Computer-assisted generation and curation of genome-scale metabolic models with case studies in the methanogen genus Methanosarcina

Methanogenic archaea, organisms that make methane as a byproduct of their metabolism, play a critical role in the global carbon cycle and have potential as a source of renewable biofuels. As a result, there has been a great deal of interest in understanding how methanogenesis works. A wide array of tools has been developed for studying methanogen metabolism, including genetic manipulation tools and efficient culturing techniques. These tools are especially well developed in model methanogens such as Methanococcus maripaludis, Methanosarcina acetivorans and Methanosarcina barkeri. Methanosarcina species are particularly attractive model organisms for methanogenesis due to their wide substrate utilization capabilities (compared to other methanogens): the diversity in metabolic capabilities for these organisms enables manipulations of methanogenesis pathways that would be lethal in other methanogens. Genetic manipulation tools have been valuable for identifying functions of individual enzymes and pathways in these organisms, but more holistic methods are needed in order to understand how these work together to accomplish observed phenotypes. Genome-scale metabolic networks allow researchers to put information on individual parts of metabolism together in a way that is useful for making novel insights. For my first Ph.D. project, I built and carefully curated a genome-scale metabolic network for Methanosarcina acetivorans and used constraint-based analysis tools to build a quantitative model based on that network. I then used the model to make predictions about how M. acetivorans utilizes carbon monoxide and the impact of the soluble heterodisulfide reductase HdrABC on its metabolic activity. While highly-curated metabolic networks are useful for studying metabolic phenotypes, the process of building them is not scalable. A genome-scale metabolic network for a single organism can take months to years to curate using the established protocols. One key reason for the lack of scalability of this process is a dearth of adequate tools to aid users in evaluating annotations and gene calls that form a bedrock for the automated generation of draft networks. The main focus of my Ph.D. has been the development of two software packages to improve the scalability of generating and curating genome-scale metabolic networks. One of these software packages, likelihood-based gap filling, uses annotation likelihood estimates for alternative gene annotations to identify pathways to fill gaps in metabolic networks that are maximally consistent with available genomic data. The other package, ITEP (Integrated Toolkit for Exploration of metabolic Pan-genomes), is a set of tools for curating and studying patterns in gains and losses of genes across groups of related organisms. In this dissertation, I describe how these tools can be used to build and to assess the quality of different parts of metabolic networks. As my final project, I have developed a new method of combining comparative genomics (using ITEP) with metabolic modeling to expose errors in both genomes and metabolic networks. I applied this method to 30 species in the genus Methanosarcina, 27 of which were newly sequenced, and demonstrated specific examples of these errors and possible ways to address them. The approach I developed makes certain classes of errors readily apparent that are not obvious when only examining individual organisms