Oxeiptosis-related transcriptional heterogeneity in lung adenocarcinoma: prognostic modeling and functional prioritization of PSMD2

Background It remains uncertain whether transcriptional variation related to oxeiptosis carries prognostic information in lung adenocarcinoma (LUAD). We investigated this question and sought a model-derived gene whose contribution could be examined experimentally. Methods Oxeiptosis-associated candidates were identified from TCGA-LUAD and used for tumor classification and survival modeling. Competing models were evaluated across six GEO cohorts, and gene contributions were examined by SHAP. We also analyzed immune-related expression and estimated drug IC 50 values with oncoPredict. The distribution of the model signature across cell populations was explored in GSE171145 using AUCell. PSMD2 was then silenced in A549 and PC9 cells, followed by qRT-PCR, wound-healing, and Transwell assays. Results The 45-gene panel separated TCGA-LUAD tumors into four expression groups, of which C3 contained greater immune and stromal components. Bidirectional stepwise Cox regression followed by ridge regression produced the best overall performance across the six evaluation cohorts (mean C-index, 0.669). The resulting model comprised PSMD2, LSG1, PTBP1, ELAVL1, ANP32B, IMMT, and HSPD1. After adjustment for age, sex, and pathological stage, each standard deviation increase in the risk score corresponded to a higher mortality risk (HR, 1.62; 95% CI, 1.41–1.85). Survival separation reached significance in TCGA and four GEO cohorts, whereas the remaining two datasets showed the same direction without statistical significance. The score tracked positively with cell-cycle and DNA-replication activity, while the high-risk group expressed lower levels of the examined HLA genes. Drug-response estimation associated lower risk with greater predicted sensitivity to ribociclib and higher risk with greater predicted sensitivity to BI-2536. At single-cell resolution, high AUCell scores occurred mainly in epithelial and myeloid populations. PSMD2 made the largest SHAP contribution, and its silencing impaired migration and invasion in both cell lines. Conclusion The seven-gene model links oxeiptosis-related expression to LUAD prognosis, immune features, and predicted drug sensitivity. The prognostic contribution of PSMD2 and the effects of its knockdown support further investigation of its function.