Association Between Mutation Context-Associated RUNX1 Transcriptional States and Immune-Stromal Heterogeneity in...

Nonsmall cell lung cancer (NSCLC) is characterized by substantial heterogeneity in driver mutation backgrounds and tumor microenvironment (TME) phenotypes, yet the transcriptional regulatory states linking genotype context to immune-stromal variation remain incompletely defined. In this study, we integrated mutation, transcriptomic, and clinical data from TCGA-LUAD/LUSC to construct a mutation-expression-matched NSCLC cohort and evaluated RUNX1 as a mutation context-associated transcriptional/regulatory state rather than as a recurrently mutated driver. Driver contexts were assessed using a prespecified six-gene panel comprising EGFR, KRAS, TP53, STK11, KEAP1, and BRAF in the overall NSCLC cohort, the primary LUAD analysis, and a descriptive LUSC sensitivity analysis. The prespecified driver-context-negative operational group was defined as tumors without mutations in any of these six genes. RUNX1 mutation status was reported separately and did not contribute to this definition. Analyses included RUNX1 distributions, six-gene overlap, the RUNX1 cutoff, comutation burden, model sample-size audits, and adjusted context models. RUNX1 expression and inferred RUNX1 target-gene-set activity were evaluated across mutation contexts, whereas TME features were characterized using MCP-counter, xCell, ssGSEA, and ESTIMATE. Single-cell RNA sequencing and Visium spatial transcriptomic datasets were further used to localize RUNX1 expression, inferred target projections, and candidate TME-supporting genes. RUNX1 was prioritized within a prespecified candidate-screening framework and was significantly associated with the inferred RUNX1 target-gene-set readout. RUNX1 expression differences required interpretation in relation to histology and comutation structure. In the primary LUAD analysis, the KRAS-mutant context was associated with higher RUNX1 expression, whereas the KEAP1-mutant context was associated with lower expression. LUSC findings were retained as a descriptive sensitivity analysis and were not given the same inferential weight. Six-gene overlap, comutation burden, the RUNX1 cutoff, and model sample-size audits further showed that driver labels were nonmutually exclusive and that the prespecified driver-context-negative group was an operational reference rather than a biologically homogeneous subtype. RUNX1-high and RUNX1-low tumors displayed mutation context-dependent immune and stromal differences, involving stromal/cancer-associated fibroblast, myeloid/dendritic cell, endothelial, immune-score, and tumor-purity axes. Single-cell analysis localized RUNX1-associated signals and target-gene projections across epithelial-like and myeloid compartments, and spatial transcriptomics supported concordance between RUNX1 and selected TME-supporting genes in NSCLC tissue sections. Public immune checkpoint inhibitor cohorts were limited by small sample size and incomplete annotations and did not support RUNX1 as a stable immunotherapy-response predictor. Overall, these findings propose a mutation-context-aware RUNX1-TME stratification framework and generate testable hypotheses for mechanistic validation and mutation-annotated immunotherapy studies.