Host-microbiome protein-protein interactions capture disease-relevant pathways

Abstract

Background: Host-microbe interactions are crucial for normal physiological and immune system development and are implicated in a variety of diseases, including inflammatory bowel disease (IBD), colorectal cancer (CRC), obesity, and type 2 diabetes (T2D). Despite large-scale case-control studies aimed at identifying microbial taxa or genes involved in pathogeneses, the mechanisms linking them to disease have thus far remained elusive.

Results: To identify potential pathways through which human-associated bacteria impact host health, we leverage publicly-available interspecies protein-protein interaction (PPI) data to find clusters of microbiome-derived proteins with high sequence identity to known human-protein interactors. We observe differential targeting of putative human-interacting bacterial genes in nine independent metagenomic studies, finding evidence that the microbiome broadly targets human proteins involved in immune, oncogenic, apoptotic, and endocrine signaling pathways in relation to IBD, CRC, obesity, and T2D diagnoses.

Conclusions: This host-centric analysis provides a mechanistic hypothesis-generating platform and extensively adds human functional annotation to commensal bacterial proteins.

Keywords: Gut microbiome; Human disease; Metagenomics; Protein-protein interactions.

Fig. 1

Human proteins differentially targeted by the microbiome in disease are enriched for relevant gene-disease associations. A The number of interspecies bacterial protein clusters (blue), human proteins (orange), and interactions (dark blue) in the human-bacteria PPI network; the number of bacterial protein clusters detected in patients from nine metagenomic studies that also have homology to experimentally verified interactors and their putative human interactors; and the number of bacterial clusters and human proteins associated with disease through our metagenomic machine learning approach, by comparing abundances in cases (gray) and control (red). B Proportions of human proteins implicated in disease, according to their GDAs (GDAs > 0.1) in DisGeNET, within: all reviewed human proteins; HBNet; human interactors with detected bacterial proteins; and those human interactors with feature importances above the 90th percentile in their respective cohorts. p values for enrichments are depicted by: * p<0.05; ** p<0.01; *** p<10⁻³; **** p<10⁻⁴ (chi-square test). Total numbers of each set are noted in the legend. C Human cellular pathways (annotated by IPA) enriched in the set of human proteins within HBNet (left) and those detected across all nine metagenomic case-control studies (right) colored according to their Benjamini-Hochberg false discovery rate (BHFDR)-adjusted p value. Only those pathways with BHFDR-adjusted < 0.05 in the disease-associated sets are shown. p values for enrichments are depicted by: * p<0.05; ** p<0.01; *** p<10⁻³; **** p<10⁻⁴ (Fisher’s exact test). D 106 species (left) with experimentally verified proteins in 3056 bacterial protein clusters are mapped to 821 bacterial species (right) with homologs detected in patients’ metagenomes (right), representing a total of 1698 clusters. Species are colored according to phylum

Fig. 2

Bacterial proteins gain access to human proteins through a variety of mechanisms. A Proportions of human proteins in the HBNet, Detected and Disease-associated subsets are plotted according to their enrichments in tissues and fluids, as annotated using DAVID. Only those with significant enrichment between any two subsets are shown. p values for enrichments are depicted by * p<0.05; ** p<0.01; *** p<0.001; **** p<0.0001 (EASE Score provided by DAVID, a modified Fisher’s exact P value; FDR-adjusted). Total numbers of each set are noted in the legend. B A schematic depicting potential opportunities for bacterial proteins to access human proteins. Interactions may involve (1) secreted human proteins, (2) bacterial proteins secreted into the extracellular space; (3) membrane vesicles that are endocytosed or can fuse with human cell membranes; (4) bacterial cellular lysate; (5) proteins injected into human cells by T3SS, T4SS, and T6SS, (6) cells and their products that translocate as a result of barrier dysfunction or “leaky gut”, and/or (7) direct contact with M cells, dendritic cells (DC), or epithelial cells. C Proportions of human proteins in the HBNet, Detected and Disease-associated subsets, are plotted according to their subcellular locations, as annotated using Gene Ontology Cellular Component, is depicted. p values for enrichments are depicted by: * p<0.05; ** p<0.01; *** p<0.001; **** p<0.0001 (chi-square test). Total percentages for these subsets are listed at right, along with p-values. Total numbers of each set are noted in the legend. D Proportions of bacterial gene clusters in the HBNet, Detected and Disease-associated subsets are plotted according to their transmembrane and secretion predictions, annotated using TMHMM, EffectiveDB, and SignalP. p values for enrichments are depicted by * p<0.05; ** p<0.01; *** p<0.001; **** p<0.0001 (chi-square test). Total numbers of each set are noted in the legend

Fig. 3

Human pathway annotation can be propagated through interactors to improve bacterial pathway annotation. A Paired stacked bar plots showing the 1102 disease-associated bacterial protein clusters according to whether they are able to be annotated by KEGG (left) and their inferred pathways according to the human proteins they target (right), as annotated by WikiPathways [77]. B Proportions of the bacterial clusters in the HBNet, Detected and Disease-associated subsets according to their COG functional categories are plotted. p values are depicted by * p<0.05; ** p<0.01; *** p<0.001; **** p<0.0001 (chi-square test). Total numbers of each set are noted in the legend

Fig. 4

Human proteins targeted by gut commensal proteins include known therapeutic drug targets. A Nafamostat, B imatinib, and C artenimol target human proteins that are differentially targeted by bacterial proteins detected in the stated metagenomic studies. Log₁₀ relative mean summed abundances of bacterial interactors in patients versus controls are provided. p-values were calculated by the Mann-Whitney rank-sum test, * p<0.05; ** p<0.01; *** p<10⁻³; **** p<10⁻⁴). Full taxa and UniRef numbers for all bacterial proteins shown are provided in Additional file 7: Table S6