Predicting Event-Free and Overall Survival after Treatment of Myeloma with the Proteasome Inhibitor Bortezomib with Pre-Treatment and 48-Hour Post-Therapy Gene Expression Patterns.

A major goal of current high-throughput molecular techniques, such as gene expression profiling (GEP), is to identify biomarkers that can act as robust and reproducible predictors of therapeutic efficacy. Recent studies demonstrated the potent anti-myeloma activity of the proteasome inhibitor Bortezomib. We sought to determine if the GEP either prior to therapy initiation and/or 48 hours after therapy initiation correlated with event-free (EFS) or overall survival (OS) in pts treated with Bortezomib (UARK 2001-37). A total of 37 pts had plasma cells (PCs)prepared for profiling at baseline and had clinical outcome information available. Of these, 25 had an event and 16 had died at the time of this study. Of the 37 patients, 20 also had PCs prepared for profiling 48 hours after therapy initiation. Of these 20, 14 had an event and 9 had died at the time of this study. RNA was isolated from PCs, labeled, and hybridized to the U95Av2 microarray containing ~12,000 genes. We used the average difference call without transformation. Cox regression analysis was used to evaluate the separate effect of each gene on EFS and OS by looking at 1) the baseline (n=37) and 2) the difference between baseline and 48 hours after therapy initiation (n=20). Without correcting for multiple comparisons, 1183 of the genes measured at baseline predicted survival in univariate anises (P <0.05). A total of 466 of the genes that exhibited a difference between baseline and 48 hours predicted survival. A total of 1817 of the genes measured at baseline predicted EFS (P <0.05) and 582 genes exhibiting a change in expression after therapy initiation predicted EFS (P <0.05). We then asked whether both baseline GEP and change in GEP simultaneously predicted OS and EFS. In particular, did over expression of a gene at baseline predict poor survival, but subsequent decrease in GEP actually lead to improved survival in the same model? Consequently, for OS and EFS we fit baseline and changes in GEP simultaneously for the 20 pts. For 166 genes, the baseline GEP and the changes in GEP were both statistically significant (P <0.05 for both variables) in the same model of OS and 263 genes were significant in the EFS model. We then ranked the most predictive models where there was an indication of the baseline going in one direction and the change in expression in the opposite direction. For OS, over-expression of 10 genes led to poor survival and a change in their expression led to improvement. We identified another 10 genes whose under expression at baseline led to poor survival and a change or increase led to an improvement. A similar analysis was performed for EFS. Some of these genes are important regulators of cell cycle, such as CDC2, TYMS, BUBI, TOP2A and KI-67. These genes were over-expressed at baseline but down-regulated in the 48 hrs follow-up after treatment in the EFS analysis suggesting that this process could be associated with improvement in outcome. In conclusion, the number of observations is too small to control for other factors. Especially for EFS, some of the identified genes could be affected by treatment and thus show a large effect on survival at baseline but improvement after 48 hours leads to better EFS. The next step will be to link the results to the treatment received, e.g. dose and schedule of the drug.