Constraint-based modeling of yeast metabolism and protein secretion

Abstract: Yeasts are extensively exploited as cell factories for producing alcoholic beverages, biofuels, bio-pharmaceutical proteins, and other value-added chemicals. To improve the performance of yeast cell factories, it is necessary to understand their metabolism. Genome-scale metabolic models (GEMs) have been widely used to study cellular metabolism systematically. However, GEMs for yeast species have not been equally developed. GEMs for the well-studied yeasts such as Saccharomyces cerevisiae have been updated several times, while most of the other yeast species have no available GEM. Additionally, classical GEMs only account for the metabolic reactions, which limits their usage to study complex phenotypes that are not controlled by metabolism alone. Thus, other biological processes can be integrated with GEMs to fulfill diverse research purposes.   In this thesis, the GEM for S. cerevisiae was updated to the latest version Yeast8, which serves as the basic model for the remaining work of the thesis including two dimensions: 1) Yeast8 was used as a template for generating GEMs of other yeast species/strains, and 2) Yeast8 was expanded to account for more biological processes. Regarding the first dimension, strain-specific GEMs for 1,011 S. cerevisiae isolates from diverse origins and species-specific GEMs for 343 yeast/fungi species were generated. These GEMs enabled explore the phenotypic diversity of the single species from diverse ecological and geographical origins, and evolution tempo among diverse yeast species. Regarding the second dimension, other biological processes were formulated within Yeast8. Firstly, Yeast8 was expanded to account for enzymatic constraints, resulting in enzyme-constrained GEMs (ecGEMs). Secondly, Yeast8 was expanded to the model CofactorYeast by accounting for enzyme cofactors such as metal ions, which was used to simulate the interaction between metal ions and metabolism, and the cellular responses to metal ion limitation. Lastly, Yeast8 was expanded to include the protein synthesis and secretion processes, named as pcSecYeast. pcSecYeast was used to simulate the competition of the recombinant protein with the native secretory-pathway-processed proteins. Besides that, pcSecYeast enabled the identification of overexpression targets for improving recombinant protein production.   When developing these complex models, issues were identified among which the lack of enzyme turnover rates, i.e., kcatvalues, needs to be solved. Accordingly, a machine learning method for kcat prediction and automated incorporation into GEMs were developed, facilitating the generation of functional ecGEMs in a large scale.