Transcriptomic immune subtyping and connectivity mapping nominate DNA methyltransferase inhibitors for immunedepleted melanoma.

Abstract

Immune-cold melanomas respond poorly to immune checkpoint blockade (ICB). We assessed whether RNA-seq-based immune subtyping and connectivity mapping could nominate repurposable compounds to shift immune-depleted melanoma toward an immune-enriched phenotype. Using RNA-seq from 432 The Cancer Genome Atlas Skin Cutaneous Melanoma cases, we assigned published molecular functional portrait subtypes and compared depleted (D, n = 169) with combined immune-enriched (IE + IE/F, n = 184) subtypes using limma-voom. The resulting 300-up/300-down signature underwent Library of Integrated Network-based Cellular Signatures L1000 connectivity mapping, with compound-level aggregation and mechanism-of-action annotation. The pipeline was rerun on two stratified split-half folds. Canonical cytotoxic and antigen-presentation genes (CD8A, PDCD1, CXCL9, CXCL10, CD3D, GZMB, PRF1) were depleted in D, validating the contrast. Raw connectivity-mapping hits were dominated by generic cytotoxic-stress compounds. After aggregation and filtering, DNA methyltransferase (DNMT) inhibitors showed the most consistent positive pattern; fdcyd, azacitidine, and RG-108, but not decitabine, were robust to sample-type sensitivity analysis. Class enrichment was not significant after correction [false discovery rate (FDR) q = 0.126]. Indoleamine 2,3-dioxygenase inhibitors trended negative (FDR q = 0.095), and mitogen-activated protein kinase kinase inhibitors showed no class-level pattern (FDR q = 0.698). Across both folds, all seven canonical genes replicated and all four DNMT inhibitors retained above-median ranks, but the cold-elevated signature module was markedly less stable than the hot-elevated module (25.7 vs. 88.7% overlap). This suggestive pattern converges with a phase II trial of azacitidine/carboplatin priming before anti-programmed death-ligand 1 rechallenge in ICB-resistant melanoma, supporting prospective evaluation.