Data-Driven Refinement of 321 Metabolic Reconstructions Representing the Human Gut Microbiota

The human gut microbiota is an important factor of human metabolism as it breaks down indigestible compounds from the diet into human-absorbable metabolites and provides its host with essential nutrients [1]. Microbial dysbiosis has been linked to several diseases [1], yet the underlying mechanisms are not well understood. Because of the complexity of the gut microbiota composition, it is difficult to determine the metabolic interactions among the individual organisms by experimental methods. One way of determining some of these interactions is using genome-scale metabolic reconstructions of individual microbes to help generate hypotheses on the possible metabolite exchanges and effects that can be tested in the laboratory. Several platforms exist that produce genome-scale metabolic reconstructions in only a few hours based on genome sequence as the only input [2]. Even though these pipelines continue to grow and improve, we still face several problems with the resulting reconstructions and manual curation of these draft models is generally required. This manual curation is often time consuming and inhibits the modelling community from keeping up with the increasing demand of multi-organism metabolic models.

We gathered 321 draft metabolic reconstructions of known human gut microorganisms from the Model SEED [3] and Kbase [4] platforms. We translated the reaction and metabolite names and abbreviations into our in-house database nomenclature, which uses descriptive names for reactions and metabolites. Using a semi-automated pipeline, we refined the draft reconstructions. For instance, we verified reaction directionalities by ensuring that the reactions were thermodynamically feasible and enabled the metabolic models to carry flux through the biomass objective function under anaerobic conditions. We performed extensive literature searches on the carbohydrate metabolism and fermentation pathways of the organisms and refined the reconstructions accordingly. In addition, we performed genome analyses on respiration pathways, quinone biosynthesis, and B-vitamin biosynthesis and refined the reconstructions according to our data. Finally, we removed any reactions that were gap-filled in the downloaded draft reconstructions but were not essential to a flux through the biomass objective function during unconstrained flux balance analysis. On average, we added 107±41 reactions to each of the 321 reconstructions and 15±6 reactions were removed. In total, the 321 reconstructions captured 2830 metabolic reactions. Even though the reconstructions are functionally different in many ways, by randomly selecting reconstructions and extracting the reaction lists we found that, on average, 95% of the 2830 pan-reaction list could be reached by selecting 119 reconstructions and that 75% of the same reaction list was reached after selecting only 12 reconstructions.

In summary, we introduce a semi-automated pipeline that performs data-driven refinement of draft metabolic reconstructions as well as correcting infeasible reaction directionalities that are often introduced by automatic metabolic reconstruction building pipelines. We also present a novel resource of refined microbial metabolic reconstructions that can be used for analyzing metabolic interactions in the human gut microbiota.

1. Nicholson, J.K., et al., Host-Gut Microbiota Metabolic Interactions. Science, 2012. 336(6086): p. 1262-1267.

2. Hamilton, J.J. and J.L. Reed, Software platforms to facilitate reconstructing genome-scale metabolic networks. Environmental Microbiology, 2014. 16(1): p. 49-59.

3. Henry, C.S., et al., High-throughput generation, optimization and analysis of genome-scale metabolic models. Nat Biotech, 2010. 28(9): p. 977-982.

4. Department of Energy Systems Biology Knowledgebase (KBase).