Genome-Wide Methylation Analysis of Patients with Diffuse Large B Cell Lymphoma Treated on the UK NCRI R-CHOP 14 Vs 21 Trial

Background:

Epigenetic alterations are a hallmark of diffuse large B cell lymphoma (DLBCL). A high degree of methylation disruption and intra-tumor methylation heterogeneity have been linked to poor outcome in DLBCL. However, high-coverage methylation data within prospective trial cohorts are warranted to characterize clinically relevant methylation subgroups of DLBCL. Here, we investigate the methylome of DLBCL cases from a large multicentre Phase III trial and its interaction with clinical and molecular characteristics.

Methods:

DNA methylation was assessed with the Illumina Infinium 450k array in archived FFPE samples of 127 newly diagnosed DLBCL patients treated with R-CHOP chemotherapy on the UK NCRI R-CHOP14v21 trial. Status of BCL2-, BCL6-, and MYC-rearrangements and amplifications was available for 111 cases. Cell-of-origin classification was available for all cases, both according to the IHC Hans algorithm and the microarray-based classification tool using Illumina DASL technology (Barrans, Br J Haematol 2012). In addition, microarray data was used to extract mean expression values of the epigenetic modifiers EZH2, MLL2, CREBBP, EP300, DNMT3A, IDH1/2 and TET2 and patients in the lowest and highest expression quartiles were compared. Analyses were performed in R 3.2.3 using the RnBeads package. Differential methylation was analyzed on region level (tiling, genes, promoters, CpG islands). P values were adjusted for multiple testing using the false discovery rate method.

Results:

After normalization and quality control 425,623 probes were included in the analysis. Median methylation levels of probes showed a bimodal distribution (unmethylated/methylated) across samples, with CpG islands and shores being mainly unmethylated, and CpG shelves and open seas being predominantly methylated. No significant association of clinical prognostic factors (age, LDH level, extranodal involvement, stage, WHO performance status) and methylation could be observed. We did not find significant differential methylation between patients who relapsed after R-CHOP therapy and those who did not. However, more detailed outcome analyses using Cox regression models for progression- and overall survival on single-site level will be provided at the meeting. There was no evidence for a difference in methylation with regards to rearrangements or amplifications of MYC or BCL6, nor with presence of double-hit abnormalities. Interestingly, cases with BCL2 rearrangement showed differential methylation of 39 promoter regions compared to cases without this abnormality (all adj. P<0.05). A strong impact on methylation patterns was seen with regards to microarray-based cell-of-origin categories (but not according to the IHC classifier). GCB vs. ABC cases showed global hypermethylation, predominantly affecting CpG islands and shores. 388 promoter regions were differentially methylated between GCB and ABC subtypes, corresponding to 226 annotated genes or miRNAs, the majority (93%) being hypermethylated in GCB cases. Differentially methylated promoter regions in GCB lymphomas were enriched for signal transduction and cell-cell communication pathway genes (adj. P<0.05) and included EZH2 target genes such as CDKN1A and SOX9. mRNA expression levels of epigenetic modifiers was not associated with significant differences in methylation, apart from DNMT3A expression. Patients with high vs. low DNMT3Aexpression showed differential methylation of 3300 promoters, which were significantly enriched for KRAS-associated genes (P=3.68E-12), cancer pathway genes (e.g. Wnt-, FGF- and hedgehog signalling; P=1.45E-08), as well as cytokine signalling genes (P=1.96E-09). Data on recurrent somatic mutations of epigenetic modifiers and correlation of methylation and mRNA expression will be provided at the meeting.

Conclusions:

The cell-of-origin as well as expression levels of DNMT3A are main drivers of methylation variability in DLBCL, whereas other molecular features like MYC seem to have little impact on methylation patterns. This comprehensive dataset of genome-wide methylation profiles from a prospective trial cohort provides important information for identifying biologically and clinically relevant epigenetic subgroups of DLBCL.