On the basis of gene expression profiling, the laboratory proposed that the most common form of lymphoma, diffuse large B cell lymphoma (DLBCL), is a composite of three molecularly distinct diseases that are indistinguishable by standard diagnostic methods. These diseases, termed germinal center B cell-like (GCB) DLBCL, activated B cell-like (ABC) DLBCL, and primary mediastinal B cell lymphoma (PMBL), arise from B lymphocytes at different stages of differentiation by distinct oncogenic pathways. The curative response of patients with DLBCL to chemotherapy is highly variable, and the DLBCL subtype distinction accounts, in part, for this heterogeneity. With CHOP multi-agent chemotherapy, the 5-year survival rates of ABC DLBCL and GCB DLBCL are 60% and 30%, respectively. This clinical disparity likely reflects the host of genetic differences between these DLBCL subtypes. A recurring theme that emerges from our molecular profiling efforts in lymphoma is that the curative response to treatment and the length of survival following diagnosis are dictated by molecular features of the tumors at diagnosis. In DLBCL, we developed a multivariate model of therapeutic outcome based on gene expression signatures, which quantitatively reflected distinct aspects of tumor biology. This study was performed in the era of CHOP chemotherapy for DLBCL, which has subsequently been supplanted by regimens that add the anti-CD20 monoclonal antibody Rituximab (R-CHOP). Since R-CHOP improves survival in DLBCL by 10-15% compared with CHOP, it was important to investigate whether our previously identified prognostic gene expression signatures were still applicable. We therefore profiled gene expression in 233 pre-treatment biopsy samples from patients with DLBCL treated with R-CHOP as well as in 181 biopsies from CHOP-treated patients. We initially demonstrated that the distinction between the ABC and GCB DLBCL subtypes remained prognostically significant in R-CHOP-treated cases, with each subtype benefiting equivalently from the addition of Rituximab. In fact, 3 of the 4 gene expression signatures that were associated with survival in the CHOP cohort retained their prognostic significances in the R-CHOP cohort. This suggested that several aspects of tumor biology that influence the curative response to therapy remain important after the addition of Rituximab. We therefore constructed an optimal survival model based on gene expression signatures identified using the CHOP-treated cohort and then tested the model in the R-CHOP-treated cohort. The final model consisted of three signatures, termed germinal center B cell, stromal-1, and stromal-2. The model was strongly associated with both overall and progression-free survival in the CHOP and R-CHOP cohorts as well as in a third CHOP cohort for which gene expression profiling data were available. The gene expression-based survival model could divide the patients in the R-CHOP cohort into quartile groups that had 3-year progression-free survival rates of 84%, 69%, 61% and 34%, respectively, indicating that the model captures much of the heterogeneity in therapeutic response. Remaining heterogeneity could be ascribed to clinical factors in the International Prognostic Index, which were statistically independent from the prognostic gene expression signatures in predicting survival. An important feature of this study was our ability to associate the prognostic gene expression signatures with particular aspects of tumor biology. The germinal center B cell signature mirrored the distinction between GCB and ABC DLBCL and therefore reflects the myriad genetic and epigenetic differences that exist between these two subtypes24. On the other hand, the stromal-1 and stromal-2 signatures reflected different aspects of the tumor microenvironment. The stromal-1 signature, which was associated with favorable outcome, identified tumors that were fibrotic and rich in histiocytic cells of the myeloid lineage. The stromal-2 signature, which was associated with poor outcome, included a host of genes that are characteristically expressed in endothelial cells and was correlated with increased tumor blood vessel density, revealing an unanticipated role for angiogenesis in DLBCL. Array-based comparative genomic hybridization was used to identify genomic changes in copy number that influenced survival. Two genomic alterations that occurred exclusively in ABC DLBCL were deletion of the INK4a/ARF tumor suppressor locus and trisomy 3. These genetic aberrations, considered separately and together, identified a subset of patients with ABC DLBCL with inferior prognosis relative to other patients with this DLBCL subtype. This ABC DLBCL subset was also characterized by oncogenic mutations in the CARD11 gene, which encodes a scaffold molecule required for NF-kB signaling downstream of the B cell receptor. These mutations are responsible for constitutive NF-kB activity in ABC DLBCLs. The co-occurrence of CARD11 mutations, INK4a/ARF locus deletions and trisomy 3 in a subset of ABC DLBCLs suggests that this subset may arise by a distinct pathogenetic pathway. We are currently investigating several platforms to deliver the molecular diagnostic and prognostic distinction to patients with lymphoma. The goal is to utilize formalin-fixed and paraffin-embedded tissue for these analyses since most lymphoma biopsies are routinely stored in this fashion. Using RNA amplification from FFPE biopsy samples and Affymetrix array-based gene expression profiling, we can recapitulate the diagnostic distinction between ABC and GCB DLBCL with 95% concordance to the results obtained using frozen biopsy specimens. The Nanostring platform for digital gene expression analysis has also proved highly effective in distinguishing ABC and GCB DLBCL. Current efforts are aimed at the development of RNA-seq as a platform for whole transcriptome analysis in DLBCL and other lymphomas.
Showing the most recent 10 out of 41 publications
