Identification of Small Molecules That Induce Therapeutic Levels of Fetal Hemoglobin for Treatment of Sickle Cell Disease By Pairing Machine Learning with High-Resolution Single Cell RNA Sequencing Maps of Adult and Fetal Human Erythropoiesis

Induction of fetal hemoglobin (HbF) to treat sickle cell disease (SCD) and beta-thalassemia has been validated as a clinical strategy to ameliorate these diseases. However, despite advances in cell and gene therapy to treat SCD, current approaches still require myeloablation and bone marrow transplantation, which carries risk and limits treatment accessibility. Currently, hydroxy urea (HU) is the only FDA approved small molecule for SCD that acts as a moderate inducer of HbF. Thus, the ability to develop a next generation small molecule with comparable efficacy to gene therapy approaches, but without the existing limitations of HU, would have immense clinical benefit. Towards the goal of identifying novel molecules that can robustly induce HbF, we generated a high-resolution single cell RNA sequencing (scRNAseq) dataset capturing pseudo-trajectories of adult and fetal erythropoiesis. This dataset allowed us to identify cell-states, cell-features, and gene-networks associated with fetal erythropoiesis in a data driven manner. In addition, by applying proprietary algorithms to this dataset, we identified a transcriptional signature targetable with a selected number of small molecules. Evaluation of these predicted small molecules in a 14-day human in vitro erythroid differentiation assay, identified a subset that induced HbF in mobilized peripheral blood (mPB) CD34+ hematopoietic stem and progenitor cells (HSPC) from healthy donors, as measured by % F-cells (flow cytometry) and % HbF (HPLC). CLT-1081 induced HbF (42.3% +17.26, n=4) above HU (17.16% +4.78, n=5) and BCL11A CRISPR knockdown (32.58% +10.66, n=5). We then measured globin gene expression to further characterize how CLT-1081 compared to HU induced changes in fetal and adult globin expression. CLT-1081 induced robust mRNA expression of both HBG1 and HBG2 and concomitant decrease in HBB transcript relative to HU. To determine if CLT-1081 acted by inducing the targeted transcriptional signature, we performed scRNAseq on a number of known HbF inducers, including CLT-1081, HU, and BCL11A CRISPR knockdown. Analysis revealed that CLT-1081 acted as predicted by our algorithms and activated the genes identified as drivers of HbF induction in the human in vitro erythroid differentiation assay. Furthermore, comparison between BCL11A CRISPR knockdown and CLT-1081 demonstrated strong correlation in terms of transcriptional response; CLT-1081 treatment reduced BCL11A expression in addition to other transcriptional changes, which elucidated a partial mechanism of action. In sum, we have used high dimensional data and computational tools in combination with established in vitro cell based assays to identify and validate a cell behavior marked by robust HbF expression and defined by a gene signature identified by our process. These findings demonstrate the power of Cellarity's platform, which augments a systems biology approach with machine learning to identify and develop drug candidates.