Evidence of different metabolic phenotypes in humans

Abstract

The study of metabolic responses to drugs, environmental changes, and diseases is a new promising area of metabonomic research. Metabolic fingerprints can be obtained by analytical techniques such as nuclear magnetic resonance (NMR). In principle, alterations of these fingerprints due to appearance/disappearance or concentration changes of metabolites can provide early evidences of, for example, onset of diseases. A major drawback in this approach is the strong day-to-day variability of the individual metabolic fingerprint, which should be rather called a metabolic "snapshot." We show here that a thorough statistical analysis performed on NMR spectra of human urine samples reveals an invariant part characteristic of each person, which can be extracted from the analysis of multiple samples of each single subject. This finding (i) provides evidence that individual metabolic phenotypes may exist and (ii) opens new perspectives to metabonomic studies, based on the possibility of eliminating the daily "noise" by multiple sample collection.

Fig. 1.

Projection of the one-dimensional ¹H NMR spectral buckets into PCA/CA subspace in the three most significant dimensions. Convex hulls of the donor-specific point clusters (37–41 points each) are shown for better visualization. Each individual donor has his/her own color code.

Fig. 2.

Dendrogram relative to cluster analysis on the 21-dimensional PCA/CA subspace. Clustering according to donor is obvious.

Fig. 3.

MC/TSV results. MM-SIMCA (a and b), PCA/K-NN (c and d), and PCA/CA/K-NN (e and f) classification approaches using the bucket table, and PCA/CA/K-NN classification approach using the signal intensity table from the 12 metabolites (g and h). (Left) Classification probabilities of individual test-set spectra into donor-specific groups. (Right) Classification probability of groups of test-set spectra into donor-specific groups by using MRC.

Fig. 4.

PCA/CA/K-NN learning curves: Box plots reporting the probability of correct classification as a function of number of spectra in model set. (a) Single-spectrum classification. (b) Classification using seven spectra and MRC. Probabilities were determined by MC/TSV using 1,000 individual runs for averaging.

Fig. 5.

Reconstruction of the one-dimensional ¹H NMR spectra by back-projection into the original space of their representations in the 21-dimensional discriminating subspace. (a) Image of the original bucket table. (b) Image of the table reconstructed from the bucket spectra representation in the 21-dimensional donor-discriminating PCA/CA subspace. (c) Median reconstructed spectra for individual donors (red, female; blue, male).