Predicting Host-Pathogen Interactions Between C. Difficile 630 and Mouse

Abstract

Clostridioides difficile (C. difficile) is the causative organism of hospital-acquired infectious diarrhea. C. difficile infection (CDI) causes more than 500,000 infections and 12,800 deaths. Antibiotic ablation of the commensal microbiome leads to colonization and infection in CDI patients. Symptomatic diarrheal infection requires the release of C. difficile toxins, TcdA and TcdB. Together, these toxins destabilize the colon epithelial cell membrane, causing fluid secretion into the colon, inflammation, and tissue damage. Molecular interaction databases describe C. difficile toxin interactions with host cell receptors, establishing these toxins as crucial virulence factors. However, there needs to be more curated and annotated information about non-toxin colonization and infection of these interactions in databases. Comprehensive identification of C. difficile toxin interactions and adhesion and colonization processes with host proteins will advance our understanding of Clostridioides difficile infection (CDI) and facilitate the design of targeted rational therapies.In this study, we predict host-pathogen interactions (HPI) between mouse and C. difficile strain 630. We select mice as the host to predict interactions as these animal models provide genetic similarity to humans and can predict human disease responses. Selecting C. difficile 630 allows for a comprehensive study due to its well-characterized genome, clinical relevance, and representation of prevalent, virulent strains. The interolog-based approach is leveraged to provide host-pathogen interactions (HPI) rapidly. This method relies on sequence-based homology to transfer experimentally validated interactions to a mouse and C. difficile 630 model. The interolog approach predicts protein-protein interactions based on the assumption that if two proteins interact between species, their orthologs will likely interact in another species' system. PSICQUIC interaction data is used to identify homologs in mouse-C. difficile. Mouse orthologs are found in human proteins using Ensembl BioMart. Next, pathogenic bacterial orthologs in C. difficile strain 630 are identified using reciprocal BLAST. This approach allows for the inference of potential interactions by mapping known interactions from one organism to another, yielding a set of 1,281 interologs; for extracellular pathogens such as C. difficile strain 630, extracellular or secreted proteins are expected to interact with host surface proteins. This pruning results in 100 interologs. CDI occurs in the colon, leading us to our next step: studying the colonic expression of host proteins. We exclude proteins not expressed in the colon as they do not contribute to the disease. Next, we manually assess C. difficile proteins in the predicted HPI for biological function and annotation. This results in 37 predicted HPIs. Network analysis of these HPIs further identifies 13 interactions with edge clustering for further investigation. One predicted HPI shows an interaction between the C. difficile strain 630 sortase protein, CD630_27180, which cleaves bacterial surface proteins for adhesion, interacting with mouse protein, Ninjurin-1, an outer surface protein in the host implicated in immune functioning. This interaction is likely as these proteins are present externally in the bacteria. Another prediction identifies an interaction between CD630_03860, a secreted sortase protein, and Col12a1, a collagen protein expressed in the colon and implicated in maintaining cell junction stability. This HPI is likely to occur as CD630_03860 is a secreted protein that may allow for the breakdown of the epithelial tight junction, giving C. difficile access to the basolateral layer of colon epithelial cells. While the interolog approach is rapid and cost-effective, it has some limitations. The interolog method relies on sequence homology to predict interactions. The predicted HPI quality in this study depends on the experimental data available on PSICQUIC. Experimental data on PSICQUIC is sourced from databases that are manually annotated. It only represents some of the available experimental data in the literature that could lead to meaningful predictions. Additionally, predicted proteins may have different functions than those in the experimental data, leading to biologically incongruent predictions in CDI. For example, C. difficile strain 630 protein flagellin (FliC) interacts with TLR5 in humans. However, this interaction cannot be predicted in mice as TLR5 in mice binds different ligands in bacterial proteins. This research advances the application of interologs to HPI prediction by using pruning methods to refine interactions specific to C. difficile infection. However, alternative approaches to predict HPIs for C. difficile should also be investigated, such as machine learning. Ensemble machine learning in predicting HPI using protein sequence has been used to predict accurate host-pathogen interaction.