(590a) Metabolic Network Reconstruction and Genome-Scale Model of Butanol-Producing Strain Clostridium Beijerinckii NCIMB 8052

Introduction: In silico reconstruction of metabolic networks—and subsequent analysis of genome-scale metabolic models—enables a global interrogation of metabolism not possible through standard experiments. Numerous successes have been demonstrated for using these models to guide rational engineering in well-studied microorganisms such as Escherichia coli and Saccharomyces cerevisiae, and models have been built for over 50 organisms to date. In this work, we use this constraint-based modeling approach to understand Clostridium beijerinckii metabolism, and to help guide rational engineering experiments to improve production of butanol—an important chemical and potential fuel additive or replacement. C. beijerinckii is an attractive organism for strain design geared towards the production of butanol because it (i) naturally produces the highest recorded butanol concentrations as a byproduct of fermentation; and (ii) can co-ferment pentose and hexose sugars (the primary sugars in lignocellulosic hydrolysate). Interrogating the metabolism of this microorganism from a systems viewpoint will allow us to simulate the global effect of various genetic modifications.

Materials and Methods: The genome-scale metabolic network for C. beijerinckii was reconstructed by merging annotation data specific to this microorganism from three major databases: KEGG, The SEED and BioCyc. The network was further refined using both computational algorithms (GapFind, GapFill) and manual curation to minimize the number of gaps, or unusable reactions. A series of balances (e.g. mass, energy) and bounds (e.g., flux capacities, thermodynamics) were applied to a matrix representation of the metabolic network to establish a computable genome-scale model. We then used Flux Balance Analysis to determine the distribution of reaction fluxes that optimize our defined objective (biomass). To assess the predictive accuracy of the model, we conducted batch fermentation experiments to determine substrate uptake and byproduct secretion rates (corresponding to model inputs and outputs, respectively). We also conducted metabolomics profiling under a variety of conditions to further complete the model and assess its predictability. Finally, the network was probed to identify and understand reactions having the largest impact on butanol formation.  

Results and Discussion:  We present the first genome-scale metabolic model (iCM926) for C. beijerinckii, containing 926 genes, 939 reactions, and 881 metabolites. The semi-automated procedure we employed to build the model provided a rapid method for bringing the genome of a lesser-characterized microorganism to life. Specifically, this procedure capitalized on annotations available in multiple genome annotation databases (KEGG, BioCyc, The SEED) to maximize the scope of the network, and utilized computational algorithms to fill network gaps and improve model completeness. Interestingly, we found that the overlap in information between the three databases was only 34%; such a small agreement highlights the importance of exercising caution when building genome-scale models, which are often based upon one primary annotation source. To ensure predictive accuracy of our iCM926 model and assess the contribution of each database, we evaluated the ability of the model to reproduce experimentally observed substrate uptake and product production rates, and examined the source database of each reaction in the resulting flux distribution. The iCM926 model, when simulated with an optimal growth objective and fixed uptake rates for only acetate and glucose, was unable to reproduce measured production rates of butyrate, acetone, butanol and ethanol. However, with additional constraints, the experimental conditions could be more accurately simulated. Notably, a significant number of actively utilized reactions in the latter simulations originated from the set of reactions that overlapped between all three databases (P = 3.52x10^-9, Fisher’s exact test). In our simulations, inhibition of the hydrogenase reaction was found to have the largest effect on butanol formation—a relationship that has been experimentally observed as well.

Conclusions:  In silico modeling of genome-scale networks offers a promising method for globally interrogating the metabolism of C. beijerinckii, a microorganism of industrial interest for butanol production. The iCM926 model is a predictive model that accurately simulates physiological behavior and provides insight into the underlying mechanisms of microbial butanol production. As such, the model will be instrumental in efforts to better understand, and metabolically engineer, this microorganism for improved butanol production. Additionally, the construction of iCM926 for a lesser-characterized microorganism highlighted important areas of investigation (e.g. genome annotations, objective functions) for model-building efforts going forward.