Multi-omics analysis of hospital-acquired diarrhoeal patients reveals … – Nature.com

Microbiota associations with extended antibiotic exposure

We first assessed HAD gut microbiota structure and diversity with respect to antibiotic treatment using alpha and beta diversity measures and analysis of compositions of microbiomes (ANCOM) differential abundance analysis. Antibiotic-associated diarrhoeal (+AAD) patients comprised 82.2% (139/169) of the cohort (Table1) and were associated with 56 unique combinations of antibiotic classes before specimen collection (Supplementary Table1), rendering analysis by antibiotic class unfeasible. For comparison, faecal samples from healthy donors recruited for faecal microbiota transplant treatment of CDI were similarly assessed (Supplementary materials). Donor samples in the study were obtained from 12 female and 8 male individuals, a smaller male cohort (40.0% male) compared to CDI patients (54.5% male) and non-CDI patients (47.4% male). FMT donors were between the ages of 18 and 65 46 with a total age of 649 years and an average of 32.45 years. While the average age of FMT donors was significantly lower than the CDI (75 years, range 55-83) and non-CDI (68 years, range 52-78) median age, we chose FMT donors purposefully as a comparison group for this study in order to assess the microbiota and metabolomes of CDI and non-CDI patients against FMT donors who are medically assessed as healthy and are actively recruited to treat recurrent and severe CDI. To analyse AAD microbiota variation, we generated two models that controlled for the period of antibiotic exposure and the number of antibiotic classes. AAD patients treated for different periods or with an increasing number of antibiotic classes did not yield significant between-group differences in alpha diversity (Fig.1a, b). However, analysis of taxonomic composition showed that in total, 35% (61/169) of samples had an elevated abundance of Enterococcus, comprising 25-99% of the gut microbiota, and that the mean abundances of Enterococcus increased with extended periods of antibiotic exposure (3 days) and an increasing number of antibiotic classes (2 classes) (Fig.1c, d). Enterococcal-dominant AAD specimens were cultured (see methods) and MALDI-TOF mass spectroscopy determined that the predominant species present in these samples was E. faecium. The final analysis was performed on 56 isolates with four removed due to poor quality sequences. The majority of E. faecium isolates identified (64%, 36/56) encoded vancomycin resistance determinants vanA or vanB, with 16% (9/56) encoding vanA, approximately 52% (29/56) encoding vanB, and approximately 3.5% (2/56) encoding both vanA and vanB (Fig.2). Approximately 46% (26/56) of E. faecium isolates identified belonged to the epidemic ST796 clonal group, all of which encoded vanB, including the two isolates that encoded both vanA and vanB (Fig.2). The next most prevalent sequence types included ST18 (10.7%, 6/56), ST1421 (7.1%, 4/56), and ST203 (5.4%, 3/56).

Violin plots of Shannon diversity indices assessed species richness and evenness among a FMT donors (n=20), non-antibiotic AAD (-AAD) (n=29), 12 days (n=29), 34 days (n=36), 57 days (n=28) and 8 days (n=37) antibiotic treatment groups, and, b FMT donors (n=20), non-antibiotic AAD (-AAD) (n=29), 1 class (n=49), 2 classes (n=44), 3 classes (n=29) and 4 antibiotic classes (n=14) treatment groups. Mean abundance of major genera colour coded and presented as stacked bar graphs present in c FMT donors, non-antibiotic AAD (-AAD), 12 days, 34 days, 57 days and 8 days antibiotic treatment groups, and, d FMT donors, non-antibiotic AAD (-AAD), 1 class, 2 classes, 3 classes and 4 antibiotic classes treatment groups. In panels a and b, data are presented as meanSD. Statistical significance was determined at p<0.05 and comparisons used KruskalWallis tests with FDR adjusted for multiple comparisons using the Benjamini and Hochberg method. Source data for panels are provided as a Source Data file.

Core genome phylogeny, sequence types, and the presence of vancomycin resistance genes vanA and vanB were determined using Nullabor v2.0 pipeline (https://github.com/tseemann/nullarbor). Analysis was performed against the reference strain, E. faecium Ef_aus00233. In the MLST column, each colour presents a visual representation of sequence type diversity. In the vanA and vanB columns, green denotes gene presence, and symbol denotes gene absence.

To determine microbiota and metabolome associations with low diversity enterococcal-dominant AAD, we stratified HAD patients into non-antibiotic (-AAD), non-enterococcal dominant antibiotic-associated diarrhoea (-Ent AAD), and enterococcal-dominant antibiotic-associated diarrhoea (+Ent AAD) groups (Fig.3a). Only patients in which the Enterococcus 16S rRNA gene amplicon sequences contributed 25% of the total microbiota were included in the +Ent AAD group.

a Summary of the HAD and antibiotic-associated diarrhoeal (+AAD) patient cohorts stratified by enterococcal proliferation. Non-antibiotic AAD (-AAD), AAD without enterococcal proliferation (-Ent+AAD) and AAD with enterococcal proliferation (+Ent AAD) whose microbiota comprised 25-99% of Enterococcus OTUs. b Violin plot of Shannon diversity indices assessed species richness and evenness among FMT donors (n=20), -AAD (n=30), -Ent AAD (n=76) and +Ent AAD (n=61) patients. Alpha diversity was estimated from Shannon diversity index (OTU abundances rarefied to 1107 reads). Statistical significance was determined at p<0.05 and comparisons used Kruskal-Wallis tests with FDR adjusted for multiple comparisons using the Benjamini and Hochberg method. Source data provided as a Source Data file. c PCoA plot based on the Bray-Curtis dissimilarity assessed microbiota differences of FMT donors (n=20), -AAD (n=30), -Ent AAD (n=76) and +Ent AAD (n=61) patients (R2=0.328, p<0.001). Statistical significance was determined at p<0.05 by PERMANOVA. The F statistic two-tailed p-value depicts the significance of the host factor in affecting the community structure, while the PERMANOVA statistic R2 depicts the fraction of variance explained by each factor.

+Ent AAD patients formed a microbially distinct subset of AAD characterised by lower diversity. Plots of alpha diversity measurements showed that compared to FMT donors, all diarrhoeal groups (-AAD, -Ent AAD and +Ent AAD) were associated with a considerable spread of alpha diversity values and significantly lower mean Shannon indices (all p<0.0001) (Fig.3b). While there was no significant difference between -AAD and -Ent AAD (p>0.05), the mean Shannon index for +Ent AAD patients was significantly reduced compared to -AAD and -Ent AAD patients (all p<0.0001) (Fig.3c). Furthermore, the ordination plot visualising Bray-Curtis dissimilarities highlighted that +AAD patients formed two distinct clusters, with -Ent AAD patients clustered left with -AAD patients while +Ent AAD patients clustered right (Fig.3c). Pairwise post-hoc PERMANOVA determined that the difference in distribution of centroids for -AAD vs -Ent AAD was insignificant (R2=0.010, p=0.415), but significant for -Ent AAD vs +Ent AAD (R2=0.325, p<0.0001).

The heatmap of the mean abundance of these 97 metabolites that best described the variation in the enterococcal-dominance PLS-DA model revealed the +Ent AAD metabolome was elevated across several classes (alcohols, amines, amino acids, primary bile acids, and sugars) and depleted in indoles, fatty acids, and phenylpropanoic acids compared to FMT donors, -AAD and -Ent AAD patients (Supplementary Fig.1A).

The PLS-DA scores plot in Supplementary Fig.1B showed that while FMT donors separated from -AAD, -Ent AAD, and +Ent AAD patients, there was no clear separation between the diarrhoeal groups. However, +Ent AAD patients (yellow) clustered further away from FMT donors and pairwise PLS-DA revealed that -Ent AAD vs +Ent AAD metabolomes were significantly different (R2Y=0.564, Q2=0.472 and p=2.031 10-14) (Supplementary Fig.1B).

Plots of mean SCFA concentrations (acetate, propionate and butyrate) revealed that compared to FMT donors, the -Ent AAD and +Ent AAD faecal metabolomes were significantly depleted in all SCFAs (Supplementary Fig.1C, E). While there was no significant difference in faecal acetate between -Ent AAD and +Ent AAD patients, the +Ent AAD metabolome was significantly reduced in propionate (p=0.037) and butyrate (p<0.0001) concentrations (Supplementary Fig.1C, E).

Individual metabolites were further assessed for their capacity to distinguish between diarrhoeal groups using the receiver operating characteristics area under the curve (ROC-AUC). Based on the AUC0.70 cut-off, several metabolites, including the amino acid L-tyrosine and its derivative desaminotyrosine, differentiated between -Ent AAD and +Ent AAD (Supplementary Table2).

We noted that while the +Ent AAD metabolome was elevated in several amino acids compared to -Ent AAD, L-tyrosine was the only amino acid significantly depleted in the +Ent AAD metabolome (p<0.0001), with a mean abundance similar to FMT donors (Fig.4a). Univariate AUC biomarker analysis revealed that reduced L-tyrosine was a possible biomarker of +Ent AAD (AUC=0.79) (Fig.4b).

a Dot plot of L-tyrosine abundance. b L-tyrosine AUC plot differentiating between -Ent AAD (n=59) from +Ent AAD (n=51) patients. c Dot plot of tyramine abundance. d Tyramine AUC plot differentiating between Ent AAD (n=59) from +Ent AAD (n=51) patients. e Dot plot of tyramine/tyrosine ratios. f Tyramine/tyrosine ratios AUC plot differentiating between -Ent AAD (n=59) from +Ent AAD (n=51) patients. Data presented as meanSD in panels a, c and e for FMT donors (n=20), -AAD (n=23), -Ent AAD (n=59) and +Ent AAD (n=51) patients. Statistical significance was determined at p<0.05 and comparisons used Kruskal-Wallis tests with FDR adjusted for multiple comparisons using the Benjamini and Hochberg method. Source data is provided as a Source Data file.

Several bacterial by-products of L-tyrosine metabolism were further analysed, with tyramine of particular interest. Decarboxylation of L-tyrosine into tyramine in the gut is associated with several genera but mainly Enterococcus, particularly E. faecium and E. faecalis31. While tyramine was significantly elevated in the +Ent AAD metabolome compared to FMT donors (p=0.003), there was no significant difference between -Ent AAD and +Ent AAD patients (Fig.4c). Furthermore, univariate AUC biomarker analysis revealed that with an AUC<0.60, tyramine was a poor biomarker differentiating +Ent AAD from -Ent AAD (Fig.4d).

The ratio of tyramine to L-tyrosine was calculated for each sample to investigate whether depleted L-tyrosine and elevated tyramine might signify enterococcal utilisation. The tyramine/tyrosine ratio was significantly higher for +Ent AAD patients compared to -Ent AAD patients (p<0.0001) (Fig.4e), and tyramine/tyrosine ratios performed substantially better in differentiating +Ent AAD with an AUC>0.80 than tyrosine or tyramine alone (Fig.4f).

Toxigenic C. difficile was detected in -AAD, -Ent AAD and +Ent AAD patients (Supplementary Fig.2A), however, our analyses showed a lack of genus-level microbiota difference between CDI and non-CDI patients (Supplementary Fig.2b, d). Despite this, we hypothesised that their metabolomes might present CDI specific-biomarkers. A heatmap of 88 metabolites that best described the variation between CDI and non-CDI patients revealed that -AAD+CDI and -Ent+CDI metabolomes were associated with reduced sugars, sugar alcohols and amino acids compared to their non-CDI counterparts (Fig.5a). Conversely, the +Ent+CDI and +Ent-CDI metabolomes were similarly enriched in a greater number of metabolites across several compound classes, including alcohols, amines, amino acids, bile acids, and sugars, and reduced in indoles, fatty acids, and phenylpropanoic acids (Fig.5a).

a Heatmap of metabolite abundances detected by untargeted GC-MS profiling that differentiated FMT donors (n=18), -AAD-CDI (n=15), -AAD+CDI (n=7), -Ent-CDI (n=48), -Ent+CDI (n=11), +Ent-CDI (n=42) and +Ent+CDI (n=9) patients. All metabolites were normalised, Pareto scaled, and log-transformed. Metabolites with VIP scores > 1.0 and p(corr) values>0.5 and<0.5 were identified as a subset of metabolites with the highest potential as biomarkers. For detailed VIP and p(corr) values, see Source Data file. Each cell corresponded to the mean abundance for each metabolite per group. Dark grey indicated the lowest and red the highest value. b PLS-DA scores plot for FMT donors (purple), -AAD-CDI (light blue), -AAD+CDI (dark blue), -Ent-CDI (red), -Ent+CDI (yellow), +Ent-CDI (green) and +Ent+CDI (orange) patients. Each point represented an individual specimen. Model cross-validation (R2Y=0.244, Q2=0.057, p=0.080 CV-ANOVA). See Source Data file for all model details. c Dot plot of acetate concentrations (g per mg of fresh weight specimen (FW). d Dot plot of butyrate concentrations (g per mg of fresh weight specimen (FW). SCFAs GC-MS profiling data are presented as meanSD in panels c and d for FMT donors (n=20), -AAD-CDI (n=21), -AAD+CDI (n=6), -Ent-CDI (n=56), -Ent+CDI (n=10), +Ent-CDI (n=49) and +Ent+CDI (n=7) patients. In panels c and d, statistical significance was determined at p<0.05 and comparisons used Kruskal-Wallis tests with FDR adjusted for multiple comparisons using the Benjamini and Hochberg method. Source data provided as a Source Data file.

While the principal component scores plot could not distinguish between CDI and non-CDI groups, multivariate ROC-AUC analysis determined that the PLS-DA classification model showed moderate to high specificity and sensitivity in differentiating each non-CDI and CDI group (AUC>0.70) (Supplementary Table3). Furthermore, between-group differences assessed by pairwise PLS-DA analysis found that while +Ent-CDI and +Ent+CDI metabolomes were not significantly different (p=1.000), the difference between -AAD-CDI vs -AAD+CDI approached statistical significance (p=0.060) and was statistically significant between -Ent-CDI and -Ent+CDI (p=0.005) (Supplementary Table3).

In addition, plots of mean SCFA concentrations (acetate, propionate and butyrate) derived from SCFA profiling revealed that -AAD+CDI and -Ent+CDI patients were elevated in acetate and butyrate, with mean concentrations similar to FMT donors (Fig.5c, d). In contrast, -Ent-CDI and +Ent-CDI patients were significantly reduced in acetate compared to FMT donors, but the only significant between-group difference observed was between -Ent+CDI and +Ent-CDI patients (p=0.016) (Fig.5d). -AAD-CDI, -Ent-CDI, +Ent-CDI and +Ent+CDI patients were significantly depleted in butyrate. Similarly, the only significant between-group difference observed was between -Ent+CDI and +Ent-CDI patients (p=0.013) (Fig.5d). Univariate AUC biomarker analysis determined that acetate and butyrate, with AUC values>0.70, were important biomarkers that could differentiate -Ent+CDI patients from -Ent-CDI patients (Supplementary Fig.3a, b).

Using ROC-AUC, individual metabolites were further assessed for their capacity to distinguish between CDI and non-CDI groups. Supplementary Tables4-6 detail the metabolites that distinguished between -AAD-CDI and -AAD+CDI, -Ent-CDI and -Ent+CDI, and +Ent-CDI and +Ent+CDI. We detected several Stickland by-products, including 5-aminovaleric acid (from L-proline), 4-methylvaleric acid (4-MPA) (from L-leucine), isovalerate (from L-leucine), isobutyrate (from L-valine) and desaminotyrosine (from L-tyrosine) (Supplementary Tables46). Non-enterococcal CDI patients were also associated with elevated indole/tryptophan ratios (Supplementary materials and Supplementary Fig.4).

5-aminovaleric acid was elevated in -Ent+CDI and +Ent+CDI patients compared to FMT donors and their non-CDI counterparts, but between-group differences were not significant (Fig.6a). However, univariate AUC biomarker analysis determined that 5-aminovaleric acid was a potential biomarker differentiating +Ent+CDI from +Ent-CDI patients (AUC=0.735) (Fig.6b). We calculated the ratio of 5-aminovaleric acid to proline for each individual to investigate whether together, depleted proline and elevated 5-aminovaleric acid might signify C. difficile utilisation. Mean 5-aminovaleric acid/proline ratios were reduced in non-CDI patients compared to their CDI counterparts but the differences in 5-aminovaleric acid/proline ratios between CDI groups and their non-CDI counterparts were not statistically significant (Fig.6c). However, univariate AUC biomarker analysis determined that 5-aminovaleric acid/proline ratios performed similarly as a biomarker of -Ent+CDI (AUC=0.718) (Fig.6d) as 5-aminovaleric acid alone.

a Dot plot of 5-aminovaleric acid abundance as detected by untargeted GC-MS profiling. b 5-aminovaleric acid AUC plot differentiating +Ent-CDI (n=42) from +Ent+CDI (n=9) patients. c Dot plot of 5-aminovaleric/L-proline ratios. d 5-aminovaleric acid/L-proline ratios AUC plot differentiating +Ent-CDI (n=42) from +Ent+CDI (n=9) patients. Data presented as meanSD in panels a and c for FMT donors (n=18), -AAD-CDI (n=15), -AAD+CDI (n=7), -Ent-CDI (n=48), -Ent+CDI (n=11), +Ent-CDI (n=42) and +Ent+CDI (n=9) patients. In panels a and c, statistical significance was determined at p<0.05 and comparisons used Kruskal-Wallis tests with FDR adjusted for multiple comparisons using the Benjamini and Hochberg method. Source data provided as a Source Data file.

Compared to FMT donors, 4-MPA was elevated with antibiotic usage and enterococcal dominance and significantly elevated in -AAD-CDI (p=0.040) and -Ent+CDI patients (p=0.040) but between-group differences for -AAD-CDI vs -AAD+CDI and -Ent-CDI vs -Ent+CDI patients were not significant (Fig.7a). However, univariate ROC-AUC determined that 4-MPA approached significance as a biomarker differentiating -Ent+CDI from -Ent-CDI patients (AUC=0.682) (Fig.7b). We calculated the ratio of L-leucine and 4-MPA for each individual to investigate whether together, depleted L-leucine and elevated 4-MPA might signify C. difficile utilisation. Compared to FMT donors, 4-MPA/L-leucine ratios were elevated in all groups, except -Ent+CDI patients who shared similarly reduced mean 4-MPA/L-leucine ratios as FMT donors (Fig.7c). The mean 4-MPA/L-leucine ratio was significantly elevated in -Ent+CDI compared to FMT donors (p=0.012) and -Ent-CDI patients (p=0.036) (Fig.7c). Univariate AUC analysis determined that 4-MPA/L-leucine ratios performed substantially better in differentiating -Ent+CDI from -Ent-CDI (AUC>0.800) (Fig.7d), than 4-MPA alone.

a Dot plot of 4-MPA abundance as detected by untargeted GC-MS profiling. b AUC plot of 4-MPA differentiating +Ent-CDI (n=42) from +Ent+CDI (n=9) patients. c Dot plot of 4-MPA/L-leucine ratios. d AUC plot of 4-MPA/L-leucine ratios differentiating +Ent-CDI (n=42) from +Ent+CDI (n=9) patients. e Dot plot of isovalerate concentrations as detected by SCFA GC-MS profiling. f Isovalerate AUC plot differentiating +Ent-CDI (n=49) from +Ent+CDI (n=7). g Dot plot of isobutyrate concentrations as detected by SCFA GC-MS profiling. h Isobutyrate AUC plot differentiating +Ent-CDI (n=49) from +Ent+CDI (n=7). Untargeted GC-MS profiling data is presented as meanSD in panels a and c for FMT donors (n=18), -AAD-CDI (n=15), -AAD+CDI (n=7), -Ent-CDI (n=48), -Ent+CDI (n=11), +Ent-CDI (n=42) and +Ent+CDI (n=9) patients. SCFA GC-MS profiling data is presented as meanSD in panels e and g for FMT donors (n=20), -AAD-CDI (n=21), -AAD+CDI (n=6), -Ent-CDI (n=56), -Ent+CDI (n=10), +Ent-CDI (n=49) and +Ent+CDI (n=7). In panels a, c, e and g, statistical significance was determined at p<0.05 and comparisons used Kruskal-Wallis tests with FDR adjusted for multiple comparisons using the Benjamini and Hochberg method. Source data provided as a Source Data file.

In a separate analysis, SCFA profiling via GC-MS detected L-leucine and L-valine Stickland fermentation products, isovalerate and isobutyrate. Compared to FMT donors, the isovalerate and isobutyrate mean concentration decreased among non-CDI patients with antibiotic treatment and enterococcal dominance but were elevated in -AAD+CDI and -Ent+CDI patients, similar to FMT donors (Fig.7e, f). However, analysis of CDI and non-CDI between-group differences determined only -Ent-CDI compared to -Ent+CDI patients were significantly reduced in isovalerate and isobutyrate (p=0.002 and p=0.0004, respectively) (Fig.7e, f). Univariate ROC-AUC analysis determined isovalerate (AUC=0.830) and isobutyrate (AUC=0.886) as significant biomarkers differentiating -Ent+CDI patients from -Ent-CDI patients (Fig.7g, h).

Desaminotyrosine was reduced in all diarrhoeal groups compared to FMT donors except for -Ent+CDI patients with a mean abundance that exceeded that of FMT donors (p=0.014) and their non-CDI counterparts (p=0.129) (Fig.8a). Univariate AUC biomarker analysis determined that desaminotyrosine was a significant biomarker differentiating -Ent+CDI patients from -Ent-CDI patients (AUC=0.720) (Fig.8b). We calculated the ratio of L-tyrosine and desaminotyrosine for each individual to investigate whether together, depleted L-tyrosine and elevated desaminotyrosine might signify C. difficile utilisation. Compared to FMT donors, desaminotyrosine/L-tyrosine ratios were reduced in all diarrhoeal groups, except in -AAD+CDI and -Ent+CDI groups who shared a similarly elevated mean desaminotyrosine/L-tyrosine ratio as FMT donors (Fig.8c). The difference in 5- desaminotyrosine/L-tyrosine ratios between CDI groups and their non-CDI counterparts was statistically significant between -Ent-CDI and -Ent+CDI groups (p=0.025) (Fig.8c). Univariate AUC biomarker analysis found desaminotyrosine/tyrosine ratios performed better in differentiating Ent+CDI patients (AUC=0.807) than desaminotyrosine on its own (Fig.8d).

a Dot plot of desaminotyrosine abundance. b Desaminotyrosine AUC plot differentiating -+Ent-CDI (n=42) from +Ent+CDI (n=9) patients. c Dot plot of desaminotyrosine/L-tyrosine ratios. d Desaminotyrosine/L-tyrosine ratios AUC plot differentiating +Ent-CDI (n=42) from +Ent+CDI (n=9) patients. Data presented as meanSD in panels a and c for FMT donors (n=18), -AAD-CDI (n=15), -AAD+CDI (n=7), -Ent-CDI (n=48), -Ent+CDI (n=11), +Ent-CDI (n=42) and +Ent+CDI (n=9) patients. Statistical significance was determined at p<0.05 and comparisons used Kruskal-Wallis tests with FDR adjusted for multiple comparisons using the Benjamini and Hochberg method. Source data provided as a Source Data file.

See the original post here:
Multi-omics analysis of hospital-acquired diarrhoeal patients reveals ... - Nature.com