Transcriptional landscape of direct reprogramming toward hematopoietic stem cells

Direct reprogramming of human fibroblasts into hematopoietic stem cells (HSCs) offers a promising strategy for generating autologous cells to treat blood and immune disorders. However, low conversion efficiency and incomplete characterization of reprogrammed states limit current protocols. To address this, we assembled a single-cell transcriptomic reference atlas spanning hematopoietic, endothelial, and fibroblast cell types, and evaluated a novel algorithmically predicted transcription factor (TF) recipe—GATA2, GFI1B, FOS, REL, and STAT5A—for induction of HSC-like states. The 5TF recipe increased conversion efficiency of fibroblasts to CD34+ cells by approximately three-fold compared to previous protocols. Long-read single-cell RNA sequencing revealed heterogeneous expression of hematopoietic- and endothelial-associated genes, consistent with partial reprogramming. To benchmark re- programmed cells, we developed an approach to map their transcriptomic positions relative to native and initial cell states within the reference atlas. Finally, we cataloged isoform diversity in reprogrammed cells, suggesting a role for alternative splicing in reprogramming dynamics. These findings highlight both the promise and complexity of direct HSC reprogramming and provide a framework for optimizing future protocols. Overall design: Direct reprogramming of human fibroblasts into hematopoietic stem cells (HSCs) was investigated using a defined set of five transcription factors (TFs): GATA2, GFI1B, FOS, STAT5A, and REL. TFs were selected using a data-guided control algorithm that identified combinations minimizing transcriptional differences between fibroblasts and HSCs. TFs were delivered via lentiviral vectors encoding doxycycline-inducible expression cassettes, and expression was induced with doxycycline for 48 hours. Cells used in control and reprogramming conditions were maintained between passages 2 and 4 at initiation, and all samples were collected before passage 10. Single-cell RNA sequencing was performed using the 10x Genomics Chromium platform, followed by long-read sequencing using Oxford Nanopore Technologies (ONT) platforms. Three libraries were generated: (1) human neonatal foreskin fibroblasts (ATCC CRL-2522), (2) cells reprogrammed with the 5TF cocktail, (3) primary human bone marrow (ATCC PCS-800-012TM, Lot 80516705). Raw sequencing data were processed using the EPI2ME Labs 'wf-single-cell' workflow with the GRCh38 reference genome (version 2024-A).