Reproducibility of Gene Expression Signatures in Diffuse Large B-Cell Lymphoma

SIMPLE SUMMARY: Multiple gene expression signatures with biological or prognostic subgroups have been published in diffuse large B-cell lymphoma (DLBCL). With exception of the cell of origin (COO) classifier, these were not validated in independent cohorts. The aim of the study was to reproduce four gene expression signatures capturing multiple biological subgroups using the NanoString platform. In addition, we aimed to identify potential associations between the signatures and portray the heterogeneity of DLBCL. We show that, in an independent cohort of clinically well-defined patients, these signatures can co-occur in the same patient and that each classifier captures a different aspect of the biological heterogenous panorama of DLBCL. Beside COO, there is clear evidence of different immune and MYC signatures. A direct comparison in our cohort showed that these signatures reflect independent biological features. More comparative studies with gene expression profiles need to be conducted to enable a further integration and to help develop new taxonomy systems for clinical utility. ABSTRACT: Multiple gene expression profiles have been identified in diffuse large B-cell lymphoma (DLBCL). Besides the cell of origin (COO) classifier, no signatures have been reproduced in independent studies or evaluated for capturing distinct aspects of DLBCL biology. We reproduced 4 signatures in 175 samples of the HOVON-84 trial on a panel of 117 genes using the NanoString platform. The four gene signatures capture the COO, MYC activity, B-cell receptor signaling, oxidative phosphorylation, and immune response. Performance of our classification algorithms were confirmed in the original datasets. We were able to validate three of the four GEP signatures. The COO algorithm resulted in 94 (54%) germinal center B-cell (GCB) type, 58 (33%) activated B-cell (ABC) type, and 23 (13%) unclassified cases. The MYC-classifier revealed 77 cases with a high MYC-activity score (44%) and this MYC-high signature was observed more frequently in ABC as compared to GCB DLBCL (68% vs. 32%, p < 0.00001). The host response (HR) signature of the consensus clustering was present in 55 (31%) patients, while the B-cell receptor signaling, and oxidative phosphorylation clusters could not be reproduced. The overlap of COO, consensus cluster and MYC activity score differentiated six gene expression clusters: GCB/MYC-high (12%), GCB/HR (16%), GCB/non-HR (27%), COO-Unclassified (13%), ABC/MYC-high (25%), and ABC/MYC-low (7%). In conclusion, the three validated signatures identify distinct subgroups based on different aspects of DLBCL biology, emphasizing that each classifier captures distinct molecular profiles.