Converging molecular evolution in acute myeloid leukaemia

Acute myeloid leukaemia (AML) is a highly heterogeneous disease. Here, we decipher the disease composition of a single AML patient through longitudinal sampling scrutinized by high-resolution genetic and phenotypic approaches, including sequencing, immunophenotyping, ex vivo drug sensitivity testing and establishment of patient-derived xenograft models. Throughout the disease course we identified patterns of both divergent and convergent molecular evolution within the leukemic compartment. We identified at least six discrete leukaemia initiating cell populations, of which five were characterised by known recurrent mutations in AML. These populations partly correlated with immunophenotypically defined cell subsets, drug sensitivity profiles and population-specific potential for engraftment in immunodeficient mice. Our results indicate that the genetic and phenotypic development are closely intertwined, and that diversity in the leukaemic gene-environment likely influences disease trajectories. SIGNIFICANCENovel therapeutic approaches in AML are characterised by targeting molecular mechanisms thought to drive leukemogenesis, but emergent evidence suggests that intra-leukemic composition may be more diverse than previously appreciated. Through in-depth genetic and phenotypic characterization of the disease course of a single AML patient, we demonstrate a high degree of inter-individual complexity that exceeds the prevailing disease conception. The temporal molecular landscape of this patient suggests that leukemogenic transitions may not be categorically monoclonal. Patterns of converging molecular evolution further imply that higher levels of biological organisation than the molecular machinery of single cells may influence leukemogenic trajectories. Disease dynamics, relational properties and causal contribution from several levels of biological organization comes into conflict with the linear monocausal explanatory model on which precision oncology is largely built. This may have implications for current precision oncology oriented prectices, including molecular categorization, molecular therapeutic targeting and predictive models.