Treatment failure in Acute Myeloid Leukemia (AML) is attributed in many cases to relapsed disease. Relapsed AML is a fundamental clinical challenge since most patients have poor clinical outcomes. The exact biological basis of AML relapse remains unclear. Genetic clonal evolution is widely believed to underlie the emergence of chemotherapy resistant clones. However, only limited, predominantly non-overlapping, somatic mutations and copy number aberrations were found to occur upon AML relapse. Furthermore, in a subset of cases, no relapse specific somatic mutations or copy number aberrations were identified. This suggests a role for other mechanisms in relapsed AML. We hypothesize that epigenetic plasticity and deregulation contributes to the pathogenesis of relapse in AML.

To explore this notion, we performed a genome scale epigenetic and genetic analysis of thirty-nine paired diagnosis and relapsed AML human patient samples using exome capture, RNA-seq and ERRBS for DNA methylation sequencing. Exome capture was performed on each patient’s germline DNA as well. Exome capture revealed only a limited number of known recurrent somatic mutations acquired upon disease relapse, in agreement with previous reports. In contrast, upon disease relapse we identified thousands of statistically significant changes in cytosine methylation patterns. Globally, the majority of patients (85%) displayed striking predominance of DNA hypermethylation (p= 1.00433e-05, binomial test for equality of proportions) upon disease relapse. Notably a smaller set of patients displayed the opposite epigenetic phenotype with prominent loss of cytosine methylation. While differential methylation in the hypermethylated group of patients localized predominantly to CpG islands, the majority of differential methylation in the hypomethylated group localized to regions lacking both CpG islands and shores. In spite of these two distinct overall cytosine methylation patterns, the majority of differentially methylated cytosines are located in intergenic regions in all cases, and a subset of promoters were hypermethylated in almost all patients at relapse. A pathway analysis indicated that the commonly hypermethylated gene promoters at relapse are involved in the Hedghog, Wnt and calcium signaling pathways (p<0.05, modified Fisher Exact test). Integration of these findings with mutational and transcriptional profiles is underway.

In order to determine whether epigenetic events linked to AML relapse could be modeled experimentally we performed a pilot study of a human AML xenograft in immunocompromised mice. Engrafted mice were treated with Ara-C at a clinically relevant dose (60mg/Kg; n=2) or vehicle alone (n=3) for five consecutive days. Human AML cells were collected at various timepoints including 28 days after Ara-C treatment where the AML had frankly relapsed in mice. Cytosine methylation profiles obtained through ERRBS revealed predominantly hypermethylated cytosines when compared to the xenotransplanted diagnostic sample (72% hypermethylated versus 28% hypomethylated). Remarkably, there was a strong overlap with gene promoters that are also aberrantly methylated in relapsed AML patients (p<0.01, hypergeometric test), including members of the Wnt signaling pathway.

We conclude that there are epigenetically distinct forms of relapsed AML. Nonetheless, there is convergent epigenetic regulation of specific gene pathways that may contribute to relapsed AML pathogenesis and xenotransplanted AML mice can serve as experimental models for further study. Finally, the genomic distribution of reprogrammed methylation suggests a role for epigenetic plasticity at distal regulatory elements. Whereas it remains unclear whether these changes represent clonal selection, their extensive and dynamic range suggest that exposure to chemotherapy may alter the fidelity of mechanisms that control cytosine methylation distribution thus permitting widespread and distant epigenetic reprogramming and contributing to disease relapse.


No relevant conflicts of interest to declare.

Author notes


Asterisk with author names denotes non-ASH members.