EHA Library - The official digital education library of European Hematology Association (EHA)

Recent profiling studies have characterized >50 recurrently mutated genes in AML. Considering the multitude of gene candidates, evaluations of the AML genomic landscape and its relationship to clinical trajectory must extend beyond single/few biomarkers studies and, importantly, involve large well-annotated datasets.

We sought to comprehensively evaluate the known driver mutations in AML; study molecular phylogenies that underpin AML transformation, and investigate how the composite genetic architecture determines phenotype and correlates to clinical response.

We study 1540 adult AML patients from three clinical trial cohorts of the AML-SG (HD98A, HD98B and 07-04) using a targeted re-sequencing approach. Taken together with cytogenetic analysis we identify 5,241 driver variants across 55 genes and 23 recurrent cytogenetic alterations. We incorporate these to diagnostic and clinical outcome data (median follow-up=2169 days), perform  classification analysis to determine molecular subgroups, and apply variable selection models to characterize prognostic biomarkers.

We find at least 1 driver mutation in 97% patients in the study, and importantly at least 2 in 85%. There was a long tail of infrequently mutated driver genes, with only 30% of the events seen in >10% of patients. The number of drivers increases with age and patients with normal karyotype have more mutations than patients with cytogenetic abnormalities. Using the fraction of reads reporting a gene mutation to distinguish subclonal from clonal mutations, we identify 690 samples with at least 2 driver mutations and subclonal heterogeneity and map recurrent preferences in  order of mutation acquisition.

We survey the composite genomic architecture and define 10 classes, which are broadly defined by inv(16); t(8;21), t(var;11q23), inv(3), t(6;9), bi-allelic CEBPA, NPM1/DNA hydroxymethylation, TP53/aneuploidies, chromatin/spliceosome and DNMT3A/IDH2 mutations, and collectively account for 85% of patients. The definition of each class extends beyond a single lesion and is rather determined by class-specific initiating events and co-operative lesions that are ordered in time. Many of these interactions extend to mutation clusters or hotspots within genes as evidenced by pairings between FLT3 TKD/ITD, NRAS c.12-13/c.61, IDH2 c.140/c.172 and CEBPA mono/bi-allelic mutations. As expected, evaluation of clinical phenotype and outcome associations identifies broad correlations between class, phenotype, response to induction chemotherapy and overall survival. Strikingly, we find that specific molecular contexts within each class (as defined by secondary and tertiary interactions) correlate with distinct and often opposing outcome trajectories and clinical phenotype ranging from excellent to highly adverse clinical outcomes. These relationships are underpinned by models of additive risk or suggestive of genetic epistasis.

Using overall survival as a primary endpoint we evaluate the known landscape of driver mutations, and genetic interactions. We derive a comprehensive set of significant factors and identify at least three genetic interactions that retain significance (q<0.01) in the presence of other clinical and demographic variables. Importantly we identify at least one prognostic predictor in 70% of the patients and observe a three-fold increase in patients with 2 or 3 prognostic predictors relative to current molecular prognostication guidelines.

Taken together, we study an extended and well-annotated clinical cohort we demonstrate the potential of recent genetic findings to deliver improved classification and risk stratification algorithms that extend beyond single biomarker correlations, capture most AML patients and that are tailored to the individual patients genomic profiles.