Global Phosphoproteome Analysis of AML Bone Marrow Reveals Predictive Markers for the Treatment with AC220

Abstract ⁷⁸⁶

Acute Myeloid Leukemia (AML) results from a combination of oncogenic events that can involve multiple signal transduction pathways including mutation-induced activation of tyrosine kinases. Kinase inhibitors are increasingly studied as promising targeted approaches either alone or in combination with other agents. However, only subsets of patients respond to respective targeted therapies. We hypothesize that the phosphorylation status of certain sets of proteins (phosphosignatures) can predict the clinical response. In recent years, advances in sample processing, mass spectrometry, and computer algorithms for the analyses of proteomics data have enabled the application of mass spectrometry-based proteomics to monitor phosphorylation events in a global and unbiased manner. These methods have become sufficiently sensitive and robust to identify and quantify thousands of phosphorylation sites in a single experiment. Furthermore, spiking-in a SILAC- (stable-isotope labelling by amino acids in cell culture) reference sample (Super-SILAC) allows for the precise quantification of phosphorylation events also in in-vivo samples. Internal tandem duplication (ITD) of FLT3 is one of the most common mutations in AML. It causes constitutive activation of FLT3. AC220 (Quizartinib®, Ambit) is a selective inhibitor of the receptor-type tyrosine-protein kinase FLT3 that is currently under development for the treatment of AML. In a recent phase II open-label study, patients with relapsed AML were treated with AC220. The results of this trial will be reported elsewhere. Here, we sought to identify phosphorylation events that predict clinical response with high accuracy, especially in the group of FLT3-ITD-positive patients. To this end, we established a global, quantitative phosphoproteome analysis of AML cells derived from patient bone marrow. We analyzed samples from 15 patients before treatment with AC220 and employed the obtained phosphoproteomics data to discover a predictive signature of phosphorylation markers. In total, we confidently identified and quantified roughly 10,000 phosphorylation sites. Unsupervised clustering of the data revealed two large clusters that are both characterized by stronger phosphorylation of associated protein kinases. Next, we correlated these clusters with the reported clinical response of the respective patients. Although the clustering was unsupervised regarding the response information, the two clusters almost perfectly separated responders (complete or partial remission) from non-responders (stable or progressive disease). Furthermore, we identified a signature comprising four phosphorylation sites that accurately predict the outcome of therapy based on the corresponding phosphorylations obtained before treatment. We evaluated the performance of this signature in a cross-validation set-up and in additional test samples that have not been used for training. Moreover, we validated the robustness of the selected predictive features. Our study supports that the phosphorylation of four proteins are candidate biomarkers for predicting response to AC220 in AML. Notably, the phosphorylation markers are more predictive than the FTL3-ITD status alone and will thus allow a more precise stratification of AML patients for treatment with AC220. In addition, our results demonstrate for the first time that identifying predictive phosphorylation signatures directly from clinical patient samples is possible.