Search

Background Nucleotide-specific 5-hydroxymethylcytosine (5hmC) remains understudied in pediatric central nervous system (CNS) tumors. 5hmC is abundant in the brain, and alterations to 5hmC in adult CNS tumors have been reported. However, traditional approaches to measure DNA methylation do not distinguish between 5-methylcytosine (5mC) and its oxidized counterpart 5hmC, including those used to build CNS tumor DNA methylation classification systems. We measured 5hmC and 5mC epigenome-wide at nucleotide resolution in glioma, ependymoma, and embryonal tumors from children, as well as control pediatric brain tissues using tandem bisulfite and oxidative bisulfite treatments followed by hybridization to the Illumina Methylation EPIC Array that interrogates over 860,000 CpG loci. Results Linear mixed effects models adjusted for age and sex tested the CpG-specific differences in 5hmC between tumor and non-tumor samples, as well as between tumor subtypes. Results from model-based clustering of tumors was used to test the relation of cluster membership with patient survival through multivariable Cox proportional hazards regression. We also assessed the robustness of multiple epigenetic CNS tumor classification methods to 5mC-specific data in both pediatric and adult CNS tumors. Compared to non-tumor samples, tumors were hypohydroxymethylated across the epigenome and tumor 5hmC localized to regulatory elements crucial to cell identity, including transcription factor binding sites and super-enhancers. Differentially hydroxymethylated loci among tumor subtypes tended to be hypermethylated and disproportionally found in CTCF binding sites and genes related to posttranscriptional RNA regulation, such as DICER1. Model-based clustering results indicated that patients with low 5hmC patterns have poorer overall survival and increased risk of recurrence. Our results suggest 5mC-specific data from OxBS-treated samples impacts methylation-based tumor classification systems giving new opportunities for further refinement of classifiers for both pediatric and adult tumors. Conclusions We identified that 5hmC localizes to super-enhancers, and genes commonly implicated in pediatric CNS tumors were differentially hypohydroxymethylated. We demonstrated that distinguishing methylation and hydroxymethylation is critical in identifying tumor-related epigenetic changes. These results have implications for patient prognostication, considerations of epigenetic therapy in CNS tumors, and for emerging molecular neuropathology classification approaches.

DNA methylation (DNAm) alterations have been heavily implicated in carcinogenesis and the pathophysiology of diseases through upstream regulation of gene expression. DNAm deep-learning approaches are able to capture features associated with aging, cell type, and disease progression, but lack incorporation of prior biological knowledge. Here, we present modular, user-friendly deep learning methodology and software, MethylCapsNet and MethylSPWNet, that group CpGs into biologically relevant capsules – such as gene promoter context, CpG island relationship, or user-defined groupings – and relate them to diagnostic and prognostic outcomes. We demonstrate these models’ utility on 3,897 individuals in the classification of central nervous system (CNS) tumors. MethylCapsNet and MethylSPWNet provide an opportunity to increase DNAm deep learning analyses’ interpretability by enabling a flexible organization of DNAm data into biologically relevant capsules.

Though aberrant cytosine modifications are prevalent in cancer, nucleotide-specific 5-hydroxymethylcytosine (5hmC) modifications remain understudied, including in pediatric CNS tumors. Brain 5-hydroxymethylation is linked with development and differentiation. We measured genome-scale nucleotide-specific 5hmC in patients with diagnoses of glioma, ependymoma, and embryonal tumors under age 18 (n=36), and in non-tumor pediatric brain tissues (n=3). DNA was processed with tandem oxidative (OxBS) and bisulfite (BS) treatments followed by hybridization to the Illumina Methylation EPIC Array that interrogates over 860,000 CpG sites. We used the OxyBS R package to determine levels of 5hmC and 5mC. Mean 5hmC levels were lower in tumors (gliomas 4.1%, ependymomas 3.9%, and embryonal tumors 3.4%) compared to nontumor tissues (5.3%). We subset to the CpGs with the 5% highest 5hmC content for downstream analyses (37,173 CpGs). These sites were enriched among regulatory elements, including TFBS (Odds Ratio 1.14 p-value 3.57E-20) and super-enhancers (OR 1.93, p-value 1.14E-126). Linear mixed-effects models adjusted for age, sex, and cell type proportions tested the CpG-specific differences in 5hmC between tumor and nontumor samples, as well as between tumor subtypes. 5hmC levels were depleted across tumors compared with nontumor brain tissues, including at CpG islands. Model-based clustering (RPMM) results indicated that patients with low 5hmC patterns have poorer overall survival and increased risk of recurrence. Our results indicate that 5hmC localizes to sites in the DNA critical to gene regulation and is associated with patient outcomes. This study offers an opportunity to potentially contribute to classification markers for childhood brain tumors.