Identification and characterization of CXCL13 producers in bone tissue in response to fasting

Abstract

The bone tissue serves as a dynamic reservoir that accommodates diverse immune cell populations during low-energy states such as fasting. Energy restriction and reduced metabolism induce a transient migration of naïve B cells from Peyer’s patches to the bone marrow in mice, a process driven by the upregulation of Cxcl13 expression within the bone niche. However, the specific cellular source of CXCL13 and the upstream signals regulating its expression remain undefined. Our analysis identified the lineage-negative fraction of the bone as the major source of CXCL13. Single-cell RNA sequencing of these cells further demonstrated that Cxcl13 was selectively expressed by osteogenic mesenchymal stromal cells (MSCs) during fasting. Consistently, immunofluorescence imaging showed an increased frequency of these CXCL13-producing cells under fasting conditions, primarily localized within the perivascular niche. Gene Ontology analysis of differentially expressed genes in these cells during fasting revealed enrichment of TGF-β and PPAR signaling. In line with this, pharmacological intervention in either the TGF-β or PPARγ signaling pathway significantly attenuated fasting-induced naïve B-cell migration and CXCL13 production in the bone. These results establish a mechanistic link between systemic metabolic cues and the bone marrow immune microenvironment. We propose that osteogenic MSCs play a vital role in orchestrating this trafficking program for naïve B cells, underscoring a dynamic crosstalk between the bone stromal cells and the immune system.

The mechanism of fasting-induced B-cell migration to the BM

Graphical Abstract

Graphical Abstract.

Introduction

Dietary interventions such as calorie restriction and fasting profoundly influence immune cell function and the distribution of lymphocytes across organs (1). We previously reported that, during fasting, germinal center B cells and IgA-producing plasma cells in Peyer’s patches (PPs) undergo apoptosis. In contrast, naïve B cells in PPs migrate to the bone marrow (BM) during fasting and are redistributed back to PPs upon refeeding. Consequently, PPs, which normally generate large amounts of IgA-producing cells, transiently shut down the IgA response during fasting. These findings suggest that intestinal IgA responses are not continuously active but are adaptively downregulated under nutrient-deprived conditions, allowing for energy conservation (2).

Moreover, Collins et al. demonstrated that CD8⁺ memory T cells are translocated to the BM during 50% calorie restriction (3). This mobilization promotes the clearance of influenza virus and Yersinia pseudotuberculosis infections in mice. Jordan et al. and Janssen et al. reported that short-term fasting facilitates the recruitment of Ly-6C^hi monocytes from the circulation to the BM (4, 5). Upon infection with Listeria monocytogenes, monocytes migrate from the BM into the bloodstream to eliminate the pathogen without causing inflammation. Thus, the BM functions as a dynamic reservoir that responds to systemic energy status by mobilizing and retaining diverse immune cell subsets.

Distinct mechanisms govern the trafficking of different immune cell subsets in the BM, depending on both cell type and metabolic state. For example, in addition to the sphingosine 1-phosphate (S1P)–S1P receptor type 1 (S1P₁) axis, chemokine–chemokine receptor pathways play crucial roles. Under homeostatic conditions, CXCL12 is constitutively expressed in the bone tissue and serves as a key regulator of hematopoietic stem cell retention and homing via CXCR4 signaling (6, 7). The CXCL12–CXCR4 axis is essential for maintaining the BM niche and bone microenvironment under steady-state conditions. CXCL12 is secreted continuously by multiple stromal populations, including CXCL12-abundant reticular (CAR) cells (8, 9). CXCR4 expression is also observed in immune subsets such as long-lived plasma cells and regulatory T (Treg) cells, and is indispensable for their localization and retention within the BM (10–12). Glucocorticoids secreted during fasting promote BM migration of certain immune cells by inducing CXCR4 expression (5).

In contrast, the mechanisms controlling naïve B cell trafficking to the BM under fasting conditions remain poorly understood. In secondary lymphoid organs, B cell trafficking is primarily directed by the CXCL13–CXCR5 axis, with the CXCL12–CXCR4 and CCL19/CCL21–CCR7 axes serving supportive roles (13–16). Fasting suppresses Cxcl13 expression in PPs, while upregulating it in bone tissue. In contrast, Cxcl12, Ccl19, and Ccl21 expression shows no significant difference between PPs and bone tissue under the same conditions. This reciprocal regulation implicates the CXCL13–CXCR5 axis as a key mediator of naïve B cell circulation between PP and the BM in response to nutritional status. However, this concept remains insufficiently validated. In secondary lymphoid organs such as PPs, CXCL13 is mainly produced by marginal reticular cells (MRCs) and follicular dendritic cells (FDCs) (16–19). However, the cellular identity of CXCL13-producing cells in bone tissue, as well as the molecular mechanisms controlling its expression, remains undefined.

In this study, we investigated the role of the CXCL13–CXCR5 axis during fasting and elucidated the cellular mechanisms within bone tissue that regulate this pathway. We demonstrate that CXCL13–CXCR5 signaling is critical for the fasting-induced migration of naïve B cells into the BM. Furthermore, we identified osteogenic mesenchymal stromal cells (MSCs) localized in the perivascular niche as a major source of CXCL13, whose expression is induced in a TGF-β- and PPARγ-dependent manner. Collectively, our findings provide new insights into how systemic metabolic changes orchestrate immune cell dynamics within the BM microenvironment.

Methods

Mice

Unless otherwise specified, 5-week-old male BALB/cA were purchased from CLEA Japan Inc. (Tokyo, Japan) and were acclimated for 1 week under specific pathogen-free conditions at the animal facilities of the National Institute of Global Health and Medicine (Chiba, Japan) or Keio University (Tokyo, Japan). We used BALB/cA mice in most experiments unless mentioned otherwise. Cxcl13-Cre/Tdtomato R26R-EYFP (abbreviated as Cxcl13-EYFP) mice were provided by Dr Burkhard Ludewig and were maintained at our facility (20). All mice were housed under a 12:12 h light–dark cycle with ad libitum access to food and water, unless otherwise stated. For fasting experiments, the mice were housed in individual plastic cages without bedding chips or bait, with a stainless-steel mesh floor to avoid coprophagia and with free access to water. A 36-h fasting period was initiated at 8:00 p.m., followed by refeeding with normal chow or tissue collection at 8:00 a.m. To neutralize CXCL13 in vivo, we administered 100 μg/mouse of anti-CXCL13 polyclonal antibodies (R&D Systems) intravenously at 0 and 18 h post-fasting. To inhibit CXCR4 in vivo, we administered 10 mg/kg of AMD3100 (Calbiochem) subcutaneously at 12 and 24 h post-fasting. To inhibit AMPK, FoxO1, Smad3, FABP4, PPARγ, PPARβ/δ in vivo, we administered 10 mg/kg of Compound C (Selleck Chemicals), 15 mg/kg of AS1842856 (Selleck Chemicals), 15 mg/kg of SIS3 (Selleck Chemicals), 15 mg/kg of BMS309403 (Selleck Chemicals), 10 mg/kg of GW9662 (Selleck Chemicals), and 10 mg/kg of DG172 dihydrochloride (Selleck Chemicals) intraperitoneally at 0 and 18 h post-fasting, respectively. Mice were euthanized at defined time points for analysis or when they reached a humane endpoint (e.g. weight loss ≥ 20%). Euthanasia was performed by cervical dislocation under deep isoflurane inhalation anesthesia (5% isoflurane), following the FELASA and AVMA guidelines for the euthanasia of animals. All animal experiments were performed according to the Institutional Guidelines for the Care and Use of Laboratory Animals in Research with approval by the local ethics committees at the National Institute of Global Health and Medicine and Keio University.

Preparation of single-cell suspension

Femurs and tibias were dissected and crushed to isolate bone cells. The crushed tissue was incubated in 20 ml RPMI 1640 medium (pH 7.2; Nacalai Tesque) supplemented with 2% fetal bovine serum (FBS; Sigma-Aldrich), 12.5 mM HEPES (Nacalai Tesque), penicillin–streptomycin mixed solution (Nacalai Tesque), 1.66 mg/ml collagenase A (Roche), 0.5 mg/ml DNase I (Sigma-Aldrich), and 1.0 mg/ml Dispase II (neutral protease, grade II; Roche) for 45 min at 37°C. The cell suspension was then passed through a 100-μm cell strainer (Greiner Bio-One) and centrifuged at 500 × g for 5 min at 4°C. Red blood cells (RBCs) were lysed using RBC lysis reagent (150 mM NH₄Cl, 1 mM EDTA, and 0.4 mM NaHCO₃), and then the cells were suspended in phosphate-buffered saline (PBS) with 2% FBS.

Splenocytes were mechanically dispersed into single-cell suspensions. RBCs were removed using RBC lysis reagent (150 mM NH₄Cl, 1 mM EDTA, and 0.4 mM NaHCO₃), and cells were suspended in PBS with 2% FBS.

Flow cytometry

Nonspecific binding was blocked with an Fc receptor antibody (clone S17011E; BioLegend) before staining with fluorochrome-conjugated antibodies for surface and intracellular staining. Most surface staining was conducted for 30 min at 4°C, whereas that for CXCR5, CXCR4, and CCR7 was conducted for 45 min at 37°C. For intracellular staining, cells were fixed, permeabilized, and stained with monoclonal antibodies using the Foxp3 staining set (Invitrogen, San Diego, CA) according to the manufacturer’s instructions. Dead cells were discriminated using 7-AAD (Thermo Fisher Scientific) or the Fixable Viability Dye eFluor 780 (FVS780; Thermo Fisher Scientific). Stained samples were analyzed using an LSRFortessa and Symphony, and were sorted using Aria III flow cytometers with DIVA software (all from BD Biosciences, San Diego, CA) and FlowJo software version 10.10.0 (FlowJo LLC, Ashland, OR). The antibodies used for flow cytometry are shown in Supplementary Table 1.

ELISA

Serum was prepared by centrifuging whole blood collected via cardiac puncture. For tissue-specific analyses, whole bone was homogenized in RIPA buffer (Nacalai Tesque) to extract proteins. Total CXCL13 concentrations were measured in both serum and bone lysates using a Mouse CXCL13/BCA1 ELISA Kit (Proteintech), according to the manufacturer’s instructions.

Immunohistochemistry

Whole-mount tissue preparation, immunofluorescence staining, and imaging of the femur were performed as previously described, with slight modification (21). Briefly, mouse femurs were isolated and immersed in 4% paraformaldehyde/PBS for 16 h at 4°C. Then, the specimens were dehydrated in 30% sucrose for 72 h at 4°C. Next, the specimens were embedded in optimal cutting temperature embedding medium and frozen in liquid nitrogen for 10 min. Specimens were preserved at −80°C until sectioned. Specimens were repeatedly sectioned with a cryostat until the surface of the BM was fully exposed along the longitudinal axis for femurs. Then, the specimens were turned over, and the same sectioning procedure was repeated on the opposite site. After the entire BM sections were generated, the sections were washed three times with PBS. Then, the sections were incubated for 1 h in a blocking solution consisting of 2% bovine serum albumin (BSA), 10% donkey serum, and 0.2% phosphate-buffered saline containing Tween (PBST). Next, the sections were stained with primary antibodies for 16 h at 4°C. Then, the sections were washed 10 times with 0.2% PBST for an hour and stained with DAPI (Invitrogen, D1306) and fluorophore-labelled secondary antibodies for 16 h at 4°C. Sections were consequently washed 10 times with 0.2% PBST for 1 h and incubated in RapiClear 1.52 for 16 h. All the primary antibodies were diluted in 2% BSA, 10% donkey serum, and 0.2% PBST. All the secondary antibodies were diluted in 10% donkey serum and 0.2% PBST. The antibodies for immunofluorescent staining are listed in Supplementary Table 2.

Confocal microscopy

Confocal microscopy was performed as previously described, with slight modifications (21). Briefly, BM slices were mounted on a microscope slide (Matsunami Glass Ind., SFRC-04) using vacuum grease (Merck, Z273554-1EA), with the BM surface covered with cover glass (Matsunami Glass Ind, C024501). Confocal microscopy immunofluorescence data were obtained with a confocal laser scanning microscope (Zeiss, LSM880). In the imaging of BM sections for this study, the z-slice thickness was set to 5–10 μm when imaged with a 10× objective, and to 2 μm when imaged with a 63× objective. The “tile experiment” mode was used for confocal imaging with 10% overlap between each tile. The acquired image fields (tiles) were stitched together with the “stitch process” mode after scanning to get an overall image of the samples.

Image analysis

All image analyses were performed using Imaris 10.0.1 software (Oxford Instruments) according to the manufacturer’s instructions. For visualization, confocal images were rendered into 3D volumes and represented as a maximum intensity projection (MIP) unless otherwise indicated. When displaying 2D images, the Ortho Slicer mode was used. For quantification, a 100-µm region (10 consecutive z-slices) was extracted from each sample along a 2500-µm segment of the femoral diaphysis, MIP-projected to generate 2D TIFF images, and imported into Imaris for analysis. Using the Surface module with identical threshold settings across samples, we segmented the BM area and the CXCL13-EYFP⁺ area within the observed field. The CXCL13-EYFP/BM area ratio was calculated by dividing the total area of CXCL13-EYFP⁺ cells by the BM area in the corresponding 2D projection.

Reverse transcription and quantitative PCR

Total RNA from the bone was extracted using TRIzol Reagent (Thermo Fisher Scientific) or Sepazol (Nacalai Tesque) according to the manufacturer’s instructions. Total RNA from magnetically isolated lineage⁺ and/or lineage⁻ bone cells was isolated using the RNeasy Micro Kit (QIAGEN, Hilden, Germany) according to the manufacturer’s instructions. RNA was reverse-transcribed using an iScript Advanced cDNA Synthesis Kit (Bio-Rad Laboratories, Hercules, CA, USA) or ReverTra Ace qPCR RT Master Mix with gDNA Remover (TOYOBO). Quantitative PCR was conducted on a CFX-Connect system (Bio-Rad) using SsoAdvanced Universal SYBR Green Supermix (Bio-Rad) along with primers in Supplementary Table 3. Gene expression was normalized against the expression of Rpl32 and showed the value of 2^−ΔCt.

Bulk RNA-seq

For the RNA-sequencing of EYFP⁺ cells in the bone, BM cells from mice fasted for 36-h and mice refed standard chow for 48-h were subjected to cell sorting using a FACSAria III (BD Biosciences) to isolate the CD45⁻ CD31⁻ Ter119⁻ EYFP⁺ cells in the bone. Sorted cells were stored in the lysis/storage buffer [1% lithium lauryl sulfate (NACALAI TESQUE), 100 mmol/L Tris-HCl (pH 7), 500 mmol/L LiCl (Sigma-Aldrich), and 10 mmol/L EDTA (NIPPON GENE)].

RNA extraction and library preparation of bulk RNA-seq of the bone cells were performed by ImmunoGeneTeqs, Inc. (Chiba, Japan) as previously described, with slight modification (22). Briefly, mRNA was isolated from total RNA by Dynabeads M-270 streptavidin with biotin-oligo (dT)25 (Thermo Fisher Scientific). First- and second-strand cDNA synthesis was performed by Superscript reverse transcriptase IV (Thermo Fisher Scientific) and Kapa HiFi DNA polymerase (Roche). While captured on the beads, cDNA was digested with fragmentase (0.5 μl for the sample; New England Biolabs). Each 3′end of the cDNA fragment (100 ng) was ligated to an adapter carrying Ion-Barcode-common sequence 1 (CS1; 1.5 μmol/L). Finally, an eight-cycle PCR step with a PCR mixture (10 μmol/L of trP1 primer, 10 μmol/L Ion_barcode, and 1× KAPA Hifi Hotstart ReadyMix) was performed to enrich for the desired cDNA library molecules using MiniAmp Thermal Cycler (Thermo Fisher Scientific), and cDNA library products were purified by size selection using AMPure XP beads (Beckman Coulter). Products were confirmed using the Agilent High Sensitivity DNA kit (Agilent Technologies).

The above 5′ end serial analysis of gene expression–sequencing libraries (input library concentration was 65 pmol/L) were used for sequencing by an Ion 540 Chef kit, an Ion 540 Chip kit, and an Ion S5 Sequencer (Thermo Fisher Scientific) according to the manufacturer’s instructions. Adapter trimming and quality filtering of sequencing data were performed by using Cutadpat-v2.4 (23) and Trimommatic-v0.36 (24). The filtered reads were mapped on Refseq mm10 using Bowtie2-2.2.5 (parameters: -t -p 11 -N 1 -D 200 -R 20 -L 20 -i S, 1, 0.50 -nofw). The mapped reads per gene (raw tag counts) were quantified as gene expression. Between-sample normalization of gene expression was performed against raw count data by using R 4.3.2 [Delaware Public Benefit Corporation (PBC), Boston, MA] with DESeq2 packages (25). Genes with adjusted P value < .05 and a fold change of ≥ 1 were identified as statistically significantly differentially expressed genes (DEG). Gene Ontology enrichment and graphical visualization of enrichment results and gene characteristics were performed with ShinyGO v0.82 (26). Genes were mapped to Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (27).

Single-cell RNA-sequence

Single-cell suspensions of bone cells were prepared as described above. After surface staining CD45, CD31, and Ter119, live CD45⁻ CD31⁻ Ter119⁻ cells were sorted by a BD IMag™ Cell Separation Magnet (BD Biosciences) and FACSAria III instrument (BD Biosciences). Sort purified CD45⁻ CD31⁻ Ter119⁻ cells were subjected to the preparation of 3’ Gene Expression libraries using a Chromium Next GEM Single Cell 3’ Reagent Kits v2 (Dual Index) (10× Genomics, Pleasanton, CA) following the manufacturer’s instructions (28). Pooled libraries were sequenced on an Illumina NovaSeq 6000 (150 bp × 2 paired-end), aiming at 800.0 M reads per sample. The sequenced paired reads were processed with the “cellranger multi” function using CellRanger v8.0.0 (10× Genomics) for alignment, filtering, barcode counting, and UMI counting. Seurat objects were generated from the filtered CellRanger output using the “Read10X” function and the “CreateSeuratObject” function (Seurat v5.1.0) (29). Low-quality cells were excluded based on the following criteria: the number of unique genes detected per cell > 500, the number of unique genes detected per cell > 8220.5, the gene expression levels > 1000, the gene expression levels < 31531.12, and the percentage of mitochondrial gene expression < 10%. Following this initial quality control, we excluded contaminating erythroid lineages (e.g. Car1⁺, Klf1⁺ early erythroid cells and Alas2⁺, Hbb-bs⁺ late erythroid cells), similar to the strategy described by Wolock et al. (30). Following these filtering steps, we obtained transcriptomic data from 3541 and 4761 high-quality cells from ad libitum-fed and fasted mice, respectively. The Seurat objects were integrated following the SCTransform integration workflow (31). The number of features in the “SelectIntegrationFeatures” function was set to “2000” to obtain as many variable genes as possible. Otherwise, the default parameter was set.

Quantification and statistical analysis

Statistical analyses were performed using GraphPad Prism 10 software (GraphPad Software, La Jolla, CA, USA). A P-value < .05 was considered statistically significant. The specific statistical test used for each experiment is detailed in the corresponding figure legend. In general, for comparisons between two groups, a two-tailed Student’s t-test was used for data assumed to have equal variances. When the variances were not homogeneous between two groups, the data were analysed by Welch’s t-test. For comparisons among three or more groups, one-way or two-way analysis of variance was performed. Post-hoc multiple comparisons were conducted using Tukey’s test (for all pairwise comparisons), Dunnett’s test (for comparisons against a single control group), as appropriate. Data are typically presented as mean ± SEM. The experiments were not randomized, and the investigators were not blinded to allocation during experiments or outcome assessment.

Results

The CXCL13–CXCR5 axis regulates naïve B cell dynamics during fasting

We first explored the contribution of the CXCL13–CXCR5 axis to BM migration of naïve B cells during fasting. The number of B220⁺IgM⁺IgD⁺ naïve B cells in the BM markedly increased after a 36-h fasting period and returned to baseline following refeeding. Similar transient expansions were also observed in CD4⁺ naïve T cells and CD8⁺ memory T cells (Fig. 1a and Supplementary Figure 1). We next analyzed the expression profiles of key chemokine receptors—CXCR4, CXCR5, and CCR7—on these immune cell subsets in the BM, blood, and spleen. Mean fluorescence intensity of CXCR5 on IgM⁺IgD⁺ naïve B cells in the BM, but not blood or the spleen, was slightly yet significantly elevated during fasting and was sustained after refeeding (Fig. 1b and Supplementary Figure 2a). In contrast, CXCR5 expression on IgM⁻IgD⁻ immature B cells, CD3ε⁺ T cells, and B220⁻CD3ε⁻ cells in the BM remained unchanged throughout the dietary regimen (Fig. 1b). CXCR4 expression was robustly upregulated in both B and T cells across the BM, blood, and spleen during fasting and subsequently returned to the basal level after refeeding (Supplementary Figure 2b). Meanwhile, CCR7 expression was either unchanged or slightly decreased across these immune cell subsets (Supplementary Figure 2c). Collectively, these data indicate that fasting specifically enhances CXCR5 expression on naïve B cells within the BM, whereas CXCR4 upregulation represents a broader systemic response among both B and T cell populations.

Figure 1.

The CXCL13–CXCR5 axis regulates naïve B cell dynamics in the bone marrow (BM) during fasting and refeeding. (a) Left: Representative flow cytometry plots show the expression of IgM and IgD on B220⁺ CD3ε⁻ B cells in the BM of mice fed ad libitum, mice fasted for 36 h, and mice refed for 48 h. The mean ± SEM of the IgM⁺ IgD⁺ naïve B cells is shown for each treatment. Right: Cell numbers of the IgM⁺ IgD⁺ naïve B cells in the BM of mice fed ad libitum, mice fasted for 36 h, and mice refed for 48 h (each group, n = 8). Data are pooled from two independent experiments. (b) Representative mean fluorescence intensity (MFI) of CXCR5 on the indicated cell subsets in the BM of mice fed ad libitum, mice fasted for 36 h, and mice refed with CE-2 for 48 h (each group, n = 4). The MFI of CXCR5 was measured by flow cytometry. Data are representative of two independent experiments. (c) Concentration of CXCL13 in bone tissue of mice fed ad libitum (n = 10), mice fasted for 36 h (n = 10), and mice refed for 48 h (n = 8). (d–e) To inhibit CXCL13 signaling, anti-CXCL13 polyclonal antibodies or IgG isotype control were administered intravenously twice (at 0 and 18 h post-fasting). Mice were analyzed by flow cytometry 36 h post-fasting. Data pooled from three independent experiments are shown (n = 4–8 per group). (d) Upper: Representative data of the expression of IgM and IgD among B220⁺ CD3ε⁻ B cells. Lower: Cell number and frequency (in live cells) of the IgM⁺ IgD⁺ naïve B cells in the BM and blood. (e) Cell numbers of the CD4⁺ T cells and CD8⁺ T cells in the BM and blood. Data represent the means ± SEM. Statistical significance was determined by one-way ANOVA followed by Tukey’s test (a), (c), (d), and (e), and by two-way ANOVA followed by multiple t-test (b). *P < .05; **P < .01.

To assess the role of the CXCL12–CXCR4 axis in the migration of immune cell subsets into the BM, we administered the CXCR4 antagonist AMD3100 to fasted mice. This treatment abrogated the fasting-induced accumulation of CD8⁺ effector memory T cells (Tem) in the BM, consistent with previous reports (3). However, CXCR4 blockade had no discernible effect on the migration of naïve B cells to the BM (Supplementary Figure 2d), suggesting that the CXCL12–CXCR4 axis is dispensable for naïve B cell trafficking during fasting.

We subsequently investigated the contribution of the CXCL13–CXCR5 axis to B cell trafficking during fasting. We confirmed that CXCL13 concentration was elevated transiently in the bone tissue upon fasting (Fig. 1c) (2). The administration of anti-CXCL13 neutralizing antibodies significantly inhibited the fasting-induced accumulation of naïve B cells in the BM (Fig. 1d). This was accompanied by an increase in the number and portion of naïve B cells in the blood of anti-CXCL13-treated mice, albeit not statistically significant (Fig. 1d). In contrast, neutralization of CXCL13 did not influence the number of CD4⁺ T cells and CD8⁺ T cells in the BM of fasted mice (Fig. 1e). These data collectively demonstrate that the CXCL13–CXCR5 axis plays a nonredundant role in mediating naïve B cell migration into the BM during fasting.

Fasting increases CXCL13-producing cells in the perivascular region of the bone

To determine the localization of CXCL13-producing cells in the bone tissue, we utilized Cxcl13-EYFP reporter mice (20). The femurs of Cxcl13-EYFP mice were co-stained for CD31 and Sca-1 to visualize endothelial cells and arteries/arterioles. Consistent with ELISA data (Fig. 1c), the frequency of CXCL13⁺ cells in the bone tissue of fasted mice increased compared to the ad libitum-fed controls (Fig. 2a and b). CXCL13⁺ cells were primarily located in the perivascular region in close proximity to endothelial cells (Fig. 2c and d, Supplementary Video 1). Notably, CXCL13⁺ cells appeared to form clusters (Fig. 2c) and were frequently found adjacent to arteries/arterioles (Sca-1⁺ CD31⁺) and sinusoidal vessels (Sca-1⁻ CD31⁺), and periosteal regions throughout the bone (Fig. 2d). Furthermore, the CXCL13-producing cells showed a stromal cell-like morphology (Fig. 2c and d).

Figure 2.

Fasting increases the number of CXCL13-producing cells residing in the perivascular region of the bone. (a) Low-magnification image of a whole-mount immunohistochemical image of the femur from CXCL13-EYFP mice under ad libitum (left) and fasting (right) conditions. (b) Quantification of CXCL13-EYFP positive area normalized to the total BM area in mice fed ad libitum (n = 4), and mice fasted for 36 h (n = 5). (c) Representative three-dimensional (3D) reconstructed images of the bone tissue from a fasted CXCL13-EYFP mouse. The image is a magnified view of the white square in panel (a). (d) Higher magnification views of the areas outlined by the dotted lines in panel (c). Arrows and filled arrowheads indicate EYFP⁺ cells adjacent to arteries/arterioles (Sca-1⁺ CD31⁺), sinusoidal vessels (Sca-1⁻ CD31⁺), respectively. Unfilled arrowheads indicate EYFP⁺ cells close to the periosteum. Data represent the means ± SEM. Statistical significance was determined by Welch’s t-test (b). **P < .01.

Osteogenic MSCs are responsible for CXCL13 production

To identify the cellular source of CXCL13 in the bone during fasting, we first fractionated bone tissue-derived cells into two fractions: lineage-positive (Lin⁺: CD45⁺, Ter119⁺, or CD31⁺) and lineage-negative (Lin⁻: CD45⁻, Ter119⁻, and CD31⁻) cells. Cxcl13 expression was largely restricted to the lineage-negative stromal cell fraction (Supplementary Figure 3a), consistent with the confocal imaging data (Fig. 2).

We further characterized Cxcl13-expressing cells by conducting single-cell RNA-seq (scRNA-seq) of the lineage-negative cell fraction of ad libitum-fed and fasted mice (Supplementary Figure 3b). Uniform Manifold Approximation and Projection analysis of the integrated dataset demonstrated eight distinct cell clusters, representing a heterogeneous population of bone stromal cells (Fig. 3a). To define these populations, we performed cluster annotation guided by marker genes established in previous studies (30, 32). Based on expression of the marker genes, each cluster was annotated as follows: MSCs and adipocyte progenitors (clusters 0), adipo-CAR cells (cluster 1), Ng2⁺ (Cspg4⁺) MSCs (cluster 2), osteo-CAR cells and osteoblasts (cluster 3), smooth muscle cells and pericytes (cluster 4), proliferating cells (cluster 5), pro-chondrocyte (cluster 6), and hematopoietic progenitors (cluster 7) (Fig. 3a and Supplementary Figure 3c). Although the relative abundance of most of the clusters was comparable between the two groups (clusters 0, 1, 3, 6, and 7) or slightly decreased in the fasted group (clusters 4 and 5), that of cluster 2 was increased in the fasted group (Fig. 3b and Supplementary Figure 3d).

Figure 3.

scRNA-seq analysis defines transcriptomic profiles of CXCL13-producing cells as osteogenic MSCs. (a) Left panel: Uniform Manifold Approximation and Projection (UMAP) plot of single-cell transcriptomes from non-hematopoietic (CD45⁻ Ter119⁻), non-endothelial (CD31⁻), and non-erythroid cells isolated from the bone marrow of ad libitum-fed (3541 cells) and fasted (4761 cells) mice. Eight distinct cell clusters are identified. Right panel: A table summarizing the functional annotation of each cluster based on marker gene expression. (b) A pie chart showing the relative frequencies of each cell cluster (left panel: ad libitum, right panel: fasting). (c) UMAP feature plots showing the expression levels of Cxcl13 and Cxcl12 across all identified cell clusters. Color intensity corresponds to gene expression level. (d) A pseudotime trajectory analysis using Monocle3 on the dataset from fasted mice. The trajectory’s origin is set to cluster 0, visualizing the differentiation path of the identified cell populations. Black lines on the UMAP plots represent the trajectory graph. (e) UMAP plot of the sub-clustered populations within cluster 2, revealing further heterogeneity. Each point represents a single cell, and colors indicate the newly assigned sub-cluster. (f) Violin plots showing the expression of key marker genes associated with pre-osteogenic mesenchymal stem cells (MSCs) within cluster 2, confirming its cellular identity. (g) UMAP feature plots showing the expression of Cxcl13, Il6, and Tgfbr2, highlighting the co-expression of these genes in specific cell populations within cluster 2.

Notably, Cxcl13 expression emerged in a distinct subpopulation within cluster 2 exclusively in the fasted group, whereas it was absent in ad libitum-fed mice (Fig. 3c). This Cxcl13-expressing population is distinct from Cxcl12-expressing CAR cells (Fig. 3c). Furthermore, cluster 2 exhibited concurrent expression of stem cell–associated genes (Cd34, Ly6a, Thy1, and Cd44) and osteogenic/chondrogenic differentiation marker genes (Igfbp6, Mfap5, S100a4, and Fmod). A pseudotime trajectory analysis using Monocle 3 confirmed that cluster 2 cells serve as progenitors of several cell subsets (Fig. 3d). Furthermore, KEGG analyses of differentially expressed genes in cluster 2 revealed an enrichment of gene sets related to ECM-receptor interaction and TGF-β signaling pathway during fasting (Supplementary Figure 4a and b).

To precisely define the cell population responsible for Cxcl13 expression, we performed subclustering analysis of cluster 2, which resolved into three distinct subclusters: chondroprogenitors and tenocytes (clusters 2-0), pre-osteogenic MSCs (clusters 2-1), multipotent MSCs (cluster 2-2) (Fig. 3e, Supplementary Figure 5a). Among these subclusters, Cxcl13-expressing cells were specifically localized to the pre-osteogenic MSC population (clusters 2-1), characterized by the expression of Mfap5, Ccn3, Cd248, and Ltbp4 (Fig. 3f and g, Supplementary Figure 5a). In addition, Cxcl13-expressing cells co-expressed Il6, and Tgfbr2 (Fig. 3g), whereas genes associated with the conventional noncanonical NF-κB pathway of CXCL13 production, including Ltbr, Tnfr, Map3k14, and Relb, were absent or expressed at low levels (Supplementary Figure 5b) (33). This finding raises the possibility that fasting-induced CXCL13 production in the bone may occur through a mechanism distinct from that in lymphoid tissues.

TGF-β and PPARγ signaling contribute to the upregulation of CXCL13 in the bone during fasting

To gain mechanistic insights into fasting-induced CXCL13 production in bone, we focused on the CXCL13-EYFP⁺ population. Quantitative PCR analysis confirmed that these fate-mapped EYFP⁺ cells exhibited significantly higher expression of Cxcl13 and cluster 2-1 marker genes, including Gsn and Cspg4, than EYFP⁻ cells (Supplementary Figure 6a). Furthermore, Cxcl13 expression in sorted EYFP⁺ cells was significantly higher in fasted mice than in refed mice (Supplementary Figure 6b). These results indicate that the CXCL13-EYFP⁺ population represents stromal cells that actively produce CXCL13 in the BM in a nutrition-dependent manner.

We then performed bulk RNA-sequencing on CD45⁻ CD31⁻ Ter119⁻ EYFP⁺ cells isolated from the bones of fasted and refed Cxcl13-EYFP reporter mice (Supplementary Figure 7a). Principal-component analysis showed distinct gene expression profiles between the two groups (Fig. 4a). Differential gene expression analysis indicated upregulation of osteogenesis-related genes (e.g. Cytl1, Fst, Pdk4, and Dio3), and lipid metabolism-related genes (e.g. Fabp4 and Angptl4) in the fasting group (Fig. 4b and Supplementary Figure 7b). KEGG analyses further demonstrated that gene subsets associated with the TGF-β signaling pathway, the PPAR signaling pathway, and low-energy-related pathways (i.e. autophagy, FoxO signaling, and AMPK signaling) were enriched during fasting (Fig. 4c).

Figure 4.

Bulk-RNA-seq analysis of CXCL13-EYFP cells in bone defines transcriptomic profiles of CXCL13-producing cells. (a–c) CD45⁻ CD31⁻ Ter119⁻ EYFP⁺ cells were isolated from the bone of Cxcl13-EYFP mice fasted for 36 h, and refed for 48 h. These cells were subjected to transcriptome analysis by bulk RNA sequencing (RNA-seq). The transcriptome data were subjected to principal-component analysis (a), volcano plot (b), and KEGG pathway enrichment analysis (c). (d–e) Mice were intraperitoneally administered Compound C (AMPK inhibitor), BMS-309403 (FABP4 inhibitor), AS1842856 (FoxO1 inhibitor), DG-172 dihydrochloride (PPARβ/δ inhibitor), GW9662 (PPARγ inhibitor), SIS3 (Smad3 inhibitor), or vehicle control twice (at 0 and 18 h post-fasting). Mice were euthanized 36 h post-fasting for flow cytometric analysis of BM cell subsets (n = 5–15 per group) (d). The concentration of CXCL13 in bone tissue was measured by ELISA (n = 3–13 per group) (e). Data are pooled from several independent experiments. Data represent the means ± SEM. Statistical significance was determined by one-way ANOVA followed by Dunnett’s test for comparison with the fasting-vehicle group (d, e). *P < .05; **P < .01.

Finally, we sought to determine the functional relevance of these pathways in fasting-induced CXCL13 production and naïve B cell migration. We treated mice with inhibitors targeting the enriched signaling pathways under fasting conditions. Specifically, we employed the following pharmacological agents: Compound C (an AMPK inhibitor), BMS309403 (a FABP4 inhibitor), DG172 dihydrochloride (a PPARβ/δ inhibitor), GW9662 (a PPARγ inhibitor), AS1842856 (a FoxO1 inhibitor), and SIS3 (a Smad3 inhibitor to block TGF-β signaling). Among the inhibitors tested, treatment with GW9662 and SIS3 partially but significantly reduced the number of naïve B cells in the BM during fasting (Fig. 4d). Consistent with these observations, bone CXCL13 production was attenuated in the mice treated with GW9662 and SIS3 (Fig. 4e). Of note, combined treatment with GW9662 and SIS3 did not result in additive or synergistic suppression, showing an inhibitory effect comparable to either inhibitor alone (Fig. 4d and e). Other inhibitors targeting AMPK, FABP4, FoxO1, and PPARβ/δ did not significantly reduce CXCL13 levels (Fig. 4e). Together, these results suggest that CXCL13 expression in bone is regulated by multiple pathways, including TGF-β and PPARγ signaling.

Discussion

We have demonstrated that fasting specifically promotes naïve B cell migration to the BM via the CXCL13–CXCR5 axis. This process is driven by upregulation of CXCL13 in a distinct population of osteogenic MSCs located within the perivascular niche. Both the TGF-β-Smad3- and PPARγ-dependent pathways play significant roles in regulating this process. These findings provide a mechanistic link between systemic metabolic changes and local chemokine production.

Immune cell trafficking is a highly coordinated process primarily governed by two major pathways: the S1P–S1P₁ axis and diverse chemokine-chemokine receptor interactions. The S1P–S1P₁ axis is well known to mediate lymphocyte egress from peripheral lymphoid tissues into the circulation under physiological conditions (34). However, in our previous study, inhibition of the S1P–S1P₁ axis by FTY720 failed to prevent fasting-induced migration of naïve B cells into the BM, indicating that the S1P–S1P₁ axis is dispensable for this process (2). This observation prompted us to focus on chemokine-chemokine receptor interactions as key regulators of naïve B cell dynamics in response to fasting. The CXCL13–CXCR5 axis is canonically known to organize B cell areas within secondary lymphoid organs like peripheral lymph nodes and the spleen (13, 15, 35). Our study uncovers a previously uncharacterized role of the CXCL13–CXCR5 axis in directing naïve B cell migration specifically into the BM during fasting. These findings broaden our understanding of the mechanisms regulating naïve B cell trafficking in response to metabolic shifts.

A variety of immune cells migrate into the BM during fasting or low-energy conditions (2–5). We also observed that 36-h fasting increases the number of naïve B cells, CD4⁺ naïve T cells, and CD8⁺ memory T cells in the BM. The CXCL12–CXCR4 axis is one of the primary mechanisms governing immune cell homing and retention within the BM niche (6, 11, 36, 37). In line with previous studies, treatment with a CXCR4 antagonist decreased the number of CD8⁺ Tem in the BM during fasting. However, this treatment failed to suppress the accumulation of naïve B cells in the BM, suggesting that the CXCL12–CXCR4 axis is not a prerequisite for their migration into the BM. In contrast, CXCL13 neutralization blocked naïve B cell migration into the BM during fasting. These observations highlight the context-dependent and cell-type-specific roles of chemokine axes.

Accumulating studies have demonstrated that CXCL13 is produced by both hematopoietic cells (e.g. T follicular helper cells, peripheral helper T cells, or lymphoid tissue inducer) and non-hematopoietic stromal cells (e.g. FDCs or MRCs) (17, 18, 38–42). We identified a specific population of osteogenic MSCs within the perivascular niche as the primary source of CXCL13 in the BM during fasting. This finding adds to the growing body of evidence that bone-resident stromal cells function as active and dynamic regulators of immune cell migration. Distinct stromal cell subsets have been reported to actively regulate specific hematopoietic and immune cell populations (43, 44). Among them, CAR cells and other stromal cells are crucial for retaining hematopoietic stem cells (8, 10, 11, 45, 46). Notably, our imaging analysis demonstrated that CXCL13-producing cells display a stromal cell-like morphology and are located adjacent to CD31⁺ vessels throughout the bone. Furthermore, their gene expression profile was distinct from that of CAR cells, highlighting the presence of a distinct stromal subset. The perivascular niche, where CXCL13-producing cells reside, serves as a dynamic hub for immune–hematopoietic interactions (47–49). For instance, BM-resident dendritic cells in the perivascular niche provide critical survival signals to B cells (48). This niche also sustains the granulocyte/macrophage progenitors undergoing proliferative bursts (49). CXCL13-producing stroma cells may contribute to directing naïve B cells to the perivascular niche. Such spatial organization likely ensures that naïve B cells have access to metabolic and survival signals from the surrounding vasculature, which may be particularly important during periods of systemic energy deprivation.

During fasting, PPs and BM exhibit reciprocal expression patterns of Cxcl13 (2), reflecting distinct cellular sources and regulatory mechanisms in each tissue. Regarding the cellular sources, CXCL13 in PPs is primarily produced by FDCs and MRCs (17, 18). This cell population resides in a bone-specific microenvironment that is fundamentally distinct from that of secondary lymphoid organs. In PPs, the production of CXCL13 is tightly regulated by the lymphotoxin-α1β2 (LTα1β2)−lymphotoxin β receptor (LTβR) signaling. LTα1β2, produced by follicular B cells and lymphoid tissue inducer cells, maintains the high Cxcl13 expression in FDCs and MRCs. We previously found that under fasting conditions, follicular B cells—the primary source of LTα1β2 (13)—undergo rapid apoptosis (2), leading to downregulation of Cxcl13. Meanwhile, the current study illustrates that CXCL13 production in the bone tissue is driven by TGF-β and PPARγ signaling pathways. Unlike PPs, the bone stromal cells are less dependent on lymphocyte-derived lymphotoxin signals for chemokine regulation. Osteogenic MSCs appear to directly sense systemic metabolic changes during fasting and activate TGF-β- and/or PPARγ signaling, resulting in upregulation of Cxcl13. Collectively, CXCL13 expression in peripheral lymphoid tissues (i.e. PPs) largely relies on LTα1β2-driven feedback, whereas metabolic sensing sustains CXCL13 expression in the primary lymphoid tissue (i.e. BM) during nutrient stress. This divergence likely underlies the reciprocal expression patterns of Cxcl13.

Our finding that naïve B cells migrate to the BM via the CXCL13–CXCR5 axis during fasting prompts a key question regarding its biological relevance. We hypothesize that this trafficking represents an adaptive survival strategy that enables naïve B cells to localize to a nutrient-rich, protective niche during fasting. The BM serves as a well-established niche for long-lived immune cells, such as plasma cells, which rely on the local microenvironment for metabolic support and protection from apoptosis (50–52). By migrating to this specialized niche, naïve B cells may be better equipped to withstand the metabolic stress of fasting, thereby preserving a vital component of the adaptive immune system. This may represent an evolutionarily conserved strategy to maintain a ready reserve of naïve lymphocytes, ensuring a rapid and robust immune response upon nutrient restoration.

A key aspect of our findings is that TGF-β signaling plays a role in promoting CXCL13 production by stromal cells. Our observation that Cxcl13-expressing osteogenic MSCs co-expressed Tgfbr2 supports this conclusion. TGF-β is a pleiotropic cytokine, and its role in inducing Cxcl13 expression has been noted in the context of tumor immunology, where TGF-β derived from cancer-associated fibroblasts drives Cxcl13 expression in activated T cells (53). Furthermore, TGF-β facilitates the differentiation of human naïve CD4⁺ T cells into CXCL13-producing CD4⁺ T cells (54). Consistently, treatment with the SMAD3 inhibitor SIS3 suppresses the generation of CXCL13-producing CD4⁺ T cells. These prior studies, however, primarily focused on its effects in pathological or inflammatory conditions. Its role in BM stromal cells under non-pathological settings, such as changes in nutritional status, had been largely unexplored. Our study demonstrates a physiological pathway in which TGF-β acts on bone-resident stromal cells to induce CXCL13 production. However, how TGF-β stimuli induce CXCL13 production in stromal cells remains an open question, and further investigation is warranted.

Moreover, the involvement of PPARγ, a master regulator of lipid metabolism, provides a critical mechanistic link between the metabolic state and CXCL13 induction. PPARγ2 binding sites in the Cxcl13 promoter are known to be involved in the activation of Cxcl13 transcription in adipose tissue (55). Given that PPARγ senses lipid-derived signals, its activation in osteogenic MSCs during fasting represents a highly plausible mechanism for translating systemic energy stress into local CXCL13 expression. However, the precise upstream signals that trigger PPARγ activity in response to fasting remain an open question and warrant further research.

Our findings indicate that pharmacological inhibition of the TGF-β and PPARγ pathways significantly attenuates fasting-induced CXCL13 production and naïve B cell migration. However, even combined inhibition did not fully restore these parameters to the baseline observed in ad libitum-fed mice, suggesting that fasting-induced naïve B cell recruitment to the BM is mediated by multiple pathways rather than a single dominant pathway. Thus, the TGF-β and PPARγ axes contribute to, but do not fully account for, CXCL13 production and naïve B cell migration during fasting.

We acknowledge limitations that warrant future investigation. Our findings open up several avenues for future research. While we identified the TGF-β-Smad3 pathway and the PPARγ as key factors for Cxcl13 expression, the downstream molecular mechanisms that cooperatively induce Cxcl13 in response to fasting remain to be elucidated. Future studies should examine whether Smad3 and PPARγ directly access the promoter and enhancer regions of the Cxcl13 locus to modulate its transcriptional activity. Moreover, the functional consequences of naïve B cell homing to the BM warrant further investigation. Do these cells exhibit enhanced survival capacity, different activation thresholds, or distinct proliferative capacity upon refeeding? It will be important to determine whether this strategic relocation protects the naïve B cell pool from metabolic stress and primes them for a more rapid and robust response once nutrient availability is restored.

In conclusion, our study uncovers a physiological mechanism linking systemic metabolic adaptation to immune homeostasis. We show that fasting induces CXCL13 production in osteogenic MSCs through TGF-β–Smad3 and PPARγ signaling, thereby directing naïve B cell trafficking into the BM. This adaptive response allows naïve B cells to reside in a metabolically protective niche during energy deprivation, highlighting a dynamic crosstalk between metabolism and immune cell positioning.

Supplementary data

Supplementary data are available at International Immunology online.

Funding

This work was supported by Grant-in-Aid for Japan Society for the Promotion of Science Fellows (JP23KJ1914) (T.O.); Keio University Doctorate Student Grant-in-Aid Program (all graduate school selection) (T.O.); a Japan Society for the Promotion of Science Research Fellowship (M.N.); Japan Agency for Medical Research and Development (22gm1310009h0003, 23gm1310009h0004, and 24gm1310009h0005) (K.H.); Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research (22K19445, 23H05482, and 25K22561) (K.H.); JST CREST (JPMJCR19H1) (K.H.); and the Fuji Foundation for Protein Research (K.H.).

Data availability

The raw data of scRNA-seq generated during this study can be found using the DDBJ Sequence Read Archive (DRA) accession number DRA026116 (scRNA-seq). This article does not report original code. Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Author contributions

T.O. and M.N. designed and performed most of the experiments. T.O. performed most of the data analysis and wrote the manuscript. K.N., R.M., and H.S. helped with the animal experiments. S.F. performed the 3D immunohistochemistry. T.O., S.K., and D.T. performed the transcriptional analysis. B.L. provided the Cxcl13-EYFP mice and discussed data. K.T., T.D., and K.S. discussed the data. K.H. and Y.I.K. provided the experimental resources and discussed the data. K.H. supervised the study, interpreted the data, and revised the manuscript.

Ethics statement

All animal experiments were performed in accordance with the Institutional Guidelines for the Care and Use of Laboratory Animals in Research, with the approval of the local ethics committees at the National Center for Global Health and Medicine and Keio University.