Single-cell multi-omics uncovers differentiation-dependent transcriptional programs shaping epigenetic subgroups in AML

Abstract Background Acute myeloid leukemia (AML) is a clinically and genetically heterogeneous disease. While traditional classification systems such as FAB relied on morphology and immunophenotype, recent classifications (WHO, ICC) emphasize genetic alterations for diagnosis and therapy. However, more directly involved in the phenotypic cellular inheritance, the epigenomic profile may offer an additional dimension for understanding AML heterogeneity. Previously, we performed ATAC-seq of 1,536 AML samples and identified 16 epigenetically defined AML subgroups (subgroups A-P), which were associated with distinct clinical, genetic, and transcriptional features (Ochi et al., ASH 2024). In the present study, we extend these findings by using single-cell RNA and ATAC-seq (scRNA/ATAC-seq) profiling to explore intra- and inter-tumor epigenetic heterogeneity, transcriptional regulation, and hierarchical differentiation trajectories across AML subgroups. Methods We performed multi-platform scRNA/ATAC-seq on 36 AML samples including all the 16 epigenomic subgroups and 4 remission samples (as controls), profiling a total of 281,167 mononuclear cells. Computational analyses included cell clustering, projection onto normal hematopoiesis reference maps, pseudotime inference, transcription factor (TF) activity analysis using SCENIC+, and leukemic stem cell (LSC) signature scoring. Results scATAC-based clustering showed that leukemic cells from individual AML samples formed distinct clusters largely separated from normal cells, typically co-clustered by subgroup, regardless of their differentiation status, indicating that leukemic cells within each subgroup share a distinct intrinsic epigenetic program. Projection onto a normal hematopoiesis reference map revealed divergent differentiation arrest among subgroups, even within genetically similar cases. For instance, all NPM1-mutant subgroups (D–F) were HOX-related but differed in differentiation states: subgroup E was arrested at the HSC stage, D at the GMP stage, and F contained both progenitors and mature cells. Single-cell analysis also refined differentiation states of subgroups F-H: while bulk ATAC-based deconvolution suggested monocyte enrichment in these subgroups, single-cell analysis revealed mature monocytes predominated in H, whereas F and G were enriched for immature promonocytes. Combined with LSC signature analysis based on gene expression, pseudotime analysis demonstrated that cells showing high LSC scores consistently mapped to early stages of the differentiation trajectory across all subgroups. These results indicate the presence of a conserved leukemic hierarchy, with stem-like cells positioned at the apex, irrespective of epigenetic subgroup, thereby underpinning AML pathogenesis. We further analyzed TF activity using the SCENIC+ program based on scRNA/ATAC-seq data, which identified key TFs enriched in each subgroup, many of which overlapped with those found in bulk RNA/ATAC-seq analysis. Pseudotime analysis further showed that TFs exhibited subgroup-specific activation dynamics. For example, HOXA9 was activated throughout the myeloid differentiation trajectory in HOX-related subgroups but peaked at different stages according to subgroups. In the RUNX1-enriched subgroup, IRF8 and BCL11A were both active but peaked at mature and immature stages, respectively. These findings indicate that key TFs exhibit subgroup-specific activation patterns at distinct stages along the myeloid differentiation trajectory. Conclusion Single-cell multi-omics analysis reveals that AML epigenetic subgroups exhibit distinct but hierarchically organized differentiation trajectories, driven by dynamically regulated TF programs. These findings refine our understanding of AML diversity and may inform more precise, differentiation-stage–specific therapeutic strategies.