ORIGINAL ARTICLE Metabolomic analysis of a human oral glucose tolerance test reveals fatty acids as reliable indicators of regulated metabolism Peter Spegel �� �� Anders P. H. Danielsson �� Karl Bacos �� Cecilia L. F. Nagorny �� Thomas Moritz �� Hindrik Mulder �� Karin Filipsson Received: 28 April 2009 / Accepted: 18 August 2009 / Published online: 1 September 2009 �� Springer Science+Business Media, LLC 2009 Abstract Gas chromatography/mass spectrometry-based metabolomics was applied to investigate dynamic changes in the plasma metabolome upon an oral glucose tolerance test (OGTT). The OGTT is a frequently used diagnostic test of glucose homeostasis and diabetes. Diabetes is diagnosed either when glucose levels C7.0 mM in the fasting state or C11.0 mM at 2 h after oral glucose intake. The accuracy of the OGTT would, however, most likely improve if additional variables could be identified. In the present study, plasma samples were drawn every 15 min for 2 h after an oral glucose load of 75 g preceded by an overnight fast in healthy individuals. Blood plasma levels of more than 200 putative metabolites were measured. Multivariate modelling was used to distinguish metabolic regulation due to the glucose challenge from that of other variability. Two data scaling methods were applied, yielding similar results when evaluated by appropriate diagnostic tools. Fatty acid levels were found to be strongly decreased during the OGTT. Also, the levels of amino acids were shown to decrease. However, technical and uninduced biological variations were found to affect the amino acid levels to a greater extent than the fatty acid levels, making the fatty acids more reliable as indicators of metabolic regulation. Levels of several metabolites corre- lated with the quadratic glucose profile and two were found having an inverse correlation. Raw data plots of all iden- tified significantly altered metabolites confirmed the excellent performance of the multivariate models. Using this approach, a better understanding of the metabolic response to an OGTT can be achieved, paving the way for inclusion of other variables describing appropriate meta- bolic control. Keywords Metabolomics Orthogonal projections to latent structures Oral glucose tolerance test Gas chromatography Mass spectrometry 1 Introduction Metabolism of carbohydrates is essential for life. Some cell types, such as neurons, depend solely on carbohydrates as energy source, and will not function properly without it. The regulation of carbohydrate metabolism is complex and involves nutrients as well as neurotransmitters and hor- mones (Heijboer et al. 2006). Difficulties of correctly studying these not yet fully understood pathways depend at least partly on the complex interplay between them, involving many counterregulatory actions. Clearly, stan- dardized conditions are required for blood glucose moni- toring and for this reason the oral glucose tolerance test Electronic supplementary material The online version of this article (doi:10.1007/s11306-009-0177-z) contains supplementary material, which is available to authorized users. P. Spegel �� (&) A. P. H. Danielsson K. Bacos C. L. F. Nagorny H. Mulder K. Filipsson Department of Clinical Sciences, Unit of Molecular Metabolism, Lund University Diabetes Centre, Malmo �� University Hospital, CRC 91:12, Entrance 72 UMAS, 205 02 Malmo, �� Sweden e-mail: Peter.spegel@med.lu.se A. P. H. Danielsson Analytical Chemistry, Faculty of Engineering LTH, Lund University, Lund, Sweden T. Moritz Umea �� Plant Science Center, Swedish University of Agricultural Sciences, Umea, �� Sweden 123 Metabolomics (2010) 6:56���66 DOI 10.1007/s11306-009-0177-z

(OGTT) was developed. The OGTT is widely used both clinically and for research purposes to diagnose diabetes and other metabolic disorders. WHO recommends that the diagnosis of diabetes should be based on either an elevated fasting plasma glucose level (C7.0 mM) or an increased plasma glucose level (C11.0 mM) 2 h after an oral glucose load of 75 g preceded by an overnight fast (WHO 2006). Hormonal regulation of the plasma glucose level after a meal is mainly exerted by insulin. Hence, several experi- mentally derived equations based on the insulin and glu- cose profiles (e.g. minimal model, homeostasis model assessment index (HOMA), and corrected insulin response (CIR)) have been developed for the estimation of insulin secretion and insulin sensitivity in both humans and ani- mals (Hanson et al. 2000). However, the action of insulin comprises a large variety of metabolic pathways and reg- ulation of several classes of plasma metabolites can be expected. Thus, a shortcoming of the study of carbohydrate metabolism is the limitation of measuring only one or a few relevant metabolic factors at the same time. Therefore, the application of metabolic markers other than glucose, indi- vidually or jointly, may potentially increase the under- standing of, as well as increasing the precision and accuracy in, the diagnosis of metabolic disorder. While metabolic regulation to some extent may be found after forming a hypothesis a priori on which metabolites will be affected, a more comprehensive approach would be to perform an unbiased determination of the whole blood plasma metabolome, i.e. metabolomics (Fiehn et al. 2000 Fiehn 2002). Metabolomics assesses the levels and changes of metabolites in biofluids or tissues it is thus an important link between genotypes and pheno- types of organisms, including humans (Fiehn et al. 2000). The total number of metabolites in human plasma is still unknown. Although the number of endogenous metabolites already is very large, different food additives, drugs and environmental contaminants further increase the number of metabolites that can be found in a biological sample. All of the metabolites are structurally different and a large num- ber remains to be identified. Multivariate analysis has proven to be an indispensable tool in the analysis of the large data sets generated in metabolomics and related approaches (Eriksson et al. 2006 Lindon and Nicholson 2008 Trygg et al. 2007 Wiklund et al. 2008). Raw plots of metabolomics data require the same number of dimensions as the number of metabolites found in the samples. Furthermore, the large data sets are often burdened with substantial amounts of noise. Multi- variate models allow for efficient noise reduction as well as reduction in the dimensionality of the data. The reduction of the dimensionality, so that the data may be graphically visualised, greatly facilitates the interpretation. Ideally, a data set containing hundreds of metabolites may be represented by two or three information-dense dimensions. Such greatly simplified representation allows for an effi- cient highlighting of metabolic regulation and identifica- tion of possible biomarkers (Wiklund et al. 2008). A crucial part of multivariate analysis is the choice of data scaling method. This is especially important in metabolo- mics, where the metabolites are present in very different concentrations. Highly abundant metabolites exhibit a much larger absolute variance than metabolites of low abundance and are likely to dominate models. If this is not corrected for by a suitable scaling method, the models would mainly reflect variation of those metabolites. Previously, the metabolic response to an OGTT has been examined by LC/MS (Shaham et al. 2008 Zhao et al. 2009). In these studies, projections to latent structures (PLS) (Shaham et al. 2008 Wold et al. 2001) or the combination of orthogonal signal correction (OSC) and PLS (Wold et al. 1998 Zhao et al. 2009) was used to identify metabolic regulation in the raw data (Zhao et al. 2009) or used to correlate a metabolite profile to insulin sensitivity (Shaham et al. 2008). Hydrophilic metabolites were not detected in one of the studies (Zhao et al. 2009) due to the use of a reversed-phase column. In the other study three different column types were employed in par- allel to cover a wider selectivity range (Shaham et al. 2008). In the present work, gas chromatography/mass spec- trometry (GC/MS)-based metabolomics (Dettmer et al. 2007 Fernandez et al. 2008 Fiehn et al. 2000) was used, to examine both the hydrophilic and the hydrophobic metabolome on a single column. Data analysis was per- formed using orthogonal projections to latent structures (OPLS) (Trygg and Wold 2002 Wiklund et al. 2008) for identification of important metabolic alterations. OPLS is a more direct method for obtaining the same results as for PLS in combination with OSC. Also, models describing gender related variations in the metabolome and variations due to instrumental drift affecting the measured metabo- lome were calculated. These three models were used and interpreted jointly to find metabolic alterations unique to the OGTT. Additionally, an OPLS based approach was applied for identification of non-linear changes in metabolite levels to identify metabolites correlated with or inversely correlated with the blood plasma insulin and glucose profiles. Different scaling methods, UV- and Pa- reto-scaling, for the raw data were used and their out- comes compared. Diagnostic tools such as the S- and SUS-plots (Wiklund et al. 2008) were applied to improve the reliability of the metabolite identifications. It was clearly shown for the present data set that the feasibility to find a unique regulation or a biomarker with multi- variate modelling using a specific scaling method relies on the use of appropriate diagnostic tools. Metabolomics of an oral glucose tolerance test 57 123