A prognostic signature model for unveiling tumor progression in lung adenocarcinoma

Abstract

A more accurate prognosis is important for clinical treatment of lung adenocarcinoma. However, due to the limitation of sample and technical bias, most prognostic signatures lacked reproducibility, and few were applied to clinical practice. In addition, understanding the molecular driving mechanism is indispensable for developing more promising therapies for lung adenocarcinoma. Here, we built an unbiased prognostic significance model to perform an integrative analysis, including differentially expressed genes and clinical data with lung adenocarcinoma patients from TCGA. Multivariable Cox proportional hazards model with the Lasso penalty and 10-fold cross-validate were used to identify the best gene signature. We generated a 17-gene signature for prognostic risk prediction based on the overall survival time of lung adenocarcinoma patients. To further test the model’s predictive ability, we have applied an independent GEO database to verify the predictive ability of prognostic signature. The model can more objectively describe several biological processes related to tumors and reveal important molecular mechanisms in tumor development by GO and KEGG analysis. Furthermore, differential expression analysis by GSEA revealed that tumor microenvironments such as ER stress, exosome, and immune microenvironment were enriched. Using single-cell RNA sequence technology, we found that risk score was positively correlated with lung adenocarcinoma marker genes and copy number variation but negatively correlated with lung epithelial marker genes. High-risk cell populations with the model had stronger cancer stemness and tumor-related pathway activation. As we expected, the risk score was in accordance with the malignancy of each cluster from tumor progression. In conclusion, the risking model established in this study is more reliable than others in evaluating the prognosis of LUAD patients.