Extracellular matrix remodeling and regulated cell-death-immune associations in adhesive capsulitis: a three-cohort transcriptomic analysis

Adhesive capsulitis (AC) is a fibroinflammatory disorder characterized by painful shoulder restriction and progressive capsular fibrosis. Whether regulated cell-death (RCD) programs are consistently altered across independent transcriptomic cohorts remains uncertain.

We reanalyzed three public bulk RNA-seq cohorts. GSE140731 was the discovery cohort (22 AC and 26 shoulder-instability surgical controls), and GSE190023 and GSE238053 were independent validation cohorts. Differential expression was estimated from filtered raw counts with PyDESeq2, with age/sex adjustment in GSE140731. Curated pyroptosis, ferroptosis, and necroptosis gene sets; immune marker-gene scores; pathway enrichment; STRING network structure; cross-cohort concordance; an exploratory three-gene classifier; and qPCR validation of four selected genes in TGF-β1-stimulated SW982 cells were evaluated.

The adjusted discovery analysis identified 439 upregulated and 113 downregulated genes at FDR < 0.05. ECM-receptor interaction was strongly positively enriched (NES = 2.299, q = 0.001), making extracellular-matrix remodeling the clearest pathway-level signal. RCD findings were mixed: pyroptosis was not different (delta = -0.105, q = 0.708), ferroptosis was lower but borderline after FDR correction (delta = -0.426, q = 0.056), and necroptosis was lower (delta = -0.320, q = 0.035); the core necroptosis gene set was also negatively enriched (NES = -1.673, q = 0.038). Six immune marker scores were FDR-significant in discovery, but none replicated at cohort-wise FDR significance in the validation cohorts. Cross-cohort gene-effect concordance was heterogeneous (Spearman rho = -0.047 for GSE190023 and 0.383 for GSE238053). The COL11A1-SPON2-PITPNC1 classifier achieved a pooled out-of-fold AUC of 0.820 (95% CI, 0.684-0.930) in GSE140731; external-cohort AUCs ranged from 0.750 to 0.812 with wide confidence intervals, and non-nested feature selection precluded claims of clinical utility. qPCR showed higher COL11A1 and SPON2 and lower PLA2G4A expression after TGF-β1 stimulation, whereas PITPNC1 did not differ significantly, providing model-specific support without establishing causal or clinical validity.

Across the three cohorts, ECM remodeling was the most reproducible feature of AC, whereas RCD- and immune-related signals were cohort-dependent and did not support uniform activation of pyroptosis, ferroptosis, or necroptosis. The classifier and RCD signatures remain hypothesis-generating. The SW982 qPCR results provide model-specific experimental support for selected candidates, but broader functional validation in primary AC tissues and cells is required.