Abstract
* Laura Corveleyn, Jolien Vanhooren and Boris Vandemoortele are shared first authors
** Vanessa Vermeirssen, Tim Lammens and Maarten Dhaenens are shared senior authors
Acute myeloid leukemia (AML) is a complex and heterogeneous hematological disease. While genomic studies have established many driver events in AML, the inherent heterogeneity still poses significant challenges in accurate diagnosis and effective treatment. Moreover, attributing the disease solely to genes is an oversimplification as the pathobiology involves numerous complex molecular components interacting at various levels to shape functional traits. Therefore, a deeper understanding of the molecular characteristics of AML has become increasingly essential in deciphering its complexity, enabling more precise diagnosis and facilitating the development of targeted treatments. In this study, we developed a unique vertical multi-omics approach, using mass-spectrometry, that integrates three distinct molecular fractions; the histone epigenome, proteome and metabolome, respectively, from a single cell pellet of 18 genetically distinct AML cell lines. Encompassing close to 100 histone post-translational modifications (hPTMs), 5700 proteins, of which 4500 are retained to minimize imputation, and over 700 confident metabolites, this dataset provides an exceptionally comprehensive image of the molecular phenotype of 18 different AML genotypes. We optimized and applied a machine-learning based multi-omics network inference algorithm to capture subtle and non-linear associations within this AML multi-omics dataset, resulting in the delineation of 99 functional protein co-expression modules with corresponding hPTM and metabolite features. As an example, one such module encompasses proteins which clearly correlate with the recently identified mito-AML phenotype (Jayavelu et al., Cancer Cell), and moreover illustrates the metabolites (such as 5'CMP, sphingosine, methylhippuric acid) and hPTMs (H1K21Ac, H3K36Bu) that are strongly regulated with it. Several other interesting modules and their associated regulators were identified, including those linked to UBTF tandem duplications and monosomy 7, as well as multiple novel modules containing previously not yet linked proteins or regulators (hPTMs, metabolites). In conclusion, although more research is needed to fully exploit the dataset's potential, this study serves as a proof-of-principle that this pipeline will facilitate future hypothesis generation and ultimately deepen our understanding of AML pathobiology.
