Genetically-Defined Diffuse Large B-Cell Lymphoma Subsets Arise By Distinct Pathogenetic Mechanisms and Predict Outcome

Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous disease that largely arises from antigen-exposed B-cells that transit through the germinal center (GC). DLBCL is further classified into transcriptional subtypes - activated B-cell (ABC) and GC B-cell (GCB). ABC DLBCLs exhibit genetic alterations in NF-kB modifiers and proximal components of the B-cell receptor (BCR) pathway and perturbed terminal B-cell differentiation. GCB DLBCLs have reported alterations in chromatin-modifying enzymes, PI3K signaling and Gα-migration pathway components and frequent translocations of BCL2 .

Despite the recognized molecular heterogeneity in DLBCL, previously published genomic analyses have largely focused on single types of alterations - mutations, copy number alterations (CNAs) or structural variants (SVs) - in smaller data sets with more limited clinical annotation. An unbiased comprehensive analysis of all three alteration types is needed to define discrete, clinically annotated subtypes of DLBCL.

We performed whole exome sequencing (WES) of 304 newly diagnosed DLBCLs, using an expanded bait set that captures recurrent SVs; 85% of study patients were uniformly treated with R-CHOP and had long-term follow-up. Somatic alterations (mutations, CNAs and SVs) and their clonality were determined with analytical pipelines developed at the Broad Institute. Notably, half of our DLBCL cohort lacked patient-matched normal samples, prompting the successful development of new methods to analyze tumor-only WES data. Significantly mutated genes (SMGs) were identified with MutSig2CV andrecurrent CNAs were defined with GISTIC2.0 .

With increased sample size and improved methodology, we identified 158 SMGs, CNAs and SVs; the 98 SMGs included ones previously unreported in DLBCL but described in other lymphoid malignancies or cancer. Additional insights into the putative biological function of newly identified alterations were obtained by overlaying the predicted protein changes onto their 3-dimensional protein structures.

We analyzed the mutational signatures in these DLBCLs and found that the majority of exome mutations were caused by spontaneous deamination at CpGs, a clock-like genetic signature associated with aging; only a minority of mutations were attributed to activation-induced cytidine deaminase (AID), an enzyme required for physiologic immunoglobulin receptor editing and aberrant somatic hypermutation.

IGH, BCL2, BCL6 and MYC were the most frequently rearranged genes (40%, 21%, 19% and 8%, respectively). There were 18 arm-level and 18 focal regions of copy gain and 2 arm-level and 32 focal regions of copy loss with frequencies ranging between 5-32%.

SMGs were significantly more likely to reside within focal CNAs (p=1x10^-44), suggesting that these driver genes were perturbed by multiple mechanisms. Individual DLBCLs had a median of 17 genetic drivers, highlighting the need for a more comprehensive analysis. Therefore, we applied an integrated clustering approach to the recurrent mutations, CNAs and SVs and delineated 5 genetically distinct DLBCL subtypes and determined the likely temporal order of alterations within each cluster.

This unbiased approach identified a previously unappreciated favorable risk ABC subset with genetic features of an extrafollicular, possibly marginal zone origin; a more tightly defined poor risk group of ABC DLBCLs with frequent BCL2 gain and concordant MYD88^L265P / CD79B mutations; a distinct poor risk subset of GCB tumors with BCL2 SVs and alterations of PTEN and epigenetic enzymes and a discrete group of good risk GCB DLBCLs with specific alterations in BCR/PI3K, JAK/STAT and BRAF pathway components and multiple histones; and a Cell-of-Origin-independent subset with biallelic inactivation of TP53, 9p21.3/CDKN2A and associated genomic instability. These findings likely explain the variability in outcome predictions with the binary ABC- vs. GCB-DLBCL transcriptional classification and the challenges of therapeutically targeting less well defined DLBCLs.

The genetically distinct DLBCL subsets provide a framework for assessing previously unrecognized heterogeneity in this disease, characterizing combinations of genetic alterations that drive DLBCL biology and guiding the development of rational single-agent and combination therapies in patients with the greatest need.