Mathematical modeling from Metagenomics - minimizing risk of enteric infections

Infections by drug-resistant enteric pathogens or by pathogen’s that thrive post antibiotic administration is a major issue in today’s healthcare. Every year, more than 2 million people in the US acquire multi-drug resistant bacterial infections leading to over 20,000 deaths, a number that dramatically grows if we consider all the other clinical conditions occurring as result of antibiotic-resistant bacteria. Either because these bacteria are directly responsible for major gastrointestinal infections (e.g., colitis, toxic megacolon) or because they are sources of antibiotic resistant genes and vectors of disease, their precise eradication from the intestinal tract is necessary to improve clinical outcome and reduce patient-to-patient transmission. While microbiome reconstitution via Fecal Matter Transplantation (FMT) methods has been proposed as possible therapeutic alternative, it suffers from being an unstandardized approach which is also difficult to optimize. In fact, determining the bacterial components that would optimally confer a desired phonotype (e.g. achieving pathogen decolonization) from N candidate strains of interest would lead to 2N-1 possible combinations to be tested, making this an experimentally intractable issue. In this project I solve this “computationally” by first developing and applying mathematical models that have been constrained to time-series data from mouse experiments and the clinics to describe the dynamics of an enteric pathogen and the host-resident microbiome. I then use the parameterized model to (in silico) predict all the microbial combinations that are refractory to invasion by this pathogen. We have successfully applied this approach to several sets of experiments with mice infected with the enteric pathogen C. difficile and to clinical data from stem cell transplantation patients colonized with C. difficile (Bucci et al., 2016; Buffie et al., 2015). Our modeling analysis allowed identification of a conserved set of microbes between mice and humans that, when introduced into mice before C. difficile challenge, were able to significantly reduce C. difficile load and increase animal survival by 50% (Buffie et al., 2015). Furthermore, coupling metabolic pathways inference and genomic analysis of the C. difficile-refractory bacteria with microbiome profiling of the available clinical samples, we identified and validated the importance of secondary bile acid biosynthesis in repressing C. difficile growth in vivo (Buffie et al., 2015). 

In addition to discovering a minimal subset of intestinal bacteria with potent inhibition against C. difficile along with the molecular mechanism responsible for this inhibition, the statistical methods that we developed under this project were applied by us to several large clinical datasets and allowed identifying what clinical predictors are responsible for the of development Antibiotic Associated Diarrheas and recurrent C. difficile infection in a large Emergency Medicine department human cohort (Haran et al., 2016, 2015). Our machine learning-based microbiome discovery tools also allowed us to determine microbiome signatures (e.g., bacteria, genes and metabolites) that are associated with anti-Mycobacterium tuberculosis treatment (Wipperman et al., 2017)  as well as those contributing to MAIT and gamma delta T cell subsets expansion during to initial M. tuberculosis infection (Vorkas e al. 2018).