2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Saturday, December 12, 2026
05:30 p.m. - 07:30 p.m.
MCCNO - Hall F
Technology and Procedures, Assays, Imaging, Omics Technologies, Bioinformatics, Cell Expansion, Computational Biology, Biological Processes, Genomics, Hematopoiesis, Immune Mechanism, Metabolism, Molecular Biology, Diseases, Myeloid Malignancies, Chronic Myeloid Malignancies, Research, Fundamental Science, Translational Research, Study Population, Human, Animal Model
PAPER

1356 Single-cell behavioromics enables fate prediction of individual hematopoietic stem cells

Alisa Chankhunthod, Rhys Sorenson-Graff2, Juliane Mercoli3, Takashi Ishida4, Stacey Dozono5, Youyi Fong6, Brandon Hadland, MD, PhD7, Helong Gary Zhao, PhD8
1 Fred Hutch Cancer Center, Seattle, Washington, United States
2 Fred Hutch Cancer Center, Seattle, Washington, United States
3 Fred Hutch Cancer Center, Seattle, Washington, United States
4 Fred Hutch Cancer Center, Seattle, Washington, United States
5 Fred Hutch Cancer Center, Seattle, Washington, United States
6 Fred Hutch Cancer Center, Seattle, Washington, United States
7 Fred Hutch Cancer Center, Seattle, Washington, United States
8 Fred Hutch Cancer Center, Seattle, Washington, United States

Prospective identification of live long-term engrafting hematopoietic stem cells (LT-HSCs) remains a significant unmet need in transplantation therapy and stem cell research. Although immunophenotyping can enrich for live HSCs, it does not identify LT-HSCs with single-cell accuracy. Single-cell transcriptomics offers high-resolution characterization of hematopoietic cells, but it neither approaches 100% accuracy in LT-HSC identification nor preserves cell viability. Such predictive inaccuracy likely originates from the mismatch between the static, point-in-time nature of current single-cell phenotyping methods and the dynamic nature of cell fate decision.  

We set out to overcome this barrier by studying the dynamic behaviors of individual hematopoietic stem and progenitor cells (HSPCs) in ex vivo niche cultures using live imaging and deep learning (DL), coupled with downstream phenotyping and ground-truth assays of long-term engraftment potential. Ex vivo niche cultures were established by index sorting immunophenotypic HSCs (CD45+GR1⁻F4/80−SCA1highEPCRhigh) from C57BL/6J mouse fetal liver (FL) to monolayers of FL endothelial cells (FL-ECs) cultured in serum-free X-VIVO 20 media with murine stem cell factor and thrombopoietin, at precisely one cell per microwell. Single HSPCs were live imaged for 72 hours in co-culture, then expanded for an additional 11 days before being subdivided for flow cytometric phenotyping, single-cell transcriptomic profiling, and assessment of engraftment potential by serial transplantation into irradiated isogenic mice. By cross-referencing end-of-expansion immunophenotypes with engraftment outcomes, we established that SCA1highEPCRhighESAM+ immunophenotype after the 14-day co-culture accurately correlated with multi-lineage serial engraftment.

Leveraging this platform, we further show that the fate of individual HSPCs can be predicted by distinctive features in their single-cell behaviorome (the totality of observed behaviors). Specifically, DL analysis of the high-dimensional totality of observable behaviors (such as motility, divisions, and morphological changes) captured by live imaging during co-culture with FL-ECs distinguished individual LT-HSCs from transiently engrafting or non-engrafting HSPCs with average accuracy of 98.5%, outperforming immunophenotyping and single-cell transcriptomics. When stratified by DL model’s latent embeddings of single-cell behavioromes, LT-HSCs formed a discrete cluster in UMAP space, separated from clusters of other HSPCs. Comparison of behavioromic features among clusters revealed that LT-HSCs exhibit reduced stochastic locomotion on FL-ECs compared to other HSPCs. These observations indicate that individual HSCs with long-term engraftment potential exhibit distinct behavioral patterns during the initial 72 hours of interaction with niche FL-ECs.

We next examined the molecular mechanisms governing this behavior-fate correlation. Integrating single-cell transcriptomic analyses of both HSPCs and FL-ECs demonstrated that the higher motilities of short-term engrafting HSCs and progenitors, compared to LT-HSCs, were associated with heightened niche-derived Activin signaling competing against TGF-beta signaling (inferred from cell-cell communication algorithms NicheNet and CellPhoneDB, and pathway activation scores) and reduced expression of cell adhesion genes downstream of canonical TGF-beta pathway. Correspondingly, antagonizing Activin signaling with recombinant Inhibin-A protein in our coculture system reduced the locomotion of HSPCs and increased the proportion of LT-HSCs amongst immunophenotypic FL CD45+GR1⁻F4/80−SCA1highEPCRhigh cells, thereby linking niche-derived signals to behavior-predictive fates.

Collectively, our studies established a method of projecting deep learning-extracted dynamic HSPC behaviors to specific signal pathways that also shape stem cell fates. The high-precision linkage between single-cell behaviorome and engraftment potential provides an opportunity to extend this approach to quantifying functional LT-HSCs in therapeutic transplant products and to tuning conditions for optimizing LT-HSC output from human pluripotent stem cell platforms. More broadly, by enabling prospective fate prediction at single-cell resolution, this behavioromic analysis framework has the potential to advance stem cell research and regenerative medicine across diverse developmental and tissue contexts.

Helong Gary Zhao, PhD: No relevant conflicts of interest to declare.