Application of Ensemble Modeling for Strain Design

Model-based analyses of metabolic networks provide an important tool for understanding and re-design of metabolic networks for production of valuable chemicals from microorganisms. Complete modeling of metabolic networks has been recognized for its importance; however, it has been hampered by the lack of enzyme kinetics. The recently developed Ensemble Modeling (EM) approach has demonstrated its potential of constructing kinetic models of metabolic networks, hence bypassing the dependence on a detailed characterization of enzyme kinetics. In the EM approach, an ensemble of models is built to span the thermodynamically allowable kinetic space. With the utilization of elementary reaction kinetics, all models in the ensemble are confined to the same steady state and the known regulations are successfully incorporated. The size of the ensemble is reduced by screening with enzyme tuning data (flux changes due to enzyme expression tuning) that are routinely generated during stain design experiments. With approximately correct network structure and steady state, the ensemble converges to a subset of models that is able to describe the known phenotypes and predictive for future experimental phenotypes. Previously, by applying EM to study the production of aromatic amino acids in Escherichia coli, we have shown that EM can successfully recognize the transketolase (Tkt) as the first limiting step for increasing the production rate of aromatics. To enhance the utility of the approach to guide strain design, it is desirable for EM to capture the responses of the system under different conditions during strain design. Therefore, we applied EM to succinate production in Escherichia coli, including the transition in strain design from aerobic to anaerobic conditions. After screening, a subset of models that demonstrated low by-product accumulation and high succinate production in an aerobic environment was obtained. This subset of models correctly predicts a further increase in succinate production that results from inactivation of the phosphotransferase system (PTS) and overexpression of non-native phosphoenolpyruvate carboxylase (PEPC). This subset derived from aerobic data also correctly predicts an increase in the succinate yield when the strain is further engineered under anaerobic conditions. Furthermore, EM can nominate future targets for experimental testing. This has been demonstrated through the application of EM for strain development of L-lysine-producing Escherichia coli. In addition to predict reported phenotypes, the retained subset of models allows for the generation of further targets for testing, for example, phosphoenolpyruvate (Ppc), aspartate aminotransferase (AspC), and glutamate dehydrogenase (GdhA). These works demonstrate that EM is capable of constructing dynamic model without detailed enzyme kinetics and its potential of becoming useful mathematical tool to guide strain design.