3074 – SINGLE CELL TRANSCRIPTOMIC ANALYSIS COMBINED WITH MACHINE LEARNING IDENTIFIES HSC-LIKE CELLS DIFFERENTIATED IN VITRO FROM HUMAN IPSCS.

Abstract

During embryonic development, haematopoietic stem and progenitor cells (HSPCs) develop in different waves at various anatomical sites. This process can be recapitulated in vitro by cytokine-mediated differentiation of human pluripotent stem cells (hPSCs) although the differentiation of haematopoietic stem cells (HSCs) remains a major challenge in the field. To depict the developmental dynamics and to study the heterogeneity of HSPCs that can be generated in vitro from hPSCs, we employed single cell RNA sequencing (scRNAseq) and cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq). We identified the transcriptome and functionally validated naïve progenitors as well as erythroid, megakaryocyte and leukocyte-committed progenitors and we associated to these the expression of CD44, CD326, ICAM2/CD9 and CD18, respectively. To identify and compare the in vitro-generated cell types to those developed in vivo, we employed a machine learning approach. We trained an artificial neural network (ANN) using a human fetal liver scRNAseq dataset of hematopoietic cells from HSCs to mature cell types. We identified a wide range of hPSCs-derived phenotypes, including a small group classified as HSC/MPP. This transient HSC-like population decreased with the progression of the differentiation protocol and was not identified by the ANN in the datasets that we generated from sorted CD43+ cells. By comparing the transcriptome of these in vitro-generated HSC/MPP-like cells with those within the fetal liver we identified transcription factors and molecular pathways that can be exploited to improve the in vitro production of HSCs in the future. Finally, when these HSC-like cells were compared to yolk sac and AGM cells we observed that in vitro derived progenitors displayed signatures of high oxidative respiration, suggesting an overall divergent metabolic profile.