Network analysis of toxin production in Clostridioides difficile identifies key metabolic dependencies

Abstract

Clostridioides difficile pathogenesis is mediated through its two toxin proteins, TcdA and TcdB, which induce intestinal epithelial cell death and inflammation. It is possible to alter C. difficile toxin production by changing various metabolite concentrations within the extracellular environment. However, it is unknown which intracellular metabolic pathways are involved and how they regulate toxin production. To investigate the response of intracellular metabolic pathways to diverse nutritional environments and toxin production states, we use previously published genome-scale metabolic models of C. difficile strains CD630 and CDR20291 (iCdG709 and iCdR703). We integrated publicly available transcriptomic data with the models using the RIPTiDe algorithm to create 16 unique contextualized C. difficile models representing a range of nutritional environments and toxin states. We used Random Forest with flux sampling and shadow pricing analyses to identify metabolic patterns correlated with toxin states and environment. Specifically, we found that arginine and ornithine uptake is particularly active in low toxin states. Additionally, uptake of arginine and ornithine is highly dependent on intracellular fatty acid and large polymer metabolite pools. We also applied the metabolic transformation algorithm (MTA) to identify model perturbations that shift metabolism from a high toxin state to a low toxin state. This analysis expands our understanding of toxin production in C. difficile and identifies metabolic dependencies that could be leveraged to mitigate disease severity.

Fig 1. C. difficile toxin production is regulated by multiple metabolic signals.

The transcription of tcdA and tcdB to synthesize toxins TcdA and TcdB is positively regulated by TcdR which is in turn negatively regulated by CcpA and CodY. Each of these components are regulated by multiple metabolic signals. Fructose bis-phosphate (FBP); branched-chain amino acids (BCAA); Guanosine triphosphate (GTP). Numbers correspond to references with evidence for the indicated regulatory mechanism.

Fig 2. Metabolic differences between toxin states are strain-specific.

(A) Summary table of the RIPTiDe contextualized models including the strain, toxin production level, and number of genes, reactions, and metabolites. (B) Normalized, absolute flux values for reactions indicated by Random Forest classifier as important for distinguishing between toxin levels. C. difficile strains 630 and R20291 are shown by light and dark purple respectively. Toxin transcript levels are shown by light (low) and dark (high) teal. Starred reactions are contextualized in panel (C). (C) Map of reactions in the metabolic model. Reactions identified by Random Forest analysis in panel (B) are starred. Arg: Arginine, Orn: Ornithine, Pro: Proline, Suc: Sucrose, UDP-Glc: UDP-Glucose, Glc1P: Glucose-1-phospate, ManNAc: N-acetyl-D-mannosamine, Guo: Guanosine, dGuo: Deoxyguanosine, G: Guanine.

Fig 3. Flux through arginine and ornithine reactions is sensitive to intracellular metabolite concentrations.

(A) Summary of the shadow pricing analysis with the top 20 reactions from the Random Forest classifier set as the objective function. The number in the "models" column (blue) corresponds to the fraction of contextualized models that were able to carry flux with the indicated reaction set as the objective function (OF). The values in the orange columns indicate the following: Increase: the number of metabolites for which an increased level results in increased flux through the OF (median shadow price > 0, range < 2); Decrease: the number of metabolites for which an increased level results in decreased flux through the OF (median shadow price < -0.1, range < 2); and Variable: the number of metabolites whose shadow price varied across RIPTiDe models (range > 2). For example, in the first row of panel (A), the OF was able to carry flux in all of the models, 294 metabolites increased flux through the OF in all of the models, 1 metabolite limited flux through the OF in all of the models, and 5 metabolites had different effects on flux through the OF across all of the models. (B) Shadow prices for limiting metabolites in arginine/ornithine and aspartate metabolism reactions. The metabolites categorized as sensitive in panel (A) for these OFs and with a shadow price < -5 are shown. Increasing negative values indicates increasing reaction flux sensitivity to the metabolite.

Fig 4. Modified MTA identifies key reaction knockouts and pathways for transformation from a high to low toxin state.

The mMTA algorithm runs a reaction KO simulation to optimize changes in reaction flux that transform the model from the reference metabolic state (high toxin) to the target metabolic state (low toxin). The reaction knockouts with the highest transformation scores are shown on the y-axis. The reactions whose flux changed under these KO conditions are shown on the x-axis. Successfully changed reactions are defined as those whose flux changed from the reference in the desired direction by a minimum threshold of significance (successful: dark blue; unsuccessful: light grey). The metabolic pathways for these reactions are shown beneath the clustering dendrogram at the top.