Development of metabolic modelling methods to determine vulnerabilities associated to metabolic reprogramming in acute myeloid leukaemia

Résumé

Metabolism refers to all the biochemical reactions that take place inside the cells of living organisms in order to sustain life. Metabolic deregulation has been observed in disease and has been established as a hallmark of cancer. The metabolic adaptation that occurs in cancer cells contributes in cancer progression, metastasis and the development of chemotherapy drug resistance. Thus, studying cancer metabolism is of great biomedical interest. The metabolic phenotype is a result of complex biological processes and regulatory mechanisms and therefore should be studied under the holistic approach of Systems Biology. Metabolic modelling provides the appropriate mathematical framework for the representation of the entirety of metabolic reactions and pathways. The amount of genomic data and the functional annotation of entire genomes has made the reconstruction of metabolic networks at a genome scale possible. Constraint-based modelling is broadly used to perform simulations on genomescale metabolic models (GSMMs), since it can integrate previously established knowledge and experimentally generated -omic data (such as transcriptomics, metabolomics and proteomics) to build highly accurate condition-specific GSMMs for predictive studies. As part of this Ph.D., we have developed computational methods to study the metabolic adaptations that emerge in Acute Myeloid Leukaemia (AML), aiming in the identification of new therapeutic and prognostic biomarkers. More specifically, a new computational platform able to integrate a high variety of –omic data into a GSMM using constraint-based methods has been developed and employed in the study of different cell line models of AML under different stresses, i.e. drug treatment or specific gene inhibition. A systematic simulation of gene knock-outs in the GSMMs led to the identification of putative metabolic-related vulnerabilities that could be exploited in novel combination therapies. Moreover, we reconstructed a consensus genome-scale model for AML and integrated patientderived transcriptomic data from The Cancer Genome Atlas (TCGA) database, resulting in the reconstruction of AML patient-specific GSMMs. We combined constraint-based modelling and machine learning dimensionality reduction and classification to perform risk-stratification of AML patients and prognostic biomarker discovery. As a result, we have introduced a bioinformatics approach focusing on personalised AML patient care and putative novel metabolic biomarkers