Spatial architecture of the synovium identifies clinically relevant phenotypes of osteoarthritis
文献摘要
Objectives Osteoarthritis (OA) is a heterogeneous joint disease characterized by marked variability in etiology and clinical symptoms. We aimed to determine whether spatial organization of synovial tissue underlies clinically relevant heterogeneity in knee OA. Methods We developed a deep learning-based semantic segmentation framework to analyze hematoxylin and eosin-stained whole-slide images of human synovium. Spatial entropy and inter-tissue connectivity metrics were incorporated to evaluate local pathology. We further applied graph neural network-based community modeling to classify tissue structures based on their microenvironment. Results The model identified key OA-associated histological features, including synovial lining hyperplasia, immune cell infiltration, angiogenesis, and fibrotic remodeling. Quantitative analysis of tissue composition revealed that enrichment of dense fibrous tissue was associated with poorer postoperative outcomes. However, tissue proportion alone did not adequately capture baseline disease status. Inter-tissue connectivity-based analysis showed that enrichment of Lining-Micro vessel interactions was associated with lower preoperative KOOS Symptom scores, indicating more severe symptoms, and elevated synovitis scores. Using features derived from graph neural network-based modeling, Micro vessels segregated into two distinct communities, one enriched in adjacent lining structures and another associated with immune-rich regions, and their relative abundances varied among knee OA patients. Integration with matched synovial single-cell RNA sequencing data showed increased predicted interactions between lining fibroblasts and endothelial cells in the Lining-Micro vessel-rich phenotype. These interactions were enriched for angiogenesis-related VEGF and ANGPTL signaling, providing supportive evidence for a distinct fibroblast-endothelial signaling state. Conclusions These findings indicate that the spatial organization of synovial architecture, beyond simple tissue composition, captures clinically relevant heterogeneity in knee OA and links histological microenvironments to cell-cell communication networks.