CXCL13⁺ CD4⁺ T cells are associated with B-cell recruitment and lesion progression in cerebral cavernous malformations

Abstract

Cerebral cavernous malformations (CCMs) are neurovascular abnormalities associated with lifelong risks of intracranial hemorrhage (ICH), seizures, and neurologic deficits, affecting approximately 0.16%~0.8% of the general population. Although immune cell infiltration is well recognized as promoting CCM progression, the mechanisms underlying their recruitment into the lesion microenvironment remain poorly understood. In our study, we employed single-cell RNA sequencing (scRNA-seq) to identify a significant enrichment of a CD4⁺ T-cell subset producing CXCL13 and decipher the cellular crosstalk within CCMs. The CXCL13-producing CD4⁺ T cells exhibited enhanced interaction with CXCR5⁺ memory B cells, which showed upregulation of genes related to activation, persistence, and targeted migration. Immunohistochemical staining supported the spatial co-localization between CXCL13⁺ CD4⁺ T cells and CXCR5⁺ CD20⁺ B cells, as well as a positive association between their abundances, reinforcing this cellular crosstalk and suggesting a potential role for CXCL13 in shaping the inflammatory microenvironment and local adaptive immune responses. Clinically, elevated CXCL13 expression in CCM lesions was associated with a higher frequency of subacute ICH and B-cell infiltration in patients. Finally, in vivo experiments showed that administration of recombinant CXCL13 increased B-cell infiltration and Prussian blue staining-defined lesion bleeding, whereas blockade of CXCL13 attenuated CCM lesion development. Thus, our study identified a potentially pathogenic CXCL13⁺ CD4⁺ T-cell subset in CCMs, providing a rationale for targeting CXCL13 signaling in CCMs.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12974-026-03950-9.

Introduction

Cerebral cavernous malformations (CCMs) are neurovascular disorders defined by tightly packed, dilated, and permeable blood vessels within the venous-capillary network [1–3]. These malformations affect approximately 0.16%~0.8% of the general population and can result in various neurological symptoms, including intracerebral hemorrhage (ICH), seizures, headaches, and focal neurological deficits [4–6]. CCMs have a well-characterized genetic origin and can manifest in either an inherited or sporadic form [1]. Genomic studies have established that loss-of-function mutations in CCM1/2/3, along with gain-of-function mutations in MAP3K3 and PIK3CA, play pivotal roles in CCM pathogenesis [7–10].

Beyond genetic drivers, accumulating evidence highlights the critical role of immune-related factors in CCM pathogenesis [11–16]. Immune cell infiltration, including T cells, B cells, and monocytes/macrophages, has been consistently observed within the CCM microenvironment [17, 18]. Previous studies have shown that B-cell infiltration and intralesional B-cell responses are associated with CCM progression: In CCMs, B cells have been characterized by in situ clonal expansion, affinity maturation, and differentiation into antibody-producing plasma cells [19–21]. Additionally, B-cell depletion has been shown to inhibit lesion development in CCM mouse models [22]. However, the mechanisms orchestrating B-cell recruitment and accumulation in CCMs remain unclear.

CXCL13 is a key immune mediator in orchestrating immune responses by recruiting CXCR5⁺ B and T cells to sites of chronic inflammation, thereby facilitating adaptive immunity [23–25]. Beyond its role in adaptive immunity, CXCL13 also modulates innate immune responses by shaping the intensity and composition of the inflammatory milieu [26]. While CXCL13 is critical for the early recruitment of lymphoid tissue inducer cells and the development of secondary lymphoid organs [27, 28], its role in shaping the lesional immune microenvironment and influencing lesion growth and clinical outcomes in CCMs remains poorly understood.

In this study, by analyzing our single-cell RNA sequencing (scRNA-seq) data from human samples, we showed a marked enrichment of T cells within CCM lesions and identified a distinct CXCL13-producing CD4⁺ T-cell subset that may contribute to the recruitment of CXCR5⁺ memory B cells and disease progression. In the CCM patient cohort, higher CXCL13 expression in CCM lesions was associated with increased B-cell infiltration and a greater frequency of subacute ICH, and functional validation in CCM mouse models supported a pathogenic role for CXCL13 in B-cell recruitment and lesion progression. These findings suggest a previously underappreciated immunoregulatory pathway in CCM pathogenesis and provide a rationale for exploring CXCL13 signaling as a potential target for precision immunotherapy.

Materials and methods

Patients and specimens

We collected 27 surgical resection specimens, 76 formalin-fixed paraffin-embedded (FFPE) tissues, and 14 peripheral blood samples from 97 CCM patients who were clinically diagnosed by medical history and neuroradiological examination (CT and/or MRI) (Table S1). Twelve non-lesional control brain specimens were obtained from epilepsy patients undergoing tailored lobectomy surgery. Patients with subacute ICH were defined radiologically on MRI by a T1 hyperintense core and a T2-hyper- or hypointense core with a surrounding hypointense rim [29]. Two board-certified neurosurgeons independently evaluated the MRI images. Concordant assessments were accepted as the final classification. In the event of a disagreement, a third senior neurosurgeon adjudicated the final result. The collection of human samples and clinical data was approved by all patients under written informed consent. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Internal Review and Ethics Committee of the Beijing Tiantan Hospital.

10× Genomics scRNA-seq

Tissue dissociation and cell purification

ScRNA-seq was performed on 10 resected specimens and 2 paired peripheral blood samples from 10 CCM patients, as well as 4 non-lesional control brain specimens from 4 epilepsy patients. Freshly resected CCM lesional biopsies and control brain specimens were washed with 10 mL ice-cold DPBS (Thermo Fisher Scientific, 14190144) to remove the residual tissue storage buffer, and then minced in a sterile culture dish on ice. We used 0.25% Trypsin (Thermo Fisher Scientific, 25200072) and 10 µg/mL DNase I (Sigma, 11284932001) dissolved in DPBS to digest the biopsies. Tissues were dissociated at 37 °C with shaking at 50 rpm for about 40 min. The enzymatic reaction was terminated by adding DMEM (Sigma-Aldrich, D5796) supplemented with 10% FBS (HyClone, SV30087.02). Peripheral blood samples were layered onto human lymphocyte separation medium (DAKEWE, 7111011) and subjected to centrifugation at 400g for 20 min at 4 °C. The leukocyte-containing buffy coat layer was carefully isolated using capillaries and washed twice with DPBS. Cell suspensions were filtered using a 40 μm cell strainer (Corning, 431750), and red blood cells were further removed by 1× red blood cell lysis buffer (Invitrogen, 00-4333-57). Dissociated cells were washed with DPBS containing 2% FBS. Cells were stained with 0.4% trypan blue (Thermo Fisher Scientific, 15250061) to assess viability and quantified using the Countess^® II Automated Cell Counter (Thermo Fisher Scientific, USA). These single-cell suspensions were used for scRNA-seq.

10× Genomics library preparation and sequencing

The scRNA-seq experiment was performed by Shanghai Biotechnology Corporation (Shanghai, China) according to the Chromium Next GEM Single Cell 3’ Reagent Kits v3.1 User Guide (CG000204 Rev D). In detail, the single-cell suspensions were loaded onto a Chromium Single Cell Controller (10× Genomics, USA) to generate Gel Beads-in-emulsion (GEMs). The GEMs were retrieved into a single tube, and reverse transcription was then performed to produce barcoded full-length cDNA from polyadenylated mRNA. Then, the GEMs were broken, and the pooled fractions were purified and subjected to second-strand cDNA synthesis, adaptor ligation, and universal cDNA amplification. Sequencing libraries were prepared using the Chromium Single Cell 3′ GEM, Library & Gel Bead Kit v3.1 according to the manufacturer’s instructions. Library quality was assessed using a High Sensitivity DNA kit (Agilent, 5067 − 4626) on the Bioanalyzer 2100 system (Agilent Technologies, USA) and quantified with the Qubit High Sensitivity DNA Assay kit (Thermo Fisher Scientific, Q32854). The libraries were sequenced on NovaSeq 6000 system (Illumina, USA) using 2 × 150 chemistry.

Data integration, batch effect correction, and unsupervised clustering

We combined the expression matrices of cells from CCM lesions, control brain biopsies, and blood samples. Data normalization (log-transformation) and scaling were performed on the integrated expression matrix using the NormalizeData and ScaleData functions in Seurat (v4.3.0) [30]. Highly variable genes (n = 2,000) were selected with the FindVariableFeatures function (selection.method = ‘vst’). Principal component analysis (PCA) was conducted with RunPCA, and dimensionality reduction was performed with UMAP using the top 20 principal components. We applied the Harmony algorithm [31] prior to downstream clustering to correct for batch effects. Clustering was then run with FindClusters, setting the resolution to 0.5 for CD4⁺ T cells. For each subset, the specific marker genes were identified using the FindAllMarkers function, and clusters were annotated based on marker genes.

Differential expression analysis

Differentially expressed genes (DEGs) between CD4_sub8 cells from CCM lesions and control biopsies were identified using Seurat’s FindMarkers function with parameters logfc.threshold = 0.1 and return.thresh = 0.01. Statistical significance was determined using the non-parametric Wilcoxon rank-sum test, and P was adjusted using the Bonferroni correction. DEGs were visualized using the volcano plot, based on log-transformed expression values.

Tissue enrichment analysis

To quantify the tissue preference of each cluster, we calculated the ratio of observed to expected cell numbers (Ro/e) across tissues, as previously described [32]. Expected cell numbers were estimated using a chi-square test. Clusters with Ro/e > 1 were considered preferentially enriched in the corresponding tissue.

Ligand-receptor signaling network and interaction analysis

To elucidate the functional characteristics of CXCL13⁺ CD4_sub8 cells, we assessed their potential interactions with B-cell subclusters using CellPhoneDB (v3.1.0) [33]. Intercellular communication was inferred from gene expression profiles, with 1,000 permutations to assess statistical significance. Only ligand–receptor pairs with permutation-derived P < 0.05 were visualized using ktplots (https://github.com/zktuong/ktplots).

Flow cytometry

Freshly resected CCM lesions were rinsed with DPBS (Thermo Fisher Scientific, 14190144) three times, cut into small pieces, and transferred to 15 mL of digestion medium containing 1 mg/mL collagenase IV (Sigma-Aldrich, C5138), 100 µg/mL hyaluronidase (Sigma-Aldrich, H2126), 20 U/mL DNase I (Sigma-Aldrich, DN25) in RPMI 1640 medium (Thermo Fisher Scientific, 11875101). Tissues were digested for 30 min at 37 °C under agitation at 200 rpm; afterward, the samples were filtered using a 150 μm nylon mesh. Peripheral blood mononuclear cells (PBMCs) were isolated from anticoagulated blood by human lymphocyte separation medium (DAKEWE, 7111011). Following the centrifugation at 400g for 20 min at 4 °C, the supernatants were discarded, and the cell pellets were resuspended in 3 mL of 1× red blood cell lysis buffer (Invitrogen, 00-4333-57) at 4 °C for 10 min. After washing with DPBS, cell pellets obtained from CCM lesions and blood samples were suspended in RPMI-1640 complete medium (Abbkine, BMM0002) containing Ionomycin (1 µg/mL, MCE, HY-13434), PMA (50 ng/mL, MCE, HY-18739), and GolgiStop (1:1500, BD, 554724). After incubation at 37 °C for 4 h, the cells were analyzed by flow cytometry.

For surface marker staining, cells were washed twice with DPBS containing 2% FBS (HyClone, SH30071.02). After Fc receptor blocking with purified NA/LE human BD Fc block (BD Biosciences, 564765), cells were incubated with the respective antibodies in the presence of Brilliant Stain Buffer (BD Biosciences, 563794) at 4 °C for 30 min. For staining of intracellular cytokines, cells were fixed and permeabilized with a fixation/permeabilization kit (BD Biosciences, 554714) after surface staining according to the manufacturer’s instructions. The permeabilized cells were then incubated with the respective antibodies at 4 °C for 30 min. For staining of transcription factors, cells were fixed and permeabilized with transcription factor staining buffer set (eBioscience, 00-5523-00) after surface staining according to the manufacturer’s instructions. The permeabilized cells were then incubated with the respective antibodies at 4 °C for 30 min. Cell viability was evaluated using fixable viability stain 700 (BD Biosciences, 564997), which can be used to exclude non-viable cells in flow cytometric assays. Flow cytometry was performed on a BD FACSCanto II (BD Biosciences, USA), and data were analyzed using FlowJo software (v10.8.1). Positive gates were set with reference to the corresponding isotype controls. The antibodies used for flow cytometry analysis are listed in Table S2.

Animals

The Animal Welfare and Ethics Committee of Beijing Neurosurgical Institute Laboratory approved all animal ethics and protocols. All mice were housed in individual, ventilated cages under a 12-hour light/dark cycle, with food and water ad libitum. All animals were housed in a pathogen-free vivarium. The Ccm2^fl/fl mice were generated on the C57BL/6J background and created by CRISPR/Cas-mediated genome engineering at Cyagen Biosciences (China). After crossing the above mice with Cdh5-CreERT2 mice (purchased from Cyagen Biosciences), iECre; Ccm2^fl/fl (Ccm2^iECKO) mice were obtained. Both male and female mice were used and were evenly distributed between the experimental and control groups.

For all neonatal mouse model experiments, at one-day post-birth (P1), pups were intragastrically injected with a 30-gauge needle using a freshly prepared 4-hydroxytamoxifen (4-OHT, Sigma-Aldrich, H7904) solution dissolved in a 9% ethanol/corn oil (volume/volume) vehicle (50 µL total volume per injection). For the recombinant mouse CXCL13 (rmCXCL13) treatment experiment, neonatal mice received 25 µg of 4-OHT. For the CXCL13 blocking treatment experiment, neonatal mice received 40 µg of 4-OHT. The CCM lesions were confirmed by a 7.0 T small-animal MRI scanner (Bruker, Germany) prior to treatment. Mice with CCM lesions were randomly assigned to groups for subsequent experiments, and the outcome assessors were blinded to group allocation.

Mouse treatment

For the in vivo experiments, Cdh5-CreERT2; Ccm2^fl/fl mice with successful CCM model induction were randomly assigned to the control and treatment groups. For the experiment of rmCXCL13 treatment, rmCXCL13 (8 µg per mouse; MCE, HY-P77911) protein was dissolved in a total of 60 µL sterile PBS (ThermoFisher, 10010023) and administered intravenously immediately after baseline MRI scanning at 4 weeks of age and twice per week thereafter until the end of the experiment. Negative control animals were simultaneously injected with sterile PBS alone. The rmCXCL13 dose was selected based on previous in vivo studies, which have generally used doses ranging from 1 to 5 µg [26, 34, 35]. In the present study, a slightly higher working dose of 8 µg was used to ensure sufficient activation of CXCL13 signaling within the experimental treatment window. CCM mouse models generally tolerate small-animal MRI at 4 weeks of age, which was used as the baseline assessment time point for CCM lesions. Because rmCXCL13 administration might accelerate lesion progression, we initiated treatment immediately after baseline MRI at 4 weeks of age to minimize the risk of excessively rapid lesion progression and associated mortality.

Although CD4⁺ T cells in murine models do not secrete CXCL13, microglia or macrophages may act as sources of CXCL13 [35–39]. To further evaluate the potential role of CXCL13 in B-cell recruitment and lesion progression in CCMs, we performed a CXCL13-blocking treatment experiment. CXCL13-blocking antibody (20 µg per mouse; R&D Systems, AF470) was dissolved in a total volume of 100 µL sterile PBS and administered intravenously immediately after baseline MRI scanning at 6 weeks of age and twice per week thereafter until the end of the experiment. Negative control animals were simultaneously injected with an isotype control (R&D Systems, AB-108-C). The dose of the CXCL13-blocking antibody was selected based on previous studies, which reported doses spanning a relatively wide range from 0.5 to 100 µg [39, 40]. Therefore, 20 µg was selected as the working dose for CXCL13-blocking antibody treatment in the present study. Since CXCL13-blocking antibody treatment might suppress lesion progression, we initiated treatment at 6 weeks of age, when CCM lesions were already established but continued to progress, thereby providing a suitable therapeutic window to evaluate the effects of CXCL13 blockade. At the indicated time points, treated mice underwent MRI examination and were then euthanized for brain tissue collection.

Magnetic resonance imaging (MRI)

MRI was performed using the 7.0 T small animal MRI scanner (Bruker, Germany) to evaluate the putative CCM lesions with the following parameters: repetition time (TR) = 600 ms, echo time (TE) = 8 ms, field of view (FOV) = 25 × 25 mm², matrix = 256 × 192, slices = 25, and slice thickness = 0.5 mm. The severity of CCM lesions was quantified using the MRI index. The MRI index = total lesion area of 25 images / whole brain area of 25 images [41]. The change in the MRI index was calculated as the difference between the post-treatment and the baseline MRI index values, divided by the baseline MRI index. The lesion area and the whole brain area were quantified using ImageJ (v1.54f).

Histological staining

Mice were anesthetized with intraperitoneal injection of sodium pentobarbital, and the brains were carefully removed from the skull. Tissue samples were fixed in 4% PFA (Thermo Fisher Scientific, J19943.K2) overnight at 4 °C, dehydrated, and embedded in paraffin. Tissue sections of 5 μm were cut from the embedded mouse brain tissues for each staining. Hematoxylin and eosin (HE) staining was performed on a Leica Autostainer XL (Leica, Germany) using an HE staining kit (Solarbio, G1120) according to the manufacturer’s instructions. Prussian blue staining was performed as described by the manufacturer with a Prussian blue iron stain kit (Solarbio, G1422). Images were acquired on a Zeiss Axio Scope A1 microscope (Zeiss, Germany). As previously described, the computational quantification of lesion area and iron deposition was performed using ImageJ (v1.54f) [42, 43].

Immunohistochemistry staining

Immunohistochemistry (IHC) was performed on FFPE sections. Slides were first dewaxed and rehydrated. Heat-induced antigen retrieval was performed with either citrate antigen retrieval buffer, pH 6.0 (ZSGB-BIO, ZLI-9065) or EDTA antigen retrieval buffer, pH 9.0 (ZSGB-BIO, ZLI-9069) for 20 min at 100 °C or 40 min at 92 °C. Primary antibodies were diluted and incubated for 120 min on slides. Slides were washed, and secondary antibodies were incubated for 30 min according to the manufacturer’s instructions. Immunoreactive cells were visualized with DAB chromogen (Genetech, GK347011) followed by a hematoxylin (BASO, BA4097) counterstain. For staining of multiple markers on FFPE sections, the Opal Polaris 7-Color Manual IHC Kit (Akoya Biosciences, NEL861001KT) was used according to the manufacturer’s instructions. Images were acquired on Zeiss Axio Scope A1 microscope (Zeiss, Germany), LSM 880 (Zeiss, Germany), or PhenoImager HT (Akoya Biosciences, USA). Primary antibodies are listed in Table S2.

Evaluation and quantification of immunohistochemical staining

Evaluation and quantification of immunohistochemical staining was performed using InForm (v3.0.0) or QuPath (v0.6.0). To avoid bias and subjective interpretation, IHC results were evaluated independently by two blinded investigators. The sections with inter-observer variation > 10% were reassessed to achieve consensus. Immunohistochemical staining of CXCL13 was scored based on staining intensity on a scale of 0 to 3 (absent = 0, weak = 1, moderate = 2, or strong = 3), and the percentage of stained cells (range, 0-100%) as previously described [26, 44]. CXCL13 IHC score (range, 0-300) = 3 × percentage of strongly staining cells + 2 × percentage of moderately staining cells + 1 × percentage of weakly staining cells. Immunohistochemical staining of cellular infiltrates was counted by examining representative high-power fields (HPFs).

The spatial relationship between CXCR5⁺ CD20⁺ or CXCR5⁻ CD20⁺ cells and CXCL13⁺ CD4⁺ cells was quantified by measuring the distance from each CXCR5⁺ CD20⁺ or CXCR5⁻ CD20⁺ cell to the nearest CXCL13⁺ CD4⁺ cell, as previously described [26, 45]. InForm (v3.0.0) and the phenoptr and phenoptrReports R packages were used for quantitative analysis. First, CXCL13⁺ CD4⁺, CXCR5⁺ CD20⁺, and CXCR5⁻ CD20⁺ cells were automatically identified using InForm. Then, the cell X/Y position coordinates and phenotype annotations were extracted from the InForm cell-segmentation output. The neighbor distance from each CXCR5⁺ CD20⁺ or CXCR5⁻ CD20⁺ cell to the nearest CXCL13⁺ CD4⁺ cell was calculated using the “find_nearest_distance” function in phenoptr on a field-by-field basis; the resulting per-cell distances were then summarized using the “nearest_neighbor_summary” function in phenoptrReports.

Statistical analysis

GraphPad Prism (v.10.1.2) was used for statistical analyses. Continuous variables are presented as mean ± SEM, and categorical variables are presented as counts and percentages. Normality of the data sets was evaluated using the Shapiro-Wilk or Kolmogorov-Smirnov tests. For normally distributed data, comparisons between two groups were performed by unpaired or paired t-tests. For non-normally distributed data, comparisons between two groups were performed using the Mann-Whitney test. Categorical variables were compared using two-sided Fisher’s exact test. For correlation analyses, Pearson correlation tests were used for normally distributed data, and Spearman correlation tests were used for non-normally distributed data. Two-sided P < 0.05 was considered statistically significant and is denoted as follows: ^* P < 0.05; ^** P < 0.01; ^*** P < 0.001; ^**** P < 0.0001.

Results

scRNA-seq identifies transcriptome heterogeneity of T-cell subclusters in human CCMs

To delineate the immune landscape of CCMs, we performed scRNA-seq on surgically resected brain lesions from 10 CCM patients, 2 matched peripheral blood samples, and 4 non-lesional control brain specimens from epilepsy patients (Fig. 1a and S1a). A total of 110,910 single cells passed quality control filters (59,658 cells from CCM lesions and 51,252 cells from control brain biopsies and blood samples). To achieve molecular cell typing, we conducted Louvain clustering [30] followed by major cell-type annotation based on established nomenclature, all of which exhibited canonical marker genes (Fig. 1b-c). Abundance comparisons showed marked differences in T/NK cells and B cells between CCM and control brain biopsies (Fig. 1d), highlighting a unique immune landscape in CCMs and suggesting their potential involvement in disease pathogenesis.

Fig. 1.

Heterogeneity of major cell types and T/NK cells in human CCM lesions and control samples. a Schematic representation of the workflow used to generate scRNA-seq data from CCM lesions (n = 10) as well as control samples, including paired peripheral blood samples (n = 2) and control brain tissues (n = 4). b Uniform manifold approximation and projection (UMAP) visualization of scRNA-seq data with main cell types. c The expression levels of common markers across main cell types. d Cell-type composition of each sample. Dot size reflects the relative proportion of each cell type within a sample, colored as in panel b. Total cell counts per sample are indicated by the bar chart above. B, paired blood sample; E, epilepsy control brain tissue; C, CCM lesion. e UMAP visualization of T/NK-cell subclusters after reclustering. f The expression levels of marker genes across T/NK-cell subclusters. g Heatmap showing the tissue preference of T/NK subclusters based on the ratio of observed to expected cell numbers (Ro/e). ++, Ro/e > 1; +, 0.2 ≤ Ro/e ≤ 1; +/−, 0 < Ro/e < 0.2; −, Ro/e = 0

To better assess the extent of phenotypic and functional heterogeneity of T/NK cells in CCMs, unsupervised clustering was then performed within T/NK cells (CCM lesions: 16,109 cells; control brain biopsies: 4,307 cells; blood samples: 9,632 cells). The fine-grained subpopulations were discerned: 5 CD4⁺ T-cell subclusters, 6 CD8⁺ T-cell subclusters, 2 γδ T-cell subclusters, and 2 NK-cell subclusters (Fig. 1e-f and Fig. S1b). Evaluation of tissue enrichment with Ro/e analysis [32] revealed notable accumulation of 2 subsets specifically within CCM lesions compared to blood samples and control brain biopsies: CD8_sub2, showing MAIT (mucosal-associated invariant T)-like characteristics (KLRB1 and SLC4A10), and CD4_sub5, exhibiting partial Treg features (FOXP3⁺ in 32.5% cells) (Fig. 1f-g). While CD4_sub5 transcriptionally resembled Tregs, its limited FOXP3 expression percentage suggests unresolved heterogeneity in CD4⁺ T cells, highlighting the need for higher-resolution analysis to elucidate its CCM-specific roles.

Enrichment of a CXCL13⁺ CD4⁺ T-cell subcluster in CCMs

We further subdivided CD4⁺ T cells (n = 15,397) to identify lesion-associated subpopulations and resolved 11 distinct subclusters (sub0–sub10, Fig. 2a-b and S2a). The evaluation of tissue enrichment using Ro/e analysis showed that, among the 11 CD4⁺ T-cell subclusters, sub8, characterized by a follicular helper T (T_FH)-like transcriptional profile, and sub6, exhibiting a Treg-like signature, were preferentially enriched in CCM lesions relative to control brain tissues and blood samples (Fig. 2b–c). Differential gene expression analysis (Wilcoxon rank-sum test) between CCM and control samples showed CXCL13 as the most significantly upregulated gene in CCMs in the sub8 cluster of CD4⁺ T cells, ranked by log₂ fold change (Fig. 2d). We also assessed the proportion of CXCL13-expressing cells and CXCL13 expression levels across major cell types in CCMs and found that both were highest in CD4⁺ T cells within the lesional microenvironment (Fig. S2b). In addition, CXCL13 expression showed a lesion-specific pattern in CCMs, with minimal expression in non-lesional control brain tissues and peripheral blood samples (Fig. 2e). Notably, CXCL13-expressing cells were predominantly enriched in the CD4_sub8 cluster among CD4 T-cell subclusters (Fig. S2c). Within the CD4_sub8 cluster, CXCL13 was expressed by 31.5% of cells from CCM lesions, compared with 2.3% of cells from control brain tissues and 0.8% of cells from blood samples (Fig. 2f). Additionally, patient-level assessment showed that the CD4_sub8 cluster was broadly detected across all CCM patients (Fig. S2d). Further stratification of the sub8 cluster into CXCL13⁺ and CXCL13⁻ subsets revealed that CXCL13⁺ cells were highly enriched in CCM lesions (Fig. 2g) and co-expressed elevated levels of markers associated with T-cell activation and co-stimulation, including PDCD1, TNFRSF4, TNFRSF18, and IL21, while these cells showed lower CXCR5 expression (Fig. 2h). The expression levels and proportions of selected genes in the sub8_CXCL13⁺ and sub8_CXCL13⁻ clusters were also provided as numeric values (Fig. S2e). These results indicated enrichment of CXCL13, primarily produced by CD4⁺ T cells exhibiting T_FH-like features, within the CCM microenvironment.

Fig. 2.

CXCL13⁺ CD4⁺ T cells are enriched in CCM lesions. a UMAP plot showing CD4 T-cell subclusters. b Dot plot showing expression of selected marker genes in CD4 T-cell subclusters. c Heatmap showing the tissue preference of CD4⁺ T-cell subclusters based on the ratio of observed to expected cell numbers (Ro/e). d Volcano plot showing differentially expressed genes in CD4_sub8 cells between CCM lesions and control brain samples. e Feature plots showing CXCL13 expression in CD4 T cells across the three sample groups. f Bar plot depicting the proportion of CXCL13⁺ cells within the CD4_sub8 subcluster across sample groups. g Heatmap showing the tissue preference of sub8_CXCL13⁺ and sub8_CXCL13⁻ cells based on Ro/e. h Dot plot showing the expression of selected genes in sub8_CXCL13⁺ and sub8_CXCL13⁻ cells. For Ro/e heatmaps in panels c and g, ++, Ro/e > 1; +, 0.2 ≤ Ro/e ≤ 1; +/−, 0 < Ro/e < 0.2; −, Ro/e = 0

Validation and phenotypic characterization of CXCL13⁺ CD4⁺ T cells in CCM lesions

To verify the presence of CXCL13⁺ CD4⁺ T cells in CCMs, we performed flow cytometry and IHC analysis. Flow cytometric analysis of freshly resected lesions (n = 12) and PBMCs (n = 12) from CCM patients revealed a significantly higher proportion of CXCL13⁺ cells among CD4⁺ T cells in CCM lesions compared to blood samples (P < 0.0001) (Fig. 3a-b and S3a-b). Consistent with this result, IHC on FFPE specimens of CCM lesions (n = 10) detected CXCL13⁺ CD4⁺ T cells, which were absent in control brain tissues (n = 8) from epilepsy patients (P < 0.0001) (Fig. 3c-d).

Fig. 3.

Validation of CXCL13⁺ CD4⁺ T cells in CCM lesions. a Representative dot plots showing the CXCL13 expression within the CD4⁺ gate of CCM lesion-infiltrating lymphocytes and PBMCs of CCM patients. b Statistical analysis of flow cytometry for CXCL13 expression within the CD4⁺ gate. Data are shown as mean ± SEM; n = 12 per group; Wilcoxon rank-sum test; ****P < 0.0001. c Representative IHC staining of CCM lesion and control brain tissue sections showing CLDN5⁺ (green), CD4⁺ (red) and CXCL13⁺ (white), with nuclei stained with DAPI (blue). Scale bar: 40 μm. d Statistical analysis of IHC for the percentage of CXCL13⁺ cells in CD4⁺ T cells. Data are shown as mean ± SEM; n = 10 and n = 8 for CCM lesions and control samples, respectively; Wilcoxon rank-sum test; ****P < 0.0001

Moreover, to further characterize the phenotype of CXCL13-expressing T cells in CCM lesions, we assessed the proportions of CD4⁺ CXCL13⁺ and CD8⁺ CXCL13⁺ cells among CD3⁺ T cells in 17 CCM lesions. The results showed that CXCL13 was detectable in both CD4⁺ and CD8⁺ T cells within CCM lesions. However, among total CD3⁺ T cells, the frequency of CD4⁺CXCL13⁺ T cells was significantly higher than that of CD8⁺CXCL13⁺ T cells (P < 0.0001, Fig. S4a-b), consistent with our scRNA-seq analysis (Fig. S2b) and supporting the conclusion that CD4⁺ T cells represent a major source of CXCL13 within the T-cell compartment. We further evaluated PD-1 expression in CXCL13⁺ CD4⁺ T cells and assessed its co-expression with additional phenotypic markers, including CD25, CCR7, and CXCR5 in 5 CCM lesion specimens. The results showed that PD-1 was highly expressed by CXCL13⁺ CD4⁺ T cells and was readily detected within CXCL13⁺ CD4⁺ T-cell subsets expressing CCR7, CD25, or CXCR5 (Fig. S4c). These findings provide protein-level support for the transcriptional profile identified by scRNA-seq and further support the existence of a CXCL13⁺ PD-1⁺ CD4⁺ T-cell population in CCM lesions. Finally, to determine whether CXCL13 was expressed by Treg cells, we performed flow cytometric analysis of 3 CCM lesion specimens and quantified the proportion of FOXP3⁺ cells among CD4⁺ CXCL13⁺ T cells. This analysis revealed that FOXP3 expression was detected in only a minor subset of CD4⁺ CXCL13⁺ T cells, indicating that CXCL13 expression was predominantly associated with non-Treg CD4⁺ T cells (Fig. S4d).

Cellular crosstalk between CXCL13⁺ CD4⁺ T cells and CXCR5⁺ B cells in CCMs

CXCL13 is a chemokine that specifically binds to the receptor CXCR5, a molecule predominantly expressed on B cells, T cells, and other immune populations [46, 47]. The CXCL13/CXCR5 axis plays an important role in immune cell recruitment and in the regulation of adaptive and innate immune responses [48, 49]. In CCM lesions, we found that CXCR5 was primarily expressed in B cells (Fig. 4a). Therefore, we further categorized B cells (CCM lesions: 4,197 cells; control brain tissues: 545 cells; blood samples: 1,945 cells) and identified 4 transcriptionally distinct subclusters (Fig. 4b–c). Among the B-cell subclusters, the Memory B_sub2 subcluster exhibited particularly high expression of CXCR5 (Fig. 4d-e), and displayed a transcriptional program characteristic of a highly activated and immunologically engaged phenotype (Fig. 4c). This subcluster showed upregulation of canonical activation markers (CD69, CXCR4, NR4A1, NR4A2, and CD83), as well as receptors for survival signals (TNFRSF13C [BAFFR], TNFRSF13B [TACI]), indicating a heightened capacity to receive trophic signals that support persistence and function in the inflamed lesion [50, 51]. The co-expression of GPR183 and CCR7 further suggests that these cells are actively navigating the inflammatory microenvironment, potentially migrating toward the T-B cell border to engage in cognate interactions or participating in the organization of ectopic lymphoid-like niches within the lesion [52, 53]. The high CXCR5 expression on these cells implies responsiveness to CXCL13, which is expressed by a specific CD4⁺ T-cell population within the lesions. In line with this observation, IHC quantification in FFPE sections of 10 CCM lesions showed a significant positive correlation between the abundances of CD4⁺ T cells and B cells within CCMs (P = 0.0121) (Fig. 4f). Moreover, in-depth ligand-receptor interaction analysis suggested enhanced signaling between CXCL13⁺ CD4⁺ T cells and CXCR5⁺ memory B cells specifically in CCM lesions compared to control samples (Fig. 4g). This suggests the identified CXCL13⁺ CD4⁺ T cells can potentially mediate the recruitment of the infiltrating B cells into CCM lesions, thereby fostering an organized inflammatory niche.

Fig. 4.

CXCL13⁺ CD4⁺ T cells are colocalized with CXCR5⁺ B cells in CCMs. a CXCR5 expression across the major cell types. The y-axis indicates the proportion of CXCR5-expressing cells within each major cell type, and the bar color reflects the scaled average expression of CXCR5. b UMAP plot showing B cell clusters. c The expression levels of representative genes across B-cell subclusters. d Expression of CXCR5 among B-cell subclusters. e Feature plots showing CXCR5 expression in B cells across the three sample groups. f Correlation between the numbers of CD4⁺ T cells and B cells in CCM lesions. n = 10; Pearson correlation test; P = 0.0121. g Cell-cell interaction network of CXCL13⁺ CD4_sub8 T cells with B-cell subclusters in CCMs. h Representative spatial relationship map of CXCL13⁺ CD4⁺ T cells (green dots), CXCR5⁺ CD20⁺ B cells (red dots), and CXCR5⁻ CD20⁺ B cells (yellow dots). (i) Distance from CXCR5⁺ CD20⁺ and CXCR5⁻ CD20⁺ B cells to the closest CXCL13⁺ CD4⁺ T cells. 26 HPFs were selected from 5 CCM samples; Paired t-test; P < 0.0001. j Representative IHC staining of CCM lesion sections from the CXCL13⁺ CD4⁺ T-cell-high and -low groups, showing CLDN5⁺ (cyan), CD4⁺ (red), CXCL13⁺ (white), CD20⁺ (green), and CXCR5⁺ (yellow), with nuclei stained with DAPI (blue). Scale bar: 20 μm. k Difference in the number of CXCR5⁺ CD20⁺ cells between the CXCL13⁺ CD4⁺ T-cell-high and -low groups. Data are shown as mean ± SEM; 24 HPFs containing CXCL13⁺ CD4⁺ cells were randomly selected from 5 CCM lesions; Wilcoxon rank-sum test; ***P = 0.0001

We next investigated the spatial interplay between CXCL13⁺ CD4⁺ T cells and infiltrating CXCR5⁺ B cells in CCMs using IHC staining for CXCL13, CD4, CXCR5, and the pan-B cell marker CD20. The result indicated that the distance between CXCR5⁺ CD20⁺ B cells and their nearest CXCL13⁺ CD4⁺ T cell was significantly shorter than that of CXCR5⁻ CD20⁺ B cells (P < 0.0001) (Fig. 4h-i). To further evaluate the functional association between CXCL13⁺ CD4⁺ T cells and CXCR5⁺ CD20⁺ B cells, we randomly selected 24 HPFs containing CXCL13⁺ CD4⁺ T cells from sections of 5 CCM lesions and quantified the numbers of CXCL13⁺ CD4⁺ T cells and CXCR5⁺ CD20⁺ B cells in each HPF. The result showed a significant positive correlation between these two cell populations in human CCM lesions (P = 0.0003) (Fig. S5a-b). Moreover, using the median number of CXCL13⁺ CD4⁺ T cells (21 cells) as the cutoff, the selected HPFs were stratified into the CXCL13⁺ CD4⁺ cell-high group (n = 12) and the CXCL13⁺ CD4⁺ cell-low group (n = 12). We then compared the number of CXCR5⁺ CD20⁺ B cells between the two groups. The results showed that CXCR5⁺ CD20⁺ B cells were significantly more abundant in the CXCL13⁺ CD4⁺ cell-high group than in the CXCL13⁺ CD4⁺ cell-low group (P = 0.0001) (Fig. 4j-k). These results provided histological evidence that the CXCL13/CXCR5 chemotactic axis is associated with B-cell recruitment and positioning within the CCM microenvironment. Considering the important role of infiltrating B cells in CCM pathogenesis [22], these findings suggest that CXCL13-mediated B-cell recruitment may contribute to CCM lesion development.

CXCL13 expression correlates with B-cell infiltration and lesion severity in CCM patients

Previous studies have shown that CXCL13 expression is associated with disease activity in autoimmune disorders [54] and patient prognosis in various cancers [47]. To investigate its clinical relevance in CCMs, we evaluated the association of CXCL13 expression with both B-cell infiltration and lesion severity in a cohort of CCM patients (Fig. 5a-b and Table S3). We first quantified CXCL13 expression levels by IHC staining in 61 CCM FFPE specimens. The CXCL13 IHC score, averaged across 10 HPFs, ranged from 0 to 26.67. Using the median value (3.39) as the cutoff, patients were stratified into low (n = 30) and high (n = 31) CXCL13 expression groups (Fig. 5a).

Fig. 5.

Elevated CXCL13 expression is associated with B-cell infiltration and lesion severity in CCM patients. a Summary of the cohort involving 61 patients evaluated for the association between CXCL13 expression and characteristics of CCM lesions, including lesion size, subacute ICH, and intralesional B-cell infiltration. b Representative CXCL13 and CD20 IHC staining and T2-weighted MRI images from the high and low CXCL13 expression groups. Red arrowheads indicate CCM lesions on MRI images. Scale bar: 50 μm. c Comparison of lesion size between patients in the high CXCL13 expression group (n = 31) and those in the low CXCL13 expression group (n = 30). Data are shown as mean ± SEM; Wilcoxon rank-sum test; ***P = 0.0002. d Correlation between lesion size and CXCL13 expression in a publicly available bulk RNA-seq dataset of 10 CCM samples. Pearson correlation test; P = 0.0005. e Comparison of the proportion of patients with subacute ICH between patients with high CXCL13 expression (n = 31) and low CXCL13 expression (n = 30). Fisher’s exact test; ****P < 0.0001. f Comparison of the proportion of CD20⁺ B cells between the high CXCL13 expression group (n = 29) and the low CXCL13 expression group (n = 28). Data are shown as mean ± SEM; Wilcoxon rank-sum test; ****P < 0.0001. Due to tissue section detachment during staining, IHC evaluation was not feasible for two patients in each of the high- and low-CXCL13 expression groups from the same batch

Next, we evaluated the lesion severity-related features by assessing lesion size (maximum lesion diameter) and the presence of subacute ICH in resected CCM lesions. The high CXCL13 expression group showed a significantly larger lesion size (35.60 ± 1.98   mm) than the low-expression group (23.97 ± 1.93  mm, P = 0.0002) (Fig. 5b-c). This positive correlation between CXCL13 expression and lesion size was independently validated using a publicly available bulk RNA-seq dataset of 10 CCM samples (GSE137596) [55] (P = 0.0005; Fig. 5d). Additionally, the proportion of patients with subacute ICH in the high CXCL13 expression group (25/31, 80.6%) was significantly higher than that in low CXCL13 expression group (9/30, 30.0%; P < 0.0001) (Fig. 5b and e). Moreover, we examined the correlation between CXCL13 expression and B-cell infiltration by performing CD20 IHC staining on the serial FFPE sections adjacent to the CXCL13-stained ones. Cases with high CXCL13 expression exhibited significantly greater infiltration of CD20⁺ B cells compared to low CXCL13 expression cases (P < 0.0001) (Fig. 5b and f). We also assessed whether age and sex were associated with the lesional immune microenvironment. A previous study suggested that the immune response within CCM lesions may vary with age, with lesions from younger subjects showing greater T-cell infiltration and lower macrophage infiltration [17]. However, analysis of our scRNA-seq and IHC data did not reveal a significant association between age and the abundance of T/NK cells, B cells, or myeloid cells (Fig. S6a-b). Similarly, no significant differences in these immune cell populations were observed between male and female patients (Fig. S6c-d). Given the limited sample size of our study, these findings should be further validated in larger cohorts.

CXCL13 is associated with increased lesion progression in the CCM mouse model

To evaluate the role of CXCL13 in the pathogenesis of CCMs, we intravenously administered rmCXCL13 (8 µg) or vehicle control to Ccm2^iECKO mice starting at 4 weeks of age. Following a 4-week treatment regimen (twice per week), brain tissues were collected for analysis (Fig. 6a). MRI scans were performed before treatment to establish baseline lesion burden and again after the 4-week treatment to monitor lesion progression. The MRI index was then calculated to quantitatively represent the severity of CCM lesions. Baseline MRI index did not differ between the treatment and control groups (P = 0.3830). After the 4-week treatment, CCM lesions were observed in both the rmCXCL13 and control treatment groups (Fig. 6b-c). However, the rmCXCL13-treated group exhibited a significantly greater increase in the MRI index compared to the control-treated group (P < 0.0001) (Fig. 6d), indicating that CXCL13 administration accelerated CCM lesion progression.

Fig. 6.

CXCL13 treatment is associated with increased lesion progression in CCM mouse models. a Schematic illustration of the experimental approach. b Representative baseline and post-treatment MRI images of the rmCXCL13 and control groups. Red arrowheads indicate CCM lesions on MRI images. c Comparison of MRI index between the control and rmCXCL13 groups at baseline and after treatment. Data are shown as mean ± SEM; n = 5 for each group; unpaired t-tests; *P = 0.0208, ns P = 0.3830. d Difference in the change rate of the MRI index between the two groups. Data are shown as mean ± SEM; n = 5 for each group; unpaired t-test; ****P < 0.0001. e Representative HE staining in the rmCXCL13 treatment group and the control group. Scale bar: 500 μm. f Difference in the lesion area detected on HE staining sections between the two groups. Data are shown as mean ± SEM; n = 5 for each group; unpaired t-test; ***P = 0.0009. g Representative Prussian blue staining in the rmCXCL13 treatment group and the control group. Scale bar: 50 μm. h and i Differences in non-heme iron deposition area and average iron density between the two groups. Data are shown as mean ± SEM; n = 5 for each group; unpaired t-tests; *P = 0.0209, **P = 0.0049. j Representative IHC staining of B220 in the rmCXCL13 treatment group and the control group. Scale bar: 50 μm. k Difference in the number of lesion-infiltrating B cells between the two groups. Data are shown as mean ± SEM; n = 5 for each group; unpaired t-test; *P = 0.0155. l Representative IHC staining of harvested brain tissue sections from the rmCXCL13-treated and control groups showing CD20⁺ (green) and CXCR5⁺ (red), with nuclei stained with DAPI (blue). Scale bar: 20 μm. m Difference in the number of lesion-infiltrating CXCR5⁺ CD20⁺ cells between the two groups. Data are shown as mean ± SEM; n = 5 for each group; unpaired t-test; **P = 0.0033. The five mice per group were obtained across three independent experiments. Specifically, in each of the first and second experiments, four successfully modeled Cdh5-CreERT2; Ccm2^fl/fl mice were randomly assigned to the control and rmCXCL13 treatment groups. In the third experiment, two successfully modeled Cdh5-CreERT2; Ccm2^fl/fl mice were evenly randomized into the control group and the rmCXCL13 treatment group

Moreover, we performed H&E, Prussian blue staining, and IHC on the harvested brain tissues to systematically evaluate the pathological features of lesions following rmCXCL13 administration. Quantitative analysis of H&E-stained sections revealed a significant increase in CCM lesion area in rmCXCL13-treated mice compared to controls (P = 0.0009) (Fig. 6e-f). Furthermore, the rmCXCL13-treated group also showed larger Prussian blue-positive areas (P = 0.0209) and higher staining density (P = 0.0049) surrounding the lesions (Fig. 6g-i), indicating more pronounced microhemorrhage of the CCM lesions. In addition, B220 IHC staining showed a significant increase in B-cell infiltration within the lesions of rmCXCL13-treated mice (P = 0.0155) (Fig. 6j-k). In parallel, the extent of B-cell infiltration was positively correlated with parameters reflecting CCM lesion size, including the MRI index (P = 0.0088) and lesion area (P = 0.0095), as well as parameters reflecting CCM-associated hemorrhage, including the non-heme iron deposition area (P = 0.0020) and average iron density (P = 0.0473) (Fig. S7). To further investigate the contribution of the CXCL13–CXCR5 axis to B-cell recruitment in CCM lesions, we examined CXCR5⁺ CD20⁺ B cells in harvested brain tissues and found that rmCXCL13 administration significantly increased the accumulation of CXCR5⁺ CD20⁺ B cells within CCM lesions (P = 0.0033) (Fig. 6l-m). Collectively, these data suggest that exogenous CXCL13 is associated with CCM progression and B-cell infiltration.

CXCL13 blocking therapy attenuates the development of CCM lesions

Given the observed pro-progression effect of rmCXCL13 and the detected perilesional CXCL13 deposition in CCM mouse models (P = 0.0453) (Fig. 7a-b and S8a), we next investigated whether targeted blockade of CXCL13 represents a potential therapeutic strategy for CCM intervention. First, we found that the perilesional CXCL13 in CCM mouse models was secreted primarily by microglia/macrophages (Fig. S8b). In human samples, although CXCL13 was predominantly enriched in T/NK cells, our scRNA-seq analysis also detected CXCL13 in microglia/macrophages (Fig. S8c), consistent with prior evidence implicating these cells as potential sources of CXCL13 [56]. Based on these observations, commencing at 6 weeks post-natal, Ccm2^iECKO mice received intravenous administration of a CXCL13-blocking antibody or isotype control, following baseline MRI. After a 2-week treatment (twice per week), MRI was repeated, and brain tissues were collected for analysis (Fig. 7c). No significant intergroup disparity was observed in baseline MRI (P = 0.9321). However, following the 2-week treatment, mice receiving CXCL13-blocking antibody exhibited an attenuated increase in MRI index compared to the control group (P = 0.0005) (Fig. 7d-f).

Fig. 7.

CXCL13 blockade attenuates lesion development in CCM mouse models. a Representative IHC staining of CXCL13 in CCM and wild-type mice. Scale bar: 50 μm. b Difference in the CXCL13-positive area between the two groups. Data are shown as mean ± SEM; n = 3 for each group; unpaired t-test; *P = 0.0453. c Schematic illustration of the experimental approach. d Representative baseline and post-treatment MRI images of the anti-CXCL13 and isotype control groups. Red arrowheads indicate CCM lesions on MRI images. e Comparison of MRI index between the isotype control and anti-CXCL13 groups at baseline and after treatment. Data are shown as mean ± SEM; n = 3 for each group; unpaired t-tests; *P = 0.0375, ns P = 0.9321. f Difference in the change rate of the MRI index between the two groups. Data are shown as mean ± SEM; n = 3 for each group; unpaired t-test; ***P = 0.0005. g Representative HE staining in the CXCL13-blocking treatment group and the isotype control group. Scale bar: 500 μm. h Difference in the lesion area detected on HE staining sections between the two groups. Data are shown as mean ± SEM; n = 3 for each group; unpaired t-test; **P = 0.0024. i Representative Prussian blue staining in the anti-CXCL13 treatment group and the isotype control group. Scale bar: 50 μm. j and k Differences in non-heme iron deposition area and average iron density between the two groups. Data are shown as mean ± SEM; n = 3 for each group; unpaired t-tests; *P = 0.0475 for non-heme iron deposition area and *P = 0.0310 for average iron density. l Representative IHC staining of B220 in the anti-CXCL13 treatment group and the isotype control group. Scale bar: 50 μm. m Difference in the number of lesion-infiltrating B cells between the two groups. Data are shown as mean ± SEM; n = 3 for each group; unpaired t-test; **P = 0.0017. n Representative IHC staining of harvested brain tissue sections from the anti-CXCL13-treated and isotype control groups showing CD20⁺ (green) and CXCR5⁺ (red), with nuclei counterstained with DAPI (blue). Scale bar: 20 μm. o Difference in the number of lesion-infiltrating CXCR5⁺ CD20⁺ cells between the two groups. Data are shown as mean ± SEM; n = 3 for each group; unpaired t-test; *P = 0.0139

Comprehensive histopathological evaluation of the excised brain specimens, via H&E, Prussian blue, and IHC staining, was also performed. Morphometric quantification of CCM lesions in H&E sections showed a statistically significant reduction in lesion area post-intervention (P = 0.0024) (Fig. 7g-h). Concurrently, the anti-CXCL13 group exhibited diminished hemosiderin deposition (P = 0.0475) and attenuated perilesional iron-positive signal intensity (P = 0.0310) compared to the control group (Fig. 7i-k). IHC-based quantification of B220⁺ cells further showed reduced B-cell infiltration following anti-CXCL13 blockade (P = 0.0017) (Fig. 7l-m). Consistently, higher levels of B-cell infiltration were associated with larger CCM lesion size and more pronounced hemorrhage-related changes, as assessed by MRI index (P = 0.0444), lesion area (P = 0.0024), non-heme iron deposition area (P = 0.0285), and average iron density (P = 0.0188) (Fig. S9). The reduced infiltration of CXCR5⁺ CD20⁺ B cells within CCM lesions following CXCL13 blockade further supported the association between the CXCL13–CXCR5 axis and B-cell recruitment (P = 0.0139) (Fig. 7n-o). These results indicated that targeting CXCL13 curtailed the exacerbation of CCM lesions.

Collectively, our results identified a potentially pathogenic CXCL13-producing CD4⁺ T-cell subcluster and suggested that the CXCL13–CXCR5 axis contributes to B-cell recruitment and hemorrhagic progression in CCMs. Importantly, therapeutic CXCL13 blockade attenuated CCM lesion growth, limited B-cell infiltration, and reduced vascular leakage (pathological process characterized by increased endothelial permeability and microhemorrhages [57, 58]). These findings identify the CXCL13 signaling pathway as a promising therapeutic target for CCM patients.

Discussion

Our study constructed a comprehensive single-cell atlas of the human CCM immune microenvironment and identified a previously uncharacterized CXCL13-producing CD4⁺ T-cell subcluster as a potential contributor to disease progression. Through integrative single-cell transcriptomics, spatial analysis, and in vivo functional validation, we showed that the CXCL13-producing CD4⁺ T-cell subcluster is selectively enriched within CCM lesions, where it engages with CXCR5⁺ memory B cells via the CXCL13-CXCR5 axis. This crosstalk may contribute to a pro-inflammatory niche that sustains chronic inflammation, thereby promoting vascular instability (pathological alterations in CCM-related vessels characterized by impaired endothelial junction integrity and an increased propensity for lesion progression and hemorrhage [59, 60]).

Elevated CXCL13 expression has been linked to increased disease activity and tissue pathology in autoimmune disorders [54, 61, 62]. Conversely, heightened CXCL13 expression was associated with better prognosis in multiple malignancies [35, 63, 64]. This divergence underscored the functional pleiotropy of CXCL13-related immune microenvironments, which could polarize toward either pathogenic inflammation or protective immunoregulation depending on the disease setting. In CCMs, our data support a role of CXCL13 in B-cell recruitment. This CXCL13-mediated circuit appears to be associated with maladaptive immune activation, as well as the persistence and severity of vascular lesions.

CXCL13 exerts its immunomodulatory effects primarily through its cognate receptor CXCR5, mediating the recruitment of lymphocytes and supporting structured immune surveillance [65]. In the context of CCMs, however, our findings revealed that this physiological axis was co-opted to establish a localized pro-inflammatory circuit. Specifically, we observed that the abundances of CXCL13⁺ CD4⁺ T cells and CXCR5⁺ B cells were positively associated and that these cells were spatially co-localized within CCM lesions, where they formed densely clustered immune aggregates distinct from canonical lymphoid structures. Moreover, a notable subset of memory B cells within the lesions showed particularly high CXCR5 expression and a transcriptional profile consistent with a highly activated, functionally specialized state. These cells upregulated a repertoire of canonical activation markers, such as CD69, CXCR4, NR4A1, NR4A2, and CD83 [66], as well as GPR183 and CCR7, two key chemokine receptors of B-cell positioning within lymphoid follicles [67, 68]. This subcluster showed significant upregulation of TNFRSF13C (BAFFR) and TNFRSF13B (TACI), receptors that bind both B-cell survival factors BAFF and APRIL, which are critical for B-cell persistence and function [69]. Collectively, the expression patterns of these molecules suggest that the CXCL13/CXCR5 axis in CCMs may contribute not only to B-cell recruitment but also to the formation of structured inflammatory niches associated with disease progression.

In addition, CXCL13-expressing CD4⁺ T cells in CCMs exhibited high expression of IL-21, a cytokine important for promoting B-cell proliferation and plasma cell differentiation [70, 71]. Consistent with this, we observed substantial plasma cell accumulation within CCM lesions, supporting the notion that CXCL13⁺ IL-21⁺ CD4⁺ T cells may contribute to in situ humoral immune activation. This finding aligns with previous reports of oligoclonal B-cell responses and in situ antibody production in CCMs [20, 21], further suggesting that these specialized CD4⁺ T cells play a role in shaping the pathological immune microenvironment of the lesions.

Functionally, this dysregulated CXCL13-CXCR5 signaling axis was associated with chronic inflammation and exacerbated lesion progression. Clinically, elevated CXCL13 expression was associated with enhanced immune infiltration and more severe disease phenotypes in patients with CCMs. In murine models, exogenous administration of CXCL13 resulted in enhanced B-cell infiltration and accelerated lesion progression, whereas CXCL13 blockade attenuated lesion progression, supporting the therapeutic potential of targeting this pathway. Notably, in the mouse model of CCM, CXCL13 secretion is predominantly attributed to microglia/macrophages, which aligns with prior reports of its expression by microglia [72] or macrophages [39, 73]. These findings highlight two distinct cellular sources that may contribute to CCM pathogenesis in mice versus humans; however, the mechanisms by which CXCL13 expression is induced in these different cellular contexts remain to be fully elucidated.

Recent studies suggested that recruited or activated immune cells may amplify local inflammatory and proteolytic injury, thereby promoting endothelial junction disruption and extracellular matrix degradation. In germinal matrix hemorrhage, activated neutrophils and monocytes have been shown to produce proteases and inflammatory mediators, including azurocidin-1, elastase, and CXCL16, thereby disrupting vascular integrity and promoting hemorrhage [74]. In line with this concept, a unified protease-threshold framework has been proposed, according to which germinal matrix hemorrhage occurs when inflammatory amplification drives proteolytic activity beyond the structural resilience of the immature germinal matrix vasculature [75]. Meanwhile, prior studies in CCMs have reported increased expression of MMP-2 and MMP-9 in lesional vascular walls and endothelial cells [76, 77], and have identified ADAMTS5-dependent versican proteolysis as a downstream pathogenic mechanism, supporting a role for protease-mediated extracellular matrix remodeling in CCMs [78]. These findings implied a plausible mechanism linking CXCL13-associated immune recruitment to vascular instability: CXCL13-associated immune cell accumulation might amplify local inflammatory and proteolytic injury, thereby compromising vascular integrity and contributing to hemorrhagic progression. Further studies will be required to validate this proposed mechanism in CCMs.

Current therapeutic strategies targeting CXCL13 primarily focus on direct blockade [27]. At the protein level, the administration of anti-CXCL13 antibodies has demonstrated considerable therapeutic efficacy by attenuating pathogenic inflammatory responses in various animal models of immune-mediated inflammatory diseases [62, 79, 80]. In this study, we applied a CXCL13-blocking antibody in mouse models of CCMs and showed that CXCL13 inhibition reduced B-cell infiltration and attenuated lesion progression, supporting CXCL13 as a promising immunotherapeutic target for CCMs.

Collectively, our findings identify an immunoregulatory axis associated with CCM pathophysiology, linking adaptive immune cell recruitment and function with clinical disease severity. Our data suggest that CXCL13 plays a disease-promoting role in shaping a pro-inflammatory lesional ecosystem and highlight the CXCL13/CXCR5 pathway as a potential therapeutic target to disrupt maladaptive immune-vascular crosstalk in CCMs. At the same time, our findings raise several important questions that warrant further investigation. First, the human cohort is relatively limited given the clinical and biological heterogeneity of CCMs, including differences in lesion stage, hemorrhagic activity, immune composition, and patient background. Therefore, the observations derived from the human dataset should be viewed as hypothesis-generating and mechanistically informative rather than definitive and will require validation in larger, more comprehensively characterized cohorts. Second, the analyzed surgical specimens predominantly represent symptomatic lesions, many of which were associated with subacute intracerebral hemorrhage. As hemorrhage itself may substantially reshape the lesional inflammatory milieu, future studies will be needed to determine the extent to which CXCL13 upregulation and lymphocyte accumulation reflect primary disease mechanisms versus secondary responses to hemorrhage-associated tissue injury. Third, although our results showed that CD4⁺ T cells represent a major source of CXCL13 in CCM lesions, CXCL13 was also detected in microglia/macrophages and in CD8⁺ T cells. Therefore, our in vivo manipulation experiments primarily address the role of CXCL13 signaling during CCM progression rather than the contribution of a specific CXCL13-producing cell type. Future studies using cell–type–specific genetic approaches will be required to define the relative contributions of distinct CXCL13-expressing populations. Finally, although the present work focuses on the CXCL13/CXCR5 axis, additional immune pathways and cellular interactions likely contribute to lesional remodeling and disease progression and remain to be systematically explored.

Conclusion

In conclusion, our study identified a CXCL13-producing CD4⁺ T-cell subcluster enriched in CCMs and implicated this subset in coordinating B-cell infiltration. Clinically, elevated lesional CXCL13 expression was associated with greater B-cell infiltration and a higher frequency of subacute ICH, while functional studies in CCM mouse models further supported CXCL13’s role in B-cell recruitment and lesion progression. These findings highlight the potential significance of the CXCL13-CXCR5 axis in shaping the inflammatory microenvironment associated with CCM progression, thereby supporting CXCL13 signaling as a promising therapeutic target for patients with CCMs.

Abbreviations

4-OHT

4-hydroxytamoxifen

CCMs

Cerebral cavernous malformations

DEGs

Differentially expressed genes

FFPE

Formalin-fixed paraffin-embedded

FOV

Field of view

HE

Hematoxylin-and-eosin

HPFs

High-power fields

ICH

Intracerebral hemorrhage

IHC

Immunohistochemistry

MRI

Magnetic resonance imaging

PCA

Principal component analysis

PBMCs

Peripheral blood mononuclear cells

rmCXCL13

Recombinant mouse CXCL13

scRNA-seq

Single-cell RNA sequencing

TE

Echo time

TR

Repetition time

Authors’ contributions

YC, DL, and JW conceived and designed the research. SZ, RH, HX, and YS designed and implemented the experiments. YY performed the computational analyses. YN collected the clinical data and human samples. HW and YJ performed the data analysis. SZ and YY wrote the manuscript with comments from all authors. YC, DL, JW, and QZ revised the manuscript. All authors approved the final version of this manuscript.

Funding

Research in the Cao lab was funded by the NSFC International Cooperation and Exchanges Program (W2411067) and the NSFC General Program (82171267). Research in the Liu lab was supported by the National Key R&D Program of China (2024YFA1306600), the National Science Foundation of China (82571989/82521102), the Hangzhou “XHMZ” Top Talent Program, and the Westlake Laboratory of Life Sciences and Biomedicine. Research in the Wang lab was supported by the Hong Kong RGC Fund (16102522), the Hong Kong ITC Fund (ITC-SKLNSD26SC01), and the Padma Harilela Professorship and Sinovac Fellowship Program (Z1541).

Data availability

Raw sequencing data have been deposited in the Genome Sequence Archive [81] (GSA, https://ngdc.cncb.ac.cn/gsa) at the National Genomics Data Center, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences [82], under accession numbers HRA018296 and HRA018294. Scripts have been uploaded to GitHub (https://github.com/sheenaseven/CXCL13_CD4T). Raw data of the bulk RNA-seq dataset of 10 CCM samples were obtained from the GEO database (GSE137596).

Declarations

Ethics approval and consent to participate

The collection of human samples was approved by all patients under written informed consent. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Internal Review and Ethics Committee of the Beijing Tiantan Hospital. The Animal Welfare and Ethics Committee of Beijing Neurosurgical Institute Laboratory approved all animal ethics and protocols.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.