Deciphering the Heterogeneity of Acute Myeloid Leukemia through Chromatin Accessibility Profiling

Background

Acute myeloid leukemia (AML) is a heterogeneous disease and is primarily defined by genetic abnormalities. Although accumulating evidence suggests the role of epigenetics in the pathogenesis of AML, it has not fully been investigated in a large cohort of patients.

Methods

We enrolled 1,563 primary AML cases from Swedish (n=1,040) and Japanese (n=523) cohorts and performed ATAC-seq (n=1,563) as well as multi-omics analysis, including targeted-capture sequencing (n=1,563), RNA-seq (n=1,398), whole-genome sequencing (n=207), ChIP-seq (n=120), drug screening (n=112), and single-cell RNA/ATAC-seq (n=31).

Results

ATAC-seq analysis identified 185K recurrent peaks, most of which were found within intergenic/intronic regions. An unbiased clustering analysis based on ATAC-seq identified 16 unique subgroups with distinct genetic drivers, transcriptome profiles, differentiation states, key transcription factors, and clinical features. Among these, three groups were well-known entities defined by t(8;21), inv(16), and t(15;17). In contrast, the remaining 13 subtypes represent a novel classification framework not defined by single genomic abnormalities.

Patients with HOX-related gene abnormalities, such as NPM1 mutation as well as KMT2A and NUP98 rearrangement, converged into four subgroups (D-G). Associated with global chromatin changes in the HOXA locus from repressive to active state, these ATAC subgroups were characterized in common by an elevated expression of the entire HOXA cluster genes, exhibiting unique clinical and molecular features. For example, subgroup D is characterized by co-occurring NPM1 and TET2/IDH1/IDH2 mutations, older age, and high WBC/blast counts, while another HOX-subgroup (E) had monocytic nature and frequent RAS pathway mutations. By contrast, subgroups I and J were enriched for bi-allelic CEBPA mutations with and without frequent bZIP domain inframe mutations, respectively. GATA2 and WT1 mutations were common in subgroup I, while subgroup J was enriched for myelodysplasia-related mutations. TP53 mutations were largely clustered into subgroups (N, O and P) characterized by erythroid, immature progenitor, and tumor microenvironment cells, respectively. Additional ATAC subgroups included those having frequent RUNX1 (K), IDH1/IDH2 (L), and DDX41 (M) mutations, or showed a CMML-like AML phenotype (H).

We next analyzed gene regulatory networks by combining RNA-seq and ATAC-seq, revealing key transcription factors (TFs) in each ATAC-subgroup. HOXA members played central roles in the HOX-related subgroups, while IRF members, including IRF4, 7, 8 and 9, were key TFs in the subgroup enriched for RUNX1 mutations (K), leading to the upregulation of the interferon pathway. ATAC subgroups also impacted patients’ survival and significantly improved the risk prediction of ELN, enabling further stratification of each ELN risk group.

Next, we performed an in vitro drug screening for 112 samples against 250 compounds and obtained a drug sensitivity profile for each ATAC-subgroup. As expected, the subgroup with frequent FLT3-ITD mutations showed a high sensitivity to a FLT3 inhibitor (quizartinib), while other subgroups (C, F, H) with monocytic differentiation and common RAS pathway mutations were sensitive to MEK inhibitors. Furthermore, we noted an unexpected sensitivity of subgroup K samples to multiple ABL inhibitors, even though they had no known ABL-related kinase mutations.

To validate these findings, we predicted ATAC-subgroups for four external AML cohorts based on gene expression and successfully reproduced an equivalent ATAC-subgroups with similar clinical, genetic, and transcriptomic features as well as drug sensitivities.

Finally, we performed single-cell RNA/ATAC-seq and profiled a total of 233,000 mononuclear cells from 31 patients. We observed that leukemic cells were separately clustered from normal cells, while cells from the same ATAC subgroups were co-clustered, supporting that leukemic cells had their own epigenetic profiles unique to each cluster.

Conclusion

Through a large-scale multi-omics analysis of AML, we revealed a comprehensive landscape of chromatin accessibility of AML, highlighting the role of epigenetic profiling as a powerful tool for deciphering heterogeneity of AML, which could be used for a better stratification of patients and therapeutics.