Biomass composition: the "elephant in the room" of metabolic modelling.

Genome-scale stoichiometric models, constrained to optimise biomass production are often used to predict mutant phenotypes. However, for Saccharomyces cerevisiae, the representation of biomass in its metabolic model has hardly changed in over a decade, despite major advances in analytical technologies. Here, we use the stoichiometric model of the yeast metabolic network to show that its ability to predict mutant phenotypes is particularly poor for genes encoding enzymes involved in energy generation. We then identify apparently inefficient energy-generating pathways in the model and demonstrate that the network suffers from the high energy burden associated with the generation of biomass. This is tightly connected to the availability of phosphate since this macronutrient links energy generation and structural biomass components. Variations in yeast's biomass composition, within experimentally-determined bounds, demonstrated that flux distributions are very sensitive to such changes and to the identity of the growth-limiting nutrient. The predictive accuracy of the yeast metabolic model is, therefore, compromised by its failure to represent biomass composition in an accurate and context-dependent manner.The authors gratefully acknowledge the financial support from the Turkish State Planning Organization (DPT09K120520 to BK), TUBITAK (106M444 to BK), BBSRC (BRIC2.2 to SGO), EU 7th Framework Programme (BIOLEDGE Contract No: 289126 to SGO).This is the final version. It was first published by Springer at http://dx.doi.org/10.1007/s11306-015-0819-