Background: Midostaurin is approved for FLT3 mutant-positive (FLT3+) acute myeloid leukemia (AML), however efficacy has also been observed in a subpopulation of FLT3 mutant-negative AML, suggesting that FLT3 mutation is not the only determinant in conferring midostaurin sensitivity. We previously described a phosphoproteomic signature significantly elevated in primary AML blasts that responded to midostaurin ex vivo (Casado et al Leukaemia 2018). This signature includes phosphorylation sites on protein kinase C delta (a midostaurin off-target) and its substrate GSK3A. In this study, we tested whether these phospho-signatures could group FLT3+ patients based on clinical responses to midostaurin plus chemotherapy. Methods: We obtained FLT3+ bone marrow (BM) and peripheral blood (PB) diagnosis specimens (n=56 cases) from the Leukemia Tissue Bank at Princess Margaret Cancer Centre. These patients were treated with standard chemotherapy plus midostaurin. Phospho-signatures quantified using mass spectrometry were analysed with a classification machine learning algorithm to group patients based on response to treatment as a function of phospho-signature status. Other features (e.g. genetic mutations, HSC-transplant) were also analysed. Differential survival analysis was carried out with Kaplan-Meier and Log Rank test methods. Phospho-signatures for BM and PB samples were analysed independently. Results: A first ML model was developed based on the signature described in the Casado et al study. Patients positive for this signature exhibited a survival probability of 243 weeks, compared to 126 weeks in signature negative patients (averages by geometric mean, Log Rank p = 9.88e-05). As the patients in the current study received chemotherapy, in addition to midostaurin, we also identified a new signature consisting of 26 phospho-sites (model 2), which partially overlapped with the first model. Patients positive for model 2 signature showed a markedly longer survival time than negative patients (269 vs 76 weeks, Log Rank p = 1.30e-05 for PB and 241 vs 56, Log Rank p = 2.13e-09 for BM specimens, Table). No other features separated survival as clearly as model 2. Conclusions: We have identified phospho-signatures with the potential to further stratify FLT3+ AML for midostaurin treatment. The presence of PRKCD signalling components in signatures provides a rationale for midostaurin activity in sensitive cases. Analysis will also be performed on FLT3 mutant-negative cases to validate the signature in this group.

Disease free survival in weeks as a function of model 2 phospho-signature. * Kaplan-Meier curves did not reach 50% reduction in disease free survival.