MFAP5+ synovial fibroblasts drive LOX upregulation to promote osteoarthritis progression Authors: Linzhu Wang, Guozhao Wei, Fenghua Ma, Xingyu Li, Huizhen Zhang, Houliang Pan, Zikang Meng, Sen Yang, Ruonan Yang, Juan Dong, Xue Xiong, Zhaoyuan Xu, Boyang Zhang, Hongyi Cheng, Baiyan Li, Lei Yang, Xiaoqiang E, Xuelian Li Published: 2026 ABSTRACT SummarySynovial fibroblasts (SFs), the predominant synovial cell population, drive osteoarthritis (OA) pathogenesis. Understanding SF reprogramming in OA progression may reveal potential targets. Transcriptomic analysis identified preferential infiltration of MFAP5+ SFs in end-stage OA. Immunohistochemistry, quantitative real-time PCR, and immunofluorescence demonstrated MFAP5 upregulation in end-stage OA synovium from rats and patients compared with early-stage OA and controls. Increased MFAP5 and vimentin colocalization was observed in the synovium of end-stage OA patients. MFAP5+ SFs showed elevated LOX expression, and both MFAP5 and LOX were upregulated in lipopolysaccharide (LPS)-stimulated rat SF cell line (iCell-0135a). MFAP5 knockdown attenuated LPS-induced LOX upregulation in iCell-0135a cells, while overexpression enhanced LOX expression. Reduced LOX expression reduced LPS-induced IL-1β upregulation. In vivo, intra-articular administration of a LOX inhibitor alleviated joint damage in OA-model rats. These findings indicate that MFAP5+ SFs with high LOX expression characterize end-stage OA, and pharmacological LOX inhibition offers therapeutic potential for OA.Subject areas: health sciences, biological sciences, cellular physiology, integrative aspects of cell biology Introduction Osteoarthritis (OA) imposes a significant economic burden on society due to its high prevalence and disability,1 especially in the context of an aging global population. OA is primarily characterized by the degeneration of articular cartilage, driven by the excessive recruitment of inflammatory cells and overexpression of metalloproteinases, resulting in permanent pathological changes throughout the entire joint.2,3 Current treatments mainly focus on symptom relief, with prosthetic joint replacement as the option for end-stage patients.4 However, there are no approved therapies to halt or reverse OA progression, underscoring a critical global unmet medical need.5 Advancing treatment therapies requires a deeper understanding of the cellular and molecular changes that occur from early to terminal stages of OA. Research on OA has frequently concentrated on articular cartilage degradation and subchondral bone sclerosis, but the synovium also undergoes pathologic changes.6 Disruption of synovial homeostasis can lead to abnormal matrix deposition, neovascularization, and resident cell alterations, ultimately resulting in fibrosis, thereby worsening OA progression.7 Synovial fibroblasts (SFs), the major tissue-resident cells in the synovium, are crucial in maintaining joint homeostasis and mediating joint damage.8 Based on their localization and function, SFs have been classified into two classical pathological subsets in rheumatoid arthritis and OA.9,10 Recent advances in single-cell RNA sequencing (scRNA-seq) have provided insights into SF heterogeneity in OA, for example, the pro-resolving mesenchymal network formed by CD200+ fibroblasts.11 However, to date, little is known about how SFs alter and the underlying molecular mechanisms throughout the progression from early-stage to end-stage OA. Herein, we utilized publicly available scRNA-seq and bulk RNA sequencing (RNA-seq) datasets to explore SF reprogramming from early-stage to end-stage in OA. Distinct functions of SF clusters were characterized along with their preferential distribution between early-stage and end-stage OA. In addition, synovial tissues from OA patients and monosodium iodoacetate (MIA)-induced OA-model rats were acquired to validate findings derived from public datasets. This study provides a detailed perspective of SFs during OA progression, while also providing valuable resources for future research. Results Highly enriched MFAP5+ SFs in end-stage OA We obtained 1,333 SFs derived from end-stage OA and 2,536 SFs derived from early-stage OA. Addressing concerns about batch effects, it was observed that cells were clustered based on cell clusters rather than samples (Figure S1B). Unsupervised graph-based clustering revealed six SF clusters with specific gene signatures, including DCN+, SLPI+, NUPR1+, MFAP5+, IGFBP3+, and CALM3+ SFs (Figures 1A and 1B). We observed continuous differentiation along SF clusters, with CALM3+ SFs showing the highest differentiation potential, while DCN+ SFs displayed the lowest (Figure 1C). We identified a cluster potentially involved in regulation of CD4+ T and αβT cells (MFAP5+ SFs), as well as a cluster exhibiting aging-related features (DCN+ SFs) (Figure 1D). NUPR1+ SFs showed high expression of apoptotic molecules and preferential infiltration in early-stage OA samples (Figures 1D and 1E). Among SF clusters, DCN+ and MFAP5+ SFs were highly abundant in end-stage samples (Figure 1E). To further explore the transcriptional reprogramming between DCN+ and MFAP5+ SFs, we found that MFAP5+ SFs exhibited lower expression of inflammation-associated genes (CHI3L1 and CXCL12) and higher expression of extracellular matrix (ECM) genes (MFAP5, TNXB, and PCOLCE2) compared to DCN+ SFs (Figure 1F), indicating their distinct effect on the progression of OA. We then employed BayesPrism to infer the presence of SF clusters in GSE176223 dataset. End-stage OA samples showed a higher proportion of MFAP5+ SFs compared to early-stage OA (Figures 1G and 1H, p = 0.051); this trend was also observed at the single-cell level. Furthermore, MFAP5 showed elevated expression in end-stage OA at bulk level (Figure 1I, p < 0.05). Single-cell landscape of SFs at different stages of OA (A) Uniform manifold approximation and projection (UMAP) visualization of SFs in GSE176308. (B) Heatmap showing the expression levels of marker genes across SF clusters in GSE176308. Expression, Z score normalized mean expression. (C) Boxplot illustrating the CytoTRACE scores across SF clusters in GSE176308. (D) Dot plot displaying the top three Biological Process pathways among the six SF clusters in GSE176308. (E) Heatmap showing tissue prevalence, as estimated by Ro/e, for each SF cluster in GSE176308. Top bar plot depicted tissue composition, and left bar plot showed cell composition. (F) Volcano plot showing differentially expressed genes (DEGs) between MFAP5+ SFs and DCN+ SFs, with top 10 genes labeled. Genes with adjusted p < 0.05 (Wilcoxon rank-sum test) and log2(fold change) > 0.1 were color coded. (G and H) Histogram (G) and boxplot (H) showing relative proportions of SF clusters predicted by BayesPrism across different samples in GSE176223. (I) Boxplot depicting MFAP5 expression in early-stage and end-stage OA synovial tissues in GSE176223. Statistical significance was calculated using hypergeometric test (D) and two-tailed Wilcoxon rank-sum test (F, H, and I). ∗p < 0.05. MFAP5+ SFs are reproducible in independent OA datasets To determine whether MFAP5+ SFs could be reproducible in other OA datasets, we collected two independent scRNA-seq datasets for further analysis. In GSE152805, through unbiased clustering and cell lineage annotation based on canonical gene expression, 10,006 single cells were categorized into 7 main lineages (Figures 2A and 2B). Cells from multiple samples were integrated into a homogeneous population (Figure S1D). The most abundant cell type was SFs (78% of the total), which were further subdivided into 5 clusters (Figure 2A). We found that C0 SFs in GSE152805 were similar to MFAP5+ SFs, which were marked by high expression of marker genes and signatures of MFAP5+ SFs (Figures 2C and 2D, p < 0.001). Other SF clusters were characterized by distinct SF signatures but not MFAP5+ SF signatures (Figures S1E–S1I, p < 0.001). In the SCP469 dataset, cell lineage labeling was performed as described in the original study (Figures 2E and S2A).12 SFs were subdivided into 5 clusters, among which C5 was annotated as MFAP5+ SFs, expressing high levels of marker genes and signatures of MFAP5+ SFs (Figures 2F and 2G, p < 0.001). SF clusters displayed heterogeneity in SCP469, for example, C1 exhibited high SLPI+ and NUPR1+ SF signatures, while C3 showed high IGFBP3+ SF signatures (Figures S2B–S2F, p < 0.001). Consistent with GSE152805 and SCP469 datasets, FBN1 and MFAP5 showed the highest expression in SFs, but their expression in others was low or undetectable (Figures 2C and 2F). MFAP5+ SFs consistently expressed high levels of THY1, a classical marker of sublining SFs. Additionally, GSN and IGFBP6, markers of adipogenic cancer-associated fibroblasts,13 were highly expressed in MFAP5+ SFs (Figures 2H–2J). MFAP5+ SFs were reproducible in other OA datasets (A) UMAP visualization of major cell lineages in GSE152805. (B) Expression levels of the marker genes across different cell lineages in GSE152805. Dot size indicated the fraction of expressing cells, and the colors represented normalized gene expression levels. (C) Violin plots showing the expression of marker genes in MFAP5+ SFs across clusters in GSE152805, with the three vertical lines indicating the median, upper, and lower quartiles. (D) Comparison of the MFAP5+ SF signature scores across five SF clusters in GSE152805. (E) UMAP view of major cell lineages in SCP469. (F) Violin plots presenting the expression of marker genes in MFAP5+ SFs across clusters in SCP469, with the three vertical lines representing the median, upper, and lower quartiles. (G) Comparison of the MFAP5+ SF signature scores across five SF clusters in SCP469. (H–J) Violin plots showing the expression levels of selected genes across SF clusters in GSE176308 (H), GSE152805 (I), and SCP469 (J). Statistical significance was calculated using two-tailed Wilcoxon rank-sum test (D and G). ∗∗∗p < 0.001. Upregulation of MFAP5 in end-stage synovium associates with aggressive synovitis In the sham rat group, synovium exhibited loose connective tissue. Mild angiogenesis and infiltration of inflammatory cells (including lymphocytes and plasma cells) were observed in early-stage MIA-induced OA-model rats. In end-stage OA-model rats, the synovial cell layer showed significant thickening, with pronounced angiogenesis and extensive inflammatory cell infiltration (Figure 3A). As shown in Figure 3B, OA-model rats exhibited a higher Krenn’s synovitis score (KSS) than sham rats, with end-stage OA rats showing the highest scores (p < 0.001). MFAP5+ SFs displayed elevated expression of inflammatory response pathway genes (Figures 3C, 3D, and S3A, p < 0.001), indicating a pro-inflammatory phenotype. Additionally, MFAP5 expression in SFs positively correlated with inflammatory response pathway signatures (Figures S3B–S3D, p < 0.001). Immunohistochemistry staining revealed the highest MFAP5 expression in end-stage OA synovial tissues, moderate expression in the early stage, and the lowest expression in sham tissues (Figures 3E–3G, p < 0.01). MFAP5 mRNA expression in synovial tissues progressively increased with OA severity, peaking in end-stage OA-model rats (Figure 3H, p < 0.01). Similarly, end-stage OA synovium tissues showed a high percentage and number of FBN1+ cells, while sham tissues showed low levels (Figures 3I–3K, p < 0.001). FBN1 mRNA expression in synovial tissues also rose with OA severity, reaching its peak in end-stage OA-model rats (Figure 3L, p < 0.05). MFAP5 upregulation in end-stage synovium of MIA-induced OA-model rats (A) Representative images of Hematoxylin and eosin (H&E) staining of synovial tissues across three groups of rats. Scale bars, 50 μm. (B) Distribution of KSS in synovial tissues among the three rat groups (n = 3–5). (C and D) Comparison of inflammatory response pathway scores between MFAP5+ SFs and other SFs in GSE176308 (C) and GSE152805 (D). (E) Representative images of MFAP5 immunohistochemistry staining of synovial tissues across the three rat groups. Scale bars, 50 μm. (F) Percentage of MFAP5+ cells in synovial tissues across the three rat groups (n = 5–6). (G) Number of MFAP5+ cells in synovial tissues across the three rat groups (n = 6). (H) The mRNA level of MFAP5 in synovial tissues across the three rat groups was measured using quantitative real-time PCR (real-time qPCR) assay (n = 7–10). (I) Representative images of FBN1 immunohistochemistry staining of synovial tissues across the three rat groups. Scale bars, 50 μm. (J) Percentage of FBN1+ cells in synovial tissues across the three rat groups (n = 6). (K) Number of FBN1+ cells in synovial tissues across the three rat groups (n = 6). (L) The mRNA level of FBN1 in synovial tissues across the three rat groups was measured using real-time qPCR assay (n = 8–15). Data were presented as mean ± standard error of mean (SEM). Statistical significance was calculated using two-tailed Wilcoxon rank-sum test (C and D) and one-way ANOVA followed by Tukey’s multiple comparisons test (B, F–H, and J–L). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. To further evaluate MFAP5 expression in human synovial tissues, we collected 14 samples from patients with knee OA or joint injuries. In the control group, synovium showed loose connective tissue. The early-stage OA group showed papillomatous hyperplasia, mild vascular hyperplasia, vasodilation, mucinous interstitial change, and slight inflammatory cell infiltration. In the end-stage OA group, significant synovial cell layer hyperplasia, vasodilation, vessel hyperemia, and extensive inflammatory infiltration, including lymphocyte aggregation, were observed (Figure 4A). Consistent with the OA-model rats, KSS was highest in end-stage OA and lowest in the control group (Figure 4B, p < 0.01). Double-immunofluorescence staining demonstrated progressively increased MFAP5 positivity in stromal cells (vimentin+), with the highest levels in end-stage OA, intermediate levels in early-stage OA, and the lowest levels in controls (Figures 4C and 4D, p < 0.05). Additionally, MFAP5 fluorescence intensity and mRNA expression were significantly higher in the end-stage OA group than in control or early-stage OA groups (Figures 4E and 4F, p < 0.05). Both MFAP5-positive intensity and mRNA levels were positively correlated with KSS (Figures 4G and 4H, p ≤ 0.05). FBN1 and MFAP5 were both markers of MFAP5+ SFs, with FBN1 exhibiting a similar expression pattern to that of MFAP5 (Figures 4I–4N, p < 0.05). However, FBN1 showed a lower colocalization rate with stromal cells in the end-stage OA group (Figures 4I and 4J, p < 0.05). This suggested that MFAP5 is a more suitable marker for the SFs that are preferentially infiltrated in end-stage OA. MFAP5 upregulation in synovium of end-stage OA patients correlated with aggressive synovitis (A) Representative images of H&E staining of synovial tissues across three groups of humans. Scale bars, 100 μm. (B) Distribution of KSS in synovial tissues across the three human groups (n = 3–5). (C) The colocalization pixel ratio of MFAP5 and vimentin was quantified in synovial tissues across three groups of humans (n = 4). (D) Representative immunofluorescence images of MFAP5 and vimentin colocalization in synovial tissues. Scale bars, 500 (top) and 20 μm (bottom). (E) Fluorescent intensity of MFAP5 in synovial tissues across the three human groups (n = 3–4). (F) The mRNA level of MFAP5 in synovial tissues across the three human groups was measured using real-time qPCR assay (n = 3–5). (G and H) Correlation between KSS and MFAP5 fluorescent intensity (G)/mRNA level (H) in synovial tissues. (I) Representative immunofluorescence images of FBN1 and vimentin colocalization in synovial tissues. Scale bars, 500 (top) and 20 μm (bottom). (J) The colocalization pixel ratio of FBN1 and vimentin was quantified in synovial tissues across three groups of humans (n = 4). (K) Fluorescent intensity of FBN1 in synovial tissues across the three human groups (n = 3–4). (L) The mRNA level of FBN1 in synovial tissues across the three human groups was measured using real-time qPCR assay (n = 3–5). (M and N) Correlation between KSS and FBN1 fluorescent intensity (K)/mRNA level (L) in synovial tissues. Data were presented as mean ± SEM. Statistical significance was calculated using one-way ANOVA followed by Tukey’s multiple comparisons test (B, C, E, F, and J–L) and Spearman's rank correlation analysis (G, H, M, and N). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. MFAP5+ SFs promote collagen deposition in synovium To further investigate the co-expression gene modules during OA development, we performed MEGENA on the GSE176308 dataset across two stages. This analysis produced a hierarchical network with 17 interconnected modules, with the c1_9 module being enriched in MFAP5+ SFs (Figures 5A and 5B). To identify the functional roles of genes within the c1_9 module, Gene Ontology pathway analysis revealed that genes were enriched in pathways related to collagen and ECM, including “ECM structural constituent,” “collagen binding,” and “collagen-containing ECM” (Figure 5C). Furthermore, in GSE152805, the C0 cluster, representing MFAP5+ SFs, showed high scores for the c1_9 module, consistent with observations from SCP469 (Figures 5D and S3E, p < 0.001). MFAP5+ SFs also exhibited significant upregulation of collagen-associated pathways, including collagen-containing ECM, “collagen formation,” and “crosslinking of collagen fibrils,” which aligned with findings in GSE152805 (Figures 5E, 5F, and S3F, p < 0.05). The expression levels of MFAP5, FBN1, and AHNAK2 were positively correlated with scores of collagen-associated pathways in SFs, whereas control groups showed weaker correlations (Figures 5G and S3G, p < 0.001). Over time, fibroblasts could deposit a composite fibronectin and collagen matrix, which eventually remodeled into a predominantly collagen-based ECM.14 MFAP5+ SFs scored higher for the “ECM binding” and “ECM organization” pathways compared to other SF clusters across all three datasets (Figures 5H and 5I, p < 0.001). Furthermore, MFAP5, FBN1, and AHNAK2 showed a positive correlation with scores of the two pathways (Figure S3H). Collectively, the results indicated that MFAP5+ SFs promoted ECM generation in OA synovial tissue. MFAP5+ SFs promoted collagen deposition in synovium (A) Dendrogram showing the hierarchy organization of MEGENA network modules in GSE176308. The top layer represented the root and largest parent modules, which branched hierarchically into smaller child modules. (B) Density plots displaying c1_9 module scores across SF clusters, with color-coded density values; red indicated higher density, and blue indicated lower density. (C) Network of pathways and genes involved in the c1_9 module, where a two-point line indicated that a gene was part of a specific pathway. (D) Raincloud plot showing c1_9 score in GSE152805. (E and F) Gene set enrichment analysis showed collagen pathways were upregulated in MFAP5+ SFs in both GSE176308 (E) and GSE152805 (F). (G) Correlation coefficients between the expression of MFAP5 and one ECM pathway score across three datasets. (H) Violin plots of ECM binding (left) and ECM organization (right) scores across SF clusters in GSE176308. Color-coded by SF clusters. (I) Density plots of ECM binding and ECM organization scores across SF clusters in SCP469 and GSE152805. (J) Comparison of significant ligand-receptor pairs from SF clusters to T cells in GSE152805. (K) Immunofluorescence images of MFAP5, CD4, and vimentin colocalization in synovial tissues. Scale bars, 200 or 20 μm. Statistical significance was calculated using two-tailed Wilcoxon rank-sum test (D and H) and Spearman's rank correlation analysis (G). ∗∗∗p < 0.001. Pathway analysis revealed that MFAP5+ SFs were enriched for the “CD4-positive, alpha-beta T cell proliferation” pathway (Figure 1D). CellChat analysis revealed that MFAP5+ SFs interacted with immune cells via collagen pathways, including COL1A2-CD44, COL6A1-CD44, COL6A2-CD44, and COL6A3-CD44 pairs (Figures 5J and S3I). Furthermore, multiplex immunofluorescence staining showed the close spatial proximity between MFAP5+ SFs and CD4 in OA synovial tissues, particularly in the sample with severe synovitis (Figure 5K). Together, these findings suggest a potential interaction between MFAP5+ SFs and immune cells in the synovium. However, this association is currently based primarily on pathway enrichment, predicted cell-cell communication, and spatial proximity and thus remains correlative. Whether MFAP5+ SFs can directly regulate immune cell behavior in the synovium requires further investigation. MFAP5+ SFs exhibit elevated TGF-β pathway activity in OA To explore shared and distinct functional features of MFAP5+ SFs across three datasets, Biological Process pathway analysis revealed their involvement in pathways such as “regulation of cellular response to TGF-β stimulus” and “regulation of cellular response to TGF-β stimulus” (Figures 6A and S4A). Compared to other SFs, MFAP5+ SFs exhibited elevated “canonical and noncanonical TGF-β signaling” pathway scores (Figure 6B, p < 0.001). Additionally, higher marker gene expression of MFAP5+ SFs was correlated with higher pathway scores both at bulk and single-cell levels (Figures 6C and S4B, p < 0.001). Among genes involved in the canonical and noncanonical TGF-β signaling pathway, TGFBR2, LOXL1, and LOX showed a strong positive correlation with MFAP5 and FBN1 (Figure 6D). TGFBR2 was highly abundant in multiple SF clusters, while LOXL1 was highly expressed in MFAP5+ SFs in certain datasets. However, unlike TGFBR2 and LOXL1, LOX was consistently expressed at high levels in MFAP5+ SFs across three datasets (Figures 6E and S4C), with elevated LOX levels observed in end-stage OA synovial tissues and a positive correlation between LOX and MFAP5/FBN1 (Figures 6F and 6G, p < 0.05). An SF inflammation model was constructed by treating the rat joint SF cell line (iCell-0135a) with lipopolysaccharide (LPS). LOX levels were significantly higher in the LPS-induced SFs compared to the control SFs (Figure 6H, p < 0.0001). Immunofluorescence staining showed that LOX knockdown in SFs suppressed LPS-induced LOX and IL-1β upregulation (Figures 6H, 6I, and S4D, p < 0. 0001). MFAP5 levels were significantly higher in the LPS-induced SFs compared to the control SFs (Figure 7C, p < 0.0001). To further investigate the regulatory effect of MFAP5 on LOX expression, we constructed an MFAP5 overexpression plasmid (OE-MFAP5) and used a short hairpin RNA to knock down MFAP5 (sh-MFAP5) (Figures S5A and S5B). Immunofluorescence staining showed that MFAP5 overexpression increased MFAP5 and LOX expression in SFs. In contrast, knockdown of MFAP5 in SFs suppressed LPS-induced MFAP5 and LOX upregulation (Figures 7A–7D, p < 0.0001). Collectively, the results suggest that MFAP5 may affect the expression of the inflammatory factor IL-1β by regulating the expression of LOX in SFs. MFAP5+ SFs exhibit elevated TGF-β pathway activity (A) Bar plot showing enriched biological process pathways in MFAP5+ SFs from GSE152805 or GSE176308, with the red dotted line indicating p = 0.05. (B) Comparison of the canonical and noncanonical TGF-β signaling scores across SF clusters in GSE176308, SCP469, and GSE152805. (C) Correlations between the scores of canonical and noncanonical TGF-β signaling and marker gene expression levels in MFAP5+ SFs from GSE107105. (D) Correlation coefficients between MFAP5/FBN1 expression and expression of genes involved in canonical and noncanonical TGF-β signaling in three datasets. (E) Expression levels of the selected genes across SF clusters in three datasets. Dot size indicated the fraction of expressing cells, and the colors represented normalized gene expression levels. (F) Boxplot showing LOX expression in early-stage and end-stage OA synovium tissues. (G) Correlations between MFAP5/FBN1 expression and LOX expression in GSE107105. (H and I) Immunofluorescence staining was used to detect the expression changes of LOX (n = 6) and IL-1β (n = 4) in iCell-0135a cells. Scale bars, 20 μm. Data were presented as mean ± SEM. Statistical significance was calculated using hypergeometric test (A), two-tailed Wilcoxon rank-sum test (B and F), Spearman's rank correlation analysis (C, D, and G), and one-way ANOVA followed by Tukey’s multiple comparisons test (H and I). ∗p < 0.05, ∗∗∗p < 0.001. ZBTB7A positively regulates the expression of LOX in SFs (A–D) Immunofluorescence staining was used to detect the expression changes of MFAP5 and LOX in iCell-0135a cells. Scale bars, 20 μm, n = 6. (E) Heatmap displaying upregulated regulons in MFAP5+ SFs from GSE176308. (F) Expression levels of selected TFs across SF clusters in three datasets, with dot size indicating the fraction of expressing cells and colors representing normalized gene expression levels. (G–J) Immunofluorescence staining was used to detect the expression changes of LOX and ZBTB7A in iCell-0135a cells. Scale bars, 20 μm, n = 6. Data were presented as mean ± SEM. Statistical significance was calculated using one-way ANOVA followed by two-tailed Student’s t test (A, G, and I) and Tukey’s multiple comparisons test (C, H, and J). ∗∗∗p < 0.001. ZBTB7A targeting LOX as a regulatory factor in MFAP5+ SFs To reveal potential upstream transcription factors (TFs), gene regulatory network inference identified 36 upregulated regulons in MFAP5+ SFs, 34 of which were significant (Figure 7E). ZBTB7A, KLF2, MAFF, YY1, and ZMIZ1 displayed both high regulon scores and elevated expression levels in MFAP5+ SFs (Figures 7E and 7F). ZBTB7A regulated epithelial-mesenchymal transition via TGF-β1 signaling.15 KLF2 governed endothelial cell quiescence through TGF-β signaling.16 The five TFs and 259 differentially expressed target genes constituted 389 regulatory relationship pairs (Figure S6A). LAMC1, one gene involved in the inflammation pathway, was a target of ZBTB7A (Figure S6A). YY1, as a TF and target gene of KLF2, ZBTB7A, and ZMIZ1, played an important role in the pathogenic mechanisms of rheumatoid arthritis through specifically promoting Th17 cell pathogenicity.17 Furthermore, among the five TFs, ZBTB7A was highly expressed in MFAP5+ SFs across three datasets (Figure 7F). LOX, regulated by ZBTB7A, was upregulated in degenerating cartilage driving enzymatic matrix cross-linking and stiffening.18 To investigate the regulatory effect of ZBTB7A on LOX expression, we performed ZBTB7A overexpression and knockdown in iCell-0135a SFs (Figures S5C and S5D). The results demonstrated that ZBTB7A overexpression upregulated LOX expression, whereas ZBTB7A knockdown attenuated LPS-induced LOX upregulation (Figures 7G and 7H, p < 0. 0001). Additionally, MFAP5 overexpression increased ZBTB7A expression, whereas MFAP5 knockdown attenuated LPS-induced ZBTB7A upregulation (Figures 7I and 7J, p < 0. 0001). LOX inhibitor improves MIA-induced joint damage LOX-IN-3 is a small-molecule inhibitor of LOX.19 To clarify the role of LOX downregulation in the OA pathogenesis, we administered LOX-IN-3 via intra-articular injection for subsequent studies. Hematoxylin and eosin (H&E) staining showed that the sham group exhibited normal joint architecture, characterized by a well-organized synovial lining, intact articular cartilage with evenly distributed chondrocytes, and negligible inflammatory cell infiltration. In contrast, the MIA+DMSO group displayed synovial hyperplasia, loss of chondrocyte columnar arrangement, cartilage matrix depletion, and extensive inflammatory infiltration, confirming successful OA induction. Treatment with LOX-IN-3 markedly attenuated these pathological changes, reducing inflammatory infiltration, synovial thickening, and cartilage degradation, with tissue morphology approaching that of the sham group. The MIA+diclofenac sodium group showed partial improvement, with milder synovitis and less severe cartilage disruption than the MIA+DMSO group (Figure 8A). Picrosirius red staining with polarized light microscopy was performed to evaluate collagen fiber morphology in joint tissues under different interventions. Compared with the sham group, the arrangement of collagen fibers in joint tissues of the MIA+DMSO group was significantly disordered, and the structure was severely damaged. LOX-IN-3 intervention restored collagen fiber alignment and tissue architecture in the MIA+LOX-IN-3 group, closely resembling the sham group and substantially reducing damage severity. The MIA+diclofenac sodium group exhibited only moderate improvement in collagen organization, with inferior tissue repair compared to LOX-IN-3 (Figure 8B). Collectively, these findings demonstrate that LOX-IN-3 exerts superior structure-modifying effects in experimental OA relative to diclofenac sodium, underscoring its promise as a disease-modifying therapeutic candidate for OA. Immunohistochemistry further showed that LOX-IN-3 reduced MIA-induced upregulation of LOX, IL-1β, MFAP5, and FBN1 in rat joint tissues, whereas diclofenac sodium only partially attenuated these changes (Figures 8C–8F, p < 0.05). These results indicate that LOX-IN-3 shows stronger inhibitory effects on MFAP5, FBN1, LOX, and IL-1β expression compared with diclofenac sodium in MIA-induced rats. LOX inhibitor improves MIA-induced joint damage (A) Representative images of H&E staining of rat joint tissues across four groups. Scale bars, 200 μm, n = 4. (B) Picrosirius red staining images of rat joint tissues across four groups. Scale bars, 100 μm. (C–F) Immunohistochemistry of LOX, IL-1β, MFAP5, and FBN1, rat joint tissues across four groups. IOD, integrated optical density. Scale bars, 400 μm, n = 4. Data were presented as mean ± SEM. Statistical significance was calculated using one-way ANOVA followed by Tukey’s multiple comparisons test (A and C–F). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Discussion In this study, we combined OA patient samples and MIA-induced OA-model rats to identify a pathological SF cluster characterized by MFAP5 overexpression. MFAP5+ SFs showed preferential infiltration in end-stage OA. Furthermore, our findings indicated that MFAP5+ SFs, a sublining SF cluster, positively correlate with the KSS. MFAP5+ SFs exhibit high LOX expression, and pharmacological blockade of LOX effectively mitigates joint damage in MIA-induced OA-model rats. We found a trend toward increased presence of MFAP5+ SFs in end-stage OA. Early-stage intervention provides the opportunity to reduce the burden of the disease by changing its course to prevent disability in OA.20 We identified six SF clusters, among which MFAP5+ SFs with MFAP5 and FBN1 high expression showed preferential presence in end-stage OA at single-cell and bulk levels. MFAP5 was involved in activating cancer-associated fibroblasts in ovarian cancer, esophageal squamous cell carcinoma, and breast cancer.21,22 FBN1, a member of fibrillin family, is known for its contributions to elastic fibers. Fibrillin gene mutations cause enhanced TGF-β signaling.23 Consistent with our bioinformatics results, we observed a distinct pattern in the expression levels of MFAP5 and FBN1, with a significant increase from control tissues to early-stage OA and remaining at the highest levels in end-stage OA both in OA-model rats and OA patients. While immunohistochemistry showed elevated MFAP5 and FBN1 expression in the synovium of end-stage OA, this increase may not be solely attributed to stromal fibroblasts, as these proteins could also be expressed by other cell types within the synovial tissue. Double-immunofluorescence staining of OA and control tissues demonstrated that stromal cells (vimentin+) expressed MFAP5 and FBN1 more intensely in the end-stage than in the early-stage and control tissues. However, the colocalization signal between MFAP5 and vimentin was stronger than that between FBN1 and vimentin. This observation led us to prioritize MFAP5 as a more reliable marker for subsequent validation experiments. To our knowledge, our present study reveals MFAP5 as an indicator of an SF cluster that is highly enriched in end-stage OA. Further studies are warranted to dissect the precise role of FBN1 in OA development and its potential synergistic interaction with MFAP5. MFAP5+ SFs were characterized by high activities of inflammation and collagen deposition with activation of the TGF-β pathway. Fibroblasts are highly plastic cells, among which myofibroblasts promote wound healing by secreting a plethora of pro-inflammatory cytokines and regulating collagen deposition.24 At single-cell resolution, we quantified the heterogeneity of the inflammation pathway across SFs and found that MFAP5+ SFs exhibited high inflammation scores, with MFAP5 expression positively correlating with these scores. This finding prompted further investigation of whether MFAP5 expression correlated with KSS in OA. Consistent with scRNA-seq data, MFAP5 expression in both OA-model rats and OA patients showed a positive correlation with KSS in synovial tissues. Synovitis is not restricted to end-stage OA, but patients with end-stage OA exhibited a high prevalence of synovitis.25 Furthermore, inflammatory infiltrates coexist with fibrotic changes in OA, especially in the end stage.26 As OA progresses, the synovium becomes hypertrophied as a result of the accumulation of excessive ECM.6 MFAP5+ SFs exhibited high expression of a co-expression gene module (c1_9) related to collagen and ECM organization, along with elevated scores of ECM pathways. SFs undergo fibrotic conversion as a result of deposition of ECM stimulated by TGF-β signaling.27 The TGF-β signaling pathway plays a vital role in maintaining joint stability in OA.28 This pathway was elevated in OA synovium29 and was higher in end-stage OA compared to early-stage OA, which suggests more severe fibrosis in end-stage synovial tissue. ZBTB7A, a regulator of TGF-β, targeted LOX in MFAP5+ SFs. ZBTB7A, part of the POK TF family, contains a POZ domain at its N terminus and a Krüppel-like zinc-finger domain at its C terminus.30 MFAP5+ SFs exhibited high activity of the ZBTB7A regulon and elevated ZBTB7A expression, suggesting that ZBTB7A may function as a key regulator of MFAP5+ SFs. Accumulated studies have demonstrated that ZBTB7A could regulate TGF-β signaling.15,30 TGFBR2, LOX, and LOXL1, three genes involved in the TGF-β pathway, positively correlated with expression levels of MFAP5 and FBN1. Furthermore, TGFBR2 and LOX, upregulated in MFAP5+ SFs, were target genes of ZBTB7A. This suggests that ZBTB7A could regulate the TGF-β pathway by targeting TGFBR2 and LOX. TGF-β not only plays a key role in the onset and persistence of synovial fibrosis but is also crucial for cartilage maintenance and repair. Seeking targets downstream of TGF-β is pivotal for preventing OA-related fibrosis without interfering with joint homeostasis.31 One research has demonstrated that LOX, COL1A1, COL5A1, and TIMP1 are most likely downstream of TGF-β.31 Given our observations of high LOX expression in synovium tissues of end-stage OA and its positive correlation with MFAP5 expression, it suggests a role in driving disease progression and serves as a potential target of the advanced stage. LOX is a collagen cross-linking enzyme whose overexpression caused synovial inflammation, whereas reduced LOX expression effectively suppressed medial meniscus-induced cartilage destruction in mice.18 MFAP5 or ZBTB7A overexpression increased LOX expression in iCell-0135a cells, while knockdown of either gene attenuated LPS-induced LOX upregulation. Reduced LOX expression suppressed LPS-induced IL-1β upregulation, suggesting that MFAP5 or ZBTB7A regulates inflammation via LOX in SFs. Given that LOX knockdown downregulated the expression of the inflammatory factor, we hypothesized that inhibiting LOX might attenuate the progression of OA. In vivo experiments demonstrated that LOX-IN-3, an LOX inhibitor, effectively alleviated joint damage in the MIA-induced OA-model rats. Limitations of the study There are several limitations in this study. First, the relatively small sample size limits our ability to comprehensively assess the influence of age and sex on MFAP5 expression and the progression of OA. Although no obvious correlation was observed between these factors and MFAP5 levels in the current cohort, future studies with larger, age- and sex-balanced patient populations are needed to validate these findings and clarify potential confounding effects. Second, Ensembl ortholog analysis revealed that the MFAP5 protein sequence identity between human and rat is approximately 56% (Figure S6B), indicating moderate conservation. This suggests that while MFAP5-related mechanisms may be broadly conserved, species-specific differences in protein structure or regulation may exist, which should be considered when extrapolating these findings to humans. Finally, FBN1 expression pattern in SFs during OA progression remains unclear. While FBN1 was highly expressed in synovium across scRNA-seq data, OA-model rats, and clinical samples, its colocalization with vimentin was limited in synovium tissues from end-stage OA. Future research incorporating spatial transcriptomics and large-scale single-cell data in OA may provide deeper insights into our questions. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Xiaoqiang E (exq_517@163.com). Materials availability This study did not generate new unique reagents or materials. Data and code availability All transcriptomic data analyzed during this study can be downloaded from public data resources. 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. Acknowledgments This work was supported by the National Key Research and Development Program of China (no. 2023YFF1204600), the Horizontal science and technology research funds (no. 2021HX026), the National Natural Science Foundation of China Joint Fund for Regional Integrated Research (no. U23AC6008), and the Natural Science Foundation of Heilongjiang Province of China (grant no. PL2025H064). Author contributions Study conception and design, L.W., B.L., L.Y., X.E., and Xuelian Li; acquisition of data, L.W., G.W., F.M., Xingyu Li, H.Z., S.Y., R.Y., J.D., H.P., and Z.M.; analysis and interpretation of data, L.W., G.W., X.X., Z.X., B.Z., and H.C. Declaration of interests The authors have declared that no conflict of interest exists. Declaration of generative AI and AI-assisted technologies in the writing process The authors used ChatGPT to improve the English syntax of a few sentences, in full compliance with the journal’s guidelines. All output was subsequently reviewed, edited, and approved by the authors, who assume full responsibility for the final content of the manuscript. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERDeposited dataGSE176308Nanus DE et al.32https://www.ncbi.nlm.nih.gov/geo/query/acc.cgiGSE152805Chou CH et al.33https://www.ncbi.nlm.nih.gov/geo/query/acc.cgiSCP469Kevin Wei et al.12https://portals.broadinstitute.org/single_cellGSE176223Nanus DE et al.32https://www.ncbi.nlm.nih.gov/geo/query/acc.cgiGSE107105Mizoguchi F et al.9https://www.ncbi.nlm.nih.gov/geo/query/acc.cgiAntibodiesMFAP5ProteintechCat# 15727-1-AP; RRID: AB_2266295FBN1ZenbioCat# 160327LOXABclonalCat# A7698; RRID: AB_2770199IL-1βABclonalCat# A1112; RRID: AB_2758416VimentinServicebioCat# GB11192; RRID: AB_2814685CD4ServicebioCat# GB150062-50Biological samplesSynovial tissues were isolated from OA or joint damage patientsDepartment of Orthopedics at the First Affiliated Hospital of Harbin Medical Universityapproval number: IRB5065923Experimental models: Organisms/Strains5 weeks old Sprague-Dawley male ratsExperimental Animal Center of the Second Affiliated Hospital of Harbin Medical Universityapproval number: IRB5065923Experimental models: Cell linesRat knee joint SF cell lineiCell Bioscience Inc.Cat# iCell-0135aChemicals, peptides, and recombinant proteinschloral hydrateMacklinCat# C804539MIASigma-AldrichCat# I4386LOX-IN-3MedChemExpressCat# HY-138625LPSSolarbioCat# SLB-IL2020Diclofenac sodiumDigenN/ADMSO vehicleBioFroxxCat# 1084ML100hydrogen peroxideElabscienceCat# E-IR-R217C3,3-diaminobenzidineElabscienceCat# E-IR-R217DDAPISolarbioCat# SLB-C0065Hematoxylin and Eosin staining KitSolarbioCat# G1120Picrosirius red staining KitServicebioCat# G1078TRIZOLInvitrogenCat# 15596026TransScript® Uni All-in-One First-Strand cDNA Synthesis SuperMixTransGen BiotechCat# 4368813QuantStudio 6 Flex systemThermo Fisher ScientificCat# ABISoftware and algorithmsR (v4.2.0)R CRANhttps://cran.r-project.orgPython (v3.7)Python Software Foundationhttps://www.python.orgGraphPad Prism (v9.0)GraphPadhttps://www.graphpad.comDoubletFinderDoubletFinder package in Rv2.0.3HarmonyHarmony package in Rv0.1.1SeuratSeurat package in Rv5.0.1GSVAGSVA package in Rv1.50.0CytoTRACECytoTRACE package in Rv0.3.3BayesPrismBayesPrism package in Rv2.2.2MEGENAMEGENA package in Rv1.3.7CellChatCellChat package in Rv1.1.3pySCENICpySCENIC in pythonv0.12.1clusterProfilerclusterProfiler package in Rv4.8.1Code reportingThis paper does not report original code. Experimental model and study participant details Generation of rat model of osteoarthritis Sprague-Dawley (SD) male (5 weeks old) rats were purchased from the Experimental Animal Center of the Second Affiliated Hospital of Harbin Medical University (Harbin, China). Rats were maintained in cages under standard conditions according to the regulations of the Institutional Animal Care and Use Committee of Harbin Medical University (approval number: IRB5065923). Rats were housed in a room with a relative humidity of 50–65% and a temperature of 25 °C under a 12-hour light/dark cycle, with unlimited access to food and water. Rats were acclimated for 7 days before the start of the experiments. SD rats were anesthetized with an intraperitoneal injection of 20% chloral hydrate (C804539, Macklin, China) at a dose of 3 mL/kg. Both knee joints were thoroughly disinfected with cotton soaked in 75% alcohol, with the operator wearing sterile gloves throughout the procedure. To induce OA lesions, 50 μL of MIA (60 mg/mL; I4386, Sigma-Aldrich, USA) was injected into the knee joints. Control mice received saline. Animals were euthanized, and knee joint specimens were collected at 1 week (early-stage OA) and 2 weeks (end-stage OA) post-injection for analysis. To determine the role of LOX-IN-3 in vivo, rats received intra-articular injections of MIA or DMSO. One week after MIA injection, the MIA-induced rats were randomly assigned to three groups. LOX-IN-3 (10 μM, 100 μL; HY-138625, MedChemExpress, China) was administered intra-articularly every 3 days for 2 weeks. Diclofenac sodium (10 μM; Digen, China) was used as a positive control, whereas DMSO vehicle (0.01%, 100 μL; 1084ML100, BioFroxx, China) served as the vehicle control, and both were administered intra-articularly. After four weeks, the rats were euthanized, and joint tissues were collected for subsequent experiments. Human synovium sample acquisition The protocols and procedures for handling human synovium samples in this study were reviewed and approved by the Ethics Committee of Harbin Medical University (approval number: IRB5065923) and conducted in accordance with the Declaration of Helsinki. Patients with OA met the 1986 American College of Rheumatology classification criteria for OA.34 Fourteen synovium samples were collected between September 2023 and July 2024 from the Department of Orthopedics at the First Affiliated Hospital of Harbin Medical University (Harbin, China), including samples from 9 knee OA patients, 3 patients with meniscal injury without OA, 1 patient with posterior cruciate ligament injury without OA, and 1 patient with anterior cruciate ligament rupture without OA. Arthritic changes in the knee joint were evaluated using Kellgren-Lawrence (K-L) grading.35 Classification of early-stage and end-stage OA was based on clinical history, KSS, and K-L grading. Early-stage OA was defined as K-L grades 1–2, KSS of 1–2, disease duration ≤ 2 years, and absence of significant joint space narrowing or osteophyte formation. End-stage OA was characterized by K-L grades 3–4, KSS of 3–4, disease duration > 5 years, and prominent radiographic features, including marked joint space narrowing, extensive osteophyte formation, and subchondral bone sclerosis. All histological and radiological evaluations were independently performed by two experienced orthopedic surgeons and radiologists. Inter-observer disagreement was resolved by consensus discussion to ensure diagnostic reliability. For detailed clinical information of patients, refer to Table S1. Written informed consent was obtained from all patients for the evaluation and publication of their anonymized data. Cell culture and treatment The rat knee joint SF cell line was procured from iCell Bioscience Inc. (iCell-0135a, China). The iCell-0135a cells were cultured according to the instructions. The cell line was authenticated using STR profiling within the last 3 years, and all the experiments were performed with mycoplasma-free cells. iCell-0135a cells were seeded in the culture flask and cultured until approximately 60% confluence was reached. Rat plasmids (MFAP5 and ZBTB7A), shRNAs constructs (sh-MFAP5, sh-ZBTB7A, sh-LOX, and sh-NC) were purchased from Gene Create. For transfection experiments, cells were treated with LPS (SLB-IL2020, Solarbio, China; 35 μg/mL) for 24 hours, followed by transfection with plasmids or shRNAs for 4-5 hours for subsequent experiments. Method details Immunohistochemistry staining Rat synovial tissue sections were embedded in paraffin after fixing with 4% paraformaldehyde. The paraffin sections were deparaffinized using xylene, rehydrated through a graded alcohol series, and subjected to heat-induced antigen retrieval in an appropriate antigen retrieval solution. After blocking endogenous peroxidase with 3% hydrogen peroxide (E-IR-R217C, Elabscience, China) at room temperature for 10 minutes, sections were incubated overnight at 4 °C with primary antibodies against MFAP5 (15727-1-AP, Proteintech, China), FBN1 (860327, Zenbio, China), LOX (A7698, ABclonal, China), and IL-1β (A27676, ABclonal, China). The following day, anti-Rabbit IgG (E-IR-R217B, Elabscience, China) was applied at room temperature for 60 minutes. Visualization was achieved using 3,3-diaminobenzidine (DAB; E-IR-R217D, Elabscience, China) staining, followed by hematoxylin counterstaining and sealing with neutral gum. The sections were dried at 37 °C overnight, and images were captured with a fluorescent microscope (Olympus). Scale bar: 50 μm. Immunofluorescent staining Synovial tissue sections from 12 individuals were fixed in 4% paraformaldehyde for 30 minutes at room temperature, followed by paraffin embedding. After permeabilization with 0.5% Triton-X100 for 60 minutes, the sections were blocked with bovine serum albumin for 30 minutes. Sections were then incubated overnight at 4 °C with primary antibodies, including vimentin (GB11192, Servicebio, China), FBN1 (860327, Zenbio, China), MFAP5 (15727-1-AP, Proteintech, China), and CD4 (GB150062-50, Servicebio, China). The following day, fluorescent secondary antibodies were added, and nuclei were stained with 4',6-diamidino-2-phenylindole (DAPI) for 15 minutes. Results were visualized under a confocal microscope (ZEISS). Colocalization pixel ratio was calculated to quantify the colocalization of MFAP5/FBN1 and vimentin, defined as the percentage of double-positive pixels relative to total positive pixels. Double-positive pixels referred to pixels with signal intensities exceeding background thresholds in both channels, whereas total positive pixels represented the union of threshold-positive pixels in either channel to avoid double-counting. iCell-0135a cells were plated in six-well plates. Following treatment, cells were rinsed three times with PBS, fixed in 4% paraformaldehyde for 30 minutes, permeabilized with 0.1% Triton X-100 in PBS at 4 °C for 40 minutes, and blocked with 4% BSA at 37 °C for 1 hour. Cells were then incubated overnight at 4 °C with primary antibodies against MFAP5 (15727-1-AP, Proteintech, China), LOX (A7698, A7698, China), and IL-1β (A27676, ABclonal, China). After three washes with ice-cold PBS, cells were incubated with secondary antibodies for 1 hour at 21–25 °C. Nuclei were counterstained with DAPI (SLB-C0065, Solarbio, China) for 25 minutes. Immunofluorescence signals were captured and quantified in four or six randomly selected fields under a fluorescence microscope. Morphological analysis The H&E staining Kit (G1120, Solarbio, China) and picrosirius red staining Kit (G1078, Servicebio, China) were used to assess morphological changes. Paraffin sections of synovial tissues were stained with hematoxylin for 5-10 minutes, rinsed with tap water, and then immersed in eosin dye for 1-3 minutes. The sections were then dehydrated through graded alcohol and cleaned with xylene. Images were acquired using a microscope (Olympus). The sections were scored based on KSS as follows: 0-1 (no synovitis), 2-4 (low-grade synovitis), and 5-9 (high-grade synovitis).36 Paraffin sections were incubated in Sirius red staining solution A at 65 °C for 30 minutes. After rinsing in water until the yellow color faded from the tissue, sections were briefly drained and stained in Sirius red solution B for 2 minutes. Following a brief rinse and draining, sections were stained in Sirius red solution C for 1 hour. Sections were then quickly rinsed in water for 1–2 seconds and rapidly dehydrated through three changes of absolute ethanol (3 seconds each). Finally, sections were cleared in clean xylene for 5 minutes and mounted with neutral resin. Images were acquired using a polarized light microscope. Quantitative real-time PCR analysis Total RNA was extracted from rat joint and human synovial tissues using TRIZOL (15596026, Invitrogen, USA). Reverse transcription was performed with TransScript® Uni All-in-One First-Strand cDNA Synthesis SuperMix (4368813, TransGen Biotech, China). Quantitative real-time PCR (qRT-PCR) was performed using a QuantStudio 6 Flex system (ABI, Thermo Fisher Scientific, USA) to measure mRNA levels, with β-actin as the internal control. Data were analyzed using the 2−ΔΔCt method. The Homo sapiens primers used in the study: MFAP5 (forward: 5’- GTT TCA GCG GAG GCC AAA TC-3’; reverse: 5’- GTT CCT ACT TTG GCT GGC GA-3’), FBN1 (forward: 5’- TTT AGC GTC CTA CAC GAG CC-3’; reverse: 5’- CTC TTC TCT TGG CCC GAC TG-3’), β-actin (forward: 5’- GTG GAT CAG CAA GCA GGA GT-3’; reverse: 5’- TGC GCA AGT TAG GTT TTG TCA-3’). The Rattus norvegicus primers used in the study: MFAP5 (forward: 5’-CTG CCA TCT GTC CCG TAG AC-3’; reverse: 5’-TAT CAG CGA GCC CGT ACA AC-3’), FBN1 (forward: 5’- GTG CAA CTG GAA ACG GGA AC-3’; reverse: 5’- GAG CCG TCC AAG TTC TGA CA-3’). Single-cell RNA-sequencing data collection and processing Two public scRNA-seq datasets of human synovial tissues from OA patients (GSE176308 and GSE152805, shown as Table S2) were individually processed to exclude low-quality cells following the thresholds provided in the corresponding published studies.32,33 For the two datasets, we used DoubletFinder (v2.0.3) with the default parameters to remove potential doublets,37 and then employed the LogNormalize method with a scale factor of 10,000 to normalize gene count expression. As for SCP469 dataset (shown in Table S2) with no available count data, the TPM matrix with filtering out low-quality cells was downloaded and used directly.12 The top 3,000 highly variable genes identified in each of the three datasets were scaled and used for PCA. Subsequently, we calculated the first 50 principal components and corrected sample-driven batch effects within each dataset using Harmony (v0.1.1) and IntegrateData function in Seurat (v5.0.1).38 We used the Local Inverse Simpson’s Index (LISI) to quantify integration and accuracy of this embedding.38 The LISI scores showed that Harmony achieved high batch-correction performance (Figure S1A), and therefore we used this method for data integration. A shared nearest-neighbor graph and clustering analysis were implemented by FindNeighbors and FindClusters functions in Seurat. The identified clusters were visualized using the Uniform Manifold Approximation and Projection (UMAP) method. Tissue distribution of clusters We calculated the ratio of observed to expected cell numbers (Ro/e) for each SF cluster to assess its tissue preference, where the expected cell numbers for each SF cluster were determined through chi-square test.39 Ro/e > 1 indicated that the cluster was enriched in a specific tissue. Differential expression and enrichment analyses Differentially expressed genes (DEGs) were identified for each SF cluster using the FindMarkers function in Seurat, employing the nonparametric Wilcoxon rank sum test with Benjamini-Hochberg correction (only.pos = T, logfc.threshold = 0.1, pct.1 > 0.5 and FDR < 0.05). In GSE176308 dataset, the signatures of each SF cluster were identified through the following steps: (1) we obtained the DEGs of a specified cluster compared to each of the others, and (2) we retained consistent DEGs as cluster-specific signatures, which were provided in Table S3 and Figure S1C. To identify overrepresented pathways in SFs from GSE176308 dataset, we performed Gene Ontology Biological Process enrichment analysis on the signatures of each SF cluster using the clusterProfiler (v4.8.1), and P < 0.01 was statistically significant. We retained and visualized the top three pathways uniquely enriched in each cluster, ranked by P value. Additionally, to identify common pathways of MFAP5+ SFs across three datasets, we conducted the same enrichment analysis on the DEGs of MFAP5+ SFs, focusing on pathways enriched in both GSE176308 and GSE152805/SCP469 datasets, or those related to T cells. We collected gene sets of collagen, ECM, and TGF-β pathways from MSigDB,40 scored single cells using AddModuleScore function in Seurat and gene set enrichment analysis. To estimate the pathway difference among bulk samples, we implemented GSVA (v1.50.0) analysis in “ssGSEA” mode. The comparison of pathway scores for different patient groups was based on two-tailed Wilcoxon rank-sum test. CytoTRACE (v0.3.3) was used to assess the state of cell differentiation.41 Correlation analysis Spearman's rank correlation analysis was used to assess correlation between marker genes of MFAP5+ SFs and scores of collagen and ECM pathways. Five control groups were generated through random sampling with seed numbers set from 1 to 5. For each control group, 30 signatures of SF clusters, excluding MFAP5+ SFs, were sampled. Correlation coefficient was calculated to assess the association between genes and pathways, and the average correlation coefficient was defined as correlation of the random control group. Deconvolution of bulk RNA sequencing We collected bulk transcriptomic data from GSE176223 dataset, comprising 11 early-stage and 12 end-stage OA samples. Cell type deconvolution was performed using the BayesPrism (v2.2.2) with GSE176308 as reference.42 The two-tailed Wilcoxon rank-sum test was used to compare the proportions of SF clusters between early-stage and end-stage OA samples. MEGENA co-expression network and cell–cell interaction analyses The GSE176308 data was used as input for gene co-expression network analysis using MEGENA (v1.3.7) with default parameters.43 Modules containing more than four genes were visualized based on their hierarchical relationships. We used CellChat (v1.1.3) to infer interactions between ‘sender’ and ‘receiver’ cell clusters.44 Ligand–receptor pairs with P < 0.05 were considered in the downstream analysis and visualization. Transcription factor regulon analysis Activated regulons in SF clusters were analyzed by SCENIC.45 The GSE176308 data was used as input for the python implementation of the SCENIC algorithm (pyscenic, v0.12.1). Briefly, the gene-gene co-expression relationships between TFs and their potential targets were inferred using the grn function. Next, the command aucell was used to calculate the regulon activity for each cell. Finally, we calculated the average activity of each regulator in each SF cluster, retaining regulators with the highest activity in MFAP5+ SFs for visualization. We retained differentially expressed TFs and targets both in GSE176308 and SCP469/GSE152805 datasets between cells labeled as MFAP5+ SFs and other SFs using two-tailed Wilcoxon rank-sum test (P < 0.05) for visualization. Quantification and statistical analysis Statistical analyses were performed using R software (v4.2.0), python software (v3.7) and GraphPad Prism software (v9.0). The experiment was conducted in a random manner. Randomization sequence was created using Stata 9.0 (StataCorp., College Station, TX) statistical software. Data were presented as the mean ± standard error of mean (SEM) of at least three independent experiments. Statistical significance was calculated using One-way ANOVA followed by Tukey’s multiple comparisons test, two-tailed Wilcoxon rank-sum test, Spearman's rank correlation analysis, two-tailed Student's t-test and hypergeometric test. P < 0.05 was statistically significant. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; ns: no significance. Footnotes Supplementary data related to this article can be found online at https://doi.org/10.1016/j.isci.2026.116286. Contributor Information Lei Yang, Email: yangray83@vip.qq.com. Xiaoqiang E, Email: exq_517@163.com. Xuelian Li, Email: lixuelian@hrbmu.edu.cn. Supplemental information Document S1. Figures S1–S6 and Tables S1–S3 References Associated Data Supplementary Materials Document S1. Figures S1–S6 and Tables S1–S3 Data Availability Statement All transcriptomic data analyzed during this study can be downloaded from public data resources. 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.