Computational modeling of the gut microbiota reveals putative metabolic mechanisms of recurrent Clostridioides difficile infection

Abstract

Approximately 30% of patients who have Clostridioides difficile infection (CDI) will suffer at least one incident of reinfection. While the underlying causes of CDI recurrence are poorly understood, interactions between C. difficile and commensal gut bacteria are thought to play an important role. In this study, an in silico pipeline was used to process 16S rRNA gene amplicon sequence data of 225 stool samples from 93 CDI patients into sample-specific models of bacterial community metabolism. Clustered metabolite production rates generated from post-diagnosis samples generated a high Enterobacteriaceae abundance cluster containing disproportionately large numbers of recurrent samples and patients. This cluster was predicted to have significantly reduced capabilities for secondary bile acid synthesis but elevated capabilities for aromatic amino acid catabolism. When applied to 16S sequence data of 40 samples from fecal microbiota transplantation (FMT) patients suffering from recurrent CDI and their stool donors, the community modeling method generated a high Enterobacteriaceae abundance cluster with a disproportionate large number of pre-FMT samples. This cluster also was predicted to exhibit reduced secondary bile acid synthesis and elevated aromatic amino acid catabolism. Collectively, these in silico predictions suggest that Enterobacteriaceae may create a gut environment favorable for C. difficile spore germination and/or toxin synthesis.

Fig 1. Clustering of 90 index samples using model-predicted metabolic capabilities.

(A) Average taxa abundances across the samples in each cluster for taxa which averaged at least 5% of the total abundance. (B) Number of recurrent, nonrecurrent and total samples in each cluster and all 90 index samples. None of the clusters contained a disproportionate number of recurrent samples (Fisher’s exact test, p > 0.25). (C) PCA plot of the abundance data with each recurrent and nonrecurrent sample labeled by its associated cluster number. (D) Intersection between samples clustered based on model-predicted metabolic capabilities and 16S-derived abundance data. The number in each box represents the number of shared samples between clusters.

Fig 2. Clustering of 119 post-index samples using model-predicted metabolic capabilities.

(A) Average taxa abundances across the samples in each cluster for taxa which averaged at least 5% of the total abundance. (B) Number of recurrent, nonrecurrent and total samples in each cluster and all 119 post-index samples. Cluster 2 contained a disproportionate number of recurrent samples (25/28) compared to the cluster 1 (14/28; p = 0.003) and the entire post-index dataset (83/119; p = 0.035). (C) PCA plot of model-predicted metabolic capabilities with each recurrent and nonrecurrent sample labeled by its associated cluster number. (D) Variance explained by PCA of 16S-derived abundance data and model-predicted metabolic capabilities. The total number of components for each dataset shown in the legend was determined by the MATLAB function pca.

Fig 3. Net maximal production rates of bile acid and aromatic amino acid metabolites in the high Enterobacteriaceae, high Bacteroides and high Enterococcus abundance clusters generated from 119 model-predicted post-index samples.

(A) Bile acid metabolites in which the average production rate was non-zero in at least one cluster. (B) Aromatic amino acid metabolites in which the average production rate was non-zero in at least one cluster. Error bars represent standard error of the mean. Metabolites abbreviations are taken from the VMH database (www.vmh.life). Full metabolite names, their associated metabolic pathways and numeric values for their average production rates in each cluster are given in S6 Table.

Fig 4. Analysis of patient samples in the high Enterobacteriaceae abundance (HEb) group.

All post-index samples from the 22 patients with at least one sample in the high Enterobacteriaceae (HEb) abundance cluster were grouped to generate the enlarged HEb group of 55 samples. (A) The number of recurrent, nonrecurrent and total patients in the HEb group compared to those in the HBo and HEc groups. All post-index samples of the 46 patients represented in the HBo cluster and the 21 patients represented in the HEc cluster were grouped to produce the 87 and 47 samples, respectively, in the HBo and HEc groups. The 66 total patients represented by these samples were allowed to reside in multiple groups. (B) Correlation between Enterobacteriaceae and other taxa in the HEb group calculated from the 55 post-index samples as measured by the proportionality coefficient ρ. The 7 taxa with the largest |ρ| values are shown. (C) Transient progression of samples from the 22 patients in the HEB group with samples denoted as 1 if contained in the HEb cluster, 2 if contained in the HBo cluster, 3 if contained in the HEc cluster and 4 if not clustered due to low abundance of modeled taxa (see Methods). (D) Average taxa abundances for an expanded HEb group that also contained pre-index and index samples to generate a dataset of 78 samples. These samples were partitioned into 35 samples prior to patients entering the HEb cluster, 28 samples during patient presence in the HEb cluster and 15 samples after patients left the HEb cluster.

Fig 5. Clustering of 40 FMT samples using model-predicted metabolic capabilities.

(A) Average taxa abundances across the samples in each cluster for taxa which averaged at least 5% in at least one cluster. (B) Correlations between Cronobacter and other taxa calculated from all 40 samples as measured by the proportionality coefficient ρ. The eight taxa with the largest |ρ| values are shown. (C) PCA plot of the model-predicted metabolic capabilities with each pre-FMT, donor and post-FMT sample labeled by its associated cluster number. (D) Number of pre-FMT, donor and post-FMT samples in each cluster and all 40 samples. Cluster 2 contained a disproportionately large number of pre-FMT samples (10/11) compared to the cluster 1 (4/29; p < 0.0001) and the entire FMT dataset (14/40; p = 0.0014).

Fig 6. Net maximal production rates of bile acid and aromatic amino acid metabolites in the high Enterobacteriaceae and high Bacteroides abundance clusters generated from 40 model-predicted FMT samples.

(A) Bile acid metabolites in which the average production rate was non-zero in at least one cluster. (B) Aromatic amino acid metabolites in which the average production rate was non-zero in at least one cluster. Error bars represent standard error of the mean. Metabolite abbreviations are taken from the VMH database (www.vmh.life). Full metabolite names, their associated metabolic pathways and numeric values for their average production rates in each cluster are given in S7 Table.