Genetically-Defined AML Subgroups Represent Diverse Epigenetic Heterogeneity Landscape

Epigenetic allele (epiallele) complexity and clonal evolution are linked to inferior clinical outcome in acute myeloid leukemia (AML). However, the source of epiallelic heterogeneity in AML is not known. Here we investigated whether the abundance and genomic distribution of epialleles might be linked to canonical recurrent somatic mutations. We performed mutational and methylome sequencing in a cohort of 119 clinically annotated AML patients curated to reflect some of the most common genetic and epigenetic lesions. Specifically, patients were selected based on presence of twelve highly recurrent genetic lesions: t(8;21), t(15;17), INV(16), t(8;21), Del(5/7q), EVI1, t(v;11q23), CEBPA double mutation, mutations in DNMT3A, IDH1/2, NPM1, and FLT3. Also included were biphenotypic patients with CEBPA silencing. We used 14 normal bone marrow CD34+ cells samples (NBM) as controls. Epiallele complexity was measured using orthogonal approaches, including "proportion of discordant reads (PDR)", epipolymorphism, and Shannon entropy. We also measured the degree of global epiallele shift compared to NBM using Methclone.

Strikingly, we found that AMLs with certain specific genetic lesions manifested characteristic and distinct genomic distribution and abundance of epialleles (p = 2.51^-10, one-way ANOVA test). Overall, epiallele diversity was linked to inferior clinical outcome (p = 0.048, log-rank test of event-free survival). Moreover, epiallele complexity was significantly associated with high-risk cytogenetic AMLs (p = 0.002, Wilcoxon rank sum test). However, certain favorable risk genetic lesions - t(15;17) and CEBPA double mutation - were linked to high levels of epiallele diversity, while others manifested low burdens, including t(8;21), INV(16), and NPM1 mutations. Similarly, certain high-risk genetic lesions - Del(5/7q), CEBPA silencing, EVI1 - manifested high levels of epigenetic diversity whereas other high-risk alleles including t(v;11q23) and DNMT3A mutations had relatively low burdens. We used tSNE and various clustering metrics to determine the relative similarity in epiallele variance and location among genetically defined AMLs. These data indicated striking differences. For example, patients with t(15;17), INV(16), t(8;21), CEBPA silencing, CEPBA double mutation and t(v;11q23) formed unique and distinct clusters, as well as patients with IDH1/2 mutations but not DNMT3A mutations. Patients with Del(5/7q) and mutations in NPM1 did not form discrete clusters suggesting that these lesions do not drive formation or diversity of specific epiallele sets.

We next wondered if cooperation between somatic mutations might increase epiallele diversity. We address this question using genetically engineered mice with uniform genetic backgrounds. We tested: i) Tet2^-/-, Flt3^ITD and Tet2^-/-+ Flt3^ITD (T2F3) and ii) Idh2^R140Q, Flt3^ITD and Idh2^R140Q+ Flt3^ITD (I2F3). In all cases, we obtained purified LSK cells at similar timepoints before development of leukemia, and normal controls. In all cases, the single mutants (Tet2^-/-, Idh2^R140Q or Flt3^ITD) manifested modest if any increase in epiallele diversity. In contrast, both the T2F3 and I2F3 mice had significantly higher levels of epiallele complexity (p=0.01 and p = 0.006, t-test, respectively compared to the single mutants). Therefore, epiallele diversity can precede transformation to acute leukemia and is enhanced by cooperation between somatic mutations.

Given that epiallele diversity is linked to inferior outcome, we wondered if it could be potentially reversed using epigenetic targeted therapy. Therefore, we treated T2F3 leukemic mice with 5'Aza, and I2F3 mice with the IDH2 inhibitor AG221. In T2F3, 5'Aza profoundly reduced epiallelic diversity (p = 0.002, t-test), whereas AG-221 was effective in I2F3 but not as strong as 5'Aza in the Tet2 leukemias (p = 0.03, t-test). Both mouse models were also treated with AC220 (FLT3 inhibitor). However, AC220 - while having some anti-leukemia activity - did not reduce epiallele diversity in either model. Our data show that epiallele complexity is potentially reversible by epigenetic therapy. Perhaps part of the benefit of epigenetic therapy is due to this effect, by reducing population fitness of AMLs. Moreover, we show that genetic lesions are a critical source of epigenetic heterogeneity driving the genomic location and abundance of epialleles.