A Multidisciplinary Model Predicts Clinical Response in Relapsed Multiple Myeloma

The future of cancer treatment lies in personalized strategies designed to specifically recognize, target, and anticipate dynamic tumor subpopulations within an individual in response to drug. Multiple myeloma (MM) is at present an incurable malignancy of bone marrow resident plasma cells with highly variable survival as a consequence of both disease- and host-specific factors. 20% of MM patients, deemed high-risk (HRMM), have shown little benefit in the era of novel agents, with an OS of less than 2 years. Intuitive treatment strategies fail to account for the complexities and evolutionary dynamics of human tumors in the face of drugs. Intuitive treatment fails to adequately account for MM evolutionary dynamics and remains a critical barrier to successful cure or, at least, long-term disease control. Reasons for therapy failure include, but are not limited to, alternation of dominant clones with each line of therapy as a consequence of Darwinian dynamics, genomic instability leading to of tumor heterogeneity, and tumor microenvironment(TME)- mediated drug resistance. We have developed an integrated computational method accounting for phenotypic tumor heterogeneity. This novel ex vivo drug screen approach, termed EMMA (evolutionary mathematical myeloma advisor), predicts patient-specific drug response in silico from fresh bone marrow biopsies within 5 days. This method utilizes longitudinal non-destructive quantification of rate and dose responses of patient-derived MM cells to drugs in an ex vivo 3D reconstruction of the bone marrow microenvironment to provide real-time personalized predictions of treatment success (percent decrease in disease burden at 90 days). The current automated 384-well plate format allows testing of 31 different drugs or combinations against a single patient sample in 5 days. An evolutionary-based computational model uses the drug sensitivity profile obtained ex vivo to detect sub-populations and their contribution to overall clinical drug response. Each drug dose is imaged once every 30 minutes for 96h. This generates 1,920 data points per drug (or combination). From these data we characterize clonal architecture as it relates to drug sensitivity as phenotypic/functional biomarker for each drug or drug combination in each MM patient sample simultaneously. We have examined the predictive accuracy of EMMA in 26 patients to date. The Pearson correlation between ex vivo model predictions and actual tumor burden changes for the 26 patients examined generated the correlation coefficient r=0.87 (P<0.0001). Further, examination of the model predictions in terms of IMWG standards revealed that 23 out of 26 patients showed agreement between model estimation and actual clinical response (88.5% concordance). The remaining 3 patients diverged by one or two stages of response: one patient presented a very good partial response (VGPR, 98.5% reduction) while the model predicted a partial response (PR, 74.5% tumor reduction); the second patient presented a partial response (PR, 74% tumor reduction) while the model predicted a complete response (CR); and the third patient presented stable disease (SD, 12% tumor reduction) and the model predicted a minimal response (MR, 30% tumor reduction). To this end, EMMA generates patient-specific clinical response predictions to individual drugs or regimens with a high degree of clinical accuracy. Beyond testing for clinical drug response, EMMA may also be used to assess dominant cell signaling pathways.  We have screened 5 patients with 25 protein kinase inhibitors (PKI) representing known signaling cascades in MM. Using heatmaps representing area under the curve (AUC) of dose-response surfaces (concentration x exposure time), we have observed both common and patient-specific sensitivities to PKIs. Together, these data demonstrate that the combination of a physiological reconstruction of the TME, a non-destructive and non-invasive cell viability assay, and mathematical models, were key to overcome the major limitations of previous predictive chemosensitivity assays. EMMA has the potential to provide precise clinical insight about treatment efficacy in a timely manner and thus become a decision support tool for oncologists based on the ever-changing clonal architecture in the face of therapy.