A multi-omics analysis-based model to predict the prognosis of low-grade gliomas

Abstract

AbstractLower-grade gliomas (LGGs) exhibit highly variable clinical behaviors, while classic histology characteristics cannot accurately reflect the authentic biological behaviors, clinical outcomes, and prognosis of LGGs. In this study, we carried out analyses of whole exome sequencing, RNA sequencing and DNA methylation in primary vs. recurrent LGG samples, and also combined the multi-omics data to construct a prognostic prediction model. TCGA-LGG dataset was searched for LGG samples. 523 samples were used for whole exome sequencing analysis, 532 for transcriptional analysis, and 529 for DNA methylation analysis. LASSO regression was used to screen genes with significant association with LGG survival from the frequently mutated genes, differentially expressed genes, and differentially methylated genes, whereby a prediction model for prognosis of LGG was further constructed and validated. The most frequently mutated diver genes in LGGs were IDH1 (77%), TP53 (48%), ATRX (37%), etc. Top significantly up-regulated genes were C6orf15, DAO, MEOX2, etc., and top significantly down-regulated genes were DMBX1, GPR50, HMX2, etc. 2077 genes were more and 299 were less methylated in recurrent vs. primary LGG samples. Thirty-nine genes from the above analysis were included to establish a prediction model of survival, which showed that the high-score group had a very significantly shorter survival than the low-score group in both training and testing sets. ROC analysis showed that AUC was 0.817 for the training set and 0.819 for the testing set. This study will be beneficial to accurately predict the survival of LGGs to identify patients with poor prognosis to take specific treatment as early, which will help improve the treatment outcomes and prognosis of LGG.