Abstract
Abstract
Background
Following the positive outcome of the RATIFY phase 3 clinical trial, the multi-kinase inhibitor midostaurin was approved for the treatment of adult patients with newly diagnosed FLT3-mutated acute myeloid leukemia (AML). However, we and others have observed that single agent midostaurin yields responses also in a substantial portion of patients not carrying FLT3 mutations. The molecular basis and the kinase targets mediating these responses are poorly understood and no biomarkers predictive of response in FLT3 wildtype (wt) AML patients exist. To identify markers distinguishing the FLT3 wt responding subset of patients, we trained machine learning multi-marker models using AML patient baseline transcriptomic and mutational data to predict ex vivo responders vs. non-responders. Further, to better understand the molecular basis of midostaurin responses and to explore the unique signaling networks modulated by midostaurin, we profiled the sensitivities of AML patient samples to midostaurin in comparison to, and in combination with, several clinically relevant oncological targeted agents of diverse mechanistic classes.
Results
Midostaurin target space is unique and it retains anti-leukemic potency under cytoprotective conditions. We have previously established that single agent midostaurin is effective ex vivo in about 25% of FLT3 wt AML patient samples and retains potency in a cytoprotective medium that masks the effects of more selective FLT3 inhibitors such as quizartinib, crenolanib and sorafenib (Karjalainen et al, Blood 2017). To further investigate the unique pathways that midostaurin, but not other FLT3 inhibitors targets, we correlated the response patterns of 87 AML patient samples in cytoprotective medium to midostaurin and 261 other kinase inhibitors in our oncology compound collection. In unsupervised cluster analysis, midostaurin showed highly similar response patterns to AZD7762, OTS167, milciclib, pacritinib, ENMD-2076 and fostamatinib. Publicly available in vitro kinase profiling (Tang et al, Cell Chem. Biol. 2018) suggested that midostaurin does not inhibit most of the primary targets of these other inhibitors, with only aurora kinases, JAK kinases and SYK appearing to be shared potent targets.
Midostaurin anti-leukemic potency is determined by the mutational background. Several multi-marker, supervised machine learning models were compared to extract biomarker signatures from either baseline transcriptomic or mutational data, in the task of predicting ex vivo midostaurin response in samples cultured in cytoprotective medium. In the full cohort (N=81), the presence of FLT3 mutations (both internal tandem repeat and tyrosine kinase domain mutations) was the strongest predictor of response. In the FLT3 wt cases (N=49), our results revealed that other select mutations correlated well with either response or non-response upon Bayesian Linear Regression analysis with cross-validation (Ammad-Ud-Din et al, Bioinformatics, 2017). Mutations in PTPN11, U2AF1, SRSF2, RUNX1, JAK2 and BCOR predicted midostaurin responders, while mutations in GATA2, WT1, NPM1 and IDH2 were enriched in non-responders (Figure 1). Baseline transcriptomic profiles, however, did not provide added value for the predictive power.
Midostaurin efficacy can be enhanced by combination with other targeted agents. Combinatorial drug screening of midostaurin in cytoprotective medium revealed several synergizing drug classes, including BCL-2 and MDM2-p53 inhibitors. Further analysis of synergizing agents in broader AML patient sample cohorts is ongoing.
Conclusions
Our results show that midostaurin may reach its biological effects through inhibition of additional kinases than just FLT3. In both FLT3 mutant and wt cases, midostaurin responses are influenced by the overall mutational background. Furthermore, our data indicates that midostaurin efficacy can be enhanced through combination with other agents. Together, we have significantly expanded the understanding of molecular determinants of midostaurin response in primary AML cells, supporting predictive biomarker discovery efforts and development of synergistic drug combinations. The emerging hypotheses from this work will have to be tested in clinical studies.
Disclosures
Porkka: Novartis: Honoraria, Research Funding; Celgene: Honoraria, Research Funding. Marques Ramos:Novartis: Employment. Pallaud:Novartis: Employment. Aittokallio:Novartis: Research Funding. Wennerberg:Novartis: Research Funding.