Contributions
Abstract: EP1303
Type: e-Poster
Background
Diffuse large B-cell lymphoma (DLBCL) accounts for 40% of all B-cell non-Hodgkin lymphomas and it is a clinically, morphologically and molecular heterogeneous disease. The cell-of-origin (COO) concept was established 20-years ago dividing DLBCL based on gene expression profiles (GEP) in an germinal center B-cell (GCB) and activated B-cell (ABC) type. Subsequently, many GEP studies have been done trying to identify biological features that can help to identify biological and clinical prognostic subgroups. Besides the COO classifier, validation is missing for most published papers.
Aims
To validate reproducibility of five published GEP signatures and whether these GEP signatures have a prognostic impact in an independent DLBCL cohort of uniformly treated DLBCL patients.
Methods
We selected a total of 309 genes to validate five previously published gene expression signatures: Lymph2Cx classifier for COO calling (Scott et al, JCO 2015), MYC activity GEP score (Carey et al, J Mol Diagn 2015), copy number variation (CNV) associated GEP signature (Chan et al, Blood 2015), consensus clustering (Monti et al, Blood 2005) and the immune-ratio signature (Keane et al, Lancet Haematol 2015). We recreated the clustering and/or classification algorithms for each of the selected original publications in 180 samples of patients treated with R-CHOP or intensified rituximab with CHOP in the HOVON 84 trial, using Nanostring gene expression profile, fluorescence in situ hybridization for MYC, BCL2 and BCL6 and CNV results data obtained by low coverage whole genome sequencing. All the analyses were performed in R environment. Results from the HOVON 84 trial have been presented and showed no benefit for the intensification of rituximab or rituximab.
Results
The Lymph2Cx COO classification algorithm was reproducible in the HOVON 84 samples. As previously reported the ABC-type was significantly associated with poorer overall survival (OS) and progression free survival (PFS). We were able to reproduce the MYC activity GEP score in the training set but had a low positive predictive value in the test group. Unlike the original paper the MYC activity GEP score was not associated with survival. The CNV data from the HOVON84 cohort resembled that of the study by Chan et al. In both data sets three genes (REL, GCB; FOXP1, ABC; NFKBIZ, ABC) with recurrent gains were consistently associated with COO subtype. Using a selected number of genes from the RCOR1 signature (30% highest and lowest) we were unable to reproduce the signature. Although the RCOR1 CN deletion impacted RCOR1 gene expression, it was not associated with survival. With the reproduction of the consensus clustering the Host Response cluster was recapitulated in our cohort with good impact in OS, but the BCR signaling and OxPhos clusters could not be reproduced. Finally, we were unable to reproduce the prognostic value of the immune-ratio GEP classifier, as proposed by Ghandi et al.
Conclusion
The reproducibility of most of the published GEP classifiers is limited and the classes as defined in our cohort are not associated to each other. Our results emphasize the need to have good quality validation cohorts when doing GEP analysis in the heterogeneous group of DLBCL.
Session topic: 20. Lymphoma Biology & Translational Research
Keyword(s): Diffuse large B cell lymphoma, Gene expression profile
Abstract: EP1303
Type: e-Poster
Background
Diffuse large B-cell lymphoma (DLBCL) accounts for 40% of all B-cell non-Hodgkin lymphomas and it is a clinically, morphologically and molecular heterogeneous disease. The cell-of-origin (COO) concept was established 20-years ago dividing DLBCL based on gene expression profiles (GEP) in an germinal center B-cell (GCB) and activated B-cell (ABC) type. Subsequently, many GEP studies have been done trying to identify biological features that can help to identify biological and clinical prognostic subgroups. Besides the COO classifier, validation is missing for most published papers.
Aims
To validate reproducibility of five published GEP signatures and whether these GEP signatures have a prognostic impact in an independent DLBCL cohort of uniformly treated DLBCL patients.
Methods
We selected a total of 309 genes to validate five previously published gene expression signatures: Lymph2Cx classifier for COO calling (Scott et al, JCO 2015), MYC activity GEP score (Carey et al, J Mol Diagn 2015), copy number variation (CNV) associated GEP signature (Chan et al, Blood 2015), consensus clustering (Monti et al, Blood 2005) and the immune-ratio signature (Keane et al, Lancet Haematol 2015). We recreated the clustering and/or classification algorithms for each of the selected original publications in 180 samples of patients treated with R-CHOP or intensified rituximab with CHOP in the HOVON 84 trial, using Nanostring gene expression profile, fluorescence in situ hybridization for MYC, BCL2 and BCL6 and CNV results data obtained by low coverage whole genome sequencing. All the analyses were performed in R environment. Results from the HOVON 84 trial have been presented and showed no benefit for the intensification of rituximab or rituximab.
Results
The Lymph2Cx COO classification algorithm was reproducible in the HOVON 84 samples. As previously reported the ABC-type was significantly associated with poorer overall survival (OS) and progression free survival (PFS). We were able to reproduce the MYC activity GEP score in the training set but had a low positive predictive value in the test group. Unlike the original paper the MYC activity GEP score was not associated with survival. The CNV data from the HOVON84 cohort resembled that of the study by Chan et al. In both data sets three genes (REL, GCB; FOXP1, ABC; NFKBIZ, ABC) with recurrent gains were consistently associated with COO subtype. Using a selected number of genes from the RCOR1 signature (30% highest and lowest) we were unable to reproduce the signature. Although the RCOR1 CN deletion impacted RCOR1 gene expression, it was not associated with survival. With the reproduction of the consensus clustering the Host Response cluster was recapitulated in our cohort with good impact in OS, but the BCR signaling and OxPhos clusters could not be reproduced. Finally, we were unable to reproduce the prognostic value of the immune-ratio GEP classifier, as proposed by Ghandi et al.
Conclusion
The reproducibility of most of the published GEP classifiers is limited and the classes as defined in our cohort are not associated to each other. Our results emphasize the need to have good quality validation cohorts when doing GEP analysis in the heterogeneous group of DLBCL.
Session topic: 20. Lymphoma Biology & Translational Research
Keyword(s): Diffuse large B cell lymphoma, Gene expression profile