UCSF

It is well established that the composition of a human’s microbiome can contribute to variations in drug metabolism. Indeed, from cumulative research efforts over the past sixty years, hundreds of drugs are now known to be altered in the presence of the gut microbiome, and many of these changed therapeutic profiles are due to direct metabolism by bacterial enzymes. While studies exploring chemical transformations within the human gut microbiome increasingly employ high-throughput methods, determining metabolite identities and the genetic elements responsible for their production is still a low-throughput process. What results are large knowledge gaps that must be overcome before the physiological and clinical relevance of a given bacterial drug metabolism event can be determined. In this thesis, I demonstrate that computational techniques can help alleviate such knowledge gaps. Attempts have been made in the past to computationally predict which bacterial species and enzymes are responsible for chemical transformations in the gut environment, but with limited utility and low accuracy.

This dissertation begins with an overview of current computational approaches for exploring single-step xenobiotic transformations in the microbiome. I highlight the strengths and weakness of current approaches and make architectural recommendations for future tools. Based on these observations and recommendations, I present an in silico approach that employs chemical and protein Similarity algorithms that Identify MicrobioMe Enzymatic Reactions (SIMMER). I show that SIMMER predicts the chemistry and responsible species and enzymes for a queried reaction with high accuracy, unlike previous methods. I demonstrate SIMMER use cases in the context of drug metabolism by predicting previously uncharacterized enzymes for 88 drug transformations known to occur in the human gut. Bacterial species containing these enzymes are enriched within human donor stool samples that metabolize the query compound. After demonstrating its utility and accuracy, I chose to make SIMMER available as both a command-line and web tool, with flexible input and output options for determining chemical transformations within the human gut. Lastly, we demonstrate an experimental use-case of SIMMER by performing the first species-level characterization of methotrexate metabolism in the microbiome of Rheumatoid Arthritis patients.

I present SIMMER as a computational addition to the microbiome researcher’s toolbox, enabling them to make informed hypotheses before embarking on the lengthy laboratory experiments required to characterize novel bacterial enzymes that can alter human ingested compounds. Beyond pharmaceutical applications, SIMMER can additionally be employed to determine bacterial enzymes responsible for breaking down non-therapeutics, such as dietary compounds or environmental pollutants. The method can also be extended in the future to make predictions on microbes in other body sites or environments.