PV1+ fenestrated capillaries at the periphery of the synovium allow access to circulating stimuli

To interrogate the synovial vasculature via which circulating stimuli gain access to the synovium, we analyzed endothelial cells from single-cell RNA-sequencing (scRNA-seq) data of mouse synovium6 (Extended Data Fig. 1a). Genes encoding adhesion molecules, such as Sele, Selp and Icam1, were mainly expressed on Ackr1+ postcapillary venules, where immune cells extravasate19 (Extended Data Fig. 1b). Capillary endothelial cells were also evident and, when considered in isolation, comprised two subsets (Fig. 1a). Plvap was exclusively expressed in cluster 1 and was among the top ten differentially expressed genes (DEGs) when comparing these two capillary cell clusters (Fig. 1b,c and Extended Data Fig. 1c). PV1 (encoded by Plvap) is a molecular component of fenestral diaphragms, conferring vascular permeability20,21, and its expression is indicative of fenestrated capillaries.

Fig. 1: PV1+ fenestrated capillaries in the L–SL interface at the peripheral area of the synovium allow circulating stimuli to access the synovium.

a, Uniform manifold approximation and projection (UMAP) visualization of synovial capillary endothelial cells extracted from CD31+ endothelial cells (Extended Data Fig. 1a,b). scRNA-seq data are from GSE145286. b, Volcano plots showing DEGs between two clusters of synovial capillary endothelial cells. c, UMAP visualization of synovial capillary endothelial cells expressing Plvap. Color bar shows the expression level. d, Representative confocal images of sections of knee joints; BM, bone marrow; M, meniscus; L, patella ligament; Syn, synovium; n = 3 mice for each group. e, Schematic diagram showing the protocol to dissect whole synovium from knee joints. The red dashed outline indicates the area of synovium dissected. Fe, femur; Ti, tibia; P, patella; Fi, fibula; Prox, proximal; Lat, lateral; Dis, distal. f, Three-dimensional reconstruction of representative confocal images of the indicated layer of whole-mount synovium. Quantification of the PV1+ area among CD31+ area at the indicated layers is shown on the bottom right; n = 5 mice for each group. g, Three-dimensional reconstruction of representative confocal images and density map of the L–SL interface of whole-mount synovium. Quantification of the PV1+ area among CD31+ area at the indicated compartments is shown on the bottom right; n = 5 mice. h, Representative confocal images of sections of knee joints from WT mice injected i.v. with 70- and 2,000-kDa dextran (300 μg of 70-kDa dextran (Dex70k) and 150 μg of 2,000-kDa dextran (Dex2000k)) 1 h before analysis. Arrowheads indicate the area where 70- and 2,000-kDa dextran merged; CL, cruciate ligament; Epi. GP, epiphyseal growth plate; scale bars, 500 and 100 μm (inset). i, Quantification of the extravasated area in each tissue; n = 4 to 6 mice for each group. j, Pie graph showing the percentage of distance between 2,000-kDa dextran and the ERTR7+ lining layer of the synovium in the section images; n = 4 mice. k, Three-dimensional reconstruction of representative confocal images of whole-mount synovium from WT mice injected i.v. with 70- and 2,000-kDa dextran 1 h before analysis; scale bar, 100 μm. l, Three-dimensional reconstruction of representative confocal images of whole-mount synovium from WT mice injected i.v. with fluorescently labeled microbeads of different sizes (25 μl of each FluoSphere carboxylate-modified microsphere dissolved in PBS) 1 h before analysis. Arrowheads indicate the sites where microbeads extravasated; scale bars, 500 and 50 μm (inset). Quantification of the area and capillary microbeads extravasated is shown on the bottom right; n = 3 to 5 mice for each group. m, Three-dimensional reconstruction of representative confocal images of whole-mount synovium from WT mice injected i.v. with OVA–AF647;rabbit polyclonal anti-OVA (RaOVA) ICs (40 μg of OVA–AF647 + 150 μg of RaOVA) 2 h before analysis. Arrowheads indicate sites where ICs extravasated; scale bars, 200 and 100 μm (z-stack images). n, Density map of a three-dimensional reconstruction of representative confocal images of whole-mount synovium from WT mice injected i.v. with microbeads or ICs. Data in d, f and l (right) were analyzed by two-tailed t-test. The center compartment was used as a control group in one-way analysis of variance (ANOVA) with a Dunnett’s test for g. Data in i and l (left) were analyzed by one-way ANOVA with Tukey’s post hoc test, and data in b were analyzed by two-tailed Wilcoxon rank-sum test. Data in d, f, g, i and l are shown as mean ± s.e.m. Images in k, m and n are representative of at least three independent experiments with similar results.

Source data

The synovium can be divided into a lining layer containing synovial fibroblasts, with overlying macrophages that interface with the synovial cavity, and a sublining layer with additional macrophage populations7,22,23,24. Sagittal section imaging showed that PV1+ capillaries were specifically located at the interface of the lining and sublining layers, termed hereafter as the lining–sublining (L–SL) interface (Fig. 1d). Although sagittal sections have been used as the gold standard to define the spatial organization of synovial cells, this method provides a limited snapshot of the synovial membrane as a whole, which stretches around the entire joint. Indeed, our knowledge of the global arrangement of synovial cells, particularly in homeostasis, is surprisingly rudimentary. To address these limitations, we developed a protocol to stereotactically dissect the entire synovium of knee joints (Fig. 1e and Extended Data Fig. 1d,e) and combined whole-mount synovial imaging with an iterative bleaching and staining protocol25 to enable multiparameter imaging. This revealed a dense PV1– vascular network in the sublining layer and confirmed that PV1 was predominantly expressed on capillaries at the L–SL interface (Fig. 1f and Extended Data Fig. 1f). When considering the L–SL interface across the entire area of the synovium, we found that PV1+ capillaries were not uniformly distributed but rather were abundant at the periphery of the synovium in proximity to adjacent bones (Fig. 1g).

To test the functional importance of the distribution of PV1 capillaries, we administered 70-kDa and 2,000-kDa dextran intravenously (i.v.) and collected organs 1 h later. Although 70-kDa dextran highly extravasated at the diaphysial side of the growth plate in bone marrow, 2,000-kDa dextran was evident in the synovium (Fig. 1h,i and Extended Data Fig. 1g,h), predominantly in the L–SL interface (Fig. 1j,k). Similarly, i.v. administration of fluorescently labeled microbeads showed that 0.2-μm microbeads extravasated exclusively from PV1+ capillaries in the L–SL interface in the periphery of healthy synovium (Fig. 1l), whereas 2-μm microbeads were excluded (Extended Data Fig. 1i). Finally, we used a clinically relevant challenge; i.v. ICs (ovalbumin (OVA) opsonized with a polyclonal anti-OVA IgG26) also extravasated into the local synovial tissue from PV1+ capillaries (Fig. 1m,n). Together, these data indicate that circulating stimuli readily gain access to the healthy synovium via highly permeable PV1+ capillaries located at the L–SL interface in the periphery of the synovium.

Three subsets of macrophages with distinct distribution patterns line synovial PV1+ capillaries

We next sought to interrogate which immune cells localized to these sites of potential vulnerability around PV1+ capillaries. ERTR7+ lining fibroblasts evident in sagittal sections (Fig. 2a) formed a uniform, tightly knit lining layer in the synovial whole-mount images (Fig. 2b), but CD68+ macrophages showed increased density in the periphery of the synovium (Fig. 2b and Extended Data Fig. 3a). No extravascular synovial T or B cells were detectable (Fig. 2c). We therefore focused on further characterizing the synovial macrophage populations.

Fig. 2: Three subsets of macrophages with distinct distribution patterns line synovial PV1+ capillaries.

a, Representative confocal images of sections of healthy knee joints; BM, bone marrow; Syn, synovium; L, patella ligament; scale bar, 200 μm. b, Three-dimensional reconstruction of representative confocal images of whole-mount synovium; scale bars, 200 (left) and 300 μm (right). c, Three-dimensional reconstruction of representative confocal images of whole-mount synovium; scale bars, 200 and 10 μm. d, Gating strategy and flow cytometric analysis of three subsets of synovial macrophages with indicated cell surface markers. Shaded regions indicate staining with isotype controls. Data are representative of at least two independent experiments with similar results. The asterisks (*) indicate that fluorophores for LYVE1 were changed to apply the indicated antibodies. Representative confocal images of each subset of macrophages are shown on the right. e, Three-dimensional reconstruction of representative confocal images of whole-mount synovium from a vertical angle. The two dashed lines are (1) the border of lining layer and L–SL interface, and (2) L–SL interface and sublining layer. Cap., capillary; Art., arteriole. f, Three-dimensional reconstruction of representative confocal images of whole-mount synovium from a vertical angle; scale bars, 100 μm. Quantification of the percentage of CD11c+MHCII+ and CD11c–MHCII+ macrophages (Mac) attached to PV1+CD31+ or PV1−CD31+ vessels; n = 3 mice for each group. Data represent mean ± s.e.m. g, Three-dimensional reconstruction of representative confocal images at the indicated layers of whole-mount synovium. The pie graphs show the percentages of the three types of macrophages in the indicated layers of the synovium; n = 3 mice. h, Three-dimensional reconstruction of representative confocal images of whole-mount synovium (left) and density maps of three subsets of macrophages (middle); scale bar, 200 μm. The numbers and densities of each macrophage type in the indicated compartment of the synovium are shown on the right; n = 4 mice for each group. i, Representative confocal images of sections of healthy knee joints; M, meniscus; C.Lig, crescent ligament; SC, synovial cavity; n = 3 mice for each group. Data represent mean ± s.e.m. Data in f and i were analyzed by two-tailed t-test. Images in a–e are representative of at least two independent experiments with similar results.

Source data

scRNA-seq data showed that synovial macrophages (Cd68+Adgre1+) can be divided into a Cx3cr1+ population (previously described as lining macrophages7), which also express Lyve1, Timd4 and Folr2, an H2-Ab1+ population (previously described as interstitial macrophages7) and a subpopulation within the H2-Ab1+ population expressing Itgax (encoding CD11c; Extended Data Fig. 2a–c). Unsupervised assessment of myeloid surface markers using flow cytometry also showed that myeloid cells in the healthy synovium can be divided into major histocompatibility complex class II+CD11c– (MHCII+CD11c–), MHCII+CD11c+ and MHCII–LYVE1+ clusters, which also express CX3CR1 (Extended Data Fig. 2d,e). These data informed the gating strategy subsequently used in our study characterizing three subsets of synovial macrophages, CD11c+IAIE(MHCII)+, CD11c–MHCII+ and MHCII–Ly6C−LYVE1+CX3CR1+cells, which were phenotypically and morphologically distinct (Fig. 2d). MHCII–LYVE1+ macrophages expressed CX3CR1 and TIM4, whereas CD206 and CD64 were expressed in all three populations.

To investigate the spatial distribution of macrophages relative to PV1+ capillaries, we performed whole-mount synovial imaging vertically at the edge of whole-mounted synovium. This showed that α-smooth muscle actin+ (αSMA+) arterioles give rise to PV1+ capillaries at the L–SL interface, which were surrounded by LYVE1+ macrophages (Fig. 2e). CD11c–MHCII+ macrophages lined both PV1– and PV1+ vasculature, whereas CD11c+MHCII+ mononuclear phagocytes (MNPs) specifically localized to PV1+ vessels (Fig. 2f). Therefore, all three macrophage subsets were found in proximity to PV1+ capillaries in the L–SL interface (Fig. 2g). When considering the synovium as a whole, lining macrophages varied in their distribution density, with far fewer cells in the central region (Fig. 2h). In the sublining layer, CD11c+MHCII+ MNPs predominantly localized in the periphery, whereas CD11c–MHCII+ macrophages were uniformly distributed throughout the synovium (Fig. 2h). Sagittal sections confirmed that the density of lining macrophages and the thickness of the lining layer were significantly greater at the peripheral area near the meniscus between the femur and tibia than at the central area (Fig. 2i).

Bulk RNA-seq of the sorted macrophage subsets showed that they were transcriptionally distinct, with LYVE1+CX3CR1+ macrophages expressing markers of lining macrophages (Cx3cr1, Vsig4 and Sparc) and CD11c+MHCII+ and CD11c–MHCII+ macrophages demonstrating transcriptional similarity to interstitial macrophages (H2-Ab1 and Retnla7; Fig. 3a–d). Compared to reference transcriptional signatures obtained from macrophages activated with different stimuli27, CD11c+MHCII+ MNPs were enriched for gene sets associated with M2 stimuli (interleukin-4 (IL-4) and IL-13), whereas LYVE1+CX3CR1+ macrophages were enriched for tumor necrosis factor (TNF) and high-density lipoprotein stimulation signatures (Fig. 3e,f). Pathway analysis showed that LYVE1+CX3CR1+ macrophages expressed genes related to endocytosis and the lysosome, whereas the MHCll+ subsets were enriched for gene sets related to cell adhesion and antigen processing and presentation (Fig. 3g).

Fig. 3: Three subsets of synovial macrophages show distinct transcriptomes and ontogenies.

a, Illustration of the experimental protocol. b, Principal component analysis (PCA) of three subsets of synovial macrophages by RNA-seq; n = 3 mice for each plot and n = 9 mice for each population. c, Volcano plots showing DEGs between LYVE1+CX3CR1+, MHCII+CD11c− and MHCII+CD11c+ macrophages from WT mice; Padj, adjusted P value. d, Heat map of the expression of canonical macrophage genes (normalized values) and dendritic cell markers in LYVE1+CX3CR1+, MHCII+CD11c− and MHCII+CD11c+ macrophages from bulk RNA-seq analysis. e, Heat map of single-sample gene set enrichment analysis (ssGSEA) of three synovial macrophage subsets by RNA-seq. The signature genes from a previously published dataset (Xue et al.27) describing the transcriptional programs activated with 28 different stimuli were used; TPP, TNF+PGE2+P3C; IFN, interferon; PA, palmitic acid; LPS, lipopolysaccharide; TNF, tumor necrosis factor; GC, glucocorticoid; HDL, high-density lipoprotein; P3C, Pam3CysSerLys4; OA, oleic acid; Lia, linoleic acid; LA, lauric acid; sLPS, standard lipopolysaccharide; upLPS, ultrapure lipopolysaccharide. f, Quantification of ssGSEA scores for signaling pathways of the indicated stimuli for each subset; n = 3 mice for each plot and n = 9 mice for each population. Data represent mean ± s.e.m. g, Heat map of ssGSEAs of three synovial macrophage subsets with KEGG enrichment analysis (scaled normalized values). h, Illustration of the experimental protocol; FCM, flow cytometry. i, Flow cytometric analysis of MS4A3–tdTomato positivity of the indicated macrophage subsets from 10-week-old mice; n = 3 mice for each group. Data represent mean ± s.e.m.; NC, negative control; mono, monocytes; PC, positive control. j, Three-dimensional reconstruction of representative confocal images of whole-mount synovium and density map of MS4A3–tdTomato. Images are representative of three animals with similar results. Data in f were analyzed by one-way ANOVA with Tukey’s post hoc test, and data in c were analyzed by Wald test.

Source data

We next investigated the kinetics of synovial macrophage replenishment by circulating monocytes using Ms4a3Cre-RosaTdT mice. Ms4a3 is specifically expressed by granulocyte–monocyte progenitors, and this model efficiently fate maps monocytes and granulocytes but not lymphocytes or tissue dendritic cells28. In accordance with published reports showing that MHCII+ macrophages are replaced faster than MHCII– macrophages in other organs28, in the synovium, almost half of MHCII+ macrophages were tdTomato+, whereas most of LYVE1+CX3CR1+ macrophages were tdTomato– (Fig. 3h,i). An equivalent proportion of MHCII+CD11c+ MNPs were tdTomato+ compared to MHCII+CD11c– macrophages, indicating that MHCII+CD11c+ MNPs have an ontogenic profile similar to macrophages rather than dendritic cells. Whole-mount synovial imaging of Ms4a3Cre-RosaTdT mice showed that tdTomato+ cells were predominantly distributed at the periphery of the synovium (Fig. 2j and Extended Data Fig. 3b).

Synovial macrophages sample circulating ICs and present antigens

Given their proximity to PV1+ capillaries, we next investigated the capacity of synovial macrophages to take up circulating ICs in vivo. Two hours following i.v. administration, ICs extravasated from PV1+ capillaries at the L–SL interface (Fig. 4a,b), and uptake was evident mainly in LYVE1+CX3CR1+ macrophages, consistent with a previous report24. CD11c+MHCII+ and CD11c–MHCII+ macrophages were also capable of internalizing ICs, with a greater IC uptake than OVA alone (Fig. 4c,d and Extended Data Fig. 3c). Notably, CD11c+MHCII+ MNPs were also able to internalize and present circulating peptide (Eα; Fig. 4e–g), consistent with their enrichment for antigen processing and presentation gene sets (Fig. 3g).

Fig. 4: Synovial macrophages sample circulating ICs, and MHCII+ macrophages present antigens.

a, Three-dimensional reconstruction of representative confocal images of the indicated layers and vertical views of whole-mount synovium from WT mice injected i.v. with OVA–AF647;RaOVA (40 μg of OVA–AF647 + 150 μg of RaOVA) 2 h before analysis. Quantification of the percentage of OVA-IC+ area within PV1+ and PV1−CD31+ area and the percentage of OVA-IC+ area in the lining layer, L–SL interface and sublining layer is shown on the right; n = 4 mice for each group. b, Schematic diagram showing the protocol. c, Pie graph showing the mean percentage of OVA-IC+ macrophage subsets among all OVA-IC+ cells; n = 3 mice. d, Scatter plots of mean fluorescence intensity (MFI) of OVA–AF647 and OVA–AF647;RaOVA; n = 3 mice for each group. e, Schematic diagram showing the protocol of antigen presentation in vivo using the Eα:YAe system. f,g, Flow cytometric analysis (f) and quantification (g) of YAe MFI of the indicated macrophage subsets from mice injected i.v. with Eα divided by that observed in macrophages from mice injected with PBS. Shaded regions indicate mice injected with PBS control; n = 3 mice for each group. h, Flow cytometric analysis of different types of synovial macrophages with indicated FcγRs. Cyan regions indicate staining with isotype controls. Scatter plots show the MFI ratio of each FcγR and isotype controls on each subset; n = 3 mice for each group. i, Three-dimensional reconstruction of representative confocal images of the indicated layers of whole-mount synovium; scale bars, 50 μm. Quantification of the percentage of FcγRllb+ area in the indicated layers is shown on the right; n = 4 mice for each group. j, Flow cytometric analysis of different types of synovial macrophages with the indicated FcγRs before and 24 h after IC injection. Gray regions indicate staining with isotype controls. The FcγR A:I ratios were calculated according to MFI ratios of activating (FcγRlll and FcγRlV) and inhibitory FcγRllb before and 24 h after IC injection on each subset; n = 3 mice for each group. k, Three-dimensional reconstruction of representative confocal images of whole-mount human synovium; scale bars, 100 (left) and 30 μm (right); L, lining layer; SL, sublining layer; SC, synovial cavity. l, Three-dimensional reconstruction of representative confocal images of whole-mount human synovium. Quantification of PV1+ area among CD31+ area in each layer; n = 3 individuals for each group. m, Three-dimensional reconstruction of representative confocal images of whole-mount human synovium. Quantification of LYVE1+ and HLA-DR+ area in the visual field is shown on the right; n = 3 individuals for each group. n, Three-dimensional reconstruction of representative confocal images of whole-mount human synovium; scale bars (right), 100 μm. Images are representative of at least two independent experiments with similar results. The arrowheads indicate the merged area for LYVE1 and CD32B. Data in a and m were analyzed by two-tailed t-test, and data in d and g–i were analyzed by one-way ANOVA with Tukey’s post hoc test. Data in a, d, g–j, l and m are shown as mean ± s.e.m.

Source data

Because Fcγ receptors (FcγRs) bind to the Fc portion of IgG to mediate the cellular effector responses to ICs, we analyzed their expression in synovial macrophages. Of note, the relative expression of activating receptors (FcγRIII/FcγRIV) and the inhibitory receptor (FcγRIIb) determines the activation threshold of a cell when encountering IgG ICs, termed the A:I ratio (Extended Data Fig. 3d)29. Flow cytometric analysis showed that LYVE1+CX3CR1+ macrophages expressed higher levels of FcγRIIb and FcγRIII than the other two subsets, but FcγRIV expression was absent in synovial macrophages (Fig. 4h). Whole-mount synovium imaging showed FcγRII/FcγRIII expression mostly at the L–SL interface on LYVE1+ macrophages (Fig. 4i and Extended Data Fig. 3e,f). Circulating ICs led to downregulation of FcγRIII expression in all three subsets, whereas the expression of the inhibitory receptor (FcγRIIb) was maintained (Fig. 4j), thereby increasing the activation threshold and preventing excessive (and potentially damaging) responses to circulating ICs.

To determine if this microarchitectural arrangement was present in human synovium, we optimized a whole-mount imaging system for human knee joint synovium using clearing-enhanced 3D (Ce3D), a tissue clearing solution (Extended Data Fig. 3g)30. Beneath the CD55+ lining layer fibroblasts6 (Fig. 4k and Extended Data Fig. 3h), an abundant PV1+ capillary network was present (Fig. 4l) and was tightly covered with LYVE1+ and HLA-DR+ macrophages and some LYVE1+ macrophages expressing CD32B (FCGR2B), whereas fat pad capillaries were scantily lined with perivascular macrophages (Fig. 4m,n), analogous to our observations in mouse synovium.

Systemic IC challenge induced distinct responses in synovial macrophage subsets

To further interrogate synovial macrophage responses to ICs, we performed bulk RNA-seq on sorted subsets following i.v. IC challenge in vivo in wild-type (WT) and FCGR2B-deficient (Fcgr2b−/−) mice (Fig. 5a). IC challenge was associated with increased Fcgr2b expression in LYVE1+CX3CR1+ macrophages in WT mice, consistent with a negative feedback loop (Fig. 5b). Less than 3% of DEGs were shared among the three macrophage subsets in both WT and FCGR2B-deficient mice, indicating subset-specific responses to circulating IC challenge (Fig. 5c). CD11c+MHCII+ and CD11c–MHCII+ macrophages shared more DEGs, including Fcgr1, Mmp19, Irf7 and Cxcl13, than LYVE1+CX3CR1+ macrophages (Fig. 5c and Extended Data Fig. 4a). Pathway enrichment analysis using upregulated genes showed that LYVE1+CX3CR1+ macrophages were enriched with gene sets related to neutrophil migration, whereas CD11c+MHCII+ macrophages were enriched with gene sets related to cell adhesion and migration after IC stimulation (Fig. 5d).

Fig. 5: Systemic IC challenge induces distinct responses in synovial macrophage subsets.

a, Schematic diagram showing the protocol for bulk RNA-seq. b, Volcano plot showing DEGs due to OVA-IC stimulation in LYVE1+CX3CR1+ macrophages from WT mice by RNA-seq; Sig., significantly. c, Venn diagram showing the number of common DEGs affected by IC stimulation between LYVE1+CX3CR1+, MHCII+CD11c− and MHCII+CD11c+ macrophages in WT and Fcgr2b−/− mice. d, Gene ontogeny (GO) analysis of DEGs specific to each macrophage type with all the DEGs of three macrophage subsets as the background gene list; commun., communication; stim., stimulation; Pos, positive; O/E, observed/expected. e, Number of DEGs in each synovial macrophage from WT and Fcgr2b−/− mice and common DEGs in both strains. f, Heat map of the expression of chemokines (scaled normalized values) with or without IC injection in LYVE1+CX3CR1+ macrophages from WT and Fcgr2b−/− mice. g, Ratio of mean Cxcl1 expression in LYVE1+CX3CR1+, MHCII+CD11c− and MHCII+CD11c+ macrophages from WT and Fcgr2b−/− mice injected i.v. with or without ICs; stim/unstim, stimulated/unstimulated. h, CXCL1 and CXCL2 enzyme-linked immunosorbent assay (ELISA) of the synovial digestion from Fcgr2b−/− mice with or without IC injection; n = 5 (CXCL1) and 4 (CXCL2) mice for each group. i, Schematic diagram showing the protocol. j, Flow cytometry quantification of synovial neutrophils (Ly6G+ gates) from WT and Fcgr2b−/− mice injected i.v. with PBS, OVA or OVA;RaOVA 6 h before analysis; n = 3 (WT) and n = 6 (Fcgr2b−/−) mice. k, Three-dimensional reconstruction of representative confocal images of whole-mount synovium depicting an MHCII+ macrophage cluster around PV1+ capillaries; scale bars, 200 (left), 50 (top right) and 100 μm (bottom right). Images are representative of at least two independent experiments with similar results. l, Number of MHCII+ macrophage clusters with a diameter of >30 μm in the whole-mount synovium in 4- and 52-week-old mice; n = 5 (4-week-old) and 6 (52-week-old) mice; w.o., weeks old; scale bars, 100 μm. m, Schematic diagram showing the protocol of systemic challenges. n,o, Number and images of MHCII+ macrophage clusters with a diameter of >30 μm in the whole-mount synovium in mice injected i.v. with PBS or OVA-IC over 2 consecutive days and analyzed 24 h after the last injection; n = 5 and 7 mice for each group. p, Number of MHCII+ macrophage clusters in mice infected orally with PBS or 5 × 106S. enterica serovar Typhimurium and analyzed after 3 weeks; n = 5 mice for each group; OG, oral gavage. The arrowheads in l, o and p mark macrophage clusters. q, Number of MHCII+ macrophage clusters in mice inoculated with two doses of 4 × 107 uropathogenic E. coli into the bladder and analyzed after 3 weeks; n = 5 mice for each group. Data in h, l and o–q were analyzed by two-tailed t-test, data in j were analyzed by one-way ANOVA with Tukey’s post hoc test, and data in b were analyzed by Wald test. Data are shown as mean ± s.e.m. in h, j, l and n–q.

Source data

FcγRIIb deficiency was associated with a greater increase in the number of DEGs in LYVE1+CX3CR1+ macrophages than in WT cells (Fig. 5e and Extended Data Fig. 4b), consistent with the high FcγRIIb expression detectable in this population. Because leukocyte activation- and neutrophil migration-associated genes were enriched in LYVE1+CX3CR1+ cells after i.v. IC challenge (Extended Data Fig. 4c), we further probed the expression of chemokines. We found a greater IC-induced upregulation of the neutrophil-recruiting chemokine Cxcl1 in LYVE1+CX3CR1+ and CD11c+MHCII+ macrophages in Fcgr2b−/− mice than in WT mice (Fig. 5f,g and Extended Data Fig. 4d), supporting the conclusion that FcγRllb is important for limiting neutrophil entry after IC challenge. We confirmed a significant increase in CXCL1 and CXCL2 protein expression and infiltrating neutrophils within the synovium of Fcgr2b−/− mice following i.v. IC challenge (Fig. 5h–j).

Interestingly, given the enrichment of cell adhesion and migration gene sets in activated CD11c+MHCII+ macrophages, we occasionally noted clusters of MHCII+ macrophages tightly entwined around PV1+ capillaries in unchallenged synovium, with LYVE1+ macrophages absent from these aggregates (Fig. 5k). The number of these aggregates increased with age (Fig. 5l). We therefore hypothesized that they may arise in response to chronic circulating stimuli, including ICs, or those present in the context of inflammation/infection in distant organs. To test this, we challenged mice with i.v. IC twice over a period of 48 h (Fig. 5m). This resulted in the appearance of MHCII+ macrophage aggregates at 72 h (Fig. 5n,o). Furthermore, oral challenge with Salmonella enterica serovar Typhimurium, a colitogenic bacteria31, also induced synovial macrophage aggregates (Fig. 5p). However, chronic urinary tract infection with uropathogenic Escherichia coli did not induce macrophage aggregates in the synovium (Fig. 5q), suggesting that organ infection variably influences joints.

We next assessed the contribution of circulating monocytes to macrophage aggregate formation. A recent study32 established an in vivo protocol to fluorescently label intravascular leukocytes that were shielded from subsequent rounds of intravascular labeling once they entered tissues. We applied this system in our chronic IC stimulation model to assess the monocyte contribution to synovial macrophage aggregates. We administered i.v. ICs together with phycoerythrin (PE)-labeled anti-CD45 48 h before synovial collection, followed by administration of i.v. ICs and AF488-labeled anti-CD45 24 h before analysis (Extended Data Fig. 5a). Although we detected both 24- and 48-h time stamp signals in the spleen, synovial macrophage aggregates were not labeled with either 24- or 48-h time stamps (Extended Data Fig. 5b). Furthermore, using monocyte reporter Ms4a3Cre-RosaTdT mice, synovial macrophage aggregates were not labeled with tdTomato (Extended Data Fig. 5c), together indicating that these synovial macrophage clusters are derived from a local macrophage pool rather than circulating monocytes.

We next asked whether fibroblasts play any role as secondary effectors, responding to cues produced by FcγR-expressing synovial macrophages after IC challenge (Extended Data Fig. 6a). Bulk RNA-seq of flow-sorted fibroblasts showed 417 IC-induced genes in synovial fibroblasts (Extended Data Fig. 6b,c), including Cxcl1, Cxcl13 and Ccl8 (Extended Data Fig. 6d). Enrichment analysis also showed that inflammatory response gene pathways, including interferon and IL-6–JAK–STAT3 signaling pathways, were upregulated following IC challenge (Extended Data Fig. 6e).

Together, these data show distinct responses of synovial macrophage subsets to circulating IC challenge, with LYVE1+CX3CR1+ macrophages poised to trigger neutrophil recruitment but held in check by FcγRllb expression. By contrast, CD11c–MHCII+ and CD11c+MHCII+ macrophages can present circulating antigens and respond to systemic immune stimuli, including ICs, by forming tight clusters around fenestrated capillaries, thus forming a physical barrier that might limit the spread of potentially harmful blood-borne cargo into the joint. In addition, synovial fibroblasts may function as secondary effectors, responding to macrophage cues.

Synovial macrophages activate nociceptors with IL-1β and nociceptors reciprocally enhance macrophage responses through CGRP

As joint pain is one of the most common features of systemic challenges, we next sought to determine the anatomical relationship between PV1+ capillaries and neurons. Sympathetic tyrosine hydroxylase+ (TH+) neurons primarily colocalized with αSMA+ arterioles in the sublining layer (Extended Data Fig. 7a), whereas calcitonin gene-related peptide+ (CGRP+) nociceptor neuronal fibers branched into the L–SL interface around PV1+ capillaries (Fig. 6a and Extended Data Fig. 7b). Quantification of the spatial location of macrophages, PV1+ vessels, PV1–CD31+ vessels and CGRP+ fibers (Extended Data Fig. 8) indicated that approximately 30–40% of each macrophage subset were in direct contact with PV1+ vessels with LYVE1+CX3