Liquid biopsy techniques based on deep sequencing of plasma cell-free DNA (cfDNA) could detect the low-frequency somatic mutations and provide an accurate diagnosis for many cancers. However, for brain gliomas, reliable performance of these techniques currently requires obtaining cfDNA from patients' cerebral spinal fluid, which is cumbersome and risky. Here we report a liquid biopsy method based on sequencing of plasma cfDNA fragments carrying 5-hydroxymethylcytosine (5hmC) using selective chemical labeling (hMe-Seal). We first constructed a dataset including 180 glioma patients and 229 non-glioma controls. We found marked concordance between cfDNA hydroxymethylome and the aberrant transcriptome of the underlying gliomas. Functional analysis also revealed overrepresentation of the differentially hydroxymethylated genes (DhmGs) in oncogenic and neural pathways. After splitting our dataset into training and test cohort, we showed that a penalized logistic model constructed with training set DhmGs could distinguish glioma patients from healthy controls in both our test set (AUC = 0.962) and an independent dataset (AUC = 0.930) consisting of 111 gliomas and 111 controls. Additionally, the DhmGs between gliomas with mutant and wild-type isocitrate dehydrogenase (IDH) could be used to train a cfDNA predictor of the IDH mutation status of the underlying tumor (AUC = 0.816), and patients with predicted IDH mutant gliomas had significantly better outcome (P = .01). These results indicate that our plasma cfDNA 5hmC sequencing method could obtain glioma-specific signals, which may be used to noninvasively detect these patients and predict the aggressiveness of their tumors.

IntroductionHigh-grade glioma (HGG) defines a group of brain gliomas characterized by contrast enhancement, high tumor heterogeneity, and poor clinical outcome. Disturbed reduction–oxidation (redox) balance has been frequently associated with the development of tumor cells and their microenvironment (TME).MethodsTo study the influence of redox balance on HGGs and their microenvironment, we collected mRNA-sequencing and clinical data of HGG patients from TCGA and CGGA databases and our own cohort. Redox-related genes (ROGs) were defined as genes in the MSigDB pathways with keyword “redox” that were differentially expressed between HGGs and normal brain samples. Unsupervised clustering analysis was used to discover ROG expression clusters. Over-representation analysis (ORA), gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA) were also employed to understand the biological implication of differentially expressed genes between HGG clusters. CIBERSORTx and ESTIMATE were used to profile the immune TME landscapes of tumors, and TIDE was used to evaluated the potential response to immune checkpoint inhibitors. Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression was used to construct HGG-ROG expression risk signature (GRORS).ResultsSeventy-five ROGs were found and consensus clustering using the expression profile of ROGs divided the both IDH-mutant (IDHmut) and IDH-wildtype (IDHwt) HGGs into subclusters with different prognosis. Functional enrichment analysis revealed that the differential aggressiveness between redox subclusters in IDHmut HGGs were significantly associated with cell cycle regulation pathways, while IDHwt HGG redox subclusters showed differentially activated immune-related pathways. In silico TME analysis on immune landscapes in the TME showed that the more aggressive redox subclusters in both IDHmut and IDHwt HGGs may harbor a more diverse composition of tumor-infiltrating immune cells, expressed a higher level of immune checkpoints and were more likely to respond to immune checkpoint blockade. Next, we established a GRORS which showed AUCs of 0.787, 0.884, and 0.917 in predicting 1–3-year survival of HGG patients in the held-out validation datasets, and the C-index of a nomogram combining the GRORS and other prognostic information reached 0.835.ConclusionBriefly, our results suggest that the expression pattern of ROGs was closely associated with the prognosis as well as the TME immune profile of HGGs, and may serve as a potential indicator for their response to immunotherapies.