Multi-cohort and single-cell multi-omic analyses reveal multicellular mechanotransduction–immune remodeling in rheumatoid arthritis synovium

Background Rheumatoid arthritis (RA) synovitis integrates immune-cell recruitment, stromal remodeling and persistent inflammation, but its relationship with tissue mechanotransduction remains unclear. We investigated whether RA involves coordinated remodeling of multicellular mechanotransduction–immune programs. Methods We constructed a 327-gene mechanotransduction compendium and used GSE89408 synovial transcriptomes from healthy controls (n = 28), patients with osteoarthritis (n = 22) and patients with RA (n = 152) for discovery. Candidates prioritized through RA-versus-Healthy and RA-versus-OA contrasts were evaluated in three patient-independent cohorts using validation-only random-effects meta-analysis. Co-expression networks, consensus clustering, locked external mapping, single-cell RNA sequencing, cell–cell communication inference, virtual knockout and single-cell chromatin-accessibility analyses (11 RA donors and one OA reference; no healthy controls) resolved network, cellular and regulatory contexts. Selected gene-expression findings were examined by quantitative PCR in independent healthy (n = 20) and RA (n = 25) synovial tissues and by overexpression and siRNA-mediated knockdown in synovial fibroblasts. Results In unadjusted bulk-tissue profiles, RA synovium showed higher expression-based scores for integrin–extracellular matrix sensing and nuclear LINC programs but lower scores for focal-adhesion force-transmission and Hippo–YAP/TAZ-associated genes. Composition adjustment attenuated several of these associations. Nine candidates met the prespecified meta-analysis criteria and mapped to four co-expression modules associated with adaptive and innate immune programs, cell cycling and mechanotransduction-related expression scores. Tissue qPCR reproduced the predicted expression directions. Two molecular states with distinct inflammatory and mechanotransduction-associated expression features were recovered in external cohorts. In exploratory analyses, TLN2 and NRXN1 expression was inversely associated with inflammatory mediators and DAS28. Exploratory single-cell analyses localized the programs to fibroblast, myeloid, lymphoid and proliferating states and inferred differences in fibroblast–myeloid communication potential. Perturbing TLN2 or NRXN1 altered selected ECM-associated transcripts and COL1A1 protein, with corresponding changes in Matrigel invasion within one donor, while transcription-factor perturbation supported candidate FOXP2–NRXN1, NFIX–NRXN1/TLN2 and FOS–TLN2 expression relationships. Conclusions Patient-derived multi-omic and experimental evidence supports coordinated transcriptional remodeling of a multicellular mechanotransduction–immune network associated with RA synovial heterogeneity.