Identification of Cancer Stem Cell-related Gene by Single-cell and Machine Learning Predicts Immune Status, Chemotherapy Drug, and Prognosis in Lung Adenocarcinoma.

Abstract

Aim: This study aimed to identify the molecular type and prognostic model of lung adenocarcinoma (LUAD) based on cancer stem cell-related genes. Studies have shown that cancer stem cells (CSC) are involved in the development, recurrence, metastasis, and drug resistance of tumors.

Method: The clinical information and RNA-seq of LUAD were obtained from the TCGA database. scRNA dataset GSE131907 and 5 GSE datasets were downloaded from the GEO database. Molecular subtypes were identified by ConsensusClusterPlus. A CSC-related prognostic signature was then constructed via univariate Cox and LASSO Cox-regression analysis.

Result: A scRNA-seq GSE131907 dataset was employed to obtain 11 cell clusters, among which, 173 differentially expressed genes in CSC were identified. Moreover, the CSC score and mRNAsi were higher in tumor samples. 18 of 173 genes were survival time-associated genes in both the TCGA-LUDA dataset and the GSE dataset. Next, two molecular subtypes (namely, CSC1 and CSC2) were identified based on 18 survival-related CSC genes with distinct immune profiles and noticeably different prognoses as well as differences in the sensitivity of chemotherapy drugs. 8 genes were used to build a prognostic model in the TCGA-LUAD dataset. High-risk patients faced worse survival than those with a low risk. The robust predictive ability of the risk score was validated by the time-dependent ROC curve revealed as well as the GSE dataset. TIDE analysis showed a higher sensitivity of patients in the low group to immunotherapy.

Conclusion: This study has revealed the effect of CSC on the heterogeneity of LUAD, and created an 8 genes prognosis model that can be potentially valuable for predicting the prognosis of LUAD and response to immunotherapy.