A multi-scale systems biology framework for understanding and engineering the human gut microbiome

Duke University, Durham NC

Investigators

Linked publications, trials & patents

Abstract

Metabolites produced by gut bacteria that shape our immune system, metabolism and behavior are dictated by complex and dynamic multi-scale interaction networks spanning gene regulatory networks (GRNs) within each cell to inter-species interactions at the community-level. A major knowledge gap is deciphering the networks driving production of these metabolites and developing the capability to predict these key functions and their effects on host phenotypes. By developing and applying a novel generalizable systems biology framework, we propose to revolutionize our ability to quantitatively study microbiomes and discover their ecological and molecular principles. We will apply this framework investigate major health-relevant metabolites including short chain fatty acids, bile acid metabolism, trimethylamine, tryptophan metabolism and hydrogen sulfide. Since combinations of gut microbial metabolites can have synergistic or antagonistic effects on host phenotypes, we will interrogate the effects of human gut communities on intestinal barrier integrity, a critical regulator of homeostasis and dysbiosis in the human gut and modulation of T-cell activity. Finally, we will use germ-free mice as a mammalian gut model to study the dynamics and functions of these communities in vivo. We will exploit the bottom-up construction of human gut communities in vitro to elucidate the quantitative contributions of environmental factors (e.g. dietary fibers), species and their interactions on community functions. Since our in vitro media is a major variable determining experimental outcomes, we will develop a systematic pipeline for media design that maximizes agreement between in vivo and in vitro community dynamics and functions. For species selections, we will use (1) a novel data driven computational model that maps genotype to function to prioritize species based on their predicted effects on target functions and (2) state-of-the-art ultrahigh-throughput techniques to screen millions of sub-communities by combining droplet microfluidics and fluorescence activated cell sorting to enrich for sub-communities containing significant inter-species interaction impacting community functions. Equipped with a designed media and human gut community, we will explore a large experimental design landscape of combinations of bacteria and environmental factors using a scalable and flexible machine learning (ML) model coupled to Bayesian optimization to maximize information from limited experiments. In addition to building a predictive model that can be used to predict system behaviors and decipher significant interactions, a major goal is to narrow the design space to elucidate the molecular principles of communities with desired community functions. Using these communities, we will deduce the nonlinear ordinary differential equations governing community dynamics to reveal interaction modalities. Finally, we will illuminate GRNs from transcriptional profiles and map these features to community functions using novel ML models. Our predictive models can elucidate complex webs of interactions linking environmental factors, species, genes, GRNs and gut microbial metabolites.