Anal. Chem. 2009, 81, 4734–4741
Pattern Recognition Analysis for the Prediction of Adverse Effects by Nonsteroidal Anti-Inflammatory Drugs Using 1H NMR-Based Metabolomics in Rats So Young Um,†,‡ Myeon Woo Chung,† Kyu-Bong Kim,† Seon Hwa Kim,† Ji Seon Oh,† Hye Young Oh,† Hwa Jeong Lee,*,‡ and Ki Hwan Choi*,† Pharmacology Department, National Institute of Toxicological Research, Korea Food and Drug Administration, 194 Tongil-ro, Eunpyung-Ku, Seoul, Korea, and Division of Life and Pharmaceutical Science, Ewha Womans University, 11-1 Daehyun-dong, Seodaemun-Ku, Seoul, South Korea Nonsteroidal anti-inflammatory drugs (NSAIDs) are commonly used to treat rheumatoid arthritis, osteoarthritis, acute pain, and fever. However, NSAIDs have side effects that include gastric erosions, ulceration, bleeding, and perforation, etc. Selective cyclooxygenase (COX)-2 inhibitors have been developed to avoid the adverse drug reaction of traditional NSAIDs. The COX-2 inhibitors have a different mechanism of action from nonselective COX inhibitors. In this study, pattern recognition analysis of the 1H nuclear magnetic resonance (NMR) spectra of urine was performed to develop surrogate biomarkers related to the gastrointestinal (GI) damage induced by NSAIDs in rats. Urine was collected for 5 h after administering the following NSAIDs at high doses: celecoxib (133 mg kg-1, po), a COX-2-selective inhibitor; and indomethacin (25 mg kg-1, po) or ibuprofen (800 mg kg-1, po), nonselective COX inhibitors. The urine was analyzed using 600 M 1H NMR for spectral binning and targeted profiling. The level of gastric damage in each animal was also determined. Indomethacin and ibuprofen caused severe gastric damage, but no lesions were observed in the celecoxibtreated rats. The 1H NMR urine spectra were divided into spectral bins (0.04 ppm) for global profiling, and 36 endogenous metabolites were assigned for targeted profiling. Multivariate data analyses were carried out to recognize the spectral pattern of endogenous metabolites related to NSAIDs using partial least-squares discrimination analysis (PLS-DA). There were different clusterings of 1H NMR spectra according to the gastric damage scores in global profiling. In targeted profiling, a few endogenous metabolites of allantoine, taurine, and dimethylamine were selected as putative biomarkers for the gastric damage induced by NSAIDs. The results of global and targeted profilings suggest that the gastric damage induced by NSAIDs can be screened in the preclinical stage of drug development using a current metabolomics study. In addition, the putative * To whom correspondence should be addressed. Phone: 82-2-380-1773 (K.H.C.); 82-2-3277-3409 (H.J.L.). Fax: 82-2-389-5225 (K.H.C.); 82-2-3277-2851 (H.J.L.). E-mail:
[email protected] (K.H.C.);
[email protected] (H.J.L.). † Korea Food and Drug Administration. ‡ Ewha Womans University.
4734
Analytical Chemistry, Vol. 81, No. 12, June 15, 2009
biomarkers might also be useful for predicting the risk of adverse effects caused by NSAIDs. Nonsteroidal anti-inflammatory drugs (NSAIDs) are commonly prescribed as analgesic, antipyretic, and anti-inflammatory drugs for the treatment of rheumatoid arthritis, osteoarthritis, acute pain, and fever. More than 30 million people are currently using NSAIDs in the United States, and over 6 billion dollars worth of NSAIDs are sold worldwide.1,2 However, these drugs are associated with significant gastrointestinal (GI) adverse reactions. Indeed, these drugs induce GI damage to animals and human. The depletion of endogenous prostaglandins (PGs) plays a major role in NSAIDinduced gastric mucosal lesions.3-5 Conventional NSAIDs inhibit both principal mammalian cyclooxygenase (COX) isozymes, COX-1 and COX-2, which reduces reduce the signs and symptoms of inflammation. GI lesions induced by NSAIDs reflect mainly the suppression of constitutive formation of “housekeeping” PGs through COX-1 inhibition, whereas limiting PG production from COX-2 is largely sufficient for pain relief. COX-2-selective inhibitors (coxibs), such as rofecoxib, celecoxib, and valdecoxib, are analgesic and anti-inflammatory drugs with good, albeit not absolute, GI tolerance.6,7 COX-2-selective NSAIDs were developed to provide better anti-inflammatory and analgesic efficacy than those of traditional NSAIDs but with less GI injury and antiplatelet activity. Endogenous metabolites are essential components of cellular regulatory and metabolic processes that respond to environmental, pathogenic, and toxicological insults.8-10 Rather than examining just one or two specific metabolites, metabolomics is a multitargeted analysis of low molecular weight, endogenous (1) (2) (3) (4) (5) (6)
(7) (8) (9) (10)
Gabriel, S. E.; Fehring, R. A. J. Clin. Epidemiol. 1992, 45, 1041–1044. Paulus, H. E. Arthritis Rheum. 1985, 28, 1168–1169. Vane, J. R. Nature (London), New Biol. 1971, 231, 232–235. Whittle, B. J. Gastroenterology 1981, 80, 94–98. Kato, S.; Takeuchi, K. Jpn. J. Pharmacol. 2002, 89, 1–6. Dhawan, V.; Schwalb, D. J.; Shumway, M. J.; Warren, M. C.; Wexler, R. S.; Zemtseva, I. S.; Zifcak, B. M.; Janero, D. R. Free Radical Biol. Med. 2005, 39, 1191–1207. Flower, R. J. Nat. Rev. Drug Discovery 2003, 2, 179–191. Goodacre, R.; Vaidyanathan, S.; Dunn, W. B.; Harrigan, G. G.; Kell, D. B. Trends Biotechnol. 2004, 22, 245–252. Nicholson, J. K.; Connelly, J.; Lindon, J. C.; Holmes, E. Nat. Rev. Drug Discovery 2002, 1, 153–161. Nicholson, J. K.; Holmes, E.; Wilson, I. D. Nat. Rev. Microbiol. 2005, 3, 431–438. 10.1021/ac9000282 CCC: $40.75 2009 American Chemical Society Published on Web 05/22/2009
metabolites.11 Indomethacin, a methylated indole derivative of the arylalkanoic acid class of NSAIDs, is a nonselective inhibitor of both COX-1 and COX-2, which participates in PGs synthesis from arachidonic acid. Ibuprofen, a member of the propionic acid group, is also a nonselective COX inhibitor. Both indomethacin and ibuprofen inhibit COX-1 and COX-2 and suppress the production of PGs in the stomach and intestines, which maintain the mucous lining of the GI tract. Therefore, these nonselective COX inhibitors can cause peptic ulcers like other nonselective COX inhibitors. Celecoxib is a member of the first generation of COX-2-selective inhibitors.12-14 COX-2 is inducible and expressed mainly in association with inflammation.15,16 COX-2-specific inhibitors have anti-inflammatory activity without the GI side effects associated with traditional NSAIDs.17,18 Celecoxib is a specific COX-2 inhibitor that does not inhibit COX-1 at the clinical dose used to treat osteoarthritis and rheumatoid arthritis.19 Metabolomics is based on highly sensitive analytical methods. Useful diagnostic information can be obtained by quantifying multiple metabolites with small molecules and comparing them to normal samples. Currently, there are two techniques commonly used to analyze metabolomes (endogenous metabolites): mass spectrometry (MS) coupled with gas chromatography (GC/MS) or liquid chromatography (LC/MS) and nuclear magnetic resonance (NMR) spectroscopy. NMR spectroscopy can be applied to biological samples, such as urine and blood, with minimal preparation or purification of metabolites and is useful for measuring concentrations with good reproducibility and nondiscriminatory detection. This can assist in discovering novel biomarkers of a drug response. NMR-based metabolomics combines proton NMR and statistical data analysis, mainly multivariate data analysis (MVDA), to provide a detailed investigation of the changes in the metabolic profile of a biofluid.20,21 (11) Bijlsma, S.; Bobeldijk, I.; Verheij, E. R.; Ramaker, R.; Kochhar, S.; Macdonald, I. A.; van Ommen, B.; Smilde, A. K. Anal. Chem. 2006, 78, 567–574. (12) Penning, T. D.; Talley, J. J.; Bertenshaw, S. R.; Carter, J. S.; Collins, P. W.; Docter, S.; Graneto, M. J.; Lee, L. F.; Malecha, J. W.; Miyashiro, J. M.; Rogers, R. S.; Rogier, D. J.; Yu, S. S.; Anderson, G. D.; Burton, E. G.; Cogburn, J. N.; Gregory, S. A.; Koboldt, C. M.; Perkins, W. E.; Seibert, K.; Veenhuizen, A. W.; Zhang, Y. Y.; Isakson, P. C. J. Med. Chem. 1997, 40, 1347–1365. (13) Ehrich, E. W.; Dallob, A.; De Lepeleire, I.; Van Hecken, A.; Riendeau, D.; Yuan, W.; Porras, A.; Wittreich, J.; Seibold, J. R.; De Schepper, P.; Mehlisch, D. R.; Gertz, B. J. Clin. Pharmacol. Ther. 1999, 65, 336–347. (14) Bennett, A.; Villa, G. Expert Opin. Pharmacother. 2000, 1, 277–286. (15) Masferrer, J. L.; Zweifel, B. S.; Manning, P. T.; Hauser, S. D.; Leahy, K. M.; Smith, W. G.; Isakson, P. C.; Seibert, K. Proc. Natl. Acad. Sci. U.S.A. 1994, 91, 3228–3232. (16) Raz, A.; Wyche, A.; Siegel, N.; Needleman, P. J. Biol. Chem. 1988, 263, 3022–3028. (17) Simon, L. S.; Weaver, A. L.; Graham, D. Y.; Kivitz, A. J.; Lipsky, P. E.; Hubbard, R. C.; Isakson, P. C.; Verburg, K. M.; Yu, S. S.; Zhao, W. W.; Geis, G. S. JAMA, J. Am. Med. Assoc. 1999, 282, 1921–1928. (18) Silverstein, F. E.; Faich, G.; Goldstein, J. L.; Simon, L. S.; Pincus, T.; Whelton, A.; Makuch, R.; Eisen, G.; Agrawal, N. M.; Stenson, W. F.; Burr, A. M.; Zhao, W. W.; Kent, J. D.; Lefkowith, J. B.; Verburg, K. M.; Geis, G. S. JAMA, J. Am. Med. Assoc. 2000, 284, 1247–1255. (19) Paulson, S. K.; Vaughn, M. B.; Jessen, S. M.; Lawal, Y.; Gresk, C. J.; Yan, B.; Maziasz, T. J.; Cook, C. S.; Karim, A. J. Pharmacol. Exp. Ther. 2001, 297, 638–645. (20) Connor, S. C.; Gray, R. A.; Hodson, M. P.; Clayton, N. M.; Haselden, J. N.; Chessell, I. P.; Bountra, B. Metabolomics 2007, 3, 29–39. (21) Slupsky, C. M.; Rankin, K. N.; Wagner, J.; Fu, H.; Chang, D.; Weljie, A. M.; Saude, E. J.; Lix, B.; Adamko, D. J.; Shah, S.; Greiner, R.; Sykes, B. D.; Marrie, T. J. Anal. Chem. 2007, 79, 6995–7004.
Metabolomics or metabolic profiling involves determining the changes in the concentration of small endogenous metabolites in biological samples caused by physiological stimuli, genetic modification, or environmental alterations.22,23 NMR-based metabolic profiling allows a rapid, simultaneous examination of complex mixtures of biomolecules present in biological samples, such as urine, and generally requires only a limited knowledge of the sample composition prior to analysis. The aim of this study was to develop surrogate biomarkers that can be used to screen the GI damage induced by NSAIDs using metabolomics in rats. In the search of a good biomarkers, 1 H NMR was used to examine the pattern recognition of endogenous metabolites by indomethacin and ibuprofen, which are nonselective COX inhibitors known to induce GI damage at high doses, and celecoxib, a COX-2-selective inhibitor that was developed as a new gastric-sparing anti-inflammatory drug. If there are good and validated biomarkers for metabolomics, biofluids, such as urine and blood, may be sufficient for analysis without the requirement for complicated animal experiments to measure the drug response or anatomical process because metabolomics is a noninvasive method that is available. This study has several advantages. The first is an analysis of the endogenous metabolites in urine as a possible substitute for endoscopy or anatomical experiments for the screening of gastric damage. The second is urinary analysis using 1H NMR to quantify 36 metabolites with reproducibility simultaneously. The profiling of endogenous metabolites reflects the environmental conditions of the rat, which is closest phenotype of an animal physical status of NSAIDs-induced GI damage. MATERIALS AND METHODS Study Design. Male Sprague-Dawley (SD) rats (body weight: 250-300 g) were kept in an accredited Korea Food and Drug Administration (KFDA, Seoul, Korea) animal facility in accordance with the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) International Animal Care Policies (Accredited Unit, KFDA: Unit No. 000996). All animals were given access to standard irradiated chow diet (Purina Mills Inc. Seoul, Korea) and water ad libitum. Upon commencement of the study, the rats were housed in a specified pathogen-free state at 23 ± 1 °C, a relative humidity of 50% ± 10%, and a 12 h light/12 h dark cycle. Each animal was transferred to a metabolic cage (designed specifically for the separate collection of urine and feces) on the day of urine collection. The rats were fasted overnight before the dosing day. Celecoxib (133 mg kg-1), indomethacin (25 mg kg-1), ibuprofen (800 mg kg-1), or vehicle (0.9% sodium chloride added with Tween-80) was administered orally once to the rats. In general, indomethacin at 25 mg kg-1 is used as positive control for a GI damage study, which is 20 times higher the daily dose commonly used. Therefore, the doses of ibuprofen and celecoxib used in this study were also 20 times higher than the commonly used daily dose. In the beginning, each group contained 10 rats each, and the vehicle group contained 16 rats. However, several rats were excluded for insufficient urine volume. In the end, the group of vehicle, (22) Dieterle, F.; Schlotterbeck, G.; Binder, M.; Ross, A.; Suter, L.; Senn, H. Chem. Res. Toxicol. 2007, 20, 1291–1299. (23) Kim, K. B.; Chung, M. W.; Um, S. Y.; Oh, J. S.; Kim, S. H.; Na, M. A.; Oh, H. Y.; Cho, W. S.; Choi, K. H. Metabolomics 2008, 4, 377–392.
Analytical Chemistry, Vol. 81, No. 12, June 15, 2009
4735
indomethcin, ibuprofen, and celecoxib were 16, 9, 8, and 10, respectively. The experimental protocol was approved by the Institutional Animal Care and Use Committees (IACUC) of KFDA. Sample Collection. After drug administration, urine samples were collected from the metabolic cages over a 5 h period. After the animals were euthanized with CO2, the stomach was removed and scored for hemorrhage damage. Scoring was performed by considering the size of all the lesions, as well as the grade of the depth, and the values were added to give an overall gastric damage score for each rat.24 All processes relating to the animals were followed as outlined in “A good practice guide to the administration of substances and removal of blood including routes and volumes”.25 The urine sample aliquots collected were stored at -74 °C until analysis. Sample Preparation for 1H NMR Analysis. The urine samples were processed according to the 1H NMR protocol after being thawed at room temperature and centrifuged at 10 000g for 10 min at 4 °C. Sixty microliters of a standard solution (5 mM 2,2-dimethyl-2-silapentane-5-sulfonate sodium salt (DSS), Sigma-Aldrich, St. Louis, MO) and 100 mM imidazole (SigmaAldrich) in D2O (purity 99.9%, Cambridge Isotope Lab. Inc., Andover, MA) and 30 µL of sodium azide solution (0.42% NaN3 (Sigma-Aldrich) in H2O) were added to the 510 µL aliquot of urine in a 1.5 mL Eppendorf tube. After vortexing, the pH was set to pH 6.8 ± 0.02. A 500 µL aliquot of the mixture was dispensed into a 5 mm glass NMR tube (Wilmad-Labglass, Co., Buena, NJ). 1 H NMR Analysis. All 1H NMR spectra were obtained on a Varian Unity Inova 600 MHz NMR spectrometer equipped with a cold probe and a 768 AS autosampler. The Noesypresat 1D 1H NMR spectra were collected at 25 °C using a presaturation pulse sequence and a spectral width of 9615.4 Hz. The time-domain data points, acquisition time, relaxation delay time, mixing time, pulse, and the number of transients in urine analysis were 76 924, 4 s, 2 s, 0.4 s, 90°, and 64, respectively. The following were used for water suppression: 1.5 s relaxation delay time, 6 power saturation, and changing the saturation frequency. The free induction decay (FID) was apodized with an exponential window function corresponding to a line broadening of 0.5 Hz and Fourier-transformed. The data was analyzed using the established method of spectral binning26 and the new technique of targeted profiling.27,28 Spectral Binning for Global Profiling. Spectral binning was performed using the Chenomx NMR Suite Professional software version 4.6 (Chenomx Inc., Edmonton, Canada). Each NMR spectrum was reduced to 0.04 ppm wide segments between 0.00 and 10.0 ppm. The spectrum regions of water (δ 4.72-5.20), urea (δ 5.56-6.00), and imidazole (δ 7.32-7.40/8.32-8.44 ppm) were removed from the analysis for all groups in order to prevent variation in each sample. The resonance from the NMR internal (24) Kim, K. B.; Chang, M. S.; Chung, Y. K.; Sohn, S. K.; Kim, S. G.; Choi, W. S. J. Pharm. Pharmacol. 1998, 50, 521–529. (25) Diehl, K. H.; Hull, R.; Morton, D.; Pfister, R.; Rabemampianina, Y.; Smith, D.; Vidal, J. M.; van de Vorstenbosch, C. J. Appl. Toxicol. 2001, 21, 15–23. (26) Holmes, E.; Antti, H. Analyst 2002, 127, 1549–1557. (27) Chang, D.; Weljie, A.; Newton, J. Pac. Symp. Biocomput. 2007, 12, 115– 126. (28) Weljie, A. M.; Newton, J.; Mercier, P.; Carlson, E.; Slupsky, C. M. Anal. Chem. 2006, 78, 4430–4442.
4736
Analytical Chemistry, Vol. 81, No. 12, June 15, 2009
standard and reference (DSS) was also excluded from the data set prior to analysis. Each NMR variable was normalized to the total area in order to allow a spectrum-to-spectrum comparison. The intensity sum was calculated after excluding of water, urea, imidazole, and DSS from the data set. In addition, the outstanding peaks by the NSAIDs, such as metabolite and drug itself, were excluded in order to remove chemical effects such as carboxypropyl phenylpropionic acid, which is the major metabolite of ibuprofen administered orally in urine. NMR Quantification for Targeted Profiling. The methyl singlet of DSS in a standard solution was used as an internal standard for the chemical shifts (set to 0 ppm) as well as for quantification. Thirty-six clearly identifiable metabolites were chosen for quantification using the Chenomx NMR Suite Professional software version 4.6 (database available at pH 7.0 for 600 MHz, Chenomx Inc.) by matching to the Chenomx Library. Chenomx Profiler, part of Chenomx NMR Suite, allowed us to rapidly identify and quantify metabolites of interest using Chenomx compound libraries representing over 270 metabolites. The reference libraries stored in the software were fitted to the urine spectra to quantify the selected metabolites. The internal DSS signal was as the reference concentration (500 µM). Typically, the concentrations of the urine metabolites are reported as ratios with the metabolite, creatinine (δ 3.03/4.04 ppm). Statistical Analysis of NMR Data. Both univariate and multivariate statistical analyses were applied to the vertically scaled data using Excel (Microsoft Corp. Seattle, WA) and SIMCA-P version 11.0.0.0, (Umetrics, AB, Umeå, Sweden), respectively. Histographic depictions were created using GraphPad Prism version 5.01 (GraphPad Software, San Diego, CA) on the raw data with the group producing gastric damage compared with the other group not causing it. Initially, principal components analysis (PCA) was applied to the clustering of all identified outliers. Partial leastsquares discrimination analysis (PLS-DA) was then advanced for global and targeted profiling from the converted spectral data. All statistical analyses were carried out using binned bucket data and the identified metabolites for spectral binning and targeted profiling. The univariate statistics were calculated using either a Student’s t test or multiple comparison, such as ANOVA (analysis of variance) followed by a Tukey’s test with each metabolite concentration normalized to the by creatinine concentration, assuming a constant rate of creatinine excretion in each urine sample. A Kruskal-Wallis test (nonparametric test) was used for comparison of gastric damage scoring. Partial Least-Squares Discriminant Analysis. The concentrations of endogenous metabolites as well as the spectral binning data were imported into SIMCA-P for global and targeted profiling. Supervised PLS-DA analysis employs independent (concentration of metabolites or the values of 0.04 ppm segments) and dependent variables (drug response groups) for the class comparisons using multivariate statistical methods and partial least-squares modeling with the latent variables to allow the simultaneous analysis of all variables. PLS-DA functions well with a large number of predictors and multicollinearity.29-32 In addition, a quantitative estimation (29) Eisen, M. B.; Spellman, P. T.; Brown, P. O.; Botstein, D. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 14863–14868. (30) Lee, Y.; Lee, C. K. Bioinformatics 2003, 19, 1132–1139. (31) Liu, Y.; Ringner, M. BMC Bioinf. 2004, 5, 70. (32) Nguyen, D. V.; Rocke, D. M. Bioinformatics 2002, 18, 1625–1632.
of the discriminatory power of each descriptor was evaluated using the VIP (variable importance for the projection) parameters provided by PLS-DA. The VIP values represent an appropriate quantitative statistical parameter ranking the descriptors according to their ability to discriminate different classes. In this study, a VIP > 1.0 was set as the threshold for a meaningful contribution (arbitrary) and compared with the results of univariate analysis to identify the significantly influential metabolites. RESULTS AND DISCUSSION 1 H NMR spectroscopy is the most powerful tool for profiling of biofluids with the following inherent advantages over other analytical techniques: (1) minimal sample preparation, (2) nondestructive, (3) nonequilibrium perturbing, (4) small (down to 1 µL) sample volumes, and (5) produces quantitative and qualitative information from the same data set. The use of 1H NMR spectroscopy to follow various biochemical responses does not require the preselection of metabolites and allows subsequent multicomponent analysis, without the bias imposed by the experimenters’ expectations. Therefore, the NMR spectra of biosamples provide important information on endogenous biochemical processes in both healthy and disease states. In general, researchers have used NMR spectroscopy and MS-based global and targeted methodologies to provide comprehensive coverage of the metabolome.33 The urine profiles change more dramatically upon a biochemical disturbance than serum, which is controlled homeostatically by renal filtration, reabsorption, and secretion.20 In the current study, urine samples were collected under identical conditions. All samples were quantified using 1H NMR, and MVDA was carried out for pattern recognition. Although each group started with 10 or 16 rats, several urine samples were omitted due to the insufficient volume because urine was only collected over a 5 h period in order to determine the correlation between the endogenous metabolites in urine and GI damage in an identical rat. An analysis of the PLS-DA scores and VIP plots of all spectra imported into SIMCA-P were performed to evaluate pattern recognition. Metabolomics using 1H NMR was carried out with global and targeted profiling to generate biomarkers for separating the groups showing NSAIDs-induced gastric damage from nondamaged groups. There were several differences in the spectra of the urine collected from the NSAIDstreated rats compared with the control group. The partial 600 MHz spectra showed a variable region corresponding to the urine collected from the rats 5 h after dosing with the NSAIDs or vehicle (Figure 1). Gastric Damage Scoring. Indomethacin and ibuprofen induce gastric damage in animals.24,34,35 In contrast, celecoxib (200 mg kg-1 orally or subcutaneously) dose not cause any GI toxicity.36 In the present study, after urine collection, the damage to the stomach removed from the rat was evaluated after preservation (33) Daykin, A. C.; Wu ¨ lfert, F. Front. Drug Des. Discovery 2006, 2, 151–173. (34) Bonabello, A.; Galmozzi, M. R.; Canaparo, R.; Isaia, G. C.; Serpe, L.; Muntoni, E.; Zara, G. P. Anesth. Analg. 2003, 97, 402–8. (35) Zamora, Z.; Gonzalez, R.; Guanche, D.; Merino, N.; Menendez, S.; Hernandez, F.; Alonso, Y.; Schulz, S. Inflammation Res. 2008, 57, 39–43. (36) Laudanno, O. M.; Cesolari, J. A.; Esnarriaga, J.; Rista, L.; Piombo, G.; Maglione, C.; Aramberry, L.; Sambrano, J.; Godoy, A.; Rocaspana, A. Dig. Dis. Sci. 2001, 46, 779–784.
in a 10% neutral buffered formalin solution for 30 min in a double-blind manner. Celecoxib, a COX-2 inhibitor, did not produce any gastric lesions in the rats even though its dose was 20 times higher (133 mg kg-1) than the daily maximum dose (6.7 mg kg-1) (Figure S-1 in the Supporting Information), which is consistent with the previous report.36 In contrast, indomethacin (25 mg kg-1) as a nonselective COX inhibitor caused gastric erosions in rats, with a significantly higher greater mean gastric damage score than the other groups. The rats treated with ibuprofen (800 mg kg-1) showed GI damages, even though the scores were lower than those of the indomethacin group (Figure 2). Statistical analysis was performed using a Kruskal-Wallis test, which does not assume a Gaussian distribution, and a Dunn’s post test (compare all pairs of groups) with the 95% confidence interval. Supporting Information Figure S-1 shows a stomach damaged by NSAIDs along with GI scoring. Spectral Binning for Global Profiling. Metabolomics researchers typically preprocess the 1H NMR spectra before carrying out statistical analyses because they contain a large volume of data. One popular preprocessing method involves the division of a spectrum into bins of equal sizes. In this equidistant binning technique, the intensities of the peaks in each bin were integrated, and this value was assigned to the bin.37 Traditional “binning” With “bucketing” uses equidistant bin sizes throughout the spectrum. Different bin sizes may be used for urine (0.04) and plasma/serum (0.01). More intelligent binning may be employed, in which a number of different schemes may be used.38,39 In this study, all samples were calculated for each spectral bin for global profiling before quantifying all the metabolites by 1H NMR spectra for targeted profiling. After binning, the spectrum dimension was reduced to 214 with the mid ppm value of each bin being considered as the new predictor, which is a significant decrease from the original peak data. The data matrix used for multivariate analysis contained 43 rows and 214 columns. Figure 3 shows a good gathering, where the rats were classified into gastric-damaged groups or non-gastricdamaged groups by NSAIDs. Although there was little difference depending on the type of NSAIDs used, dispersed spots could be divided into two clusters, as shown in Figure 3. When Figure 3 was compared with Figure 4, a scores plot of PLS-DA from the spectral binning showed similar patterns to the targeted profiling. Interestingly, the clustering from celecoxib was close to that of the control in global profiling, which is obviously different from the other groups such as indomethacin and ibuprofen which are known to induce GI damage. Global metabolic profiling has long been used to determine biomarkers to help assess the pathophysiological health status of patients.40,41 Metabolomics is referred to collectively as global metabolic profiling or, simply, metabolic profiling. Global meta(37) Schnackenberg, L. K.; Beger, R. D. Drug Discovery Today: Technol. 2007, 4, 13–16. (38) Davis, R. A.; Charlton, A. J.; Godward, J.; Jones, S. A.; Harrison, M.; Wilson, J. C. Chemom. Intell. Lab. Syst. 2007, 85, 144–154. (39) De Meyer, T.; Sinnaeve, D.; Van Gasse, B.; Tsiporkova, E.; Rietzschel, E. R.; De Buyzere, M. L.; Gillebert, T. C.; Bekaert, S.; Martins, J. C.; Van Criekinge, W. Anal. Chem. 2008, 80, 3783–3790. (40) Fiehn, O. Comp. Funct. Genomics 2001, 2, 155–168. (41) German, J. B.; Bauman, D. E.; Burrin, D. G.; Failla, M. L.; Freake, H. C.; King, J. C.; Klein, S.; Milner, J. A.; Pelto, G. H.; Rasmussen, K. M.; Zeisel, S. H. J. Nutr. 2004, 134, 2729–2732.
Analytical Chemistry, Vol. 81, No. 12, June 15, 2009
4737
Figure 1. The aliphatic region of the 600 MHz 1H NMR urine spectra from a rat dosed with vehicle (a), indomethacin (IM) (25 mg kg-1) (b), ibuprofen (IBU) (800 mg kg-1) (c), or celecoxib (CELE) (133 mg kg-1) (d), respectively.
bolic profiling will also help the medical community understand the complex interactions between metabolites, genes, and proteins on the overall health status.37 Targeted Profiling. A comparison with the vehicle revealed several signals to be unique for indomethacin; niacinamide, hippurate, fumarate, trigonelline, taurine, trimethylamine, 2-oxo4738
Analytical Chemistry, Vol. 81, No. 12, June 15, 2009
glutarate, citrate, succinate, and acetate. These compounds include the surrogate biomarkers for separating the gastric-damaged group from the non-gastric-damaged group by NSAIDs. Table 1 shows a list of the metabolites targeted by the Chenomx NMR Suite in rat urine. With Chenomx library, endogenous metabolites were identified from the NMR spectra, and then, each metabolite
Figure 2. Gastric damage score induced by the oral administration of indomethacin (25 mg kg-1, IM), ibuprofen (800 mg kg-1, IBU), or celecoxib (133 mg kg-1, CELE) in rat. Gastric damage score: vehicle and celecoxib were 0, indomethacin and ibuprofen were 7.6 ( 1.6 and 2.6 ( 1.2, respectively. * P < 0.05, tested by Kruskal-Wallis. Each column represents the mean ( SEM.
Figure 4. Targeted profiling of the gastric-damaged groups (damaged) and nondamaged groups (nondamaged) after the NSAIDs treatment including the vehicle-treated groups using PLS-DA in urine samples. The PLS-DA models show the metabolic distinction between the stomachs damaged by ibuprofen (circle) or indomethacin (triangle) and not damaged by celecoxib (diamond) or vehicle (star). The 3D scores plot shows the separation of clustering between the gastricdamaged and nondamaged groups (R 2X ) 0.891, R 2Y ) 0.909, Q 2 ) 0.678).
Figure 3. Separation of the gastric-damaged groups (damaged) and nondamaged groups (nondamaged) by NSAIDs including vehicle (veh)-treated groups using PLS-DA in global profiling of urine samples. The PLS-DA models illustrate the metabolic distinction between the stomachs damaged by ibuprofen (circle) or indomethacin (triangle) and not damaged by celecoxib (diamond) or vehicle (star). Gastric-damaged groups treated with the nonselective COX inhibitors show the visual separation of clustering from the nondamaged groups treated with celecoxib, as a COX-2-selective inhibitor, and vehicle in the t1 vs t2 vs t3 3D scores plot (R 2X ) 0.632, R 2Y ) 0.898, Q 2 ) 0.836).
was quantified. The several metabolites related with clustering were selected after profiling with SIMCA-P. Histopathology revealed 36 small molecules as identifiable endogenous metabolites from adverse drug reactions in the stomach, which are expected to be related to the drug-induced changes in the urine. Pattern Recognition Analysis. All the samples were quantified with 1H NMR spectra for targeted profiling, after spectral binning for global profiling. According to the class of NSAIDs, the endogenous metabolites were clustered in PLS-DA (Figures 3 and 4). The results of Figures 2-4 show that the nonselective COX inhibitors among the NSAIDs induced stomach damage and their endogenous metabolites were different from those of
celecoxib, as a COX-2-selective inhibitor. After endogenous metabolites identified with Chenomx were analyzed with score plot and loading plot, it could be known whether GI-damaged group are clustering and separated from the non-gastric-damaged group, or not. In PLS-DA, the endogenous metabolites of rats with GI damage were clearly classified from the rats without GI damage. VIP revealed several endogenous metabolites that would influence the categorization by GI-damaged and nondamaged in PLS-DA (Figure S-2, parts a and b, in the Supporting Information). PLS-DA is a statistical method that allows a graphical overview of large multivariate data sets and provides good discrimination between the two groups. This pictorial representation of the full data in one plot simplifies the detection of sample grouping because each sample is represented by one point on the plot. The samples that appear close together on the graph are said to cluster and are similar in biochemical composition.20 Good separation in both global profiling and targeted profiling was observed in the gastricdamaged groups and non-gastric-damaged groups after treatment with indomethacin, ibuprofen, celecoxib, or vehicle (Figures 3 and 4). Consequently, the pattern of endogenous metabolites depends on the level of gastric damage induced by NSAIDs. This suggests that several endogenous metabolites can be used to predict the COX-2 selectivity of NSAIDs and NSAIDs-related GI damage. VIP analysis showed the order of metabolites according to their critical influence on clustering (Figure S-2a in the Supporting Information). The key metabolites, which were essential for distinguishing between the gastric-damaged group and nondamaged group, were selected from the results of the important metabolites (VIP > 1.00) and univariate statistical analysis (p-value 1.0) (Figure 5). All the metabolites, such as allantoin, citrate, 2-oxoglutarate, taurine, acetate, hippurate, and dimethylamine were decreased or increased significantly by NSAIDs compared with the vehicle. After identification and quantification by NMR, concentrations of endogenous metabolites 4740
Analytical Chemistry, Vol. 81, No. 12, June 15, 2009
Figure 5. Concentrations of the endogenous metabolites (normalized against the creatinine concentration) in the gastric-damaged groups by NSAIDs and vehicle-treated group. An ANOVA-Tukey’s multiple comparison test was performed to assess the statistical significance. The error bars are expressed as the SEM * p < 0.05 vs veh, # p < 0.05 vs indo.
Figure 6. Concentrations of endogenous metabolites (normalized against the creatinine concentration) for the celecoxib group and vehicle-treated group. An unpaired t test was performed to assess the statistical significance between each control group and celecoxib treatment group. The error bars are expressed as the SEM * p < 0.05 vs veh.
were normalized with creatinine, and then, we calculated the average in each group. Fluctuation of all of compounds detected using NMR and significant difference were evaluated compared with the vehicle group. However, some of them, citrate, 2-oxoglutarate, acetate, and hippurate, were decreased by celecoxib, occurred without GI damage, and were also used as the negative control for GI damage (Figure 6). Only allantoin, taurine, and dimethylamine were not changed significantly by celecoxib compared with vehicle, as opposed to indomethacin and ibuprofen. Some urine metabolites frequently change in response to the administration of toxicants, regardless of the nature of the toxicant, its mechanism of action, or its target.42 A further study will refine the predictive potential of these metabolites for NSAID development. An additional investigation will focus on the metabolites (42) Robertson, D. G. Toxicol. Sci. 2005, 85, 809–822.
that are corrected by a treatment with NSAIDs, but not by administration of the agent itself. Statistical identification of the biological processes associated with NSAID will enhance the robustness of these findings. Nevertheless, metabolic profiling allowed the identification of a drug-induced endogenous metabolite. Moreover, monitoring drug-induced disturbances in the homeostasis of the basal metabolites by metabolomics is more accessible, less expensive, more sensitive, and less biased than the classical methods used to characterize the drug safety and efficacy in drug development. Overall, these putative biomarkers can be used to predict the risk of adverse effects caused by NSAIDs. CONCLUDING REMARKS This study classified the metabolic differences in urine of gastric-damaged NSAIDs groups and nondamaged groups. Several putative biomarkers for predicting NSAIDs-related GI damage were found using 1H NMR quantification and statistical analyses, including increased allantoin, decreased taurine, and increased dimethylamine, compared with the vehicle. Endoscopy is used to examine NSAIDs-related gastric damage, but it is expensive, time-consuming, requires specific expertise, and has limited patient acceptability. A further study should be performed to
confirm and validate this biomarker for a diagnosis of gastric damage, such as gastric ulcer, more conveniently and noninvasively. Moreover, additional research is needed to confirm these biomarkers because the other side effects of NSAIDs can be induced simultaneously at this dose. These biomarkers could be used to screen large populations for GI damage related to NSAID use. The strength of this study is that urine analysis with metabolomics is simple and noninvasive. In addition, metabolomics with a surrogate biomarker can be used as a more rapid and easier preliminary screening tool to determine if new compounds induce GI damage. ACKNOWLEDGMENT This study was supported by a 2007 Grant (07151KFDA631) from the Korea Food and Drug Administration. SUPPORTING INFORMATION AVAILABLE Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review January 6, 2009. Accepted April 30, 2009. AC9000282
Analytical Chemistry, Vol. 81, No. 12, June 15, 2009
4741
