A Transcriptional Signature of Metabolic-Immune Conflict Fails to Provide Independent Prognostic or Predictive Value in Melanoma

Abstract

BackgroundAn inverse relationship between glutamine-associated and interferon-{gamma} (IFN-{gamma})-associated transcription in bulk melanoma could indicate metabolic antagonism or simply different cellular origins. We tested these alternatives and evaluated prognostic and immunotherapy-response associations. MethodsContinuous GSVA scores were analyzed in TCGA-SKCM (469 tumors), an independent melanoma survival cohort (GSE65904; 214 tumors), a pretreatment nivolumab cohort (GSE91061; 49 tumors), and single-cell RNA sequencing from 19 patients (GSE72056; 4,645 cells). Cox models used harmonized endpoints and clinical covariates. Bulk composition was examined with ABSOLUTE purity, EPIC, and MCP-counter. Single-cell scores were summarized by patient and compartment; paired tests treated patients, not cells, as independent units. Response models used Firth logistic regression, leave-one-out discrimination, and simulation-based precision analysis. ResultsIn TCGA, the candidate scores were inversely correlated before adjustment (r=-0.397; 95% CI -0.470 to -0.318) but progressively attenuated after purity/sample-type adjustment (r=-0.218), EPIC adjustment (r=-0.146), and MCP-counter adjustment (r=0.019). Single-cell analysis localized higher glutamine-associated scores to malignant cells than paired T, NK, or B-cell compartments, while IFN-{gamma}-associated scores were lower in malignant cells than paired T-cell and macrophage compartments. IFN-{gamma} was associated with lower mortality in TCGA (adjusted hazard ratio [HR] per SD 0.648; 95% CI 0.555-0.757) and GSE65904 (HR 0.667; 95% CI 0.534-0.835). Glutamine-associated scores were not independently prognostic in TCGA (HR 0.966; p=0.673) and were uncertain in GSE65904 (HR 0.823; p=0.084). The interaction was null in TCGA, nominal for the candidate sets in GSE65904, and null with curated sets. No response association was supported in GSE91061; approximately an odds ratio of 3.25 per SD was required for 80% simulated power. ConclusionsCellular composition accounts for much of the bulk glutamine-IFN-{gamma} transcriptional axis in melanoma. IFN-{gamma}-associated transcription is a reproducible prognostic correlate, but glutamine-associated transcription provides no stable incremental outcome information. These data do not establish metabolic competition or treatment prediction; spatial or functional measurements and larger treatment-comparative cohorts are needed.