Comprehensive Evaluation of Two Genome-scale Metabolic Network Models of Scheffersomyces Stipitis

Due to its important role in the production of lignocellulosic ethanol, a second generation biofuel, Scheffersomyces stipitis has drawn tremendous research interest. This is because S. stipitis has the highest native capacity to ferment xylose into ethanol and how to efficiently utilize the pentose sugars in lignocellulosic hydrolysate remains one of the major barriers in producing cost-effective lignocellulosic ethanol through biological conversion. Although S. stipitis can produce ethanol at a yield close to the theoretical maximum under microaerobic condition, its xylose consumption rate is only half of that under fully respiratory conditions. In addition, low ethanol tolerance and no growth under anaerobic condition also limit its direct utilization in industrial applications. To address these limitations, attempts have been made to improve xylose pick up rate of S. stipitis under microaerobic condition, as well as introducing xylose fermentation pathways possessed by S. stipitis to engineered Saccharomyces cerevisiae. However, these attempts have not been entirely successful: for engineered S. stipitis strains, they still suffer from low xylose consumption rates, lower ethanol tolerance as well as incapable to grow anaerobically; for engineered S. cerevisiae mutants, in general, the expression of target genes from S. stipitis has not been very effective due to different regulatory mechanisms in S. stipitis (Crabtree-negative) and S. cerevisiae (Crabtree-positive), as well as the redox imbalance introduced by the xylose fermentation pathway under microaerobic condition.
Because of the complexity involved in the two yeast strains, a better understanding on cellular metabolism of S. stipitis would be helpful for improving both S. stipitis and engineered S. cerevisiae. Specifically, genome-level understanding of xylose metabolism would provide valuable insights on identifying effective strategies for design of the engineered mutant strains. The complete genome of S. stipitis has been sequenced, which provides the foundation for genome-scale metabolic network reconstruction. Because genome-scale metabolic models connect the genotype with phenotype, they provide a holistic view of the cellular metabolism. Once validated, genome-scale models provide a platform to effectively interrogate cellular metabolism, such as characterizing metabolic resource allocation, predicting phenotype, designing experiments to verify model predictions, as well as designing mutant strains with desired properties. More importantly, genome-scale models allow systematic assessment of how a (genetic or environmental) perturbation would affect the organism as a whole.
For metabolic network models, it is clear that the quality of the model determines the outcome of the application. Therefore, it is critically important to determine how accurate a metabolic network model is, particularly a genome-scale model, in describing the microorganismâ??s cellular metabolism. Currently, model validation is done mainly through wet lab experiments, i.e., comparing model predictions with experimental measurements. Due to the cost and other difficulties associated with getting intracellular measurements, most experimental measurements are limited to cross membrane fluxes such as substrate pickup rates, production secretion rates, and cell growth rate. However, due to the scale and complexity involved in genome-scale models, a good agreement between measured and computed cross-membrane fluxes does not necessarily indicate that the model quality is high. Therefore, new methods are needed to extract qualitative biological knowledge embedded in the quantitative in silico experiments based on the metabolic network models in order to evaluate the model quality. In this work, we present a system identification based framework to examine metabolic network models in a systematic way.
Recently, two genome-scale models have been published: iSS8441 and iBB8142. iSS844 was constructed semi-automatically by integrating automatic reconstruction with S. cerevisiae as a reference framework and manual curation and modification. iBB814 was constructed manually by following a published protocol for generating a high-quality genome-scale metabolic reconstruction. iSS844 includes 1332 reactions, 922 metabolites and 4 compartments; iBB814 includes 1371 reactions, 644 metabolites and 3 compartments. Both models have fair amount of experimental validations and both performed well in matching the model predictions with some experimental measurements. However, after careful examination, we found that both models have certain limitations. For example, neither of them predicts
the production of xylitol under any condition, aerobic, microaerobic or anaerobic. However, we know that xylitol is a key byproduct produced by S. Stipitis under microaerobic condition due the redox imbalance caused by xylose fermentation step (XR and XDH). Therefore, being able to predict xylitol production under microaerobic condition is very important for a model to identify candidates for balancing cellular redox potential.
In this work, we conduct comprehensive evaluations of the two genome-scale models mentioned above, with the aim to identify a model that agrees better with existing biological knowledge to serve as the basis for further model improvement. The evaluation consists of three steps: first is the manual comparison of the reactions included in both models; second is the conduction of in silico experiments to examine modelâ??s global behavior through phenotype phase plane analysis, and to compare the model predictions with additional experimental results reported in literature; Final step is the application of the proposed system identification framework to extract biological knowledge from carefully designed in silico experiments, and to examine whether the extracted knowledge agrees with existing understandings.
From manual examination, we found that the two models have large discrepancies, as there are very limited number of overlapping reactions from both models. In addition, we identified several key errors that each model contain. But overall, iBB814 agrees better with existing biological knowledge on S. Stipitis. However, through additional experimental validation, iSS884 performs noticeably better compared to iSS814, even though the global examination through in silico experiments reveals some unexpected behavior of iSS884. Finally, with the system identification based framework, we obtain information on how the perturbation of increased oxygen pickup rate under microaerobic condition propagates through the whole network, and concluded that iBB814 yielded biological information with better accuracy. In addition, through the framework, a futile cycle was identified in the nucleotides subsystem of iSS884, which was used to provide required energy (ATP supply) for cell growth. When the futile cycle is eliminated from the model, the model fails to predict any cell growth.
In conclusion, iBB814 is the better model because it aligns better with existing biological knowledge. In addition, we demonstrated that the proposed system identification based framework is a powerful tool for extracting biological information embedded in the quantitative simulation results produced by the metabolic network models. This framework bridges the gap between the complicated numerical results generated from genome-scale models and qualitative biological knowledge that can be easily understood by biologists.

References:

1. Caspeta, Luis, Saeed Shoaie, Rasmus Agren, Intawat Nookaew, and Jens Nielsen. "Genome-scale metabolic reconstructions of Pichia stipitis and Pichia pastoris and in silico evaluation of their potentials." BMC systems biology 6, no. 1 (2012): 24.
2. Balagurunathan, Balaji, Sudhakar Jonnalagadda, Lily Tan, and Rajagopalan Srinivasan. "Reconstruction and analysis of a genome-scale metabolic model for Scheffersomyces stipitis." Microb Cell Fact 11, no. 1 (2012): 27.