Evaluation of the impact of glycolysis-related gene signatures on prognosis and therapeutic targeting in lung adenocarcinoma.

Abstract

Abnormal glycolysis is one of the hallmarks of cancer and plays a significant role in its progression. This study investigates the association between glycolysis genes and the progression of lung adenocarcinoma (LUAD). Utilizing various bioinformatics techniques, the research explores the heterogeneity of glycolysis genes in different LUAD cell types, identifies glycolysis-related prognostic signatures (GRPS). We obtained one training set for model construction from the Cancer Genome Atlas (TCGA) database, and also obtained four LUAD gene expression datasets as validation sets from the Gene Expression Omnibus (GEO) database. The single-cell RNA sequencing (scRNA seq) data also comes from the GEO database. Firstly, the "limma" R package was used to identify differentially expressed glycolysis related genes, and a machine learning computational framework composed of multiple combinations was used to preliminarily screen for glycolysis related prognostic markers (GRPS) in LUAD. Based on these GRPS, prognostic features were developed and validated through survival analysis, column chart development, and ROC curve analysis. The ssGSEA algorithm, ESTIMATE algorithm, and seven integrated computational algorithms from the TIMER 2.0 database were used to analyze the immune cell infiltration patterns of different risk groups. Analyze scRNA seq data to evaluate the distribution of GRPS and intercellular communication among various cell types, and further determine the core GRPS through the "hdWGCNA" and "ConstructNetwork" packages. In addition, we also evaluated the responsiveness of high and low-risk groups to 198 drugs using the "OncoPredict" software package. Result: We found that the glycolytic activity score of tumor tissue was significantly higher than that of normal tissue, and a total of 49 upregulated genes and 15 downregulated genes were selected from the total. Based on a machine learning computational framework, a total of 8 GRPS were screened, which constitute the prognostic features of LUAD patients. This feature demonstrates strong prognostic value, as confirmed by univariate and multivariate Cox regression analysis. Significant differences in tumor microenvironment (TME) immune infiltration were observed between high and low-risk groups. ScRNA seq revealed the distribution and expression of cell type specific GRPS, particularly in T cells, epithelial cells, and fibroblasts, while also revealing the strong cell-cell communication ability of the high GRPS group. The hdWGCNA analysis ultimately identified five core GRPS, namely DDIT4, FKBP4, CHPF, EFNA3, and B3GNT3. In addition, there are significant differences in sensitivity to most drugs between high-risk and low-risk cohorts, with WIKI4 and Lapatinib negatively correlated with risk scores, while Doramapimod and Niraparib positively correlated with risk scores. This study established a GRPS based risk feature for LUAD, demonstrating strong predictive power for prognosis assessment. The drug sensitivity results also provide drug guidance for the clinical application of this feature, all of which provide important clinical utility for the prognosis of LUAD. At the same time, the intercellular communication network was plotted based on the GRPS score, providing insights into the pathogenesis of LUAD and offering new ideas for developing targeted therapies and precision medicine methods.