Identification of small molecules that selectively target JAK2V617F driven cytokine-independent megakaryopoiesis by leveraging single cell RNA sequencing maps of myelofibrosis patients samples and a deep learning framework

Abstract In myeloproliferative neoplasms including myelofibrosis (MF), the accumulation of driver mutations like JAK2V617F in the hematopoietic stem cell (HSC) compartment results in constitutive activation of JAK2 signaling resulting in dysregulated hematopoiesis. MF is characterized by bone marrow fibrosis and inflammation, resulting in splenomegaly, anemia, and eventual bone marrow failure. While allogenic bone marrow transplantation remains the only curative treatment, it is not a viable option for most patients due to therapy related morbidity and the challenges associated with managing older patients with cytopenia and bone marrow fibrosis post-transplantation. Currently, JAK2 inhibitors like ruxolitinib manage disease symptoms, but often cause dose-limiting cytopenias due to activity against both wildtype and mutant JAK2. Based on the lack of disease-modifying therapies that target the mutant hematopoietic stem cell, we leveraged high dimensional data to identify gene networks that are selectively induced in the JAK2V617Fmutant clone. We built a transcriptomic disease map encompassing over 150,000 cells from 50 MF patients, capturing the hematopoietic stem and progenitor compartment (HSPC) and deployed a deep learning framework leveraging transcriptomics to identify and target features unique to the mutant HSPCs. In parallel, we developed an iPSC model harboring the JAK2V617F mutation and generated iHSC to faithfully recapitulate the disease biology observed in patient samples. Single cell RNA sequencing of the mutant iHSCs confirmed reproduction of the clinical disease signatures observed in patient cells across the hematopoietic compartment, including HSPC, megakaryocyte erythroid progenitor (MEP), and megakaryocyte progenitors (MKP). Furthermore, JAK2V617F iHSCs recapitulated the MF phenotype of cytokine-independent megakaryopoiesis as demonstrated by the differentiation of CD42+ megakaryocyte progenitors in the absence of thrombopoietin (TPO). Consistent with TPO-independent megakaryopoiesis, we demonstrated that platelet factor 4 (PF4) is a biomarker that reliably measures aberrant megakaryopoesis in vitro. We then leveraged our proprietary machine learning (ML) guided drug discovery platform to predict small molecules that target the disease signature with the aim to reverse the disease phenotype by targeting the stem cell compartment. Our ML-guided screen identified small molecules that efficiently decreased both PF4 levels and the number of CD42+ MKPs in our iHSC JAK2V617Fdisease model. Biochemical binding assays and single cell transcriptomics demonstrated that our lead molecules did not bind JAK2 protein and had different transcriptional signatures in JAK2V617F mutant cells compared to the JAK2 inhibitor ruxolitinib. Erythroid differentiation studies using healthy human CD34+ showed that our lead molecule was more than 10-fold less potent than ruxolitinib at inhibiting normal erythropoiesis demonstrating the selectivity of our hit molecules in targeting the mutant HSPCs. Finally, we demonstrated dose-dependent inhibition of TPO-independent megakaryopoieis in primary MF CD34+ samples with our lead compound. We have leveraged single cell transcriptomics, ML and iPSC models to identify small molecules that can selectively target aberrant JAK2 mutant HSPCs by using transcriptomics to target distinct pathways downstream of hyperactive JAK2 signaling. Moreover, our strategy decouples mutant JAK2 signaling sparing normal hematopoiesis and reducing the anemia risk seen with current clinical approaches.