MINIREVIEW Metabonomics in pharmaceutical R & D John C. Lindon, Elaine Holmes and Jeremy K. Nicholson Biomolecular Medicine, Faculty of Medicine, Sir Alexander Fleming Building, Imperial College London, UK Introduction Metabonomics has been formally defined [1] and can be understood as the comprehensive and simultaneous systematic determination of metabolite levels in whole organisms and their changes over time as a conse- quence of stimuli such as diet, lifestyle, environment, genetic effects, and pharmaceutical interventions, both beneficial and adverse. For mammalian systems, this involves the analysis of biofluids and tissues, and the complex datasets are usually interpreted using chemo- metric techniques [1,2]. The approach builds on meta- bolic analyses using NMR spectroscopy [3,4] and mass spectrometry [5] first reported around 20 years ago, and indeed it goes back to the concept suggested by Pauling et al. in 1971 [6]. In this minireview, the main technologies used in metabonomics are summarized, brief details of the types of samples used are given, and the current phar- maceutical applications of metabonomics are des- cribed, but the important aspect of the measurement of metabolic fluxes using stable isotope labeling is not covered here. Some prospects for the future are dis- cussed later. Metabonomics studies of pharmaceutical relevance generally use biofluids, or cell or tissue extracts. Urine and plasma are easily obtained, essentially noninvasive- ly, and hence can be readily used for disease diagnosis Keywords biomarkers; diagnostics; drug safety; metabonomics; spectroscopy Correspondence J. C. Lindon, Biomolecular Medicine, Faculty of Medicine, Biomedical Sciences Division, Imperial College London, Sir Alexander Fleming Building, South Kensington, London, SW7 2AZ, UK Fax: +44 20 75 943 066 Tel: +44 20 75 943 194 E-mail: j.lindon@imperial.ac.uk (Received 20 October 2006, revised 20 November 2006, accepted 30 November 2006) doi:10.1111/j.1742-4658.2007.05673.x This minireview is based on a lecture given at the First Maga Circe Confer- ence on metabolomics held at Sabaudia, Italy, in March 2006 in which the analytical and statistical techniques used in metabonomics, efforts at stan- dardization and some of the major applications to pharmaceutical research and development are reviewed. Metabonomics involves the determination of multiple metabolites simultaneously in biofluids, tissues and tissue extracts. Applications to preclinical drug safety studies are illustrated by the Consortium for Metabonomic Toxicology, a collaboration involving several major pharmaceutical companies. This consortium was able, through the measurement of a dataset of NMR spectra of rodent urine and serum samples, to build a predictive expert system for liver and kidney toxicity. A secondary benefit was the elucidation of the endogenous bio- chemicals responsible for the classification. The use of metabonomics in disease diagnosis and therapy monitoring is discussed with an exemplifica- tion from coronary artery disease, and the concept of pharmaco-meta- bonomics as a way of predicting an individual’s response to treatment is exemplified. Finally, some advantages and perceived difficulties of the metabonomics approach are summarized. Abbreviations CE, capillary electrophoresis; CLOUDS, classification of unknowns by density superposition; COSY, correlation spectroscopy; CPMG, Carr–Purcell–Meiboom–Gill; CSF, cerebrospinal fluid; DA, discriminant analysis; FT, Fourier transform; IBS, irritable bowel syndrome; LC-PUFA, long chain polyunsaturated fatty acid; MAS, magic angle spinning; OSC, orthogonal signal correction; PCA, principal component analysis; PLS, partial least squares; QC, quality control; SHY, statistical hetero-spectroscopy; STOCSY, statistical total correlation spectroscopy; TOF, time of flight; TSP, trimethylsilylpropionic acid sodium salt; UPLC, ultra performance liquid chromatography. 1140 FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS and in a clinical trials setting for monitoring drug ther- apy. However, there is a wide range of fluids that has been studied, including seminal fluid, amniotic fluid, cerebrospinal fluid, synovial fluid, digestive fluids, blis- ter and cyst fluids, lung aspirates and dialysis fluids. In addition, a number of metabonomics studies have used analysis of intact tissue biopsy samples and their lipid and aqueous extracts [7]. This particular approach can also be used to characterize in vitro cell systems such as tumor cells [8] and tissue spheroids [9]. Metabonomics analytical technologies The principal analytical techniques that are employed for metabonomic studies are based on NMR spectro- scopy and mass spectrometry (MS). MS requires a separation of the metabolic components using either gas chromatography (GC) after chemical derivatiza- tion, or liquid chromatography (LC), with the newer method of ultra performance 1 LC (UPLC) being used increasingly. Some users have advocated direct injec- tion MS especially with the use of Fourier transform mass spectrometers. The use of capillary electrophor- esis (CE) coupled to MS has also shown some prom- ise. Other more specialized techniques such as Fourier transform infra-red (FTIR) spectroscopy and arrayed electrochemical detection have been used in some cases [10,11]. The main limitation of these is the low level of detailed molecular identification that can be achieved. However, the combination of retention time and redox properties can serve as a basis for database searching of libraries of standard compounds and the separation output can also be directed to a mass spectrometer for additional identification experiments. All metabonomics studies result in complex multiva- riate datasets that require visualization software and chemometric and bioinformatic methods for interpret- ation. The aim of these procedures is to produce bio- chemically based fingerprints that are of diagnostic or other classification value. A second stage, crucial in such studies, is to identify the substances causing the diagnosis or classification, and these become the com- bination of biomarkers that define the biological or clinical context. NMR spectroscopy Standard commercial NMR spectrometers can be used for metabonomics, and for large scale pharma- ceutical studies, automatic sample preparation is often employed. This can involve addition of buffer to sta- bilize the pH, and D 2 O as a magnetic field lock signal for the spectrometer. NMR spectra typically take only around 5 min to acquire using robotic flow-injection methods. For large scale studies, barcoded vials con- taining the biofluid can be used and the contents of these can be transferred and prepared for analysis using robotic liquid handling technology into 96-well plates under laboratory information management sys- tem control. Alternatively, for more precious samples or for those of limited volume, conventional 5 mm or capillary NMR tubes are usually used, either individu- ally or using a commercial sample tube changer and automatic data acquisition. The large interfering NMR signal arising from water in all biofluids is eliminated by use of standard NMR solvent suppression pulse sequences. The reference compound used in aqueous media is usually the sodium salt of 3-trimethylsilylpropionic acid (TSP), with the methylene groups deuterated to avoid giving rise to peaks in the 1 H NMR spectrum. Absolute con- centrations can be obtained if the sample contains an added internal standard of known concentration, or if a standard addition of the analyte of interest is added to the sample, or if the concentration of a substance is known by independent means (e.g., many metabolites can be quantified by conventional biochemical assays). Whilst a 1 H NMR spectrum of urine contains thou- sands of sharp lines from predominantly low molecular mass metabolites, blood plasma and serum contain both low and high molecular mass components, and these give a wide range of signal line widths. Broad bands from protein and lipoprotein signals contribute strongly to the 1 H NMR spectra, with sharp peaks from small molecules superimposed on them [12]. Standard NMR pulse sequences, where the observed peak intensities are edited on the basis of molecular diffusion coefficients or on NMR relaxation times, can be used to select only the contributions from macro- molecules, or alternatively to select only the signals from the small molecule metabolites, respectively. It is also possible to use these approaches to investigate molecular mobility and flexibility, and to study inter- molecular interactions such as the reversible binding between small molecules and proteins. The development of high resolution 1 H magic angle spinning (MAS) NMR spectroscopy has allowed the acquisition of high resolution NMR data on small pieces of intact tissues with no pretreatment [7,13]. Rapid spinning of the sample (typically at  4–6 kHz) at an angle of 54.7° relative to the applied magnetic field serves to reduce the loss of information caused by line broadening effects seen in nonliquid samples such as tissues. MAS NMR spectroscopy has straightfor- ward, but manual, sample preparation. NMR spectro- scopy on a tissue sample in an MAS experiment is the J. C. Lindon et al. Metabonomics in pharmaceutical R & D FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS 1141 same as solution state NMR and all common pulse techniques can be employed in order to study meta- bolic changes and to perform molecular structure elu- cidation and molecular dynamics studies. Some typical 1 H NMR spectra are given in Fig. 1 showing the different profiles from rat liver tissue (using MAS NMR spectroscopy), urine and blood plasma. Two-dimensional NMR spectroscopy can be useful for increasing signal dispersion and for elucidating the connectivities between signals, thereby aiding bio- marker identification. Those of principal use include 1 H- 1 H 2D J-resolved spectroscopy, which attenuates the peaks from macromolecules and yields information on the multiplicity and coupling patterns of resonances and 1 H- 1 H spin connectivity experiments known as correlation spectroscopy (COSY) and total correlation spectroscopy (TOCSY), giving information on which hydrogens in a molecule are close in chemical bond terms. Use of other types of nuclei, such as naturally abundant 13 Cor 15 N, or where present 31 P, through heteronuclear correlation experiments, can be import- ant to help assign NMR peaks. These experiments benefit from the use of so-called inverse detection, where the lower sensitivity or less abundant nucleus NMR spectrum (such as 13 C) is detected indirectly using the more sensitive ⁄ abundant nucleus ( 1 H). The commercialization of cryogenic probes where the detector coil and preamplifier (but not the samples) are cooled to around 20K is already proving useful for metabonomics studies. This has provided an improve- ment in spectral signal ⁄ noise ratios of up to a factor of five by reducing the thermal noise in the electronics of the spectrometer. Conversely, a reduction in data acquisition times by up to a factor of 25 become poss- ible for the same amount of sample. NMR spectro- scopy of biofluids detecting the much less sensitive 13 C nuclei which also only have a natural abundance (1.1%) also becomes possible because of the increase in signal-to-noise ratio [14]. This technology also makes the use of tissue-specific microdialysis samples more feasible [15]. A B 7.5 7.0 6.5 6.0 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 δ 1 H 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 δ 1 H 5.56.08.28.4 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 δ 1 H C Fig. 1. (A) 600 MHz standard solvent sup- pression pulse 1 H NMR spectrum of rat urine. (B) 600 MHz Carr–Purcell–Meiboom– Gill (CPMG) 1 H NMR spectrum of rat blood plasma. (C) 600 MHz CPMG 1 H MAS NMR spectrum of the left lateral lobe of a rat liver. Metabonomics in pharmaceutical R & D J. C. Lindon et al. 1142 FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS Mass spectrometry Mass spectrometry coupled to a chromatographic separ- ation has also been widely used in metabolic fingerprint- ing and metabolite identification. Although most studies to date have been on plant extracts and model cell sys- tem extracts [16], its application to mammalian studies is increasing. MS is inherently considerably more sensi- tive than NMR spectroscopy provided the metabolite ionizes, but it is generally necessary to employ different separation techniques for different classes of substances. Analyte quantitation by MS in complex mixtures of highly variable composition can be impaired by variable ionization and ion suppression effects. For plant meta- bolic studies, most investigations have used chemical derivatization to ensure volatility and analytical repro- ducibility, followed by GC-MS analysis. Some approa- ches using MS rely on more targeted studies, for example by detailed analysis of lipids [17]. For metabonomics applications on biofluids, an HPLC chromatogram is generated with MS detection, usually using electrospray ionization, and both positive and negative ion chromatograms can be measured. At each sampling point in the chromatogram there is a full mass spectrum and so the data is three-dimen- sional in nature, i.e., retention time, mass and inten- sity. Given this very high resolution it is possible to cut out any mass peaks from interfering substances such as drug metabolites, essentially without affecting the structure of the dataset. The problem of ion suppression is minimized by improving the efficiency of the chromatography and this has been achieved using UPLC. This is a combina- tion of a 1.7 lm reversed-phase packing material, and a chromatographic system, operating at around 827.4 bar 2 . UPLC provides around a 10-fold increase in speed and a three- to five-fold increase in sensitivity compared to a conventional stationary phase. UPLC- MS has already been used for metabolic profiling of urines in a number of rodent studies [18]. A compar- ison of data generated using both HPLC-MS and UPLC-MS is given in Fig. 2. CE coupled to mass spectrometry has also been explored as a possible technology for metabonomics studies [19]. Metabolites are first separated by CE based on their charge and size and then selectively detected using MS, and the technique has been applied to studies of bacterial growth. HPLC UPLC 2.5 min 5 4 6 8 7.5 10 150 300 450 600 750 m/z 150 300 450 600 750 m/z 9000 7500 6000 4500 3000 1500 0 cm 1750 1500 1000 1250 750 500 250 0 cm Fig. 2. Three-dimensional plots of retention time, m ⁄ z and intensity from control white male mouse urine using (left) HPLC-MS with a 2.1 cm · 100 mm Waters 5 Symmetry 3.5 lm C18 column (Milford, MA, USA), eluted with 0–95% linear gradient of water with 0.1% (v ⁄ v) formic acid:acetonitrile with 0.1% (v ⁄ v) formic acid over 10 min at a flow rate of 0.6 mLÆmin )1 and (right) UPLC-MS with 2.1 cm · 100 mm Waters ACQUITY 1.7 lm C18 column, eluted with the same solvents at a flow rate of 0.5 mLÆmin )1 . In both cases, the column eluent was monitored by ESI orthogonal acceleration 6 -TOF-MS from 50 to 850 m ⁄ z in positive ion mode. Reproduced with permission from Wilson et al. [18]. J. C. Lindon et al. Metabonomics in pharmaceutical R & D FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS 1143 For biomarker identification, it is also possible to separate out substances of interest on a larger scale from a complex biofluid sample using techniques such as solid phase extraction or HPLC. For metabolite identification, directly coupled chromatography-spectro- scopy methods can also be used. The most general of these ‘hyphenated’ approaches is HPLC-NMR-MS [20] in which the eluting HPLC peak is split, with parallel analysis by directly coupled NMR and MS techniques. This can be operated in on-flow, stopped-flow and loop-storage modes, and thus can provide the full array of NMR and MS-based molecular identification tools. These include 2D NMR spectroscopy as well as MS-MS for identification of fragment ions and Fourier transform (FT)-MS or time of flight (TOF)-MS for accurate mass measurement and hence derivation of molecular empirical formulae. Chemometrics methods One common objective in metabonomics is to classify a sample based on identification of inherent patterns of peaks in a dataset (usually a spectrum) and secondly to identify those spectral features responsible for the clas- sification. The approach can also be used for reducing the dimensionality of complex datasets, for example by 2D or 3D mapping procedures, to enable easy visual- ization of any clustering or similarity of the various samples. Alternatively, in what are known as ‘super- vised’ methods, multiparametric datasets can be mod- elled so that the class of separate samples (a ‘validation set’) can be predicted based on a series of mathematical models derived from the original data or ‘training set’. One popular technique that has been used exten- sively in metabonomics is principal components analy- sis (PCA). Each PC is a linear combination of the original data parameters (e.g., intensity values for a range of ion m ⁄ z-values from MS) and each successive PC explains the maximum amount of variance poss- ible, not accounted for by the previous PCs. Each PC is orthogonal and therefore independent of the other PCs and so the variation in the spectral set is usually described by many fewer PCs than comprise the num- ber of original data point values, because the less important PCs describe the noise variation in the spec- tra. Conversion of the data to PCs results in two mat- rices known as scores and loadings. Scores, the linear combinations of the original variables, can be regarded as the new variables, and in a scores plot each point represents a single sample spectrum. The PC loadings, where each point represents a different spectral inten- sity, define the way in which the old spectral variables are linearly combined to form the new variables and show those variables carrying the greatest weight in determining the positions of the points in the scores plot. In addition, there are many other visualization (or unsupervised) methods such as nonlinear mapping and hierarchical cluster analysis. To illustrate PCA, Fig. 3 shows the scores and loa- dings plots for PC1 versus PC2 for data from a series of 1 H NMR spectra of rat urine in a toxicity study. In Fig. 3. Results of a principal components analysis based on NMR spectra of urine from rats treated with control dosing vehicle, or one of the two liver toxins, a-naphthylthioisocyanate (ANIT) or hydrazine. (A) PC scores plot (PC1 versus PC2) where each point corresponds to a single urine sample, showing clear clustering of the samples from control urine and from the toxin-treated animals. The liver toxins form separate clusters because they have different biochemical mechanisms of action and hence different biochemical profiles in the urine. (B) The corresponding PC loadings plot where each point corresponds to a specific NMR spectral region, leading to the possibility of identifying biomarkers of the toxicological effect. For example, in the scores plot, the points corresponding to urines from hydrazine-treated animals appear in the lower left quad- rant and in the corresponding loadings plot, this region indicated that NMR peaks from 2-AA (2-aminoadipate) were important, and thus this is a biomarker of the hydrazine-induced toxicity. Metabonomics in pharmaceutical R & D J. C. Lindon et al. 1144 FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS the scores plot, each point represents a single NMR spectrum and the clustering of points shows the differ- ent biochemical effects of the two different toxins rel- ative to the control group. In cases where samples are collected over time, onset and recovery trajectories can be observed. The loadings plot indicates which regions of the NMR spectra are responsible for the clustering. If a predictive model is required, one widely used supervised method (i.e. using a training set of data with known outcomes) is projection on latent structures (PLS). This is a method that relates a data matrix con- taining variables from samples, such as spectral inten- sity values (an X matrix), to a matrix containing outcome variables (e.g., measurements of response, such as toxicity scores) for those samples (a Y matrix). PLS can also be used to examine the influence of time on a dataset, which is particularly useful for biofluid NMR data collected from samples taken over a time course of the progression of a pathological effect. PLS can also be combined with discriminant analysis (DA) to establish the optimal position to place a discriminant surface that best separates classes. It is important to build and test such chemometric models using independent training data and validation datasets. Extensions of this approach allow the evaluation of those descriptors that are completely independent (orthogonal) to the Y mat- rix of end-point data. This orthogonal signal correction (OSC) can be used to remove irrelevant and confusing parameters and has been integrated into the PLS algo- rithm [21]. If the Y matrix contains continuous data, then PLS regression is a very useful approach. There is a variety of other methods that use nonlinear combinations of the data variables and these include genetic algorithms, machine learning, Bayesian mode- ling and artificial neural networks. In these, a training set of data is used to develop algorithms, which ‘learn’ the structure of the data and can cope with complex functions. For example, probabilistic neural networks have shown promise for predicting toxicity from NMR- based metabonomics data [22]. It should be emphasized that both unsupervised and supervised approaches can be useful in metabonomics. Unsupervised methods provide information on the nat- ural structure of the data, whilst for supervised meth- ods it is vital to carry out proper validation of the models generated, involving training datasets and blind validation and test sets of data. Statistical spectroscopy for biomarker identification Recently, a new method for identifying multiple NMR peaks from the same molecule in a complex mixture, hence providing a new approach to molecu- lar identification, has been introduced. This is based on the concept of statistical total correlation spectro- scopy and has been termed STOCSY [23]. This takes advantage of the colinearity of many of the intensity variables in a set of spectra (e.g., 1 H NMR spectra) so that a pseudo-2D NMR spectrum can be calcula- ted that displays the correlation among the intensities of the various peaks across the whole sample. This method is not limited to the usual connectivities that are deducible from more standard 2D NMR spectro- scopic methods, such as TOCSY. Added information is available by examining lower correlation coeffi- cients or even negative correlations because this leads to connection between two or more molecules involved in the same biochemical pathway. In an extension of the method, the combination of STO- CSY with supervised chemometrics methods offers a new framework for analysis of metabonomic data. In a first step, a supervised multivariate discriminant analysis can be used to extract the parts of NMR spectra related to discrimination between two sample classes. This information is then combined with the STOCSY results to help identify the molecules responsible for the metabolic variation. The applica- bility of the method is illustrated in Fig. 4, where a spin system of two triplets can be noticed at d 2.91 and d 2.51. This spin system is strongly correlated to others resonances in the aromatic region of the spec- trum, although not spin-coupled. Computing only the correlation between one of the data points represent- ing the maximum of one of the triplets, and all the other variables leads to a single vector, which has the size of the number of variables used. Then, by selecting the spectrum with the maximum value of this selected variable, it is possible to plot it with a colour code corresponding to the correlation between the selected resonance and all the other points of the spectra. Correlations can be observed between reso- nances with no NMR-based spin-coupling connectiv- ity. Thus, in the aromatic region, shown in Fig. 4, it is possible to recognize the resonances of a meta- substituted benzene ring (one triplet, two doublets, and one singlet). Thus, this molecule can be identi- fied as a derivative of a meta-substituted phenylprop- anoic acid and is probably 3-hydroxyphenylpropionic acid. The approach is not limited to NMR spectra alone and has been extended to other forms of data. It has recently been applied to coanalysis of both NMR and mass spectra from a metabonomic toxicity study [24]. This allowed better assignment of biomarkers of the toxin effect by using the correlated but complementary J. C. Lindon et al. Metabonomics in pharmaceutical R & D FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS 1145 information available from the NMR and mass spectra taken on a whole sample cohort. Standards and reporting needs in metabonomics Several initiatives have been under way to investigate the reporting needs and standardization of reporting arrangements for metabonomics studies. The Standard Metabolic Reporting Structures group (http:// www.smrsgroup.org) has produced a draft policy document covering all of those aspects of a metabolic study that are recommended for recording, from the origin of a biological sample, the analysis of material from that sample, and chemometric and statistical approaches to retrieve information from the sample data, and a summary publication has appeared [25]. The various levels and consequent detail for reporting needs, including journal submissions, public databases and regulatory submissions have also been addressed. In parallel, a scheme called ArMet for capturing data and metadata from metabolic studies has been pro- posed and developed [26]. These activities were fol- lowed up in August 2005 with a workshop and discussion meeting sponsored by the US National Institutes of Health, from which plans are being devel- oped to define standards in a number of areas relevant to metabonomics, including characterization of sam- ple-related metadata, technical standards and related data, metadata and quality control matters for the analytical instrumentation, data transfer methodologies and schema for implementation of such activities, and development of standard vocabularies to enable trans- parent exchange of data [27]. For details of the current activity in this area the reader is referred to the Meta- bolomics Standards Initiative (http://msi-workgroups. sourceforge.net/). Pharmaceutical R & D metabonomics applications Physiological and gut microfloral effects A good understanding of normal biochemical profiles is a prerequisite for evaluation of metabolic changes caused by xenobiotics or disease. Thus metabonomics has been used to identify metabolic differences, in experimental animals such as mice and rats, caused by a range of inherent and external factors [28]. These dif- ferences may help explain differential toxicity of drugs between strains and interanimal variation within a study. Many effects can be distinguished, including male ⁄ female differences, age-related changes, estrus cycle effects in females, diet, diurnal effects, differenti- ation of wildtype and genetically modified animals, and interspecies differences and similarities using both NMR- and MS-based approaches. The importance of the symbiotic relationship between mammals and their gut microfloral popula- tions has been recognized [29] and highlighted in sev- eral studies. These include a study in which axenic (germ free) rats were allowed to recolonize their gut microflora in normal laboratory conditions with their urine biochemical profiles being monitored for 21 days using 1 H NMR spectroscopy [30], and the combined effects of gut bacteria and gut parasites on metabolic profiles [31]. A B C 1.5 14 12 10 8 6 4 2 0 Intensity (a.u.) –2 109976543210 0.75 0 2.9 0.8 0.4 0 7.4 7.3 7.2 HO HO O 1 0.95 0.85 0.75 0.65 0.7 0.8 0.9 δ 1 H (p.p.m.) r 2 δ 1 H (p.p.m.) 2.8 2.7 2.6 2.5 Fig. 4. One-dimensional STOCSY analysis to identify peaks correlated to that at the chemical shift, d 2.51. The degree of correla- tion across the spectrum has been colour- coded and projected on the spectrum. (A) Full spectrum; (B) partial spectrum between d 7.1–7.5; (C) partial spectrum between d 2.4–3.0. The STOCSY procedure enabled the assignment of this metabolite as 3-hydroxyphenylpropionic acid. Adapted from Cloarec et al. [23] Metabonomics in pharmaceutical R & D J. C. Lindon et al. 1146 FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS The metabolic consequences of early life maternal separation stress have been investigated in rats, either alone or in combination with secondary acute water avoidance stress [32]. The effect of a long chain poly- unsaturated fatty acid (LC-PUFA) enriched dietary intervention (postulated to be beneficial) in stressed animals was also assessed. Systematic changes in meta- bolic biochemistry were evaluated using 1 H NMR spectroscopy of blood plasma and multivariate pattern recognition techniques. The biochemical response to stress was characterized by decreased levels of total lipoproteins and increased levels of amino acids, glucose, lactate, creatine and citrate. Secondary acute water avoidance stress also caused elevated levels of O-acetyl glycoproteins in blood plasma. LC-PUFA dietary enrichment did not alter the metabolic response to stress but did result in a modified lipoprotein pro- file. This work indicated that the different stressor types resulted in some common systemic metabolic responses that involve changes in energy and muscle metabolism, but that they are not reversible by dietary intervention. Irritable bowel syndrome (IBS) is a common multi- factorial intestinal disorder for which the aetiology remains largely undefined, and recently, using a Trichi- nella spiralis-induced model of postinfective IBS, the effects of probiotic bacteria on gut dysfunction have been investigated using metabonomics [33]. Mice were divided into four groups: an uninfected control group and three T. spiralis-infected groups, one as infected control and the two other groups subsequently treated with either Lactobacillus paracasei or L. paracasei-free medium. Plasma, jejunal wall and longitudinal myen- teric muscle samples were collected at day 21 postinfec- tion, and NMR spectroscopy was used to characterize these and plasma metabolic profiles. T. spiralis-infected mice showed an increased energy metabolism, fat mobilization and a disruption of amino acid meta- bolism due to increased protein breakdown, which were related to the intestinal hypercontractility. Increased levels of taurine, creatine and glycerophos- phorylcholine in the jejunal muscles were associated with the muscular hypertrophy and disrupted jejunal functions. L. paracasei treatment normalized the mus- cular activity and the disturbed energy metabolism as evidenced by decreased glycogenesis and elevated lipid breakdown in comparison with untreated T. spiralis- infected mice. Changes in the levels of plasma metabo- lites (glutamine, lysine, methionine) that might relate to a modulation of immunological responses were also observed in the presence of the probiotic treatment. The work suggested that probiotics may be beneficial in patients with IBS. Pre-clinical drug candidate safety assessment In vivo preclinical drug safety assessment remains one of the main bottle-necks in pharmaceutical R & D and is a prime target for improving efficiency in drug development. Having defined the metabolic hyperspace occupied by normal animals, metabonomics can be used for rapid classification of a biofluid sample as normal or abnormal. Classification of the target organ or region of toxicity, the biochemical mechanism of a toxin, the identification of combination biomarkers of toxic effect and evaluation of the time course of the effect, e.g., the onset, evolution and regression of toxicity, can all be determined. There have been many studies using 1 H NMR spectroscopy of biofluids to characterize drug toxicity going back to the 1980s [3], and the role of metabonomics in particular, and magnetic reson- ance in general in toxicological evaluation of drugs has been comprehensively reviewed [34]. The situation is now changing with the introduction of the combined use of NMR spectroscopy and HPLC-MS for toxicity studies. The usefulness of NMR-based metabonomics for the evaluation of xenobiotic toxicity effects has recently been comprehensively explored by the successful Con- sortium for Metabonomic Toxicology. This was con- ducted by five pharmaceutical companies and Imperial College, London, UK [35], and its aim was to develop methodologies for the acquisition of metabonomic data using 1 H NMR spectroscopy of urine and blood serum from rats and mice for preclinical toxicological screening of candidate drugs, to build databases of spectra and to develop an expert system for predicting target organ toxicity. To assess the levels of analytical and biological vari- ation that could arise through the use of metabonom- ics on a multisite basis, a feasibility study was carried out at the start of the project, using the same detailed protocol and using the same model toxin, across all company sites. The biological variability was evaluated by a detailed comparison of the ability of the compan- ies to provide consistent urine and serum samples for an in-life study of the same toxin. There was a high degree of consistency between samples from the various companies and dose-related effects could be distinguished from intersite variation. An intersite NMR analytical reproducibility study also revealed a high degree of robustness giving a multivariate coeffi- cient of regression between paired samples of only about 1.6% [36]. To achieve the project goals, new methodologies for analyzing and classifying the complex datasets were J. C. Lindon et al. Metabonomics in pharmaceutical R & D FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS 1147 developed. The predictive expert system that was developed takes into account the metabolic trajectory over time, and so a new way of comparing and scaling these multivariate trajectories was developed [37]. A novel classification method for identifying the class of toxicity based on all of the NMR data for a given study was also developed. This has been termed ‘Clas- sification Of Unknowns by Density Superposition (CLOUDS)’ [38] and is a novel non-neural implemen- tation of a classification technique developed from probabilistic neural networks. This consortium showed that it is possible to con- struct predictive and informative models of toxicity using NMR-based metabonomic data, delineating the whole time course of toxicity. Curated databases of spectral ( 35 000 NMR spectra) and conventional (clinical chemistry, histopathology, etc.) results for 147 model toxins and treatments that served as the basis for computer-based expert systems for toxicity predic- tion were also produced. All of the project goals inclu- ding provision of multivariate statistical models (expert systems) for prediction of toxicity, initially for liver and kidney toxicity in the rat and mouse were achieved, and the predictive systems and databases were transferred to the sponsoring companies [39]. Clinical pharmaceutical applications Many examples exist in the literature on the use of NMR-based metabolic profiling to aid human disease diagnosis, such as the investigation of diabetes using plasma and urine, neurological conditions such as Alzheimer’s disease using cerebrospinal fluid, arthritis using synovial fluid and male infertility using seminal fluid. In addition, analysis of urine has been used in the investigation of drug overdose, renal transplanta- tion and various renal diseases. NMR spectroscopy of urine and plasma has been used extensively for the diagnosis of inborn errors of metabolism in chil- dren [40]. Most of the earlier studies using NMR spectroscopy have been reviewed previously [41], but more recent studies include cerebrospinal fluid sample analysis using NMR spectroscopy to distinguish var- ious types of meningitis infection (bacterial, viral and fungal) [42] and to investigate subarachnoid haemor- rhage [43]. Human serum samples have been analysed using NMR spectroscopy to develop a diagnostic method for coronary artery disease [44]. One area of disease where progress is being made using NMR- based metabonomics studies of biofluids is cancer. This is highlighted by a publication on the diagnosis of epithelial ovarian cancer based on analysis of serum [45]. Pharmacometabonomics For personalized healthcare, an individual’s drug treat- ments must be tailored so as to achieve maximal effic- acy and avoid adverse drug reactions. One of the approaches has been to understand the genetic make- up of different individuals (pharmacogenomics) and to relate these to their varying abilities to handle pharma- ceuticals both for their beneficial effects and for identi- fying adverse effects. Very recently, an alternative approach to understanding such intersubject variability has been developed using metabonomics, and used to predict the metabolism and toxicity of a dosed sub- stance, based solely on the analysis and modeling of a predose metabolic profile [46]. Unlike pharmaco- genomics, this approach, which has been termed ‘phar- macometabonomics’, is sensitive to both the genetic and modifying environmental influences that determine the metabolic fingerprint of an individual. This new approach has been illustrated with studies of the toxic- ity and metabolism of compounds with very different modes of action (allyl alcohol, galactosamine and acet- aminophen) administered to rats. Integration of -omics results The value of obtaining multiple NMR spectroscopic and ⁄ or LC-MS datasets from various biofluid samples and tissues of the same animals collected at different time points has been demonstrated. This procedure has been termed ‘integrated metabonomics’ [2] and can be used to describe the changes in metabolism in different body compartments affected by exposure to, for exam- ple, toxic xenobiotics. If profiles are obtained over time, they provide extra information and are character- istic of particular types and mechanisms of pathology. Samples from multiple sources give a more complete description of the biochemical consequences than can be obtained from one fluid or tissue alone. Although this review concentrates on metabolic ana- lyses, there is a requirement to integrate information at the transcriptomic, proteomic and metabonomic levels, despite these different levels of biological control showing very different time scales of change. This is because some time courses can be very rapid, such as gene switching, some require much longer time scales, e.g. protein synthesis, or in the case of metabolic chan- ges, can encompass enormous ranges of time scales. Biochemical changes do not always occur in the order, transcriptomic, proteomic, metabolic, because, for example, pharmacological or toxicological effects at the metabolic level can induce subsequent adaptation effects at the proteomic or transcriptomic levels. One Metabonomics in pharmaceutical R & D J. C. Lindon et al. 1148 FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS important potential role for high-throughput and highly automated metabonomics methods therefore could be to direct the timing of more expensive or labour-intensive proteomic and transcriptomic analyses in order to maximize the probability of observing meaningful and relevant biochemical changes using those techniques. In addition, overlaid with this temporal complexity, is the fact that environmental and lifestyle effects have a large effect at all levels of molecular biology. Gene and protein expression effects and metabolite levels can be altered by such factors, and this variation has to be incorporated into any analysis as part of inter- sample and interindividual variation. Even healthy ani- mals and man can be considered as ‘super-organisms’, with an internal ecosystem of diverse symbiotic gut microflora that have metabolic processes that interact with the host and for which, in many cases, the gen- ome is not known. The complexity of mammalian bio- logical systems and the diverse features that need to be measured to allow ‘-omics’ data to be fully interpreted have been reviewed recently [47] and it has been argued that novel approaches will continue to be required to measure and model metabolic processes in various compartments from such global systems with different interacting cell types, and with various geno- mes, connected by cometabolic processes. Integration of metabonomics data with that from other multivariate techniques in molecular biology such as from gene array experiments or proteomics is now becoming a reality. Pharmaceutically related examples include phenotypic differences [48] and toxic- ity studies of acetaminophen [49], bromobenzene [50,51], a-naphthylisothiocyanate [52], hydrazine [53] and methapyriline [54]. Future promise In summary, it is clear that metabonomics will have an impact in pharmaceutical R & D but some potential disadvantages of the approach include the use of mul- tiple analytical technologies with different sensitivities, dynamic ranges and metabolite detection abilities and the complexity of the resulting datasets. Through the inappropriate application of chemometrics, it is poss- ible to over-interpret the data, but this is easily avoi- ded by correct statistical rigour. There remains a need for the regulatory agencies to be trained in the inter- pretation of the data and for the availability of more well trained practitioners generally. However, on the other hand, the analytical proce- dures used are stable and robust, and have a high degree of reproducibility, and although advances will obviously be made in the future, current data will always be readable and interpretable. In contrast to other -omics, metabonomics enjoys a good level of biological reproducibility and the cost per sample and per analyte is relatively low. It has the advantage of not having to preselect analytes, and generally it is minimally invasive with hypothesis generation studies being easily possible. Metabolic biomarkers are closely identifiable with real biological endpoints and provide a global systems interpretation of biological effects, including the interactions between multiple genomes such as humans and their gut microflora. One major potential strength of metabonomics is the possibility that metabolic biomarkers will be more easily used across species than transcriptomic or proteomic bio- markers and this should be important for pharmaceuti- cal studies. For complex disease and drug effect evaluation, combinations of biomarkers are likely to be necessary and thus there will be many opportunities for metabo- nomics that are as yet under-explored, such as its use in environmental toxicity studies, its use in directing the timing of transcriptomic and proteomic experi- ments, and its use for deriving theranostic biomarkers. It will surely be an integral part of any multiomics study where all the datasets are combined in order to derive an optimum set of biomarkers. References 1 Nicholson JK, Lindon JC & Holmes E (1999) Metabo- nomics’: Understanding the metabolic responses of liv- ing systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data. Xenobiotica 29, 1181–1189. 2 Nicholson JK, Connelly J, Lindon JC & Holmes E (2002) Metabonomics: a platform for studying drug toxicity and gene function. Nat Rev Drug Disc 1, 153–162. 3 Nicholson JK & Wilson ID (1989) High-resolution pro- ton magnetic resonance spectroscopy of biological fluids. Prog NMR Spectrosc 21, 449–501. 4 Gartland KPR, Beddell CR, Lindon JC & Nicholson JK (1991) The application of pattern recognition meth- ods to the analysis and classification of toxicological data derived from proton NMR spectroscopy of urine. Mol Pharmacol 39, 629–642. 5 van der Greef J, Tas AC, Bouwman J, Ten Noever de Brauw MC & Schreurs WHP (1983) Evaluation of field- desorption and fast atom-bombardment mass spectro- metric profiles by pattern recognition techniques. Anal Chim Acta 150, 45–52. 6 Pauling L, Robinson AB, Teranishi R & Cary P (1971) Quantitative analysis of urine vapor and breath using J. C. Lindon et al. Metabonomics in pharmaceutical R & D FEBS Journal 274 (2007) 1140–1151 ª 2007 The Authors Journal compilation ª 2007 FEBS 1149 [...]... Jain SJ, Moulder K, Mennerick S, Crowley JR, Townsend RR & Gordon JI (2005) An integrated functional genomics and metabolomics approach for defining poor prognosis in human neuroendocrine cancers Proc Nat Acad Sci USA 102, 9901–9906 Bollard ME, Xu JS, Purcell W, Griffin JL, Quirk C, Holmes E & Nicholson JK (2002) Metabolic profiling of the effects of d- galactosamine in liver spheroids using 1 H NMR and... J, Smith AR, Taylor J, Fiehn O, Goodacre R, Bino RJ, Hall R et al (2004) A proposed framework for the description of plant metabolomics experiments and their results Nature Biotechnol 22, 1601–1606 Castle AL, Fiehn O, Kaddurah-Daouk R & Lindon JC (2006) Metabolomics standards workshop and the development of international standards for reporting metabolomics experimental results Brief Bioinform 7, 159–165... C & Griffin JL (2005) Metabolites from cerebrospinal fluid in aneurysmal subarachnoid haemorrhage correlate with vasospasm and clinical outcome: a pattern-recognition H-1 NMR study NMR Biomed 18, 24–33 44 Brindle JT, Antti H, Holmes E, Tranter G, Nicholson JK, Bethell HWL, Clarke S, Schofield M, McKilligin E, Mosedale DE et al (2002) Rapid and noninvasive diagnosis of the presence and severity of coronary... Clin Biochem 40, 16–24 41 Lindon JC, Nicholson JK & Everett JR (1999) NMR spectroscopy of biofluids Ann Rep NMR Spec 38, 1–88 42 Coen M, O’Sullivan M, Bubb WA, Kuchel PW & Sorrell T (2005) Proton nuclear magnetic resonance-based metabonomics for rapid diagnosis of meningitis and ventriculitis Clin Infect Dis 41, 1582–1590 Metabonomics in pharmaceutical R & D 43 Dunne VG, Bhattachayya S, Besser M, Rae... liver, urine, and plasma in the rat: an integrative metabonomic approach Chem Res Toxicol 14, 1401–1412 53 Kleno TG, Kiehr B, Baunsgaard D & Sidelmann UG (2004) Combination of ‘omics’ data to investigate the mechanism(s) of hydrazine-induced hepatotoxicity in rats and to identify potential biomarkers Biomarkers 9, 116–138 54 Craig A (2004) Parallel metabonomic and genomic characterisation of experimental... induced hepatic necrosis Toxicol Pathol 33, 425–433 51 Waters NJ, Waterfield CJ, Farrant RD, Holmes E & Nicholson JK (2006) Integrated metabonomic analysis of bromobenzene-induced hepatotoxicity: novel induction of 5-oxo-prolinosis J Proteome Res 5, 1448–1459 52 Waters NJ, Holmes E, Williams A, Waterfield CJ, Farrant RD & Nicholson JK (2001) NMR and pattern recognition studies on the time-related metabolic... the link between genotypes and phenotypes Plant Mol Biol 48, 155–171 Morris M & Watkins SM (2005) Focused metabolomic profiling in the drug development process: advances from lipid profiling Curr Op Chem Biol 9, 407–412 Wilson ID, Nicholson JK, Castro-Perez J, Granger JH, Johnson KA, Smith BW & Plumb RS (2005) High resolution ‘ultra performance’ liquid chromatography coupled to oa-TOF mass spectrometry... 1 H NMR and MAS NMR spectroscopy Chem Res Toxicol 15, 1351–1359 Kaderbhai NN, Broadhurst DI, Ellis DI & Kell DB (2003) Functional genomics via metabolic footprinting: monitoring metabolite secretion by Escherichia coli tryptophan metabolism mutants using FT-IR and direct injection electrospray mass spectrometry Comp Funct Genom 4, 376–391 Gamache PH, Meyer DF, Granger MC & Acworth IN (2004) Metabolomic... human brain tumor specimens Cancer Res 58, 1825–1832 Keun HC, Beckonert O, Griffin JL, Richter C, Moskau D, Lindon JC & Nicholson JK (2002) Cryogenic probe 13 C NMR spectroscopy of urine for metabonomic studies Anal Chem 74, 4588–4593 Price KE, Vandaveer SS, Lunte CE & Larive CK (2005) Tissue targeted metabonomics: metabolic profiling by microdialysis sampling and microcoil NMR J Pharm Biomed Anal 38, 904–909... A, Barton RH, Trygg J, Hudson J, Blancher C, Gauguier D, Lindon JC, Holmes E & Nicholson JK (2005) Statistical total correlation spectroscopy: An exploratory approach for latent biomarker identification from metabolic 1H NMR data sets Anal Chem 77, 1282–1289 Crockford DJ, Holmes E, Lindon JC, Plumb RS, Zirah S, Bruce SJ, Rainville P, Stumpf CL & Nicholson JK (2006) Statistical HeterospectroscopY (SHY), . under way to investigate the reporting needs and standardization of reporting arrangements for metabonomics studies. The Standard Metabolic Reporting Structures. For example, in the scores plot, the points corresponding to urines from hydrazine-treated animals appear in the lower left quad- rant and in the corresponding