Abstract
Abstract Abstract 619 Diffuse Large B-cell Lymphoma (DLBCL) is a complex biological entity with heterogeneous genetic, biological and clinical features. Gene expression profiling studies have attempted to resolve some of this heterogeneity. For example, DLBCL patients harboring gene signatures associated with expression of germinal center B-cell genes (GCB) or activated B-cell genes (ABC) vary in their response rate to standard chemo-immunotherapy regimens. Since epigenetic gene regulation can play a fundamental role in determining the phenotype of normal and malignant tissues we asked whether ABC and GCB DLBCLs also display unique epigenetic signatures that might be clinically useful and further explain the biology of these tumors. For this purpose we examined the DNA methylation level of 50,000 cytosine residues distributed among 14,000 gene promoters in a cohort of 159 patients with DLBCL, all of whom were uniformly treated with R-CHOP, using the HELP assay and high-density oligonucleotide microarrays. For a subset of these patients (n=69), we also performed Affymetrix gene expression profiling. First, a Bayesian predictor of ABC/GCB subtypes was trained from a published expression profiling study of 203 DLBCL patients. The predictor was then applied to our cohort of 69 patients. At a probability cutoff of 0.9, 20 patients were classified as ABC, 40 were classified as GCB and 9 could not be classified (i.e. “type III DLBCL”). As expected from published studies, the differences in progression free survival (PFS) and overall survival (OS) of these ABC vs. GCB patients thus determined was highly significant, with p=0.0026 (log rank) and p=0.043 (log rank) respectively, with a much worse prognosis for ABC patients compared to GCB ones. We then asked whether the ABC and GCB subtypes could be predicted from the DNA methylation profiles of the same 69 patients. By performing a t-test we found that 239 genes were differentially methylated between ABC and GCB (p<0.0001) and also displayed >30% difference in methylation level. This DNA methylation signature was incorporated into a new Bayesian predictor, which we showed to predict ABC and GCB DLBCL subtypes from DNA methylation profiles with a 91% accuracy. Using a cross-validation procedure, we estimated that the classification performance on independent cases to be ∼87%. The predicted ABC and GCB cases retained the clinical predictive power of the gene expression profile when applied to the remaining 90 patients that did not have gene expression profiles, confirming its clinical relevance (difference in PFS p=0.0148, log rank). Gene set enrichment analysis showed that the ABC DNA methylation signature was enriched in genes involved in antigen dependent B and T-cell responses and in TNF inflammatory responses (p<1.01E-4 and <6.01E-4 respectively). A computational analysis of promoter DNA sequences of the genes involved in this signature revealed over-representation of binding sites for transcription factors including MYB, STAT5A, MAZ, and JUN and Sp1; many of these factors have a known role in B cell development and function. The role of Sp1 in these tumors is under further examination. Among the 239 genes that were differentially methylated and the 411 genes that were differentially expressed between ABC and GCB there was an overlap of 16 genes (greater than expected by chance Fisher Exact p=0.005). A predictor based on the methylation profiles of these 16 genes was on its own 92% accurate in identifying ABC vs. GCB cases among our cohort of DLBCLs. Although there was a general trend for inverse correlation in expression between the 239 differentially methylated genes, these 16 overlapping genes displayed marked and extreme inverse correlation. This was validated by single locus quantitative methylation sequencing (MassArray) and QPCR. The 16 genes included genes known to play critical roles in B-cell differentiation, proliferation and metabolism but not previously implicated in DLBCL. Gain and loss of function assays of a subset of these genes in ABC and GCB DLBCL cell lines show that they have tumor suppressor functions in DLBCL. Our results demonstrate for the first time that ABC and GCB DLBCLs are epigenetically distinct diseases; they identify new biological differences and candidate tumor suppressor genes between these tumors and demonstrate that a DNA methylation classifier can be used to distinguish GCB and ABC DLBCL subtypes. Disclosures: No relevant conflicts of interest to declare.