Large-scale prediction of phenotype: concept

The capability to gather organism wide data has far outstripped the ability to understand it. Transforming large-scale data into a "better" cell requires tools that integrate physiology with its environment. One such tool is large-scale mathematical models that marry stoichiometry and kinetics with metabolic regulation and control. It is straightforward to determine stoichiometry (at least for central pathways), and kinetics can be roughly approximated where need be. However, the molecular details of the "metabolic wiring" managing the cell are often missing. Presented here is a surrogate for these missing details based on a simple premise; over evolutionary time, biological systems have developed objective-based programs that frugally manage gene expression and enzyme activity. Mathematically, this notion can be represented as sets of nonlinear control or "management" problems which, when solved in parallel with the model balances, offer a prediction of how gene expression and enzyme activity are modulated, in the absence of specific mechanistic details. We present a model of Escherichia coli central carbon metabolism, describing batch aerobic growth on glucose, in which transcription, translation, and activity of the gene products of 45 genes is "managed" using this approach. The model consists of 122 species (metabolites, enzymes, mRNA pools, and biomass) and describes 46 reactions (17 reversible). The model is identified (kinetic parameters as well as management structure) from metabolic flux ratio (METAFoR) analysis and physiological measurements. Simulations of a pyruvate kinase knockout strain are compared with experiments and it is shown the model is capable of accurately capturing the metabolic reprogramming resulting from the deletion. Analysis of the mRNA expression pattern, translational pattern and enzyme activity pattern of the wild-type versus mutant indicates a combination of expression and specific activity shifts are responsible for observed differences. While being only a first step toward large-scale physiological modeling, this work is important in two ways. First, it strengthens the hypothesis that unknown mechanism can be reasonably approximated using objective-based management criteria. Second, it provides a dynamic means to couple large-scale analysis technologies with physiology at the single-gene, single-protein level.