Abstract
Bacterial pathogens pose a major threat to human health around the globe, causing tens of millions of deaths annually. Traditionally, the clinical standard of care for treating bacterial infections are broad-spectrum antibiotics, which target universal bacterial cellular processes. While antibiotics can be effective at resolving bacterial infections, they can also disrupt beneficial populations of bacteria in our microbiomes. Additionally, the over-prescription and misuse of antibiotics is contributing to the rapid development of antimicrobial resistance phenotypes in bacterial populations. Antimicrobial resistant bacterial infection can be non-responsive to one or multiple antibiotic treatments and is a significant clinical issue that is only increasing in severity. Because of this, it is imperative that we identify more innovative and creative ways to discover alternatives to traditional antimicrobials. By using computational approaches, we can accelerate this process; identifying therapeutic candidates in silico can significantly reduce resources spent on in vitro experimentation.
Specifically, we can leverage genome-scale metabolic network reconstructions (GENREs) to simulate metabolism in bacteria with strain-level specificity. Machine learning algorithms can then be employed to identify notable patterns in the output from the metabolic simulations. Ultimately, identifying these metabolic patterns can help uncover novel targets and alternatives to traditional antimicrobial therapies, that can be validated with in vitro experimental analyses. Following this approach, we first created a python-compatible, pFBA-based, automated software tool for the generation of high-quality GENREs, Reconstructor.
Secondly, we leveraged Reconstructor to generate a large collection of GENREs of all known pathogens. We used this collection of GENREs to gain a deeper understanding of unique metabolic function in pathogen subgroups, and determined that stomach-specific pathogens exhibit the most unique metabolic phenotypes. We then identified a gene that is uniquely essential to stomach pathogens in silico (thyX), identified an existing compound (Lawsone) known to target thyX, and validated in vitro that Lawsone is a selective growth inhibitor of stomach-associated pathogens.
Finally, we used a metabolic network modeling approach to explore the functional landscape across bacterial species in over-the-counter probiotic supplements. We compared the range of metabolic functions across probiotic, commensal, and pathogenic bacterial species, to identify functional gaps in probiotic species. Finally, through in silico analyses and in vitro assays, we identified commensal vaginal species that could serve as possible members of a novel therapeutic probiotic consortium to support women’s vaginal health.
Together, the work presented in this Dissertation demonstrates that genome-scale metabolic network modeling can be a valuable tool for identifying and developing therapeutic alternatives to broad-spectrum antibiotics. As antimicrobial resistant bacterial infection becomes a more severe clinical issue, it is essential that alternatives to traditional antibiotics are developed for treating bacterial infections. Through improvements in genome sequencing and annotation technologies and further refinement of automated GENRE creation tools, this metabolic network modeling approach could prove essential for developing new clinical standards of care in the future.
