Identification of an Autophagy-Related 10-lncRNA-mRNA Signature for Distinguishing Glioblastoma Multiforme from Lower‑Grade Glioma and Prognosis Prediction

Background: Autophagy provides the nutrients and energy for tumor growth, invasion and metastasis. Theoretically, autophagy-related mRNAs and regulatory long non-coding RNAs (lncRNAs) may represent promising biomarkers to predict tumor progression and poor prognosis. Our study aims to develop an autophagy-related signature to distinguish glioblastoma (GBM) from lower-grade gliomas (LGG) and predict overall survival (OS). Methods: The expression profile of GBM and LGG was collected from the Chinese Glioma Genome Atlas (CGGA) database that was used to identify differentially expressed genes (DEGs) and lncRNAs (DELs). The autophagy-related genes were obtained from the Human Autophagy Database. The autophagy-related DELs were identified by a co-expression network with DEGs. These DEGs and DELs underwent univariate and multivariate analyses to screen prognostic genes. They were entered into the Logit regression model to identify the GBM feature genes. The prognostic signature was evaluated by survival curve analyses and validated using The Cancer Genome Atlas (TCGA) dataset. The prognostic model and clinicopathological parameters were integrated to construct the nomogram. Results: A total of 131 autophagy-related DEGs and 54 autophagy-related DELs were identified. Ten of them were demonstrated as independent prognostic factors and could distinguish GBM from LGG, with the accuracy of 0.891 using CGGA dataset and 0.790 using TCGA dataset. The risk score was established based on these 10 genes. Patients with higher risk score were at an increased risk of developing GBM (49.7% vs 21.3%) and worse OS prognosis than those in low risk group. The predictive accuracy was 0.840 and 0.744 for CGGA and TCGA dataset, respectively. Multivariate analysis showed age, recurrence, IDH mutation and risk score status were independent prognostic factors and thus they were used to build a nomogram which showed the highest predictive power than other factors. Conclusion: The established nomogram may aid the clinical decision making of personalized treatment.