A strategy for modelling dynamic responses in metabolic samples characterized by GC/MS Pa¨r Jonsson, a Hans Stenlund, a Thomas Moritz, b Johan Trygg, a Michael Sjo¨stro¨m, a Elwin R. Verheij, c Johan Lindberg, d Ina Schuppe-Koistinen, d and Henrik Antti a, * a Research Group for Chemometrics, Organic Chemistry, Department of Chemistry, Umea ˚ University, SE-901 87 Umea ˚, Sweden b Umea ˚ Plant Science Center, Department of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences, SE-901 87 Umea ˚, Sweden c TNO Pharma, PO Box 3603700 AJ Zeist, Netherlands d Molecular Toxicology, Safety Assessment, AstraZeneca R&D, SE-141 85 So ¨derta ¨lje, Sweden Received 7 March 2006; accepted 10 May 2006 A multivariate strategy for studying the metabolic response over time in urinary GC/MS data is presented and exemplified by a study of drug-induced liver toxicity in the rat. The strategy includes the generation of representative data through hierarchical multivariate curve resolution (H-MCR), highlighting the importance of obtaining resolved metabolite profiles for quantification and identification of exogenous (drug related) and endogenous compounds (potential biomarkers) and for allowing reliable comparisons of multiple samples through multivariate projections. Batch modelling was used to monitor and characterize the normal (control) metabolic variation over time as well as to map the dynamic response of the drug treated animals in relation to the control. In this way treatment related metabolic responses over time could be detected and classified as being drug related or being potential biomarkers. In summary the proposed strategy uses the relatively high sensitivity and reproducibility of GC/MS in combination with efficient multivariate curve resolution and data analysis to discover individual markers of drug metabolism and drug toxicity. The presented results imply that the strategy can be of great value in drug toxicity studies for classifying metabolic markers in relation to their dynamic responses as well as for biomarker identification. KEY WORDS: GC/MS; metabolomics; metabonomics; batch modelling; toxicology; hepatotoxicity; curve resolution; chemometrics. 1. Introduction The measurement and modelling of endogenous metabolite fluxes in biological fluids and tissues have found widespread application in the study of toxico- logical events in animal models (Lindon et al., 2003; Robertson, 2005) and is now starting to press forward in the areas of clinical diagnosis and biomarker identifi- cation of drug effects and toxicity (Brindle et al., 2002; Nicholson et al., 2004; van der Greef et al., 2004). In order to continue on the path towards understanding and predicting complex biological processes, higher demands will be put on analytical chemistry techniques as well as data processing, modelling and interpretation. As the development of more sensitive high-throughput analytical techniques for metabolite profiling is accel- erating, issues such as data handling, information extraction, interpretation, validation and compound identification will become even more important in order to utilize the potential of the acquired data, but also to manifest the link between the data and the biology of the investigated system. At this stage of the development it is important to realize that it might not always be the most sensitive analytical technique that will produce the most reliable and biologically relevant results. This can be due to issues related to analytical reproducibility as well as the fact that there is still a lot of work to be done with identifying compounds (metabolites) and to decide their significance in a statistical and biological context. Logically, and likely to be the case in the future, the more metabolites that can be detected by an analytical technique, the more information about the studied sys- tem should be retained. However, the reality today is that the lack of control of the data acquired by the high sensitivity techniques greatly increases the risk for pro- ducing spurious results when comparing samples. This is a problem that recently has been highlighted and dis- cussed in the area of clinical diagnosis (Baggerly et al., 2004). In our opinion a better foundation for successful results and increased understanding of complex biolog- ical systems is to be able to, in a reproducible way, detect better resolved and, to a higher extent, identified compounds. In drug toxicity studies this will have implications for detection and identification of drug metabolites and biomarkers, for facilitating the def- inition of biomarker significance and for aiding mechanistic explanations. NMR, GC/MS and LC/MS are the most frequently used instrumental platforms for metabolite profiling of * To whom correspondence should be addressed. E-mail: henrik.antti@chem.umu.se Metabolomics, Vol. 2, No. 3, September 2006 (Ó 2006) DOI: 10.1007/s11306-006-0027-1 135 1011-372X/06/0900–0135/0 Ó 2006 Springer ScienceþBusiness Media, Inc.