Systems Biology Methods to Model Cancer Metabolism and Epigenetics for Drug Discovery

Abstract

Cancer progression involves coordinated changes in several biological networks including gene expression and metabolic flux. Despite decades of research dedicated towards understanding the cellular machinery underpinning cancer, there are network-level effects that are still not completely clear. Multi-modal data generated from high throughput experiments are now available to understand these complex processes. New data-driven methods that can leverage these large datasets to help explain complex biological processes and predict clinically relevant outcomes are needed. This thesis focuses on the algorithms and analytical approaches I developed that utilize multiple omics sources to develop a holistic model of cancer metabolism. First, I explored the important biochemical and network attributes that facilitate cancer metabolic rewiring in Chapter 2. We assembled a data model from various databases that represented the metabolic network as three categories of features: 1) biochemical features, 2) static network features, and 3) dynamic network features. Our approach called MetOncoFit trains tissue-specific cancer models to predict cancer outcomes. Our model identified metabolic enzymes with high catalytic activity are frequently upregulated in many tumors and are associated with poor survival. MetOncoFit also identified metabolites that are hot spots of dysregulation, based on their position in the metabolic network. Together, this approach illustrates how enzyme activity and the metabolic network topology influences tumorigenesis. Second, we examined how metabolism changes over the evolution of a cancer cell from a benign state to a metastatic state through a process called the epithelial-to-mesenchymal transition (EMT). In Chapter 3, I describe my analysis of multiple time-course EMT omics data using COnstraint-Based Reconstruction and Analysis (COBRA) to model changes in metabolic activity during EMT. We identified temporal metabolic dependencies in glycolysis and glutamine metabolism. Additionally, we experimentally validated an isoform-specific dependency on Enolase 3 for cell survival during EMT. Together, this approach uncovered temporally regulated cell-state-specific metabolic dependencies in cancer undergoing EMT. Finally, during our investigation of the metabolic changes that occur during EMT, we found that several metabolic reactions that are linked to acetylation were differentially active in different EMT states. Given that epigenetic modifications such as histone post-translational modifications (PTMs) regulate expression of essential genes in cancer and are associated with nutrient status, we sought to develop a predictive model of the epigenetic-metabolic network. I describe this work in Chapter 4, where we used machine learning to map the relationships between metabolism, histone post-translational modifications, and the impact metabolic status has drug sensitivity to different histone modifying drugs using multi-modal data from the Cancer Cell Line Encyclopedia. Further, we experimentally validated the metabolic link with drug sensitivity and identified four metabolites that had either synergistic or antagonistic effects on Vorinostat and GSK-J4 treatment. Together, this work describes a predictive model that suggests new associations between metabolic status, epigenetic drug therapy, and specific epigenetic states. This dissertation presents new algorithms to model cancer metabolic activity and predict biological outcomes by leveraging publicly available sources of multi-modal omics data. Through the algorithms and machine learning models I have developed, we identify potential metabolic drivers of metabolic rewiring from a network perspective and how metabolism changes during EMT. Finally, we discuss how metabolic rewiring and nutrient status impacts the epigenome and drug sensitivity. The work presented in this thesis could lead to new therapeutic strategies that target metabolic processes in various cancers.