A Diffuse Large B-Cell Lymphoma Classification System That Associates Normal B-Cell Subset Phenotypes with Prognosis

Background

Recent findings have introduced biological classification of non-Hodgkin lymphomas as exemplified by the “activated B-cell-like” (ABC) and “germinal-center B-cell–like” (GCB) subgroups of diffuse large B-cell lymphoma (DLBCL).

Aims

The goal is to generate a refined cell of origin (COO) classification that includes B-cell subset associated gene signatures (BAGS) from the normal B-cell hierarchy.

Methods

We have combined fluorescence activated cell sorting and gene expression profiling by Gene Chip HG U133 Plus 2.0 to generate five BAGS for naïve, germinal centrocytes and centroblasts, post germinal memory B-cells, and plasmablasts of normal human tonsils. Clinical data sets are the Aalborg Project CHEPRETRO (N=89), the Lymphoma/Leukemia Molecular Profiling Project LLMPP (N=233), the International DLBCL Rituximab-CHOP Consortium MD Anderson Project IDRC (N=460), the Mayo Clinic, Brigham & Women Hospital, and Dana-Farber Cancer Institute Project MDFCI (N=88) available on the GEO website. All statistical analyses were done with R version 3.0.1 and full documentation is provided by a Sweave document.

Results

First, we verified the quality of the sampled B-cell subsets based on the expression patterns of differentiation molecules, transcription factors, and genes matching biological knowledge. Next, we constructed a BAGS-classifier provided by 77-115 gene probe sets, capable of assigning samples to each of the five COO subtypes.

Second, we assigned individual lymphoma cases in 5 patient cohorts including a total of 1063 patient. BAGS identified COO subtypes with frequencies of 2-7 % naïve (N), 35-41 % centrocytes (CC), 18-21 % centroblasts (CB), 4-15 % memory (M), 12-18 % plasmablast (PB), and 15-16 % unclassified (UC) subtypes. The frequencies was not different between cohorts (p=0.41).

Third, the BAGS subtypes was associated significantly with overall survival and time to progression for R-CHOP–treated patients in clinical cohorts from the LLMPPN (p=0.0441/0.0358) and the IDRC Program (p=0.002/8e-04).

Fourth, we found a very high fraction of GCB samples to be of CC or CB subtypes. On the contrary, a major fraction of BAGS-unclassified subtypes were of the ABC class. In a multiple Cox proportional hazards model we identified the BAGS subtypes to be a prognostic variables independent of ABC/GCB subtypes but not of IPI and age. The most significant impact was observed within the GCB subclass, where the GCB-CC subtype had superior prognosis compared to the GCB-CB subtype, in accordance with individual assignments for drug specific sensitivity to hydroxydaunomycin and vincristine.

Fifth, we performed genetic evaluation of the BAGS subtypes by mutation analysis within the CHEPRETRO cohort for EZH2, CD79B, and MYD88 identifying frequencies of 6.3%, 10.1% and 14.7%, respectively. The EZH2 mutation was only identified in the GCBN-CC and -CB subtypes. Mutations of CD79B and MYD88 were preferential in ABC, present in all subtypes. The CC subtype had high p53 mutation and indel frequencies, whereas the CB subtype had high Chr12q15 amplification frequencies and a complex genotype. Finally, the CC subtype expressed LMO2, NF-κB targets, CD58, Stroma1, and MHCII genes, known to have prognostic impact.

Summary

In summary, this study addresses an unmet medical need for a new diagnostic platform for individual DLBCL classification of “cell of origin” phenotyping attempting to design new strategies and more effective disease management.