Functional role of skull lymphoid structures in CNS immunosurveillance

Recent findings have highlighted the direct connection between skull bone marrow (BM) and the dura mater through osseous channels, enabling cerebrospinal fluid (CSF) to reach the skull BM^1,2,3,4,7. This connection is functionally important, as the skull BM is a primary lymphoid organ that supplies immune cells—including developing and immature B cells and myeloid cells—to the CNS and CNS border regions under homeostatic and neuroinflammatory or injurious conditions^5,6. Recent studies also suggested that skull BM contains tumour-specific CD8⁺ T cells in human patients with glioblastoma and regulatory T cells in mouse tumour models^8,9. Although the BM is a primary lymphoid organ, adaptive immune responses also occur within it under certain conditions^10,11,12. Mature antigen-presenting cells (APCs), T cells and B cells migrate to the BM, where residing naive T and B cells can encounter antigens and form clusters^{13,14,15,16,17,18,19}. While previous studies have shown that BM can support adaptive immune responses, whether and how skull BM contributes to CNS antigen recognition and immunosurveillance has not yet been elucidated.

Here we demonstrate that the skull BM of mice contains cellular components that are characteristic of peripheral lymphoid organs. APCs, T cells and B cells form clusters with germinal centre (GC)-like structures. These skull lymphoid structures participate in CNS immune surveillance under both homeostatic and brain cancer conditions in mice. By modulating their presence and activity in vivo, we show that these structures contribute to anti-tumour immunity.

We compared T cell phenotypes in the skull BM to those of other BM sites in mice to characterize the adaptive immune composition of skull BM. The frequency of CD4⁺ and CD8⁺ T cells within total immune cells was similar across BM sites, with a higher frequency in the sternum BM, but overall lower than in secondary lymphoid organs (SLOs) (Extended Data Fig. 1a,b). The frequencies of naive and effector T cells were also comparable. However, skull BM displayed a higher proportion of central memory cells compared with other BM sites (Extended Data Fig. 1c,d), suggesting a distinct T cell environment within the skull BM.

To further examine T cell phenotypes in BM, we conducted 10x single-cell RNA sequencing (scRNA-seq) analysis of T cells from mouse skull and sternum BMs. On the basis of clustering, we identified diverse CD4⁺ T cell subsets, including naive (Sell and Ccr7), activated (Cd44), interferon-stimulated gene (ISG)-expressing (Isg15), regulatory (Foxp3) and differentiated helper (Tbx21, Gata3 and Rorc) cells (Fig. 1a and Extended Data Fig. 1e). Notably, we observed follicular helper T (T_FH) cells marked by the expression of genes such as Pdcd1, Cxcr5, Bcl6 and Il21 (ref. ²⁰) (Fig. 1a,b, and Extended Data Fig. 1e,f).

To validate these findings at the protein level, we performed flow cytometry analysis of BM from IL-21–VFP (vivid Verde fluorescent protein) mice, and identified IL-21⁺CXCR5⁺ T_FH cells as a distinct subset (Fig. 1d). We confirmed that CD4⁺ T cells are the primary source of IL-21 in the BM (Extended Data Fig. 1g), with the majority of T_FH cells in an effector stage and some in a central memory stage (Extended Data Fig. 1h). Protein-level expression of markers on T_FH cells was higher than on IL-21⁻CXCR5⁻ double-negative cells (Fig. 1c and Extended Data Fig. 1i,j). These T_FH cells also expressed T helper type 1 (T_H1) markers, such as CXCR3 and T-bet, and the majority of activated non-regulatory T (T_reg) cells and non-T_FH cells were T_H1 cells (Extended Data Fig. 1i–k). The presence of T_FH cells in the skull BM was also confirmed by confocal microscopy (Extended Data Fig. 1l,m).

Skull BM exhibited a higher percentage of T_FH cells than other BM sites (Fig. 1d and Extended Data Fig. 1n). Analysis of public scRNA-seq data from human skull BM (GSE233304)⁸ also supported the presence of T_FH-like cells, despite their low abundance (Extended Data Fig. 1o–r). Confocal imaging analysis of whole-mounted calvarium demonstrated that T_FH cells form clusters with B cells in the BM of occipital and interparietal regions of the skull (Fig. 1e and Extended Data Fig. 1s). However, these cells were rare in the BMs of parietal and frontal skull bones as well as in sternum BM.

In summary, T_FH cells are present in the BMs of various bones, and skull BM contains a higher proportion of these cells compared with other sites, with preferential forming of clusters with B cells under homeostatic conditions.

On the basis of the observation of a higher number of T_FH cells in the skull BM, we hypothesized that the skull BM might exhibit stronger B cell responses compared with other BM sites. Analyses of an available scRNA-seq dataset (GSE184766)⁴ identified a skull-BM-specific cluster that was absent in the tibia BM (Extended Data Fig. 2a–d). While both BMs contained IgA-expressing plasma cells (PCs), the skull BM contained a distinct population of IgG⁺ plasmablasts (PBs) (Extended Data Fig. 2c–e). A pseudobulk comparison of differentially expressed genes between skull and tibia BM immune cells revealed that most differentially expressed genes in skull BM immune cells were related to antibody responses (such as Jchain and Ighg3) (Extended Data Fig. 2f), suggesting that the skull BM may provide a favourable environment for B cell activity.

We next performed scRNA-seq analysis of isolated B cells and antibody-secreting cells (ASCs) from mouse skull and sternum BM. Most B cells were in early developmental stages (Fig. 2a and Extended Data Fig. 2g). Notably, there was a cluster of B cells expressing GC-related signatures (for example, Fas, Ada and Bcl6) and ASC signatures (such as Jchain, Sdc1 and Prdm1) (Fig. 2b and Extended Data Fig. 2h). These cells expressed class-switched immunoglobulins but lacked memory markers and Ighd expression. Reclustering this group, we identified four subsets based on marker expression: Epcam⁺ PCs, Tigit⁺ PCs, PBs and GC-like B cells, which expressed genes associated with GCs, such as Bcl6, Il21r, Mki67 and Top2a (Fig. 2c and Extended Data Fig. 2i,j). The skull BM had a higher proportion of cells in the GC and ASC (GC/ASC) cluster than the sternum BM (Extended Data Fig. 2k).

While overall B cells, including developing B cells, exhibited similar transcriptional patterns across BMs, mature B cells, memory B cells and GC/ASC clusters showed distinct profiles between skull and sternum BM (Fig. 2d and Extended Data Fig. 2l). Consistent with previous results, the skull BM contained a higher proportion of PBs (Fig. 2d). Although GC-like B cells were present in the sternum BM, they were more abundant in the skull BM, where they exhibited higher GC marker expression, suggesting closer resemblance to GCs (Fig. 2d and Extended Data Fig. 2m,n).

We confirmed these observations at the protein level. While the total B cell frequency among immune cells was similar across BMs, the skull BM had a higher percentage of IgD^–IgM^– developing B cells or class-switched cells and fewer IgD⁺IgM⁺ mature B cells than the sternum and femur BMs (Extended Data Fig. 2o,p). Using both protein markers (IgD^–CD38^–FAS⁺) and activation-induced cytidine deaminase (AID)-Ai14 reporter mice, we confirmed the presence of GC-like B cells, which were more abundant in the skull BM than in other BM sites, while the frequency of class-switched memory B cells was comparable (Fig. 2e and Supplementary Fig. 1a–d). Similarly, the skull BM contained more ASCs than other BM sites (Fig. 2f and Supplementary Fig. 1e–h).

In conclusion, the skull BM exhibits a higher activation status and a greater abundance of GC-like B cells and ASCs compared with other BMs.

Canonical B cell activation and differentiation typically require CD4⁺ T cell-derived cytokines, such as IL-21, and co-stimulatory molecules, such as CD40L²¹. To assess whether T_FH cells support canonical B cell activation and differentiation in the skull BM, we first depleted CD4⁺ T cells (or CD8⁺ T cells) using antibodies (Supplementary Fig. 2a,b). CD4⁺ T cell depletion resulted in a significant reduction in ASC differentiation in the skull BM; CD8⁺ T cell depletion had no significant effect on ASC differentiation (Supplementary Fig. 2c). As the cytokine IL-21 is crucial for early B cell activation and is commonly produced by T_FH cells²², we examined skull BM of IL-21 receptor (Il21r)-deficient mice and found that skull GC-like B cells and ASCs were dependent on IL-21 signalling, similar to splenic cells (Supplementary Fig. 2d–g). However, total B cells and other immune cells in the skull BM were unaffected by deletion of Il21r (Supplementary Fig. 2h). Moreover, we showed that skull GC-like B cells rely on the CD40–CD40L axis, as systemic CD40L neutralization using anti-CD40L antibodies reduced GC-like B cells without affecting the total T or B cell numbers (Supplementary Fig. 3a–c).

To assess the local requirement of T–B cell interactions within the skull BM, we delivered a hydrogel containing anti-CD40L antibody subcutaneously (s.c.) under the scalp. Delivered antibodies preferentially bound to T_FH cells in the skull BM, but were under detection threshold in other tissues, supporting a relatively localized effect (Supplementary Fig. 3d). This treatment reduced the number of GC-like B cells and ASCs specifically in the skull BM, while these cells were unaffected in the sternum BM, dura and spleen (Supplementary Fig. 3e–j). On the other hand, peripheral hydrogel delivery of anti-CD40L into the flank did not affect GC-like B cells in the skull BM, whereas draining inguinal lymph nodes (LNs) showed depletion (Supplementary Fig. 3k). Together, these findings suggest that T_FH cells support B cell activation and differentiation within the skull BM.

We observed that IL-21-expressing T_FH cells form clusters with B cells in the skull BM (Fig. 1e), suggesting the presence of lymphoid structures²⁰. To investigate further, we performed confocal imaging of cleared mouse calvarium to examine clustering among MHC-II⁺ APCs, CD4⁺ T cells and CD20⁺ B cells. B cells formed clusters with APCs and T cells predominantly in the posterior regions of the skull BM, such as the occipital and interparietal areas of the calvaria bone, compared with other areas of the skull BM and sternum BM (Fig. 3a and Extended Data Fig. 3a). However, such immune cell organization was not observed in non-marrow regions, such as the diploic vein channels and sutures (Extended Data Fig. 3b,c).

A doublet assay also confirmed that the skull BM had more frequent T–B cell interactions than the sternum BM (Fig. 3b), with skull doublets expressing FAS and IL-21R, markers that are absent in doublets from the sternum and anterior skull BM (Fig. 3c and Extended Data Fig. 3d). Femur BM exhibited substantial amounts of FAS⁺IL-21R⁺ doublets; however, the percentage was lower than in the posterior skull BM. These findings indicate that the skull BM hosts closely interacting T and B cells, forming GC-like clusters in the presence of APCs, which are more common in the posterior than anterior skull BM.

To further determine whether these structures exhibited GC-like characteristics, we investigated multiple GC components. First, the expression of the B cell chemoattractant CXCL13 and B cell activating factor (BAFF) was co-localized within these clusters (Fig. 3d,e); second, we observed S1PR2-expressing B cell clusters (Fig. 3f); third, we used AID-Ai14 reporter mice to confirm GC-like structures (Extended Data Fig. 3e); and finally, we observed follicular dendritic cells (FDCs) within the B cell-rich area (Extended Data Fig. 3f), and flow cytometry confirmed that the skull BM contained FDCs while other BM sites did not (Extended Data Fig. 3g,h).

Moreover, single-cell BCR-sequencing analysis revealed that GC-like B cells and ASCs displayed clonal expansion and evidence of somatic hypermutation (Extended Data Fig. 3i–k). Some BCR clones were shared between the skull and sternum BM, skull-specific clones of GC-like B cells and ASCs were also identified (Extended Data Fig. 3l).

Notably, GC-like B cells were already present at 7 days postnatal, with substantial numbers, although ASCs were significantly less abundant at this early stage (Extended Data Fig. 4a,b). In 1-year-old mice, the percentage of overall GC-like B cells did not increase, while the posterior skull section exhibited a slight increase in S1PR2⁺ GC cells (Extended Data Fig. 4c–f) and ASCs increased in both the skull and sternum BM (Extended Data Fig. 4g–j).

In summary, the skull BM contains lymphoid structures with GC-like features that are established from the early postnatal period and into ageing.

Under homeostasis, lymphoid structures in various organs may respond to self-antigens from tissues and antigens from microbiota and diet²³. To investigate the antigens that activate cells within the skull lymphoid structures, we compared specific pathogen-free (SPF) and germ-free (GF) mice, and GF mice on an amino-acid-based antigen-free (AF) diet. As expected, IgA⁺ ASCs in Peyer’s patches were reduced in the GF and AF mice compared with in the SPF mice (Extended Data Fig. 4k). Similarly, IgA⁺ ASCs in the BM are also gut microbiota dependent²⁴. Consistently, IgA⁺ ASCs and GC-like B cells in the skull BM were decreased in GF and AF mice compared with in SPF mice (Extended Data Fig. 4l,m). However, a fraction of GC-like B cells remained, and IgG⁺ and IgM⁺ ASCs were not affected across groups (Extended Data Fig. 4m,n). Given that GC-like B cells and ASCs in the skull BM underwent somatic hypermutation and clonal expansion (Extended Data Fig. 3i–k), we hypothesized that the remaining GC-like B cells and ASCs in GF and AF mice are responding to self-antigens^25,26,27,28.

To test this, we used an adeno-associated virus (AAV) delivery system with a human synapsin 1 (hSyn) promoter to induce the expression of model antigens in neurons. We used AAV-hSyn-OVA-mCherry-P2A-eGFP, with AAV-hSyn-OVA-P2A-eGFP as a control. In this system, eGFP is retained in neurons, whereas ovalbumin (OVA)-conjugated mCherry can be detected on release. AAV-mediated gene expression was confirmed by eGFP and/or mCherry signals in the brain 1 month after infection. OVA–mCherry was detected not only at the injection site but also in the perivascular spaces and leptomeninges, whereas eGFP signals remained restricted to neurons. Importantly, mCherry signals were not detected leaking through the injection tract (Extended Data Fig. 5a).

OVA–mCherry was detected as expected^4,27,29 in the rostral rhinal hub of the dura (Extended Data Fig. 5b), deep cervical LNs (dCLNs) (Extended Data Fig. 5c) and the skull BM, where occipital BM showed the strongest signal, whereas the burr hole was negative (Extended Data Fig. 5d). Similarly, flow cytometry also revealed OVA–mCherry⁺ immune cells in the dura and skull BM, but not in other tissues (Extended Data Fig. 5e). Tracer experiments confirmed that craniotomy injury does not make an artificial conduit for CSF, and the absence of AAV-derived eGFP mRNA in the skull BM and dura excluded leakage of AAVs from craniotomy surgery (Supplementary Fig. 4).

Together, these data demonstrate that intracranial AAV delivery does not induce artificial antigen leakage and that antigens reach brain border tissues, probably through known physiological routes.

We next examined antigen-specific responses. One month after injection of AAV-OVA (or hen-egg lysozyme (HEL)-OVA), we were able to detect OVA antigen in the brain homogenate and CSF, but not in the serum (Extended Data Fig. 6a,b). We adoptively transferred naive CD4⁺ T cells from OT-II mice and B cells from MD4 mice after 1 month of AAV injection, and analysed their activation in various organs 1 week later (Fig. 4a). Transferred cells were predominantly found in different tissues, with minimal presence in the circulation (Extended Data Fig. 6c).

HEL-OVA expression in neurons resulted in minor changes in overall immune cell numbers in the skull BM (Extended Data Fig. 6d–f). Notably, skull OT-II T cells were activated, as indicated by CD44 expression with enhanced differentiation into T_FH cells. MD4 B cells also displayed more GC marker expression compared with the control group (Fig. 4b,c and Extended Data Fig. 6g,h). Similarly, neuronal HEL-OVA expression did not change the overall immune cell numbers in whole-brain tissues, while the number and activation marker expression of OT-II and MD4 cells were increased compared with the control group (Extended Data Fig. 6i,j). Transferred cells were not detected within the brain parenchyma (Extended Data Fig. 6k). Moreover, no significant differences in OT-II and MD4 cells were observed in the dura, dCLNs, spleen or sternum BM (Extended Data Fig. 6l–o). These findings suggest that neuron-derived antigens can specifically activate and differentiate T and B cells in the brain and skull BM under non-disease conditions.

We next examined antigen-specific T and B cell activation in the skull BM when antigens are derived from brain tumour cells. We engrafted either the syngeneic CT2A mouse glioma cell line or the CT2A line expressing HEL-OVA (CHELLO) into immunocompetent mice. First, we confirmed that tumour cells remained within the parenchyma without exiting along the injection route (notably, a subset of mice exhibited invasion into the proximal leptomeninges; Extended Data Fig. 7a,b). Second, CHELLO engraftment did not increase overall immune cell numbers in the skull BM or within the tumour mass when compared with the control CT2A group, suggesting that CHELLO engineering did not lead to notable immunogenicity (Extended Data Fig. 7c–f). However, after 1 week of OT-II and MD4 cell transfer, cell numbers and expression of activation markers of transferred cells in the skull BM were significantly increased (Fig. 4d–f and Extended Data Fig. 7g,h). Transferred cells were preferentially localized to the posterior skull BM (Extended Data Fig. 7i,j). CHELLO engraftment induced only minor changes in OT-II and MD4 cells in the tumour region and other tissues (Extended Data Fig. 7k–o). On the other hand, peripheral (into flank) CT2A injection elicited responses in the draining inguinal LNs but not the skull BM (Extended Data Fig. 7p–r). Thus, CNS antigen-specific anti-tumour T and B cell responses can occur locally in the skull BM. Moreover, when examined as early as day 2 after transfer, antigen-specific OT-II responses were detected only in the skull BM, whereas GC responses required a longer time to develop, suggesting that the skull BM is among the first brain barriers to respond to brain-derived antigens (Extended Data Fig. 8a–k).

Finally, we tested whether glioma grafts induce endogenous GC responses in the skull BM by comparing sham and CT2A-injected groups. CT2A injection increased the number of GC-like B cells in the skull BM, but it did not affect those in the sternum BM, dCLNs, spleen and brain, while the dura exhibited a modest increase in cell number (Fig. 4g and Extended Data Fig. 8l–p). Moreover, CT2A engraftment increased the number of B cells and effector CD8⁺ T cells in the skull BM (Extended Data Fig. 8q,r). Shallow tumour implantation, which enhances meningeal invasion, drove similar T and B cell responses in the skull BM in comparison to striatal tumours (Extended Data Fig. 8s).

To determine whether skull lymphoid responses occur independently of SLO responses, we first performed dCLN ligation surgery in CHELLO mice (Extended Data Fig. 9a). While OT-II responses in the tumour grafts were not affected by dCLN ligation, the cell number and activation of tumour-infiltrating MD4 B cells and total polyclonal B cells were reduced, suggesting that dCLNs support intratumoural B cell responses (Extended Data Fig. 9b–d). However, the cell number and activation of OT-II CD4⁺ T cells and MD4 B cells in the skull BM were not reduced by dCLN ligation (Extended Data Fig. 9e). Moreover, antigen-specific T and B cell responses were still detected in the skull BM of splenectomized lymphotoxin-α (LTA)-deficient mice (Extended Data Fig. 9f–k). These data suggest that immune responses can occur within the skull lymphoid structures, including those against local antigens, in the absence of SLOs.

To rule out the possibility that craniotomy injury contributes to our observations, we performed intrahindbrain tumour injection through the cisterna magna without craniotomy, which reproduced the previous results of skull-specific OT-II/MD4 responses (Extended Data Fig. 10a–d).

To address the role of antigen clearance from the brain in driving immune response, we used Aqp4-deficient mice, which are impaired in CSF perfusion of the brain (glymphatic system)³⁰. Aqp4-deficient mice showed reduced antigen uptake in the dura and skull BM, impaired antigen-specific responses in the skull BM and decreased survival in the CT2A brain tumour model (Extended Data Fig. 10e–g), whereas dCLN ligation did not affect antigen uptake in these tissues (Extended Data Fig. 10h). These data suggest that active, glymphatic-driven antigen efflux from the brain is required for skull lymphoid responses.

In summary, immune cells in the skull BM can be activated and expand GC-like structures in response to brain-derived antigens.

Finally, we investigated whether skull lymphoid structures have a functional role in models of brain tumours. Although B cells are known to have mixed roles in brain tumours^31,32, we confirmed that antibody responses are essential for optimal anti-tumour responses by depleting ASCs (Extended Data Fig. 11a–c). Subcutaneous delivery of hydrogel containing anti-CD40L antibodies disrupted skull lymphoid responses (Supplementary Fig. 3d–j). This partial blockade of skull lymphoid structures significantly reduced overall survival in CT2A-bearing mice (Fig. 5a), accompanied by decreased lymphocyte infiltration (Fig. 5b,c) and attenuated intratumoural CD8⁺ T cell activation (Fig. 5d).

To examine the therapeutic potential of skull BM lymphoid structures, we enhanced this niche using a CD40 agonist antibody. As CD40 agonism alone is known to induce regulatory B cells³³, we added IL-21 and IFNγ to promote GC and IgG responses, respectively (Supplementary Fig. 2d–h and Extended Data Fig. 11d). Systemic administration of the CD40 agonist alone led to a slight, non-significant increase in survival of CT2A glioma-bearing mice; however, administering all three agents together significantly extended their survival (Extended Data Fig. 11e). Local administration of a CD40 agonist mixed with cytokines and hydrogel under the scalp enhanced skull lymphoid responses but not in other tissues (Extended Data Fig. 11f,g), and significantly prolonged survival (Fig. 5e). However, local treatment with anti-CD40 agonist alone resulted in earlier mortality, possibly due to the induction of regulatory B cells. This local triple therapy also successfully induced B cell activation and IgG⁺ GC responses in the skull BM, while the change in intratumoural B cell activity was minor (Fig. 5f and Extended Data Fig. 11h,i), suggesting that this local treatment primarily targets skull BM B cells. Moreover, it was ineffective in B-cell-deficient muMT mice and ASC-deficient Jchain-DTA mice, indicating that its efficacy relies on B cells and antibody responses of ASCs (Extended Data Fig. 11j,k). Local triple therapy also exerted anti-tumour effects in mice with craniotomy-free intrahindbrain tumour injections, indicating that these results were not driven by surgery-induced damage artefacts (Extended Data Fig. 11l). Moreover, the efficacy of treatment was not affected by dCLN removal (Extended Data Fig. 11m), and this local treatment prolonged survival in splenectomized LTA-deficient mice, suggesting that skull BM-directed local therapy can retain therapeutic efficacy even when classical SLO function is compromised (Extended Data Fig. 11n). However, control splenectomized LTA-deficient mice exhibited shorter survival compared with wild-type (WT) mice (Fig. 5e and Extended Data Fig. 11n), and antigen-specific responses were generally weaker in magnitude than those of WT mice (Fig. 4d–f and Extended Data Fig. 9f–k).

Taken together, these results suggest that skull BM contains early-responding functional lymphoid structures that participate in anti-tumour immunity independently of peripheral SLOs, while the peripheral immune compartments contribute at later stage to amplify and sustain these responses.

In the tumour microenvironment, diverse immune cells, including myeloid cells, B cells, natural killer (NK) cells and CD8⁺ T cells, expressed the Fcγ receptor (FcγR; CD16/32), potentially mediating IgG-driven immune responses (Extended Data Fig. 12a). The triple-therapy group had an increased number of overall intratumoural CD45⁺ immune cells and microglia, while the CD45^high myeloid composition remained largely unchanged (Fig. 5g and Extended Data Fig. 12b).

Triple therapy reduced IL-10 expression of several immune cells such as PBs, PCs and T_reg cells (Extended Data Fig. 12c,d), and markedly enhanced cytotoxicity of NK cells and CD8⁺ T cells, as evidenced by elevated GZMB expression, while changes in the total cell number were minor (Fig. 5h-j and Extended Data Fig. 12e). Moreover, increased polyfunctionality by triple therapy treatment was specifically observed in FcγR⁺CD8⁺ T cells (Extended Data Fig. 12f). FcγR⁺ NK and CD8⁺ T cells also exhibited enhanced GZMB expression (Extended Data Fig. 12g,h), while overall cytokine production of total NK and T cells was comparable across groups (Extended Data Fig. 12i,j), potentially suggesting FcγR-mediated activation. Given the shared BCR repertoire between tumour-infiltrating and skull BM PNA⁺ B cells, these findings suggest that intratumoural immunity is, at least in part, shaped by skull-derived B cell responses, potentially through IgG–FcγR interactions (Extended Data Fig. 12k).

Overall, these data indicate that skull-targeting CD40 agonism combined with administration of the IL-21 and IFNγ supports IgG responses in the skull BM and subsequently enhances intratumoural microglial, NK and CD8⁺ T cell anti-tumour responses while attenuating suppressive immune cells.

Here we found functional lymphoid structures within the skull BM, encompassing T_FH cells, FDCs and GC-like structures. These structures respond to CNS-derived antigens and mount antigen-specific adaptive immune responses in disease-free and brain tumour model context, thereby partaking in CNS immune surveillance.

The skull BM is a primary lymphoid tissue that supplies immune cells, including myeloid cells and immature B cells, to the brain and its border regions^5,6. BM is also known to contain mature immune cells, including memory and regulatory T cells^9,13,19. Although the presence of these cells suggests that BM can support adaptive immune responses against circulating antigens^14,16,18—particularly during inflammation^8,11,15, or when peripheral SLOs are nonfunctional¹⁰—such responses are generally limited under healthy conditions. Here we demonstrate that lymphoid structures, including GCs, are localized within the skull BM, not observed to the same extent in other BM sites. T_FH cells promote constitutive B cell activation through cytokines and co-stimulatory molecules, enabling a sustained immune activity. This activity may be facilitated by continuous exposure to CNS-derived antigens delivered via the CSF through osseous channels^1,2,3,4.

Our OVA–mCherry-tracing experiments showed that brain-derived antigen can reach the skull BM, where it supports local antigen presentation within lymphoid structures. OVA distribution along perivascular spaces and brain borders, together with AQP4 dependency, supports drainage through physiological glymphatic pathways. Selective OVA detection in the skull BM, but not in the sternum BM or spleen, suggests likely preferential use of skull channels^1,2,3,4 rather than systemic dissemination, although the latter cannot be fully excluded. In the dCLNs, OVA was detected using whole-mount imaging but was not robustly detected within immune cells by flow cytometry. This discrepancy may reflect limited internalization, fluorophore loss or degradation during longer-distance drainage. The limited OVA accumulation in the dCLNs compared with previous intracisternal tracer studies may reflect differences between tracer delivery into the CSF and physiological antigen release from the brain parenchyma, as also discussed previously³⁴. Indeed, ref. ³⁴ similarly observed accumulation of neuron-derived proteins in the dCLNs by imaging with limited intracellular detection, and found that their drainage patterns differed from those of injected tracers. Their study proposed a nearest-exit principle whereby CNS-derived proteins preferentially drain toward proximal border compartments. Although striatal antigens preferentially used basal and nasal routes, this does not preclude access to the posterior skull BM, where OVA was directly detected in our study. Moreover, ref. ³⁴ found slower tracer turnover in the skull BM than in other regions, suggesting prolonged antigen retention. Together with pre-existing lymphoid structures enriched in the occipital skull BM, these findings suggest that, in addition to antigen drainage, the local immune environment may further facilitate antigen-specific immune responses.

While our data indicate that skull lymphoid structures are among the earliest responders to brain-derived antigens and contribute substantially to CNS immunosurveillance and anti-tumour immunity, the contribution of other border tissues and peripheral organs cannot be excluded. SLO-deficient mice still permitted antigen-specific responses within the skull BM and retained therapeutic responses to skull-targeting triple therapy. However, these responses were reduced in magnitude, and tumour-bearing SLO-deficient mice exhibited shorter survival compared with WT mice. These data suggest that skull BM provides rapid immune responses, potentially owing to its close anatomical proximity to the CNS, while peripheral immune compartments may contribute to amplification and propagation of the responses. Further studies are needed to define the complex interactions among CNS border tissues and peripheral immune compartments, as well as their relative contributions and functional hierarchy within the neuroimmune axis.

Recent work identified effector memory CD8⁺ T cells in the skull BM of patients with glioblastoma⁸. These cells possess cytotoxic functionality and share TCRs with tumour-infiltrating T cells, suggesting that the skull BM may participate in anti-tumour immune responses in humans. The presence of T_FH-like cells in human skull BM further supports the potential relevance of these findings to human biology; however, additional studies will be required to determine how these mechanisms translate to human patients. In particular, the magnitude and organization of skull lymphoid responses may vary depending on disease contexts, including tumour location, meningeal involvement and patterns of CNS antigen drainage. As shown in our data, a subset of mice exhibited leptomeningeal invasion as tumours progressed. The extent of such involvement may vary depending on tumour type, location and growth kinetics. Although our comparison of shallow tumours, which increase the likelihood of meningeal involvement, and striatal tumours did not reveal differences in overall T and B cell responses across tissues, tumour invasion into the meninges may directly alter border immune microenvironments and antigen drainage pathways, potentially leading to immune responses distinct from those observed in tumours confined to the parenchyma. Defining how meningeal involvement influences antigen delivery and skull-associated responses will therefore be important for understanding the broader applicability of our findings. Future studies across broader pathological settings and human specimens will be also important to define the generalizability and clinical relevance of these responses.

In conclusion, this study identifies functional lymphoid structures in the skull BM that support CNS antigen-specific immune responses under both homeostatic and pathological conditions. Our findings suggest that the skull BM represents an adaptive immune niche at the CNS borders and may provide insights into neuroimmune interactions and inform future therapeutic strategies for brain tumours and other neurological diseases.

Animals

SPF WT C57BL/6J (JAX, 000664) mice were obtained from The Jackson Laboratory. The following strains were also purchased from The Jackson Laboratory: Il21-VFP (B6.Cg-Il21^tm1.1Hm/DcrJ, JAX, 030295)³⁵, Il21r-knockout (B6.129-Il21r^tm1Kopf/J, JAX, 019115)³⁶, OT-II (B6.Cg-Tg(TcraTcrb)425Cbn/J, JAX, 004194)³⁷, MD4 (C57BL/6-Tg(IghelMD4)4Ccg/J, JAX:002595)³⁸, Jchain-creERT2 (B6(129)-Jchain^{em1(icre/ERT2)Deep}/J, JAX, 035764)³⁹, ROSA-DTA (B6.129P2-Gt(ROSA)26Sor^tm1(DTA)Lky/J, JAX, 009669)⁴⁰, Ai9 (B6.Cg-Gt(ROSA)^{26Sortm9(CAG-tdTomato)Hze}/J, JAX, 007909)⁴¹, Thy1.1 (B6.PL-Thy1a/CyJ, JAX, 000406), AID-cre (B6.129P2-Aicda^tm1(cre)Mnz/J, JAX, 007770)⁴², Ai14 (B6.Cg-Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze}/J, JAX, 007914)⁴¹, UBC-GFP (C57BL/6-Tg(UBC-GFP)30Scha/J, JAX, 004353)^43,44, Ifng KO (B6.129S7-Ifng^tm1Ts/J, JAX, 002287)⁴⁵, Lta KO (B6.129S2-Lta^tm1Dch/J, JAX, 002258)⁴⁶, Aqp4 KO (B6(Cg)-Aqp4<tm1.1Tsna>, RIKEN, RBRC10053)⁴⁷ and muMT (B6.129S2-Ighm^tm1Cgn/J, JAX, 002288)⁴⁸ mice. All animals were housed in a temperature-controlled (22 °C) and humidity-controlled (33–39%) environment under a 12 h–12 h light–dark cycle and had ad libitum access to food and water.

Gnotobiotic mice were housed in plastic flexible film isolators in the gnotobiotic facility, also under a 12 h–12 h light–dark cycle. These mice were weaned onto autoclaved food, with ad libitum access to food and water. GF mice were fed γ-irradiated AAD (Teklad custom diet, TD.130916) for 4 weeks to establish AAD-fed GF cohort. Unless otherwise specified, 8-week-old male mice were used for experiments. All procedures were approved by the Institutional Animal Care and Use Committee of Washington University in St Louis (23-0145).

Cell lines and AAVs

The mouse brain cancer cell line GL261 was provided by G. Dunn⁴⁹. The mouse glioma cell line CT2A was purchased from Sigma-Aldrich (SCC194). To generate HEL-OVA-expressing CT2A cell lines, full-length OVA DNA was obtained from Addgene (64599). The DNA encoding HEL-linker-OVA-IRES-eGFP was inserted into pLX208 vectors, and the cell lines were transduced with lentiviruses. Cells were selected with 100 µg ml⁻¹ hygromycin (Invitrogen) and maintained with 50 µg ml⁻¹ hygromycin supplementation.

Cell lines were passaged using trypsin/EDTA (Gibco, 25300) and maintained in Dulbecco’s modified Eagle’s medium (DMEM, Corning, 10-013-CM) supplemented with 10% FBS (Gibco, 26140) and 1% penicillin–streptomycin (Gibco, 15140). All cell lines were confirmed to be free of mycoplasma contamination. Cell lines were authenticated by short-tandem-repeat analysis.

pENN-AAV-hSyn-cre-WPRE-hGH (Addgene, 105553-AAV9), rAAV2/2-hSyn-DIO-eGFP-WPRE-hGH, rAAV2/2-hSyn-DIO-OVA-Flag-2A-eGFP-PA, rAAV2/9-hSyn-DIO-OVA-HEL-2A-eGFP-PA (Biohippo), AAV2-hSyn-OVA-P2A-eGFP-WPRE3-SV40polyA, and AAV2-hSyn-OVA-mCherry-P2A-eGFP-WPRE3-SV40polyA (AAVnerGene) were used.

Tumour and AAV injections

For intracranial and intrahindbrain injection, mice were anaesthetized and secured in a stereotaxic instrument (Kopf) using isoflurane (5% for induction, 1–2% for maintenance) or an intraperitoneal (i.p.) injection of a ketamine (100 mg per kg) and xylazine (10 mg per kg) cocktail. Ophthalmic solution was applied to the eyes to prevent dehydration.

For intracranial AAV injection, after a small craniotomy, 0.5 μl of 1 × 10¹² viral genomes per ml of AAVs diluted in Dulbecco’s phosphate-buffered saline (DPBS; Gibco, 14190) was injected into the striatum (coordinates: anteroposterior (AP), 1.5 mm; mediolateral (ML), 1.5 mm; dorsoventral (DV), −2.5 mm) using a Nanoliter 2020 injector (World Precision Instruments) with a fine glass capillary at a rate of 100 nl min⁻¹.

For the intracranial tumour injection model, 1 × 10⁵ cancer cells in 2 μl of DPBS were injected into the brain (coordinates: AP, 2 mm; ML, 2 mm; DV, −3 mm) using a 26 G syringe (Hamilton, 80300) and an injector (KD Scientific)^50,51. After intracranial injection, the burr hole was sealed with bone wax (Surgical Specialities Corporation, 901). For intrahindbrain injection, the posterior scalp and neck of anaesthetized mice were shaved. The head was secured in a stereotaxic frame with the neck tilted. A midline incision was made, and the posterior nuchal musculature was divided to expose the occipital bone and the dura overlying the cisterna magna. In total, 1 × 10⁵ cancer cells suspended in 2 μl of DPBS were injected into the hindbrain with a 26 G syringe through the cisterna magna. For peripheral tumour injection, anaesthetized mice were shaved, and 5 × 10⁵ cancer cells in 100 μl of DPBS were injected s.c. into the flank.

Injections

For tamoxifen-induced Cre recombination, recipient mice were injected i.p. with 0.1 mg per g of tamoxifen (Sigma-Aldrich, T5648) dissolved in corn oil per day for five consecutive days. For T cell depletion, mice were treated with 200 μg of anti-CD4 (GK1.5, BioXcell, BE0003-1), anti-CD8a (2.43, BioXcell, BE0061) or rat IgG2b isotype control antibody (LTF-2, BioXcell, BE0090). For GC enhancement, 100 μg (i.p.) or 20 μg (s.c.) of anti-CD40 antibody (FGK4.5, BioXcell, BE0016) or rat IgG2a isotype control antibody (2A3, BioXcell, BE0089) was administered, with or without 10 μg (i.p.) or 2 μg (s.c.) of recombinant IFNγ (BioLegend, 575308) and 1 μg (i.p.) or 0.2 μg (s.c.) of recombinant IL-21 (BioLegend, 574506), either i.p. or transcranially. For GC blockade, 200 μg (i.p.) or 20 μg (s.c.) of anti-CD40L antibody (MR-1, BioXcell, BE0017-1) or polyclonal Armenian hamster IgG control antibody (BioXcell, BE0091) was injected i.p. or transcranially. For transcranial injections, materials were mixed with 10 mg ml⁻¹ of carboxymethylcellulose hydrogel (Sigma-Aldrich) in a total volume of 300 μl and injected s.c. under the scalp. For peripheral injections, anti-CD40L antibody was mixed with hydrogel and injected s.c. into the flank.

For intracranial OVA injections, 0.5 μl of 1 mg ml⁻¹ OVA-647 (Invitrogen) in DPBS was injected into the striatum in 1 min (coordinates: AP, 1.5 mm; ML, 1.5 mm; DV, −2.5 mm). For intracisternal OVA injections, 5 μl of 1 mg ml⁻¹ OVA-647 in DPBS was injected into the cisterna magna as described previously²⁷.

Splenectomy

Lta-KO mice were anaesthetized by i.p. injection of ketamine (40–80 mg per kg) and xylazine (5–10 mg per kg). The skin over the left abdomen was sterilized with iodine solution, and a 0.5 cm incision was made with sterile scalpel. The spleen was exposed using smooth forceps, and vessels and attachments were ligated with absorbable 5-0 sutures before excision of the spleen. The midline fascial defect was closed with 4-0 absorbable sutures, and the skin was closed with interrupted 4-0 non-absorbable sutures.

dCLN ligation and removal

dCLN ligation was performed as previously described^29,52. In brief, mice were anaesthetized with ketamine–xylazine cocktail. An incision was made 5 mm above the clavicle. The sternocleidomastoid muscles were retracted, and afferent lymphatic vessels were ligated with a nylon suture. Sham-surgery group received the incision and muscle retraction but were not ligated. For lymphadenectomy, dCLNs were removed during this process. Mice were sutured and recovered on a heating pad until responsive. After surgery, mice were given analgesic and prophylactic antibiotics.

Adoptive transfer

Spleens from donor mice were collected, and primary naive or whole CD4⁺ T cells and B cells were isolated using the EasySep Mouse Naive CD4⁺ T Cell Isolation Kit (StemCell Technologies, 19765), mouse CD4⁺ T cell isolation kit (StemCell Technologies, 19852) and the Mouse B Cell Isolation Kit (StemCell Technologies, 19854), respectively. A total of 2–5 × 10⁵ naive CD4⁺ T cells and 0.5–1 × 10⁶ B cells were resuspended in DPBS and intravenously injected through the retro-orbital route.

Single-cell isolation and flow cytometry

For i.v. immune labelling, mice were anaesthetized and received 7.5 μg of CD45-BV750 (30-F11, BD Biosciences) intravenously through the retro-orbital route. Then, 3–5 min later, mice were euthanized. Mice were euthanized using a CO₂ chamber and transcardially perfused with DPBS. Single-cell suspensions were prepared from each organ. Bones were collected and cleaned to remove attached soft tissues, and the dura was peeled from the skull. Single-cell suspensions from BM were obtained as previously described^4,5. In brief, bones were mechanically dissociated by chopping with scissors in Roswell Park Memorial Institute (RPMI) 1640 medium (Gibco, 11875) containing 2% FBS.

Dura, chopped LNs, brain and brain tumour tissues were digested with a solution containing 1 mg ml⁻¹ collagenase VIII (Sigma-Aldrich, C2139) and 0.5 mg ml⁻¹ DNase I (Thermo Fisher Scientific, EN0521) at 37 °C for 30 min. Whole spleens or single-cell suspensions from various organs were passed through 70-μm strainers. Cell suspensions from brain and tumour tissues were centrifuged with 30% Percoll (Sigma-Aldrich, 17-0891-01) to remove myelin debris. Red blood cells were removed using ammonium–chloride–potassium (ACK) lysis buffer (Gibco, A1049201).

Single-cell suspensions were treated with anti-CD16/32 antibody (93, BioLegend) to block Fc receptors, then stained with antibodies targeting the following molecules: CD45 (30-F11, BioLegend), CD3 (17A2, BioLegend), CD4 (GK1.5, BD Biosciences), CD8a (53-6.7, BioLegend), CD11b (M1/70, BD Biosciences), CD11c (N418, BioLegend), I-A/I-E (M5/114.15.2, BioLegend), NK1.1 (PK136, Thermo Fisher Scientific), CD19 (1D3, BD Biosciences), CD45R/B220 (RA3-6B2, BD Biosciences), IgD (11-26c.2a, BioLegend), IgM (II/41, Thermo Fisher Scientific), IgA (mA-6E1, Thermo Fisher Scientific), IgG (Poly4053, BioLegend), CD138 (281-2, BD Biosciences), TACI (8F10, BD Biosciences), CD38 (90, BioLegend), FAS (SA367H8, BioLegend), CXCR5 (L138D7, BioLegend), CXCR3 (CXCR3-173, BioLegend), CD154 (MR1, BD Biosciences), CXCR4 (L276F12, BioLegend), ICOS (7E.17G9, Thermo Fisher Scientific), PD-1 (RMP1-30, BD Biosciences), CD44 (IM7, BD Biosciences), CD69 (H1.2F3, BD Biosciences), CD62L (MEL-14, BioLegend), IL-21R (4A9, BioLegend), Thy1.2 (53-2.1, BioLegend), IgMa (DS-1, BD Biosciences), IgMb (AF6-78, BD Biosciences), TCR Va2 (B20.1, BioLegend), TCR Vβ5.1/5.2 (MR9-4, BD Biosciences), CD31 (390, BioLegend), PDPN (8.1.1, BioLegend), CR1/2 (7E9, BioLegend), CD16/32 (93, BioLegend), CD25 (PC61, BioLegend), Ly6C (AL-21, BD Biosciences), Ly6G (1A8, BD Biosciences) and CD103 (2E7, BioLegend).

For intracellular staining, cells were fixed and permeabilized using the FoxP3 staining kit (BioLegend) according to the manufacturer’s instructions, and antibodies targeting T-bet (4B10, BioLegend), BCL6 (K112-91, BD Biosciences), IFNγ (XMG1.2, BioLegend), TNF (MP6-XT22, BioLegend), GZMB (NGZB, Thermo Fisher Scientific), IL-10 (JES5-16E3, BD Biosciences) and FOXP3 (FJK-16s, Thermo Fisher Scientific) were used. Live cells were gated using Zombie NIR fixable viability dye (BioLegend). For intracellular cytokine staining, single-cell suspensions were stimulated with PMA/ionomycin (Cell Stimulation Cocktail; 00-4970-93, Thermo Fisher Scientific) in the presence of Protein Transport Inhibitor Cocktail (00-4980-93, Thermo Fisher Scientific) for 4 h at 37 °C. Antibodies were diluted 1:200 for staining. All samples were acquired on an Aurora spectral flow cytometer (Cytek), and data were analysed using FlowJo software v.10.8.1 (TreeStar). The gating strategies and FMO control are shown in Supplementary Figs. 5 and 6.

qPCR with reverse transcription

Tissues were isolated as described above. Total RNA was extracted from the brain and skull using TRIzol (Sigma-Aldrich, T9424), while RNA from dura was isolated using the RNAqueous-Micro Total RNA Isolation Kit (Thermo Fisher Scientific, AM1931) according to the manufacturer’s instructions. Purified RNA was reverse transcribed into cDNA using the iScript Reverse Transcription Supermix for quantitative PCR (qPCR) (Bio-Rad, 1708841). qPCR was performed on the QuantStudio 6 Flex system (Applied Biosystems).

The following primers were used to detect eGFP. Forward: 5′-AAGGGCATCGACTTCAAGG-3′; reverse: 5′-TGCTTGTCGGCCATGATATAG-3′. Transcript levels were normalized to Actb, and relative gene expression was calculated using the comparative ΔΔC_t method.

Immunofluorescence

Mice were perfused with Dulbecco’s phosphate-buffered saline (DPBS) followed by 4% paraformaldehyde (PFA) through intracardiac injection. The samples were then fixed in 4% PFA. Bone samples were decalcified in 5% EDTA solution at 4 °C for 3 days, with daily EDTA changes. For cryosectioning, samples were dehydrated in 30% sucrose solution at 4 °C for 3 days, then embedded in Fisher Healthcare Tissue-Plus OCT compound (Fisher Healthcare, 23-730-571). The blocks were frozen and sectioned at 20–50 μm thickness using a cryostat (Leica Biosystems).

For immunostaining, tissues were blocked in 0.2% Triton X-100 in DPBS with 5% goat or donkey serum, then stained with primary antibodies, including: Armenian hamster anti-mouse CD3e (145, Thermo Fisher Scientific), rabbit anti-mouse CD20 (SP32, Thermo Fisher Scientific), APC-conjugated mouse-anti-mouse IgMa (MA-69, BioLegend), rat anti-mouse BLIMP1 (6D3, Thermo Fisher Scientific), rabbit polyclonal anti-mouse S1PR2 (Proteintech), Alexa Fluor 647-conjugated rat anti-mouse PD-1 (29 F.1A12, BioLegend), Alexa Fluor 594-conjugated rat anti-CD4 (GK1.5, BioLegend), Alexa Fluor 488-conjugated rat anti-mouse CD20 (SA275A11, BioLegend), polyclonal goat anti-mouse CXCL13 (R&D Systems), polyclonal goat anti-mouse BAFF (R&D Systems), CD35 chimeric recombinant rabbit antibody (8C12, Thermo Fisher Scientific), rat anti-mouse FDC (FDC-M1, BD Biosciences), rabbit polyclonal anti-mouse BCL6 (Thermo Fisher Scientific) and eFluor 660-conjugated rat anti-mouse B220 (RA3-6B2, Thermo Fisher Scientific), polyclonal anti-GFP (Thermo Fisher Scientific, A10262). Secondary antibodies included: Alexa Fluor 488-conjugated goat anti-Armenian hamster IgG (Jackson Laboratory, 127-545-099), Alexa Fluor 594-conjugated donkey anti-rabbit IgG (Thermo Fisher Scientific, A32754), Alexa Fluor 647-conjugated donkey anti-goat IgG (Thermo Fisher Scientific, A32849), Alexa Fluor 647-conjugated donkey anti-rat IgG (Thermo Fisher Scientific, A78947), Alexa 488-conjugated donkey anti-rat IgG (Thermo Fisher Scientific, A48269), Alexa Fluor 647-conjugated donkey anti-rabbit IgG (Thermo Fisher Scientific, A32795TR), Alexa Fluor 488-conjugated donkey anti-rabbit IgG (Thermo Fisher Scientific, A-21206).

For in vivo labelling, the following antibodies were injected intravenously 40 min before perfusion: 30–100 μg of Alexa Fluor 488-conjugated anti-mouse CD20 (SA275A11, BioLegend), Alexa Fluor 594-conjugated anti-mouse I-A/I-E (M5/114.15.2, BioLegend), coralite594-conjugated anti-mouse Thy1.2 (30-H12, Thermo Fisher Scientific), Alexa Fluor 594-conjugated anti-mouse CD31 (390, BioLegend) and eFluor 660-conjugated anti-mouse B220 (RA3-6B2, Thermo Fisher Scientific), Alexa Fluor 594-conjugated anti-mouse CD4 (GK1.5, BioLegend) and Alexa Fluor 647-conjugated anti-mouse CD4 (GK1.5, BioLegend). Nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI, Sigma-Aldrich, D9542).

For whole-mount staining, the ADAPT-3D protocol was used for tissue clearing⁵³. Whole-mount staining of the dura was performed as previously described²⁷. Images were acquired using the wide-field microscopy (Olympus SLIDEVIEW VS200) or Stellaris confocal microscope (Leica) with ×10, ×20 or ×40 objectives with Leica Application Suite v.4.2.1.23810 and Nikon Elements v.5.2.0 and analysed with Fiji software v.2.14.0/1.54j. The colocalization plugin in Fiji was used to identify co-localized cells. Cell counting and density calculation were performed with QuPath-0.6.0 software⁵⁴. The intensity and coverage of fluorescent proteins were analysed using Fiji software. The background fluorescence was determined using negative controls, such as non-fluorescent samples or regions. Values were not normalized. Representative single-plane images of whole mounted skulls are provided in Supplementary Fig. 7.

Single-cell transcriptomic analysis

Public data were downloaded from the Gene Expression Omnibus (GEO) under accession numbers GSE184766 (ref. ⁴) and GSE233304 (ref. ⁸). For T and B cell scRNA-seq analysis, single-cell suspensions from the skull and sternum of healthy 8-week-old C57BL/6J male mice were prepared as described above. Single-cell suspensions were treated with anti-CD16/32 (2.4G2) to block Fc receptors. For T cell isolation, cells were stained with PerCP/Cy5.5-conjugated anti-CD45 (30-F11, BioLegend), PE/Cy7-conjugated anti-CD11b (M1/70, BD Biosciences), PE-conjugated anti-Thy1.2 (30-H12, Thermo Fisher Scientific), eFluor 660-conjugated anti-B220 (RA3-6B2, Thermo Fisher Scientific), FITC-conjugated anti-CD4 (RM4-4, BD Biosciences) and Alexa Fluor 488-conjugated CD8a (4SM16, Thermo Fisher Scientific). Dead cells were labelled with DAPI. Live CD45⁺B220⁻CD11b⁻Thy1.2⁺CD4/CD8⁺ cells were sorted. For B cell and ASC isolation, cells were stained with PerCP/Cy5.5-conjugated anti-CD45 (30-F11, BioLegend), FITC-conjugated anti-CD3 (17A2, Thermo Fisher Scientific), Alexa Fluor 700-conjugated anti-CD19 (6D5, BioLegend), APC-conjugated anti-CD138 (281-2, Thermo Fisher Scientific), PE-conjugated anti-TACI (8F10, BD Biosciences) and Pacific Blue-conjugated anti-IgD (11-26 c.2a, BioLegend). Dead cells were labelled with Zombie-NIR viability dye (BioLegend). Live CD45⁺CD3⁻CD138⁻CD19⁺ B cells and live CD45⁺IgD⁻CD3⁻CD138⁺TACI⁺ ASCs were isolated using the FACSAria II (BD Biosciences) system. For GC B cell scRNA-seq, single-cell suspensions from skull and tumour tissues were prepared as described above, 14 days after tumour injection. GC B cells were isolated using the GC B cell (PNA) microbead kit (Miltenyi Biotec). Dead cells were labelled with DAPI, and DAPI⁻ live cells were sorted using the FACSAria III (BD Biosciences) system.

Cells were loaded onto a 10x Genomics Chromium platform for gel beads-in-emulsions. Single Cell 5′ Library & Gel Bead Kit and Mouse B Cell Chromium V(D)J Reagents Kits were used to generate cDNA libraries targeting BCRs and gene expression. Libraries were sequenced using the Illumina NovaSeq 6000 system. Data were aligned to the mm10 genome using the CellRanger software pipeline (10x Genomics; v.1.1.0). Matrices were loaded into Seurat v.4 or v.5 for analysis⁵⁵, and R v.4.4.0 was used for statistical analyses. Cells with unique feature counts <200 or >6,000 (for T cell data), <200 or >7,500 (for B cell data), and <200 or >8,000 (for GC B cell data) with >20% (for T and B cell data) and >5% (skull) or >10% (tumour tissue) (for GC B cell data) mitochondrial gene expression were excluded for quality control. Doublets were also removed. Filtered and normalized data were analysed with the Seurat pipeline. Objects were merged and principal component analysis (PCA) and UMAP analysis were conducted using an elbow plot. Shared nearest neighbour (SNN) clustering was optimized with the Louvain algorithm in the FindClusters function. V(D)J receptor annotation data were generated with the CellRanger vdj command. Filtered contig CSV files were imported into R, and productive barcodes were filtered. The scRepertoire pipeline was used for scBCR/RNA-seq data analysis⁵⁶. Cluster markers were identified using the FindMarkers function, and cluster identities were determined manually. The Nebulosa package was used to visualize gene expression density⁵⁷. Somatic hypermutation frequencies were calculated using BCR sequence data processed with the Immcantation pipeline (https://immcantation.readthedocs.io/en/stable/)^58,59 and the IgBLAST and IMGT germline sequence databases. The Change-O package was used to process V(D)J annotations in BCR data files, and clonal family sizes were calculated using the Alakazam package. BCR mutation frequencies of IGHV were calculated using the SHazaM package.

OVA measurement

AAV-eGFP-, AAV-OVA-infected brains were chopped, homogenized with 70 μm strainer. After centrifugation, supernatants were collected and stored at −80 °C. CSF from AAV-eGFP- or AAV-HEL-OVA-infected mice was collected as described before⁵². OVA was measured by enzyme-linked immunosorbent assay (ELISA) using the sandwich OVA ELISA kit (LS Bio, LS-F9540-1) according to the manufacturer’s instructions. BioTek Gen5 software v.3.11 (Agilent) was used.

Statistical analysis

All data are presented as the mean ± s.e.m. Differences between two experimental groups were analysed using a two-tailed paired or unpaired Student’s t-test and Mann–Whitney U-test. For comparisons involving more than two groups, one-way ANOVA with Tukey’s multiple comparison test was used. For paired samples, Dunnett’s correction was applied. Two-way ANOVA with Sidak’s correction was also used, with paired comparisons used when samples were paired. Survival data were analysed using the log-rank test, and if multiple comparisons were required, the Holm–Sidak method was used for correction. All data are representative of results from at least two independent experiments. scBCR/RNA-seq data were analysed and visualized using R software (v.4.4.0) with RStudio, and all other data were analysed and plotted with GraphPad Prism (v.10.3.1).

Reporting summary

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

1. Herisson, F. et al. Direct vascular channels connect skull bone marrow and the brain surface enabling myeloid cell migration. Nat. Neurosci. 21, 1209–1217 (2018).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

2. Cai, R. et al. Panoptic imaging of transparent mice reveals whole-body neuronal projections and skull-meninges connections. Nat. Neurosci. 22, 317–327 (2019).

   Article  CAS  PubMed  Google Scholar 

3. Yao, H. et al. Leukaemia hijacks a neural mechanism to invade the central nervous system. Nature 560, 55–60 (2018).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

4. Mazzitelli, J. A. et al. Cerebrospinal fluid regulates skull bone marrow niches via direct access through dural channels. Nat. Neurosci. 25, 555–560 (2022).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

5. Cugurra, A. et al. Skull and vertebral bone marrow are myeloid cell reservoirs for the meninges and CNS parenchyma. Science https://doi.org/10.1126/science.abf7844 (2021).

6. Brioschi, S. et al. Heterogeneity of meningeal B cells reveals a lymphopoietic niche at the CNS borders. Science https://doi.org/10.1126/science.abf9277 (2021).

7. Ringstad, G. & Eide, P. K. Molecular trans-dural efflux to skull bone marrow in humans with CSF disorders. Brain 145, 1464–1472 (2022).

   Article  PubMed  PubMed Central  Google Scholar 

8. Dobersalske, C. et al. Cranioencephalic functional lymphoid units in glioblastoma. Nat. Med. https://doi.org/10.1038/s41591-024-03152-x (2024).

9. Cooper, E. et al. Childhood brain tumors instruct cranial hematopoiesis and immunotolerance. Nat. Genet. 58, 317–328 (2026).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

10. Tripp, R. A., Topham, D. J., Watson, S. R. & Doherty, P. C. Bone marrow can function as a lymphoid organ during a primary immune response under conditions of disrupted lymphocyte trafficking. J. Immunol. 158, 3716–3720 (1997).

    Article  CAS  PubMed  Google Scholar 

11. Agbay, R. et al. Characteristics and clinical implications of reactive germinal centers in the bone marrow. Hum. Pathol. 68, 7–21 (2017).

    Article  CAS  PubMed  Google Scholar 

12. Dilosa, R. M., Maeda, K., Masuda, A., Szakal, A. K. & Tew, J. G. Germinal center B cells and antibody production in the bone marrow. J. Immunol. 146, 4071–4077 (1991).

    Article  CAS  PubMed  Google Scholar 

13. Sapoznikov, A. et al. Perivascular clusters of dendritic cells provide critical survival signals to B cells in bone marrow niches. Nat. Immunol. 9, 388–395 (2008).

    Article  CAS  PubMed  Google Scholar 

14. Cariappa, A. et al. Perisinusoidal B cells in the bone marrow participate in T-independent responses to blood-borne microbes. Immunity 23, 397–407 (2005).

    Article  CAS  PubMed  Google Scholar 

15. Siracusa, F. et al. Nonfollicular reactivation of bone marrow resident memory CD4 T cells in immune clusters of the bone marrow. Proc. Natl Acad. Sci. USA 115, 1334–1339 (2018).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

16. Feuerer, M. et al. Bone marrow as a priming site for T-cell responses to blood-borne antigen. Nat. Med. 9, 1151–1157 (2003).

    Article  CAS  PubMed  Google Scholar 

17. Zhao, E. et al. Bone marrow and the control of immunity. Cell. Mol. Immunol. 9, 11–19 (2012).

    Article  CAS  PubMed  Google Scholar 

18. Cavanagh, L. L. et al. Activation of bone marrow-resident memory T cells by circulating, antigen-bearing dendritic cells. Nat. Immunol. 6, 1029–1037 (2005).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

19. Mazo, I. B. et al. Bone marrow is a major reservoir and site of recruitment for central memory CD8⁺ T cells. Immunity 22, 259–270 (2005).

    Article  CAS  PubMed  Google Scholar 

20. Song, W. & Craft, J. T follicular helper cell heterogeneity. Annu. Rev. Immunol. 42, 127–152 (2024).

    Article  CAS  PubMed  Google Scholar 

21. Akkaya, M., Kwak, K. & Pierce, S. K. B cell memory: building two walls of protection against pathogens. Nat. Rev. Immunol. 20, 229–238 (2020).

    Article  CAS  PubMed  Google Scholar 

22. Belanger, S. & Crotty, S. Dances with cytokines, featuring T_FH cells, IL-21, IL-4 and B cells. Nat. Immunol. 17, 1135–1136 (2016).

    Article  CAS  PubMed  Google Scholar 

23. Randall, T. D. & Mebius, R. E. The development and function of mucosal lymphoid tissues: a balancing act with micro-organisms. Mucosal Immunol. 7, 455–466 (2014).

    Article  CAS  PubMed  Google Scholar 

24. Wilmore, J. R. et al. Commensal microbes induce serum IgA responses that protect against polymicrobial sepsis. Cell Host Microbe 23, 302–311 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

25. Fitzpatrick, Z. et al. Venous-plexus-associated lymphoid hubs support meningeal humoral immunity. Nature https://doi.org/10.1038/s41586-024-07202-9 (2024).

26. Park, J. H., Alexander, J. F., Smyth, L. C. D. & Kipnis, J. DALT: the brain’s border patrol. Cell Res. 34, 603–604 (2024).

    Article  PubMed  PubMed Central  Google Scholar 

27. Rustenhoven, J. et al. Functional characterization of the dural sinuses as a neuroimmune interface. Cell 184, 1000–1016 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

28. Cohen, M. et al. Meningeal lymphoid structures are activated under acute and chronic spinal cord pathologies. Life Sci. Alliance https://doi.org/10.26508/lsa.202000907 (2021).

29. Louveau, A. et al. Structural and functional features of central nervous system lymphatic vessels. Nature 523, 337–341 (2015).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

30. Iliff, J. J. et al. A paravascular pathway facilitates CSF flow through the brain parenchyma and the clearance of interstitial solutes, including amyloid beta. Sci. Transl. Med. 4, 147ra111 (2012).

    Article  PubMed  PubMed Central  Google Scholar 

31. Lee-Chang, C. et al. Myeloid-derived suppressive cells promote B cell-mediated immunosuppression via transfer of PD-L1 in glioblastoma. Cancer Immunol. Res. 7, 1928–1943 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

32. Lee-Chang, C. et al. Activation of 4-1BBL⁺ B cells with CD40 agonism and IFNγ elicits potent immunity against glioblastoma. J. Exp. Med. https://doi.org/10.1084/jem.20200913 (2021).

33. van Hooren, L. et al. Agonistic CD40 therapy induces tertiary lymphoid structures but impairs responses to checkpoint blockade in glioma. Nat. Commun. 12, 4127 (2021).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

34. Chayama, Y. et al. Physiological brain clearance architecture revealed by neuronal protein tracing. Cell https://doi.org/10.1016/j.cell.2026.04.048 (2026).

35. Marnik, E. A. et al. Precocious interleukin 21 expression in naive mice identifies a natural helper cell population in autoimmune disease. Cell Rep. 21, 208–221 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

36. Frohlich, A. et al. IL-21 receptor signaling is integral to the development of Th2 effector responses in vivo. Blood 109, 2023–2031 (2007).

    Article  CAS  PubMed  Google Scholar 

37. Barnden, M. J., Allison, J., Heath, W. R. & Carbone, F. R. Defective TCR expression in transgenic mice constructed using cDNA-based α- and β-chain genes under the control of heterologous regulatory elements. Immunol. Cell Biol. 76, 34–40 (1998).

    Article  CAS  PubMed  Google Scholar 

38. Mason, D. Y., Jones, M. & Goodnow, C. C. Development and follicular localization of tolerant B lymphocytes in lysozyme/anti-lysozyme IgM/IgD transgenic mice. Int. Immunol. 4, 163–175 (1992).

    Article  CAS  PubMed  Google Scholar 

39. Wong, R. et al. Affinity-restricted memory B cells dominate recall responses to heterologous flaviviruses. Immunity 53, 1078–1094 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

40. Voehringer, D., Liang, H. E. & Locksley, R. M. Homeostasis and effector function of lymphopenia-induced “memory-like” T cells in constitutively T cell-depleted mice. J. Immunol. 180, 4742–4753 (2008).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

41. Madisen, L. et al. A robust and high-throughput Cre reporting and characterization system for the whole mouse brain. Nat. Neurosci. 13, 133–140 (2010).

    Article  CAS  PubMed  Google Scholar 

42. Robbiani, D. F. et al. AID is required for the chromosomal breaks in c-myc that lead to c-myc/IgH translocations. Cell 135, 1028–1038 (2008).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

43. Schaefer, B. C., Schaefer, M. L., Kappler, J. W., Marrack, P. & Kedl, R. M. Observation of antigen-dependent CD8⁺ T-cell/dendritic cell interactions in vivo. Cell Immunol. 214, 110–122 (2001).

    Article  CAS  PubMed  Google Scholar 

44. Liu, S., Lockhart, J. R., Fontenard, S., Berlett, M. & Ryan, T. M. Mapping the chromosomal insertion site of the GFP transgene of UBC-GFP mice to the MHC locus. J. Immunol. 204, 1982–1987 (2020).

    Article  CAS  PubMed  Google Scholar 

45. Dalton, D. K. et al. Multiple defects of immune cell function in mice with disrupted interferon-γ genes. Science 259, 1739–1742 (1993).

    Article  ADS  CAS  PubMed  Google Scholar 

46. De Togni, P. et al. Abnormal development of peripheral lymphoid organs in mice deficient in lymphotoxin. Science 264, 703–707 (1994).

    Article  ADS  PubMed  Google Scholar 

47. Kitaura, H. et al. Activity-dependent glial swelling is impaired in aquaporin-4 knockout mice. Neurosci. Res. 64, 208–212 (2009).

    Article  CAS  PubMed  Google Scholar 

48. Kitamura, D., Roes, J., Kuhn, R. & Rajewsky, K. A B cell-deficient mouse by targeted disruption of the membrane exon of the immunoglobulin μ chain gene. Nature 350, 423–426 (1991).

    Article  ADS  CAS  PubMed  Google Scholar 

49. Bowman-Kirigin, J. A. et al. The conventional dendritic cell 1 subset primes CD8⁺ T cells and traffics tumor antigen to drive antitumor immunity in the brain. Cancer Immunol. Res. 11, 20–37 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

50. Park, J. H. et al. Tumor hypoxia represses γδ T cell-mediated antitumor immunity against brain tumors. Nat. Immunol. 22, 336–346 (2021).

    Article  CAS  PubMed  Google Scholar 

51. Park, J. H. et al. Protocol to analyze antitumor immunity of orthotopic injection and spontaneous murine high-grade glioma models using flow cytometry and single-cell RNA sequencing. STAR Protoc. 3, 101607 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

52. Kim, K. et al. Meningeal lymphatics-microglia axis regulates synaptic physiology. Cell 188, 2705–2719 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

53. Lee, D. D. et al. ADAPT-3D:accelerated deep adaptable processing of tissue for 3-dimensional fluorescence tissue imaging for research and clinical settings. Sci. Rep. 15, 31841 (2025).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

54. Bankhead, P. et al. QuPath: open source software for digital pathology image analysis. Sci. Rep. 7, 16878 (2017).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

55. Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

56. Borcherding, N., Bormann, N. L. & Kraus, G. scRepertoire: an R-based toolkit for single-cell immune receptor analysis. F1000Res 9, 47 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

57. Alquicira-Hernandez, J. & Powell, J. E. Nebulosa recovers single-cell gene expression signals by kernel density estimation. Bioinformatics 37, 2485–2487 (2021).

    Article  CAS  PubMed  Google Scholar 

58. Vander Heiden, J. A. et al. pRESTO: a toolkit for processing high-throughput sequencing raw reads of lymphocyte receptor repertoires. Bioinformatics 30, 1930–1932 (2014).

    Article  Google Scholar 

59. Gupta, N. T. et al. Change-O: a toolkit for analyzing large-scale B cell immunoglobulin repertoire sequencing data. Bioinformatics 31, 3356–3358 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

Download references

Authors and Affiliations

1. Brain Immunology and Glia (BIG) Center, Washington University in St Louis, St Louis, MO, USA

   Jang Hyun Park, Daviti Abramishvili, Gustavo Gastão Davanzo, Ruben Silva, Xingxing Gu, Siling Du, Igor Smirnov & Jonathan Kipnis

2. Department of Pathology and Immunology, School of Medicine, Washington University in St Louis, St Louis, MO, USA

   Jang Hyun Park, Daviti Abramishvili, Gustavo Gastão Davanzo, Ruben Silva, Xingxing Gu, Siling Du, Daniel D. Lee, Bernd H. Zinselmeyer, Jackson S. Turner, Gwendalyn J. Randolph, Igor Smirnov & Jonathan Kipnis

Contributions

J.H.P. conceptualized, designed and performed the experiments, analysed and interpreted the data, generated figures, wrote and edited the manuscript. D.A. and R.S. assisted in performing experiments. G.G.D. assisted in performing experiments, performed scRNA-seq analysis of T cells. X.G. generated engineered cell lines. S.D. assisted in preparing samples for scRNA-seq. D.D.L., B.H.Z. and G.J.R. helped with clearance methods and microscopy. J.S.T. guided the design of experiments related to GC-like B cells. I.S. performed animal surgeries. J.K. conceptualized, provided resources, guided the analysis and interpretation of the generated data, and wrote and edited the manuscript.

Corresponding authors

Correspondence to Jang Hyun Park or Jonathan Kipnis.

Competing interests

J.K. is a co-founder of Pranas Neuro and Rho Bio.

Peer review information

Nature thanks the anonymous reviewer(s) for their contribution to the peer review of this work.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions