Omics-driven and constraint-based modelling of microbial community metabolism

Abstract

As microbial ecology research has benefited from advancements in omics technologies, the increasing complexity and volume of generated data has in turn stimulated developments in computational biology approaches for analysis and interpretation. This is where my research lies: at the intersection of microbial ecology, metagenomics, and metabolic modelling. More specifically, this thesis outlines the development and applications of omics-driven metabolic modelling approaches to understand microbial community metabolism in diverse ecological contexts. I begin by providing background on systems biology applications of microbial ecology, metagenomic methods, and genome-scale metabolic modelling approaches. Next, I outline the development of metaGEM, a workflow designed to reconstruct context-specific genome-scale metabolic models (GEMs) via metagenome-assembled genomes (MAGs) and predict nutritional dependencies within communities directly from shotgun metagenomic samples. I applied this workflow to five datasets: synthetic lab cultures, human gut, plant associated, bulk soil, and ocean metagenomes, reconstructing over 14,000 MAGs and corresponding GEMs. Having developed this omics-driven metabolic modelling approach, I applied these methods to three case studies including: i) the distribution of auxotrophies across microbial communities, ii) the role of microbial interactions in cheese flavour formation, and iii) the pathogen-suppressing ability of a defined gut sub-community. In the first case study, I discuss two collaborations aiming to understand the prevalence, role, and origin of auxotrophies across microbes. In the first collaboration, I observed a high frequency of predicted amino acid auxotrophs across microbial communities in a large dataset, with a higher percentage of total auxotroph relative abundance in host-associated compared to free-living samples, as well as evidence for higher drug tolerance in auxotrophs. In the second collaboration, I analysed data from sequenced soil isolates and metagenomes which were also screened in the lab for individual amino acid auxotrophies by colleagues. A metagenomics-driven metabolic modelling analysis found that models largely under-estimated auxotrophies, while genomic annotations alone overestimated them. I also found evidence for the enrichment of insertion sequences in auxotrophic genomes, suggesting a possible mechanism for regulation or gene loss. In the second case study, I teamed up with industrial collaborators to analyse microbial interactions that shape cheese flavour formation. Starting with genomes assembled from fermentation cultures, I reconstructed metabolic models and simulated them under varying media conditions, revealing evidence for amino acid auxotrophies across three community members which could be rescued by a fourth microbe, as well as strain-specific metabolisms contributing to flavour formation. In the third case study, I discuss my collaboration with a team of researchers interested in the metabolic interactions that affect a defined 14-member community's ability to suppress Clostridioides difficile. For this study, I reconstructed and simulated GEMs from context-specific genomes, carried out a metabolomics data analysis between suppressive versus non-suppressive samples, and identified genomic signatures that may explain observed differences in the metabolomic landscape. Moving beyond applications of omics-driven metabolic modelling, I outline the development and application of gutDBX, a metagenomic-driven database containing 141,556 non-dereplicated microbial sequences associated with xenobiotic or pharmacologically relevant metabolism from both human and mouse gut. In short, I used a manually curated subset of relevant KEGG modules to filter sequences based on the eggNOG annotations of two catalogues, while validating the resulting database by mapping against the MetaCardis cohort. Interestingly, I found varying correlations between the number of gutDBX hits and serum cholesterol, Shannon-diversity index, and total number of prescribed drugs. Finally, I conclude by highlighting further applications of metaGEM and detailing possible future directions for omics-driven metabolic modelling, including studying the global microbiome plastic-degrading potential, long read sequencing of understudied environments, and paleometagenomics.