Patients were allocated into SS and non-SS (NC) groups according to published criteria (Chen et al., 2010a). The loss of ROM greater than 25% was defined as SS (Sun et al., 2018). Patients with rotator cuff tears were allocated into NC group.

MRI evaluation was performed using a 1.5 T Magnetic resonance scanner (UNITED IMAGING, uMR560) with an interval of 3 mm for each sagittal slice.

Shoulder capsule samples were obtained following standardized procedure: (Cole et al., 2009): Patient was placed with the affected shoulder abducted at 60° (Kim et al., 2021). Standard arthroscopic portals were established, followed by glenohumeral joint exploration and appropriate debridement (Lahti et al., 2025). Tissue was harvested from the joint capsule using the basket forceps. The sample was immediately transferred into liquid nitrogen.

8-week-old male C57BL/6J mice were randomly allocated to NC group, Model group, and LY294002 group (4 mouses for each group). The SS model was constructed by joint immobilization for 3 weeks (Qi et al., 2025). Mice in Model group and LY294002 group received intra-articular injections of 20 μL PBS or LY294002 (1.5 mg/mL) at one and 2 weeks postoperatively. NC group received sham surgery. Mice were kept under a 12 h light/dark cycle and were free to food and water. All animals were euthanized by overdose CO₂ 3 weeks postoperatively. Shoulder joint samples were then collected. ROM assessment was performed using a published methodology (Luo et al., 2022).

Total RNA was extracted from 12 capsules (6 SS and 6 NSS). Agilent 4,200 Bioanalyzer was used to assess the quality and size distribution of cDNA libraries. Sequencing was performed and data were analyzed using Seqtk (v1.3) and aligned to the GRCh38/hg38 reference genome. Transcript quantification was carried out and normalized. Data was analyzed by EdgeR (v3.42.0) in R (v4.4.3). Differentially expressed genes (DEGs) were defined at |log₂FC| > 1 and p < 0.05.

Gene expression data related to adhesive capsulitis (AC) were obtained from the Gene Expression Omnibus (GSE140731), comprising 48 samples (26 AC cases and 22 controls). Data analysis was conducted using the same statistical criteria (|log₂FC| > 1 and p < 0.05) to identify significant DEGs.

Gene Ontology (GO) enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were performed (Mu et al., 2024). The lists of DEGs from SS vs. NSS, AC vs. control, and the common DEGs between them were submitted separately. For the common DEGs, GO enrichment was conducted separately for the upregulated and downregulated gene subsets.

The common DEGs was submitted to the STRING database. A minimum interaction score of 0.4 was set. The resulting network was imported uploaded to Cytoscape (v3.9.1). The Molecular Complex Detection (MCODE) plugin was used to identify highly interconnected clusters.

The top 5 candidate hub genes were determined by aggregating the results from all 12 calculation algorithms available in cytoHubba. The Receiver Operating Characteristic (ROC) curve was constructed and the Area Under the Curve (AUC) was calculated.

Murine fibroblast NIH3T3s were purchased from Servicebio. Cells were maintained in high-glucose DMEM medium with 10% fetal bovine serum +1% penicillin/streptomycin at 37 °C with 5% CO₂. For in vitro experiments, cells were treated with FGF9 or LY294002 of 50 ng/mL and 3 μg/mL for 24 h, respectively.

When cells reached 80% confluence, a linear scratch was created by a pipette tip. Then, different interventions were administered for 24 h. After that, cells were observed and imaged under a microscope. The migration ability was quantified according to the recovery rate.

Cells were labeled with EdU solution for 2 h. Next, cells were fixed and permeabilized. After that, click reaction regent was employed. Finally, images were acquired using fluorescence microscope.

Joint samples were embedded and sectioned at 8 μm. Then, sections were stained with hematoxylin and eosin for 5 min each. Masson staining was conducted using a commercial kit (G1006, Servicebio). Images were captured by microscope.

Sections were treated with antigen retrieval solution (G0142, Servicebio) and blocking buffer (P0102, Beyotime). NIH3T3s were fixed and blocked. Next, sections and cells were incubated with corresponding primary and secondary antibodies. DAPI was used to locate nuclei. Images were captured by fluorescence microscope.

GraphPad Prism was used for data analysis. Data was presented as mean ± SD. Student’s t-test and one-way ANOVA were employed for comparisons. P < 0.05 was considered as statistical significance.

A total of 12 patients (6 for SS and 6 for non-SS) were included in this study, and the shoulder joint capsules were collected. The baseline data and ROM measurement results were demonstrated in Table 1. There was no significant difference in gender, age, BMI, or symptom duration between SS and non-SS groups. However, a significantly decrease of ROM was observed in SS patients, indicating joint stiffness. Additionally, there were higher VAS scores and lower Constant scores in the SS group, reflecting severe pain and functional impairment. MRI results showed significant thickening and edema in the capsule of SS patients (Figures 1A,C). Pronounced capsule hyperplasia and swelling were observed in the SS group under arthroscopy (Figures 1B,D). These results suggested that there were significant joint stiffness, capsule fibrosis and function impairment in SS patients.

The transcriptomic analysis on the patients’ capsules identified 821 DEGs, with 375 upregulated and 446 downregulated. The result was demonstrated by volcano plot (Figure 2A). GO analysis showed that DEGs were enriched in “cell periphery”, “plasma membrane” and “extracellular space” (Figure 2B). KEGG pathway analysis indicated that DEGs were enriched in “Cell adhesion molecules” and “Cytokine-cytokine receptor interaction” (Figure 2C). In addition, the transcriptomic analysis between AC and negative controls identified 1,025 DEGs, with 692 upregulated and 333 downregulated (Figure 2D). The results of GO and KEGG analysis were shown in Figures 2E,F.

Next, we conducted cross-analysis based on the datasets between SS and AC, and identified 100 common DEGs with 80 upregulated and 20 downregulated (Figure 3A). GO enrichment showed that DEGs were highly enriched in “extracellular matrix” and “collagen metabolic process” (Figures 3B,C). KEGG analysis showed predominant enrichment in “Cytokine-cytokine receptor interaction” and “TGF-beta signaling pathway” (Figure 3D).

After that, these DEGs were imported into the String database to construct a PPI network (Figures 4A,B). The MCODE algorithm was employed to identify key cluster modules and revealed a densely interconnected module consisting of 11 nodes and 24 edges, with a high cluster score of 4.8. These results indicated the robust and biologically significant interactions among these proteins (Figure 4C). To further investigate the interaction relationships among hub genes, the cytoHubba algorithm was applied and identified the top five genes including CXCL9, LOX, FGF9, POSTN, MMP9. Additionally, the ROC curves were generated and the AUC with a 95% CI was calculated (Figure 4D). Among these five genes, FGF9 was selected as the candidate pathogenic molecule.

To validate the multi-omics findings, we performed histological analysis on capsules from SS patients and established a mouse SS model. HE and Masson staining showed that the cell infiltration and collagen expression in SS group was significantly greater than NC group (Figures 5A,C). Similarly, the capsules of mouse SS models also exhibited significantly enhanced cell infiltration and capsule thickening (Figures 5B,D,E). The ROM of SS mice was markedly decreased compared to NC (Figure 5F). These results indicated that both clinical and mouse SS capsules develop characteristic fibrosis.

Subsequently, we assessed the FGF9 expression in the SS capsules. Immunofluorescence staining showed that the FGF9 were significantly upregulated in the capsules of both SS patients and SS mice (Figures 6A–D). These results aligned with our multi-omics analysis, suggesting that FGF9 played a critical role in SS pathogenesis.

In order to unveil the effect of FGF9 on fibroblasts, we employed LY294002, a specific inhibitor of the PI3K/Akt pathway. Wound healing assay showed that FGF9 significantly enhanced the migration ability of NIH3T3s (Figures 7A,B). Additionally, immunofluorescence staining and EdU assay demonstrated that the α-SMA expression and proliferation activity of NIH3T3s were markedly upregulated (Figures 7C–E). However, this activation was significantly reversed by LY294002.

We further evaluated the therapeutic effect of LY294002 in vivo. HE and Masson staining demonstrated that LY294002 significantly reduced capsule thickness when compared to Model group (Figures 7F,G). There was also a significant restore in ROM after LY294002 treatment (Figure 7H). These results indicated that FGF9 induced fibroblasts activation and capsule fibrosis through PI3K/Akt signaling pathway. Targeted inhibition of PI3K/Akt signaling could be a promising therapeutic strategy for SS.