Multi-omics analysis of DNMT3A- and NPM1-mutated acute myeloid leukemia

Acute myeloid leukemia (AML) represents a genetically heterogeneous group of aggressive myeloid malignancies arising from clonal expansion of aberrant, myeloid-primed hematopoietic stem or progenitor cells. Intensive chemotherapy efficiently targets proliferating blasts and achieves remission in the majority of patients. However, most patients relapse, likely due to persisting, slowly proliferating leukemic stem cells (LSCs). A novel flow cytometry sorting strategy was recently developed in-house to enrich five different leukemic populations, two of them enriched for LSCs (GPR56+NKG2DL-). This strategy was applied to a genetically harmonized DNMT3A and NPM1 double-mutant AML cohort. Despite identical driver mutations, one group presented with early relapse (ER) while the other achieved long-term remission (LTR). Multi-omics profiling (RNA-seq, DNA methylation, and genetic information) allowed me to deeply characterize these sorted leukemic populations and identify biological processes associated with ER. This analysis confirmed xenotransplantation experiments and demonstrated that the LSC-enriched populations exhibited indeed more stem-like characteristics. Still, LSC-enriched populations showed a higher cell cycle activity compared to non-engrafting, more differentiated AML populations. The LSC-enriched populations were transcriptionally similar, but the CD34+ population retained also healthy hematopoietic stem cells (HSCs) while the CD34- population contained exclusively leukemic (stem) cells. This was particularly reflected by the distinct mutant allele frequencies of the DNMT3A- and NPM1-mutations. By analyzing the LSC-enriched populations, I demonstrated a higher transcriptomic instability in ER LSCs compared to LTR LSCs that may be initiated by increasing hypomethylation associated with an earlier onset of the DNMT3A mutation. Moreover, ER LSCs exhibited a more stem-like phenotype, characterized by higher activity of mitochondrial oxidative phosphorylation compared to LTR LSCs, which presented enhanced glycolytic activity instead. The difference in energy metabolism was partially confirmed by untargeted metabolomics analyses. In a technical development project, I also implemented an interactive R shiny app (MetaboExtract) and an R package (MetAlyzer) to infer suitable extraction protocols for metabolomics studies. In addition, I trained an outcome prediction expression signature to stratify patients based on their risk of relapse and hence long-term chemotherapy sensitivity. This signature was highly predictive in different AML cohorts and was able to stratify AML patients with poor and more favorable overall survival. In summary, my work revealed biological mechanisms associated with an early relapse in LSC-enriched AML populations and generated a novel outcome prediction signature to stratify patients