD4 and CD8 T cells, regulatory T cells (T_reg cells), T_FH cells, innate lymphoid cells (ILCs), γδ T cells, mucosal-associated invariant T cells (MAIT cells) and three subsets of NK cells marked by expression of CD16, CD56 and/or KLRC3 (Fig. 4a and Extended Data Fig. 5a). Overall, CD4 T cells were enriched in nasal samples, comprising 80% of CD3⁺ cells in NALT (Fig. 4b, Extended Data Fig. 5b, confirmed by flow cytometry, and Extended Data Fig. 5c). Tissue-resident memory (T_RM) CD8 T cells, T_FH cells and ILCs were almost exclusively found in NALT samples, whereas CD16⁺ NK cells, CD8 T effector cytotoxic (CTL) and CD8 T effector memory cells were predominantly found in blood (Fig. 4b,c and Extended Data Fig. 5d,e). Spatially, programmed cell death protein 1 (PD-1)-expressing T_FH cells were located within B cell follicles (Fig. 4d), whereas CD8 T cells dominated within the epithelium (Fig. 4e), including CD8⁺CD103⁺ T_RM cells, which were also scattered in the B cell follicles and interfollicular regions (Fig. 4f,g and Extended Data Fig. 5f). In COVID-19, there was an expansion of NALT CD8 T cells compared with healthy controls, particularly naive (T_n), central memory (T_CM) and T_RM cells (Fig. 4c). The last is consistent with studies of T_RM cells in murine lymph nodes, demonstrating that, although most tissue CD8 T_RM cells remain indefinitely resident, a small number of emigrants re-locate to local draining lymph nodes, becoming broadly distributed to enable rapid expansion and defense on local rechallenge³¹.

**Fig. 4: NALT CD8 T_RM cells exhibit clonal expansion and prolonged activation.**

As with myeloid cells, nasal T cell or innate lymphocyte subsets demonstrated greater infection-associated transcriptional changes compared with their circulating counterparts in acute COVID-19, particularly ‘INFα response’ and ‘INFγ response’ gene sets (Fig. 5a and Extended Data Fig. 5g). In convalescent COVID-19, the INFα response gene-set expression returned to healthy control levels in all NALT CD8 T cell subsets except for CD8 T_RM cells (Fig. 5a and Extended Data Fig. 6a). Indeed, CD8 T_RM cells showed evidence of ongoing activation in convalescent participants with significant enrichment of several immune-relevant gene sets (Fig. 5a and Extended Data Fig. 6b), including cytotoxicity-effector programs, with IFNG and IL26 among notable leading-edge genes (Fig. 5b and Extended Data Fig. 6c,d). Of relevance, IL-26 binds to DNA released from damaged cells to promote damage-associated inflammation³² and may also act directly on epithelial cells, promoting CXCL8 production³³. In convalescence, NALT T_RM cells also expressed tissue repair-associated molecules including TGFB1 and IL32 (Fig. 5b). Single-cell T cell receptor (TCR) analysis showed the lowest levels of diversity in the nasal CD8 T_RM cell population in convalescent COVID, as measured by Shannon entropy (Fig. 5c), suggestive of clonal expansion. Indeed, comparing healthy control, acute and convalescent COVID-19, a progressive increase in clonally expanded T cells (defined as TCR clonotype with ≥2 cells) was uniquely detectable in T_RM cells among nasal CD8 T cells (Fig. 5d and Extended Data Fig. 6e,f). Expression of canonical CD8 T_RM cell transcripts was also restricted to the clonally expanded population (Extended Data Fig. 6g). Using 44,139 unique major histocompatibility complex (MHC)-I-matched SARS-COV-2-specific TCR CDR3 regions from a combination of the ImmunoCODE³⁴ and VDJdb³⁵ databases, we used GLIPH2 (ref. ³⁶) as an in silico method to identify SARS-CoV-2-specific CD8 T cells through CDR3 sequence matching. This indicated an increase in virus-specific CD8 T cells in peripheral blood in acute COVID samples, but a later expansion in convalescent NALT CD8 T_RM cells (Extended Data Fig. 6h). In murine models, expression of ID2 and ID3 has been linked to the establishment of short- and long-term memory CD8 T cells, respectively³⁷. CD8 T cells in both NALT and peripheral blood upregulated ID2 in acute and convalescent COVID-19, but upregulation of the long-term memory marker ID3 was significant only in convalescent patients (Extended Data Fig. 6i).

**Fig. 5: NALT CD8 T_RM cells exhibit clonal expansion and prolonged activation.**

Persistent CD8 T cell activation can lead to exhaustion. Overall, exhaustion-associated genes³⁸ and inhibitory molecules, such as TIGIT, LAG3, VSIR and PDCD1 (encoding PD-1), were upregulated in convalescent compared with acute COVID-19 (Fig. 5e and Extended Data Fig. 6e,j). In the present study, CD8 T_RM cells again differed from other cytotoxic nasal CD8 T cell subsets, with less enrichment of exhaustion gene-set enrichment and lower expression of TIGIT and PDCD1 (Fig. 5e and Extended Data Fig. 6e).

Together, our data reveal marked induction of IFN-induced antiviral programs in NALT T cells in active COVID-19 infection that rapidly subsides in convalescence. The exception to this was CD8 T_RM cells, which alone showed clonal expansion and evidence of ongoing cytotoxic activity in convalescence, as well as production of IFNγ, antimicrobial peptides and pro-repair molecules. Remarkably, despite this persistent activation, T_RM cells had reduced transcriptional evidence of exhaustion compared with other CD8 T cell subsets. Although T_RM cells have previously been described in human secondary lymphoid organs³⁹, our study provides a unique insight into how NALT T_RM cells respond during viral challenge, suggesting that they play parallel roles in defense, the coordination or recruitment of other T cell subsets and tissue repair.

Antiviral GC responses in SARS-CoV-2 infection

In the B cell compartment, we identified naive, nonswitched and switched memory, FCRL4⁺ mucosa-associated memory B cells, immunoglobulin (Ig)A/IgG plasma cells (PCs) and early plasmablasts, as well as distinct clusters of GC-associated populations, including dark zone, light zone and memory precursors^4,40,41 (Fig. 6a and Extended Data Fig. 7a–c). Cells annotated as B GCs showed expression of canonical GC markers (AICDA and BCL6), but failed to enrich for either a light or dark zone signature specifically.

**Fig. 6: GC and PC expansion in NALT after COVID infection.**

Most B cells profiled originated from nasal samples (Fig. 6b,c), confirmed by flow cytometry (Extended Data Fig. 7d,e). The nasal B cell compartment was proportionally expanded in both acute and convalescent COVID-19 compared with healthy controls, with a prominent increase in GC populations in convalescent samples (Fig. 6b). FCRL4⁺ memory cells decreased in COVID-19 and there was also a proportional reduction in nasal switched memory B cells in convalescence compared with acute disease (Fig. 6b), confirmed by flow cytometry (Extended Data Fig. 7e).

Spatially, PCs were layered along the entire subepithelium in convalescence, with many IgG⁺ and IgA⁺ PCs co-localizing here, along with occasional IgD⁺ PCs (Fig. 6d and Extended Data Fig. 7f,g), the latter previously described in tonsils and the nasopharynx⁴². Expression of the polyclonal Ig receptor (PIGR), required for transepithelial shuttling of IgA⁴³, was evident in NALT epithelium and increased in convalescent COVID-19 (Fig. 6e), augmenting the capacity to transport IgA generated by adjacent PCs into the nasal space. Transepithelial transport of IgG requires the neonatal Fc receptor (FcRn)⁴⁴. FCGRT transcripts were detectable in nasal epithelial cells but did not show an increased post-COVID-19 infection and, indeed, even decreased (Fig. 6e), potentially indicating that IgG produced in NALT may bolster systemic rather than nasal IgG. Most IgG⁺ PCs, as well as many IgA⁺ PCs, expressed CXCR4 (Fig. 6f), enabling localization to a CXCL12-expressing niche. The remaining IgA⁺ PCs expressed CCR10 and CXCR3 (Fig. 6f), with NALT stromal cells the only detectable source of CCL28 (the ligand for CCR10), but myeloid cells the major source of the CXCR3 ligands CXCL9, CXCL10 and CXCL11 (Fig. 6g and Extended Data Fig. 7h). Indeed, macrophages and DCs also expressed TNFSF13 and TNSF13B (encoding APRIL and BAFF, respectively; Fig. 6g and Extended Data Fig. 7h), survival factors for PCs⁴⁵, and co-localized with PC clusters (Fig. 6h).

Nasal B and plasmablast cells showed induction of IFNα response pathway genes in acute COVID-19 (Fig. 6i and Extended Data Fig. 8a). Notably, GC B cells showed substantial expression of several type I IFN-dependent antiviral genes, including MX1, ISG15, IRF7 and IFITM1/2 (Fig. 6j). MX1 directly inhibits viral ribonucleoprotein complexes and has been implicated in SARS-CoV-2 defense⁴⁶ and its expression by GC B cells indicates that these specialized adaptive immune cells retain cell-autonomous antiviral capacity while simultaneously differentiating to produce progeny with class-switched, somatically mutated B cell receptors (BCRs). Indeed, among immune cells, in acute COVID-19, light zone B cells showed the highest expression of IFN-induced protein 20 (ISG20), an RNA exonuclease with broad antiviral properties⁴⁷ (Fig. 6j).

BCR analysis showed an increase in CDR3 junctional length in GC B cell populations in convalescent COVID-19 compared with other groups, suggesting increased V(D)J rearrangement and selection (Extended Data Fig. 8b). The Gini index (a measure of clonal selection) was also significantly increased in convalescent COVID-19 (Fig. 7a). Analysis of class switching in expanded clones (more than two) showed a marked expansion of IgM⁺ clones (predominantly nonswitched memory B cell clones) and IgG1 and IgG2 clones (predominantly GC B cells in active COVID-19, with IgG clones more numerous in convalescence) (Fig. 7b). Thus, class switching to IgG rather than IgA appears to be the favored local mucosal response to SARS-CoV-2 infection. In acute COVID, we found an expansion of memory B cell clones, with expanded GC clones more prominent in convalescence (Fig. 7c).

**Fig. 7: Type I IFN response and GC cell fate and progression.**

Next, we sought to identify SARS-CoV-2-specific BCRs, comparing our single-cell BCR data to a reference COVID antibody database (CoV-AbDab)⁴⁸. First, using GLIPH2 (Grouping of Lymphocyte Interactions by Paratope Hotspots v.2) to perform heavy chain matching with CoV-AbDab, SARS-CoV-2-binding BCR sequences were defined as cells with an identical CDR3-L sequence and a paired CDR3-H with significant motif similarity to an antibody within the CoV-AbDab (Extended Data Fig. 9a,b). Putative SARS-CoV-2-binding BCRs (n = 768) comprised around 5% of GC B cell BCRs in patients with COVID-19, but were not detectable in control GC B cell BCRs (Fig. 7d,e). Clonal expansion of SARS-CoV-2-specific B cells was also principally evident among GC B cells in convalescent COVID-19 participants (Fig. 7f and Extended Data Fig. 9c,d), supporting the conclusion that this lymphoid tissue generates early local antiviral antibodies in COVID-19. As GLIPH2 is typically used for TCR matching, we also matched SARS-CoV-2-specific B cells identified by GLIPH2 with SARS-CoV-2-specific cells identified by Levenshtein distance similarity to CoV-AbDab BCRs. Although this identified a smaller number of SARS-CoV-2-specific BCRs, the pattern of expansion of SARS-COV-2-specific B cells mirrored the results generated by GLIPH2 (Extended Data Fig. 9e,f).

The cues that determine which cellular fate a GC B cell adopts and what signals initiate GC exit to a memory B cell or PC fate in humans remain unclear. Murine studies suggest that this decision is influenced by a combination of BCR antigen affinity, local cytokine environment and the provision of co-receptor signals by T_FH cells. A temporal model has been proposed, where the GC response undergoes a switch in its output as it matures, with memory B cells preferred at early time points and PCs later⁴⁹. Temporal studies of GC progression in lymphoid organs in humans are lacking, so, to investigate this, we used Slingshot, pseudotime, cell-fate trajectory analysis⁵⁰. To specify the root of this trajectory analysis, we used cellular entropy, a marker of transcriptional diversity that has been proposed as inversely proportional to cellular differentiation⁵¹. After dark zone GC cells, which are known to increase their genetic diversity through somatic hypermutation, this identified the B GC light zone cluster (B_GC_LZ) as the least differentiated (Extended Data Fig. 10a). Pseudotime trajectory analysis rooted in the light zone GC cluster revealed three distinct cell trajectories, one terminating in the antibody-secreting cell (ASC) cluster, the second in the switched memory B cell cluster and the third in the GC dark zone population, which we termed a ‘GC re-entry’ fate (Fig. 7g). When comparing disease groups and control, we found a paucity of cells in the terminal stages of the GC re-entry trajectory in acute COVID-19, in contrast to convalescent COVID-19, where many cells were evident in the late GC re-entry trajectory. Cells from convalescent COVID-19 patients also showed an early peak in density in the ASC trajectory compared with acute COVID-19 and controls (Fig. 7h). Taken together with our analysis of expanded clones, our pseudotime data are consistent with a temporal model of GC progression.

Murine studies suggest that the switch in GC output over time may be mediated by changes in the transcriptional characteristics of T_FH cells, with the ratio IL-21:IL-4 decreasing over time, and a switch toward IL-4 production. In our dataset, T_FH cells showed increased expression of IL21 in convalescent COVID-19 compared with acute disease and control samples, with little IL4 detectable (Fig. 8a). T_FH cell–light zone GC B cell interaction prediction using CellPhoneDB also showed a temporal increase in several interactions that would promote GC progression, including CXCL13 CXCR5, TNFRSF13C–TNFSF13B, IL21R–IL21 and CD40–CD40L in convalescent COVID (Fig. 8b). Using cell2TCR⁵², we matched T cell subsets in the NALT to an experimentally derived T cell activation signature generated from a SARS-CoV-2 human infection model⁵², and used this to plot T cell subset activation profiles in NALT over the course of infection. This showed a significant increase in T_FH and follicular regulatory T (T_FR) cell activation in convalescent disease, the latter T cell expressing increased levels of TNFRSF18 (GITR), CTLA-4 and PD-1 (Extended Data Fig. 10c).

**Fig. 8: Type I IFN response and GC cell fate and progression.**

We hypothesized that, during viral infection, type I IFNs may provide an environmental cue that influences GC output. Assessment of gene expression showed persistent induction of type I IFN response genes in cells across the memory B cell trajectory, in contrast to cells progressing along a GC re-entry or ASC trajectory (Fig. 8c). It is interesting that, in GC B cells, there was a negative correlation between type I IFN response genes and HMGB2 and BACH2 in SARS-CoV-2 infection, particularly in convalescent disease (Fig. 8d and Extended Data Fig. 10d,e). In NALT T_FH cells, type I IFN response genes showed a negative correlation with BCL6 in COVID-19 (Fig. 8e). Altogether, this is consistent with the conclusion that type 1 IFN may inhibit the progression of the GC reaction, potentially providing an explanation as to why severe COVID-19 is associated with limited GC formation.

Discussion

Although there is macroscopic involution of the pharyngeal tonsils with age, our study showed that the postnasal space (PNS) contains lymphoid tissue in adults that is responsive to viral infection and can be sampled to profile the cellular molecular processes occurring during the generation of adaptive immune responses in living participants. We showed that these NALT biopsies can be collected in an outpatient clinic setting, providing a means of longitudinally assessing secondary lymphoid tissue in humans. NALT immune responses are particularly relevant to pathogens that enter via the nasal mucosa, including SARS-CoV-2, and to nasal vaccines^10,11. We demonstrated that NALT profiling enables the study of cell types that are central to adaptive immunity, but not present in blood, particularly GCs and stromal cells. A recent study using PNS swabs similarly captured adenoid GC populations in the context of SARS-CoV-2 infection⁵³. This methodology has the advantage of enabling longitudinal sampling but does not provide spatial information or include stromal or significant numbers of myeloid cells.

Lymphoid tissues contain several myeloid cell populations. We found that, in acute COVID-19, monocytes formed a superficial lining in NALT, expressing the antibacterial protein calprotectin²¹, as well as self- and neutrophil-recruiting chemokines. Neutrophils were also evident in this zone, some of which had histone citrullination, required for chromatin decondensation and NET formation²². NETs can also entrap and kill bacteria⁵⁴, potentially acting together with monocytes to defend the underlying follicles and T cells from nasal bacteria during a period of vulnerability when overlying epithelial cells are damaged or killed by invading virus.

Murine studies show that subcapsular sinus macrophages play important roles in viral defense⁸. Furthermore, lymphatic sinus-lining macrophages produce IL-18 in response to bacterial infection, stimulating innate lymphocytes to produce IFNγ, which enhanced macrophage antibacterial function⁹. The pharyngeal tonsils do not have afferent lymphatic drainage via a subcapsular sinus, but rather antigens and pathogens come directly from the overlying nasal space. In the present study, we found that human NALT contains a dense network of macrophages, including a FOLR2⁺ population, which expanded in convalescent COVID-19, expressing IL18, and pro-repair molecules, including type VI collagen.

Murine models of upper respiratory tract influenza challenge have shown CD8 T_RM cell expansion in the nasal tissue⁵⁵ and SARS-CoV-2-specific CD8 T_RM cells have also been described in human nasal mucosa⁵⁶. Our study confirmed their presence in NALT and showed that, unlike other CD8 T cell subsets, T_RM cells clonally expand and show prolonged cytotoxic gene expression in convalescence, with little evidence of exhaustion. This is in contrast to previous reports suggesting T_RM cell exhaustion in active COVID-19 infection⁵⁷. It is interesting that NALT T_RM cells also expressed lymphocyte-recruiting chemokines and TGFB1, suggesting that they play additional roles beyond immediate antiviral defense.

We found reciprocal changes in the expression of IgG and IgA transporters in NALT epithelium, which suggest an increase in IgA transport to the nasal mucosa, but that NALT IgG contributes to systemic immunity. We did not take nasal secretion or saliva samples to directly measure Ig levels; however, studies delineating secretory IgA, IgA and IgG responses in SARS-CoV-2 support this finding^58,59.

Our longitudinal analysis of GC progression in SARS-CoV-2 infection provides unique insights into the dynamics of GC responses in humans. Although studies in mice suggest a temporal model where memory B cells emerge earlier in responses, this has been difficult to confirm in humans. We found increased representation of memory B cells in expanded clones in acute COVID-19, which decreased in convalescence, and our pseudotime analysis showed a peak in cell density in the ASC trajectory in convalescence. Together, these analyses are supportive of a temporal model of GC progression. We also found that type I IFN may provide an environmental cue that influences GC output, with memory cells showing the highest expression of type I IFN response genes throughout their trajectory.

In the present study, we showed that adults with mild COVID-19 disease do form GCs in response to SARS-CoV-2. This enabled us to probe potential mechanisms of GC collapse described in fatal disease¹⁹. We found that high IFN response gene expression was associated with reduced expression of BACH2 and BCL6 in GC B cells and reduced expression of CD40L in T_FH cells. This implies that an excess of type I IFNs, as has been described in severe disease, may inhibit the progression of the GC reaction.

A limitation to our study is that the acute and convalescent samples were not taken from the same participants and two out of eight of the acute COVID-19 patients were on corticosteroids, one on corticosteroids and baricitinib and one on corticosteroids and tocilizumab, compared with the five convalescent participants, none of whom had severe disease or were on immunomodulating medications, which may influence the acute and convalescent comparisons.

In summary, our study provides a unique insight into how nasal adaptive immune responses are generated and defended during SARS-CoV-2 infection. Beyond COVID-19, our study provides proof of principle that this sampling strategy will enable long-sought efforts to interrogate longitudinal changes in lymphoid organs in humans in disease and after therapeutic intervention.

Methods

The list of reagents used in the present study can be found in Supplementary Table 2.

Human participant recruitment

COVID-19 participants were recruited from a single site—Addenbrooke’s Hospital, Cambridge, UK—between May 2020 and July 2021. Healthy control individuals were recruited between August 2019 and November 2019 from Cambridge, UK. Ethical approval was given through the Cambridge Central Research Ethics Committee (reference no. 08/H0308/267, IRAS project no. 194217), administered through Cambridge University Hospitals NHS Foundation Trust. All participants provided informed written consent and, thus, patients unable to provide this, including those undergoing invasive ventilation, were excluded. The following participants were also excluded: aged <18 years, with a chronic systemic infection (for example, HIV, viral hepatitis), with a diagnosis of active malignancy, previous head and neck radiotherapy, a diagnosis of systemic autoimmune disease (for example, rheumatoid arthritis, ANCA-associated vasculitis) or a known nasal or postnasal anomaly precluding access or safe biopsy (for example, severe septal deviation, nasal vascular malformation). All active COVID-19 participants, with the exception of participant no. 1, were recruited within 7 d of their first positive swab (quantitative RT–PCR) result and within 14 d of symptom onset. Participant no. 1 was recruited on the basis of radiological changes consistent with COVID-19, in the presence of positive COVID-19 serology and the onset of symptoms consistent with COVID-19 within the past 14 d. All convalescent COVID-19 participants were recruited between 21 d and 28 d after a positive COVID-19 swab result. No COVID-19 participants had received a COVID-19 vaccine at the time of recruitment.

Sample collection

Nasal samples were collected by an otolaryngologist under direct endoscopic visualization. Briefly, the nasal cavity was anesthetized with topical local anesthetic and epinephrine (lidocaine 5%:phenylephrine 0.5%) administered via spray and neurosurgical patties (Codman). Nasal brushings and microcurettage samples were collected from the inferior turbinate under direct vision with the aid of a Thudichum nasal speculum. The PNS was inspected transnasally with a 0° nasendoscope and one to three biopsies taken from the PNS mucosa using Tilley–Henkel, Takahashi fine-cupped or Blakesley forceps, and placed directly into Roswell Park Memorial Institute (RPMI)-1640 medium on ice. Blood samples were collected contemporaneously into sodium citrate tubes using the vacutainer system (BD)

Sample processing to single-cell suspension

All samples were processed fresh, within 1 h of collection. PNS samples were washed in RPMI-1640 medium, diced and placed into a 15-ml Falcon tube with 5 ml of RPMI-1640 medium and a digestion solution of 62.5 μl of Liberase (stock concentration 1 mg ml⁻¹; Sigma-Aldrich) and 250 μl of DNase I (stock concentration 2.5 mg ml⁻¹; Roche). The sample tube was placed into a shaking incubator at 37 °C for 30 min at 220 r.p.m. The sample was then pushed through a 70-μm cell strainer (Falcon) into a 50-ml Falcon tube and washed in 50 ml of RPMI-1640 medium at 500g for 6 min with the brake on. The cell pellet was resuspended in 10 ml of 44% Percoll solution and centrifuged at 800g for 20 min and 20 °C with no brake. The supernatant was aspirated. If the cell pellet had visible blood staining, 2 ml of ACK red cell lysis buffer was added for 2 min, before quenching with 20 ml of 1× phosphate-buffered saline (PBS) and centrifuging at 500g and 4 °C for 6 min (brake on). After aspirating the supernatant, the cell pellet was resuspended in 100 μl of 1× PBS. Nasal brushing heads were vigorously agitated 15× in 5 ml of RMPI-1640 collection fluid. The brush head was cut off and left in the collection fluid while it was centrifuged at 500g for 6 min at 4 °C (brake on). The brush head and supernatant were removed and the cell pellet resuspended in 100 μl. Nasal microcurettes were agitated in 5 ml of RPMI-1640 collection fluid before centrifugation at 500g for 6 min (4 °C, brake on). The resulting cell pellet was resuspended in a digestion fluid of Liberase or DNase I at the same concentration as the PNS samples and placed into a shaking incubator for 20 min at 37 °C and 220 r.p.m. After digestion, samples were disassociated by pipette mixing and washed at 500g for 6 min (brake on) and resuspended in 100 μl of 1× PBS. Then, 18 ml of blood from the sodium citrate collection tubes was diluted 1:1 with 1× PBS and 15 ml of Histopaque 1077 (Sigma-Aldrich) was added to each of two 50-ml Falcon tubes, with the blood–PBS mixture layered on top. Each tube was centrifuged at 800g for 20 min at room temperature (low brake). The resulting PBMC layer was aspirated with a Pasteur pipette and washed twice in 50 ml of PBS before being resuspended in 500 μl of 1× PBS.

The samples were counted with Trypan Blue staining to assess cell counts and viability, before downstream applications.

ScRNA-seq library generation

PNS samples were pooled with curettage and brushing cells at a ratio of 2:1:1. Where nasal curettage and/or nasal brushing cell counts were insufficient or unavailable, only PNS samples were used. Blood and nasal samples were diluted to a concentration of 2 × 10⁶ cells per ml and loaded on to a 10x Chromium controller in parallel lanes with a targeted cell recovery of 14,000 cells. Single-cell libraries + TCR and BCR VDJ libraries were created using Chromium Next GEM (Gel Beads in Emulsion) Single cell 5′ kit v.1.1 (10x Genomics), in accordance with the manufacturer’s user guide. Samples were sequenced using a Novaseq 6000 S4 system (Illumina).

Flow cytometry

After scRNA-seq GEM creation, the remaining single-cell suspensions were diluted to a concentration of 2 × 10⁶ cells in 100 μl of 1× PBS. If this concentration was not achievable, all remaining cells were used. Cells were blocked with 5 μl of FcR blocking reagent (Miltenyi-Biotech) before undergoing surface staining with fluorophore-labeled antibodies. The antibodies used in the present study are listed in Supplementary Table 2. The samples were washed and resuspended in 1× PBS and run on a Fortessa II flow cytometer (BD). Data were gated into populations using FlowJo (Treestar/BD). All populations used in this publication were gated on a starting population of CD45⁺ live singlets. Gating strategies are available in Supplementary Fig. 1.

Immunofluorescence microscopy

PNS biopsies were collected as described above and fixed for 15–30 min (depending on tissue size) in AntigenFix (DiaPath, cat. no. P0016), followed by 8 h in 30% sucrose in PBS before embedding in OCT (CellPath, cat. no. KMA-0100-00A). Then, 15-µm sections were mounted on Polysine slides (Thermo Fisher Scientific, cat. no. J2800AMNZ) and permeabilized and blocked in 0.1 M Tris, pH 7.4 containing 0.1% Triton (Sigma-Aldrich, cat. no. 93426-100ML), 1% normal mouse serum (Invitrogen, cat. no. 10410), 1% normal donkey serum (Abcam, cat. no. ab7475) and 1% bovine serum albumin (R&D Systems, cat. no. DY995) for 1 h at 20 °C. Samples were stained for 2 h at 20 °C (primary antibodies) or 1 h at 20 °C (secondary antibodies) in a humid chamber with the appropriate antibodies, washed 3× in PBS and mounted in Fluoromount-G (Southern Biotech, cat. no. 0100-01). Images were captured using a TCS SP8 (Leica) inverted microscope, on a ×40, 1.1 numerical aperture, water immersion objective. Raw imaging data were processed using Imaris (Bitplane/Oxford Instruments). All antibodies used in the present study are listed in Supplementary Table 2.

Iterative staining of PNS biopsies was performed as described by Radtke et al.^61,62. Staining and imaging were performed as described above. After imaging, the coverslip was removed and the slides washed 3× in PBS to remove any mounting medium. Bleaching of the fluorochromes was achieved using 1 mg ml⁻¹ of a solution of lithium borohydride (Acros Organics, cat. no. 206810050) in water for 15 min at 20 °C. The slides were then washed 3× in PBS before staining as described above with a different set of antibodies. The process was repeated up to 7×. Raw imaging data were processed using Imaris with either Hoechst or CD19 as fiducial for alignment of subsequent images.

Quantification and statistical analysis

ScRNA-seq data processing

Sequenced data were aligned to hu38 human reference genome using Cellranger v.6 (10x Genomics). Removal of ambient RNA and assigning of droplets were undertaken using Cellbender⁶³. Automated doublet detection and filtering were performed using Scrublet⁶⁴ (P < 0.1), as well as manually by gene co-expression. Quality control and filtering were performed in Scanpy⁶⁵, using the sc-dandelion pre-processing module (dandelion.pp,recipe_scanpy_qc) with max. genes = 6000, min. genes = 200 and a Gaussian mixture model distribution to determine mitochondrial filtering values. Batch correction was performed with Harmony⁶⁶ and Leiden clustering was also performed. Cell labels were assigned manually by gene expression profiles (Supplementary Fig.1c), with iterative rounds of subclustering to assign cell subtype labels. Assigned cluster labels were checked against signatures of flow-sorted, sequenced immune cells⁶⁷ using SingleR⁶⁸. Differential gene expression was performed by cell type and condition using FindMarkers in Seurat⁶⁹ (logfc.threshold = 0.1, Min.pct = 0.1, min.cells.group = 10). GSEA⁷⁰ was performed on lists of DEGs using fgsea⁷¹ (minSize = 15) and plotted in ggplot2, with P values calculated by permutation using the fgsea package intrinsic functions. Gene-set expression scores were calculated as the difference between the average expression levels of each gene set and randomly sampled pool of all (control) genes for each cell, using scanpy.tl.score_genes: an implementation of Seurat’s AddModuleScore function⁷². INFα response gene expression correlations were calculated using Pearson’s correlation coefficient implemented through the scipy package, with two-sided P values reported. Correlations were calculated on named cell subtypes with named gene expression values >0. Cell proportion plots were performed using ggplot2, with calculation of P values and confidence intervals (CIs) using a permutation and bootstrapping approach from the scProportionTest package. Trajectory analysis was performed using Slingshot⁵⁰. For B cell trajectories, memory B cells, GC B cells and PCs were subset from the main object and reclustered. B cell Slingshot lineages were inferred with B GC light zone (B_GC_LZ) assigned as the starting cluster. For MNP cell trajectories, monocytes, macrophages and cDCs were subset from the main object and reclustered. MNP cell Slingshot lineages were inferred with classic monocytes assigned as the starting cluster. CellPhoneDB analysis was performed as previously described⁷³, with additional plotting of results using ktplots. Single-cell TCR analysis was performed using Scirpy⁷⁴. Single-cell BCR analysis was performed using sc-Dandelion⁷⁵ with V(D)J gene reannotation and parsing to AIRR format performed in the sc-Dandelion singularity container (sc-dendelion_latest.sif). Gini’s index was calculated in sc-Dandelion, with P values calculated using Tukey’s honestly significant difference test implemented through the scikit-posthocs package. For identification of putative BCR sequences with capacity to bind SARS-CoV-2 surface antigens, we downloaded CDR3 sequences and germline assignments of known SARS-CoV-2-binding antibodies from the Coronavirus Antibody Database (CoV-AbDab, updated 26 July 2022)⁴⁸ and filtered for antibodies of B cell origin. We defined putative SARS-CoV-2-binding BCR sequences in our dataset as B cells with complete CDR3-L match and significant motif similarity for CDR3-H compared with the CoV-AbDab database. Motif enrichment analysis between BCR CDR3-H sequences in our data and CoV-AbDab was performed using GLIPH2 (ref. ³⁶). First, we compiled a reference database of naive BCR CDR3 sequences using published CDR3 sequence from COVID-naive B cells from multiple sources, for the GLIPH2 pipeline. Briefly, 1,437,743 CDR3-H sequences from bulk BCR sequencing (downloaded from iReceptor gateway 30 July 2022)⁶⁰ and 55,444 CDR3-H sequences from three single-cell GEX + V(D)J-seq studies^75,76,77 of control COVID-naive B cells were included in the reference. For GLIPH2 analysis, we considered only motifs that were more than six residues in length. We assessed motifs biased toward the target data (CoV-AbDab + BCR-seq V(D)J data combined) versus the naive reference data (Fisher’s exact test P < 0.05), to identify sequences enriched in our cohort. Shared motifs were visualized using ggseqlogo⁷⁸. We recombined CDR3-H and CDR3-L sequences using their single-cell barcodes and filtered for single cells where both CDR3-H and CDR3-L sequences matched the corresponding CoV-AbDab antibody.

Harmonization and label transfer with a publicly available, airway brushing dataset were performed with scANVI⁷⁹, using seed labels from the annotated NALT and PBMC sample cell-type labels^{80,81,82,83,84}.

Reporting summary

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