Microenvironmental Signatures Reveal Biological Subtypes of Diffuse Large B-Cell Lymphoma (DLBCL) Distinct from Tumor Cell Molecular Profiling

Research in DLBCL pathogenesis has largely focused on the lymphoma cells that defined molecular subtypes. To elucidate the role of the lymphoma microenvironment (LME) in this process, we developed and deconvoluted transcriptomics signatures of LME cells and pathways from 3,026 DLBCLs from 13 datasets including our new cohort of 127 pts. Mutations were available for 562 pts of the datasets and for 22 pts from our cohort (WES with matched normal). Applying density-based clustering we identified 4 LME signatures, independent of reported transcriptional and genetic classifications based on lymphoma cells:

LME-1 DLBCLs (n=726, 24% - GCB/ABC: 35%/36%) were characterized by an "immunosuppressive" ME enriched for Tregs, myeloid-derived suppressor cells, CD8^PD1high, natural killer and macrophages type 2 and prevalence of genetic mechanisms of immune escape in malignant cells such as mutations in B2M (33%) and CD70 (10%). Malignant cells in LME-1 DLBCL presented high activity of NF-kB and JAK/STAT signaling pathways, likely due to high frequency of co-occurring MYD88^L265and CD79B mutations (40%) and the presence of a cytokine rich milieu including high expression of IL10, IL6 and TNFS13B. Good outcome genetic groups BN2 and EZB constituted 46% of pts

LME-2 DLBCLs (n=484, 16% - GCB/ABC: 55%/30%) were characterized by an "anti-tumor immunity" ME enriched for T cells, follicular T_H and follicular dendritic cells (FDC). Lymphoma cells presented the highest number of BCL2 translocations and EZH2 mutations (40%, p=0.001 vs. other LME groups) and activation of cell motility and chemotaxis pathways that likely account for the advance stage at presentation (70% were stages III/IV p=0.02 vs. 40-50% in the other LME groups). Lymphoma cells expressed higher levels of CCL20, CCR6 and CXCR5. Good outcome genetic groups BN2 and EZB constituted 74% of pts. Notably, ABC LME-2 DLBCLs had better prognosis that any other ABC DLBCL (p<0.007 n=848) and similar to GCB DLBCLs

LME-3 DLBCLs (n=847, 28% - GCB/ABC: 56%/25%) were characterized by a "mesenchymal" ME enriched for cancer-associated fibroblasts (CAFs), reticular DC, FDC and endothelial cells. Lymphoma cells showed higher mutations in BCR/Pi3K signaling intermediates SGK1 and GNA13 (20 and 13%) and activation of the TGFB signaling and matrix remodeling pathways. LME-3 DLBCLs expressed higher levels of MMP9, MMP2, TIMP1 and TIMP2. Good outcome genetic groups BN2 and EZB constituted 75% of pts, and were no patients harboring NOTCH1 mutations. LME-3 DLBCL also had the higher proportion of non-cellular LME component represented by the extracellular matrix (ECM). Patients with DLBCL with higher ECM proportion had better progression free survival (p=0.0004). We specifically tested the ECM effect in a murine DLBCL model by analyzing changes in ECM proteins associated with disease progression by serial proteomics. DLBCL progression was accompanied by increase in lymphoma cells in detriment of CAFs and ECM proteins. Among them, we identified the small proteoglycan decorin (DCN). Accordingly, parental administration of recombinant DCN to DLBCL mice (n=10 vs vehicle n=10) decreased tumor volume (p<0.001) and lymphoma cell proliferation while improving the LME immune infiltrate

LME-4 DLBCLs (n=969, 32% - GCB/ABC: 46%/38%) were characterized by a "depleted" ME with increased proportion of lymphoma cells with mutations in MYD88, PIM1 and HLA-C (36, 38 and 13%), higher genomic instability and epigenetic heterogeneity (by DNA methylation). Good outcome genetic groups BN2 and EZB constituted 68% of pts. Lymphoma cells showed activation of Pi3K signaling and hypermethylation and low expression of the TGFB mediator SMAD1 (p<0.01 vs other LME groups). In agreement with our demonstration that pharmacologically reversible SMAD1 hypermethylation drives chemoresistance in DLBCL (Cancer Discovery 2013), patients with LME-4 DLBCL had less favorable outcomes in overall survival (HR=1.65 P<0.005 n=1,980) and response to immunochemotherapy (p<0.0001 n=810) than any other LME groups; likely reflecting the acquisition of ME-autonomous survival mechanisms by lymphoma cells. The bad outcome of LME-4 DLBCLs was not dependent on MYC/BCL2 or BCL6 double translocations, or the overall mutational load.

In sum, ME signatures in DLBCL associate with clinical outcomes independently of existing molecular subtypes, contribute to explain DLBCL biology and provide potential novel therapies