Computing abundance constraints in Saccharomyces cerevisiae’s metabolism

Doctoral thesis, 2019

The unicellular eukaryotic organism Saccharomyces cerevisiae (budding yeast) is routinely used for production of high-value chemical compounds in the biotechnology industry. To improve production yields, it is fundamental to understand cellular metabolism, i.e. all biochemical reactions that occur inside the cell. In the past 20 years, genome-scale metabolic models (GEMs) have risen as computational tools for simulating all possible metabolic phenotypes that the cell can attain, while respecting constraints such as mass balances and reaction reversibilities. However, the number of metabolic states bound to only those constraints is infinite; therefore, it becomes necessary to include additional condition-specific constraints. Moreover, we would like these constraints to reflect physical limitations inside the cell, avoiding arbitrary ad-hoc bounds.

In this thesis, approaches for including abundance constraints (i.e. constraints based on absolute abundances of different biomolecules) are evaluated in a GEM of S. cerevisiae. First, the GEM approach and how it has been used in S. cerevisiae is reviewed, identifying key areas for development. Afterwards, the concepts of sustainable model development and multi-layer experimental data generation are presented as foundation stones for constructing integrative analysis. Regarding the first concept, a systematic way of recording changes in a GEM using a version-controlled system is introduced, allowing reproducibility and open collaboration from the community. Regarding the second concept, a multi-omics dataset of yeast grown under different temperature, osmotic and ethanol stresses is presented and used throughout the thesis for studying metabolism.

The major part of this work focuses on the integration into GEMs of abundance data of two types of bio-molecules: lipids and enzymes. First, a method for integrating lipid requirements in an unbiased way (SLIMEr) is presented and implemented for yeast, to show that lipid metabolism can be re-arranged without spending high amounts of energy. Secondly, a method for adding so-called “enzyme constraints” into a GEM (GECKO) is developed. These enzyme constraints limit reaction rates by the absolute abundance of enzymes, and prove to be crucial for explaining yeast physiology and computing enzyme usage in metabolism. Thirdly, the quantification technique used for estimating enzyme abundances is analyzed in terms of accuracy and precision, and further improved by varying the normalization and scaling steps. Finally, GECKO is used on the stress dataset to create enzyme-constrained models of yeast representing each stress condition. This allows comparing the distribution of enzyme usage within and between conditions, highlighting enzymes that play an important role in the metabolic response to stress.