Recurrent microRNA–gene associations with tumour cytolytic activity and their sensitivity to immune composition

Background Cytolytic activity (CYT), derived from granzyme A and perforin 1 expression, reflects cytotoxic immune activity but is sensitive to tissue composition. Objective To identify recurrent microRNA (miRNA)–gene associations linked to CYT within individual cancers and to assess their cellular-context dependence and reproducibility. Methods Matched expression and clinical data from 9,286 primary tumours across 31 cancer types in The Cancer Genome Atlas were analysed; purity-adjusted discovery used consensus purity estimates (available for 7,523 tumours) in 20 eligible cancers. Pairs required the miRNA–CYT, miRNA–gene and gene–CYT associations to meet |ρ| ≥ 0.30 and a false discovery rate < 0.05 in the same cancer, in at least three cancers. Fixed pairs were evaluated for sensitivity to cellular composition and against correlation-matched non-target controls and then assessed in an independent breast-cancer cohort; survival, conditional and cluster analyses were exploratory. Results Forty-seven pairs involving six miRNAs and 40 genes met the criteria. Associations were heterogeneous across cancers; none of 44 evaluable pairs retained recurrence after T cell adjustment, and none outperformed correlation-matched non-target controls after correction. Cellular-context checks retained 31 pair–cancer–population combinations representing 19 pairs. All 47 pairs met the three-arm conjunction and effect-size criteria in 101 external breast tumours and, after exome-derived purity adjustment, in 96. Directions agreed with the unadjusted and purity-adjusted Cancer Genome Atlas breast-cancer references for 47 and 46 pairs, respectively. Conclusions Recurrent immune-associated expression patterns were reproducible in one external breast-cancer cohort. Composition sensitivity and matched-control results limit claims of direct targeting; causal mechanisms and clinical utility remain unestablished.