Background
Grade 4 glioma is the most aggressive and currently incurable brain tumor with a median survival of one year in adult patients. Elucidating novel transcriptomic and epigenetic contributors to the molecular heterogeneity underlying its aggressiveness may lead to improved clinical outcomes.
Conclusions
The 5hmC-based prognostic model could offer a robust tool to predict survival in patients with grade 4 gliomas, potentially outperforming existing prognostic factors such as IDH1 mutations. The crosstalk between 5hmC and gene expression revealed another layer of complexity underlying the molecular heterogeneity in grade 4 gliomas, offering opportunities for identifying novel therapeutic targets.
Methods
To identify grade 4 glioma -associated 5-hydroxymethylcytosine (5hmC) and transcriptomic features as well as their cross-talks, genome-wide 5hmC and transcriptomic profiles of tissue samples from 61 patients with grade 4 gliomas and 9 normal controls were obtained for differential and co-regulation/co-modification analyses. Prognostic models on overall survival based on transcriptomic features and the 5hmC modifications summarized over genic regions (promoters, gene bodies) and brain-derived histone marks were developed using machine learning algorithms.
Results
Despite global reduction, the majority of differential 5hmC features showed higher modification levels in grade 4 gliomas as compared to normal controls. In addition, the bi-directional correlations between 5hmC modifications over promoter regions or gene bodies and gene expression were greatly disturbed in grade 4 gliomas regardless of IDH1 mutation status. Phenotype-associated co-regulated 5hmC-5hmC modules and 5hmC-mRNA modules not only are enriched with different molecular pathways that are indicative of the pathogenesis of grade 4 gliomas, but also are of prognostic significance comparable to IDH1 mutation status. Lastly, the best-performing 5hmC model can predict patient survival at a much higher accuracy (c-index = 74%) when compared to conventional prognostic factor IDH1 (c-index = 57%), capturing the molecular characteristics of tumors that are independent of IDH1 mutation status and gene expression-based molecular subtypes. Conclusions: The 5hmC-based prognostic model could offer a robust tool to predict survival in patients with grade 4 gliomas, potentially outperforming existing prognostic factors such as IDH1 mutations. The crosstalk between 5hmC and gene expression revealed another layer of complexity underlying the molecular heterogeneity in grade 4 gliomas, offering opportunities for identifying novel therapeutic targets.