Skip to main content
PMC home page
eLife logoLink to eLife
. 2022 Jan 27;11:e72801. doi: 10.7554/eLife.72801

Multi-omics investigation of Clostridioides difficile-colonized patients reveals pathogen and commensal correlates of C. difficile pathogenesis

Skye RS Fishbein 1,2, John I Robinson 3, Tiffany Hink 4, Kimberly A Reske 4, Erin P Newcomer 1,2, Carey-Ann D Burnham 2,5,6, Jeffrey P Henderson 3, Erik R Dubberke 4,✉, Gautam Dantas 1,2,5,7,✉
Editors: Peter J Turnbaugh8, Gisela Storz9
PMCID: PMC8794467  PMID: 35083969

Abstract

Clostridioides difficile infection (CDI) imposes a substantial burden on the health care system in the United States. Understanding the biological basis for the spectrum of C. difficile-related disease manifestations is imperative to improving treatment and prevention of CDI. Here, we investigate the correlates of asymptomatic C. difficile colonization using a multi-omics approach. We compared the fecal microbiome and metabolome profiles of patients with CDI versus asymptomatically colonized patients, integrating clinical and pathogen factors into our analysis. We found that CDI patients were more likely to be colonized by strains with the binary toxin (CDT) locus or strains of ribotype 027, which are often hypervirulent. We find that microbiomes of asymptomatically colonized patients are significantly enriched for species in the class Clostridia relative to those of symptomatic patients. Relative to CDI microbiomes, asymptomatically colonized patient microbiomes were enriched with sucrose degradation pathways encoded by commensal Clostridia, in addition to glycoside hydrolases putatively involved in starch and sucrose degradation. Fecal metabolomics corroborates the carbohydrate degradation signature: we identify carbohydrate compounds enriched in asymptomatically colonized patients relative to CDI patients. Further, we reveal that across C. difficile isolates, the carbohydrates sucrose, rhamnose, and lactulose do not serve as robust growth substrates in vitro, consistent with their enriched detection in our metagenomic and metabolite profiling of asymptomatically colonized individuals. We conclude that pathogen genetic variation may be strongly related to disease outcome. More interestingly, we hypothesize that in asymptomatically colonized individuals, carbohydrate metabolism by other commensal Clostridia may prevent CDI by inhibiting C. difficile proliferation. These insights into C. difficile colonization and putative commensal competition suggest novel avenues to develop probiotic or prebiotic therapeutics against CDI.

Research organism: Human, Other

Introduction

Clostridioides difficile infection (CDI) remains a significant cause of morbidity and mortality in the health care setting and in the community (Guh et al., 2020). Antibiotic treatments, among other risk factors associated with weakened colonization resistance, increase susceptibility to CDI (Dubberke and Olsen, 2012; Eze et al., 2017). C. difficile residence in the human gastrointestinal (GI) tract may result in a spectrum of clinical manifestations, from asymptomatic colonization to severe CDI-related colitis and fatal toxic megacolon (Crobach et al., 2018). Diagnosis of CDI relies on detection of the protein toxin, most commonly by enzyme immunoassay (EIA), or the detection of the toxin-encoding genes tcdA and tcdB, by nucleic acid amplification test (NAAT). These diagnostic tools serve as rough benchmarks for assessing the severity of disease. Discrepancies between the results of these assays, as in the case of patients with clinically significant diarrhea (CSD) who are EIA negative (EIA-) but NAAT positive for toxigenic C. difficile (Cx+), highlight the complexity of states in which C. difficile can exist in the GI tract. Because CDI is a multi-factorial interaction between the host, pathogen, and microbiome, clarifying the differences in biological correlates between asymptomatic colonization (Cx+/EIA-) and CDI (Cx+/EIA+) is critical for identifying mechanisms of colonization resistance, and for defining novel probiotic or prebiotic avenues for treatment or prevention of CDI (Kondepudi et al., 2012; Rätsep et al., 2017).

C. difficile enters the GI tract as a spore, germinates in the presence of primary bile acids, and replicates through consumption of amino acids and other microbiota or host-derived nutrients (Hryckowian et al., 2017). Notably, many of these metabolic cues are characteristic of a perturbed microbiome (Nagao-Kitamoto et al., 2020; Battaglioli et al., 2018). The hallmark of C. difficile pathogenesis is the expression of the toxin locus encoded on the tcd operon; this locus is tightly regulated by nutrient levels (Martin-Verstraete et al., 2016). Correspondingly, it is hypothesized that an environment replete of nutrients induces toxinogenesis, allowing C. difficile to restructure the gut environment and acquire nutrients through inflammation (Fletcher et al., 2018; Fletcher et al., 2021). The instances of patients who are colonized but have no detectable C. difficile toxin in their stool suggests that these patients’ microbiomes may be less permissive towards CDI development. Identification of metabolic traits within the microbiome of asymptomatic, C. difficile-colonized patients could reveal a number of potential therapeutic pathways toward precise amelioration of symptomatic C. difficile disease.

A multitude of probiotic and prebiotic approaches have demonstrated efficacy to curb C. difficile proliferation in vivo (Rätsep et al., 2017; Chen et al., 2020; Pereira et al., 2020). While restoration of the microbiota through fecal microbiota transplantation can provide colonization resistance (Laffin et al., 2017), the molecular mechanisms of how this resistance is conferred remain unclear. Recent studies using a murine model of infection have indicated that the administration of carbohydrates (both complex and simple) in the diet can be used to curb or prevent CDI (Mefferd et al., 2020; Schnizlein et al., 2020; Hryckowian et al., 2018). Paradoxically, integrated metabolomics and transcriptomics data collected during murine C. difficile colonization indicates that simple carbohydrates are imperative for pathogen replication (Fletcher et al., 2018). It is critical to understand the mechanism by which catabolism of specific carbohydrates could inhibit C. difficile proliferation in the human GI tract.

Here, we perform a multi-level investigation of two relevant patient populations, those colonized with C. difficile but EIA negative (asymptomatically colonized) and those who are EIA positive (CDI) to understand the microbial and metabolic features that may underlie protection from CDI. First, we use microbiome analyses to identify a number of non-C. difficile, clostridial species that are negatively correlated with C. difficile in asymptomatically colonized individuals. Secondly, interrogation of a metabolomics dataset from the same patient population (Robinson et al., 2019) reveals increased abundance of a number of carbohydrate metabolites in asymptomatically colonized patients. Finally, we show that some metabolites enriched in asymptomatically colonized individuals are largely non-utilizable by C. difficile isolates. Together, these datasets reveal that asymptomatically colonized patients are defined by an interaction of clostridial species and carbohydrate metabolites that may serve as a last-line of resistance against CDI in colonized patients.

Results

The clinical manifestation of C. difficile colonization in a host gastrointestinal tract is determined by a multi-factorial interaction between the host, their microbiome, and the pathogen. We hypothesized that, among these factors, natural variation in C. difficile strains infecting patients might differentiate asymptomatic from CDI patients (Dubberke et al., 2018). Through retrospective analysis of a human cohort of 124 patients (Supplementary file 1) with clinically significant diarrhea (CSD) and stool submitted for C. difficile toxin testing, we defined two cohorts: those diagnosed with CDI (Cx+/EIA+) or those asymptomatically colonized (Cx+/EIA-) (Robinson et al., 2019). EIA status (EIA+ or EIA-) was determined by the result of the clinical toxin EIA performed on the stool specimen, and a positive toxigenic culture (a C. difficile isolate with with tcdA and/or tcdB; Cx+) (Dubberke et al., 2018). In-depth analysis of C. difficile isolate factors related to EIA status was performed on the isolates corresponding to the 102 metagenomic samples analyzed (see Materials and Methods, Supplementary file 2). Multiplex PCR was used to identify isolates with cdtAB, the binary toxin locus (Cowardin et al., 2016). Notably, there was a significant enrichment of isolates with cdtAB in the stools of patients with CDI (Figure 1A; p = 0.0012, Fisher’s exact test). Additionally, there were differences in the distribution of C. difficile strains associated with the two patient cohorts; CDI patients were more likely to be infected by a C. difficile isolate of the ribotype 027 lineage (Figure 1A; p = 0.0058, Fisher’s exact test), a clade likely to contain virulent members (Merrigan et al., 2010). Interestingly, of the isolates positive for cdtAB (22 out of 102 isolates), 36% were considered a ribotype 027 strain. Given these genetic indicators of potential differences in virulence, we asked if strains from both groups were capable of producing toxin, using culture supernatants from in vitro broth culture. We found that 56% of isolates expressed detectable TcdA/B, with no significant different (p = 0.86) in the capacity of strains from Cx+/EIA- stools (24 out of 54 isolates) or Cx+/EIA+ (24 out of 48 isolates) to elaborate toxin (Figure 1—figure supplement 1A). Predictably, differences in genetic indicators of strain virulence (as indicated by prevalence of both a prominent ribotype and a second toxin locus) were significant correlates of EIA status.

Figure 1. Pathogen and microbiome determinants of C. difficile-colonized patients.

(A) Clostridioides difficile isolate distribution based on PCR and ribotyping data for each isolate cultured from patient stools. **, p < 0.001 as measured by a Fisher’s exact test. (B) Principal coordinate analysis (PCoA) of weighted Unifrac distances between stool microbiomes. Colors indicate EIA status. Groups were not significantly different as measured by a PERMANOVA (p = 0.69). (C) Significant microbial taxa associated with disease state, where a positive coefficient is associated with Cx+/EIA+ state and a negative coefficient is associated with Cx+/EIA- state. Colors indicate taxonomic Class of the microbial feature, and the size of circle corresponds to magnitude of statistical significance. Features with q-value of <0.25 were plotted. (D) Network of features associated with antibiotic exposure or EIA status. Species nodes are connected to metadata nodes by edged colored with the feature weight (coefficient) computed using linear mixed modeling (MaAslin2). All taxa displayed had a q-value of <0.25 in respective analyses.

Figure 1—source data 1. Raw absorbance value for in vitro toxin ELISA of 102 C.difficile isolates.

Figure 1.

Figure 1—figure supplement 1. Microbiome configuration of C. difficile-colonized patients.

Figure 1—figure supplement 1.

(A) Presence of protein toxins TcdA and TcdB, based on ELISA of in vitro culture supernatants. There was no significant difference in the number of isolates able to produce toxin (p = 0.86, Fisher’s exact test). (B) Faith’s diversity between microbiomes of asymptomatically-colonized (Cx+/EIA-) and CDI (Cx+/EIA+) patients. Diversity between groups was not significantly different (p = 0.1602, Wilcoxon rank sum test). PCoA of weighted UniFrac distances between stool metagenomes, colored by the presence of the binary toxin (CDT) locus (C) or corresponding isolate ribotype (D). (E) Phylum-level relative abundances between CDI and asymptomatically-colonized patients. (F) Relative abundance of significantly altered clostridial taxa from MaAsLin2 analysis. p-values in (E) and (F) calculated using a Wilcoxon rank sum test and corrected for multiple testing.

Figure 1—figure supplement 2. Kraken analysis of metagenomic data.

Figure 1—figure supplement 2.

(A) Significant microbial taxa associated with disease state, where a positive coefficient is strongly associated with CDI state. Colors indicate taxonomic Class of microbial feature, and size of circle corresponds to magnitude of statistical significance. Features with q-value of <0.25 were plotted. (B) Relative abundance of significantly altered taxa from MaAsLin2 analysis with a q-value of <0.05. p-Values indicate a corrected Wilcoxon rank sum test. (C) Relative abundance of C. difficile stratified by whether the stool contained an isolate with cdtAB or not. (D) Relative abundance of C. difficile stratified by the corresponding isolate ribotype.
Open in a new tab

As antibiotics are a well-known risk factor for CDI, we analyzed previous inpatient antibiotic orders (within one month prior to diagnosis) for patients in the Cx+/EIA- and Cx+/EIA+ cohort, as a proxy for antibiotic exposure (Table 1). Fitting antibiotic exposure to a logistic regression model (McFadden’s R2 = 0.306) revealed that CDI was significantly associated with cephalosporin exposure. Analysis of potential antibiotic exposures in our patient cohort confirms the risk that antibiotics pose for CDI development (Mullish and Williams, 2018; Webb et al., 2020).

Table 1. Logistic regression coefficients for antibiotic exposures associated with Cx+/EIA+ in patient cohort.

Antibiotic Coefficient Standard error p-Value
Cephalosporin 2.68 0.74 2.70E-04
Fluoroquinolone 0.19 1.09 0.86
Carbapenem 0.34 1.12 0.76
Metronidazole 1.11 0.95 0.24
Vancomycin (intravenous) 1.44 0.95 0.13
Open in a new tab

*Hosmer and Lemeshow Goodness of fit test p = 0.7536.

Antibiotics increase susceptibility to CDI through disruption of colonization resistance, mainly conferred to the host via the gut microbiome (Theriot et al., 2014). To determine the microbial correlates of disease state, we performed shotgun metagenomic sequencing on patient stool samples from the asymptomatic (n = 54) and CDI (n = 48) groups, and classified species using MetaPhlAn2. Given the strong association with antibiotic exposure in our CDI cohort, we hypothesized that our asymptomatically colonized patients would have increased microbiome-mediated colonization resistance relative to CDI patients. We examined community structure in stool metagenomes and found that there was no significant difference in Faith’s diversity (Faith and Baker, 2007), a measure of alpha-diversity that incorporates phylogenetic relationships, between patient groups (Figure 1—figure supplement 1B, Wilcoxon rank-sum test, p = 0.1602). There were no significant differences in beta-diversity, as measured by weighted Unifrac distance, between EIA status (p = 0.233, permutational analysis of variance test [PERMANOVA]) (Figure 1B). Although we hypothesized that increased virulence (through additional toxin allele or ribotype) associated with EIA+ could affect microbiome structure, we found no significant association between beta-diversity and cdtAB presence (Figure 1—figure supplement 1C; p = 0.799, PERMANOVA) or ribotype distribution (Figure 1—figure supplement 1D; p = 0.982, PERMANOVA). Previous comparative microbiome studies have revealed phylum-level differences in Bacteroides and Firmicutes in CDI cases versus controls not colonized with C. difficile (Kachrimanidou and Tsintarakis, 2020). In contrast, we found no significant differences in relative abundance of bacterial phyla between asymptomatically colonized patients and patients with CDI (Figure 1—figure supplement 1E). These data indicate that there were no gross differences in microbiome structure related to either EIA status or pathogen features.

Instead, we hypothesized that differences between these states may manifest at higher resolution. We used a multivariable regression model, as implemented by MaAslin2 (Mallick et al., 2021) to identify microbial taxa predictive of either group. Interestingly, species from class Clostridia were most enriched in taxa significantly altered by EIA status (Fisher’s exact test, p = 0.0022). C. difficile was the strongest predictor of CDI state, whereas non-C. difficile clostridial taxa were predictive of asymptomatic state (Figure 1C, FDR < 0.25). Correspondingly, we saw increased C. difficile relative abundance in CDI patients and increased levels of a number of non-C. difficile clostridial species, including Eubacterium spp., Dorea spp., and Lachnospiraceae spp. in asymptomatic patients (Figure 1—figure supplement 1F). Given our inability to detect C. difficile in all sequenced stools (70 out of 102 culture-positive stool samples), we utilized an alternative metagenomic species classifier Kraken (Wood and Salzberg, 2014), to validate our findings. Using Kraken, we detected C. difficile in nearly all stool metagenomes (101 out of 102). Using the identical linear mixed modeling approach (MaAsLin2), we recapitulated data indicating that C. difficile abundance was the strongest predictor of EIA status and increased in Cx+/EIA+ patients. Additionally, a number of commensal clostridial taxa from the Eubacterium genus and Anaerostipes genus were strongly associated with EIA- status, confirming prior MetaPhlAn2 predictions (Figure 1—figure supplement 2A,B).

Using our microbiome data, we examined the association between C. difficile levels and pathogen markers previously associated with EIA status. We found that C. difficile relative abundance was not significantly different when stratified by isolate CDT status (Figure 1—figure supplement 2C; p = 0.3, Wilcoxon rank sum) or isolate ribotype (p = 0.78, Kruskal-Wallis). Notably, there was a slight, yet insignificant increase in C. difficile abundance in microbiomes associated with a ribotype 027 isolate (Figure 1—figure supplement 2D) relative to microbiomes associated with C. difficile isolates of other ribotypes. We also interrogated taxonomic features that were predictive of antibiotic exposure. Expectedly, we found that taxonomic features predictive of CDI state were also associated with antibiotic exposure (Figure 1D). Our data indicate that patients with asymptomatic C. difficile colonization or CDI do not have grossly different gut microbiome community structures but instead have distinctive alterations in a subset of species from class Clostridia and class Bacilli in the microbiota.

C. difficile pathogenesis is heavily affected by carbohydrate, amino acid, and bile acid levels in the gastrointestinal tract, related to the metabolism of competitive commensals (Sorbara and Pamer, 2019). To identify metabolic pathways in other clostridia that might enable them to outcompete C. difficile, we defined metabolic potential in patient microbiomes using HUMAnN2 to quantify microbial pathway abundances. We found no significant differences in alpha- or beta-diversity between overall metabolic pathway composition in the two patient microbiome groups (Figure 2—figure supplement 1A,B; p = 0.2393, Wilcoxon rank sum and p = 0.054, PERMANOVA). Therefore, we trained an elastic net model to identify specific pathways associated with EIA status (Figure 2A). We found a number of carbohydrate degradation pathways and amino acid biosynthetic pathways associated with the asymptomatically-colonized (Cx+/EIA-) patients, including sucrose degradation III and fucose and rhamnose degradation. Investigation of the genera that encode such pathways revealed that the sucrose degradation III pathway was increased in asymptomatic patients, largely due to Blautia spp. and Faecalibacterium spp. of the class Clostridia (Figure 2B). Interestingly, the fucose and rhamnose degradation pathways were entirely defined by Escherichia spp., presumably E. coli. This suggests that metabolic functions such as fucose and rhamnose degradation may be confined to a smaller number of taxa than carbohydrate degradation pathways such as sucrose degradation. Using the HUMAnN2 (Franzosa et al., 2018) gene family information, we used linear mixed modeling to identify carbohydrate-active enzymes differentially associated with EIA status (Figure 2C). Supporting the pathway analysis, we found an increased abundance of a subset of glycoside hydrolase genes, specifically involved in sucrose and starch metabolism in the asymptomatically colonized patients. Our metabolic pathway analyses highlight differentially abundant carbohydrate degradation processes in clostridial taxa that could contribute to colonization resistance against C. difficile in patient microbiomes.

Figure 2. Carbohydrate metabolic processes present in asymptomatic patient microbiome.

(A) Significant pathways associated with EIA status, as derived from the elastic net model. Mean-prediction AUC for the elastic net model was 0.825. (B) Relative abundance of taxa in pathways associated with asymptomatic patients, where each patient’s metagenomic relative abundance is depicted by a single bar. Bars are colored by genera predicted to encode pathway. (C) Relative abundance of glycosidic hydrolase genes significantly associated with EIA status (q-value of <0.25) in stool metagenomes, where circles represent KEGG pathway classification.

Figure 2.

Figure 2—figure supplement 1. Compositional measurements of metabolic pathways and metabolites.

Figure 2—figure supplement 1.

(A) Alpha-diversity (measured by richness) of metabolic pathways in patient groups. (B) Non-metric multi-dimensional scaling (NMDS) ordination of Bray-Curtis dissimilarity measurements between stool microbiome metabolic pathways. (C) NMDS ordination of Euclidean distances between stool metabolomes. A permutation test was used to quantify centroid distances between groups (p = 0.01).
Open in a new tab

We hypothesized that differences in metabolic potential of fecal microbiome communities might be reflected in metabolomic profiles, and therefore sought to identify metabolites that are altered in CDI patients relative to those asymptomatically colonized with C. difficile (Robinson et al., 2019). Ordination of Euclidean distances between Cx+/EIA- and Cx+/EIA+ stool metabolomes revealed no significant differences in metabolome structure (Figure 2—figure supplement 1C, PERMANOVA = 0.426). We again used MaAslin2 to determine metabolites associated with each disease state. Consistent with previous analysis, a number of end-product Stickland fermentation metabolites (4-methypentanoic acid and 5-aminovalerate) were associated with CDI patients. While 4-hydroxyproline was the strongest predictor of asymptomaticallycolonized patients, many of the significant metabolites that were associated with asymptomatic patients were predicted to be carbohydrates (Figure 3A, FDR < 0.25; Supplementary file 3). Putative metabolite identities were initially annotated by matching metabolite spectra to the NIST14 GC-MS spectral library. The preponderance of carbohydrates in asymptomatically colonized patients and the substantial similarity of carbohydrate spectra prompted us to rigorously validate the identities of these metabolites by comparing EI spectra and GC retention times against authentic standards, where commercially available (Supplementary file 3, Figure 3—figure supplement 1). These data reveal a carbohydrate signature that is depleted in CDI patients. Notably, fructose and rhamnose are either substrates or products of the sucrose degradation III and fucose and rhamnose degradation pathways, which we found to be enriched in asymptomatically colonized patients. The co-occurrence of these microbial pathways and their corresponding metabolites in asymptomatically colonized patients suggests that a commensal carbohydrate catabolism may contribute to suppression of C. difficile pathogenesis.

Figure 3. Asymptomatically colonized patients are defined by carbohydrate species.

(A) Significant metabolites associated with EIA state, where a positive coefficient is strongly associated with Cx+/EIA+ metabolomes. Colors indicate human metabolic database (HMDB) sub-classification, and size of circle corresponds to magnitude of statistical significance. * indicates closest potential match; ** indicates two peaks from the same compound; contaminant indicates mass spectrometry contaminant. (B) Clinical and reference strains grown in C. difficile minimal medium (CDMM) with equimolar amounts of carbohydrate sources added. Growth was measured by taking maximum absorbance values over 48 hr. Each point represents the mean of two technical replicates of a unique isolate (top). Growth of C. difficile 630, from same conditions as above (bottom).

Figure 3—source data 1. Growth curve data for C. difficile isolates.
elife-72801-fig3-data1.xlsx (101.8KB, xlsx)

Figure 3.

Figure 3—figure supplement 1. Validation of significantly associated metabolites.

Figure 3—figure supplement 1.

(A) EI spectra and GC retention times of identified sugar and sugar alcohol features (red) compared to spectra of authentic standards (blue). (B) EI spectrum of 4-hydroxyproline feature compared to NIST14 library spectrum.
Open in a new tab

Our examination of taxa, metabolic pathways, and metabolites revealed a number of carbohydrates which we predict are undigestible by C. difficile or are end-products of a more complex commensal metabolism that is exclusionary to C. difficile. Using a set of clinical C. difficile isolates cultured from this patient cohort (8 isolates representing six different ribotypes), we examined growth of C. difficile on carbohydrates associated with asymptomatically colonized patients. Using a defined minimal media (CDMM)( Karasawa et al., 1995) to test nutrient utilization, we found that C. difficile isolates grew robustly on fructose as expected (median maximum A600 of 0.90), but did not proliferate on rhamnose or lactulose (median maximum A600 of 0.19 and 0.24, respectively). Notably, in the case of sorbitol, we found that a subset of strains, including the reference strain C. difficile 630 and C. difficile VPI10643, grew to a maximum A600 of greater than 0.47 (Figure 3B). Given that we had found sucrose degradation as a metabolic pathway enriched in asymptomatically colonized patients, we hypothesized that C. difficile would be unable to use this carbohydrate. Indeed, when grown on sucrose as the sole carbon source, strains achieved a median maximum A600 ~4.7-fold less than that of growth on fructose. C. difficile’s restricted carbohydrate metabolism, coupled with the presence of commensal Clostridia could hamper progression to CDI.

We hypothesized that the differential abundance of identified stool metabolites in these patient cohorts is related to the metabolism of specific microbes or host processes. We performed a sparse partial least-squares-discriminatory analysis (sPLS-DA) with the mixOmics package to define relationships between the most predictive features of patient metabolomes and microbiomes. We optimized the number of latent components (Figure 4—figure supplement 1A) and number of variables (Figure 4—figure supplement 1B). Our final model contained two latent components, with the first one composed of 15 metabolites and 25 microbial species. Of the largest metagenomic variable weights, four out of five species (C. difficile, a Lachnospiraceae spp., Anaerostipes hadrus, and Clostridium clostridioforme) were also significantly associated with an EIA state (Figure 1). Of the metabolomic variable weights (Figure 4A), the 10 highest-weighted metabolites were also discovered by previous analyses (Figure 2). The predictive value of each of the components per block was greater that an area under the curve (AUC) of 0.85, with the second metagenomic block component having the best performance (AUC = 0.94, Figure 4—figure supplement 1C). The strong performance of the latent components in classifying samples via EIA status validated our previous findings. Using the variables defining the first latent component, we performed correlational analyses (Figure 4B) and found a number of striking correlations. C. difficile abundance was positively correlated with a number of well-known Stickland metabolites (5-amino-valeric acid and 4-methylpentanoic acid, rho = 0.48 and 0.36, respectively)(Robinson et al., 2019), whereas C. difficile had negative correlations with fructose, rhamnose, and hydroxyproline (rho = –0.27, 0.36, and –0.34, respectively). Given our metagenomic data suggesting that Kraken metagenomic profiling yielded more sensitive estimates of C. difficile abundance, we performed an independent multi-omics analysis on the same dataset using the Kraken metagenomic data. Using a similar process of model building as above, the final model consisted of two latent components, with 15 metabolites and 15 microbes in the first component (Figure 4—figure supplement 1C). C. difficile was also most positively correlated to 5-amino-valeric acid and 4-methylpentanoic acid, with corresponding negative correlations to fructose, rhamnose, and hydroxyproline (Figure 4—figure supplement 1D). These microbe-metabolite relationships highlight the known pathophysiology of CDI, and identify novel C. difficile-carbohydrate relationships that define asymptomatic colonization.

Figure 4. Multi-omics signature of C. difficile-colonized patients reveals C.difficile-metabolite relationships.

(A) Correlation circle indicating the contribution of each variable (microbe or metabolite) to latent component of sparse partial least-squares-discriminatory analysis (sPLS-DA) using MetaPhlAn2 data. (B) Heatmap of Spearman correlations between metagenomic and metabolomic variables from the first latent component using MetaPhlAn2 data.

Figure 4.

Figure 4—figure supplement 1. Multi-omics analysis performance.

Figure 4—figure supplement 1.

(A) Examination of error rate dependent on number of components included in model construction, related to Figure 4. (B) Balanced error rate dependent on the number of features used in each block of each component; the left number indicates the number of metagenomic features used, and the right number indicates the number of metabolomic features used. (C) ROC curves with area under the curve measurements indicated for each block and component of the MetaPhlAn2 model. (D) Correlation circle indicating the contribution of each variable (microbe or metabolite) to latent component of sparse partial least-squares-discriminatory analysis (sPLS-DA) using Kraken data. (E) Heatmap of Spearman correlations between metagenomic and metabolomic variables from the first latent component using Kraken data.
Open in a new tab

Given the anticorrelation between C. difficile and rhamnose, we sought to explain the enrichment of this carbohydrate in asymptomatically colonized patients. Though C. difficile cannot grow on rhamnose as the sole carbohydrate, in other organisms rhamnose has substantial transcriptional influence over carbon catabolite gene clusters (Egan and Schleif, 1993; Hirooka et al., 2015). We wanted to rule out the possibility that rhamnose may impact C. difficile through possibly cryptic transcriptional reprogramming, perhaps contributing to C. difficile repression in vivo. Accordingly, we performed whole transcriptome RNA sequencing on C. difficile cultures exposed to a metabolizable substrate, fructose, or a non-metabolizable substrate, rhamnose. In the presence of fructose, we found 555 genes significantly altered (adjusted p-value < 0.05 and |fold-change| > 2) (Supplementary file 4). Some of the most altered genes were indicative of carbon catabolite repression of sugar transport and upregulation of glycolytic processes to metabolize fructose. In contrast, we found only three genes significantly increased in the rhamnose condition. The lack of striking systems-level or targeted (toxin expression, sporulation) regulation by rhamnose, and C. difficile’s inability to utilize it, leads us to conclude that its association with asymptomatically colonized patients’ microbiomes is not through direct interaction or suppression of C. difficile. Instead, we speculate that rhamnose may be the byproduct of a complex commensal metabolism of other dietary polysaccharide substrates, which could exclude C. difficile from the GI tract.

Discussion

Factors affecting the outcome of C. difficile colonization

Susceptibility to CDI is the result of a complex interaction between host factors (variation in bile acid metabolism, adaptive immunity) and abiotic factors such as antibiotic treatment and diet (Mullish and Allegretti, 2021; Littmann et al., 2021). These variables largely affect colonization resistance in the gut microbiome community and influence pathogen proliferation (germination rate, variation in toxin activity, and metabolic capacity). Our study endeavored to identify gut microbiome signatures (both taxonomic and metabolic), bacteriologic traits, and antibiotic exposure histories that might help explain Cx+/EIA- C. difficile colonization. Clinically, this manifestation is an intermediate state on the spectrum of C. difficile- associated disease and correspondingly, a diagnostic conundrum. One limitation of this study is our inability to assess dietary histories of patients leading up the diagnostic event. The metabolomic data provides a snapshot in time. While we hypothesize the increase in monosaccharides is due to an increase in carbohydrate degradation within the community, it is unclear whether the carbohydrate signature is due to microbial community differences in cross-feeding rates or differences in host diet.

Another important limitation to this study is our inability to control for C. difficile strain differences and correspondingly, heterogeneity in processes such as spore germination, nutrient utilization, and toxin expression in vivo (Kumar et al., 2019; Hunt and Ballard, 2013). Strains infecting Cx+/EIA+ patients were more likely to contain the cdtAB toxin locus, and the distribution of ribotypes was qualitatively different between the two cohorts (indicating significant pathogen variation). In vitro examination of toxin production (TcdA and TcdB) using a commercial ELISA indicated that over half of isolates expressed detectable levels of toxins. Toxin expression is well-known to be regulated by nutrient conditions and although our in vitro data indicate that both cohorts contain similar numbers of strains capable of producing toxin in vitro, such conditions are considered inadequate to predict in vivo levels of toxin production (Burnham and Carroll, 2013; Akerlund et al., 2006). Further, we found that a diverse set of clinical C. difficile strains might have variation in their ability to utilize nutrients such as sorbitol, which contrasts with reports of model C. difficile strains harboring more flexibility in their ability to utilize nutrients (Theriot et al., 2014; Jenior et al., 2017; Scaria et al., 2014). The outcome of strain level differences in metabolism and virulence is further complexified by in vivo conditions that might influence pathogen proliferation. Yet, we speculate that certain gastrointestinal environments both encourage some growth of C. difficile and discourage the elaboration of toxin, as toxin expression is actively repressed in nutrient-rich conditions.

Antibiotic treatment is the most well-understood risk factor for CDI (Stevens et al., 2011; Deshpande et al., 2013), and antibiotic exposure in our cohort likely results in loss of the species we find depleted from CDI patients. Here, we confirm that exposure to a number of antibiotics is associated with CDI patients, including cephalosporins (significantly associated) and intravenous vancomycin (weakly associated). Clindamycin and quinolones, two antibiotics also associated with CDI in other human cohorts (Teng et al., 2019) are likely not significantly associated in our population due to the low prevalence of their exposure. Our microbiome data reveals decreased levels of Streptococcus, Ruminococcus, and Eubacterium spp. in CDI patients. Findings from both human cohorts and mouse models of antibiotic treatment indicate that a number of clostridial taxa are depleted upon administration of a variety of antibiotic treatments (Palleja et al., 2018; Rashid et al., 2015). It is also posited that some of these taxa are integral to protection from CDI (Mills et al., 2018). Given the attempts to use FMTs or Firmicutes-enriched probiotics to prevent CDI, we hypothesize that the restoration of lost species from class Clostridia after high-risk antibiotic treatment could be a novel avenue for CDI prevention (McGovern et al., 2021).

Gut metabolites as markers of C. difficile proliferation and the microbiome

While metabolites associated with CDI and correlated with C. difficile abundance (4-methyl-pentanoic acid and 5-amino-valeric acid) clearly reflect C. difficile proliferation (Akerlund et al., 2006), the metabolites associated with Cx+/EIA- patients could reflect a number of non-mutually exclusive biological scenarios, indicating either the absence of C. difficile proliferation or the presence of a stable community where C. difficile pathogenesis is prevented by community metabolic elements.

In the one scenario, we reference two metabolites, 4-hydroxyproline and sorbitol, which have been considered host products of collagen degradation and inflammation (Fletcher et al., 2021; Pruss and Sonnenburg, 2021). The abundance of 4-hydroxyproline in the stools of Cx+/EIA- and its anticorrelation with C. difficile levels would suggest that it is a substrate consumed by C. difficile during pathogenesis. In a mouse model of CDI, sugar alcohols and amino acids observed before infection were considered representative of a ‘pre-colonized state’ (Fletcher et al., 2018; Theriot et al., 2014), as these nutrients declined as CDI progressed. However, we restricted our cohort to patients who were not on their way to developing CDI, by excluding patients with EIA- stool if they were subsequently diagnosed with CDI or received empiric CDI treatment within 10 days of initial stool collection (Dubberke et al., 2018).

In another scenario, the overlap of signatures between pathways, metabolites, and microbes highlights a number of possible metabolic pathways that might be exclusionary to C. difficile, namely starch/sucrose degradation and rhamnose degradation. The combination of our microbiome data, which shows enrichment of number of commensal Clostridia such as Eubacterium spp.(Desai et al., 2016), starch/sucrose degradation pathways, and our in vitro data highlights a possible microbe-metabolite combination that could prevent C. difficile proliferation. Rhamnose is a major component of plant and some bacterial cell-wall polysaccharides (Silva et al., 2020). Metabolic pathway profiling revealed an enrichment of fucose and rhamnose degradation pathways in asymptomatically colonized patients, represented by Enterobacterales taxa. Therefore, we propose that the detected rhamnose is a byproduct of commensal catabolism of more complex polysaccharides containing rhamnose (Porter and Martens, 2017; Mistou et al., 2016). These findings are of course limited by the scope of the in vitro experiment and the correlative nature of our microbiome data. Future work examining in vivo competition between diverse C. difficile isolates and commensal isolates with critical metabolic elements would be required.

Lactulose was a carbohydrate associated with asymptomatically-colonized patients and not a robust growth substrate for C. difficile. Interestingly, lactulose has been previously associated with a decrease in C. difficile-related diarrhea (Maltz et al., 2020) and decreased risk of CDI (Maltz et al., 2020; Agarwalla et al., 2017). Lactulose is a disaccharide product from heat treatment of lactose (a common sugar in dairy products), but it is also a component of some laxatives (Adachi, 1958). However, patients were screened and excluded from this cohort if they were prescribed laxatives in the 24 hr prior to sample collection. In addition to this screening/exclusion criteria, lactulose is almost exclusively prescribed to liver failure patients (there were none reported in this study), thus it is more likely to be present from consumption of heated milk (containing lactose). Other in vitro work demonstrates that addition of ‘non-digestible’ oligosaccharides, such as lactulose, provides a competitive advantage to Bifidobacterium spp. over C. difficile (Kondepudi et al., 2012; Hopkins and Macfarlane, 2003). While we do not recommend lactulose, a laxative, as such a prebiotic, there are a number of other ‘non-digestible’ oligosaccharides that might serve similar purposes in future interventions (Hopkins and Macfarlane, 2003). Taken together, these data emphasize the potential for synthetic or natural prebiotic interventions to shift a vulnerable microbiota away from CDI.

Strategies to ameliorate toxigenic C. difficile proliferation

Our multi-omics analyses of a colonized asymptomatic patient population support a growing body of literature concerning commensal metabolism as a tool against C. difficile. Evidence from both mouse models of disease and human studies indicate that administration of polysaccharides or ‘microbial accessible carbohydrates’ may prevent C. difficile proliferation or decrease the risk of CDI (Mefferd et al., 2020; Schnizlein et al., 2020; Hryckowian et al., 2018; Maltz et al., 2020; Lewis et al., 2005). Recently, a probiotics-based attempt to design a consortium of mucosal sugar utilizers revealed its ability to decrease C. difficile colonization in vivo (Pereira et al., 2020), indicating that increasing mucosal metabolism, or carbohydrate catabolism, may be another route to strengthening commensal resistance to C. difficile. Interestingly, previous attempts to combinatorically assemble species and nutrient combinations that might inhibit C. difficile indicate that success is afforded by species able to competitively utilize carbohydrates such as sorbitol and mannitol (Ghimire et al., 2020). Given the plethora of prebiotics and probiotics explored in the C. difficile field, we emphasize the need for an approach that harnesses both probiotic- and prebiotic-based components to inhibit the proliferation of C. difficile and toxin-mediated pathogenesis.

Materials and methods

Patient cohort analysis

A previous retrospective cohort study was conducted to understand C. difficile colonization. In that study, C. difficile isolates were cultured from patient stool as described. Ribotyping was performed using the DiversiLab Bacterial Barcodes software (bioMerieux) (Dubberke et al., 2018; Westblade et al., 2013). Analysis of isolate genetic traits and in vitro toxin production was performed on the 102 isolates for which we had corresponding metagenomic sequencing data (see below). Data concerning isolate ribotype was aggregated into the three most abundant ribotypes (ribotype 027, ribotype 106, ribotype 14/20), where all other ribotypes or unclassified strains were grouped into ‘Other’. For the purposes of this study, data concerning inpatient antibiotic orders were retrospectively collected from the electronic medical informatics database for patients with toxin EIA positive (Cx+/EIA+) stool (n = 62) or toxin EIA negative (Cx+/EIA-) stool (n = 62). The presence of antibiotic orders was classified into three dichotomous groups by timing of exposure: antibiotics in 0–7 days before stool collection (1 week), antibiotics in >7–14 days before stool collection (2 weeks), and antibiotics in >14–30 days before stool collection (1 month). To understand the specific antibiotics associated with EIA status in our patient cohort, raw antibiotic exposure data was aggregated by time. Additionally, low-prevalent antibiotics ( < 10% exposure in patients) were removed from analysis. Logistic regression analysis was performed using the glm function in R. To understand overall antibiotic exposure as it relates to EIA status, any antibiotic exposure was considered ‘1’ and zero antibiotic exposure in a patient was considered ‘0’. The binary antibiotic exposure variable was then used in linear mixed modeling analysis to understand species associated with antibiotic exposure.

Metagenomic sequencing and analysis of patient stool

Metagenomic DNA was extracted from patient stools as previously described (Fishbein et al., 2021a). C. difficile was isolated from patient stools as previously described (Fishbein et al., 2021a). Illumina libraries of patient stool metagenomic DNA were prepared and pooled as previous described (Fishbein et al., 2021a; Baym et al., 2015). Fecal metagenomic libraries were submitted for 2 × 150 bp paired-end sequencing on an Illumina NextSeq High-Output platform. Reads were binned by index sequences and reads were trimmed and quality filtered using Trimmomatic v.0.38 (Bolger et al., 2014) to remove adapter sequences and DeconSeq (Schmieder and Edwards, 2011) to remove human sequences. Samples that were less than 15% bacterial DNA during initial sequencing were discarded, and all samples were sequenced to a depth of at least 5 million reads. Sample loss due to low bacterial DNA resulted in a smaller cohort than originally reported (Dubberke et al., 2018), with the final set of metagenomes representing 54 Cx+/EIA- and 48 Cx+/EIA+ patients.

We performed taxonomic profiling of metagenomic sequences using MetaPhlAn2 (Truong et al., 2015), and functional pathway profiling using HUMAnN2 (Franzosa et al., 2018). MetaCyc pathway abundances were normalized to relative abundances using the humann2_renorm.py function. The humann2_barplot.py function was used to assess taxonomic composition of metabolic pathways. Custom python scripts were used to parse MetaPhlAn2 ‘_profiled_metagenome.txt’ and HUMAnN2 ‘pathwayabundance.txt’ files. Data were imported to R to analyze community composition and differential associations. To analyze carbohydrate-active enzymes, we used humann2_regroup.py and humann2_rename.py function to reannotate the ‘_genefamilies.txt’ files and identify genes with the enzyme classification number EC:3.2.1.*, representing glycosidases, enzymes that participate in carbohydrate degradation (Ghimire et al., 2020).

Metagenomic data analysis

For both microbiome and metabolomic data, the nearZeroVar function of the caret package was used to remove low-prevalent or invariant taxa/pathways/metabolites (Kuhn, 2008). These filtered data sets were analyzed for differential association and multi-omics modeling. Alpha-diversity and beta-diversity were calculated using the vegan package. Weighted UniFrac distance was used as a beta-diversity metric for microbial taxa and Bray Curtis dissimiliarity was used as a beta-diversity metric for metabolic pathways, while Euclidean distance was used as a beta-diversity metric for metabolomes. The MaAslin2 package was used for linear mixed modeling to identify microbial taxa, gene families, and metabolites associated with EIA/antibiotic exposure status.

To analyze HUMAnN2 pathways enriched in cohorts, we used statistical inference of associations between microbial communities and host phenotypes (SIAMCAT) (Wirbel et al., 2021), using the siamcat package in R, to fit an elastic net model to the data. We used the following parameters: log.std normalization, 10 folds and 10 resamples for data splitting. The model.interpretation.plot function was used to display features weights for features used in >70% of models generated in training.

Determination of candidate metabolites

Putative identification of metabolites of interest (Supplementary file 3) was initially performed through spectral matching against the NIST14 electron ionization spectrum library. Several features were previously identified by our group (see Robinson et al., 2019). Features predicted to be sugars or sugar alcohols were compared to a panel of authentic standards (D-sorbitol, D-mannitol, D-fructose, L-rhamnose, L-fucose, lactulose, glucose, mannose, D-galactose, D-talose, myo-inositol, and L-sorbose). Because isomeric sugars generate very similar spectra, we utilized both spectral similarity and retention time to identify sugar metabolites (Figure 3—figure supplement 1).

Multi-omics analysis

The metagenomic relative abundance data was imputed with min(abundance >0)/2, and the metabolomic data was imputed with a value of 1. For both filtered datasets, a centered log-ratio transformation was used to analyze filtered metagenomic and metabolomic datasets above. The mixOmics (Rohart et al., 2017) package in R was used for multi-omics analysis of both MetaPhlAn and Kraken metagenomic relative abundance data. To avoid over-fitting on the large number of variables in our datasets, we utilized sPLS-DA. Briefly, to determine the number of variables from each dataset to keep in the final model, we estimated model error rates for all combinations of seq(15,30,5) variables for both metagenomic and metabolomic datasets, using the function tune.block.splsda (10-fold cross-validation, repeated 50 times, “max.dist” distance metric). Spearman correlations were calculated between CLR-transformed microbial taxa and metabolite abundances, from the variables defining the first latent components, and plotted using the cim package.

Bacteriology and in vitro growth assays

C. difficile strains were isolated from patient stools by plating on cycloserine-cefoxitin fructose agar as previously described; strains were stored at –80°C (Fishbein et al., 2021a). C. difficile VPI10643 and C. difficile 630 reference strains were purchased from ATCC, and included in the assays described below using the same conditions as clinical isolates. For in vitro growth assays, CDMM was prepared as previously described (Karasawa et al., 1995) and 20 mM of specified carbohydrates were added. Clinical isolates were inoculated into tryptone-yeast extract (TY) broth and grown for 16 hr, then washed with PBS and diluted 1:100 into media with different carbohydrates sources. Growth was measured in a shaking, 96-well plate at 37°C for 48 hours.

In vitro ELISAs to assess toxin production in each isolate were performed on using TGCbiomics kits for ‘Simultaneous detection of TcdA and TcdB’ and ‘C. difficile GDH detection kit’ as a control ELISA. Cultures were grown for 24 hr in TY media in deep 96-well plate. Following, cultures were spun down and culture supernatants were diluted 1:5 in dilution buffer and loaded onto ELISA plates for detection of both toxin and control protein (GmbH), per manufacturer’s instructions. Isolates were considered positive for toxin if they had greater absorbance than that of the negative control.

RNA sequencing and data analysis

Five mL of each strain (in biological triplicate) were grown to log-phase (OD600 ~0.4)in TY and exposed to TY- rhamnose or TY-fructose (with each carbohydrate at 30 mM). Cells were harvested by adding one volume of 1:1(v/v) acetone/ethanol to the culture to arrest growth and RNA degradation. Sample were spun at 4000 x g for 5 min. The cell pellet was washed with 500 µl TE buffer (0.5 M EDTA, 1 M Tris pH 7.4) and spun down to remove the supernatant. The cell pellet was resuspend in one mL Trizol and two rounds of bead-beating at 4500 rpm for 45 s were performed. A total of 300 µl of chloroform was added to the suspension, lysates were vortexed, and centrifuged at 4000 rpm for 10 min at 4°C. The aqueous layer was removed and RNA was precipitated using isopropanol, washed with 70% ethanol, and resolubilized in TE buffer. Total RNA was treated with Turbo DNase (for two rounds of digestion). rRNA depletion was performed using the QiaFastSelect kit (Hilden, Germany), following manufacturer’s instructions. Libraries were prepared using the rRNA-depleted RNA as input for NEBNext Ultra II RNA Library Prep Kit (NEB, Ipswich, MA). Libraries were pooled and submitted for 2 × 150 bp paired-end sequencing on an Illumina NextSeq High-Output platform at the Center for Genome Sciences and Systems Biology at Washington University in St. Louis.

Raw reads were trimmed using Trimmomatic v. 0.38, and aligned to a C. difficile VPI10643 reference genome (GCF_000155025.1) using Bowtie2. SAM files were converted to BAM format and indexed using samtools. Read counts for each gene feature were obtained using the featureCounts function of subread-1.6.5 package. Counts were manually imported into R, and DEseq2 was used to identify differentially expressed gene products in the case of TY-fructose relative to TY and TY-rhamnose relative to TY.

Data deposition

Metagenomic reads were deposited under BioProject accession number PRJNA748262 and RNA sequencing reads were deposited under BioProject accession number PRJNA748261. All R code and metadata used to generate figures is deposited at https://github.com/srsfishbein/2021EIACdiff_multiomics, (Fishbein, 2021b copy archived at swh:1:rev:0c2a33d873e43194afb5818733e46c6ff28d6947).

Acknowledgements

The authors are grateful for members of the Dantas lab for their helpful feedback on the data analysis and preparation of the manuscript. The authors are specifically grateful to Drew J Schwartz for his insightful feedback. The authors would also like to thank the Edison Family Center for Genome Sciences and Systems Biology staff, Eric Martin, Brian Koebbe, MariaLynn Crosby, and Jessica Hoisington-López for their expertise and support in in sequencing/data analysis.

This work was supported in part by awards to ERD from the CDC Broad Agency Announcement, contract 200-2017-96178). JPH was supported by CDC (Broad Agency Announcement, contract (Broad Agency Announcement, contract 200-2019-05950) and the National Institute of Diabetes, Digestive, and Kidney Diseases of the National Institutes of Health (RO1DK111930). GD received support from the National Center for Complementary and Integrative Health (NCCIH: https://nccih.nih.gov/) of the National Institute of Health (NIH) under award number R01AT009741; the National Institute for Occupational Safety and Health (NIOSH: https://www.cdc.gov/niosh/index.htm) of the US Center for Disease Control and Prevention (CDC) under award number R01OH011578l, and the Congressionally Directed Medical Research Program (CDMRP: https://cdmrp.army.mil/prmrp/default) of the US Department of Defense DOD under award number W81XWH1810225. SRSF is supported by the T32 Pediatric Gastroenterology Research Training Program under the National Institute of Child Health and Human Development (NICHD: https://www.nicdhd.nih/gov) of the NIH under award number T32DK077653 (PI: P.I. Tarr).

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Contributor Information

Erik R Dubberke, Email: edubberk@wustl.edu.

Gautam Dantas, Email: dantas@wustl.edu.

Peter J Turnbaugh, University of California, San Francisco, United States.

Gisela Storz, National Institute of Child Health and Human Development, United States.

Funding Information

This paper was supported by the following grants:

  • National Institute of Diabetes and Digestive and Kidney Diseases RO1DK111930 to Jeffrey P Henderson.

  • Centers for Disease Control and Prevention 200-2017-96178 to Erik R Dubberke.

  • National Center for Complementary and Integrative Health R01AT009741 to Gautam Dantas.

  • National Institute for Occupational Safety and Health R01OH011578l to Gautam Dantas.

  • Congressionally Directed Medical Research Programs W81XWH1810225 to Gautam Dantas.

  • Eunice Kennedy Shriver National Institute of Child Health and Human Development T32 HD004010 to Skye RS Fishbein.

Additional information

Competing interests

No competing interests declared.

E.R.D. has received research support from Synthetic Biologics, Pfizer and Ferring, and has been a consultant for Summit, Merck, Ferring, Pfizer and Seres Therapeutics, all unrelated to this study.

Author contributions

Conceptualization, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing – original draft, Writing – review and editing.

Formal analysis, Investigation, Methodology, Validation, Visualization, Writing – review and editing.

Data curation, Methodology.

Data curation, Methodology, Project administration, Writing – review and editing.

Investigation, Writing – review and editing.

Data curation, Investigation, Supervision, Writing – review and editing.

Conceptualization, Funding acquisition, Investigation, Methodology, Supervision, Writing – review and editing.

Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Resources, Supervision, Writing – review and editing.

Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Validation, Visualization, Writing – review and editing.

Additional files

Supplementary file 1. Patient demographic data.
elife-72801-supp1.xlsx (9.7KB, xlsx)
Supplementary file 2. Fecal metagenomics metadata file with isolate information.
elife-72801-supp2.xlsx (12.5KB, xlsx)
Supplementary file 3. MaAsLin2 output of metabolites associated with EIA status in addition to metabolite validation information.
elife-72801-supp3.xlsx (18.4KB, xlsx)
Supplementary file 4. DEseq output of in vitro rhamnose-exposed C difficile transcriptomic profiling.
elife-72801-supp4.xlsx (714.2KB, xlsx)
Transparent reporting form

Data availability

Metagenomics reads were deposited under BioProject accession number PRJNA748262 and RNA sequencing reads were deposited under BioProject accession number PRJNA748261.

The following dataset was generated:

Fishbein SRS. 2021. Fecal metagenomes of C. difficile colonized patients. NCBI BioProject. PRJNA748262

Fishbein SRS. 2021. C. difficile carbohydrate transcriptomics. NCBI BioProject. PRJNA748261

References

  1. Adachi S. Formation of lactulose and tagatose from lactose in strongly heated milk. Nature. 1958;181:840–841. doi: 10.1038/181840a0. [DOI] [PubMed] [Google Scholar]
  2. Agarwalla A, Weber A, Davey S, Hamilton K, Goldberg D, Rhim AD, Yang YX. Lactulose Is Associated With Decreased Risk of Clostridium difficile Infection in Decompensated Cirrhosis. Clinical Gastroenterology and Hepatology. 2017;15:953–954. doi: 10.1016/j.cgh.2017.01.012. [DOI] [PubMed] [Google Scholar]
  3. Akerlund T, Svenungsson B, Lagergren A, Burman LG. Correlation of disease severity with fecal toxin levels in patients with Clostridium difficile-associated diarrhea and distribution of PCR ribotypes and toxin yields in vitro of corresponding isolates. Journal of Clinical Microbiology. 2006;44:353–358. doi: 10.1128/JCM.44.2.353-358.2006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Battaglioli EJ, Hale VL, Chen J, Jeraldo P, Ruiz-Mojica C, Schmidt BA, Rekdal VM, Till LM, Huq L, Smits SA, Moor WJ, Jones-Hall Y, Smyrk T, Khanna S, Pardi DS, Grover M, Patel R, Chia N, Nelson H, Sonnenburg JL, Farrugia G, Kashyap PC. Clostridioides difficile uses amino acids associated with gut microbial dysbiosis in a subset of patients with diarrhea. Science Translational Medicine. 2018;10:464. doi: 10.1126/scitranslmed.aam7019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Baym M, Kryazhimskiy S, Lieberman TD, Chung H, Desai MM, Kishony R. Inexpensive multiplexed library preparation for megabase-sized genomes. PLOS ONE. 2015;10:e0128036. doi: 10.1371/journal.pone.0128036. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Bolger AM, Lohse M, Usadel B. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics. 2014;30:2114–2120. doi: 10.1093/bioinformatics/btu170. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Burnham C-AD, Carroll KC. Diagnosis of Clostridium difficile infection: an ongoing conundrum for clinicians and for clinical laboratories. Clinical Microbiology Reviews. 2013;26:604–630. doi: 10.1128/CMR.00016-13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Chen K, Zhu Y, Zhang Y, Hamza T, Yu H, Saint Fleur A, Galen J, Yang Z, Feng H. A probiotic yeast-based immunotherapy against Clostridioides difficile infection. Science Translational Medicine. 2020;12:567. doi: 10.1126/scitranslmed.aax4905. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Cowardin CA, Buonomo EL, Saleh MM, Wilson MG, Burgess SL, Kuehne SA, Schwan C, Eichhoff AM, Koch-Nolte F, Lyras D, Aktories K, Minton NP, Petri WA. The binary toxin CDT enhances Clostridium difficile virulence by suppressing protective colonic eosinophilia. Nature Microbiology. 2016;1:16108. doi: 10.1038/nmicrobiol.2016.108. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Crobach MJT, Vernon JJ, Loo VG, Kong LY, Péchiné S, Wilcox MH, Kuijper EJ. Understanding Clostridium difficile Colonization. Clinical Microbiology Reviews. 2018;31:e00021-17. doi: 10.1128/CMR.00021-17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Desai MS, Seekatz AM, Koropatkin NM, Kamada N, Hickey CA, Wolter M, Pudlo NA, Kitamoto S, Terrapon N, Muller A, Young VB, Henrissat B, Wilmes P, Stappenbeck TS, Núñez G, Martens EC. A Dietary Fiber-Deprived Gut Microbiota Degrades the Colonic Mucus Barrier and Enhances Pathogen Susceptibility. Cell. 2016;167:1339–1353. doi: 10.1016/j.cell.2016.10.043. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Deshpande A, Pasupuleti V, Thota P, Pant C, Rolston DDK, Sferra TJ, Hernandez AV, Donskey CJ. Community-associated Clostridium difficile infection and antibiotics: a meta-analysis. The Journal of Antimicrobial Chemotherapy. 2013;68:1951–1961. doi: 10.1093/jac/dkt129. [DOI] [PubMed] [Google Scholar]
  13. Dubberke ER, Olsen MA. Burden of Clostridium difficile on the healthcare system. Clinical Infectious Diseases. 2012;55 Suppl 2:S88–S92. doi: 10.1093/cid/cis335. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Dubberke ER, Reske KA, Hink T, Kwon JH, Cass C, Bongu J, Burnham CAD, Henderson JP. Clostridium difficile colonization among patients with clinically significant diarrhea and no identifiable cause of diarrhea. Infection Control and Hospital Epidemiology. 2018;39:1330–1333. doi: 10.1017/ice.2018.225. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Egan SM, Schleif RF. A regulatory cascade in the induction of rhaBAD. Journal of Molecular Biology. 1993;234:87–98. doi: 10.1006/jmbi.1993.1565. [DOI] [PubMed] [Google Scholar]
  16. Eze P, Balsells E, Kyaw MH, Nair H. Risk factors for Clostridium difficile infections - an overview of the evidence base and challenges in data synthesis. Journal of Global Health. 2017;7:010417. doi: 10.7189/jogh.07.010417. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Faith DP, Baker AM. Phylogenetic diversity (PD) and biodiversity conservation: some bioinformatics challenges. Evolutionary Bioinformatics Online. 2007;2:121–128. [PMC free article] [PubMed] [Google Scholar]
  18. Fishbein SRS, Hink T, Reske KA, Cass C, Struttmann E, Iqbal ZH, Seiler S, Kwon JH, Burnham CA, Dantas G, Dubberke ER. Randomized Controlled Trial of Oral Vancomycin Treatment in Clostridioides difficile-Colonized Patients. MSphere. 2021a;6:e00936-20. doi: 10.1128/mSphere.00936-20. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Fishbein SRS. 2021EIACdiff_multiomics. swh:1:rev:0c2a33d873e43194afb5818733e46c6ff28d6947Software Heritage. 2021b https://archive.softwareheritage.org/swh:1:dir:bc7e2445e6c16ffcedd04e6c0566d7fd5e3cd192;origin=https://github.com/srsfishbein/2021EIACdiff_multiomics;visit=swh:1:snp:e3b72c68e69c68cee04167d5124e03f6d6413000;anchor=swh:1:rev:0c2a33d873e43194afb5818733e46c6ff28d6947
  20. Fletcher JR, Erwin S, Lanzas C, Theriot CM. Shifts in the Gut Metabolome and Clostridium difficile Transcriptome throughout Colonization and Infection in a Mouse Model. MSphere. 2018;3:e00089-18. doi: 10.1128/mSphere.00089-18. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Fletcher JR, Pike CM, Parsons RJ, Rivera AJ, Foley MH, McLaren MR, Montgomery SA, Theriot CM. Clostridioides difficile exploits toxin-mediated inflammation to alter the host nutritional landscape and exclude competitors from the gut microbiota. Nature Communications. 2021;12:462. doi: 10.1038/s41467-020-20746-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Franzosa EA, McIver LJ, Rahnavard G, Thompson LR, Schirmer M, Weingart G, Lipson KS, Knight R, Caporaso JG, Segata N, Huttenhower C. Species-level functional profiling of metagenomes and metatranscriptomes. Nature Methods. 2018;15:962–968. doi: 10.1038/s41592-018-0176-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Ghimire S, Roy C, Wongkuna S, Antony L, Maji A, Keena MC, Foley A, Scaria J. Identification of Clostridioides difficile-Inhibiting Gut Commensals Using Culturomics. Phenotyping, and Combinatorial Community Assembly. MSystems. 2020;5:e19. doi: 10.1128/mSystems.00620-19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Guh A, Mu Y, Winston LG, Johnston H, Olson D, Farley MM, Wilson LE, Holzbauer SM, Phipps EC, Dumyati GK, Beldavs ZG, Kainer MA, Karlsson M, Gerding DN. Emerging Infections Program Clostridioides difficile Infection Working, Trends in U.S. Burden of Clostridioides Difficile Infection and Outcomes. N Engl J Med. 2020;382:1320–1330. doi: 10.1056/NEJMoa1910215. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Hirooka K, Kodoi Y, Satomura T, Fujita Y. Regulation of the rhaEWRBMA Operon Involved in l-Rhamnose Catabolism through Two Transcriptional Factors, RhaR and CcpA, in Bacillus subtilis. Journal of Bacteriology. 2015;198:830–845. doi: 10.1128/JB.00856-15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Hopkins MJ, Macfarlane GT. Nondigestible oligosaccharides enhance bacterial colonization resistance against Clostridium difficile in vitro. Applied and Environmental Microbiology. 2003;69:1920–1927. doi: 10.1128/AEM.69.4.1920-1927.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hryckowian AJ, Pruss KM, Sonnenburg JL. The emerging metabolic view of Clostridium difficile pathogenesis. Current Opinion in Microbiology. 2017;35:42–47. doi: 10.1016/j.mib.2016.11.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Hryckowian AJ, Van Treuren W, Smits SA, Davis NM, Gardner JO, Bouley DM, Sonnenburg JL. Microbiota-accessible carbohydrates suppress Clostridium difficile infection in a murine model. Nature Microbiology. 2018;3:662–669. doi: 10.1038/s41564-018-0150-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Hunt JJ, Ballard JD. Variations in virulence and molecular biology among emerging strains of Clostridium difficile. Microbiology and Molecular Biology Reviews. 2013;77:567–581. doi: 10.1128/MMBR.00017-13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Jenior ML, Leslie JL, Young VB, Schloss PD. Clostridium difficile Colonizes Alternative Nutrient Niches during Infection across Distinct Murine Gut Microbiomes. MSystems. 2017;2:e00063-17. doi: 10.1128/mSystems.00063-17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Kachrimanidou M, Tsintarakis E. Insights into the Role of Human Gut Microbiota in Clostridioides difficile Infection. Microorganisms. 2020;8:E200. doi: 10.3390/microorganisms8020200. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Karasawa T, Ikoma S, Yamakawa K, Nakamura S. A defined growth medium for Clostridium difficile. Microbiology. 1995;141 (Pt 2):371–375. doi: 10.1099/13500872-141-2-371. [DOI] [PubMed] [Google Scholar]
  33. Kondepudi KK, Ambalam P, Nilsson I, Wadström T, Ljungh A. Prebiotic-non-digestible oligosaccharides preference of probiotic bifidobacteria and antimicrobial activity against Clostridium difficile. Anaerobe. 2012;18:489–497. doi: 10.1016/j.anaerobe.2012.08.005. [DOI] [PubMed] [Google Scholar]
  34. Kuhn M. Building Predictive Models in R Using the caret Package. Journal of Statistical Software. 2008;28:1–26. doi: 10.18637/jss.v028.i05. [DOI] [Google Scholar]
  35. Kumar N, Browne HP, Viciani E, Forster SC, Clare S, Harcourt K, Stares MD, Dougan G, Fairley DJ, Roberts P, Pirmohamed M, Clokie MRJ, Jensen MBF, Hargreaves KR, Ip M, Wieler LH, Seyboldt C, Norén T, Riley TV, Kuijper EJ, Wren BW, Lawley TD. Adaptation of host transmission cycle during Clostridium difficile speciation. Nature Genetics. 2019;51:1315–1320. doi: 10.1038/s41588-019-0478-8. [DOI] [PubMed] [Google Scholar]
  36. Laffin M, Millan B, Madsen KL. Fecal microbial transplantation as a therapeutic option in patients colonized with antibiotic resistant organisms. Gut Microbes. 2017;8:221–224. doi: 10.1080/19490976.2016.1278105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Lewis S, Burmeister S, Brazier J. Effect of the prebiotic oligofructose on relapse of Clostridium difficile-associated diarrhea: a randomized, controlled study. Clinical Gastroenterology and Hepatology. 2005;3:442–448. doi: 10.1016/s1542-3565(04)00677-9. [DOI] [PubMed] [Google Scholar]
  38. Littmann ER, Lee JJ, Denny JE, Alam Z, Maslanka JR, Zarin I, Matsuda R, Carter RA, Susac B, Saffern MS, Fett B, Mattei LM, Bittinger K, Abt MC. Host immunity modulates the efficacy of microbiota transplantation for treatment of Clostridioides difficile infection. Nature Communications. 2021;12:755. doi: 10.1038/s41467-020-20793-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Mallick H, Rahnavard A, McIver LJ, Ma S, Zhang Y, Nguyen LH, Tickle TL, Weingart G, Ren B, Schwager EH, Chatterjee S, Thompson KN, Wilkinson JE, Subramanian A, Lu Y, Waldron L, Paulson JN, Franzosa EA, Bravo HC, Huttenhower C. Multivariable association discovery in population-scale meta-omics studies. PLOS Computational Biology. 2021;17:e1009442. doi: 10.1371/journal.pcbi.1009442. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Maltz C, Miskovitz PF, Hajifathalian K. Lactulose may reduce Clostridium difficile-related diarrhea among patients receiving antibiotics. JGH Open. 2020;4:1088–1090. doi: 10.1002/jgh3.12390. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Martin-Verstraete I, Peltier J, Dupuy B. The Regulatory Networks That Control Clostridium difficile Toxin Synthesis. Toxins. 2016;8:E153. doi: 10.3390/toxins8050153. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. McGovern BH, Ford CB, Henn MR, Pardi DS, Khanna S, Hohmann EL, O’Brien EJ, Desjardins CA, Bernardo P, Wortman JR, Lombardo M-J, Litcofsky KD, Winkler JA, McChalicher CWJ, Li SS, Tomlinson AD, Nandakumar M, Cook DN, Pomerantz RJ, Auninš JG, Trucksis M. SER-109, an Investigational Microbiome Drug to Reduce Recurrence After Clostridioides difficile Infection: Lessons Learned From a Phase 2 Trial. Clinical Infectious Diseases. 2021;72:2132–2140. doi: 10.1093/cid/ciaa387. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Mefferd CC, Bhute SS, Phan JR, Villarama JV, Do DM, Alarcia S, Abel-Santos E, Hedlund BP. A High-Fat/High-Protein, Atkins-Type Diet Exacerbates Clostridioides (Clostridium) Difficile Infection in Mice, Whereas a High-Carbohydrate Diet Protects. MSystems. 2020;5:e19. doi: 10.1128/mSystems.00765-19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Merrigan M, Venugopal A, Mallozzi M, Roxas B, Viswanathan VK, Johnson S, Gerding DN, Vedantam G. Human hypervirulent Clostridium difficile strains exhibit increased sporulation as well as robust toxin production. Journal of Bacteriology. 2010;192:4904–4911. doi: 10.1128/JB.00445-10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Mills JP, Rao K, Young VB. Probiotics for prevention of Clostridium difficile infection. Current Opinion in Gastroenterology. 2018;34:3–10. doi: 10.1097/MOG.0000000000000410. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Mistou M-Y, Sutcliffe IC, van Sorge NM. Bacterial glycobiology: rhamnose-containing cell wall polysaccharides in Gram-positive bacteria. FEMS Microbiology Reviews. 2016;40:464–479. doi: 10.1093/femsre/fuw006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Mullish BH, Williams HR. Clostridium difficile infection and antibiotic-associated diarrhoea. Clinical Medicine. 2018;18:237–241. doi: 10.7861/clinmedicine.18-3-237. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Mullish BH, Allegretti JR. The contribution of bile acid metabolism to the pathogenesis of Clostridioides difficile infection. Therapeutic Advances in Gastroenterology. 2021;14:17562848211017725. doi: 10.1177/17562848211017725. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Nagao-Kitamoto H, Leslie JL, Kitamoto S, Jin C, Thomsson KA, Gillilland MG, Kuffa P, Goto Y, Jenq RR, Ishii C, Hirayama A, Seekatz AM, Martens EC, Eaton KA, Kao JY, Fukuda S, Higgins PDR, Karlsson NG, Young VB, Kamada N. Interleukin-22-mediated host glycosylation prevents Clostridioides difficile infection by modulating the metabolic activity of the gut microbiota. Nature Medicine. 2020;26:608–617. doi: 10.1038/s41591-020-0764-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Palleja A, Mikkelsen KH, Forslund SK, Kashani A, Allin KH, Nielsen T, Hansen TH, Liang S, Feng Q, Zhang C, Pyl PT, Coelho LP, Yang H, Wang J, Typas A, Nielsen MF, Nielsen HB, Bork P, Wang J, Vilsbøll T, Hansen T, Knop FK, Arumugam M, Pedersen O. Recovery of gut microbiota of healthy adults following antibiotic exposure. Nature Microbiology. 2018;3:1255–1265. doi: 10.1038/s41564-018-0257-9. [DOI] [PubMed] [Google Scholar]
  51. Pereira FC, Wasmund K, Cobankovic I, Jehmlich N, Herbold CW, Lee KS, Sziranyi B, Vesely C, Decker T, Stocker R, Warth B, von Bergen M, Wagner M, Berry D. Rational design of a microbial consortium of mucosal sugar utilizers reduces Clostridiodes difficile colonization. Nature Communications. 2020;11:5104. doi: 10.1038/s41467-020-18928-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Porter NT, Martens EC. The Critical Roles of Polysaccharides in Gut Microbial Ecology and Physiology. Annual Review of Microbiology. 2017;71:349–369. doi: 10.1146/annurev-micro-102215-095316. [DOI] [PubMed] [Google Scholar]
  53. Pruss KM, Sonnenburg JL. C. difficile exploits a host metabolite produced during toxin-mediated disease. Nature. 2021;593:261–265. doi: 10.1038/s41586-021-03502-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Rashid MU, Zaura E, Buijs MJ, Keijser BJF, Crielaard W, Nord CE, Weintraub A. Determining the Long-term Effect of Antibiotic Administration on the Human Normal Intestinal Microbiota Using Culture and Pyrosequencing Methods. Clinical Infectious Diseases. 2015;60 Suppl 2:S77–S84. doi: 10.1093/cid/civ137. [DOI] [PubMed] [Google Scholar]
  55. Rätsep M, Kõljalg S, Sepp E, Smidt I, Truusalu K, Songisepp E, Stsepetova J, Naaber P, Mikelsaar RH, Mikelsaar M. A combination of the probiotic and prebiotic product can prevent the germination of Clostridium difficile spores and infection. Anaerobe. 2017;47:94–103. doi: 10.1016/j.anaerobe.2017.03.019. [DOI] [PubMed] [Google Scholar]
  56. Robinson JI, Weir WH, Crowley JR, Hink T, Reske KA, Kwon JH, Burnham C-AD, Dubberke ER, Mucha PJ, Henderson JP. Metabolomic networks connect host-microbiome processes to human Clostridioides difficile infections. The Journal of Clinical Investigation. 2019;129:3792–3806. doi: 10.1172/JCI126905. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Rohart F, Gautier B, Singh A, Lê Cao K-A. mixOmics: An R package for ’omics feature selection and multiple data integration. PLOS Computational Biology. 2017;13:e1005752. doi: 10.1371/journal.pcbi.1005752. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Scaria J, Chen JW, Useh N, He H, McDonough SP, Mao C, Sobral B, Chang YF. Comparative nutritional and chemical phenome of Clostridium difficile isolates determined using phenotype microarrays. International Journal of Infectious Diseases. 2014;27:20–25. doi: 10.1016/j.ijid.2014.06.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Schmieder R, Edwards R. Fast identification and removal of sequence contamination from genomic and metagenomic datasets. PLOS ONE. 2011;6:e17288. doi: 10.1371/journal.pone.0017288. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Schnizlein MK, Vendrov KC, Edwards SJ, Martens EC, Young VB. Dietary Xanthan Gum Alters Antibiotic Efficacy against the Murine Gut Microbiota and Attenuates Clostridioides difficile Colonization. MSphere. 2020;5:e00708-19. doi: 10.1128/mSphere.00708-19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Silva J, Ferraz R, Dupree P, Showalter AM, Coimbra S. Three Decades of Advances in Arabinogalactan-Protein Biosynthesis. Frontiers in Plant Science. 2020;11:610377. doi: 10.3389/fpls.2020.610377. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Sorbara MT, Pamer EG. Interbacterial mechanisms of colonization resistance and the strategies pathogens use to overcome them. Mucosal Immunology. 2019;12:1–9. doi: 10.1038/s41385-018-0053-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Stevens V, Dumyati G, Fine LS, Fisher SG, van Wijngaarden E. Cumulative antibiotic exposures over time and the risk of Clostridium difficile infection. Clinical Infectious Diseases. 2011;53:42–48. doi: 10.1093/cid/cir301. [DOI] [PubMed] [Google Scholar]
  64. Teng C, Reveles KR, Obodozie-Ofoegbu OO, Frei CR. Clostridium difficile Infection Risk with Important Antibiotic Classes: An Analysis of the FDA Adverse Event Reporting System. International Journal of Medical Sciences. 2019;16:630–635. doi: 10.7150/ijms.30739. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Theriot CM, Koenigsknecht MJ, Carlson PE, Hatton GE, Nelson AM, Li B, Huffnagle GB, Z Li J, Young VB. Antibiotic-induced shifts in the mouse gut microbiome and metabolome increase susceptibility to Clostridium difficile infection. Nature Communications. 2014;5:3114. doi: 10.1038/ncomms4114. [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Truong DT, Franzosa EA, Tickle TL, Scholz M, Weingart G, Pasolli E, Tett A, Huttenhower C, Segata N. MetaPhlAn2 for enhanced metagenomic taxonomic profiling. Nature Methods. 2015;12:902–903. doi: 10.1038/nmeth.3589. [DOI] [PubMed] [Google Scholar]
  67. Webb BJ, Subramanian A, Lopansri B, Goodman B, Jones PB, Ferraro J, Stenehjem E, Brown SM. Antibiotic Exposure and Risk for Hospital-Associated Clostridioides difficile Infection. Antimicrobial Agents and Chemotherapy. 2020;64:e02169-19. doi: 10.1128/AAC.02169-19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Westblade LF, Chamberland RR, MacCannell D, Collins R, Dubberke ER, Dunne WM, Burnham C-AD. Development and evaluation of a novel, semiautomated Clostridium difficile typing platform. Journal of Clinical Microbiology. 2013;51:621–624. doi: 10.1128/JCM.02627-12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Wirbel J, Zych K, Essex M, Karcher N, Kartal E, Salazar G, Bork P, Sunagawa S, Zeller G. Microbiome meta-analysis and cross-disease comparison enabled by the SIAMCAT machine learning toolbox. Genome Biology. 2021;22:93. doi: 10.1186/s13059-021-02306-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Wood DE, Salzberg SL. Kraken: ultrafast metagenomic sequence classification using exact alignments. Genome Biology. 2014;15:R46. doi: 10.1186/gb-2014-15-3-r46. [DOI] [PMC free article] [PubMed] [Google Scholar]

Editor's evaluation

Peter J Turnbaugh 1

Not everyone colonized by C. difficile has gut symptoms, but the reasons why are unclear. This article uses the combination of sequencing and mass spectrometry to compare patients with or without symptoms, revealing links between specific gut bacteria and diet, which could lead to diet or bacterial treatment or prevention strategies.

Decision letter

Editor: Peter J Turnbaugh1
Reviewed by: Peter J Turnbaugh2

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Decision letter after peer review:

Thank you for submitting your article "Multi-omics investigation of Clostridioides difficile-colonized patients reveals protective commensal carbohydrate metabolism" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Peter J Turnbaugh as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Gisela Storz as the Senior Editor.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) Need to discuss the ability to exclude alternative hypotheses, including variations between C. difficile strains, dietary intake, differences in host physiology, and bile acid production/metabolism. The former seems like a critical point – are these individuals colonized by similar strains of C. difficile? Are they all toxin positive? It is critical to test if C. difficile from the Cx+/EIA- samples are actually capable of producing toxin. This is important to discern whether there are facets of the microbiome/metabolome which turn toxin off in Cx+/EIA- samples or if C. difficile in these patients have mutations which make them unable to produce toxin.

2) Given the compositional nature of the sequencing data it is possible that differences in C. difficile are responsible for some of the observed differences in community structure. Please mask C. difficile reads and re-run the key analyses to check if they hold up.

3) Please discuss the literature precedent for C. difficile growth on different carbohydrates and ideally include data for the type strain.

4) Please check if the conclusions are impacted by removing the Cx+/EIA- samples with metagenomically undetectable C. difficile from the computational analyses used in Figures 1-3. The concern is whether these samples are driving the perceived differences between Cx+/EIA+ patients and Cx+/EIA- patients (does C. difficile abundance or metabolite abundance still differentiate Cx+/EIA+ patients from Cx+/EIA- patients?).

Reviewer #1:

The mechanisms that protect some individuals from C. difficile-associated colitis remain poorly understood; however, recent data has implicated both diet and the microbiome. Here, the authors use paired metagenomic and metabolomic analysis to identify differences in asymptomatic and symptomatic patients, suggesting that competition between clostridial species for carbohydrate metabolism may play a role. The data is clearly presented and provides clear hypotheses for future studies aimed at understanding the complex interactions between enteric pathogens, the gut microbiome, and host pathophysiology.

Strengths of this study include its unique cohort, rigorous analysis and presentation, inclusion of some initial in vitro validation work, and potential for inspiring future hypothesis-driven experiments.

Weaknesses include the lack of consideration or ability to control for alternative hypotheses, including variations between C. difficile strains, dietary intake, differences in host physiology, and bile acid production/metabolism. The effect sizes are also modest with no high-level differences in the microbiome or metabolome between groups. Finally, there is no evidence of generalizability to other patient cohorts. Given these caveats, it is important to be clear throughout that this is a hypothesis generating exercise and that the degree to which commensal carbohydrate metabolism is protective against C. difficile infection requires further clinical and mechanistic data.

Comments for the authors:

1. Need to discuss the ability to exclude alternative hypotheses, including variations between C. difficile strains, dietary intake, differences in host physiology, and bile acid production/metabolism. The former seems like a critical point – are these individuals colonized by similar strains of C. difficile? Are they all toxin positive? I was unclear how asymptomatic carriage is defined, this is critical to the current paper and should be included in the main text and methods, not as a citation.

2. Given the compositional nature of the sequencing data it is possible that differences in C. difficile are responsible for some of the observed differences in community structure. I'd recommend masking C. difficile reads and re-running the key analyses to check if they hold up.

3. The in vitro validation is helpful, but I'm unclear as to whether it is new information. If any prior studies have been done they should be cited here.

Reviewer #2:

The manuscript by Fishbein et al. examines an exciting and timely question about the microbial and metabolomic factors in the gastrointestinal tract that determine if C. difficile remains dormant as a colonizer or triggers infection. They do this through the re-analysis of their previously published cohort wherein they can separate the colonization vs infection state on the basis of qPCR and toxin detection (EIA) which provides a unique opportunity to address their question in an excellently phenotyped cohort.

Using appropriate and current approaches, the authors find relatively subtle differences in the microbiome and metabolome of colonized/infected participants including certain carbohydrates which are elevated in asymptomatic individuals. They demonstrate these carbohydrates are not substrates for C. difficile; however, this line of experimentation sought to find negative results and it remains to be determined if they have any relevance in vivo or in complex communities. The authors could consider an additional experiment to probe this observation in more depth such as ex vivo fecal incubations or strain-strain competition experiments to provide more direct evidence for how they may influence the suppression of C. difficile.

Comments for the authors:

Given that this manuscript is primarily computational, it would be beneficial if the code for the analysis was shared in a public repository.

PRJNA748262 does not appear to be publicly available.

Line 66: would asymptomatic colonization be on the disease spectrum?

Reviewer #3:

This manuscript reveals exciting details of the lifestyle of C. difficile in individuals who are asymptomatic carriers and contrasts these with individuals with active CDI. It will be an important contribution to the C. difficile research field and more broadly to people interested in targeted manipulation of microbial communities.

The study team leveraged a unique patient cohort, coupled with metagenomic and metabolomic analyses, and bacterial culture techniques to suggest that the gut microbiome (namely, commensal clostridia) inhibit C. difficile proliferation. Much of the paper relies on the idea that C. difficile burdens are lower in asymptomatically colonized individuals versus those with C. difficile infection (CDI). However, many individuals who are asymptomatically colonized with C. difficile have burdens that are comparable to those who have CDI (and there may be a bimodal distribution of asymptomatically colonized individuals – those that have C. difficile detectable by metagenomic sequencing and those that do not). I worry that their conclusions are skewed by an apparent binary distribution of C. difficile burdens in both patient groups.

The patients are binned based on initial C. difficile diagnostic tests (asymptomatic individuals carry toxin-encoding C. difficile as detected by PCR but have no detectable toxin via immunoassay). I am unsure how sensitive the standard diagnostic immunoassay for C. difficile is but do know C. difficile toxin production is controlled by a complex regulatory network and abundance of these toxins can change with respect to parameters such as nutrient availability, temperature, pH, and cell density. I wonder if the isolates from the study are all capable of producing functional toxin. These strains could have mutations (in regulatory elements, for example) that may impact toxin production.

The authors perform a culture-based experiment to support their findings that certain metabolites present in the gastrointestinal tracts of asymptomatic individuals do not support the growth of C. difficile. They use isolates from the study participants but it is unclear if these isolates were taken from asymptomatic individuals or those with CDI but do not leverage strains commonly used in the research community (C. difficile 630 and C. difficile R20291).

Lines 67-69. What is the limit of detection of EIA? It's striking to me that, while C. difficile burdens are a predictor of Cx+/EIA+ vs. Cx+/EIA- patients, there are ~23 Cx+/EIA- patients whose relative abundance of C. difficile essentially superimposes with those of the Cx+/EIA+ group. There are fibroblast rounding assays that quantify C. difficile toxins. The limit of detection of these assays may be lower than kits used in diagnostic labs. Can the authors perform one of these assays on the samples in this study to better calibrate readers expectations of whether toxigenic C. difficile is or is not expressing toxin? As the paper stands, I assume that production of toxin in C. difficile is all-or-nothing, but prior work on regulation of toxin production in C. difficile has demonstrated that many regulatory cues influence toxin production (Martin-Verstraete et al. 2016) so I expect this to be non-binary. In addition, and considering the regulatory networks that control toxin production, are the C. difficile isolates from asymptomatic patients capable of producing toxin in vitro? Just because the gene is there doesn't mean that it's functional.

Line 140-142. "Given that C. difficile levels were an overt predictor of CDI…" Are C. difficile levels still a predictor of the Cx+/EIA+ state if the samples where C. difficile was not detectable by metagenomic sequencing were removed (see Figure S1C [bottom right hand corner of the graph], n=2 Cx+/EIA+ samples and n=? Cx+/EIA- samples)? The burdens of C. difficile in these individuals seem to be at least an order of magnitude lower than patients with low-but-detectable burdens and appear to be bimodally distributed. If there is no change to the conclusions of the study if these samples are removed, are there features (metabolites, taxa) that differentiate individuals who do or do not have metagenomically detectable C. difficile?

Line 144. There is no Figure S1D in the manuscript.

Lines 228-230. I really like the idea that rhamnose may be the by-product of microbiome metabolism or other dietary polysaccharides which could exclude C. difficile from the gut. What is known about rhamnose metabolism by gut microbes? Do any of the taxa that associate with Cx+/EIA+ conditions metabolize rhamnose? Do any of the taxa that associiate with Cx+/EIA- conditions metabolize rhamnopolysaccharides? This could be done by mining the metagenomes for CAZymes or by performing growth assays with isolates of relevant bacteria.

Lines 248-251: "The observed increased abundance of monosaccharides…" If this is true, shouldn't there be fewer sugars in both high diversity conditions and low diversity conditions? That is, in a high diversity microbiome, some microbes would metabolize complex polysaccahrides and others would cross feed on the waste products, whereas in a low diversity microbiome, there may be fewer degraders of polysaccharides, but maybe also fewer cross-feeders? Further expansion on this complexity or references to support the assertion in the sentence on lines 248-251 would be useful.

Lines 253-256: Given the previous reports cited in the manuscript and the range of metabolic capabilities illustrated by the clinical isolates used in this study, it would be worthwhile to assay the sugars from Figure 4 against strains of C. difficile commonly used in the C. difficile research community (at least, C. difficile 630 and difficile R0291).

Lines 262-264: This is really interesting. How many patients were excluded? Assaying toxin production in these patients (see comment on Lines 67-69, above) before and after diagnosis, or sequencing C. difficile from these patients at each of these time points, would be helpful in thinking about the transitions between asymptomatic carriage and symptomatic CDI. This is, of course, beyond the scope of the current study and only mentioned because it would be of scientific and clinical interest.

Line 320. How many Cx+/EIA- samples were included in the study?

Line 397. Please clarify in this section of the same conditions were used for VPI10643.

Lines 438-441. What percent of the variation in the data are expressed in this beta diversity metric? Is the conclusion different if a phylogenetically-aware metric (e.g. Weighted/Unweighted UniFrac) is used?

Figure S1. Please include p values for taxa noted in Figure S1C. How many patients' stool samples have undetectable C. difficile via metagenomic analysis?

Figure 4. OD readings are typically plotted on a log scale for growth curve data.

eLife. 2022 Jan 27;11:e72801. doi: 10.7554/eLife.72801.sa2

Author response

Essential revisions:

1) Need to discuss the ability to exclude alternative hypotheses, including variations between C. difficile strains, dietary intake, differences in host physiology, and bile acid production/metabolism. The former seems like a critical point – are these individuals colonized by similar strains of C. difficile? Are they all toxin positive? It is critical to test if C. difficile from the Cx+/EIA- samples are actually capable of producing toxin. This is important to discern whether there are facets of the microbiome/metabolome which turn toxin off in Cx+/EIA- samples or if C. difficile in these patients have mutations which make them unable to produce toxin.

We thank the reviewers for their critical insight into the complexity of CDI. To address the points raised, we analyzed the ribotype distribution across C. difficile isolates in our Cx+/EIA- and Cx+/EIA+ patients and also used diagnostic PCR data to characterize the presence or absence of the binary toxin loci (cdtAB). We found a significant enrichment of the binary (CDT) toxin loci in isolates of Cx+/EIA+ patients and a significant enrichment of isolates from the RT027 lineage, considered a hypervirulent lineage of C.

difficile (Figure 1A). We also found that CDT+/EIA+ stools had slightly increased abundances of C. difficile (Supplementary Figure 2C, P=0.2); this is consistent with previous data that suggests that the presence of the binary toxin is associated with severe disease [1]. We incorporated these pathogen variables in revised community analyses (Figure 1, Supplementary Figure 1, Supplementary Figure 2).

Clinically, toxigenic C. difficile is defined as a C. difficile isolate with PCRdetectable tcdAB. Yet, as the reviewer has indicated it is very possible that the isolate has the locus present but does not express the toxin due to multiple possible levels of genetic/epigenetic variation. To answer the question of whether the Cx+/EIA- express toxin (TcdA and TcdB), we performed in vitro toxin ELISAs on all 102 C. difficile isolates for which we had corresponding stool metagenomic data. We found that approximately half of the isolates expressed detectable levels of toxin, and found no significant differences in the proportion of isolates that express toxin between the two study groups (Supplementary Figure 1A, P=0.86, Fisher’s exact test). As has previously been reported [2, 3], toxin expression is highly regulated by environmental cues, and can be difficult to detect during in vitro broth culture. While our in vitro data would indicate that there is no difference in the toxin expression capacity of isolates from the two cohorts, future experiments, outside the scope of this paper, will examine the variability in clinical isolate toxin expression in vivo and under different environmental cues.

2) Given the compositional nature of the sequencing data it is possible that differences in C. difficile are responsible for some of the observed differences in community structure. Please mask C. difficile reads and re-run the key analyses to check if they hold up.

We appreciate the reviewer’s statistical concern, and believe that these concerns stem from the apparent bimodality of C. difficile abundance. We addressed this concern both by performing analyses that the reviewer asked for, as well as by using an independent metagenomic taxonomic classifier which is more sensitive to C. difficile reads.

We masked C. difficile abundance (derived from Metaphlan metagenomic classification) and renormalized the taxonomic relative abundances to understand whether this changed the outcome of analyses from Figure 1. We found that beta diversity (using UniFrac distances) was still not significantly different between EIA+ and EIA- metagenomes (PERMANOVA, P=0.2). We performed MaAsLin2 linear mixed modeling on this dataset and found that the same commensal clostridia (from the C. difficile unmasked analysis) were associated with EIA- patients (Author response image 1).

Author response image 1. Maaslin2 analysis of metagenomes with C. difficile reads removed and compositional dataset renormalized.

Author response image 1.

Open in a new tab

In parallel, we also used an alternate metagenomic taxonomic sequence classifier, Kraken, which uses a k-mer based approach to map reads to marker sequences. This method is more sensitive than MetaPhlAn2, and accordingly detected more C. difficile abundance in our NAAT+ metagenomic samples (finding C. difficile in 101/102 samples vs. 70/102 samples from MetaPhlAn2). Importantly, however, we observed highly correlated measures of C. difficile abundance (Pearson’s rho=0.94) from the MetaPhlAn2 and Kraken abundance calculations. Furthermore, we found that C. difficile and a number of commensal clostridia from the Eubacterium and Anaerostipes were significantly associated with EIA status in both MetaPhlAn2 and Kraken analyses (Figure 1 and Supplementary Figure 2). We were also able to demonstrate that the magnitude of correlations between C. difficile and metabolites was preserved between both analyses. Briefly, in our MetaPhlAn2 analysis, C. difficile was positively correlated with 5-aminovaleric acid and 4-methylpentanoic acid (rho=0.48 and 0.36, respectively), and negatively correlated with fructose, rhamnose, and hydroxyproline (rho=-0.27, -0.36, and -0.34, respectively). Using Kraken metagenomic taxonomic classifications (which detects more C. difficile in our stool microbiomes), C. difficile was also well-correlated with 5-aminovaleric acid and 4-methylpentanoic acid (rho of 0.51 and 0.41 respectively), and anticorrelated to fructose, rhamnose, and hydroxyproline (rho of -0.29,-0.35 and -0.37 respectively). These results are displayed in Figure 4 and Supplementary Figure 5.

Based on these findings, we do not believe that the bimodality of C. difficile abundances observed using MetaPhlAn2 has a major influence on our findings.

3) Please discuss the literature precedent for C. difficile growth on different carbohydrates and ideally include data for the type strain.

We have included a discussion of the literature, concerning sorbitol growth and growth on other carbohydrates to the Discussion (lines 311-313), and highlight that a number of these substrates have not been rigorously validated in clinical isolates. Additionally, we used C. difficile 630 as the reference strain for this experiment (Figure 3B,C).

4) Please check if the conclusions are impacted by removing the Cx+/EIA- samples with metagenomically undetectable C. difficile from the computational analyses used in Figures 1-3. The concern is whether these samples are driving the perceived differences between Cx+/EIA+ patients and Cx+/EIA- patients (does C. difficile abundance or metabolite abundance still differentiate Cx+/EIA+ patients from Cx+/EIA- patients?).

We thank this reviewer for bringing up this point, and we have addressed it in detail above (Essential revision #2). Per the reviewer’s suggestion, we removed samples that had no metagenomically detectable C. difficile (using the MetaPhlAn2 classifier, leaving 70 samples) and reran analyses in Figure 1 to assess if the microbiome differences were preserved. We expected that this biased removal of a substantial number of samples (the samples with the lowest C. difficile abundance) would reduce statistical power in our analyses due to uneven reduction of sample size of our cohort, and would likely primarily affect the significance of the association with C. difficile abundance in our cohorts. As predicted, when we perform the MaAsLin2analyses on this reduced dataset, C. difficile was no longer a predictor of EIA status (FDR<0.25) by this analysis (Author response image 2). However, Anaerostipes hadrus and Lachnospiraceae_bacterium_5_1_63FAA were still the most predictive taxa associated with EIA state (with equivalent FDR values of 0.0033 and 0.0077, respectively).

Author response image 2. Maaslin2 analysis of stool metagenomes with detectable C. difficile.</Author response image 2 title/legend>.

Author response image 2.

Open in a new tab

However, as we have clarified in the results, all stools were found to contain viable tcdAB-positive C. difficile isolates via selective culture, and therefore it appears the stool samples with the lowest C. difficile abundance are below the metagenomic sequence limit of C. difficile detection using MetaPhlAn2. We hypothesized that if we used an alternative taxonomic classifier with a lower limit of detection, we might be able to detect C. difficile in the same metagenomic data. Accordingly, we performed the same computational analyses using Kraken taxonomic classifications instead, which detected C. difficile in 101/102 samples vs. 70/102 samples from MetaPhlAn2. We detail the results of these analyses above and in Figure 1, Supplementary Figure 2, Supplementary Figure 5, and Figure 3, showing that our major conclusions remain unchanged from those based on Metaphlan 2 classifications. Based on our culturing data and Kraken analyses, which we explain in our revised manuscript (lines 178-179, lines 262-270), we believe it is appropriate to retain all samples in our analyses.

Reviewer #1:

[…]Comments for the authors:

1. Need to discuss the ability to exclude alternative hypotheses, including variations between C. difficile strains, dietary intake, differences in host physiology, and bile acid production/metabolism. The former seems like a critical point – are these individuals colonized by similar strains of C. difficile? Are they all toxin positive? I was unclear how asymptomatic carriage is defined, this is critical to the current paper and should be included in the main text and methods, not as a citation.

We appreciate the reviewer’s critical insight into a weakness of our original submission: our lack of consideration/discussion of other factors contributing to C. difficile pathogenesis, which we have striven to address in our revision. To clarify the basis of our comparisons, we have added statements in the Introduction and expanded on our definition of asymptomatic carriage in the beginning of the Results section, including the diagnostic pathways that led to identification of these strains as toxigenic C. difficile (lines 119-126). We focused on the most accessible data to us, namely data on the strain type and toxin identity of the pathogen. We performed a series of analyses on the microbiome data to assess the contribution of strain identity (using the proxy of ribotype and cdtAB toxin allele) to microbial community structure and present these figures in the supplementary data (Figure 1 and Supplementary Figure 1); there we highlight critical differences between strain type and cdtAB allele status in EIA cohorts. Additionally, we included a paragraph in the Discussion concerning the contribution of strain heterogeneity to variation in clinical outcome (lines 302-317). Finally, we discuss our limited capacity to account for dietary intake and host physiology in the Discussion section (lines 290-294).

2. Given the compositional nature of the sequencing data it is possible that differences in C. difficile are responsible for some of the observed differences in community structure. I'd recommend masking C. difficile reads and re-running the key analyses to check if they hold up.

We addressed this response in detail above (Essential revision 2). Briefly, we masked the reads, renormalized the relative abundances for each patient, and revealed that commensal Clostridia species were still significantly associated with Cx+/EIA- patients.

3. The in vitro validation is helpful, but I'm unclear as to whether it is new information. If any prior studies have been done they should be cited here.

We thank the reviewer for this consideration and acknowledge that while information on fructose and sorbitol usage have been studied in reference or commonly used strains, here we present new data on nutrient utilization by clinical isolates. This revealed the while some strains appear to grow as robustly as the reference isolates (C. difficile VPI10643 and C. difficile 630), some do not appear to grow as well on sorbitol as the sole carbon source. Fructose and sorbitol served as well-known utilizable carbohydrate substrates such that we could draw conclusions about the ability of C. difficile to utilize carbohydrates such as lactulose, rhamnose, and sucrose. Finally, we have added to the Discussion section (lines 311-314) to highlight the preceding literature to our work and the limitations of our nutrient utilization data.

Reviewer #2:

[…] Comments for the authors:

Given that this manuscript is primarily computational, it would be beneficial if the code for the analysis was shared in a public repository.

We have generated a.Rmd file and deposited this to Github.

PRJNA748262 does not appear to be publicly available.

We have made this data public.

Line 66: would asymptomatic colonization be on the disease spectrum?

We have revised this wording to clarify this point, specifically by changing the phrase “disease spectrum” to “clinical manifestation” (line 72).

Reviewer #3:

[…] Lines 67-69. What is the limit of detection of EIA? It's striking to me that, while C. difficile burdens are a predictor of Cx+/EIA+ vs. Cx+/EIA- patients, there are ~23 Cx+/EIA- patients whose relative abundance of C. difficile essentially superimposes with those of the Cx+/EIA+ group. There are fibroblast rounding assays that quantify C. difficile toxins. The limit of detection of these assays may be lower than kits used in diagnostic labs. Can the authors perform one of these assays on the samples in this study to better calibrate readers expectations of whether toxigenic C. difficile is or is not expressing toxin? As the paper stands, I assume that production of toxin in C. difficile is all-or-nothing, but prior work on regulation of toxin production in C. difficile has demonstrated that many regulatory cues influence toxin production (Martin-Verstraete et al. 2016) so I expect this to be non-binary. In addition, and considering the regulatory networks that control toxin production, are the C. difficile isolates from asymptomatic patients capable of producing toxin in vitro? Just because the gene is there doesn't mean that it's functional.

We sincerely appreciate this concern and have addressed it in detail (Essential revision #1). Briefly, our in vitro ELISA data indicates that half of isolates were positive for the toxin (this frequency was not different between EIA- and EIA+, Supplementary Figure 1A). This data suggests that there is no difference in the number of isolates that produce toxin, and namely, that a significant proportion of EIA- isolates are able to express toxin in vitro. However, as the reviewer is no doubt aware, this is an inadequate method of assessing a strain’s in vivo capacity to elaborate toxin [2, 3]. We have highlighted this limitation in our discussion (lines 310-313).

Line 140-142. "Given that C. difficile levels were an overt predictor of CDI…" Are C. difficile levels still a predictor of the Cx+/EIA+ state if the samples where C. difficile was not detectable by metagenomic sequencing were removed (see Figure S1C [bottom right hand corner of the graph], n=2 Cx+/EIA+ samples and n=? Cx+/EIA- samples)? The burdens of C. difficile in these individuals seem to be at least an order of magnitude lower than patients with low-but-detectable burdens and appear to be bimodally distributed. If there is no change to the conclusions of the study if these samples are removed, are there features (metabolites, taxa) that differentiate individuals who do or do not have metagenomically detectable C. difficile?

We have addressed this in Essential revision 2 and 4. If samples with metagenomically undetectable levels of C. difficile are removed, as expected, the C. difficile association was no longer statistically significant. Based on our culture data and the use of an alternative metagenomic taxonomic classifier (Kraken), we hypothesize that samples with no metagenomically detectable C. difficile (by MetaPhlAn2) are those that contain the lowest abundance of the organism. To support the conclusions of our manuscript, we performed identical analysis of differentially abundant taxa (using Metaphlan in Figure 1 and Kraken in Figure S2), and found that increased C. difficile is associated with Cx+/EIA+ patients and a number of commensal Clostridia are associated with Cx+/EIA- patients in both analyses of the same sequencing data. Additionally, we performed identical analyses of multi-omic correlates of these cohorts (Figure 3 and Figure S5) and found the findings were preserved. Briefly, in our Metaphlan analysis, C. difficile was positively correlated with 5-amino-valeric acid and 4-methylpentanoic acid (rho=0.48 and 0.36, respectively), and negatively correlated with fructose, rhamnose, and hydroxyproline (rho=-0.27, -0.36, and -0.34, respectively). Using Kraken-classified metagenomic data (which detects more C. difficile in our stool microbiomes), C. difficile was well-correlated with 5-amino-valeric acid and 4-methylpentanoic acid (rho of 0.51 and 0.41 respectively), and anti-correlated to fructose, rhamnose, and hydroxyproline (rho of -0.29,-0.35 and -0.37 respectively).

Line 144. There is no Figure S1D in the manuscript.

We have corrected this error.

Lines 228-230. I really like the idea that rhamnose may be the by-product of microbiome metabolism or other dietary polysaccharides which could exclude C. difficile from the gut. What is known about rhamnose metabolism by gut microbes? Do any of the taxa that associate with Cx+/EIA+ conditions metabolize rhamnose? Do any of the taxa that associate with Cx+/EIA- conditions metabolize rhamnopolysaccharides? This could be done by mining the metagenomes for CAZymes or by performing growth assays with isolates of relevant bacteria.

In response to multiple reviewer and editorial comments, we have stepped back from hypotheses concerning specific metabolites. Given the complexity of CDI and the cross-sectionality of our data, we have refocused the manuscript more broadly on trying to understand asymptomatic colonization using pathogen and microbiome data. Yet, we felt it was an important exercise to examine the presence of CAZymes in our metagenomic data. Thus, we utilized Humann2 functionally-annotated gene family abundance data, and identified all KEGG-annotated genes with the enzyme commission number EC 3.2.1 [4], as these represent glycoside hydrolases that could be involved in carbohydrate degradation. We found a number of genes putatively involved in starch and sucrose metabolism that were increased in EIA- patients, supporting our MetaCyc pathway analysis in Figure 2. We are extremely interested in identifying species metabolite utilization profiles, specifically in the case of starch and rhamnopolysaccharides, but feel that it is outside the scope of this manuscript. Yet, we have added a speculative note concerning Clostridia metabolism to the Discussion section (line 348-349).

Lines 248-251: "The observed increased abundance of monosaccharides…" If this is true, shouldn't there be fewer sugars in both high diversity conditions and low diversity conditions? That is, in a high diversity microbiome, some microbes would metabolize complex polysaccahrides and others would cross feed on the waste products, whereas in a low diversity microbiome, there may be fewer degraders of polysaccharides, but maybe also fewer cross-feeders? Further expansion on this complexity or references to support the assertion in the sentence on lines 248-251 would be useful.

We are extremely thankful for the reviewer’s insight and curiosity concerning the relationship between monosaccharide levels and microbiome diversity. Based on these comments, we hypothesized that there might be a relationship between monosaccharide levels and microbiome diversity. To directly examine this relationship, we summed levels of representative (and biochemically confirmed) monosaccharides: fructose, rhamnose, and glucose. We then measured correlation between the sugar sum and 3 different measures of alpha diversity and found no strong correlation, as measured by Pearson’s rho: Shannon diversity(-0.083), richness(-0.045), and Faith’s diversity (-0.081). We have also stepped back from speculating as to the meaning behind the level of monosaccharides. As we have not accounted for differences in host diet, we note that we cannot attribute the differences in levels monosaccharides to differences in host diet or differences in cross-feeding rate in the community (lines 299-301).

Lines 253-256: Given the previous reports cited in the manuscript and the range of metabolic capabilities illustrated by the clinical isolates used in this study, it would be worthwhile to assay the sugars from Figure 4 against strains of C. difficile commonly used in the C. difficile research community (at least, C. difficile 630 and difficile R0291).

We have redone these experiments with C. difficile 630, a commonly used reference strain, and also kept C. difficile VPI10643, which is a commonly used reference strain for animal experiments [5, 6]

Lines 262-264: This is really interesting. How many patients were excluded? Assaying toxin production in these patients (see comment on Lines 67-69, above) before and after diagnosis, or sequencing C. difficile from these patients at each of these time points, would be helpful in thinking about the transitions between asymptomatic carriage and symptomatic CDI. This is, of course, beyond the scope of the current study and only mentioned because it would be of scientific and clinical interest.

We appreciate the reviewer’s excitement, and we are in the process of trying to study such transitions both in animal models and in patients.

Line 320. How many Cx+/EIA- samples were included in the study?

We have added numbers to the text for precision.

Line 397. Please clarify in this section of the same conditions were used for VPI10643.

We have clarified this in the text.

Lines 438-441. What percent of the variation in the data are expressed in this beta diversity metric? Is the conclusion different if a phylogenetically-aware metric (e.g. Weighted/Unweighted UniFrac) is used?

More variation is incorporated in the principal coordinate analysis of weighted UniFrac distance (a phylogenetically-aware metric, Author response image 3B), compared to that of Bray Curtis dissimilarity (Author response image 3A). We had previously quantified beta-dispersion between groups, but found that it would be more appropriate to quantify the permutational analysis of variance (PERMANOVA) using the adonis package and found that using phylogenetically-aware measures such as unweighted UniFrac distance and weighted UniFrac dissimilarity yielded insignificant differences in community structure (P=654, P=0.233), compared to that of Bray Curtis (P=0.415). For this reason, we have decided to present weighted Unifrac dissimilarity in the main text.

Author response image 3.

Author response image 3.

Open in a new tab

Figure S1. Please include p values for taxa noted in Figure S1C. How many patients' stool samples have undetectable C. difficile via metagenomic analysis?

We have included p-values. Additionally, 32 out of 102 samples have undetectable C. difficile via MetaPhlAn2 metagenomic analysis; however, please see response to Essential revision #2 above details regarding culture-based C. difficile detection and Kraken-based metagenomic detection (C. difficile detected metagenomically in 99% of samples).

Figure 4. OD readings are typically plotted on a log scale for growth curve data.

We have redone this figure and plotted it on a log-scale.

References:

1. Cowardin, C.A., E.L. Buonomo, M.M. Saleh, M.G. Wilson, S.L. Burgess, S.A. Kuehne, C. Schwan, A.M. Eichhoff, F. Koch-Nolte, D. Lyras, K. Aktories, N.P. Minton, and W.A. Petri, Jr., The binary toxin CDT enhances Clostridium difficile virulence by suppressing protective colonic eosinophilia. Nat Microbiol, 2016. 1(8): p. 16108.

2. Burnham, C.A. and K.C. Carroll, Diagnosis of Clostridium difficile infection: an ongoing conundrum for clinicians and for clinical laboratories. Clin Microbiol Rev, 2013. 26(3): p. 604-30.

3. Akerlund, T., B. Svenungsson, A. Lagergren, and L.G. Burman, Correlation of disease severity with fecal toxin levels in patients with Clostridium difficile-associated diarrhea and distribution of PCR ribotypes and toxin yields in vitro of corresponding isolates. J Clin Microbiol, 2006. 44(2): p. 353-8.

4. Tanes, C., K. Bittinger, Y. Gao, E.S. Friedman, L. Nessel, U.R. Paladhi, L. Chau, E. Panfen, M.A. Fischbach, J. Braun, R.J. Xavier, C.B. Clish, H. Li, F.D. Bushman, J.D. Lewis, and G.D. Wu, Role of dietary fiber in the recovery of the human gut microbiome and its metabolome. Cell Host Microbe, 2021. 29(3): p. 394-407 e5.

5. Theriot, C.M., C.C. Koumpouras, P.E. Carlson, Bergin, II, D.M. Aronoff, and V.B. Young, Cefoperazone-treated mice as an experimental platform to assess differential virulence of Clostridium difficile strains. Gut Microbes, 2011. 2(6): p. 326-34.

6. Erikstrup, L.T., M. Aarup, R. Hagemann-Madsen, F. Dagnaes-Hansen, B. Kristensen, K.E. Olsen, and K. Fuursted, Treatment of Clostridium difficile infection in mice with vancomycin alone is as effective as treatment with vancomycin and metronidazole in combination. BMJ Open Gastroenterol, 2015. 2(1): p. e000038.

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Data Citations

    1. Fishbein SRS. 2021. Fecal metagenomes of C. difficile colonized patients. NCBI BioProject. PRJNA748262
    2. Fishbein SRS. 2021. C. difficile carbohydrate transcriptomics. NCBI BioProject. PRJNA748261

    Supplementary Materials

    Figure 1—source data 1. Raw absorbance value for in vitro toxin ELISA of 102 C.difficile isolates.
    elife-72801-fig1-data1.xlsx (14.3KB, xlsx)
    Figure 3—source data 1. Growth curve data for C. difficile isolates.
    elife-72801-fig3-data1.xlsx (101.8KB, xlsx)
    Supplementary file 1. Patient demographic data.
    elife-72801-supp1.xlsx (9.7KB, xlsx)
    Supplementary file 2. Fecal metagenomics metadata file with isolate information.
    elife-72801-supp2.xlsx (12.5KB, xlsx)
    Supplementary file 3. MaAsLin2 output of metabolites associated with EIA status in addition to metabolite validation information.
    elife-72801-supp3.xlsx (18.4KB, xlsx)
    Supplementary file 4. DEseq output of in vitro rhamnose-exposed C difficile transcriptomic profiling.
    elife-72801-supp4.xlsx (714.2KB, xlsx)
    Transparent reporting form
    elife-72801-transrepform1.pdf (336.9KB, pdf)

    Data Availability Statement

    Metagenomics reads were deposited under BioProject accession number PRJNA748262 and RNA sequencing reads were deposited under BioProject accession number PRJNA748261.

    The following dataset was generated:

    Fishbein SRS. 2021. Fecal metagenomes of C. difficile colonized patients. NCBI BioProject. PRJNA748262

    Fishbein SRS. 2021. C. difficile carbohydrate transcriptomics. NCBI BioProject. PRJNA748261

    Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

    RESOURCES