Utilization of single-cell RNA-Seq and genome-scale modeling for investigating cancer metabolism

Abstract: Cancer remains a leading cause of death worldwide, and its dysregulated metabolism is a promising target for therapy. However, metabolism is complex to study – the metabolism of a cell involves the interplay of thousands of chemical reactions that are combined in different ways across tissues and cell types. Genome-scale metabolic models (GEMs), where the reaction networks of cells are described using a mathematical formulation, have been developed to help in such studies. In this thesis, methods were developed for determining the active metabolic network (the context-specific model) in individual cell types, followed by studies of cancer metabolism. To enable identification of the active metabolic network per cell type, single-cell RNA sequencing (scRNA-Seq) was employed to detect the presence of individual genes. However, the technical and biological variation in scRNA-Seq data poses a major challenge to the identification of the active reaction network in a cell type. The variability of gene expression due to technical and biological factors was therefore examined, concluding that data from thousands of cells is often required to provide enough stability for robust model generation. An improved quantification method for scRNA-Seq data, called BUTTERFLY, was also developed and implemented as part of the kallisto-bustools scRNA-Seq workflow. A new optimized version of tINIT, which enables generation of context-specific models, was also developed. It allowed for generation of models based on bootstrapped cell populations, which were used to acquire the statistical uncertainty of models generated from scRNA-Seq data. Finally, the method was applied to a lung cancer dataset, identifying both known and unknown features of cancer metabolism. To further explore cancer metabolism, a study was conducted to investigate the most optimal metabolic behavior under different degrees of hypoxia. To this end, a diffusion-based model for estimating nutrient availability was developed, as well as a light-weight version of the tool GECKO that enables constraining the total enzyme usage in the model. The model could explain the glutamine addiction phenomenon in cancers and was used to show that metabolic collaboration between cell types in tumors is likely not important for growth.