Evaluation of 1H NMR metabolic profiling using biofluid mixture design

A strategy for evaluating the performance of quantitative spectral analysis tools in conditions that better approximate background variation in a metabonomics experiment is presented. Three different urine samples were mixed in known proportions according to a {3, 3} simplex lattice experimental design and analyzed in triplicate by 1D (1)H NMR spectroscopy. Fifty-four urinary metabolites were subsequently quantified from the sample spectra using two methods common in metabolic profiling studies: (1) targeted spectral fitting and (2) targeted spectral integration. Multivariate analysis using partial least-squares (PLS) regression showed the latent structure of the spectral set recapitulated the experimental mixture design. The goodness-of-prediction statistic (Q(2)) of each metabolite variable in a PLS model was calculated as a metric for the reliability of measurement, across the sample compositional space. Several metabolites were observed to have low Q(2) values, largely as a consequence of their spectral resonances having low s/n or strong overlap with other sample components. This strategy has the potential to allow evaluation of spectral features obtained from metabolic profiling platforms in the context of the compositional background found in real biological sample sets, which may be subject to considerable variation. We suggest that it be incorporated into metabolic profiling studies to improve the estimation of matrix effects that confound accurate metabolite measurement. This novel method provides a rational basis for exploiting information from several samples in an efficient manner and avoids the use of multiple spike-in authentic standards, which may be difficult to obtain.