Abstract 1753: Prospective stratification using gene expression arrays in a randomised trial of R-CHOP +/− Bortezomib in diffuse large B-cell lymphoma (DLBL): The UK NCRI REMoDL-B study, ISRCTN 51837425

Abstract

Abstract DLBL subtypes are recognised by patterns of gene expression, corresponding to Germinal Center (GCB) or activated peripheral blood (ABC) B-cells. They differ in frequency of mutation in the B-cell receptor signalling pathway and levels of expression of NF-kB target genes. Results of treatment with standard (R-CHOP) therapy are inferior for ABC type, and this study tests the use of bortezomib to downregulate NF-kB and reverse the deficit. This trial is the first to use gene expression profiling (GEP) to stratify cases into GCB and ABC at entry, with adaptive statistical design to analyse the outcome by subtype at predefined timepoints. Patients (pts) newly diagnosed with DLBL undergo staging and commence R-CHOP. During the first 3 week cycle formalin-fixed paraffin-embedded (FFPE) tissue undergoes extraction of messenger RNA for GEP of 24 000 probe sets using Illumina cDNA-mediated Annealing, Selection, extension, and Ligation (DASL®) array at the central laboratory. Cases are allocated to GCB, ABC or other type before the 2nd cycle. Thereafter, all pts are randomized to continue conventional R-CHOP or to add bortezomib 1.3mg/m2 days 1 + 8. The study will randomize a minimum 260 ABC type, to detect a difference in progression free survival (PFS) of 10% with bortezomib, with two-sided significance 5% & 90% power. The design allows closure of randomization for GCB cases if 6 month PFS is <80% after 55 receive R-CHOP-B. A second analysis for futility in GCB is planned after 73 treated with R-CHOP-B are followed for one year: if PFS is <85%, further exploration of bortezomib in this group is not warranted. Prior to trial opening, FFPE material from 200 pts with DLBL was analysed by DASL® protocol. Of 27 genes in a published (LLMPP) classification, 20 are present on the platform. Array data were standardised by converting to z-score. Four decision tree machine learning tools were trained on LLMPP data and combined into a meta-classifier using the Vote scheme with average of probabilities. This showed DLBL reproducibly classified using this platform with outcomes similar to those previously reported. Additional analyses of DNA extracted from tumour material will detect mutations in NF-kB pathway genes by combining the Fluidigm Access Array™ PCR System with next generation sequencing. Since trial opening in 2011, samples have been analysed from 39 pts. 30 (77%) biopsies were successfully classified; 3 cases had poor quality of RNA and in 6 there was insufficient tumour in the block. Cases classified include both surgical and needle core biopsies. Maximum laboratory turnround time was 12 working days. This study demonstrates it is possible to carry out GEP at diagnosis in a multicenter trial. It will address the relationship between response to treatment and molecular biomarkers, and serve as a model for future study design. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 1753. doi:1538-7445.AM2012-1753