470265 Constructing Predictive Kinetic Models of Metabolism for Guiding Strain Design

Wednesday, November 16, 2016: 8:30 AM
Continental 8 (Hilton San Francisco Union Square)
Ali Khodayari1, Satyakam Dash2, Evert K. Holwerda3, Daniel Olson3, Lee R. Lynd4 and Costas Maranas5, (1)Chemical Engineering, The Pennsylvania State University, University Park, PA, (2)Chemical Engineering, The Pennsylvania State University, State College, PA, (3)Thayer School of Engineering, Dartmouth College, Hanover, NH, (4)Thayer School of Engineering, Department of Biological Sciences, Dartmouth College, Hanover, NH, (5)Department of Chemical Engineering, The Pennsylvania State University, University Park, PA

Existing computational strain design tools operating at a genome-scale primarily rely on a stoichiometric description of metabolism. Despite many successes, designed mutants often fail as stoichiometric models do not directly account for enzyme level, metabolite concentration and substrate-level regulatory barriers. Developing kinetic models of metabolism at a genome-scale that faithfully recapitulate the effect of multiple genetic interventions would be transformative in our ability to reliably design novel overproducing microbial strains. Here, we introduce k-ecoli457, a genome-scale kinetic model of Escherichia coli metabolism that satisfies fluxomic data for a wild-type and 25 mutant strains for different substrates (glucose, acetate, and pyruvate) and growth (aerobic vs. anaerobic) conditions [1]. Model parameterization is carried out using a machine-learning genetic algorithm by simultaneously imposing all available fluxomic data. Overall, the predicted product yields by k-ecoli457 achieve significantly higher value of correlation with experimental data (i.e., Pearson’s correlation coefficient of 0.84) than FBA, MOMA or maximization of product yield (i.e., 0.18, 0.37 and 0.11, respectively).

In addition, we reconstructed a second-generation genome scale metabolic model (iCth446) for C. thermocellum by correcting cofactor dependencies, restoring elemental and charge balances and updating GAM and NGAM values to improve phenotype predictions. The iCth446 model was next used as a scaffold to develop k-ctherm118, a core kinetic model of C. thermocellum metabolism [2]. k-ctherm118 parameterization was carried out by simultaneously imposing fermentation data in lactate, malate, acetate and hydrogen production pathways for 19 measured metabolites spanning a library of 22 distinct single and multiple gene knock-out mutants. k-ctherm118 pointed at a systemic effect of limiting nitrogen resulting in increased yields for lactate, pyruvate and amino acids and an increase in ammonia and sugar phosphates concentrations (~1.5 fold) due to down-regulation of fermentation pathways under ethanol stress. The changes in the concentrations for fourteen out of eighteen metabolites in a ∆ldh mutant compared to wild-type were correctly predicted by k-ctherm118. These results quantitatively demonstrate that the developed kinetic models can reliably be used to predict genetically perturbed phenotypes under different growth conditions with a higher accuracy than any other earlier modeling effort.

[1] Khodayari, A., Maranas, C. k-ecoli457: A genome-scale Escherichia coli kinetic metabolic model satisfying flux data for multiple mutant strains. submitted.

[2] Dash, S., Khodayari, A., Lynd, L., Maranas, C. Characterization of Clostridium thermocellum strains with disrupted fermentation end-product pathways. in preparation.

Extended Abstract: File Not Uploaded
See more of this Session: In silico Systems Biology
See more of this Group/Topical: Food, Pharmaceutical & Bioengineering Division