The development and validation of a m6A-lncRNAs based prognostic model for overall survival in lung squamous cell carcinoma.

Abstract

No biomarkers have been identified for the prognosis of lung squamous cell carcinoma (LUSC). Risk models based on m6A-lncRNAs help to predict survival in some cancers. However, very few studies have reported m6A-lncRNA risk models in LUSC. We aimed to construct a prognostic model based on m6A-lncRNAs in LUSC. The clinical and RNA-sequencing information of 504 LUSC patients were downloaded from The Cancer Genome Atlas (TCGA) database. Prognostic m6A-lncRNAs were identified by a Pearson correlation analysis and univariate Cox regression analysis. The ConsensusClusterPlus algorithm was used to cluster the prognostic m6A-lncRNAs. The overall survival (OS) and clinicopathological characteristics of the 2 clusters were compared. A gene set enrichment analysis (GSEA) analysis was performed to analyze the genes enriched in the 2 clusters. A least absolute shrinkage and selection operator (LASSO) Cox regression analysis was used to construct the risk-score model. Two hundred and forty eight patients were randomly chosen from TCGA-LUSC cohort for the training set. The receiver operating characteristic (ROC) curve analysis was used to assess the predictive ability of the model. The clinical characteristics and OS in the high- and low-risk groups were compared. The independent prognostic value of the model was tested by Cox regression analyses. Thirteen m6A-lncRNAs were identified as prognostic lncRNAs and classified into cluster A and cluster B. The OS of patients in cluster A was better than that of patients in cluster B (P<0.001). Patients in cluster B had higher expressions of immune checkpoints. Immune score, stromal score, and ESTIMATE score were higher in cluster B (P<0.001). Seven of the 13 lncRNAs were used to construct the risk-score model. Patients in the high-risk group had a worse OS. ROC curves showed a under the curve (AUC) of 0.639 in the training set and 0.624 in the validation set. A high risk was associated with cluster B, a high immune score, and stage III-IV disease. Patients in the high-risk group had increased expressions of immune checkpoints. The Cox regression analyses showed that the risk-score model had independent prognostic value for OS. The risk-score model retained its prognostic value in different subgroups. The m6A-lncRNA risk-score model is an independent prognostic factor for OS in LUSC patients. However, the risk-score model need to be further tested clinically.