Integrated single-cell and bulk RNA analysis identifies glycolysis-associated epithelial cell signatures and prognostic model in non-small cell lung cancer

Metabolic reprogramming, particularly enhanced glycolysis, is a major feature of cancer and has been associated with tumor progression and therapeutic resistance in non-small cell lung cancer (NSCLC). However, the heterogeneity of glycolytic activity at the cellular level and its clinical significance in NSCLC remain incompletely understood. Single-cell and bulk RNA sequencing datasets from multiple NSCLC cohorts were integrated to investigate glycolysis-related epithelial cell heterogeneity. Glycolytic activity was evaluated using single-sample gene set enrichment analysis (ssGSEA). Non-negative matrix factorization (NMF) clustering was used to identify glycolysis-associated epithelial cell subpopulations. A prognostic risk model was established using LASSO-Cox regression and validated in independent cohorts. In addition, tumor immune microenvironment characteristics and predicted drug sensitivity were analyzed between different risk groups. Glycolysis scores were significantly elevated in NSCLC tissues and were associated with poorer patient survival. Among different cell populations, epithelial cells exhibited the highest glycolytic activity. Seven glycolysis-related epithelial cell states were identified, among which DSC2 + Epi-C2 showed the strongest glycolytic activity and its marker signature was associated with unfavorable prognosis. Based on glycolysis-associated epithelial cell signature genes, a 19-gene prognostic model was constructed and evaluated in the training, internal validation, and external cohorts. Histology-specific analyses showed shorter survival in the high-risk groups of both LUAD and LUSC, although model discrimination was lower in LUSC. High-risk patients exhibited distinct immune microenvironment features and differences in transcriptome-predicted drug sensitivity. This study characterized glycolysis-associated epithelial cell heterogeneity in NSCLC and identified a DSC2 + epithelial cell state associated with poor prognosis. The proposed prognostic model may provide additional value for risk stratification, but prospective validation is required, and the computational drug analyses do not establish therapeutic efficacy.