Metabolomics
Metabolites are the substrates, products and signals of biochemistry: sugars, amino acids, nucleotides, lipids, cofactors, drugs and microbial products. They change within seconds and integrate genes, environment, diet and microbiome, which makes them the omics layer closest to phenotype.
Research questions
- Which metabolites or pathways distinguish disease, and can they be biomarkers?
- How do genes (mQTLs), diet, drugs and gut microbes shape circulating metabolites?
- Where does carbon flow through central metabolism in a tumour or immune cell (isotope tracing)?
- What is this unknown feature? (structure elucidation is the bottleneck of the field)
- What has a person been exposed to (exposome)?
Technologies
| Platform | Strengths | Weaknesses | Typical use |
|---|---|---|---|
| LC-MS (reversed phase, HILIC) | Broad coverage, sensitive, MS/MS for identification | Ion suppression, batch drift, many features per metabolite (adducts, isotopes, in-source fragments) | Untargeted discovery, lipidomics |
| GC-MS | Reproducible retention, large EI spectral libraries (NIST) | Needs derivatisation and volatility | Central carbon metabolism, organic acids |
| NMR | Quantitative, non-destructive, highly reproducible, structural information | ~100× less sensitive than MS | Large cohorts (e.g. Nightingale lipoprotein panels), flux |
| Targeted triple-quad MRM | Absolute quantification with standards | Only predefined metabolites | Validation, clinical assays (Biocrates kits) |
| MS imaging (MALDI, DESI) | Spatial distribution in tissue | Identification and quantification hard | Spatial metabolomics |
How metabolomic data are generated
- Collect & quenchStop metabolism instantly: cold methanol, liquid nitrogen; process plasma within ~30 min on ice
- ExtractMethanol/acetonitrile for polar metabolites; MTBE or Folch/Bligh–Dyer for lipids; add internal standards
- PrepareDry and reconstitute; derivatise for GC-MS; make a pooled QC from all samples
- Run orderRandomise samples; inject blanks and pooled QCs every 5–10 samples; MS/MS on pooled QCs for identification
- LC-MS / GC-MS / NMRReversed phase (lipids, non-polar) and HILIC (polar), positive and negative ionisation
- Raw outputVendor files → mzML; NMR free-induction decays → spectra
Untargeted LC-MS workflow
- Design: randomised injection order, pooled QC samples every 5–10 injections, blanks, internal standards, and long-term reference material (NIST SRM 1950 plasma).
- Conversion: vendor files to open formats (mzML) with ProteoWizard msconvert.
- Feature detection: find chromatographic peaks in m/z–retention-time space (XCMS centWave, MZmine 3 ADAP, MS-DIAL, OpenMS FeatureFinderMetabo).
- Alignment and grouping across samples (retention-time correction), gap filling.
- Annotation of feature redundancy: group adducts ([M+H]+, [M+Na]+), isotopes and fragments of the same compound (CAMERA, MS-FLO, IPA).
- Drift and batch correction: QC-based LOESS or random forest (QC-RLSC, SERRF), and filter features with high CV in QCs (>20–30%).
- Identification: match accurate mass, retention time and MS/MS spectra to libraries (MoNA, MassBank, GNPS, NIST, mzCloud) or predict structures (SIRIUS with CSI:FingerID, CANOPUS compound classes, MS-FINDER, MetFrag). Molecular networking (GNPS/FBMN) clusters related spectra so annotations propagate.
- Statistics: log or generalised-log transform, scaling (Pareto, auto), univariate tests with FDR, multivariate PCA, PLS-DA and OPLS-DA (with permutation testing and cross-validation; PLS-DA overfits easily).
- Pathway analysis: over-representation on KEGG, SMPDB or Reactome (MetaboAnalyst), or annotation-free mummichog / PIUMet that work directly from m/z values.
Level 1: confirmed with an authentic standard (same RT and MS/MS on the same system). Level 2: probable structure from library spectrum match. Level 3: tentative compound class. Level 4: unknown. Schymanski et al. (2014) added finer levels for high-resolution MS. Always report the level. Most untargeted features remain at level 3–4.
Flux and metabolic models
Lipids are measured with the same instruments but have their own extraction, standards, nomenclature and quantification rules; they are covered in the Lipidomics chapter, with a focus on targeted methods. Stable-isotope tracing follows 13C-glucose or 13C-glutamine into downstream metabolites; isotopologue distributions are corrected for natural abundance (IsoCor, AccuCor) and fluxes estimated with INCA or 13CFLUX2. Genome-scale metabolic models (Recon3D, Human1/Human-GEM, AGORA2 for gut microbes) predict fluxes by flux balance analysis under constraints (COBRApy, COBRA Toolbox, RAVEN); omics data constrain them to tissue or patient-specific models (GIMME, iMAT, tINIT).
Tools
Benchmarks
- The "dark metabolome": most MS/MS spectra in untargeted studies are unannotated. Structure prediction, in silico libraries and ML-based class annotation are active areas.
- Cross-lab reproducibility and harmonisation of feature tables.
- Absolute quantification at scale and combining metabolomics with fluxes.
- Distinguishing host-, microbe- and diet-derived metabolites in integrated host–microbiome models.
- Spatial metabolomics at single-cell resolution.