Control Engineering Perspective on Genome-Scale Metabolic Modeling

Abstract

Fossil fuels impart major problems on the global economy and have detrimental effects to the environment, which has caused a world-wide initiative of producing renewable fuels. Lignocellulosic bioethanol for renewable energy has recently gained attention, because it can overcome the limitations that first generation biofuels impose. Nonetheless, in order to have this process commercialized, the biological conversion of pentose sugars, mainly xylose, needs to be improved. Scheffersomyces stipitis has a physiology that makes it a valuable candidate for lignocellulosic bioethanol production, and lately has provided genes for designing recombinant Saccharomyces cerevisiae. In this study, a system biology approach was taken to understand the relationship of the genotype to phenotype, whereby genome-scale metabolic models (GSMMs) are used in conjunction with constraint-based modeling. The major restriction of GSMMs is having an accurate methodology for validation and evaluation. This is due to the size and complexity of the models. A new system identification based (SID-based) framework was established in order to enable a knowledge-matching approach for GSMM validation. The SID framework provided an avenue to extract the metabolic information embedded in a GSMM, through designed in silico experiments, and model validation is done by matching the extracted knowledge with the existing knowledge. Chapter 2 provides the methodology of the SID framework and illustrates the usage through a simple metabolic network. In Chapter 3, a comprehensive examination was carried out on two published GSMMs of S. stipitis, iSS884 and iBB814, in order to find the superior model The conventional validation experiments proved to be unreliable, since iSS884 performed better on the quantitative experiments, while iBB814 was better on the qualitative experiments. The uncertainty of which model was superior was brought to light through the SID framework. iBB814 showed that it agreed with the existing metabolic knowledge on S. stipitis better than iSS884. Chapter 4 showed that the errors in iBB814 were eliminated by refining iBB814 to construct a modified model known as iAD828. The SID framework was used to guide model refinement, which is typically a labor some and time intensive process. SID framework eradicates the trial-and-error approach, but rather has the power to uncover the reaction errors. iAD828 predicts xylitol production under oxygen-limited conditions, which is in agreement with experimental reports. This was a significant improvement, since iSS884 and iBB814 does not have this capability and now iAD828 can be used to properly engineer recombinant strains. Also the SID framework results of iAD828 show noteworthy improvement relative to iSS884 and iBB814. The superior performance of iAD828 propelled the use of this model for strategies to increase ethanol production. Understanding cofactor balance during fermentation is crucial in obtaining high quality strains for ethanol overproduction. Recently much work has been done on cofactor imbalance of the first two reactions of xylose metabolism-xylose reductase and xylitol dehydrogenase. There is not a clear understanding in S. stipitis how the cofactor preference of xylose reductase affects the metabolism. The cofactor preference of xylose reductase was varied and an optimal phenotype was determined. Analysis from this guided in silico metabolic engineering strategies resulted in elevated production of ethanol. This information can be found in Chapter 5. In Chapter 6, SID enhanced PhPP analysis was developed as a tool to overcome the limitations that PhPP analysis imposed. The power of this tool was shown by applying it to an illustrated example and an E. coli core model. Here the traditional PhPP analysis was unable to uncover the metabolic knowledge that the SID enhanced PhPP analysis was able to accomplish. The traditional PhPP analysis used shadow prices to determine the different phenotypes. This proved to be problematic for the E. coli core model. SID enhanced PhPP analysis was able to detect a “missing” phenotype that PhPP analysis failed to uncover. Also as the size of the metabolic model increases, the shadow price from PhPP analysis decreases to the point of having only miniscule meaning. Error was shown in the shadow price of the formate exchange flux. SID enhanced PhPP analysis provides a powerful tool for understanding metabolic phenotypes. Chapter 7 describes the conclusions and the future work of this study.