Journal of Chromatography A, 1262 (2012) 130137

Contents lists available at SciVerse ScienceDirect

Journal of Chromatography A
journal homepage: www.elsevier.com/locate/chroma

Comprehensive analysis of commercial willow bark extracts by new technology platform: Combined use of metabolomics, high-performance liquid chromatographysolid-phase extractionnuclear magnetic resonance spectroscopy and high-resolution radical scavenging assay
Sara Agnolet a,b , Stefanie Wiese b , Robert Verpoorte c , Dan Staerk a,b,
a

Department of Drug Design and Pharmacology, Faculty of Health and Medical Sciences, University of Copenhagen, Universitetsparken 2, DK-2100 Copenhagen, Denmark Department of Basic Sciences and Environment, Faculty of Life Sciences, University of Copenhagen, Thorvaldsensvej 40, DK-1871 Frederiksberg, Denmark c Natural Products Laboratory, Institute of Biology, Leiden University, NL-2300RA Leiden, The Netherlands
b

a r t i c l e

i n f o

a b s t r a c t
Here, proof-of-concept of a new analytical platform used for the comprehensive analysis of a small set of commercial willow bark products is presented, and compared with a traditional standardization solely based on analysis of salicin and salicin derivatives. The platform combines principal component analysis (PCA) of two chemical ngerprints, i.e., HPLC and 1 H NMR data, and a pharmacological ngerprint, i.e., high-resolution 2,2 -azino-bis(3-ethylbenzothiazoline-6-sulfonate) radical cation (ABTS + ) reduction prole, with targeted identication of constituents of interest by hyphenated HPLCsolidphase extractiontube transfer NMR, i.e., HPLCSPEttNMR. Score plots from PCA of HPLC and 1 H NMR ngerprints showed the same distinct grouping of preparations formulated as capsules of Salix alba bark and separation of S. alba cortex. Loading plots revealed this to be due to high amount of salicin in capsules and ampelopsin, taxifolin, 7-O-methyltaxifolin-3 -O-glucoside, and 7-O-methyltaxifolin in S. alba cortex, respectively. PCA of high-resolution radical scavenging proles revealed clear separation of preparations along principal component 1 due to the major radical scavengers (+)-catechin and ampelopsin. The new analytical platform allowed identication of 16 compounds in commercial willow bark extracts, and identication of ampelopsin, taxifolin, 7-O-methyltaxifolin-3 -O-glucoside, and 7-Omethyltaxifolin in S. alba bark extract is reported for the rst time. The detection of the novel compound, ethyl 1-hydroxy-6-oxocyclohex-2-enecarboxylate, is also described. 2012 Elsevier B.V. All rights reserved.

Article history: Received 11 July 2012 Received in revised form 31 August 2012 Accepted 3 September 2012 Available online 10 September 2012 Keywords: Liquid chromatography Metabolomics Nuclear magnetic resonance Multivariate data analysis Willow bark Antioxidant activity

1. Introduction Extracts from willow bark have since ancient times been used for their analgesic and anti-inammatory properties, which have been attributed to the content of salicin and salicin derivatives, i.e., salicylates, as prodrugs of salicylic acid [1]. The efcacy of willow bark extracts has been demonstrated in several clinical trials [25], and such extracts could serve as valuable alternatives to aspirin or other nonsteroidal anti-inammatory drugs, with better tolerability and patients acceptance. This is in particular important for use in longterm treatment of conditions such as back pain or arthritis [6]. However, pharmacological studies indicate that the content of salicylates alone cannot explain the analgesic and anti-inammatory

Corresponding author at: Department of Drug Design and Pharmacology, Faculty of Health and Medical Sciences, University of Copenhagen, Universitetsparken 2, DK-2100 Copenhagen, Denmark. Tel.: +45 3533 6177; fax: +45 3533 6001. E-mail address: ds@sund.ku.dk (D. Staerk). 0021-9673/$ see front matter 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.chroma.2012.09.013

effect of willow bark extracts; and other constituents may contribute to the overall effect through different mechanisms [7,8]. Several studies have focused on phytochemical investigation of Salix species used for preparing the nal willow bark products, and, e.g., Salix daphnoides, Salix pentadra, Salix purpurea, Salix alba, and Salix fragilis have been investigated for their content of phenolic glycosides [911]. In addition to salicylates, avonoids and condensed tannins constitute the major groups of secondary metabolites in Salix species, and these compounds are believed to contribute to the analgesic and anti-inammatory effect of willow bark [12,13]. Commercial preparations of willow bark extract are produced from hydroalcoholic extracts of bark from different salicylate rich Salix species, and the preparations are standardized to contain at least 5% of total salicin measured after hydrolysis of salicin derivatives (according to the European Pharmacopoeia [14]). This allows the ratio between salicin, salicin derivatives and other constituents to be highly variable; a factor that might alter bioavailability as well as compromise efcacy and safety. Until now, only two studies with commercial willow bark preparations have been published. One

S. Agnolet et al. / J. Chromatogr. A 1262 (2012) 130137

131

study reported development of a LCMS/MS method for targeted identication of a series of reference compounds [15], and the other reported development of a HPLC-ngerprint analysis with emphasis on preprocessing methods and not on the chemical composition of the preparations [16]. Thus, there is a need for fast and efcient techniques that provide non-selective and comprehensive chemical, and preferably also biochemical/pharmacological, information of as many constituents as possible. Metabolomics, i.e., the systematic and comprehensive study of the unique chemical ngerprints generated by all cellular processes, is performed by applying multivariate data analysis to chemical ngerprints. Metabolomics based on highperformance liquid chromatography coupled with ultraviolet detection (HPLCUV) and mass spectrometry (LCMS) has proven to be an efcient method that provides comprehensive analysis of the global composition of complex extracts [17,18]. However, a major disadvantage is the non-linear response factor due to differences in molar absorptivity (HPLCUV) and ionization (LCMS); and the lack of detailed structural information from the UV-spectra limits structural identication of HPLC analytes to known compounds based on spiking with reference material. 1 H NMR based metabolomics has been used extensively for evaluation of chemical variation due to species differences, chemical diversity of cultivars, biotic and abiotic stress, etc. [1921]. Recently 1 H NMR metabolomics has also been applied for evaluation of the global composition of herbal remedies [22,23]; and the technique has proven superior to multivariate data analysis based on HPLC ngerprints [24,25]. NMR spectroscopy has the advantage of being a non-destructive and fast method. In addition to this, detailed structural information and absolute quantication can be obtained without the need of reference material; two characteristics that, in spite of the lower sensitivity, constitute a major advantage over other analytical techniques. However, plant extracts represent very complex mixtures, and signal overlap can decrease the ability to prole secondary metabolites. In these cases the use of twodimensional NMR experiments [26] or the hyphenation of NMR and chromatographic separation [25,27], vide infra, are necessary for the nal identication of compounds. HPLCSPENMR [28] has proven to be an efcient hyphenated HPLCNMR technique, and is based on automated adsorption (trapping) of analytes on SPE cartridges after post-column dilution of the HPLC eluate with water. Cartridges are subsequently dried with a stream of nitrogen gas to remove the non-deuterated HPLC eluate and eluted into the NMR ow cell or into small-volume NMR tubes (HPLCSPEtube transfer NMR, HPLCSPEttNMR) with fully deuterated organic solvents. HPLCSPENMR systems are commercially available, and have been widely used in natural products research [29,30]. Recently HPLCSPENMR also proved very useful for identication of metabolites in metabolomics research [25]. Metabolomics combined with HPLCSPENMR can potentially provide a comprehensive chemical analysis of all constituents in plant extracts. However, information about (bio)activity of the individual constituents is missing, and in this context further coupling of HPLCSPENMR with high-resolution screening, i.e., (bio)assays coupled with HPLC, are promising developments. HPLCSPENMR coupled with direct on-line monitoring of radical scavenging by post-column addition of a relatively stable colored radical like diphenylpicrylhydrazyl radical (DPPH ) or ABTS + constitute an example of further coupling with high-resolution (bio)activity screening [31], but dedicated equipment is needed for this purpose. In this study, HPLC separation has been coupled with the easy implementable microplate-based screening for radical scavenging activity, as described by Wiese and Staerk [32], for generation of ABTS + reduction proles for multivariate data analysis as well as for pinpointing individual metabolites with radical scavenging activity.

In this work, we present a new approach that combines information from multivariate data analysis of chemical (HPLC and 1 H NMR) and pharmacological (ABTS + reduction proles) ngerprints, for selecting preparations for subsequent targeted HPLCSPEttNMR analysis. 2. Experimental 2.1. Materials Six commercial samples containing willow bark extract were purchased from health food- and pharmacy stores in Denmark and Italy during December 2009. Two samples (preparations 1 and 2) were hydroalcoholic solutions of S. alba bark extract. Preparation 3 was a sample of S. alba cortex sold for the preparation of decoctions whereas preparation 4, obtained as a powder, was an extract from the bark of S. purpurea. Preparations 5 and 6 contained extract of S. alba bark, and were formulated as capsules. These samples were different batches from the same vendor. HPLC-grade methanol was purchased from Fischer Scientic (Slangerup, Denmark), water was puried by deionization and 0.22 m membrane ltration (Millipore). Acetonitrile-d3 (99.8% D) and methanol-d4 (HDO + D2 O < 0.03%) for NMR experiments were obtained from CortecNet (Voisins-Le-Bretonneux, France) and Eurisotop (Gif-surYvette, France), respectively. ABTS (2,2 -azino-bis(3-ethylbenzothiazoline-6-sulfonic acid) diammonium salt), sodium phosphate dibasic and sodium phosphate monobasic dihydrate were purchased from SigmaAldrich (Steinheim, Germany). 2.2. Sample preparation Preparations 1 and 2 were evaporated to dryness at reduced pressure at temperatures below 40 C. Preparation 3 was pulverized, and 74 g of material was extracted three times with 100 mL of 70% methanol. After ltration the combined extracts were evaporated to dryness at temperatures below 40 C. For preparations 46, 400 mg of powder was added to 10 mL of 70% methanol, left overnight, subsequently centrifuged 5 min at 13,500 rpm, and nally the supernatant was evaporated to dryness at temperatures below 40 C. Samples for 1 H NMR based metabolomics were prepared in methanol-d4 to a nal concentration of 50 mg/mL. Samples were prepared in triplicate, and were centrifuged at 13,500 rpm for 5 min before transferring 600 L to 5-mm NMR tubes. Samples for HPLC analysis were prepared in methanolwater 1:1 to a nal concentration of 10 mg/mL. After HPLC analysis, a number of samples for each preparation were subjected to alkaline hydrolysis. Thus, 200 L of 0.1 M NaOH was added directly to the HPLC vial and the mixture were heated on a water bath at 60 C for 1.5 h. The sample was neutralized with 20 L of 1 M HCl and subsequently analyzed by HPLC. The method has been adapted after [11]. Injection volumes were 20 L for samples containing herbal preparations as well as for salicin standard solutions. Five replicates were made for each sample for quantitative analysis of salicin and for chromatographic ngerprint analysis, whereas three replicates were used for the quantitative analysis of salicin after hydrolyzation. Samples for HPLCSPEttNMR were dissolved in methanolwater 1:1 to a nal concentration of 30 mg/mL and injection volumes of 30 L were used. 2.3. HPLC analysis HPLC separations were performed on an Agilent 1200 system consisting of a G1367C high performance autosampler, a G1311A quaternary pump with G1322A degasser, a G1316A thermostatted column compartment, a G1315C diode array detector, and

132

S. Agnolet et al. / J. Chromatogr. A 1262 (2012) 130137

a G1364C fraction collector. All operations were controlled by ChemStation revision B.03.02 software. The analyses were performed at 40 C on a 150 mm 4.6 mm i.d. Phenomenex C18 (2) with a ow rate of 0.8 mL/min using a Luna column (3 m, 100 A) mixture of watermethanol 95:5 + 0.1% formic acid (eluent A) and methanolwater 95:5 + 0.1% formic acid (eluent B) using the following linear gradient elution prole: 0 min, 0% B; 25 min, 100% B; 30 min, 100% B; 31 min, 0% B; 45 min, 0% B. Reference material of salicin was isolated from preparation 5 by dissolving the content of 1 capsule in 4 mL of methanolwater 1:1, centrifuging at 13,500 rpm for 3 min, and injecting 20 L of the supernatant. The separation was performed as described, and threshold-based collection from 30 consecutive separations yielded 18.1 mg of pure salicin as determined by HPLC and 1 H NMR spectroscopy. A stock solution of salicin was prepared by dissolving 10.0 mg of salicin in 2.0 mL of methanolwater 1:1. 2.4. Quantitative analysis of salicin The HPLC method described above was used for the quantication of salicin and validated for precision (intra-day and inter-day repeatability tests), accuracy, linearity, and sensitivity (limit of detection, LOD, and limit of quantitation, LOQ) [11]. Thus, intra-day and inter-day reproducibility tests were performed by quantitative analysis of three concentrations of salicin (0.5, 1.0 and 2.0 mg/mL) on three consecutive days. Accuracy was determined by the standard addition method for recovery studies. Thus, three samples containing extract of preparation 4 were spiked with 50, 100, and 150 L salicin standard solution, corresponding to 0.25, 0.5 and 0.75 mg of salicin. All analyses were performed in triplicate. Linearity was determined by quantitative analysis of dilutions of the salicin stock solution, giving nal concentrations of 3.0, 2.0, 1.0, 0.5, and 0.1 mg/mL. The HPLC calibration curve was constructed plotting the peak areas (A) against the concentration of each standard solution (c). Limit of detection (LOD), dened as concentration giving peak signal-to-noise ratio of 3, and limit of quantication (LOQ), dened as concentration giving peak signal-to-noise ratio of 10, were found by linear regression of HPLC peak areas. The analyses were performed with 20 L injections of sample-concentrations of 0.1, 0.05, 0.01 and 0.005 mg/mL obtained by dilution of the above 0.1 mg/mL salicin solution with methanolwater 1:1. The LOD of salicin was 0.005 mg/mL and the LOQ was 0.016 mg/mL. All validated parameters are reported in Table 1.

2.5. High-resolution radical scavenging screening Radical scavenging capacity was detected following the reduction of the radical cation ABTS + spectrophotometrically as described by Pellegrini et al. [33]. ABTS + was prepared by incubating ABTS (2.5 mM) with 0.875 mM potassium persulfate in millipore water and allowing the mixture to stand in darkness for 1216 h before use. 0.1 M sodium phosphate buffer (pH 7.4) was used to neutralize acidied HPLC eluents. ABTS + solution was diluted ve times with sodium phosphate buffer prior to use. Coupling of HPLCphoto-diode array (HPLC-PDA) with the radical scavenging activity assay was performed by fractionating the HPLC eluate into 96 well plates in 100 L steps from 0 to 24 min. 200 L ABTS + solution were then added to each fraction. Absorbance was measured at 620 nm every minute for 10 min using Thermo Scientic Multiskan FC Microplate Photometer. The method has been described by Wiese and Staerk [32]. HPLC conditions were as described above. The experiments were performed for all the samples in triplicate. Samples were prepared as described above to a nal concentration of 10 mg/mL. The resulting ABTS + reduction proles were baseline corrected with a polynomial function in MATLAB, version 7.6.0, software (Matworks, Natick, MA). Data were reduced to the time range from 5 to 20 min and used for principal component analysis after mean-centering using SIMCA-P version 12 software (Umetrics, Ume, Sweden). 2.6. HPLC ngerprint analysis Chromatographic PDA matrices were sampled at a rate of 0.25 s1 , and the chromatogram at 270 nm was extracted from all datasets, and imported into MATLAB. Data were reduced to the chromatographic time range from 5 to 20 min and baseline corrected with a polynomial function. Finally, all UV traces were collected in a unique matrix (30 2250 data points) and aligned using a correlation optimized warping algorithm [34]. The dataset was imported into SIMCA-P version 12 software (Umetrics, Ume, Sweden) and PCA was performed on mean-centered data without any normalization. 2.7. HPLCPDASPE-ttNMR analysis HPLCPDASPEttNMR experiments were performed with the Agilent 1200 Series HPLC system described in Section 2.3, a Knauer K100 Wellchrom pump for post-column dilution, a Spark Holland Prospekt 2 solid-phase extraction device, and a Gilson 215 Liquid Handler equipped with a 1-mm needle for automatic lling of 2.5-mm Bruker Match tubes. All operations were controlled using Bruker HyStar version 3.2 and Bruker PrepGilson ST version 1.2 softwares. For preparations 1, 3, and 5 a 30 mg/mL solution was prepared and 50 L injection volumes were used for chromatographic separations using the above-mentioned gradient elution prole. Eight repeated separations were performed with preparations 3 and 5, whereas 15 repeated separations were performed with preparation 1. Analytes were trapped on general-purpose poly(divinylbenzene)-based resin (GP-phase) SPE cartridges (from Spark Holland, 10 mm 2 mm i.d.; post-column dilution with water, 2 mL/min) using both threshold-based absorbance at 270 nm and time-based collection for selection of analytes. The adsorbed material was eluted to 2.5-mm NMR tubes within 140 L of acetonitrile-d3 and 1 H NMR spectra were acquired in automation on a Bruker Avance spectrometer (1 H frequency 300.13 MHz) equipped with a 5-mm broadband observe (BBO) probe and a BACS sample changer. Spectra were acquired at ambient temperatures, and a total of 256 transients were recorded as 64k data points with a spectral width of 20 ppm using 30 pulses, 5.5 s acquisition time, and an inter-pulse delay of 7 s. 1 H NMR spectra of selected

Table 1 Validation parameters for the quantitative analysis of salicin: precision (intra-day and inter-day repeatability), accuracy, linearity and sensitivity (limit of detection, LOD, and limit of quantication, LOQ). Precision Salicin standard 0.5 mg/mL Salicin standard 1.0 mg/mL Salicin standard 2.0 mg/mL Accuracy 0.25 mg (50 L standard solution) 0.50 mg (100 L standard solution) 0.75 mg (150 L standard solution) Intra-day precisiona 0.507 0.000 1.004 0.000 2.037 0.002 Inter-day precisiona 0.508 0.003 1.003 0.002 2.042 0.010 Mean recoveriesa 103.08 1.07 102.67 3.93 101.61 0.92

Linearity A = 3.83c + 0.124 r2 = 0.9990 (A, peak areas; c, concentration of standard solutions; r2 , squared correlation coefcient) Sensitivity LOD = 0.005 mg/mL, LOQ = 0.016 mg/mL
a

Mean SD, n = 3.

S. Agnolet et al. / J. Chromatogr. A 1262 (2012) 130137

133

samples were acquired on a Bruker Avance III NMR spectrometer (1 H frequency 799.96 MHz) equipped with a 5-mm TCI cryoprobe operated at 298.15 K. For each sample, 64 transients were recorded as 64k data points with a spectral width of 20 ppm using 30 pulses, 2.0 s acquisition time, and an inter-pulse delay of 3 s. All spectra were manually phase corrected, baseline corrected, and calibrated to residual acetonitrile signal ( 1.94) in TopSpin version 1.3 software. 2.8. NMR-based metabolomics analysis All 1 H NMR spectra were acquired immediately after preparation of the samples using the 300 MHz Bruker Avance spectrometer described in Section 2.7. Spectra were acquired in automation using the same acquisition parameters as described above, except that a total of 128 transients were acquired. The free induction decays were Fourier transformed to 128k data points using line broadening of 0.1 Hz. All spectra were manually phase corrected, baseline corrected, and calibrated to residual methanol signal ( 3.31) in TopSpin version 1.3. The 1 H NMR spectra were reduced to 213 sequentially integrated spectral regions (buckets) of 0.04 ppm width, from 0.5 to 9 using AMIX version 3.6.8 software (Bruker BioSpin, Germany). The NMR data reduction le from AMIX was imported into a Microsoft Excel spreadsheet for addition of labels. Principal component analysis was performed with SIMCA-P after mean-centering of data and exclusion of regions corresponding to the residual methanol signal ( 3.323.36), water ( 4.845.0), ethanol ( 1.161.20 and 3.63.64), methylene envelope of fatty acids ( 1.241.36), and acetic acid ( 1.881.96). 3. Results and discussion 3.1. Quantication of salicin and salicin derivatives In order to assess the ratio between free salicin and salicin derivatives, a HPLC method for quantication of salicin was developed and validated using a salicin standard solution (Table 1). Thus, quantitative analysis of salicin was performed before and after alkaline hydrolysis of the extract, and the difference between the obtained values was calculated as a measure of the amount of salicin derivatives in the samples. The results are reported in Table 2, and this shows that there is a very large variation in the content of free salicin and salicin derivatives. The content of free salicin ranges from 1.68% to 25.59%, whereas the content of salicin derivatives ranges from 0.3% to 4%. Interestingly, preparations 5 and 6 that contain a high amount of free salicin, are totally depleted of salicin derivatives. The absence of salicin derivatives could indicate degradation of salicin derivatives to salicin during the manufacturing process or during storage; or that the preparations were enriched with salicin during their production. Furthermore only preparations 1, 4, 5 and 6 contain the minimum amount of 5% of total salicin as dened in the European Pharmacopoeia [14].
Table 2 Information about the commercial products included in the study. Sample Prep1 Prep2 Prep3 Prep4 Prep5 Prep6 % of free salicin (mean SD, n = 5) 6.81 1.68 2.75 4.73 25.59 24.07 0.20 0.11 0.05 0.43 0.69 1.04 % of salicylates expressed as salicin (mean SD, n = 3) 4.03 1.04 0.73 0.31 0.21 0.02 0.02 0.17 Amount of extract (mg/day)a,b (mean SD, n = 3)b 55.9 94.9 8.0 145.1 1033.4 1025.3 1.0 1.2 0.2b 3.8 7.3 12.3 Total salicin intake (mg/day)a (mean SD, n = 3) 6.1 2.6 0.3 7.3 264.4 246.8 0.2 0.1 0.01 0.7 7.4 11.1

Fig. 1. Chromatographic (A) and 1 H NMR spectroscopic (B) ngerprints used for principal component analysis. Only the rst series of the replicates for preparations 16 are shown.

Based on the recommended indications found in the supplied leaets, the amount of extract and total salicin were calculated for each preparation (Table 2). For preparations 14, the amount of total salicin intake per day varies from 0.3 to 7.3 mg, which is considerably lower than the 247264 mg obtained for preparations 5 and 6. Daily doses of 60120 mg or 120240 mg of total salicin have been recommended from the Commission E of the German Federal Agency and The European Scientic Cooperative on Phytotherapy, respectively [35,36]. Likewise, daily doses of 120 mg or 240 mg of total salicin have been used in clinical trials [2,4,6]. Thus, our results show that only the preparations formulated as capsules, e.g., preparations 5 and 6, would contain a dose of total salicin necessary for the analgesic and anti-inammatory activity. 3.2. HPLC- and 1 H NMR-based metabolomics Metabolomics analysis was performed with HPLC traces acquired at 270 nm as well as with 1 H NMR spectra. Overlaid spectra of these chemical ngerprints used for multivariate data analysis are shown in Fig. 1. Thus, HPLC traces in the time range from 5 to 20 min were used for multivariate data analysis after baseline correction and correlation optimized warping alignment performed according to Skov et al. [34]. Principal component analysis (PCA) was subsequently carried out on the centered dataset without any pretreatment. A model with six principal components (PC) explaining 99.8% of the variance was obtained and the resulting score and loading plot for PC1 vs PC2 are shown in Fig. 2. The samples are separated in three distinct groups. Thus, the two preparations

a Total amount of extract and total salicin intake per day were calculated based on the indication on the packages. Indications were 40 drops 3 times per day for preparation 1 and preparation 2, one spoon per day (400 mg) for preparation 4, and 3 capsules daily (each 400 mg) for preparations 5 and 6. b Indication for preparation 3 would be a spoon of bark material (corresponding to 2 g) to prepare one cup of decoction per day.

134

S. Agnolet et al. / J. Chromatogr. A 1262 (2012) 130137

Fig. 2. Score (A) and loading (B) plots of PC1 vs PC2 for the six-component model of the chromatographic ngerprints of preparations 16 [ 270 nm, tR 520 min, baseline corrected and aligned, data centered].

formulated as capsules (preparations 5 and 6), are separated along PC1 in the score plot (Fig. 2A), and the loadings (Fig. 2B) reveals this to be due to their higher content of salicin. Preparation 3 is instead separated in the left upper part of the score plot along PC2, which in the loadings plot is seen to be due to the presence of 6 intense characteristic peaks (tR 7.6, 11.3, 13.9, 16.5, 17.3, and 18.3 min, respectively) that were not detected in any of the other samples. The fact that the freshly prepared extract of S. alba bark, i.e., preparation 3, is different from the commercially available products is not unexpected, because the industrial manufacturing process of herbal remedies often leads to removal of some compounds. The last distinct group consists of preparations 1, 2 and 4, which are grouped together in the lower left hand side of the plot. They display a common pattern formed by salicin (tR 8.9 min) and three major peaks (tR 9.5, 10.6, and 11.0 min) with different relative amounts. Interestingly, preparation 4 should be derived from the bark of S. purpurea, but show a metabolite prole very similar to preparations 1 and 2 that are hydroalcoholic extracts of S. alba bark. 1 H NMR metabolomics analysis was performed using three replicates for each sample. After phase correction, baseline correction and calibration to methanol residual signal ( 3.31), the spectra in the range from 0.5 to 9.0 ppm were sequentially integrated using 213 buckets of 0.04 ppm width. Other data integration protocols, i.e., 0.02 and 0.01 ppm bucketing as well as full resolution data, were evaluated, but these methods showed the same overall PCA plots ensuring that no information were lost during the unsupervised global evaluation of the extracts. The regions corresponding to the methanol residual signals ( 3.31, p, CD2 HOD and 3.35, s, CH3 OH), ethanol residual signals ( 3.61, 2H, q, J = 10.9 Hz; 1.18, 3H, t, J = 10.9 Hz), water ( 4.845.0), methylene envelope of fatty acids ( 1.29, br s), and a very intense singlet at 1.93 ppm that was assigned to acetic acid (present in high amount in preparations 5 and 6) were removed. All the above-mentioned signals were assigned based on literature data [37]. Principal component analysis was then performed on the centered dataset without any pretreatment. This resulted in a model with seven principal components accounting for 99.5% of the variance. The corresponding score and loading plots for PC1 vs PC2 are shown in Fig. 3A and B, respectively. The score plot obtained from 1 H NMR metabolomics analysis displays features similar to the score plot obtained from HPLC traces. The two preparations formulated as capsules (5 and 6) are separated from the remaining preparations along PC1 for their higher amount of salicin. Other major signals that contribute to the differentiation of these samples, have, based on the loading plot, been assigned to formic acid (singlet at 8.5) and sucrose (anomeric signals at 5.40, 1H, d, J = 3.8 Hz and 4.10, 1H, d, J = 8.5 Hz). Preparations 14, grouping together along PC1 in the left part of the plot,

are separated along PC2. The samples are characterized by an overall similar NMR prole; but with different relative intensities of the signals. Thus, loadings reveal that preparation 4 is separated along PC2 due to the general lower intensities of a large number of signals that are common to preparations 14. Unfortunately, assignment of these major common signals is complicated by signal overlap. However, loadings corresponding to two signals at 5.92 (d, J = 2.0 Hz) and 5.84 (d, J = 2.0 Hz) are consistent with H-6 and H-8 of catechin; although an unambiguous assignment is not possible based on the 1 H NMR spectra alone. Thus, preparation 4 contains less of the compound, tentatively assigned to catechin, compared to preparations 13. Further loadings responsible for separation of preparation 3 along PC2 were observed. Thus, besides the signals at 5.92 (d, J = 2.0 Hz) and 5.84 (d, J = 2.0 Hz), these loadings correspond to 6.056.06 (d, J = 2.0 Hz) and 6.116.12 (d, J = 2.0 Hz), together with a singlet at 6.54 ppm. All these signals are consistent with the presence of at least two avonoids in addition to catechin. 3.3. HPLCPDASPEttNMR guided by high-resolution ABTS + reduction proles Structural information based on the loadings in Fig. 2A, i.e., tR values, are limited due to the lack of reference material. Likewise, unambiguous identication of compounds based on loadings in Fig. 3A, i.e., 1 H NMR spectral data, are hampered by overlap of closely related analogs. In addition, neither of these methods provide information about the potential bioactivity of the metabolites. For these reasons, a new platform combining high-resolution ABTS + reduction proles and hyphenated HPLCSPEttNMR was used for analysis of preparations 16. The microplate-based highresolution radical scavenging activity screening was performed using a decolorization assay based on reduction of the radical cation ABTS + for assessment of radical scavenging activity in each microplate well after time-sliced collection of HPLC eluate. The method has been described by Wiese and Staerk [32], and provided (i) high-resolution ABTS + reduction proles for metabolomics and (ii) information for guiding HPLC-PDASPEttNMR analysis towards bioactive constituents. The high-resolution ABTS + reduction proles overlaid with the UV trace at 270 nm from HPLC separation of preparation 3 is shown in Fig. 4; and similar proles for all preparations are shown in Supplementary Fig. 1. The high-resolution ABTS + reduction proles were subjected to principal component analysis, which resulted in a three-component model explaining 88% of the variance. The resulting score and loading plot for PC1 vs PC2 are displayed in Fig. 5A and B, respectively. Preparations 1 and 3 are separated from the rest of the samples along PC1, and the loadings reveals this to be due to a

S. Agnolet et al. / J. Chromatogr. A 1262 (2012) 130137

135

Fig. 3. Score (A) and loading (B) plots of PC1 vs PC2 for the seven-components PCA model of the region 0.59.0 of the 1 H NMR spectral data [300 MHz, 0.04 ppm bucket, data mean centered]. In the loading plot the identied buckets are color coded as follow: salicin (gold), catechin (green), ampelopsin (brown), benzoic acid and derivatives (red), sucrose (violet) and formic acid (orange). (For interpretation of the references to color in this gure legend, the reader is referred to the web version of the article.)

Fig. 4. HPLC chromatogram of preparation 3 (top) overlaid the high-resolution ABTS + reduction prole (bottom).

common radical scavenger (tR 9.5 min), which is also present in smaller amount in preparations 2 and 4, as well as a major radical scavenger (tR 11.3 min) observed in preparation 3. On the other hand, no radical scavenging activity is detected in preparations 5 and 6 grouped together in the lower left hand side of the plot. Furthermore, preparation 3 is clearly separated along PC2 in the lower right side of the score plot, and loadings reveals this to be due to the 6 distinct radical scavengers (tR 7.6, 11.3, 13.9, 16.5, 17.3, and 18.3 min, respectively) that are not present in any of the other samples. Principal component analysis performed with HPLC-, 1 H NMR-, and ABTS + reduction proles (Figs. 2, 3 and 5, respectively)

showed in each cases three distinct clusters. In order to choose one preparation from each cluster from all three score plots, preparations 1, 3 and 5 were selected for HPLCSPEttNMR analysis. These three samples contain all metabolites present in all preparations, and the HPLCSPEttNMR analysis resulted in identication of 16 compounds. The identity and distribution of the isolated compounds are listed in Table 3 and spectral data are reported in Supplementary Table 1. For preparation 1, six peaks were trapped on GP-phase cartridges, and subsequent NMR analysis led to identication of salicin (tR 8.8 min), (+)-catechin (tR 9.5 min), syringin (tR 10.6 min), triandrin (tR 11.0 min), ethyl 1-hydroxy-6oxocyclohex-2-enecarboxylate (tR 12.5 min), and benzoic acid (tR 16.1 min). The identication of these compounds in preparation 1 is marked with lled circles in Table 3, and other occurrences of the compounds in other preparations, based on comparison of HPLC-PDA data, is marked with open circles. Thus salicin and syringin were identied in all 6 preparations. Ethyl 1-hydroxy-6oxocyclohex-2-enecarboxylate was only found in the hydroalcholic preparations, i.e., preparations 1 and 2. 1-Hydroxy-6-oxocyclohex2-enecarboxylic acid is commonly encountered in Salix as subunit of compounds such as tremulacin or salicortin, and as a methyl ester. Recently, the methyl ester was also isolated from Dovyalis abyssinica [38]. The compound isolated here is thus likely to be the product of the trans-esterication of the previously isolated methyl ester kept in an ethanolic solution. Benzoic acid was also present in all preparations except preparation 3, and even though it is a well-known plant constituent, it could originate from

Fig. 5. Score (A) and loading (B) plots of PC1 vs PC2 for the six-component model using high-resolution ABTS + reduction proles [tR 520 min, baseline corrected and aligned, data centered].

136

S. Agnolet et al. / J. Chromatogr. A 1262 (2012) 130137

Table 3 List of compounds identied with HPLCSPEttNMR experiments performed on three samples chosen based on the HPLC ngerprint analysis.a Compound tR (min) Preparation 1 Salicin (+)-Catechin p-Hydroxybenzoic acid Syringin Triandrin Ampelopsin Ethyl 1-hydroxy-6-oxocyclohex-2-enecarboxylate p-Hydroxycinnamic acid (=p-coumaric acid) Taxifolin Benzoic acid p-Methoxybenzoic acid 7-O-methyltaxifolin-3 -O-glucoside 7-O-methyltaxifolin trans-p-Methoxycinnamic acid cis-p-Methoxycinnamic acid Cinnamic acid 8.8 9.5 10.1 10.6 11.0 11.3 12.5 13.8 13.9 16.1 16.1 17.3 18.3 18.3 18.3 20.0 2 3 4 5 6

a The symbol is used for indicating where the compounds have been identied by HPLCSPEttNMR and for indicating the other samples in which the compounds were identied by comparison with tR and UV spectrum from HPLC-PDA experiments.

degradation of tremulacin, tremuloidin or other compound esteried with a benzoyl moiety. Preparation 3 showed characteristic peaks with radical scavenging activity (Fig. 4), and HPLC-PDASPE-ttNMR analysis led to the identication of the avonoids ampelopsin (tR 11.3 min), taxifolin (tR 13.9 min), 7-O-methyltaxifolin-3 -O-glucoside (tR 17.3 min), and 7-O-methyltaxifolin (tR 18.3 min). Ampelopsin represents, together with the other identied avonols, the major constituents of S. alba cortex included in this study. These compounds are reported for the rst time in S. alba cortex, whereas the avanones eriodictyol and naringenin has previously been isolated from S. alba bark extract [39]. HPLCSPEttNMR analysis showed that preparations 5 and 6 contained a large amount of salicin and to a smaller extent syringin, in common with all other preparations. However, the analysis did not show the presence of salicin derivatives. Instead, the HPLCSPEttNMR analysis showed the presence of a series of small phenolic acids.

of our knowledge the rst comprehensive investigation of willow bark commercial products, which is emphasized by the identication of ampelopsin, taxifolin, 7-O-methyltaxifolin-3 -O-glucoside, and 7-O-methyltaxifolin in S. alba bark extract for the rst time and identication of (+)-catechin as a major constituent responsible for the radical scavenging activity in willow bark preparations. Although this proof-of-concept study was performed with a limited number of samples, it highlights the need for better quality control of herbal products, and the presented technology platform represents a complete set of modern analytical methods that can be applied for future comprehensive investigations of herbal remedies. Acknowledgements The Drug Research Academy is thanked for a PhD scholarship to Sara Agnolet. The 800 MHz NMR experiments were obtained using a Bruker Avance III spectrometer at the Danish Instrument Center for NMR Spectroscopy of Biological Macromolecules. HPLCSPEtube transfer equipment was obtained via a grant from The Carlsberg Foundation. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.chroma. 2012.09.013. References
[1] J.G. Mahdi, A.J. Mahdi, A.J. Mahdi, I.D. Bowen, Cell Prolif. 39 (2006) 147. [2] S. Chrubasik, E. Eisenberg, E. Balan, T. Weinberger, R. Luzzati, C. Conradt, Am. J. Med. 109 (2000) 9. [3] S. Chrubasik, O. Knzel, A. Black, C. Conradt, F. Kerschbaumer, Phytomedicine 8 (2001) 241. [4] S. Chrubasik, O. Knzel, A. Model, C. Conradt, A. Black, Rheumatology 40 (2001) 1388. [5] S.Y. Mills, R.K. Jacoby, M. Chackseld, M. Willoughby, Br. J. Rheumatol. 35 (1996) 874. [6] B. Schmid, R. Ldtke, H.K. Selbmann, I. Ktter, B. Tschirdewahn, W. Schaffner, L. Heide, Phytother. Res. 15 (2001) 344. [7] B.L. Fiebich, S. Chrubasik, Phytomedicine 11 (2004) 135. [8] B. Schmid, I. Ktter, L. Heide, Eur. J. Clin. Pharmacol. 57 (2001) 387. [9] N. Frster, C. Ulrichs, M. Zander, R. Ktzel, I. Mewis, Gesunde Panzen 61 (2009) 129. [10] B. Meier, D. Lehmann, O. Sticher, A. Bettschart, Pharm. Acta Helv. 60 (1985) 269. [11] L. Poblocka-Olech, A.-M. van Nederkassel, Y. Vander Heyden, M. Krauze Baranowska, D. Gld, T. Baczek, J. Sep. Sci. 30 (2007) 2958.

4. Conclusions In this study a targeted quantitative analysis of the content of salicin in six commercial samples of willow bark extracts was performed, but contrary to the standard quality control as dened by the European Pharmacopoeia, the analysis was repeated both before and after alkaline hydrolysis providing a measure of the relative amount of free and total salicin. Furthermore, with the aim of a more comprehensive evaluation of the composition of the complex extracts, principal component analysis of two chemical ngerprints, i.e., HPLC- and 1 H NMR data, and a pharmacological prole, i.e., high-resolution ABTS + reduction prole was carried out. Subsequent targeted structural identication of 16 analytes was effectively performed with HPLCSPEttNMR; demonstrating the usefulness of the automated solid-phase extraction unit interfacing HPLC and NMR. The score plots from PCA of HPLC- and 1 H NMR data showed the same overall clustering of the preparations (preparations 1, 2 and 4; preparation 3; preparations 5 and 6), whereas the score plot based on ABTS + reduction proles separated preparation 1 and preparation 3 from the remaining preparations. This demonstrates the value of using chemical as well as pharmacological ngerprints for multivariate data analysis when studying complex herbal extracts, and the presented techniques thus represents a new promising instrumental platform for future investigations of other herbal products. This study is to the best

S. Agnolet et al. / J. Chromatogr. A 1262 (2012) 130137 [12] A. Nahrstedt, M. Schmidt, R. Jggi, J. Metz, M.T. Khayyal, Wien. Med. Wochenschr. 157 (2007) 348. [13] S. Enayat, S. Banerjee, Food Chem. 116 (2009) 23. [14] E. Pharmacopoeia, European Pharmacopoeia, 6th edition, 2008. [15] B. Kammerer, R. Kahlich, C. Biegert, C.H. Gleiter, L. Heide, Phytochem. Anal. 16 (2005) 470. [16] M.M.W.B. Hendriks, L. Cruz-Juarez, D. De Bont, R.D. Hall, Anal. Chim. Acta 545 (2005) 53. [17] P. Chen, M. Ozcan, J. Harnly, Anal. Bioanal. Chem. 389 (2007) 251. [18] L.-W. Yang, D.-H. Wu, X. Tang, W. Peng, X.-R. Wang, Y. Ma, W.-W. Su, J. Chromatogr. A 1070 (2005) 35. [19] M. Frdrich, Y.H. Choi, L. Angenot, G. Harnischfeger, A.W.M. Lefeber, R. Verpoorte, Phytochemistry 65 (2004) 1993. [20] Y.H. Choi, H.K. Kim, A. Hazekamp, C. Erkelens, A.W.M. Lefeber, R. Verpoorte, J. Nat. Prod. 67 (2004) 953. [21] O. Hendrawati, Q. Yao, H.K. Kim, H.J.M. Linthorst, C. Erkelens, A.W.M. Lefeber, Y.H. Choi, R. Verpoorte, Plant Sci. 170 (2006) 1118. [22] B. Rasmussen, O. Cloarec, H. Tang, D. Staerk, J.W. Jaroszewski, Planta Med. 72 (2006) 556. [23] J. Qu, Q. Liang, G. Luo, Y. Wang, Anal. Chem. 76 (2004) 2239. [24] N.J.C. Bailey, J. Sampson, P.J. Hylands, J.K. Nicholson, E. Holmes, Planta Med. 68 (2002) 734. [25] S. Agnolet, J.W. Jaroszewski, R. Verpoorte, D. Staerk, Metabolomics 6 (2010) 292. [26] S.Y. Yang, H.K. Kim, A.W.M. Lefeber, C. Erkelens, N. Angelova, Y.H. Choi, R. Verpoorte, Planta Med. 72 (2006) 364.

137

[27] Y.-Z. Liang, P. Xie, K. Chan, J. Chromatogr. B 812 (2004) 53. [28] D. Staerk, M. Lambert, J.W. Jaroszewski, in: O. Kayser, W.J. Quax (Eds.), Medicinal Plant Biotechnology: From Basic Research to Industrial Applications, vol. 1, Wiley-VCH, Weinheim, Germany, 2006, p. 29. [29] M. Lambert, D. Staerk, S.H. Hansen, J.W. Jaroszewski, Magn. Reson. Chem. 43 (2005) 771. [30] D. Staerk, J.R. Kesting, M. Sairaanpour, M. Witt, J. Asili, S.A. Emami, J.W. Jaroszewski, Phytochemistry 70 (2009) 1055. [31] T.A. van Beek, K.K.R. Tetala, I.I. Koleva, A. Dapkevicius, V. Exarchou, S.M.F. Jeurissen, F.W. Claassen, E.J.C. van der Klift, Phytochem. Rev. 8 (2009) 387. [32] S. Wiese, J. Nielsen, D. Staerk, Coupling HPLC-SPE-NMR with a MicroplateBased High-Resolution Antioxidant Assay for Efcient Analysis of Antioxidants in Food Validation and Proof-of-Concept Study with Caper Buds. Submitted to Food Chemistry. [33] R. Re, N. Pellegrini, A. Proteggente, A. Pannala, M. Yang, C. Rice-Evans, Free Radic. Biol. Med. 26 (1999) 1231. [34] T. Skov, F. van den Berg, G. Tomasi, R. Bro, J. Chemometr. 20 (2006) 484. [35] M. Blumenthal, in: Complete German Commission E Monographs (Ed.), Therapeutic Guide to Herbal Medicines, American Botanical Council/Integrative Medicine Communications, Austin/Boston, 1998. [36] ESCOP, Salicis cortex Willow Bark, in: ESCOP Monographs, Fascicule 4, European Scientic Cooperative of Phytotherapy, Exeter, UK, 1997. [37] H.E. Gottlieb, V. Kotlyar, A. Nudelman, J. Org. Chem. 62 (1997) 7512. [38] B. Rasmussen, A.J. Nkurunziza, M. Witt, H.A. Oketch-Rabah, J.W. Jaroszewski, D. Staerk, J. Nat. Prod. 69 (2006) 1300. [39] Q. Du, G. Jerz, P. Winterhalter, J. Liq. Chromatogr. Relat. Technol. 27 (2004) 3257.