A Gene Expression-Based Risk Stratification Model Developed in Newly Diagnosed Multiple Myeloma Treated with High Dose Therapy Is Predictive of Outcome in Relapsed Disease Treated with Single Agent Bortezomib.

Using data derived from the U133Plus2.0 microarray (U2), we recently constructed a 17-gene model predictive of high-risk multiple myeloma (MM). In the model, 13% of newly diagnosed cases were considered to have high-risk MM with 24 month overall survival estimates of 50% and 90% in the high- and low-risk groups, respectively (p<0.0001). Although validated in a separate cohort of newly diagnosed cases also receiving high dose therapy, it is not known whether this model has broader utility. To address this issue we applied the model to a dataset from a pharmacogenomic effort as part of a multi-center phase III trial testing the efficacy of the proteasome inhibitor bortezomib compared to dexamethasone. For purposes of evaluating the robustness of our model, this dataset is particularly useful in that it differs from the UAMS dataset in several important areas, such as relapsed vs. newly diagnosed disease, single agent bortezomib or dexamethasone vs. multi-agent induction therapy and high-dose melphalan-based tandem transplants, multiple centers vs. single center, delayed processing of bone marrow aspirates vs. immediate processing, negative vs. positive selection of plasma cells, and microarray platform U133A/B (UA) vs. U2 in the Millennium and UAMS datasets, respectively. A total of 144 of the first 351 UAMS cases applied to the U2 had also been applied to UA and were analyzed together with the 188 Millennium samples. Of the 17 genes on U2 we were able to find exact matches for 16 genes on UA. We applied the UA signal intensities of the 16 genes and the exact multivariate stepwise discriminant analysis (MSDA) model used to derive the 17-gene model using U2 data to the 144 UAMS cases. Risk scores for the 144 UAMS cases using the 17-gene U2- and 16-gene UA-derived models were calculated and a strong correlation observed (r=0.89; P<0.001). Using a score of greater than 1.6 as the high-risk cut-point with both models, a confusion matrix revealed a 96.5% agreement between high- and low-risk annotations among the 144 cases. Kaplan-Meier survival analysis of high- and low-risk groups defined by the 17- and 16-gene models revealed similar survival differences between high- and low-risk MM (U2: P=0.0046, HR=2.49; UA: P=0.0026; HR=2.58). These data suggests that the risk model is highly stable across these platforms. We then applied the 16-gene model to 188 cases of relapsed disease treated with bortezomib. Using the same discriminant score cut-off of 1.6 on the Millennium data, the 16-gene model defined 17.6% of cases as high-risk and this subset had significantly shorter overall survival times (P<0.0001; HR=2.93). The estimated 24-month survival was 9% in the high-risk vs. 40% in the low-risk cohort. The model also identified high-risk disease in a dexamethasone treated cohort (P<0.0001; HR=3.01). These data suggest that the 17-gene high-risk model is robust and not confounded by variables distinguishing these two datasets. These data also reveal a common molecular signature of high-risk MM in patients with newly diagnosed and relapsed disease pointing to a potential common molecular mechanism of drug resistance in MM that is revealed in this signature. We are currently developing and validating the 17-gene model using technology that quantifies tumor RNA directly in purified cells.