Abstract
Acute myeloid leukemia (AML) is driven by diverse genetic alterations in hematopoietic precursor cells and has a high rate of relapse. Among the most common changes are mutations in NPM1, encoding nucleophosmin. Recent work suggests that mutant NPM1 sustains leukemogenic transcription of HOXA/B genes and MEIS1 by coordinating with RNA Polymerase II and the MLL/menin complex as well as by inhibiting histone deacetylases. It is hence critical to better understand the epigenetic mechanisms governing leukemogenesis and treatment resistance. Here, we sought to comprehensively characterize the gene regulatory landscape of NPM1-mutantAML at diagnosis and relapse by analyzing the chromatin accessibility, gene expression, and surface protein abundance of AML cells at single-cell resolution – together with NPM1 mutational status.
We developed GoT-DOGMA-seq, a single-cell method integrating Genotyping of Transcriptomes (Nam et al., 2019)for NPM1 genotyping with DOGMA-seq (Mimitou et al., 2021), which captures chromatin accessibility, gene expression, and surface protein abundance information from the same individual cells. We applied this method to paired bone marrow mononuclear cells (BMMCs) from three patients at diagnosis of AML (n = 16,793 cells) and after relapse following chemotherapy or venetoclax (n = 22,191 cells) and compared to BMMCs from healthy controls (n = 33,375 cells across 7 individuals).
Malignant cells, identified through NPM1 genotyping, immunophenotyping, copy number alterations, and assessment of AML gene expression signatures, exhibited pronounced heterogeneity at diagnosis and relapse, reflecting varied myeloid cell types. Cell states characterized by chromatin accessibility and gene expression were correlated, but there were distinctions. For example, HSC/MPP-like malignant cells subdivided into states related to differential stemness, myeloid chromatin priming, and transcription factor (TF) activity, as well as copy number alterations (e.g., chr17p loss), inflammatory signaling (e.g., TNF-alpha and IL-1 signaling), and mitochondrial energetics (e.g., oxidative phosphorylation).
Overall, chromatin accessibility data revealed 167,959 candidatecis-regulatory elements (cCREs). These cCREs covered 88% of ENCODE elements, suggesting consistency with the database as well as recognition of novel cCREs. 98,698 cCREs (59%) were uniquely accessible in HSC/MPP-like malignant cells and not in control cells. Nearly 51% of these elements were located in introns, 27% were in distal intergenic loci, ~11% were in promoters, ~7% were in exons, and the remainder were in untranslated regions or other downstream loci. We hypothesized that these identified cCREs help to regulate myeloid leukemogenesis or pathways of treatment resistance.
Of the leukemia-associated cCREs, diagnosis-specific cCREs (13%; n = 12,957) were enriched for motifs of transcription factor (TF) regulators of normal and abnormal myeloid differentiation, including HOXA/B, MEIS1, as well as CEBP, RUNX, and AP-1 TFs. Chemotherapy relapse-specific cCREs (12%; n = 11,980) were enriched for IRF, STAT1, ZEB1, and SNAI1 motifs, and linked to the expression of genes involved in interferon signaling, altogether suggesting a shift toward inflammatory, resistance, and self-renewal pathways in relapse. cCREs significantly more highly accessible in venetoclax relapse cells compared to controls were linked to the expression of genes in the NFkB, RAS-RAF-MEK-ERK, and PI3K-AKT-mTOR signaling pathways, including MCL-1, which are known to mediate acquired resistance to venetoclax.
We have analyzed NPM1-mutantAML samples at single-cell resolution and shown that AML cells at diagnosis and relapse are highly heterogeneous both epigenetically and transcriptionally. We have demonstrated that integrating chromatin accessibility and gene expression data enables more precise annotation of malignant cell states. We have also identified cCREs that may help to coordinate gene expression programs associated with leukemogenesis and treatment resistance, highlighting potential regulatory vulnerabilities.
