Pooled Sample Strategy in Conjunction with High ... - ACS Publications

Apr 13, 2010 - Biotransformation, Bristol-Myers Squibb Research and Development, Pennington, New Jersey 08534, and Toxicology and Clinical Pathology ...
0 downloads 0 Views 347KB Size
Anal. Chem. 2010, 82, 3834–3839

Pooled Sample Strategy in Conjunction with High-Resolution Liquid Chromatography-Mass Spectrometry-Based Background Subtraction to Identify Toxicological Markers in Dogs Treated with Ibipinabant Haiying Zhang,*,† Laura Patrone,‡ John Kozlosky,‡ Lindsay Tomlinson,§ Greg Cosma,‡ and Joseph Horvath‡ Biotransformation, Bristol-Myers Squibb Research and Development, Pennington, New Jersey 08534, and Toxicology and Clinical Pathology, Bristol-Myers Squibb Research and Development, New Brunswick, New Jersey 08903 Metabolomics with chromatography-mass spectrometry is often challenging and relies on statistical tools to discern changes in a metabolome. A pooled sample strategy was proposed, consisting of (1) identification of potential marker candidates by detecting changes of metabolites in a few pooled samples between treated and control groups and (2) validation of markers of statistically significant changes with a large set of individual samples. This strategy was enabled by applying a thorough background subtraction approach based on high-resolution mass spectrometry. In a proof-of-principle study, plasma samples were generated and pooled in a 6-week investigational study to identify potential toxicological markers for an observed muscle toxicity associated with the treatment of ibipinabant in dogs. With pooled control samples as backgrounds, potential marker candidates were revealed in the background-subtracted profiles of the pooled ibipinabant-treated samples. After further cleaning with the use of mass defect filtering to exclude drug metabolites and the comparison of profiles between pooled treated samples to eliminate inconsistent peaks, the major biomarker candidates in the profiles were identified to be 19 acylcarnitines. A total of 3 of the 19 acylcarnitines were measured on the set of individual samples to allow for statistical analysis. The results confirmed the significance of acylcarnitine elevations in ibipinabant-treated dogs and indicated that the acylcarnitines could be early markers for the dog-specific toxicity. The advantages of the pooled sample strategy and its potential limitations for metabolomics are discussed. Metabolomics is a methodology to identify unique chemical fingerprints or metabolic changes that specific cellular or physi* To whom correspondence should be addressed. Haiying Zhang, Ph.D., Bristol-Myers Squibb Research and Development, 311 Pennington-Rocky Hill Road, Pennington, NJ 08534. E-mail: [email protected]. Phone: (609) 8183537. † Biotransformation. ‡ Toxicology. § Clinical Pathology.

3834

Analytical Chemistry, Vol. 82, No. 9, May 1, 2010

ological processes leave behind.1,2 The fingerprints that metabolomics targets are typically metabolites of small molecular weights and can be of a wide chemical diversity. This is very different from transcriptomics or proteomics which focuses on only a specific type of macromolecules. The metabolites identified in a metabolomic study are potential biomarkers that can be used to understand or track a biological process. For example, metabolite markers identified in urine or plasma samples can be used to monitor physiological changes caused by toxic insult of a chemical. In many cases, such changes can be related to specific syndromes, e.g., a specific lesion in muscle. This is of particular relevance to toxicological assessment of potential drug candidates and can be important for lead compound discovery, clinical trial testing, and/ or postapproval drug monitoring.3,4 The methodology of metabolomics has been extensively reviewed.3-16 Typically, a metabolomic study involves the analysis of a large set of individual samples and relies on statistical techniques to discern significant changes of metabolites from typical biological variations among individuals. Such statistics(1) Daviss, B. Scientist 2005, 19, 25–28. (2) Nicholson, J. K.; Lindon, J. C.; Holmes, E. Xenobiotica 1999, 11, 1181– 1189. (3) Wishart, D. S. Drugs R&D 2008, 9, 307–22. (4) Schnackenberg, L. K.; Beger, R. D. Toxicol. Mech. Methods 2008, 18, 301– 311. (5) Portilla, D; Schnackenberg, L. K.; Beger, R. D. Sem. Nephrol. 2007, 27, 609–620. (6) Werner, E; Heilier, J. F.; Ducruix, C.; Ezan, E.; Junot, C.; Tabet, J. C. J. Chromatogr., B 2008, 871, 143–163. (7) Dunn, W. B. Phys. Biol. 2008, 5, 11001. (8) Gomase, V. S.; Changbhale, S. S.; Patil, S. A.; Kale, K. V. Curr. Drug Metab. 2008, 9, 89–98. (9) Gowda, G. A.; Zhang, S.; Gu, H.; Asiago, V.; Shanaiah, N.; Raftery, D. Expert Rev. Mol. Diagn. 2008, 8, 617–633. (10) Lewis, G. D.; Asnani, A.; Gerszten, R. E. J. Am. Coll. Cardiol. 2008, 52, 117–23. (11) Kim, Y. S.; Maruvada, P.; Milner, J. A. Future Oncol. 2008, 4, 93–102. (12) Clarke, C. J.; Haselden, J. N. Toxicol. Pathol. 2008, 36, 140–147. (13) Feng, X.; Liu, X.; Luo, Q.; Liu, B. F. Mass Spectrom. Rev. 2008, 27, 635– 660. (14) Roberts, L. D.; McCombie, G.; Titman, C. M.; Griffin, J. L. J. Chromatogr., B 2008, 871, 174–181. (15) Novotny, M. V.; Soini, H. A.; Mechref, Y. J. Chromatogr., B 2008, 866, 26–47. (16) Allwood, J. W.; Ellis, D. I.; Goodacre, R. Physiol. Plant. 2008, 132, 117–35. 10.1021/ac100287a  2010 American Chemical Society Published on Web 04/13/2010

based change detection is particularly challenging for data generated from chromatography-mass spectrometry, because the complexity of the technology necessitates that data of all individual samples be appropriately processed before the statistical detection can succeed.17,18 We propose a pooled sample strategy to shift the burden of change detection from the statistical analysis. In this strategy, potential markers of changes are identified from the analysis of a few pooled samples of a treated group and a control group. Statistics on individual samples is used for confirmation purposes instead of for change detection. A somewhat similar strategy has been used in the areas of transcriptomics19,20 and proteomics4,5,21,22 to reduce the time and resource demand of analyzing a large set of individual samples. In the area of metabolomic profiling, the implementation of a pooled sample strategy is difficult because of the vast chemical diversity of the metabolome. Pooled metabolomic samples have been reported for quality control purposes only to overcome the problem of monitoring preparation and analytical performance of a large set of samples.23-26 Nevertheless, we believe that a pooled sample strategy, if feasible, should be of value to a metabolomic study to both reduce resource requirements and to alleviate data processing complexity. The advance of modern high-resolution mass spectrometry brings new opportunities to glean useful information out of samples containing a wide chemical diversity. For example, a mass defect filtering technique has been developed which leverages highresolution liquid chromatography-mass spectrometry (LC-MS) data and predictable mass defects of drug metabolites to remove interference ions in a sample matrix to reveal drug metabolite ions of interest.27,28 Recently, a systematic background subtraction approach was proposed, leveraging high-resolution data sets to discern drug metabolite peaks of interest in complex biological samples.29 The novelty of this background subtraction approach is that ions in a drug-treated sample are thoroughly evaluated, based on mass precision, for their presence in a broader chromatographic time window in the control sample(s). Provided that adequate control samples are obtained, drug metabolites (17) Broeckling, C. D.; Reddy, I. R.; Duran, A. L.; Zhao, X.; Sumner, L. W. Anal. Chem. 2006, 78, 4334–4341. (18) Ghosh, S; Grant, D. F.; Dey, D. K.; Hill, D. W. BMC Bioinf. 2008, 9, 38. (19) Agrawal, D.; Chen, T.; Irby, R.; Quackenbush, J.; Chambers, A. F.; Szabo, M.; Cantor, A.; Coppola, D.; Yeatman, T. J. J. Nat. Cancer Inst. 2002, 94, 513–521. (20) Jolly, R. A.; Goldstein, K. M.; Wei, T.; Gao, H.; Chen, P.; Huang, S.; Colet, J.-M.; Ryan, T. P.; Thomas, C. E.; Estrem, S. T. Physiol. Genomics 2005, 22, 346–355. (21) Zhang, J; Goodlett, D. R.; Peskind, E. R.; Quinn, J. F.; Zhou, Y.; Wang, Q.; Pan, C.; Yi, E.; Eng, J.; Aebersold, R. H.; Montine, T. J. Neurobiol. Aging 2005, 26, 207–227. (22) Zhang, J.; Goodlett, D. R.; Quinn, J. F.; Peskind, E.; Kaye, J. A.; Zhou, Y.; Pan, C.; Yi, E.; Eng, J.; Wang, Q.; Aebersold, R. H.; Montine, T. J. J. Alzheimers Dis. 2005, 7, 125–133. (23) Sangster, T; Major, H; Plumb, R; Wilson, A. J.; Wilson, I. D. Analyst 2006, 131, 1075–1078. (24) Gika, H. G.; Macpherson, E.; Theodoridis, G. A.; Wilson, I. D. J. Chromatogr., B 2008, 871, 299–305. (25) Gika, H. G.; Theodoridis, G. A.; Wingate, J. E.; Wilson, I. D. J. Proteome Res. 2007, 6, 3291–3303. (26) Li, X.; Xu, Z.; Lu, X.; Yang, X.; Yin, P.; Kong, H.; Yu, Y.; Xu, G. Anal. Chim. Acta 2009, 633, 257–262. (27) Zhang, H.; Zhang, D.; Ray, K.; Zhu, M. J. Mass Spectrom. 2009, 44, 999– 1016. (28) Huang, Y. Drug Discovery Dev. 2009, 12, 29–31. (29) Zhang, H.; Yang, Y. J. Mass Spectrom. 2008, 43, 1181–90.

present in the treated sample can be effectively detected as major peaks in a background-subtracted chromatographic profile.29-32 In this article, we explore using this high-resolution mass spectrometry-based background subtraction approach to enable pooled sample strategy in metabolomic profiling to identify potential biomarker candidates. The goal is to identify robust markers that can be of practical value to indicate a cellular or physiological change. The model study used was one to identify potential toxicological markers for an observed muscle toxicity associated with the treatment of ibipinabant in dogs. Ibipinabant (SLV-319, BMS-646256) is a potent and selective cannabinoid receptor 1 antagonist that was being developed for the treatment of obesity.33 MATERIALS AND METHODS Strategy. The pooled sample strategy for a metabolomic study consists of two parts: (1) identification of potential biomarker candidates through the analysis of pooled treated and control samples; and (2) quantitative analysis of the identified marker candidates with sets of individual samples to determine their statistical significance. In the first part, two or more pooled samples are generated for each sample group. The samples are subject to chromatography-high resolution mass spectrometry analysis. The data of each pooled treated sample is backgroundsubtracted with that of the pooled control samples or vise versa. The background-subtracted profiles of the same sample group are compared to eliminate any inconsistent peaks. Additional processing techniques may be used to eliminate components that are not of interest. For example, mass defect filtering may be used to remove drug metabolites in pooled drug-dosed samples. The remaining major peaks are characterized to establish their identity. These are the potential biomarker candidates identified and they are subject to the second part of the process for statistical validation purposes. In the second part, an optimal analytical condition is developed for the analysis of the identified biomarker candidates with the large set of individual samples. The analysis can be a targeted one on the identified biomarkers only or a global one encompassing a wider range of metabolites including the major biomarker candidates of interest. Chemicals. Ibipinabant was supplied by Bristol-Myers Squibb. All other chemicals used were commercially available and were of either guaranteed grade or HPLC grade. Six-Week Oral Investigative Study in Dogs. Purebred female beagle dogs (body weight, 5-7 kg) were purchased from Marshall Farms (Nose Rose, NY). Animals were assigned to the ibipinabant-treatment group and the control group, with six animals in each group. Suspension of ibipinabant (300 mg/kg/ day) in 1% methylcellulose and 0.5% poloxamer 188 in water or 1% methylcellulose and 0.5% poloxamer 188 in water alone (control) were administered orally to the animals once daily for 6 (30) Zhang, H.; Ma, L.; He, K.; Zhu, M. J. Mass Spectrom. 2008, 43, 1191–200. (31) Zhang, H.; Grubb, M.; Wu, W.; Josephs, J.; Humphreys, W. G. Anal. Chem. 2009, 81, 2695–2700. (32) Zhu, P.; Ding, W.; Tong, W.; Ghosal, A.; Alton, K.; Chowdhury, S. Rapid Commun. Mass Spectrom. 2009, 23, 1563–1572. (33) Lange, J. H.; Coolen, H. K.; van Stuivenberg, H. H.; Dijksman, J. A.; Herremans, A. H.; Ronken, E.; Keizer, H. G.; Tipker, K.; McCreary, A. C.; Veerman, W.; Wals, H. C.; Stork, B.; Verveer, P. C.; den Hartog, A. P.; de Jong, N. M.; Adolfs, T. J.; Hoogendoorn, J.; Kruse, C. G. J. Med. Chem. 2004, 47, 627–643.

Analytical Chemistry, Vol. 82, No. 9, May 1, 2010

3835

weeks. Plasma samples (0.5 mL aliquots) were collected into EDTA-tubes weekly for 6 weeks during the dosing period. Plasma samples were prepared by centrifugation. The study was conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals (NIH Publication No. 85-23, revised 1996) and was approved by the Bristol-Myers Squibb Site Institutional Animal Care and Use Committee. Pooling of Dog Plasma Samples for Identification of Potential Biomarkers. Pooled plasma samples were generated for week 2 and week 3 samples for both the ibipinabant-treated group and the control group. This was done by pooling 50 µL from each of the six samples within the same animal group. A volume of 100 µL of pooled plasma samples were mixed with 200 µL of cold methanol. The suspension was vortexed and centrifuged at 2000g for 10 min. Aliquots of the supernatants were subject to LC-MS analyses. The injection volume for the plasma samples was 5 µL. High-Resolution LC-MS Analysis of Pooled Samples. The high-resolution LC-MS data were obtained using a Finnigan LTQ FT mass spectrometer (ThermoFinnigan, San Jose, CA) coupled with an Acquity ultrahigh-performance liquid chromatography (UHPLC) module (Waters, Milford, MA). The UHPLC was performed using an Acquity BEH C18 column (2.1 mm × 100 mm, 1.7 µm, Waters, Milford, MA). The column temperature was maintained at 40 °C, and the autosampler temperature at 4 °C. A gradient with eluent A (0.1% formic acid in water) and eluent B (0.1% formic acid in acetonitrile) was used at a flow rate of 0.6 mL/min. The gradient was linearly increased from 10 to 95% B over 7 min, then held at 95% B for 2 min and then brought back to initial condition and held for 2.5 min. The mass spectrometer was operated in positive ionization electrospray mode with a capillary temperature of 320 °C. The Fourier transform-mass spectrometry (FT-MS) data were acquired at a resolving power of 12 500 in a range of m/z 85-850. This resolving power was considered adequate for molecules in the range and yielded a duty cycle that was compatible with UHPLC. Data Analysis of Pooled Samples for Biomarker Identification. The high-resolution LC-MS data of the pooled ibipinabanttreated samples were processed with background subtraction using the data of the pooled control samples as controls. The background subtraction algorithm has been described elsewhere.29-31 In the following statements, an analyte scan denotes a mass spectrometric data acquisition event at a chromatographic time point on a treated sample; a control scan denotes a similar event on a control sample. In brief, the software defined a range of control scans based on a specified time window (±0.5 min in this study) around the chromatographic time point of an analyte scan so that the accurate mass data of the control sample contained within that window could be considered for matrix ion checking. The dynamic “range of control scans” algorithm was looped throughout the analyte scans in an LC-MS data set. Once the control scans were defined for each analyte scan, the actual subtraction of any ion in the spectrum of the analyte scan was performed by first identifying the same ion in the spectra of the control scans (masses in analyte vs control data were matched as long as they fell within a specified mass tolerance window around the analyte masses, which was set to ±20 ppm in this study). The highest intensity of the identified ion in the spectra of the control scans 3836

Analytical Chemistry, Vol. 82, No. 9, May 1, 2010

was then determined and multiplied with a specified scaling factor (2) and was subtracted from that of the ion in the spectrum of the analyte scan. The background-subtracted data were divided into two parts using mass defect filtering. The mass defect filter window was set to -40 to +50 mDa around the mass defect (0.0738) of the protonated ion of ibipinabant over a mass range of ±50 Da around the mass of ibipinabant (m/z 487). The MDF-retained data and the filtered-out data were each output into separate files. This process was done with an in-house script described elsewhere.34,35 Preparation of Individual Dog Plasma Samples. Aliquots (100 µL) of each individual plasma sample was mixed with 200 µL of cold methanol containing ∼3 ng/mL D9-C14:0 acylcarnitine (internal standard). The suspension was vortexed and centrifuged at 2000g for 10 min. Aliquots of the supernatants were subject to LC-MS analyses. LC-MS Analysis of Acylcarnitine Levels in Individual Dog Plasma Samples. An LC-MS method was developed to specifically measure the levels of the major identified acylcarnitines in individual plasma samples of the 6 week study. The HPLC was performed on an XTerra MS C18 column (2.1 mm × 50 mm, 5 µm, Waters, Milford, MA). The column temperature was maintained at 40 °C, and the autosampler temperature at 4 °C. A gradient with eluent A (10 mM ammonium bicarbonate in water) and eluent B (80% acetonitrile containing 10 mM ammonium bicarbonate) was used at a flow rate of 1.0 mL/min. The gradient was held at 2% B for 1 min and then linearly increased from 2 to 100% B over 4 min, then held at 100% B for 2 min, and then brought back to initial conditions and held for 2 min. The mass spectrometry analysis was performed on a Micromass Quattro Ultima (Waters, Milford, MA) in positive electrospray mode. Specific multiple reaction monitoring (MRM) transitions were set to monitor C12:0 acylcarnitine (m/z 344 to m/z 85), C14:0 acylcarnitine (m/z 372 to m/z 85), C16:0 acylcarnitine (m/z 400 to m/z 85), and D9-C14:0 acylcarnitine as an internal standard (m/z 381 to m/z 85). The collision energy was set to 27 eV. RESULTS Results from the Analysis of Pooled Samples for the Identification of Potential Biomarkers. In Figure 1, base peak chromatograms are displayed showing profiles of the original, unprocessed high-resolution LC-MS data of the pooled plasma samples. The profiles of all four samples look similar to each other, and potential biomarker peaks cannot be easily discernible between the ibipinabant-treated and control groups. In Figure 2, chromatograms are displayed showing profiles of the weeks 2 and 3 data of the pooled ibipinabant-treated samples each being background-subtracted with data of the two pooled control samples. A number of potential biomarkers are revealed in both profiles as distinct peaks. It is noteworthy that these peaks are of 2% or less intensity relative to the maximum intensity of the unsubtracted data shown in Figure 1, and thus they were buried under the baseline when unprocessed. In addition to the potential biomarker peaks that exhibit in both of the pooled drugtreated samples, a few extra peaks (e.g., those marked with “×”) (34) Zhang, H.; Zhang, D.; Ray, K. J. Mass Spectrom. 2003, 38, 1110–1112. (35) Zhang, H.; Zhu, M.; Ray, K. L.; Ma, L.; Zhang, D. Rapid Commun. Mass Spectrom. 2008, 22, 2082–2088.

Figure 1. Original unprocessed base peak chromatograms of pooled plasma samples.

Figure 2. Base peak chromatograms of weeks 2 and 3 pooled plasma samples from the ibipinabant-treated group processed with background subtraction.

exist in one profile but not in another. Because of such inconsistency, these peaks are excluded from consideration for potential biomarkers. The background-subtracted data of the pooled treated samples were further processed with mass defect filtering to separate out drug metabolite peaks. In Figure 3a,b, chromatograms are displayed showing profiles of the data that were “filtered out” by mass defect filter, in which most peaks in Figure 2 are still present but a few are removed. These few peaks are presented in the profiles of the mass defect filter-retained data as major distinct peaks (Figure 3c,d). Their identities were confirmed to be ibipinabant and its desmethyl and hydroxyl metabolites, based on their accurate mass m/z values and data from a follow-up listdependent tandem mass spectrometry (MS/MS) experiment. The potential biomarker peaks shown in Figure 3a,b were investigated to determine their identities. Except for the retention time (RT) 7.9 peak, all other major peaks were determined to be

acylcarnitines. This was based on their accurate mass m/z values and data from a follow-up list-dependent MS/MS experiment. The RT 7.9 peak was a cluster of m/z 648.4204 and m/z 664.4171 ions. The intensity differences of the RT 7.9 peak between the treated and control sample groups were small and subtle, and a scale factor of 3 (instead of 2) in the background subtraction process would remove this peak in the profiles. Therefore, no further efforts were spent to identify it because of the lack of practical value. Results from the Measurement of Individual Samples to Validate the Acylcarnitine Markers. Further study was conducted to investigate the statistical relevance and validity of the acylcarnitine peaks as biomarkers related to the observed muscle toxicity. This was done by measuring acylcarnitine levels in individual plasma samples for both the ibipinabant-treated and control groups. The measurements were carried out with MRM experiments by monitoring specific acylcarnitine-to-carnitine headAnalytical Chemistry, Vol. 82, No. 9, May 1, 2010

3837

Figure 3. Base peak chromatograms of pooled plasma samples from the ibipinabant-treated group processed with background subtraction followed by mass defect filtering.

Figure 4. Levels of C12:0, C14:0, and C16:0 acylcarnitines measured relative to D9-C14:0 acylcarnitine standard. The error bar is standard error (N ) 6).

group (m/z 85) transitions. Because of duty-cycle constraints, only C12:0, C14:0, and C16:0 acylcarnitines were measured as representatives, although a total of 19 acylcarnitines were identified in the pooled samples. In Figure 4, a plot is shown to depict the relative levels of C12:0, C14:0, and C16:0 acyl carnitines between the treated and the control groups in the duration of the 6-week study. The acylcarnitine levels in the treated group were elevated over the control group as early as week 1 with statistical significance. The acylcarnitines were most elevated in week 2 and remained significantly increased through week 6. DISCUSSION The first part of the pooled sample strategy for metabolomics is the analysis of a few pooled samples to identify potential marker 3838

Analytical Chemistry, Vol. 82, No. 9, May 1, 2010

candidates. This is difficult to implement for a number of reasons including the chemical diversity of small molecule metabolites, the confounding factor of biological variation to the detection of true changes of potential biomarker candidates, and the intrinsic complexity of delineating the chromatography-mass spectrometry data to extract useful information. In this article, however, we were able to detect potential biomarker candidates by analyzing a few pooled samples using a background subtraction approach. With high-resolution mass spectrometry data, this background subtraction approach was found thorough and effective. By applying a small scale factor setting in the background subtraction operation, we were able to discern biomarker candidates amid a multitude of biological interferences. Additional techniques were used to enhance the profiling, including the use of mass defect filtering to remove drug metabolite components and the comparison of profiles between pooled samples to eliminate inconsistent peaks. The comparison of profiles between two or more pooled samples inagroup,similartothesubpoolingmethodusedintranscriptomics,36-38 helped to improve the statistical viability of the pooled sample strategy. The overall success of the pooled sample strategy lies in the availability of good control samples. The opportunity to obtain good control samples requires inclusion of collection in the study design and the appropriate nature of samples to be analyzed. It is assumed that by pooling, typical biological variations among individuals are generally averaged out, and therefore the majority of metabolites are of similar concentrations between the treated and control groups. For urine metabolomics where metabolite concentrations may shift because of volume, the volume to be (36) Peng, X.; Wood, C. L.; Blalock, E. M.; Chen, K. C.; Landfield, P. W.; Stromberg, A. J. BMC Bioinf. 2003, 4, 26. (37) Kendziorski, C; Irizarry, R. A.; Chen, K. S.; Haag, J. D.; Gould, M. N. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 4252–4257. (38) Kendziorski, C. M.; Zhang, Y.; Lan, H.; Attie, A. D. Biostatistics 2003, 4, 465–477.

pooled from each individual samples may have to be adjusted to “normalize” the content of the resultant pooled samples.39 After the identification of potential marker candidates with pooled samples, statistical validation needs to be followed with sets of individual samples. The ability to decouple these two parts of analysis with the pooled sample strategy is advantageous because statistical analysis is now used as a confirmation tool instead of a change-detection tool and thus can be conducted more effectively. In addition, the data acquisition for statistical analysis can be optimized and be more efficient as well. Another advantage of decoupling the two parts is that the first part of marker candidate identification is done with a few pooled samples and is therefore more amenable to a thorough analytical interrogation than a large set of individual samples would be. This is true for both data processing and data generation. In our case, for example, should the LC-MS condition fail to detect any significant marker candidates, a different analytical method would have been tried to focus on a different spectrum of the chemical diversity. A potential limitation of the pooled sample strategy is that it only identifies metabolites of change according to the predefined treatment and control groups; this may limit its use for a study investigating multiple biological parameters. Of course, this can be addressed by generating pooled samples of different treated and control groups in the study design stage. Another likely limitation of the pooled sample strategy is that only robust changes with relatively small biological variations are identified. Subtle changes (e.g., less than 2-fold difference) or changes with large biological variations may not be picked up or may be averaged out in the sample pooling process. This would not be an issue for identifying robust markers of practical values, but it may be a concern for mechanism studies where subtle changes of relatively (39) Warrack, B. M.; Hnatyshyn, S; Ott, K.-H.; Reily, M. D.; Sanders, M.; Zhang, H.; Drexler, D. M. J. Chromatogr., B 2009, 877, 547–552. (40) Shchelochkov, O.; Wong, L. J.; Shaibani, A.; Shinawi, M. Muscle Nerve 2009, 39, 374–382. (41) Minkler, P. E.; Kerner, J; North, K. N.; Hoppel, C. L. Clin. Chim. Acta 2005, 352, 81–92. (42) Vianey-Liaud, C.; Divry, P.; Gregersen, N.; Mathieu, M. J. Inher. Metab. Dis. 1987, 10, 159–198 (Suppl. 1).

large variability are still of value to identify. Because of this concern, it is desirable that, after the identification of marker candidates with pooled samples, the analysis of individual samples should be performed in a global mode so that the data may still be used for statistical detection of additional changes when necessary. In a practical sense though, as in this case study, the identification and confirmation of major metabolite biomarkers in itself will typically lead to additional studies to investigate the related components. In summary we have demonstrated that, with a pooled sample strategy enabled by high-resolution LC-MS-based background subtraction, we were able to identify toxicology markers in ibipinabant-treated dogs in a metabolomic study. The statistical confirmation of the elevations of the identified acylcarnitines in dog plasma contributed to the understanding of the mechanism of the observed dog muscle toxicity. The drastic elevations of acylcarnitine levels in circulation appear to be early markers of an altered lipid metabolism,40-42 and histopathology findings indicated that such elevations preceded muscle damage. The acylcarnitines were eventually validated to be early markers of the species-specific toxicity. The methodology demonstrated in this article should be applicable to similar evaluations to detect potential biomarkers in an altered metabolome. ACKNOWLEDGMENT We would like to thank Xin Wang for developing and implementing the background subtraction algorithm. We thank Bethanne Warrack for assisting some of the UHPLC experiments, and Jonathan Josephs and William Humphreys for a critical reading of this manuscript. Part of this work is an actualization of a concept presented at the 53rd ASMS conference (Zhang, H.; Warrack, B.; Hnatyshyn, S.; Aranibar, N.; Friedrichs, M.; Ott, K.H.; Ray, K.; Sanders, M. 53rd ASMS Conference on Mass Spectrometry and Allied Topics, San Antonio, TX, June 5-9, 2005). Received for review February 1, 2010. Accepted March 31, 2010. AC100287A

Analytical Chemistry, Vol. 82, No. 9, May 1, 2010

3839