Metabonomics study on the effects of the ginsenoside Rg3 in a β-cyclodextrin-based formulation on tumor-bearing rats by a fully automatic hydrophilic interaction/ …

Metabonomics Study on the Effects of theGinsenoside Rg3 in a �-Cyclodextrin-BasedFormulation on Tumor-Bearing Rats by a FullyAutomatic Hydrophilic Interaction/Reversed-PhaseColumn-Switching HPLC-ESI-MS Approach

Yuan Wang,† Jiangshan Wang,† Ming Yao,‡ Xinjie Zhao,† Jens Fritsche,§

Philippe Schmitt-Kopplin,| Zongwei Cai,⊥ Dafang Wan,‡ Xin Lu,† Shengli Yang,†,‡,# Jianren Gu,‡

Hans Ulrich Haring,¶ Erwin D. Schleicher,¶ Rainer Lehmann,*,¶ and Guowang Xu*,†

National Chromatography Research and Analysis Center, Dalian Institute of Chemical Physics, Chinese Academy ofSciences, 16023 Dalian, China, State Key Laboratory of Oncogenes and Related Genes, Shanghai Cancer Institute,Shanghai Jiao Tong University, 200032 Shanghai, China, Immatics Biotechnologies GmbH, 72076 Tuebingen,Germany, Helmholtz-Zentrum MuenchensGerman Research Center for Environmental Health, Institute for EcologicalChemistry, Ingoldstaedter Landstrasse 1 D-85764 Neuherberg, Germany, Department of Chemistry, Hong KongBaptist University, Kowloon, Hong Kong SAR, China, Shanghai Research Center of Biotechnology, Chinese Academyof Sciences, Shanghai 200032, China, and Department of Internal Medicine 4, Division of Clinical Chemistry andPathobiochemistry (Central Laboratory), University Hospital Tuebingen, D-72076 Tuebingen, Germany

The goal of this study was the application of a novel, fullyautomatic column-switching approach in a metabonomicsstudy combining the orthogonal selectivities of hydrophilicinteraction chromatography (HILIC) and reversed-phasechromatography. The temporal, pharmacodynamic effectsof the ginsenoside Rg3 on the metabonome in urine ofhealthy and liver-tumor-bearing rats have been investi-gated. Within a total analysis time of 52 min we detected5686 polar, and on the second column an additional1808 apolar, urinary metabolite ions. The administrationof a single, high dose of Rg3 in a �-cyclodextrin-basedformulation led to a considerable change of the metabolicpattern in cancer rats during 3 days studied. Seventeenbiomarker candidates including three apolar metabolites,which were not retained on the HILIC column, weredetected. Overall, the results suggest that the developedliquid chromatography-mass spectrometry strategy is apromising tool in metabonomics studies for global analy-sis of highly complex biosamples. It may not only increasethe number of discovered biomarkers but consequentlyimprove the comprehensive information on metabolicchanges in a fully automatic manner.

Over the past decade, the application of metabonomicsanalyses of body fluids increased considerably in various research

fields like pharmaceutical discovery and development,1,2 nutritionand epidemiology investigations,3 research projects for the evalu-ation of drug toxicity,4 studies for the diagnosis and treatment ofdiseases,5 and discovery of novel biomarkers.6–10 For the analysisof body fluids urine is the most favorable sample material, becauseit can be collected noninvasively. Furthermore, urine contains agreat number of metabolites indicating the metabolic state of theorganism5 and reflecting specific traits to perturbations.11–17 Onthe basis of the aqueous character of urine all excreted metabolites

* Corresponding authors. Fax: +86-411-84379559 (G.X.), +49-7071-5348 (R.L.).E-mail: [email protected] (G.X.), [email protected] (R.L.).

† Dalian Institute of Chemical Physics.‡ Shanghai Jiao Tong University.§ Immatics Biotechnologies GmbH.| Institute for Ecological Chemistry.⊥ Hong Kong Baptist University.# Shanghai Research Center of Biotechnology.¶ University Hospital Tuebingen.

(1) Robertson, D. G.; Reily, M. D.; Baker, J. D. J. Proteome Res. 2007, 6, 526–539.

(2) Lindon, J. C.; Holmes, E.; Nicholson, J. K. FEBS J. 2007, 274, 1140–1151.(3) Stella, C.; Beckwith-Hall, B.; Cloarec, O.; Holmes, E.; Lindon, J. C.; Powell,

J.; van der Ouderaa, F.; Bingham, S.; Cross, A. J.; Nicholson, J. K. J. ProteomeRes. 2006, 5, 2780–2788.

(4) Ekman, D. R.; Keun, H. C.; Eads, C. D.; Furnish, C. M.; Rockett, J. C.; Dix,D. J. Metabolomics 2006, 2, 63–73.

(5) Chen, J.; Zhao, X.; Fritsche, J.; Yin, P.; Schmitt-Kopplin, P.; Wang, W.; Lu,X.; Haring, H. U.; Schleicher, E. D.; Lehmann, R.; Xu, G. Anal. Chem. 2008,80, 1280–1289.

(6) Holmes, E.; Cloarec, O.; Nicholson, J. K. J. Proteome Res. 2006, 5, 1313–1320.

(7) Marchesi, J. R.; Holmes, E.; Khan, F.; Kochhar, S.; Scanlan, P.; Shanahan,F.; Wilson, I. D.; Wang, Y. L. J. Proteome Res. 2007, 6, 546–551.

(8) Jia, L. W.; Wang, C.; Kong, H. W.; Cai, Z. W.; Xu, G. W. Metabolomics 2006,2, 95–104.

(9) Jia, L. W.; Wang, C.; Kong, H. W.; Yang, J.; Li, F. L.; Lv, S.; Xu, G. W.J. Pharm. Biomed. Anal. 2007, 43, 646–654.

(10) Yang, J.; Zhao, X. J.; Liu, X. L.; Wang, C.; Gao, P.; Wang, J. S.; Li, L. J.; Gu,J. R.; Yang, S. L.; Xu, G. W. J. Proteome Res. 2006, 5, 554–561.

(11) Ebbels, T. M. D.; Keun, H. C.; Beckonert, O. P.; Bollard, M. E.; Lindon,J. C.; Holmes, E.; Nicholson, J. K. J. Proteome Res. 2007, 6, 4407–4422.

(12) 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.

(13) Seiwert, B.; Karst, U. Anal. Chem. 2007, 79, 7131–7138.(14) Lee, S. H.; Woo, H. M.; Jung, B. H.; Lee, J.; Kwon, O. S.; Pyo, H. S.; Choi,

M. H.; Chung, B. C. Anal. Chem. 2007, 79, 6102–6110.(15) Maher, A. D.; Zirah, S. F. M.; Holmes, E.; Nicholson, J. K. Anal. Chem.

2007, 79, 5204–5211.

Anal. Chem. 2008, 80, 4680–4688

10.1021/ac8002402 CCC: $40.75 2008 American Chemical Society4680 Analytical Chemistry, Vol. 80, No. 12, June 15, 2008Published on Web 05/09/2008

are more or less polar. But most of the current chromatographic-driven metabonomics studies performed the analysis of the uri-nary metabonome using reversed-phase (RP) chromatography16–24

where polar compounds like organic acids, pyrimidines, smallpeptides, and many others often exhibit little or no retention.Additives, like ion pair reagents, which may solve this analyticalproblem are inappropriate since they disturb the mass spectro-metric (MS) analysis. A useful alternative is hydrophilic interactionchromatography (HILIC), because polar compounds can beretained using organic/water eluents and it is well suited for liquidchromatography-mass spectrometry (LC-MS) hyphenation. Upto now some metabonomics applications using HILIC have beenpublished.25–28 In HILIC the elution is in the order of increasinghydrophilicity, which may lead to the loss of nonretained, weaklypolar and apolar compounds excreted in the urine and, conse-quently, to the loss of potential biomarkers and information onmetabolic pathway alterations. To circumvent this problem, veryrecently data of the same rat urine samples achieved on an HILICand on an RP column were combined,27,28 thereby accepting twosample injections and a decrease in analytical speed and samplethroughput.

To utilize the advantages of the complementary analyticalpower of HILIC and RP-LC-MS and to circumvent the necessityof two injections, loss of speed, and sample throughput wedeveloped a novel on-line HILIC/RP-LC-MS system based onsingle sample injection and fully automatic column-switchingtechnique.29 Here we demonstrate the practical use of this LC-MSsystem studying the effects of a single, high dose administrationof the ginsenoside Rg3 on the urinary metabonome of tumor-bearing male Sprague-Dawley (SD) rats.

EXPERIMENTAL SECTIONChemicals and Reagents. HPLC-grade acetonitrile (ACN),

formic acid, and ammonium acetate were purchased from Tedia

(Fairfield, OH). Ammonium tribasic phosphate was purchasedfrom Kemio Chemical Reagent (Tianjin, China). Ammonia wasfrom Lianbang Reagent Factory (Shenyang, China).

Creatinine, butyrobetaine, creatine, 5-hydroxytryptamine, andkynurenic acid were purchased from Sigma (Sigma-Aldrich, St.Louis, MO). N,N-Dimethylglycine and pipecolinic acid werepurchased from Alfa Aesar (Tianjin, China). Ginsenoside Rg3 wasprovided by Dalian industrial university.

Sample Collection and Preparation. Rg3 was dissolved in5% �-cyclodextrin (�-CD) by ultrasonification. Six male SD rats,which had been inoculated with Walker 256 tumor, were perfusedwith a single dose of 100 mg/kg of the Rg3 solvent (marked as“cancer+Rg3”). The perfusion was performed as described be-fore.30 Six healthy male SD rats were dosed in the same way(“control+Rg3”), and groups of six cancer (“cancer”) and sixhealthy (“control”) male SD rats were perfused with the 5% �-CDsolution without Rg3. Twenty-four hour urines from the rats werecollected before perfusion (day 0) and for 3 consecutive days (day1 to day 3). In total, 94 urine samples were obtained. Two samplesfrom untreated cancer rats at day 3 failed to be collected properly.

One milliliter of each sample was deproteinized by adding 1.5mL of organic solvent (ACN/MeOH, 3/1, v/v) followed bycentrifugation (12 000 rpm, 15 min). The supernatant was lyoph-ilized and resolved in 500 µL of ACN/MeOH/formic acid, 650/350/2 (v/v/v) and filtered before injection (0.22 µm filter).

Chromatographic Equipment. The column-switching HPLCsystem was equipped with an auto sampler (SIL-20A, Shimadzu,Kyoto, Japan), five solvent delivery pumps (P1-P5; LC-20AB orLC-20AT from Shimadzu), two analytical columns (column 1 andcolumn 2), a transfer column (column T), an electronicallycontrolled 10-port 2-position valve (7725i, Rheodyne, Rohnert Park,CA), a 6-port 2-position valve (FCV-20AH2, Shimadzu), a 20A SUSmixer (Shimadzu), and two system controllers (CBM-20Alite,Shimadzu). The separation results were recorded by one massspectrometer (API Q trap linear ion trap; Applied Biosystems, PaloAlto, CA). Details about the column-switching process are givenin Table 1.

Separation Conditions. An HILIC Si column (3 µm, 100 Å,150 mm × 2.1 mm i.d.) equipped with a guard column (10 mm ×2.1 mm i.d.) both from Waters, MA was used as column 1. Thetransfer column and column 2 were identical (C-18 Hypersil ODS2,5 µm, 100 Å, 150 mm × 4.6 mm i.d., Elite, Dalian, China). Themobile phase A for column 1 (HILIC) was 10 mM CH3COONH4

dissolved in ACN/water (9/1; v/v), and mobile phase B was 10mM CH3COONH4 in ACN/water (1/1; v/v). The gradient elutionstarted from 12.5% mobile phase B up to 100% B in 15 min, whichwas kept for 10 min. The flow rate was 250 µL/min, which wasreduced to 180 µL/min by a splitting device in front of the MS.The transfer solvent contained 3% ACN with 0.1% formic acid (flowrate of 0.75 mL/min in step 1 and 2.0 mL/min in step 2). Forcolumn 2, the mobile phase A was 0.1% aqueous formic acid andmobile phase B was 100% ACN. Gradient elution started from 25%B for 25 min, then increased to 95% B in 10 min, which was keptfor 15 min, then decreased to 25% B in 2 min (flow rate 1.0 mL/min on-column; decreased to 120 µL/min by postcolumn splitting).

(16) Hodson, M. P.; Dear, G. J.; Roberts, A. D.; Haylock, C. L.; Ball, R. J.; Plumb,R. S.; Stumpf, C. L.; Griffin, J. L.; Haselden, J. N. Anal. Biochem. 2007,362, 182–192.

(17) Lutz, U.; Lutz, R. W.; Lutz, W. K. Anal. Chem. 2006, 78, 4564–4571.(18) Forshed, J.; Idborg, H.; Jacobsson, S. P. Chemom. Intell. Lab. Syst. 2007,

85, 102–109.(19) Jonsson, P.; Bruce, S. J.; Moritz, T.; Trygg, J.; Sjostrom, M.; Plumb, R.;

Granger, J.; Maibaum, E.; Nicholson, J. K.; Holmes, E.; Antti, H. Analyst2005, 130, 701–707.

(20) Plumb, R. S.; Stumpf, C. L.; Gorenstein, M. V.; Castro-Perez, J. M.; Dear,G. J.; Anthony, M.; Sweatman, B. C.; Connor, S. C.; Haselden, J. N. RapidCommun. Mass Spectrom. 2002, 16, 1991–1996.

(21) Wagner, S.; Scholz, K.; Donegan, M.; Burton, L.; Wingate, J.; Volkel, W.Anal. Chem. 2006, 78, 1296–1305.

(22) Wagner, S.; Scholz, K.; Sieber, M.; Kellert, M.; Voelkel, W. Anal. Chem.2007, 79, 2918–2926.

(23) Cloarec, O.; Campbell, A.; Tseng, L. H.; Braumann, U.; Spraul, M.; Scarfe,G.; Weaver, R.; Nicholson, J. K. Anal. Chem. 2007, 79, 3304–3311.

(24) Crockford, D. J.; Lindon, J. C.; Cloarec, O.; Plumb, R. S.; Bruce, S. J.; Zirah,S.; Rainville, P.; Stumpf, C. L.; Johnson, K.; Holmes, E.; Nicholson, J. K.Anal. Chem. 2006, 78, 4398–4408.

(25) Cubbon, S.; Bradbury, T.; Wilson, J.; Thomas-Oates, J. Anal. Chem. 2007,79, 8911–8918.

(26) Bajad, S. U.; Lu, W. Y.; Kimball, E. H.; Yuan, J.; Peterson, C.; Rabinowitz,J. D. J.Chromatogr., A 2006, 1125, 76–88.

(27) Idborg, H.; Zamani, L.; Edlund, P. O.; Schuppe-Koistinen, I.; Jacobsson,S. P. J. Chromatogr., B 2005, 828, 9–13.

(28) Kind, T.; Tolstikov, V.; Fiehn, O.; Weiss, R. H. Anal. Biochem. 2007, 363,185–195.

(29) Wang, Y.; Lu, X.; Xu, G. In Proceedings of Dalian International Symposiaand Exhibition on Chromatography (30th ISCC, 4th GCxGC, NISEC); Xu,G., Sandra, P., Zhang, Y., Marriott, P., Eds.; Chemical Industry Press;Beijing, China, 2007; pp 213-214.

(30) Qian, T. X.; Cai, Z. W.; Wong, N. S. R.; Mak, N. K.; Jiang, Z. H.J. Chromatogr., B 2005, 816, 223–232.

4681Analytical Chemistry, Vol. 80, No. 12, June 15, 2008

The total run time, including separation and equilibration time,for one sample was 52 min.

The results were recorded by MS in the positive enhancedMS (EMS) mode in the m/z range 100-800. The sourcetemperature was set at 250 °C with nitrogen at a curtain gaspressure of 25 psi, the nebulization gas pressure of 30 psi, andauxiliary gas pressure of 35 psi. The ion spray voltage was 5000V. The declustering potential was 40 V. The collision energy was10 eV, and collision gas was set at low. The LIT fill time was 20ms with no Q0 trapping. The scan rate was 1000 amu/s.

HPLC-MS/MS Experiment. Information-dependent acquisi-tion (IDA) mode was applied to perform survey scans of EMSand dependent enhanced product ions (EPI) were sequentiallyrecorded during the whole HPLC run to obtain tandem MS spectraof the first three ions with the highest intensity in a scan cycle.The HPLC and EMS parameters were the same as mentionedabove, but the scan rate was 4000 amu/s to get a higher dutycycle. In the EPI scan mode of each cycle, the resolution of Q1was set at “low” for higher sensitivity. Among the detectedpotential biomarkers were also ions with a lower intensity thanthe first three ions in a cycle. Consequently, the product ions ofthese metabolites as well as the structural information could notbe obtained through the IDA mode. Structural information of theseions by EPI scans was achieved separately for each biomarkerapplying different collision energy levels (from 25 to 60 eV).

Peak Alignment. Peak alignment and peak list extraction wereprocessed with the MarkerView software (version 1.1, AppliedBiosystems, U.S.A.). Briefly, the spectra of each sample weredivided into equidistant segments (bins) of 0.5 amu. Then,intensities of the ions in each bin were summed up; zero was givenif no intensity was in the bin. After that, peaks were determinedby centroiding mass and retention time, and variables cor-responded to superset of all peaks in all samples. The tolerancesof both m/z value and retention time for the LC-MS data wereapplied for alignment process. All aligned peaks with an intensityof more than 50 000 (3-fold of noise) were included in the peaklist.

Multivariate Data Analysis. Normalization. The urine volumevaries depending on diet and drinking. Therefore, it is advanta-geous to express the urinary amount of compounds as a functionof creatinine concentration.31 Therefore, we divided the intensityof the metabolites by the individual creatinine concentration in

each sample to ensure direct comparability of the data. Thecreatinine concentration was determined by LC-UV as recentlydescribed.32

Missing Values. According to the “80% rule”,16,33 only variableshaving more than 80% nonzero measurement values in each groupwere handled in the peak list. After that, the matrices for hydrophilicand hydrophobic metabolites were merged for the following dataanalysis, containing 606 polar and 185 apolar metabolites, respectively.

Multivariable Data Analysis. The principle componentanalysis (PCA) and partial least-squares-discriminant analysis(PLS-DA) were performed with SIMCA-P 11.0 software (Umetrics,UMEÅ, Sweden).

PCA. As an unsupervised multivariate data analysis method,PCA34 is used to give an overview of the clustering trends in high-dimensional data. The data matrix X containing the LC-MSintensities for each metabolite in each observation is decomposedinto so-called principal components: X ) T ·PT. These can bedisplayed as a set of scores (T) focusing on the observations oras a set of loadings (P) focusing on the metabolites. The firstprincipal component (t1, p1) explains most of the variation in thedata matrix X, whereas the second (t2, p2) describes the maximumof the remaining variation, and so on. Therefore, it is possible touse only a small set of components to visualize inherent trendsof the data without loosing important information. To evaluate theresponse of the tested rats to administration of Rg3 at healthy orcancer situations and to investigate day-to-day variations, the PCAscore plot of the first two principal components was used.

PLS-DA. For the exploration of the difference among the testedgroups, supervised methods like PLS-DA34 are more powerful thanthe unsupervised ones by incorporating the known classification.This means that a group label also called response gets associatedwith each observation. PLS-DA is an extension of PCA in the sensethat it calculates principal components that maximize the explainedcovariance between data and response rather than the variancein the data alone. Similar to PCA, the PLS-DA model can bevisualized as score or loading plot. In the present study, the PLS-DA model was constructed using the detected LC-MS data (X

(31) Gehrke, C. W.; Kuo, K. C.; Waalkes, T. P.; Borek, E. Cancer Res. 1979,39, 1150–1153.

(32) Jen, J. F.; Hsiao, S. L.; Liu, K. H. Talanta 2002, 58, 711–717.(33) Bijlsma, S.; Bobeldijk, L.; Verheij, E. R.; Ramaker, R.; Kochhar, S.;

Macdonald, I. A.; van Ommen, B.; Smilde, A. K. Anal. Chem. 2006, 78,567–574.

(34) Eriksson, L.; Johansson, E.; Kettaneh-Wold, N.; Trygg, J.; Wikstrom, C.;Wold, S. Multi- and Megavariate Data Analysis; Umetrics Academy: Umea,Sweden, 2006.

Table 1. Principle of the Fully Automatic Column-Switching Technique

analysis time 10-port valvea 6-port valvea

flow rate of thetransfer solvent

(mL/min) column 1 column T column 2

step 1 (0-2.7 min) 0 0 0.75 25 min gradient(polar compounds)

trapping nonretainedapolar compoundsfrom column 1

pre-equilibrium

step 2 (2.7-4.0 min) 1 0 2.0 waiting forsolutes transferstep 3 (4.0-25.0 min) 1 0 0

step 4 (25.0-43.0 min) 0 1 0 pre-equilibrium forthe next injection

transferring apolarcompounds to column 2

25 min gradient(apolar compounds) and2 min re-equilibriumstep 5 (43.0-52.0 min) 0 0 0.75 pre-equilibrium for the

next injection

a “0” and “1” represent the positions of the dual-position valves.

4682 Analytical Chemistry, Vol. 80, No. 12, June 15, 2008

matrix) and the four tested groups (control, control+Rg3, cancer,and cancer+Rg3) as response (Y matrix).

Validation of PLS-DA models is crucial to ensure that the modelis not overfitted and robust, as well as to exclude models that arejust due to chance. After such a validation, PLS-DA models canbe used to predict the class membership of new data. To validatethe predictive ability of the constructed model, cross-validation34

(7-fold) was performed. In an iterative process, the data set isdivided into seven parts of which six are used for modeling andone to test the model. The error calculated during this test iscombined to the cross-validation parameter Q2 describing howwell the model explains the relationship between the responseand the data. To exclude any by-chance relationship, a responsepermutation test was performed. In this test, the given member-ship of a sample to one of the four groups is randomly permutated(Y permutation). If random permutations yield high Q2 values andtherefore good models, the model for the real class membershipmight also be due to chance.

To identify potential biomarkers in metabonomics studies, itis important to know how much individual metabolites contributeto the principal components of the PLS-DA model. These quanti-ties can be expressed as VIP (variable importance in the projec-tion) values35 or visualized using the S-plot.36 The S-plot visualizesboth the covariance and correlation between the metabolites andthe modeled class designation identifying statistically significantand potentially physiological relevant metabolites. Those metabo-lites showing a small correlation coefficient and large confidenceinterval for their VIP values were excluded from the potentialbiomarker list.

Detection of Exact Mass of the Metabolites. To elucidatethe elemental composition of biomarker candidates a sample rerunwas performed and fractions were collected. High-resolution massspectra of each faction were acquired on a Bruker (Bremen,Germany) APEX Qe Fourier transform ion cyclotron resonanceMS (FTICR/MS) equipped with a 12 T superconducting magnetand an Apollo II ESI source in the positive ion mode. Spectra wereexternally calibrated on clusters of arginine (10 mg/L in metha-nol); calibration errors in the relevant mass range were alwaysbelow 100 ppb.

RESULTS AND DISCUSSIONAnalytical Conditions and Robustness of the Complemen-

tary LC-MS System. A typical base peak chromatogram (BPC)of rat urine applying this complementary HILIC and RP-LC-MSanalysis in the positive ion mode is given in Figure 1A. Figure 1Bshows the corresponding two-dimensional contour plot. The peaksbelow 25 min analysis time represent signals of hydrophilicmetabolite ions from the HILIC column, and the peaks above 25min correspond to hydrophobic metabolites eluted from the RPcolumn. The column-switching technique automatically combininganalytical, transfer, and equilibration steps of all three columnsallows high-throughput analysis in a total run time of 52 min forone sample, followed by the automatic injection of the next sample

(see Table 1). The retention times of 23 typical peaks in the BPCs(13 on the HILIC and 10 on the RP column) showed a relativestandard deviation (RSD) of 0.3-3.2% on the HILIC column and0.05-0.4% on the RP column (n > 100 runs). Sample-dependentnegative effects on the ionization efficiency or the MS pattern,e.g., based on differences in the osmolality or ionic compositionof the urines, were not significant.

Effect of the Ginsenoside Rg3 in a �-CD-Based Formula-tion on the Urinary Metabolic Pattern in Healthy and CancerRats. Rg3 is a minor ginsenoside in Panax ginseng, which hasbeen reported to show various medicinal effects.37–41 First, westudied the effects of Rg3 applying PCA on the data to follow themetabolic patterns over a time period of 3 days subsequent tothe administration. Initially, the urinary metabonome of healthyand cancer rats, both treated at day 0 with Rg3, was investigated(Figure 2A). The PCA analysis revealed a clear influence on theurinary metabonomes of the hepatic tumor-bearing rats. Of note,the “control+Rg3” group did not show any relevant metabolicchanges. Additionally, to exclude any effects of the excipient onthe metabolism, the urinary metabonome of healthy and cancer

(35) Wold, S.; Johansson, E.; Cocchi, M. In 3D QSAR in Drug Design: Theory,Methods and Applications; Kubinyi, H., Ed.; ESCOM: Leiden, The Nether-lands, 1993.

(36) Wiklund, S.; Johansson, E.; Sjostrom, L.; Mellerowicz, E. J.; Edlund, U.;Shockcor, J. P.; Gottfries, J.; Moritz, T.; Trygg, J. Anal. Chem. 2008, 80,115–122.

(37) Wang, C. Z.; Zhang, B.; Song, W. X.; Wang, A. B.; Ni, M.; Luo, X. J.; Aung,H. H.; Xie, J. T.; Tong, R.; He, T. C.; Yuan, C. S. J. Agric. Food Chem. 2006,54, 9936–9942.

(38) Zhang, Q. Y.; Kang, X. M.; Zhao, W. H. Biochem. Biophys. Res. Commun.2006, 342, 824–828.

(39) Lee, H. U.; Bae, E. A.; Han, M. J.; Kim, D. H. Biol. Pharm. Bull. 2005, 28,1992–1994.

(40) Lee, S. H.; Jung, B. H.; Kim, S. Y.; Lee, E. H.; Chung, B. C. Pharm. Res.2006, 54, 46–49.

(41) Joo, S. S.; Won, T. J.; Kim, M. S.; Lee, D. I. Fitoterapia 2004, 75, 337–341.

Figure 1. Simultaneous analysis of polar and apolar metabolites inurine of a healthy rat treated with the ginsenoside Rg3 in a �-CD-based formulation on automatically hyphenated hydrophilic interactionand reversed-phase HPLC columns followed by mass spectrometricdetection. (A) Base peak chromatogram and (B) the correspondingcontour plot. Details on the separation conditions are given in theExperimental Section.

4683Analytical Chemistry, Vol. 80, No. 12, June 15, 2008

Figure 2. Time-dependent changes of the metabolic pattern in urine of (A) healthy (open labels) and cancer (closed labels) rats bothtreated with a single high oral dose of Rg3 in a �-CD-based formulation and (B) healthy and cancer rats both treated with a single oraldose of solely the excipient �-CD. The analyses were performed by principal component analysis (PCA). d0 ) day of Rg3 administration;d1 ) 1 day after Rg3 administration; d2 ) 2 days after Rg3 administration; d3 ) 3 days after Rg3 administration.

4684 Analytical Chemistry, Vol. 80, No. 12, June 15, 2008

rats infused at day 0 solely with �-CD was analyzed (Figure 2B).Unexpectedly, we detected a comparable effect on the metabolicchanges in the cancer rats solely treated with the excipient, similarto that found in the cancer rats infused with Rg3 in a �-CD-basedformulation. Again, no remarkable alterations in the healthycontrols were detected (Figure 2B). The PCA results led us toconclude that a single dose of a 5% solution of �-CD alone exertsgreat effects on the metabolic pattern of tumor-bearing SD rats.On the basis of this finding we searched the literature and foundthat CDs are utilized in pharmaceutical formulations to increasesolubility, reduce bitterness, and enhance stability.42 But mostimportant CDs enhance oral bioavailability of drugs.42,43 World-wide about 30 different pharmaceutical products contain CDs42,43

including �-CD as an excipient for ginsenosides.30,44 Furthermore,in the context with cancer, there is a report demonstrating aneffect of methyl-�-CD by the modification of cell membraneintegrity and cell morphology in a multidrug resistant humancancer cell line.45 The same group showed in animal models theaccumulation of methyl-�-CD in tumor, liver, and kidney tissueafter a single injection.46

Although PCA works to explain the variation between samples,PLS-DA includes the membership (i.e., healthy control, healthy+Rg3, cancer, and cancer+Rg3 rats) of each sample in order toinvestigate variations between the different classes, e.g., changes

caused by Rg3. A comparison of the urines collected at day 1 andday 2 was performed, because the greatest metabolic differencesbetween untreated cancer state (represented by day 0 in the PCAplot) versus treated cancer rats versus healthy controls occurredbetween day 0 and day 1 with less variation between day 1 andday 2 (Figure 2). A model using PLS-DA and randomized sampleorder was constructed analyzing the pooled data from day 1 and2, i.e., the set of samples of each group consisted of the metabolicpattern of 12 urines. The PLS-DA score plot of the first and secondPLS components clearly divided Rg3-treated tumor-bearing ratsfrom healthy controls and from the cancer rats treated with theexcipient resulting in three distinct clusters (Figure 3). In themetabonome of healthy rats treated with Rg3 or excipient onlyslight differences were obvious suggesting that besides the effectsof �-CD alone the perturbation with Rg3 had also a remarkableinfluence on the metabolism of the cancer rats in contrast tohealthy rats. Of note, the PLS-DA analysis of the same samplesperformed with the data on the single column (HILIC or RP) ledto similar PLS-DA score plots (data not shown), suggesting thatthe information achieved by a single-column analysis in a meta-bonomics project may be sufficient to cluster individuals basedon differences in their metabolic state, but nonretained metabolitesand consequently important information about metabolic alter-ations will be lost using only one analytical column (see thefollowing section).

According to the goodness of fit and the predictive power formodels based on the first two, three, four, and five PLS compo-nents, the final model was set to contain four components in orderto achieve an optimal balance between fit and predictive ability.To estimate the predictive ability of our model, we used 7-foldcross-validation, resulting in a goodness of fit of R2 ) 0.842 anda goodness of prediction of Q2 ) 0.652 for the four-component

(42) Carrier, R. L.; Miller, L. A.; Ahmed, I. J. Controlled Release 2007, 123, 78–99.

(43) Loftsson, T.; Vogensen, S. B.; Brewster, M. E.; Konradottir, F. J. Pharm.Sci. 2007, 96, 2532–2546.

(44) Cai, Z. W.; Qian, T. X.; Wong, N. S. R.; Jiang, Z. H. Anal. Chim. Acta 2003,492, 283–293.

(45) Grosse, P. Y.; Bressolle, F.; Vago, P.; Simony-Lafontaine, J.; Radal, M.;Pinguet, F. Anticancer Res. 1998, 18, 379–384.

(46) Grosse, P. Y.; Bressolle, F.; Pinguet, F. Br. J. Cancer 1998, 78, 1165–1169.

Figure 3. PLS-DA score plot of healthy controls (∆), healthy but Rg3-treated (2), cancer (0), and cancer but Rg3-treated (9) rats. A modelusing PLS-DA and randomized sample order was constructed analyzing the pooled data from day 1 and day 2; therefore, the set of samplesof each group consisted of the metabolic pattern of 12 urines.

4685Analytical Chemistry, Vol. 80, No. 12, June 15, 2008

model. A response permutation test (Y scrambling) was used toassess the significance of the predictive ability and to exclude thatour model might be overfitted due to chance correlation.47

According to Eriksson et al.,47 the R2 intercept should not exceed0.4 and the Q2 intercept should not exceed 0.05 for a valid model.We calculated for our model a R2 intercept of 0.390 and a Q2

intercept of -0.235, indicating that the PLS-DA model was valid.Additionally, we tested the predictive ability of our model morerigorously using an external validation set, i.e., to validate themodel with a set of subjects that were not used for building themodel. Eight samples (the “training set”) were selected at randomto construct a PLS-DA model. This model was then used to predict

Figure 4. (A) PLS-DA loading plot. The variables are labeled with m/z and retention time. (B) Potential biomarkers (square circled) in theS-plot and (C) the corresponding loading plot with jack-knifed confidence intervals. The metabolites (E1-E3) marked by an asterisk were excludedfrom the potential biomarker list (Table 2).

Table 2. Potential Biomarkers Sorted by Retention Time

no. m/z (retention time)a,bm/z by FTICR/MS(deviation in ppm) product ions potential biomarker

1 190.2 (3.20) 190.04991 (0.2) 89, 116, 144, 162, 172, kynurenic acidc

2 318.4 (4.56) 318.30028 (0.03) 102, 256, 258, 300 phytosphingosine3 168.2 (6.04) 168.11321 (0.4) 61, 109, 126, 150 6-amino-2-methyl-5-propyl-4-pyrimidinol4 160.2 (6.89) [M + H - NH3]+ n.a.d 115, 117, 132, 142, 160 5-hydroxytryptamine (serotonin)c

5 679.7 (6.92) 679.51094 (0.5) 322, 435, 548, 661 not identified6 130.2 (11.67) n.a. 84 pipecolic acidc

7 212.2 (11.86) 212.10300 (0.2) 109, 124, 148, 194 not identified8 124.2 (11.95) n.a. 83 nicotinic acid/isonicotinic acid9 104.1 (12.55) n.a. 58 N,N-dimethylglycinec

10 132.2 (12.72) n.a. 90 creatinec

11 245.3 (14.28) 245.16088 (0.3) 115, 157, 172, 203 ethyl N2-acetyl-L-argininate12 229.3 (14.88) 229.15470 (0.1) 70, 114, 142, 170 leucylproline13 146.2 (20.02) n.a. 87 butyrobetainec

14 143.2 (20.39) n.a. 84, 114 alloxan15 145.1 (27.38); RP column n.a. 102, 128, 130 4-OH-cyclohexylcarboxylic acid16 205.2 (27.52); RP column 205.08599 (0.3) 85, 102, 128, 145 not identified17 279.3 (31.16); RP column 279.15915 (0.2) tetramethyl-2-(2′-carboxyethyl)-6-hydroxychroman

a Positive mode. b Biomarkers detected on the second, i.e., the RP column show retention times above 25 min. c Identified based on retentiontime and MS/MS spectrum of an authentic standard. d Not analyzed by FTICR/MS.

4686 Analytical Chemistry, Vol. 80, No. 12, June 15, 2008

the class membership in a test set comprising four urines of eachgroup. The PLS-DA model calculated from mass spectrometricdata for the training set predicted accurately the health or cancerstate in 100% of cases in the “test set” and furthermore thetreatment or nontreatment with Rg3 for the cancer rats (data notshown).

Detection and Identification of Biomarker Candidates.The loading plot (Figure 4A) shows the metabolite ions respon-sible for the separation of the different groups. Additionally, an

S-plot was calculated (Figure 4B), visualizing both the covarianceand correlation between the metabolites and the modeled classdesignation,35 which supports the identification of potential biom-arkers based on both the contributions to the model as well astheir reliability. Furthermore, the variable importance for projec-tion (VIP) plot summarizing the contribution of a metabolite ionto the generation of the model has been applied.48 The mostimportant metabolites for the discrimination were assigned bysorting the ions according to their VIP value to select among the791 variables. Metabolite ions with VIP value >1.0 were consid-

(47) Eriksson, L.; Antti, H.; Gottfries, J.; Holmes, E.; Johansson, E.; Lindgren,F.; Long, I.; Lundstedt, T.; Trygg, J.; Wold, S. Anal. Bioanal. Chem. 2004,380, 419–429.

(48) Wold, S.; Sjostrom, M.; Eriksson, L. Chemom. Intell. Lab. Syst. 2001, 58,109–130.

Figure 5. (A) Exact mass determination of the metabolite ion m/z 229.15470 of the loading plot shown in Figure 4A measured with a 12 Telectrospray ionization Fourier transform mass spectrometer in positive mode (attribution of the formula [C11H20N2O3 + H]+ with an accuracy of100 ppb); (B) positive product ion spectrum of m/z 229; (C) proposed fragmentation scheme of the theoretical fragmentation mechanism ofleucylproline in the positive ESI mode; (D) time-dependent effect of a single oral dose of solely the excipient �-CD or Rg3 in a �-CD-basedformulation on leucylproline in the urine in healthy and cancer rats. Significant differences (p < 0.01) are labeled with an asterisk.

4687Analytical Chemistry, Vol. 80, No. 12, June 15, 2008

ered as signals from biomarker candidates. VIP values of allvariables were calculated, and among the top 50 VIPs we detectedalso six apolar compounds which were retained and analyzed onthe second RP column. As shown in Figure 4C, some metabolites(marked with an asterisk) showed great confidence intervals,suggesting that their contribution to the PLS-DA model might becaused by analytical variation. Such metabolites were excludedfrom list. Finally we generated a list of potentially relevantmetabolite ion masses (Table 2).

To identify these metabolites we determined the elementalcomposition (exact mass, the nitrogen rule, and the 13C detectionby positive ESI FTICR/MS), considered the retention behaviors(weak/strong on the HILIC or RP column), investigated production spectra generated by MS/MS experiments, and searchedliterature and metabolite databases. The elucidated molecularstructures were confirmed with authentic standards, if com-mercially available. Exemplarily, the FTICR/MS to elucidate theelemental composition of m/z 229 is given in Figure 5 A. Theexact mass was determined to be 229.15470 (100 ppb accuracy)composed of [C11H20N2O3 + H]+, which revealed by online data-base search six candidate compounds. Consequently, we re-evaluated the product ion spectra (Figure 5B). The product ionsat m/z 70, 114, 142, and 170 led to the unequivocal identificationof leucylproline. A fragmentation scheme of the theoreticalfragmentation mechanism of leucylproline in the positive ESImode is given in Figure 5C. In total, 14 biomarker candidates hadbeen identified (Table 2).

To investigate the effects of the applied treatment on theamount of leucylproline we compared the signal intensity in theurines of all four groups studied (Figure 5D). Interestingly,following a single perfusion solely with the excipient resulted ina slight (but not significant) increase at day 1 and 2 in the cancerrats. In cancer rats treated with Rg3 in a �-CD-based formulationthe increase was strong and significant (Figure 5D). This increasein leucylproline occurred neither in the Rg3-treated healthy ratsnor in the healthy rats perfused with the �-CD solvent (Figure5D). These results indicate that Rg3-dependent alterations of themetabolism in cancer rats may occur until day 2 after administra-tion. Furthermore, it implies that although in the organism thepharmacokinetic described for Rg3 is much shorter (reported half-life is 18.5 min44), Rg3 exerts an alteration on the metabolism fora much longer period in cancer rats most possibly achieved bythe altered bioavailability from the metabolic products of Rg3mediated by the excipient. Because these effects cannot bedetected in Rg3-treated healthy control, we speculate that thedetected urinary metabolic changes mediated by Rg3 (or itsmetabolite Rh2) may be related to a direct influence either onthe metabolism in the tumor tissue or be associated with positiveeffects on tumor-related dysregulation of the general metabolismin mammals. Anticancer effects,37,38,49 hepatoprotective effects,39

neuroprotective effect,40 and hematopoietic effect41 had beendescribed for Rg3.

Moreover, as summarized in Table 2, we detected biomarkercandidates derived from tryptophan metabolism (5-hydroxy-tryptamine, kynurenic acid), sphingolipid metabolism reflecting

apoptotic processes50 (phytosphingosine), lysine degradation(pipecolic acid, butyrobetaine), purine metabolism (alloxan),glycine metabolism (dimethylglycine, creatine), and nicotinatemetabolism (nicotinic acid/isonicotinic acid). Furthermore, thewater-soluble tocopherol metabolite tetramethyl-2-(2′-carboxy-ethyl)-6-hydroxychroman, as well as 6-amino-2-methyl-5-propyl-4-pyrimidinol, ethyl N2-acetyl-L-argininate and the organic acid 4-OH-cyclohexylcarboxylic acid (4-HCH) were identified. 4-HCH is anorganic acid detected in human urine, which was proved to be acompound from bacterial origin possibly from gastrointestinaltract.51,52 The Rg3-mediated increase in 4-HCH led us to speculateabout a potential influence of Rg3 on the gut flora. Of note, afteroral intake Rg3 itself is metabolized to the ginsenoside Rh2 byhuman fecal microflora,53 most potently by Bacteroides sp.,Bifidobacterium sp., and Fusobacterium sp.53 Important to note,this useful information about a potential influence of Rg3 treatmenton the gut flora, detected by the increase in 4-HCH in cancer rats,would not have been detected applying single-column analysisbecause 4-HCH was not retained on the HILIC column (seeTable 2).

CONCLUSIONSA novel, complementary on-line hyphenated HPLC-MS ap-

proach based on column-switching technique for the simultaneousanalysis of polar and apolar compounds combining hydrophilicinteraction and reversed-phase chromatography is presented. Theresults clearly demonstrate the capability of this fully automaticHPLC-MS as a novel tool for metabonomics studies. In contrastto a single-column approach, nonretained metabolites can bedetected on the second, complementary column thereby leadingto an increase in the number of discovered metabolites andbiomarker candidates, as well as extended knowledge aboutmetabolic traits under the studied conditions, which was validatedin this study investigating the effects of the ginsenoside Rg3 in a�-CD-based formulation on the metabolism of liver-tumor-bearingrats.

ACKNOWLEDGMENTThis study has been supported by the foundation for Distin-

guished Young Scholars from the National Natural ScienceFoundation of China (No. 20425516), the 11th five year of planprojects from the Ministry of Science and Technology of China(2007DFA31060), the Knowledge Innovation Program of theChinese Academy of Sciences (K2006A12, K2006A14), and theSino-German Center for Research Promotion (DFG and NSFC,GZ 364).

Received for review February 1, 2008. Accepted April 4,2008.

AC8002402

(49) Liu, W. K.; Xu, S. X.; Che, C. T. Life Sci. 2000, 67, 1297–1306.

(50) Nagahara, Y.; Shinomiya, T.; Kuroda, S.; Kaneko, N.; Nishio, R.; Ikekita,M. Cancer Sci. 2005, 96, 83–92.

(51) Kronick, J. B.; Mamer, O. A.; Montgomery, J.; Scriver, C. R. Clin. Chim.Acta 1983, 132, 205–208.

(52) Swell, A. C.; Bohles, H. J. Clin. Chem. 1991, 37, 1301–1302.(53) Bae, E. A.; Han, M. J.; Kim, E. J.; Kim, D. H. Arch. Pharm. Res. 2004, 27,

61–67.

4688 Analytical Chemistry, Vol. 80, No. 12, June 15, 2008