Opening up the "Black Box": metabolic phenotyping and metabolome-wide association studies in epidemiology

Abstract

Background: Metabolic phenotyping of humans allows information to be captured on the interactions between dietary, xenobiotic, other lifestyle and environmental exposures, and genetic variation, which together influence the balance between health and disease risks at both individual and population levels.

Objectives: We describe here the main procedures in large-scale metabolic phenotyping and their application to metabolome-wide association (MWA) studies.

Methods: By use of high-throughput technologies and advanced spectroscopic methods, application of metabolic profiling to large-scale epidemiologic sample collections, including metabolome-wide association (MWA) studies for biomarker discovery and identification.

Discussion: Metabolic profiling at epidemiologic scale requires optimization of experimental protocol to maximize reproducibility, sensitivity, and quantitative reliability, and to reduce analytical drift. Customized multivariate statistical modeling approaches are needed for effective data visualization and biomarker discovery with control for false-positive associations since 100s or 1,000s of complex metabolic spectra are being processed.

Conclusion: Metabolic profiling is an exciting addition to the armamentarium of the epidemiologist for the discovery of new disease-risk biomarkers and diagnostics, and to provide novel insights into etiology, biological mechanisms, and pathways.

Figure 1. Large scale metabonomic screening of human populations. Scores plot for a PCA model derived from the ¹H NMR urine data from INTERMAP (n=4,630 participants) colored by population sample

The ¹H NMR measurement for a given urine specimen is plotted as a single point. The plot shows four outlying phenotypes based on latent information extracted from the spectral data sets, attributable to high urinary glucose, trimethylamine-N-oxide, ethanol and acetaminophen. Representative spectra of participants with characteristic spectral profiles for these metabolites are also highlighted.

Figure 2. Flow chart of the data pre-processing and data analysis strategy for large-scale population studies