Multi-Omic Analysis Identifies Epigenetic Evolution in Relapsed Acute Myeloid Leukemia

Introduction: Acute myeloid leukemia (AML) is associated with a poor prognosis even with aggressive treatments including high dose chemotherapy. While most patients enter clinical remission, these remissions are often short-lived leading to chemotherapy-resistant relapsed disease that accounts for the majority of deaths. We undertook a meta-analysis of published datasets consisting of 142 genotyped paired diagnosis-relapse AML samples to understand the genetic evolution of AML between the two disease states. This analysis determined that a plurality of cases exhibited the same mutations at diagnosis and relapse, and that genetically stable clones were associated with an increased probability of relapse. The finding that many cases exhibited no clonal genetic evolution upon relapse, yet exhibited chemotherapy resistance, lead us to hypothesize that epigenetic evolution plays a significant role in AML relapse. Here, our objective was to investigate the epigenetic evolution and cis and trans regulatory elements that correlate with AML relapse. Methods: We identified 27 paired diagnosis and relapse specimens from patients treated at Stanford with high dose chemotherapy regimens. Leukemic blasts, and in some cases leukemia stem cell (LSC)-enriched fractions, were purified by FACS. Cells were then analyzed through a multi-omic platform including genotyping with a myeloid malignancy targeted panel, RNA-seq, and ATAC-seq to obtain a molecular and chromatin accessibility profile of each sample. The resulting data set was analyzed to investigate epigenetic evolution in relapsed AML. Results: Genotyping analysis of banked AML specimens identified a similar pattern of genetic evolution as our meta-analysis, with several samples exhibiting the same mutations at diagnosis and relapse. We used an epigenetic matrix of chromatin accessibility data obtained from purified cell populations within the hematopoietic hierarchy and implemented this with the CIBERSORT algorithm to map the regulatory programs active in diagnosis and relapsed AML blasts. This analysis revealed a general trend of epigenetic states associated with more primitive cells (such as hematopoietic stem and progenitor cells) active at relapse, as opposed to more differentiated myeloid cell programs active at diagnosis. Focusing further on samples with no genetic changes between the two disease states, we observed several samples with substantial epigenetic evolution at relapse, with AML blasts shifting from a