Harmony Alliance Provides Novel Insights into Acute Myeloid Leukemia Based on a Pan-European NGS Data Collection

Background: To fully capture the molecular heterogeneity underlying hematologic malignancies (HMs), large cohorts need to be analyzed to unravel the impact of genetic aberrations on treatment outcome. In accordance, we have built a large “Big Data for Better Outcome” platform for HMs within the HARMONY Alliance, a pan-European private-public partnership that aims to put together over 100000 cases of AML, ALL, CLL, MM, MDS, NHL, and pediatric HMs.

Aims: We report first results of our “proof-of-principle” AML study based on the first 4986 AML cases comprised in the HARMONY platform.

Methods: First, we implemented a de facto anonymization and a data harmonization process using the Observational Medical Outcomes Partnership (OMOP) common data model to include AML into our “Big Data for Better Outcome” platform. For data analysis, we have implemented gene-gene interaction analyses for co-occurrence and mutual exclusivity, a hierarchical Dirichlet process for class discovery, and a Bradley-Terry analysis to estimate clonal evolution. Finally, to assess the effects of genomic data on clinical outcome, i.e. rates of remission, relapse and survival, we have fitted prognostic multistage models.

Results: The proof-of-principle analysis was based on patients (pts) with combined clinical and molecular information available. Male to female ratio was 53% vs. 47%, and the median age was 52 (18-91) years. The ELN 2017 risk groups were well represented (favorable: 27%, intermediate: 41%, adverse: 32%), and 43% of pts were treated with an allogeneic stem cell transplantation (alloSCT), whereas 57% of pts received conventional consolidation. Known patterns of co-occurrence and mutual exclusivity were confirmed by gene-gene interaction analysis. For example, we could provide additional evidence for the co-occurrence of RUNX1 mutations with aberrations in SRSF2, SF3B1, and STAG2, genes involved in splicing processes. Examining the variant allele frequency (VAF) of mutations using a Bradley-Terry analysis, we could further refine the model of clonal evolution and generate additional evidence that epigenetic driver mutations in genes affecting DNA methylation (e.g. DNMT3A, TET2, IDH1/2) are earlier events than mutations in histone modifying enzymes (e.g. KMT2D, EZH2, ASXL1, EP300), which usually occur later. With regard to the detection of outcome relevant mutational patterns, an unsupervised cluster genomic aberration based analysis allowed the subcategorization of “distinct” ELN risk groups, such as cases harboring an inv(16). In this core binding factor AML subclass, different inv(16) outcome cohorts could be delineated based on mutational patterns characterized by either NRAS mutations, or KRAS, KIT, FLT3-ITD and CBL mutations. Similarly, supervised outcome analyses could indicate the power of mutations to predict overall survival (OS) following an alloSCT. Our results confirmed that many patients with high-risk genotypes, such as e.g. patients harboring a TP53 mutation, do only benefit little from an alloSCT (median OS of 90 days without alloSCT vs. 382 days following alloSCT, p<0.001). However, distinct higher risk genotype constellations, such as DNMT3A in combination with PTPN11 mutations, can have a much larger survival benefit from an alloSCT (median OS 427 days without alloSCT vs. 1493 days following alloSCT, p=0.027). Finally, we could investigate FLT3-TKD mutations in n=190 cases, and define FLT3-TKD mutant AML with concomitant NPM1 mutation as “favorable” intermediate risk ELN genotype.

Summary/Conclusion: Results the HARMONY AML proof-of-principle study clearly demonstrate the benefit in combining OMOP harmonized data sets within a Pan-European “Big Data for Better Outcome” platform. In accordance, the HARMONY Alliance proves that big data sets might enable us to significantly improve individual patient outcomes and thus to further individualize patient management. By the time of the ASH Annual Meeting we will present the data of our ongoing analyses, which will included findings of a validation data set comprising additional molecularly well-defined AML cases that are currently submitted to the HARMONY platform.

(*HD and GO contributed equally to the work).