Immune–metabolic heterogeneity and prognostic associations in breast cancer: development of an eight-gene risk model and preliminary functional characterization of NAT2

Abstract

Abstract Breast cancer is highly heterogeneous, and immune–metabolic interactions within the tumor microenvironment may contribute to differences in clinical outcome. In this study, transcriptomic, clinical, and somatic mutation data from TCGA-BRCA were analyzed, and the prognostic model was evaluated in GSE199633, a retrospective external transcriptomic cohort. Immune- and metabolism-related genes were integrated to define molecular subtypes and construct a prognostic model. Tumor microenvironment features, tumor mutational burden (TMB), and NAT2-related findings were further examined. Two molecular subtypes derived from the integrated immune- and metabolism-related candidate pool were identified. The 15-gene subtype panel was predominantly immune-related, with a smaller metabolism-related component. An eight-gene signature stratified patients into high- and low-risk groups across the training, internal testing, and retrospective external transcriptomic cohorts. Low-risk tumors showed an immune-hot-like phenotype, with higher immune and stromal scores, higher estimated cytotoxic lymphocyte abundance, increased immune checkpoint expression, and higher immunophenoscores, whereas high-risk tumors showed proliferation-related programs and relatively lower estimated immune-cell abundance. Additional correlation, network, and enrichment analyses showed reproducible correlation patterns and associations with immune, proliferative, and metabolic features rather than evidence of a single mechanistic pathway. The association between TMB and overall survival varied across risk-score groups. NAT2 was consistently upregulated in breast cancer, and its silencing reduced colony formation, migration, and invasion in vitro. Overall, the signature showed retrospective cross-cohort prognostic reproducibility, and NAT2 emerged as a risk-associated candidate gene with preliminary functional support; however, the generalisability of the model to prospectively recruited populations and routine clinical settings remains uncertain.