Expanding a Yeast Genome Scale Model to Account for Enzymatic Limitations

Genome scale models (GEMs) are a recurrent approach for modelling metabolic fluxes and integrating omic data. In the case of Saccharomyces cerevisiae(budding yeast), proteomics has been by far the less explored type of omic data [1]. Moreover, the few studies so far that have included proteomics in their approach (specifically enzymes), have done it through correlations but not directly constraining the metabolic fluxes. The latter could give improved and more realistic predictions, especially when protein resources are scarce (e.g. at high growth rates or during stress). A current challenge therefore in yeast GEMs is to expand them to account for enzymes and constrain them using experimental data. 

Here we present the enhancement of a yeast GEM [2] to account for enzymatic constraints. This was done with the constraint-based approach [3] extended to include enzymes as part of the reactions, therefore using mass balances for both metabolites and enzymes. With this formalism reaction fluxes can be limited by the maximum flux the enzyme is able to catalyse (v_max), which can be estimated from the enzyme’s abundance inside the cell together with the enzyme’s turnover number (k_cat). Enzyme abundances were either experimentally measured or retrieved from online databases. Turnover numbers were also queried from databases, but several consistency checks were followed to cope with the high variability of data. All involved enzymes in the model were inferred using the GEM’s gene level and querying gene – protein databases. Non–ideal cases such as isozymes, promiscuous enzymes and complexes were taken into account using logical transformations [4]. The presented approach shares elements from previous studies [5,6], but stands out as the first time that this is done in yeast at a genome scale, using real turnover numbers and proteomic measurements. 

The model was first tested limiting total protein abundance, which was sufficient to predict metabolic strategies in yeast such as ethanol production at high growth rates (the Crabtree effect) and under temperature stress. Afterwards, proteomic experiments with all enzyme concentrations measured were used to further constrain the model, allowing certain flexibility by introducing inconsistency scores in order to assure a feasible solution. This gave additional insight on enzyme usage, and significantly reduced the solution subspace (maintaining a biologically relevant flux distribution). The model is expected to be of great use in metabolic engineering applications, specifically for predicting enzymatic costs of producing highly-valued chemical compounds. 

References: 

[1] B. J. Sánchez and J. Nielsen, Integr. Biol., 2015. 

[2] H. W. Aung, S. A. Henry, and L. P. Walker, Ind. Biotechnol., 9, 4, 215–228, 2013. 

[3] N. E. Lewis, H. Nagarajan, and B. Ø. Palsson, Nat. Rev. Microbiol., 10, 4, 291–305, 2012. 

[4] C. Zhang, B. Ji, A. Mardinoglu, J. Nielsen, and Q. Hua, Bioinformatics, 2015. 

[5] R. Adadi, B. Volkmer, R. Milo, M. Heinemann, and T. Shlomi, PLoS Comput. Biol., 8, 7, 2012. 

[6] E. J. O’Brien and B. Ø. Palsson, Curr. Opin. Biotechnol., 34, 125–134, 2015.