Dissecting Genetic and Phenotypic Heterogeneity to Map Molecular Phylogenies and Deliver Personalized Outcome and Treatment Predictions in AML

For many years, clinical management of Acute Myeloid Leukemia (AML) has relied on patient classification into molecular groups, mostly defined by fusion genes. Recent insights of AML genomes have uncovered extended heterogeneity implicating >100 recurrently mutated genes, many of which are infrequently mutated. In each patient, multiple mutations are present defining unique genetic and clonal constellations. This genetic diversity significantly complicates the translation of molecular findings into routine clinical practice. We present our full analysis on the genomic characterization of 1540 AML patients enrolled in clinical trials of the German-Austrian AML Study Group. Together with cytogenetic profiling we map 5234 pathogenic lesions across 77 genomic loci. Amongst these, we characterise a cluster of hotspot mutations in the MYC oncogene. Overall we find ≥1 driver mutation in 96% patients, and ≥2 in 85%. The earliest mutations in AML evolution implicate genes mutated in age-related clonal hematopoiesis (DNMT3A, ASXL1, TET2) or fusion genes, followed by ordered acquisition of mutations in transcription, chromatin or splicing regulators. RTK/RAS mutations frequently represent late events with evidence of parallel evolution in 14% of AML. We formally model genomic structure and find that AML is subdivided in at least 11 molecular and clinically distinct classes defined by t(15;17), t(8;21), inv(16)/t(16;16), t(6;9), inv(3)/t(3;3), AML defined by MLL- rearrangements, CEBPAbi-allelic, NPM1, TP53/complex karyotype, AML with chromatin/splicing factor mutations, and provisionally AML with <3 aneuploidies. ~87% of patients with acquired mutations are molecularly classified. Each class is defined by a distinct subset of genetic lesions, with evidence of preferred order in mutation acquisition, thus guiding future development of minimal residual disease and combination therapy protocols. 19% (n=291) of patients were classified in the chromatin/spliceosome class. In this group, mutations in splicing factor genes and/or RUNX1 cluster with mutations in chromatin modifiers (ASXL1, EZH2, STAG2, MLLPTD). Patients in this group mostly represented Intermediate risk AML (ELN recommendations), were older, with lower WBC/blasts, inferior response rates to induction chemotherapy, poor long-term clinical outlook, higher rates of secondary AML and MDS-related morphology. Compared to classes defined by fusion genes, classes defined by genes are considerably more complex. We explore whether variability of clinical response (complete remission, relapse, relapse related mortality and overall survival) is at least in part accounted for by the extended genomic landscape. We find that the recurrent secondary and tertiary genotypes (often implicating rare genes/mutation-hotspots) markedly redefine clinical response and long-term curability beyond those predicted by single classifier lesions. To this effect, we apply global statistical models to calculate the contributions of genomic variables to overall risk whilst taking into account demographic, diagnostic and treatment factors. We find that gene-by-gene interactions are associated with additive as well as epistatic effects to patients risk, and contribute ~10% of relapse related mortality risk. We build prognostication models tailored to individual patients molecular, demographic and clinical variables at time of diagnosis and deliver more accurate risk predictions. For example, on the basis of the composite genomic and clinical profiles subsets of patients categorized as Favorable/Intermediate risk AML show risk estimates associated with adverse prognosis. Such patients are evaluated for therapeutic protocol selection tailored to higher risk groups (transplant at first CR instead of relapse), and ascertained for overall survival benefit. We apply same approaches for high-risk patients associated with favorable profiles and collectively deliver a paradigm of personally tailored risk assessment coupled with appropriate selection of therapeutic intervention. Taken together comprehensive genome profiling shows that genetic heterogeneity in AML is not random. Characterization of the extended genetic framework beyond single classifier lesions, informs future strategies for personalized prognostication, minimal residual disease monitoring and combination therapy protocols.