Diffuse Large B-cell Lymphoma (DLBCL) is the most common non-Hodgkin lymphoma worldwide. It is a heterogeneous disease in which one third of patients either do not respond to treatment or relapse within five years after chemotherapy. It is unclear whether epigenetic alterations are responsible for B cell lymphomas relapse phenotypes, such as increased aggressiveness and chemoresistance. To investigate how the B cell lymphoma epigenome evolves upon chemotherapy, we used Enhanced Reduced Representation Bisulfite Sequencing (ERRBS) to profile DNA methylation genome-wide in primary matched diagnosis-relapse DLBCL samples.

We interrogated 13 pairs of DLBCL diagnosis tumors and their matched relapse samples. In addition, we performed methylation profiling of normal tonsilar B cell subsets (Naïve and germinal center B cells) from two healthy human individuals. ERRBS provided DNA methylation levels at 3-4M CpG sites. When combining methylation levels from all interrogated CpG sites, we observed increased DNA methylation levels at CpG-islands (CGIs; p=3.5e-9, t-test) in DLBCLs compared to normal B cells, and stable or slightly decreasing methylation levels outside of CGIs (>10 kb away from known CGIs; p=0.057, t-test). There was no significant change in average DNA methylation levels from diagnosis to relapse either at CGIs or outside of CGIs. However, when we investigated DNA methylation changes at gene promoters, we identified 107 consistently differentially methylated promoters between diagnosis and relapse (> 10% DNA methylation alteration and p < 0.05, paired t-test). Pathway analysis of the corresponding genes using iPAGE identified several pathways and processes associated with either hyper or hypo-methylated genes in relapse, demonstrating that methylation changes associated with relapse are functionally coherent. For example, several genes with TGF-beta receptor activity displayed lower DNA methylation in relapse.

Taking advantage of single CpG resolution and high coverage provided by ERRBS, we then sought to investigate the extent of allele-specific methylation (ASM) levels in normal tissues and DLBCL patients. We found increased ASM levels in DLBCLs compared to normal tissues (p=0.0011, t-test) confirming observations in solid tumors. There was no significant change in ASM levels from diagnosis to relapse (p=0.24, t-test). These results suggest that methylation changes associated with lymphomagenesis might frequently involve one allele only, perhaps due to differential nuclear localization of individual chromosomes. However allele-specific methylation may not play a key role in lymphoma progression.

Finally, we investigated whether intra-tumor methylation heterogeneity at diagnosis would predict whether a DLBCL patient would relapse. We quantified intra-tumor methylation heterogeneity using a statistical approach based on the probability that two randomly sampled DNA molecules from the tumor cell populations differ from each other in their methylation pattern. We found decreased intra-sample methylation heterogeneity in DLBCLs compared to normal germinal center B cells (p=1.9e-4, t-test), consistent with the clonal origin of tumors. 12 out of 13 pairs also displayed decreased methylation heterogeneity from diagnosis to relapse, which is also consistent with clonal selection upon chemotherapy treatment. We then performed ERRBS on primary tumors from 8 DLBCL patients who have not relapsed five years after treatment. We found that non-relapse patients displayed significantly lower intra-tumor methylation heterogeneity as compared to that of the relapsed patients (p=0.047, t-test), which suggests that increased epigenetic diversity within a population of tumor cells at diagnosis might fuel the Darwinian evolutionary process underlying relapse. We also looked at genetic clonal heterogeneity based on next-generation sequencing of somatic hypermutation profiles in IGH VDJ sequences, but found no differences between relapsed and not relapsed patients (p=0.22, Wilcoxon test). This suggests that epigenetic heterogeneity plays a more substantial role than clonal heterogeneity in fueling the relapse phenotype.

In summary, this study provides the first comprehensive characterization of aberrations in DNA methylation in relapse DLBCLs and identified epigenetic diversity in DLBCLs as a potential predictor of relapse.


No relevant conflicts of interest to declare.

Author notes


Asterisk with author names denotes non-ASH members.