Alterations of 5-hydroxymethylation in circulating cell-free DNA reflect molecular distinctions of subtypes of non-Hodgkin lymphoma

The 5-methylcytosines (5mC) have been implicated in the pathogenesis of diffuse large B-cell lymphoma (DLBCL) and follicular lymphoma (FL). However, the role of 5-hydroxymethylcytosines (5hmC) that are generated from 5mC through active demethylation, in lymphomagenesis is unknown. We profiled genome-wide 5hmC in circulating cell-free DNA (cfDNA) from 73 newly diagnosed patients with DLBCL and FL. We identified 294 differentially modified genes between DLBCL and FL. The differential 5hmC in the DLBCL/FL-differentiating genes co-localized with enhancers marks: H3K4me1and H3K27ac. A four-gene panel (CNN2, HMG20B, ACRBP, IZUMO1) robustly represented the overall 5hmC modification pattern that distinguished FL from DLBCL with an area under curve of 88.5% in testing set. The median 5hmC modification levels in signature genes showed potential for separating patients for risk of all-cause mortality. This study provides evidence that genome-wide 5hmC profiles in cfDNA differ between DLBCL and FL and could be exploited as a non-invasive approach. We obtained genome-wide 5hmC profiles using the 5hmC-Seal technique in cfDNA samples from 48 DLBCL patients and 25 FL patients at the University of Chicago Medical Center from 2010 to 2013. Patients were followed through March 18, 2019. The normalized 5hmC-Seal count data of 17,698 gene bodies with non-zero variance from the total 73 samples were used to identify informative genes with differential modification between FL and DLBCL. The elastic net regularization was applied to the multivariate logistic regression models with the 10-fold cross validation using the glmnet package in the R Statistical Environment, to select the most representative genes with differential modifications. To explore the gene regulatory relevance of 5hmC in cfDNA, we also summarized the 5hmC-Seal data to various genomic features from the Roadmap Epigenomics Project B-cell data. Functional relevance of the differentially modified genes to biological pathways and processes were explored using the Database Visualization and Integrated Discovery (DAVID) tool. The prognostic value of a given gene was assessed using the multivariate Cox Proportional Hazards model, controlling age at the time of initial diagnosis and gender.