Evaluating Microbial Community Interactions, using Ecology and Flux Balance Analysis

Abstract

This work is motivated by an interest in microbial interactions and their role in human health. I wanted to answer the question of how associations between therapeutic interventions and pathology intersected with the gut microbiome; I explore this in the association between proton pump inhibitor (PPI) usage and C. difficile infection (CDI) by assessing changes in the gut microbiome during PPI treatment in Chapter 1 of my thesis. I found that observed species diversity decreased while on PPI, but that I was unable to establish a mechanistic or metabolic framework from this data in line with previously published in vitro studies on C. difficile. I explore COnstraint Based Reconstruction and Analysis (COBRA) techniques to transform a taxonomic call (“Bifidobacterium longum”) into a whole-genome metabolic model and explore the limitations of the technique using custom constraints in Chapter 2 of my thesis. I find that the absence of custom constraints can create growth tradeoffs to fulfill a systems level objective of increasing biomass, but aggressively constraining a system towards a single point of balanced growth is not experimentally validated. After satisfying my understanding of the limitations of COBRA, I explore implementation of a sum of metabolic models pairwise framework, benchmarking a biomass only ab initio framework and a rate-of-change framework that requires a starting relative abundance. I benchmark this against existing framework MICOM and conclude that our framework does not predict for overly stable communities, especially at long timepoints into the future, making our framework a poor replacement for time-longitudinal prediction. In Chapter 4, I seek to answer the question of how to improve predictions such as those from Chapter 3; noticing that macromolecular degradation was not well implemented I implement granular macromolecular degradation but see no significant improvement in prediction accuracy. Lastly in Chapter 5, I take the pairwise community interaction framework established in Chapter 3 and ask if this framework can be used to predict probiotic candidates for host communities. I establish a framework that tests pairwise interactions of a probiotic with Host Community members and pathogens, then develop a scoring metric to identify organisms that integrate with host communities and exclude C. difficile. I explore my data for consistency with existing probiotic literature, finding Bifidobacterium strains that can competitively exclude C. difficile; I verify metabolic interactions by recapitulating a Low Risk/High Risk category from published literature, demonstrating that high risk microbiota indeed are more favorable to C. difficile engraftment, consistent with a publication-pending finding. I conclude by noting that microbe-microbe interactions can be described in ecological terms such as loss of diversity, but note that with community flux balance analysis, we can describe changes in diversity based on intermicrobial competition between pairs, which cumulatively results in ecologically relevant community rearrangement.