The Ecological and Temporal Genetic Architecture of The Microbiome in Health and Disease

Abstract

Most computational studies in the microbiome field run a single model to associate known microbial taxa or pathways with a disease or phenotype of interest. This approach ignores the fact that many pathways and strains of the microbiome have yet to be discovered. It also leads to results confounded by arbitrary analytical choice made during the analysis including the features to adjust for and the algorithm used. Here we use microbial genes discovered through de novo assembly to make robust, interpretable, associations that utilize the vast diversity of the microbiome. In Chapter 1 we used de novo assembly to learn about the diversity of microbial genes across microbiomes from different hosts and ecosystems. We created a gene catalog of 117 million genes from 17 different microbial ecosystems. We found that most genes only appear in a single sample, but there is also a small group of genes conserved across all types of microbiomes, but not necessarily every microorganism. We also demonstrated that the microbial genes enriched in different environments are characteristic of the ecologies or hosts they were in. In Chapter 2, we used Specification Curve Analysis, a technique used to evaluate the effect of analytical choices on associations, to demonstrate that microbial genes, taxa, and pathways are not robust predictors of future type 1 diabetes risk. We ran 5,291 models and showed great heterogeneity in prediction accuracy depending on the analytical choices made. However, we saw that using the number of autoantibodies present to predict type 1 diabetes consistently outperformed the microbiome. In Chapter 3, we use joint single-cell RNA-seq and TCR sequencing to identify T cell clones common to both the pancreatic islets and circulating blood. We found these matching T cells were clonally expanded as well indicating that they were activated upon interaction with self-antigens. Future studies will have to examine whether these T cells also react to microbial antigens as well, demonstrating a potential link between gut microbiota and type 1 diabetes. In Chapter 4 we examined how the microbiota changes over generations in response to a high fat diet. We fed C57BL/6 mice a high fat diet over 4 generations and saw that most obese mice did not have progenies, but lean, obesity resistant mice did. The progenies of lean mice were also obesity resistant, resulting in a bottleneck effect where mice got healthier over time despite the high fat diet. Because the mice were genetically identical, we suspected the microbiome may contribute to variation in weight. We showed that microbiome diversity was negatively correlated with weight and mice with more rich microbiomes were more likely to have progeny. We hypothesize that microbial diversity contributes to obesity resistance and provides an adaptive advantage for the host. To test whether the microbiome adapts to a high fat diet over time we looked for genes under positive selection from mice on a high fat diet. We found a small number of genes under selection, but future studies are needed to validate this.