Development of an integrated metabolomic profiling approach for infectious diseases research

Abstract

Metabolomic profiling offers direct insights into the chemical environment and metabolic pathway activities at sites of human disease. During infection, this environment may receive important contributions from both host and pathogen. Here we apply an untargeted metabolomics approach to identify compounds associated with an E. coli urinary tract infection population. Correlative and structural data from minimally processed samples were obtained using an optimized LC-MS platform capable of resolving ~2300 molecular features. Principal component analysis readily distinguished patient groups and multiple supervised chemometric analyses resolved robust metabolomic shifts between groups. These analyses revealed nine compounds whose provisional structures suggest candidate infection-associated endocrine, catabolic, and lipid pathways. Several of these metabolite signatures may derive from microbial processing of host metabolites. Overall, this study highlights the ability of metabolomic approaches to directly identify compounds encountered by, and produced from, bacterial pathogens within human hosts.

Fig 1

A global metabolic profiling platform for human urine.

Fig 2. Typical base peak intensity (BPI) chromatograms and contour plots of a urine sample analyzed with different LC columns

(A) Ascentis Express phenyl-hexyl (100 mm × 2.0 mm, 2.7 μm, Supelco) (B) Shim-pack XR-ODS (100 mm × 2.0 mm, 2.2 μm, Shimadzu); (C) Betasil C18 (100 mm × 2.1 mm, 5.0 μm). BPI chromatograms are depicted at left and at right are relative contour plots with ion signal intensity represented by a color scale from light blue (low) to red (high).

Fig 3. Comparison of extracted ion chromatograms (XIC) ofm/z539.0-539.5 from different column types. Green

The fused core phenyl-hexyl column resolves four well-defined metabolite features; Red: The Betasil C18 column (100 mm × 2.1 mm, 5.0 μm, Thermo Scientific) resolves two metabolite features; Blue: The Shim-pack XR-ODS column (100 mm × 2.0 mm, 2.2 μm, Shimadzu) exhibits prolonged retention and resolves two metabolite features. The highest peak intensity was observed with the fused core phenyl-hexyl column, which is nearly twice that of the Shim-pack XR-ODS column, and four times that from the Betasil C18 column.

Fig 4. Typical positive and negative ionization mode BPI chromatograms of a urine sample analyzed under otherwise identical LC/MS conditions

Although overall signal intensity is higher in positive mode, more metabolite features were distinguished in negative ionization mode.

Fig 5. Reproducibility with repeated measures

(A) Global overlay of BPI chromatograms from a pooled QC sample; (B) Local zoom overviews of the overlaid BPI chromatogram; (C) Score scatter plots of PCA analyses of pooled QC samples alongside analyzed samples.

Fig 6. Phenotypic differentiation of urine from control subjects and UTI associated with E.coli, mixed Gram positives, or both from urinary culture

(A) 2D score scatter plot resulting from PLS-DA analysis; (B) 3D score scatter plot resulting from PLS-DA analysis.

Fig 7. Global metabolic urinary phenotype differentiation of E.coli UTI patients from controls

(A) Heat map derived from unsupervised hierarchical clustering of metabolite signals (rows) grouped by sample type (columns); (B) Z-score plot for reference group (controls); (C) Z-score plot for the metabolites normalized to the mean of the control samples; (D) Volcano plot mapped by the log2 mean value of patients/controls vs p value of t-test of patients vs controls (x: log 2 mean value of patients/controls; y: p value of t-test of patients vs controls); Metabolites with significantly higher and lower levels among E.coli UTI patients are enclosed by the red or blue squares, respectively. (E) Scatter score plot resulting from PLS_DA model; (F) Scatter score plot resulting from OPLS_DA model. Further goodness of fit analysis for PLS_DA and OPLS_DA are provided in Supplemental Fig 6.

Fig 8. Multiple discovery and validation measures contribute to consensus identification of UTI-associated metabolites

(A) 14 differentiable metabolites were confirmed by loading plot analysis of OPLS_DA data. (B) The S-plot analysis of the same OPLS_DA data identifies 12 differentiable metabolites. (C) VIP plot of data from the OPLS model identifies 13 metabolites shared with either the loading and/or S-plots. (D) A Venn diagram depicts contributions from these three analyses to the consensus list of differentiated metabolites.

Fig 9. Identification of tetrahydroaldosterone-3-glucuronide as a differentiable metabolite in E.coli UTI

(A) The extracted ion chromatogram for m/z 539.28 (B) The matched formula confirmed by MS and MS/MS analysis. (C) MS/MS spectrum. (D) The chemical structure of tetrahydroaldosterone-3-glucuronide. (E) Proposed fragmentation pathway based on collisionally induced dissociation (CID). (F) Overlaid chromatograms reveal the relative distribution of tetrahydroaldosterone-3-glucuronide between controls and E.coli UTI patients. (G) A bar plot of relative tetrahydroaldosterone-3-glucuronide levels between controls and E.coli UTI patients.

Fig 10. Pathway enrichment and topology analysis

A. Summary of pathway impact based on KEGG pathway networks; (B) TCA cycle with a bar plot of 3-carboxy-1-hydroxypropyl-ThPP (p<0.05); (C) Terpenoid backbone biosynthesis with a bar plot of diphosphomevalonate (p<0.05); (D) Amino sugar and nucleotide sugar metabolism with a bar plot of N-acetylneuraminic acid (p<0.05); (E) Starch and sucrose metabolism with a bar plot of tetrahydroaldosterone-3-glucuronide (p<0.05); (F) Arachidonic acid metabolism with a bar plot of 6-keto-prostaglandin F1a (p<0.05); (G) Pentose and glucuronate interconversions with a bar plot of tetrahydroaldosterone-3-glucuronide (p<0.05); (H) Steroid hormone biosynthesis with bar plot of 21-hydroxypregnenolone (p<0.05). Ctrl\: controls; EcUTI: E.coli UTI patient.