Integration of Deep Multi-Omics Profiling Veals New Insights into the Biology of Poor-Risk Acute Myeloid Leukemia

Background:

The poor-risk cytogenetic subgroup of acute myeloid leukaemia (AML) includes various chromosomal aberrations and represents a heterogeneous population of patients with a dismal 10-year overall survival. While the success of genetic landscaping studies is encouraging, it is debatable whether genomics, or indeed any single-omics platform alone, is sufficient to capture the biology of a disease that continues to evade our existing treatments so effectively. Instead, we need to develop a much better understanding of the complexity of this subgroup of AMLs: the relationship and interdependencies across biochemical pathways, how these may differ between patients and their impact on the leukemia and normal stem cell compartments. To launch this process, we have completed a multi-omics profiling programme to shed new light on the genetic and biochemical features of poor-risk AML (https://poor-risk-aml.bham.ac.uk/).

Aims: Application of multi-omics and integrative approaches to decipher the complexities of cytogenetically poor-risk AML

Methods: Sample inclusion criteria were based on cytogenetics and availability of sufficient diagnostic bone marrow or peripheral blood material for analysis. The 50 primary AMLs included 17 cases with complex karyotype, 13 -7/del(7), 11 KMT2A rearrangements (with the exception of t(9;11)), 4 t(6;9), 3 -5/de(5), 1 del(17) and 1 inv(3). Profiles consisted of a combination of genomics (whole genome sequencing (WGS, 60X for tumour and 30X for germ-line controls), targeted sequencing of 54 myeloid loci, and total RNA-seq (100 million reads per bulk sample), mass spectrometry proteomics and phosphoproteomics (with >6,000 proteins and > 25,000 phosphorylation sites detected and quantified), mass cytometry (CyTOF, 39 markers), drug screening (ranging from 200-500 approved or investigational compounds) and the selective generation of patient-derived xenograft (PDX) models.

Results: Near complete datasets have been compiled on all 50 primary AMLs, with the exception of WGS analysis where profiling was restricted to cases where corresponding germline DNA was available. Integration of WGS and RNA-seq data identified 122 genes having notable allele-specific expression (ASE) in ≥ 5 samples supported by ≥ 3 SNPs and these included the transcription factor GATA2 and the DNA topoisomerase TOP1MT. Use of RNA fusion capture tools resolved novel inter- and intra- chromosomal gene rearrangements that were confirmed by WGS. The four t(6;9)(p23;q34)/DEK-NUP214 cases, with a mean age of diagnosis of 43.5 years and all harboring FLT3-ITD mutations, arose from the most immature hematopoietic compartment (CD34+CD117+ enrichment) and demonstrated a unique transcriptomic signature, which included upregulation of FOXO3 and GRP56. Collectively, t(6;9) primary samples also showed a selective drug sensitivity to XPO1 (selinexor and eltanexor) and JAK inhibitors (ruxolitinib, tofacitinib and momelotinib) compared to other cytogenetic risk groups. On the other hand, a comparison of in vitro drug sensitivity data with genomic data of our entire cohort of patients demonstrated that TP53 wt AMLs (n=37) were more sensitive to all four MDM2 inhibitors (AMG-232, idasanutlin, SAR405838 and NVP-CGM097) compared to TP53 mutated cases (n=13). Comparisons of transcriptomics with the in vitro sensitivity to drugs included in early/late phase AML clinical trials, identified signatures of response associated with MDM2 and Aurora B kinase (AZD1152-HQPA) inhibitors. Phosphoproteomics analysis and machine learning modeling separated KMT2A rearranged leukemias into 2 discrete groups (group A: MLLT4, MLLT10 and TET1; group B with MLLT6, ELL and SEP9 fusion partners). Functionally, group A presented with elevated HOXA10 protein expression and enhanced in vitro response to genotoxic drugs and cell cycle inhibitors when compared to group B leukemia.

Conclusions: Our study demonstrates the feasibility of simultaneously generating omics data from several different platforms and highlights that a combination of genetic and proteomic profiles may help to inform the choice of therapies based on the underlying biology of a patient’s AML.