(585j) Comprehensive Evaluation of Two Genome-Wide Metabolic Network Models On Scheffersomyces Stipitis

In the past few decades, significant progress has been made in the development of computational and high-throughput experimental approaches. These advancements enable the reconstruction of genome-scale metabolic network models, which provides a holistic view of the organism’s metabolism and the genotype with phenotype of the organism. These models, and constraint-based metabolic flux analysis methods, such as flux balance analysis (FBA), have been used extensively to study genome-wide cellular metabolic networks as they provide quantitative descriptions of the cellular metabolism without the knowledge of kinetic information which is often difficult to obtain¹.

Various applications of genome-scale models have been reported, such as hypothesis-driven discovery, genetic identification for overproduction of bio-value chemicals, and drug targeting. For example, the genome scale model for the bacterium Mannheimia succiniciproducens was used to guide the improved production of succinic acid by genetic engineering².  The network structure of genome-scale models can also be used to provide mechanistic knowledge under different conditions, as variation in fluxes is often coupled to changes in transcription level of genes.  A recent study showed that genome-scale models can be used to determine reactions that are transcriptional controlled³. 

Despite the genome-scale metabolic networks models enormous potentials, there is a large gap between the in silico results and comprehensible biological information that can be easily understood by biologists for genetic engineering, which have limited their more extensive use in many application.  Because these models typically predict over thousands of fluxes, there is a need for an effective method to decipher the complex network and to determine whether a model provides an accurate description of the organism.  In this work, instead of examining the massive array of intracellular fluxes, we have developed a subspace identification based approach to analyze and study cellular metabolism where genome scale models are used as high-fidelity simulators to investigate how a perturbation propagates through the whole network.  The proposed framework is developed to study how a specific factor, such as increases oxygen pick-up rate, regulates a particular enzyme and subsequently affects the metabolic network, by carrying out the following three steps: 1. Design and conduct a series of in silico experiments to reflect the desired change that would be introduce to the cell; 2. Apply multivariable analysis tools to analyze the high dimensional in silico experimental results, and identify the key component of the metabolic network that are affected by the factor; 3. Visualize the analysis results to show how the network is affected.

This approach is carried out on Scheffersomyces stipitis, which is a yeast strain that has the highest native capacity for xylose fermentation, and the potential to significantly impact the production of biomass-derived biofuels and other value-added chemicals⁴. Two recently published genome-scale metabolic network models^5,6 have been evaluated comprehensively, which for simplicity will be named Model A⁵ and B⁶ respectively. The manual examination shows that there are explicit differences between the models in terms of the model structure and reactions included, but both models predicted very similar phenotypes on growths under various oxygen and xylose conditions. Therefore, it is difficult to pin-point how those differences contribute to the model prediction, and it is difficult to determine which model provides a better description of the cellular metabolism of the strain.

By applying the proposed approach, we found that the two models have significant discrepancies mechanistically when responding to oxygen pickup rate under both oxygen-limited and aerobic conditions. These results provide a clear perspective on the metabolic rearrangement when varying oxygenation levels, and what reactions are being up-regulated or de-regulated under certain signaling.  Our analysis results clearly indicate that Model B is better than Model A, because the model predictions comply better with general qualitative knowledge of the strain.  This work demonstrates that the proposed system identification based approach has the capability to dissect large scale genome-wide metabolic network models, and can significantly expedite the understanding of complex large scale metabolic networks.

References

1. Lewis, N.E. et al. (2012) Constraining the metabolic genotype–phenotype relationship using a phylogeny of in silico methods. Nature Reviews Microbiology , 10, 291.
2. Oberhardt, M.A. et al. (2009) Applications of genome-scale metabolic reconstructions. Molecular System Biology, 5, 320
3. Osterlund, T. et al. (2013) Mapping condition-dependent regulation of metabolism in yeast through genome-scale modeling, BMC Systems Biology, 7, 36
4. Rumbold, K et al. (2009) Microbial production host selection for converting second-generation feedstocks into bioproducts. Microb Cell Fact, 8, 64
5. Caspeta, L et al. (2012) Genome-scale metabolic reconstructions of Pichia stipitis and Pichia pastoris and in silico evaluation of their potentials. BMC System Biology, 6, 24
6. Balagurunathan, B et al. (2012) Reconstruction and analysis of a genome-scale metabolic model for Scheffersomyces stipites.  Microb Cell Fact 11, 27