Candidate MicroRNA Regulatory Axes in Melanoma CD8+ T Cell Exhaustion: A Network-Based In Silico Stratification

Background Anti-programmed cell death protein-1 (anti-PD-1) immunotherapy has transformed the treatment of advanced melanoma, but durable benefit remains limited to a subset of patients. CD8+ T cell exhaustion contributes to immune escape in the melanoma tumor microenvironment, while the post-transcriptional regulation of exhaustion-associated genes by microRNAs (miRNAs) remains incompletely understood. Objective We characterized miRNA-mRNA regulatory associations in CD8+ T cell exhaustion-enriched melanoma transcriptomes, classified inverse associations as loss-of-repression (LoR) or active suppression (AS), extended the network to candidate long non-coding RNA (lncRNA)-miRNA-mRNA relationships, and examined whether network-derived transcriptomic scores were associated with anti-PD-1 outcomes in independent cohorts. Methods The TCGA-SKCM bulk transcriptomes were filtered by single-sample gene set enrichment analysis (ssGSEA), yielding 121 CD8+ T cell exhaustion-enriched cases and 115 cases with paired miRNA and mRNA measurements. Differentially expressed miRNAs (DEmiRNAs) were identified between fixed high- and low-exhaustion tertiles (n = 38 each) using two-sided Mann-Whitney U tests with Benjamini-Hochberg correction (|log₂FC| ≥ 0.5; false discovery rate (FDR) ≤ 0.05). Database-supported inverse Spearman's correlations (ρ ≤ -0.30; FDR ≤ 0.05) were assembled into a bipartite network. Edges were classified as LoR or AS, and a competing endogenous RNA (ceRNA) extension incorporated DIANA-LncBase/ENCORI lncRNA-miRNA interactions. Nine biologically anchored axes underwent continuous-score analysis and HC3 regression adjusted for tumor purity, CD8, fibroblast/CAF, myeloid, interferon-gamma, and sample type, together with four sensitivity analyses. Exploratory clinical testing used GSE78220 and the pre-PD-1 biopsy subset of the DFCI melanoma cohort (cBioPortal study identifier: mel_dfci_2019). Results Twenty-six DEmiRNAs (20 upregulated and 6 downregulated) formed 326 inverse miRNA-mRNA edges, comprising 263 AS and 63 LoR associations. Six of nine focused axes met the adjusted-support criterion, and all six were AS-classified including upregulated miR-155-5p as the dominant hub, with inverse associations involving FOXO3 (ρ = -0.305, adjusted p = 0.020) and MEIS1, the strongest priority edge (ρ = -0.470, adjusted p < 0.0001). NEAT1, MALAT1, and XIST emerged as the top-degree lncRNA hubs, all classified as AS-sponge type. Exclusion of one solid-tissue-normal specimen left 114 paired tumors, retained 25 DEmiRNAs, and supported seven of nine axes in the tumor-only sensitivity analysis. Relational integration with lncRNA-miRNA records yielded 29,274 candidate chains, including 790 containing a focused miRNA-mRNA edge. Transcriptomic score analyses were negative and exploratory in GSE78220 (mRNA proxy area under the curve (AUC) = 0.631; Mann-Whitney p = 0.269) and in the DFCI melanoma cohort (mRNA topology AUC = 0.453, p = 0.451; lncRNA topology AUC = 0.388, p = 0.0695). Conclusions The LoR/AS framework offers a transparent means of organizing correlative miRNA-mRNA hypotheses in melanoma bulk transcriptomes. The adjusted results favored an AS pattern among the focused axes, but they did not establish CD8+ T-cell-intrinsic regulation, direct miRNA targeting, ceRNA activity, or clinical predictive utility. Experimental testing in sorted or single-cell melanoma CD8+ tumor-infiltrating lymphocyte systems is required.