RNA Sequencing of Newly Diagnosed Multiple Myeloma to Identify a Predictive Signature for Proteasome Inhibitor-Based Therapy

A major challenge in modern myeloma management is the selection of the best initial treatment from a plethora of available therapies. PAD (bortezomib, doxorubicin, and dexamethasone) is a highly effective regimen which may obviate the need for an upfront high-dose melphalan stem cell transplant (HDSCT). However, it is not clear a priori which patients will benefit from this treatment. In the phase 2 PADIMAC trial, patients with newly diagnosed myeloma were treated with six cycles of PAD and stratified to receive HDSCT in partial remission (PR) or watch and wait in very good partial remission (VGPR) or better. To identify predictive biomarkers, we extracted somatic RNA prior to treatment and performed massively parallel RNA sequencing (RNAseq).

Basic quality control metrics suggested a high-quality RNAseq dataset. Furthermore, spiked expression of IgH fusion partners was consistent with fluorescent in situ hybridization (FISH) data for each sample. All t(4;14) and t(4;16) cases were correctly identified by RNAseq on the basis of the relevant fusion transcript. As a final quality control measure, recurrent variants (e.g. NRAS, KRAS, TP53, FAM46C, DIS3) were identified in the expected proportions.

We performed a pre-planned analysis of differential gene expression between those patients achieving a VGPR or better, sustained for at least 12 months without HDSCT (excellent responders), and those patients who achieved less than a PR (poor responders). 85 genes were upregulated in poor responders and 225 in excellent responders. Interestingly, there was no significant overlap between our PAD-predictive signature and the Arkansas 70 gene poor prognosis signature, the Arkansas 17 gene poor prognosis signature, or the EMC92 prognostic signature. This is likely to reflect in part, the greater sensitivity of RNAseq for gene expression compared to microarrays. However, it also implies that there is a selective signature for PAD responsiveness.

We employed the Gage and Pathview packages to identify non-redundant pathways associated with PAD responsiveness. Significantly upregulated genes in poor responders were those involved in: DNA replication, base excision repair, and the Fanconi anaemia pathway. Genes upregulated in excellent responders included those involved in: several signalling pathways (including RAS, NF-Kappa B, FOXO and JAK-STAT pathways); chemokine and cell adhesion; protein processing in the endoplasmic reticulum; and immune activation pathways such as T-cell receptor signalling and natural killer cell-mediated cytotoxicity.

Finally, we investigated the ability of expression signatures to stratify patients on the basis of PAD responsiveness. Our analysis suggests that RNASeq-defined transcriptional profiles could aid in the selection of patients for PAD therapy (leading to an increase in overall response rate from 70% to 86%), as well as the stratification of patients in CR to a non-HDSCT protocol. We anticipate that our current analysis could easily incorporate specific response signatures for other regimens and could make personalized therapy for myeloma a reality in the foreseeable future.