Construction and validation of an anoikis-related prognostic model for lung adenocarcinoma based on bulk and single-cell transcriptomic data

Abstract

Lung adenocarcinoma (LUAD) is a highly aggressive lung cancer with poor prognosis due to lack of reliable biomarkers. Resistance to anoikis drives tumor progression and metastasis. This study aims to develop and validate an anoikis-related prognostic model for LUAD. We employed univariate Cox regression analysis, LASSO regression, and random forest algorithms to identify anoikis-related genes (ARG) from bulk transcriptomic datasets, and establish a 7-gene prognostic signature, validated in two LUAD cohorts from GEO database. We evaluated immune infiltration, molecular functions, and genomic alterations between risk groups and analyzed single-cell RNA sequencing data. IHC and mIF validated TIMP1 expression and its interaction with Treg cells. We developed a 7-gene prognostic model (LDHA, PLK1, TRAF2, ITGB4, SLCO1B3, TIMP1, ZEB2) using machine learning to predict survival in LUAD patients. The model accurately predicted 1-year survival rates (GSE31210: AUC = 0.805; GSE30219: AUC = 0.787), 2-year survival rates (GSE31210: AUC = 0.769; GSE30219: AUC = 0.681), and 3-year survival rates (GSE31210: AUC = 0.695; GSE30219: AUC = 0.735) and correlated with clinical features, immune infiltration, and tumor microenvironment (TME) remodeling. Single-cell sequencing data showed that LUAD patients exhibited an immunosuppressive TME phenotype, which was exacerbated by high TIMP1 expression in epithelial cells, promoting Treg cell activity. The 7-gene ARG prognostic model established in this study shows promising potential as a clinically applicable tool for decision-making.