A comprehensive strategy using chromatographic profiles combined with chemometric methods: Application to quality control of Polygonum cuspidatum Sieb. et Zucc

Accepted Manuscript Title: A comprehensive strategy using chromatographic profiles combined with chemometric methods: application to quality control of Polygonum cuspidatum Sieb. et Zucc Author: Fangyuan Gao Zihua Xu Weizhong Wang Guocai Lu Yvan Vander Heyden Tingting Zhou Guorong Fan PII: DOI: Reference:

S0021-9673(16)31127-X http://dx.doi.org/doi:10.1016/j.chroma.2016.08.050 CHROMA 357850

To appear in:

Journal of Chromatography A

Received date: Revised date: Accepted date:

19-5-2016 15-8-2016 22-8-2016

Please cite this article as: Fangyuan Gao, Zihua Xu, Weizhong Wang, Guocai Lu, Yvan Vander Heyden, Tingting Zhou, Guorong Fan, A comprehensive strategy using chromatographic profiles combined with chemometric methods: application to quality control of Polygonum cuspidatum Sieb.et Zucc, Journal of Chromatography A http://dx.doi.org/10.1016/j.chroma.2016.08.050 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

A comprehensive strategy using chromatographic profiles combined with chemometric methods: application to quality control of Polygonum cuspidatum Sieb. et Zucc

Fangyuan Gao

a, b,1

, Zihua Xu

a, 1

, Weizhong Wang a, Guocai Lu b, Yvan Vander

Heyden c, Tingting Zhou a, *, Guorong Fan a, d, e, **

a

Shanghai Key Laboratory for Pharmaceutical Metabolite Research, School of Pharmacy,

Second Military Medical University, No. 325 Guohe Road, Shanghai 200433, China. E-mail address: [email protected] (F. Y. Gao), [email protected] (Z. H. Xu), [email protected] (T. T. Zhou), [email protected] (G. R. Fan) b

Department of Health Toxicology, Faculty of Tropical Medicine and Public Health, Second

Military Medical University, No. 800 Xiangyin Road, Shanghai 200433, China c

Department of Analytical Chemistry and Pharmaceutical Technology, Center for

Pharmaceutical Research, Vrije Universiteit Brussel - VUB, Laarbeeklaan 103, B-1090 Brussels, Belgium d

Department of Clinical Pharmacy, Shanghai General Hospital, School of Medicine,

Shanghai Jiaotong University, No. 100 Haining Road, Shanghai 200025, China e

Tongji University School of Medicine, No. 1239 Siping Road, Shanghai 200120, China

*,**

Corresponding author: School of Pharmacy, Second Military Medical University, No. 325

Guohe Road, Shanghai 200433, China. Department of Clinical Pharmacy, Shanghai General Hospital, School of Medicine, Shanghai Jiaotong University, No. 100 Haining Road, Shanghai 200025, China. Tel./Fax: +86 021 81871266.

E-mail address: [email protected] (T. T. Zhou), [email protected] (G. R. Fan) 1

These authors contributed equally to this work.

1

Highlights 

A comprehensive strategy was established for quality control of herbs.



The comparison between three chemometric methods was studied for the first time.



Fifteen markers were selected for quality control according to chemometric results.



Variances resulted from different calculation theories of the chemometric methods.



An easily calculated parameter was adopted to reflect quality fluctuations.

Abstract For the strict quality control of herbs, a comprehensive strategy based on chromatographic profiles and chemometric methods which could reliably select quantitative indices, robustly quantitate multi-markers and systematically compare different chemometric methods was proposed and successfully applied to the quality analysis of P. cuspidatum. Based on the construction of chromatographic profiles by an efficient accelerated solvent extraction (ASE) and reliable high-performance liquid chromatography-ultraviolet (HPLC-UV) methods, different chemometric methods were employed, namely similarity analyses (SA), hierarchical clustering analysis (HCA) and linear discriminant analysis (LDA). The differences in classification of herb samples were studied for the first time. To reasonably determine the quality of 2

herbs and evaluate different chemometric methods, a comprehensive strategy containing three key steps was performed including selection of quantitative indices, development of a reliable quantification method and adoption of an easily calculated and visible parameter. The quantitative method which was acceptable with good linearity with correlation coefficients > 0.9995 and satisfactory repeatability (RSD < 1.5%), precision (RSD < 2.84%), reproducibility (RSD < 2.88%), stability (RSD < 2.85%) and recoveries (91.5% - 105.6%, RSD < 2.83%) was applied to quality evaluation of fourteen batches of the P. cuspidatum samples through simultaneous quantitative determination of fifteen marker compounds. The limits of quantitation of fifteen compounds ranged from 1-60 μg/ml. From the results of the quality evaluation, it was found that the different calculation theories of the chemometric methods resulted in the variation of classifiers of samples: SA classified samples through the mean values and HCA & LDA classified similar objects.

Keywords: Quality control, Chromatographic profiles, Quantitative analysis, Chemometric methods, Polygonum cuspidatum

1. Introduction

Quality of herbal medicines is closely related to the multiple bioactive chemical ingredients, which is critical to the therapeutic or preventable effects of herbs [1]. The methods for quality assessment and control of herbal medicines are various. Therefore, 3

the selection and application of appropriate quality analysis methods is the key to ensure the quality of herbs.

Chromatographic fingerprint, as commonly used strategy for quality control of herbs, is able to provide global and basic information of the distribution of multi-chemical constituents in a complex system and has great potential for further identification, authentication, characterization and classification of herbal medicines [2]. However, at present studies on the quality control of herbs are insufficient in chromatographic profiles and limited on the selection of qualitative and quantitative markers [3-5]. Some chemical markers in chromatographic profiles are usually selected within the confine of few compounds with high contents or known structures, which could not comprehensively reflect the synergistic action of multi-constituents in herbs and possess less statistical and practical values on the establishment of quality standard. To overcome the limits, chromatographic fingerprint combined with some chemometric methods was adopted to select characteristic chemical components (chemical markers), to distinguish herbs from different origins, and then to comprehensively evaluate the quality of medicinal herbs [6-10].

Initially, similarity analysis (SA) was performed to accurately calculate the correlation coefficient of original data for classification of similarity and dissimilarity of samples and recognize the common peaks of chromatographic profiles as quantitative indices [11]. Moreover, hierarchical cluster analysis (HCA) and linear discriminant analysis (LDA) were combined to find the underlying clustering 4

tendency of objects, group the different sources of samples into different clusters through converting the entire chromatograms into statistical structures and then generate the linear discriminant functions of each group membership based on the observed characteristics of each cluster [12]. According to the results of LDA, the value variables (peaks) were selected as quantitative calculation references (discriminate peaks).

The chromatographic profiles combined with the chemometric methods can provide quantitative calculation markers (common peaks in SA, discriminate peaks in HCA & LDA) and allow classification of the sources of herbs. However, according to the best of our knowledge, differences between SA and HCA & LDA have been observed in some reports [13, 14], and the significances of variances had never been concerned and investigated, although these methods are widely employed in the quality control of herbal medicines. Therefore, an authoritative analysis method is necessary to compare the differences between SA and HCA & LDA. In our study, practical quantitative determination of characteristic chemical markers coupled with an easily calculated and visible parameter are adopted to reflect quality fluctuations of herbs from different origins and further compare the differences between SA and HCA & LDA. Furthermore, since the complex calculation and professional statistical software were avoided, the analysis method was easily understood and grasped. Rhizoma Polygoni Cuspidati, the dried root and rhizome of Polygonum cuspidatum Sieb. et Zucc (Polygonaceae), which is widely distributed throughout China

5

(HuZhang in Chinese), Japan (Ko-Jo-Kon in Japanese) and North America (Mexican bamboo) [15], is a popular herbal medicine for the treatment of trauma, atherosclerosis, hypertension, hyperlipidemia, asthma, hepatitis, diarrhea and cancer [16-23]. Modern phytochemistry and pharmacology studies reported that P. cuspidatum contains a variety of prominent bioactive compound classes, including stilbenes, anthraquinones, napthalenes and their glycosides [24], each of which exhibits characteristic pharmaceutical properties. For example, resveratrol and its glycoside polydatin, as stilbenes, are widely used in medicine, health products and cosmetic industries for the anti-oxidant, antiinflammarory, anticancer and cardioprotective activities [25-28] and emodin, physcion, torachrysone and their glycosides, as anthraquinones and napthalenes, exerts antitumor and antibacterial pharmaceutical activities [29]. Since P. cuspidatum from different origins have extremely distinct chemical ingredients in qualitative and quantitative aspect, it was chosen as an object of our study to verify the comprehensive strategy based on chromatographic profiles combined with different chemometric methods. In this study, P. cuspidatum samples from different origins were studied using the new strategy. The chromatographic profiles were firstly constructed through the accurate and reliable accelerated solvent extraction (ASE) and high-performance liquid

chromatography-ultraviolet

(HPLC-UV)

method.

And

then

different

chemometric methods including SA and HCA combined with LDA were adopt to distinguish the samples from different origins and select characteristic chemical markers, respectively. This study provided a new and comprehensive strategy on 6

comparison of different chemometric methods and the quality evaluation, which was a valuable reference for the accuracy and reliability of herbal medicines on the quality control. 2. Materials and methods 2.1. Chemicals and reagents P. cuspidatum from 14 different regions in China were purchased from authentic local Traditional Chinese Medicine (TCM) pharmacies, and further confirmed by Dr. Luping Qin (Department of Pharmacognosy, School of Pharmacy, Second Military Medical University, Shanghai, China). Reference compounds, polydatin, resveratrol, emodin, physcion and 2-methoxystypandrone were isolated from plants P. cuspidatum by ourselves using High speed counter current Chromatography (HSCCC), and their purities were over 95% checked by high-performance liquid chromatography-diode array detection (HPLC-DAD). Their structures were unambiguously identified by comparison of their 1H-NMR, MS spectra with the literature data. The other nine reference compounds, gallic acid, epicatechin and epicatechin gallate were purchased from

Shanghai

yuanye

Bio-Technology

Co.,

Ltd,

resveratrol-4’-O-β-D-(6’’-O-β-D-galloy)-glucopyranoside (R6GG), emodin-8-O-β-Dglucoside, emodin-1-O-β-D-glucoside, physcion-8-O-β-D-glucoside from Shanghai Sunny Biotech Co., Ltd, torachrysone and torachrysone-8-O-β-D-glucoside from BioBioPar Co., Ltd. The cinnamic acid was selected as internal standard (IS) which was purchased from the National Institute for Control of Pharmaceutical and 7

Biological Products (Beijing, China). The HPLC grade organic solvent acetonitrile was purchased from Merck (Darmstadt, Germany). Analytical grade methanol, ethanol and acetic acid were obtained from China Medicine Group Shanghai Chemical Reagent Corporation (Shanghai, China). Double distilled (DI) water was produced by a Milli-Q academic water purification system (Millipore, USA). 2.2. Sample preparation A total of 14 batches of the dried P. cuspidatum samples (shown in Table 1) were ground into fine powder with a motor grinder, and then sieved through a 50 mesh sieve with a particle diameter of less than 0.355 mm. Finally, the powder was stored in sealed bags at room temperature till extraction. The sample of P. cuspidatum from Zhejiang Province was used in method development studies. ASE was performed on a Dionex ASE 350 (Dionex Corp., Sunnyvalue, CA) system. A portion of 1.0 g of P. cuspidatum powder was packed in a 10 ml Dionex stainless steel extraction cell. Each cell was equipped with stainless steel screw caps and a circular glass microfiber filter at the bottom to avoid the collection of suspended particles in the collection vessels. The extraction cell was arranged in the cell tray and was extracted using conditions obtained from single factor experiments. Then the cell was rinsed with fresh extraction solvent (60% of the extraction cell volume) and purged with nitrogen to expulse rest of solvent in the cell (100 s). Extracts were collected into 66 ml glass vessels. After extraction, each resulting extract obtained by ASE was filtered through filter 8

paper and the residue was rinsed extensively with extraction solvent three or four times to ensure removal of all bioactive constituents. The filtrate and washings were quantitatively transferred into a 50 ml volumetric flask and make up to volume with extractant. Cinnamic acid (400 μg/ml) was used as internal standard for quantitative determination of analytes. After membrane filtration, an aliquot of 20 μL was injected into the HPLC-UV system for analysis. All experiments were performed in triplicate. 2.3. HPLC-UV Chromatographic conditions The HPLC-UV quantitative analysis was performed on a Shimadzu HPLC chromatographic system (Shimadzu Corporation, Kyoto, Japan), which is equipped with two LC-10AT pumps, a SCL-10A system controller, a model DGU-10A degasser unit, an SIL-10AD autosampler, a CTO-10AS column oven and an SPD-10A UV-vis detector. The UV-vis detector was set at 230 nm. The chromatographic conditions were: DiamonsilTM C18 column (4.6 mm×200 mm, 5 μm, Dikma Technologies, Beijing, China); sample injection volume, 20μL; temperature of column oven, 25 ℃; flow rate, 1.0 ml/min. The mobile phases consisted of 0.01% acetic acid (A) and acetonitrile (B). A gradient elution was performed according to the following profile: 0-20 min, 15-30% B; 20-35 min, 30-50% B; 35-45 min, 50-80% B; 45-55min, 80-80% B. Data were processed with Shimadzu LC-S software. 2.4. Mixed standard solution preparation Dissolve cinnamic acid into methanol in order to get 8 mg/ml internal standard solution. The mixed standard stock solution containing gallic acid, epicatechin, 9

polydatin,

epicatechin

gallate,

resveratrol-4’-O-β-D-(6’’-O-β-D-galloy)-

glucopyranoside, emodin-1-O-β-D-glucoside, resveratrol, torachrysone-8-O-β-Dglucoside,

emodin-8-O-β-D-glucoside,

physcion-8-O-β-D-glucoside,

2-methoxystypandrone, emodin, torachrysone and physcion was prepared and diluted to a series of appropriate concentrations, adding 400 μg/ml internal standard for the construction of calibration curves. The mixed standard solutions were stored at 4 ℃ in darkness and brought to room temperature for HPLC analyses. 2.5. Data analysis Data analysis for chromatographic fingerprinting was performed using the professional software named “Similarity Evaluation System for Chromatographic Fingerprint of TCMs” (China Committee of Pharmacopeia, 2004A version), and then the HPLC chromatographic data were converted to data matrix of peak retention time and peak area. The peak area of each peak was imported into the SPSS 10.0 software package (Chicago, IL, USA) for HCA and LDA analyses. 3. Results and discussion 3.1. Optimization of HPLC-UV conditions To achieve chromatograms with satisfactory separation of chemical compounds in P. cuspidatum, we investigated and optimized a series of HPLC parameters including column, mobile phases, gradient elution and detection wavelength. Different columns were tested and it was found that some structural analogues could be eluted with baseline separation during shorter analysis time using DiamonsilTM C18 column. 10

Acetonitrile instead of methanol was selected as organic eluent in mobile phase system in order to gain more powerful elution ability and separation efficiency. A low concentration acetic acid was added to the mobile phase A for the improvement of peak shape of analytes. Additionally, since the components of P. cuspidatum is complex and there are some similar polarity of ingredients, a gradient elution mode was highly needed for the complete separation of structural analogous to avoid interference, restrain the broadening and overlapping of target peaks. After optimization of chromatographic conditions, a sufficiently large number of peaks on the chromatogram were achieved within 55 min when the column was eluted in gradient with acetonitrile and water containing 0.01% acetic acid. The most appropriate detection wavelength was set at 230 nm because most of characteristic compounds in P. cuspidatum have satisfactory sensitive at this UV wavelength. 3.2. Optimization of ASE conditions In order to optimize the extraction conditions, four important parameters controlling extraction yield, including the extraction solvent and the extraction temperature, extraction time and static cycles of ASE were studied. The intensities of three main peaks (peak 15, peak 26 and peak 34 shown in Fig. 1A) from the optimized HPLC-UV conditions were used to evaluate the extraction conditions. Water, 25% ethanol, 50% ethanol, 75% ethanol and 95% ethanol were investigated. The results demonstrated that 75% ethanol was the most suitable extraction solvent. Since temperature affects strongly the extraction efficiency, a series of experiments at 50, 75, 100, 125, 150 and 165 ℃, with 5 min of extraction time and two static cycles, 11

were performed. It was found that the highest extraction yields were obtained at 125 ℃. Then, different extraction times (3, 5, 7, 10 and 13 min) were evaluated along with fixing the other two variables at 125 ℃ and doubled static cycles. The compounds were almost completely extracted within 5 min. The static cycle of ASE was determined by performing consecutive accelerated solvent extractions on the same samples three times at 125 ℃ and 5 min. The yields of compounds were no longer increase after two cycles of extraction of the same matrix. Therefore, the extraction temperature, the extraction time and the static cycle as optimum for the extraction of P. cuspidatum were 125 ℃, 5 min and two cycles. 3.3. Chromatographic profiles of 14 batches of P. cuspidatum The variation of the compositions in different P. cuspidatum samples might depend on varieties, climate, agricultural areas, harvesting seasons and so on. For the purpose of controlling the quality of the herbs, chromatographic profiles are effectively applied to reveal chemical information of the botanical products and discriminate different P. cuspidatum samples. Under the optimal conditions, the HPLC profiles of fourteen batches of P. cuspidatum from various regions of China were obtained. The chromatograms of samples 1-14 were inputted into the “Similarity Evaluation System for Chromatographic Fingerprint of TCMs” for further analysis. It is found that a total of 38 peaks were shown in chromatographic profiles (Fig. 1A), nineteen peaks among which had been identified in our previous study [30]. 3.4. Chemometric methods of P. cuspidatum samples 12

3.4.1. SA of P. cuspidatum samples SA, as a statistic method, has been widely applied in fingerprint data to evaluate the similarity between two profiles. The State Food and Drug Administration (SFDA) suggests that chromatograms of herbal samples should be evaluated in terms of similarity by calculating the correlation coefficient of original data compared with one reference profile. Using “Similarity Evaluation System for Chromatographic Fingerprint of TCM” software, the common peaks could be discriminated which provided a direction for the next qualification of active ingredients. Based on the theory, a reference chromatogram was firstly constructed by median data (Fig. 1B). The reference fingerprint is a chromatographic pattern of some common kinds of components in all 14 samples which are pharmacologically active and chemically characteristic in practice. A total of thirteen common peaks in the reference fingerprint were determined, which were gallic acid (peak 5), polydatin (peak 15), resveratrol-4’-O-β-D-(6’’-O-β-D-galloy)-glucopyranoside (R2GG, peak 17), R6GG (peak 18), emodin-1-O-β-D-glucoside (peak 22), resveratrol (peak 24), torachrysone-8-O-β-D-glucoside (peak 25), emodin-8-O-β-D-glucoside (peak 26), physcion-8-O-β-D-glucoside (peak 28), 2-methoxystypandrone (peak 33), emodin (peak 34), torachrysone (peak 35) and physcion (peak 38). R2GG was tentatively identified according to UV absorption, m/z of their quasimolecular ions, MSn fragmentation patterns. The thirteen common peaks were regarded as quantitative indexes tentatively.

13

On the other hand, the chromatograms of 14 P. cuspidatum samples were compared with the reference chromatogram respectively, and their similarities were evaluated with correlation coefficients (Table 2). The results of similarities indicated that the samples shared the similar chromatographic patterns with the similarity indexes higher than 0.950, except for samples 2, 3, 4, 6 and 14. Generally, the closer the values were to 1, the more similar two chromatograms were. Therefore, based on similarity analysis results, the samples with correlation coefficient higher than 0.950 were clustered into a group, and those below 0.95 were sorted to another group. Moreover, we observed that the RSD values of relative peak area with reference of 13 common peaks from 14 samples were very high (11.21-67.70%). To obtain a more comprehensive evaluation for the quality of 14 batches of P. cuspidatum samples, additional Chemometrics is needed. 3.4.2. HCA combined with LDA of P. cuspidatum samples Since the SA classified the P. cuspidatum samples according to results of similarity calculated based on the relative value using common peaks as a reference chromatogram, minor differences between very similar chromatograms might not be distinguished. For the investigation of influence of all peaks on the classifiers, another well-known unsupervised pattern recognition method of data analysis, HCA, was carried out to assign a set of different P. cuspidatum samples into clusters by converting the entire chromatograms into statistical structures. Meanwhile LDA was used to create a model classifier between clusters using linear discriminant functions for group prediction purposes. Through the linear discriminant functions, the value 14

variables could be regarded as discriminate peaks to distinguish different samples. The HCA computation was implemented by performing singular value decomposition on the data array of the profiles, which consisted of a total of 14×38 data matrix. Each row represented a plant sample and each column contained the values of 38 peaks areas. The results of the cluster dendrograms were calculated by hierarchical and squared euclidean distance (Fig. 2). Clearly all P. cuspidatum samples were grouped into two main clusters. Cluster A consisted of samples 1, 5, 6, 7, 8, 10, 12, 13 and 14; Cluster B consisted of samples 2, 3, 4, 9 and 11. The groups were not consistent with that of SA. Then, based on the results of HCA, the linear discriminant functions of the two clusters were generated by LDA, which were as follows:

A  4.670E 4Peak 7  4.753E 4Peak 16  2.004E 2Peak 17  8.151E 3Peak 20  4.834E  4Peak 25  2.589E  2Peak 27  9.396E  3Peak 30  1.176E  3Peak 33  4.268E  2Peak 37  9.131E  3Peak 38  1.691E 4

B  7.660E 4Peak 7  7.763E 4Peak 16  3.282E 2Peak 17  1.335E 2Peak 20  7.954E  4Peak 25  4.258E  2Peak 27  1.554E  2Peak 30  1.931E  3Peak 33  7.008E  2Peak 37  1.500E  2Peak 38  4.557E 4

It can be seen that only ten variables were used to generate the linear discriminant functions. The ten variables demonstrated to have good discrimination ability for the classification of samples, which were epicatechin (peaks 7), epicatechin gallate (peak 16), R2GG (peak 17), peak 20, torachrysone-8-O-β-D-glucoside (peak 25), 15

citreorosein

(peak

27),

emodin-8-O-β-D-(6’-acetyl)-glucoside

(peak

30),

2-methoxystypandrone (peak 33), chrysophanol (peak 37), physcion (peak 38), respectively.

Emodin-8-O-β-D-(6’-acetyl)-glucoside

and

chrysophanol

were

tentatively identified according to UV absorption, m/z of their quasimolecular ions, MSn fragmentation patterns. 3.4.3. Selection of quantitative markers by SA and HCA & LDA The apparent differences between the results from SA and HCA & LDA were observed (Fig. 2 and Table 2). Based on the fact, the quantification of multi-ingredients was necessary for reasonable determination of the quality of herbal medicine and evaluation of different chemometric methods. A total of thirteen common peaks obtained by SA and ten discriminate peaks by HCA & LDA should be selected as quantitative indices. Due to the lack of standards of peak 27, peak 30 and peak 37 and failure to identify peak 20, a total of fifteen compounds were quantified, namely gallic acid (peak 5), epicatechin (peaks 7), polydatin (peak 15), epicatechin gallate (peak 16), R2GG (peak 17), R6GG (peak 18), emodin-1-O-β-D-glucoside (peak 22), resveratrol (peak 24), torachrysone-8-O-β-D-glucoside (peak 25), emodin-8-O-β-D-glucoside (peak 26), physcion-8-O-β-D-glucoside (peak 28), 2-methoxystypandrone (peak 33), emodin (peak 34), torachrysone (peak 35) and physcion (peak 38). Since R2GG and R6GG were isomers, the standard of R6GG was used to quantify the two compounds. 3.5. Validation of HPLC-UV method 16

The HPLC-UV chromatograms of mixed standard solution, P. cuspidatum extract and P. cuspidatum extract spiked with IS are shown in Fig. 3. For a specific, accurate and sensitive quantitative assay, the validation of the HPLC was clarified through a series of tests. 3.5.1. Linearity, limits of quantification The calibration curves of each reference compound were performed with six appropriate concentrations in triplicate. Linear regression analyses were carried out by plotting the peak area ratio of the analyte against the internal standard versus the analyte concentration of the mixed standard solutions which are shown in Table 3. It was observed that linear calibration curves were constructed in a wide concentration range corresponding to their levels in samples and the methods exhibited excellent linearity with high correlation coefficients (R2). The limits of quantification (LOQs) of the fourteen compounds were determined by diluting the sample matrix when the signal-to-noise ratios (S/N) of analytes were about 10 with the performance of repeatability, precision and reproducibility of method at the concentration. Table 3 summarizes LOQs for the fourteen standards, which indicated that the analytical method was accepted with sufficient sensitivity, accuration and precisions. 3.5.2. Repeatability, precision, reproducibility and stability The repeatability of retention time and peak area is the most important parameter for confident target compound quantification. The intra-day and inter-day precisions of the method were determined by injecting the sample solution for six replicates on 17

the same day and by measuring it twice a day for three consecutive days, respectively. The reproducibility was assessed by analyzing six samples of P. cuspidatum prepared independently. The sample stability was estimated by re-analyzing one sample during 12 hours. The precision of retention time and peak area were found to be better than 1.5% (RSD, n=6) for the target components and IS, indicating that the HPLC produced good repeatability from run to run. The RSD values of intra-day and inter-day precisions, reproducibility and stability for the fourteen compounds were listed in Table 3. The results demonstrated that the values were within acceptable range and the method provided an accurate, precise and stable measurement. 3.5.3. Recovery test The recovery test for reflecting accuracy was determined by the standard addition approach. The test was performed by adding known amounts of the fourteen reference compounds corresponding to their contents in sample into a certain amount (1.0 g) of P. cuspidatum separately. The spiked samples were then extracted by ASE under the optimum condition, processed and quantified by HPLC-UV mentioned above. Six replicates were performed for the test. The recovery was figured out according the formula: recovery (%) = (amount detected–original amount)/amount spiked ×100%. All the results were estimated on the ground of relative standard deviation (RSD). The results of recovery test are summarized in Table 3, indicating that the HPLC method was precise and accurate for the 14 standards with satisfactory recoveries. Validation tests demonstrated that the established HPLC-UV approach was suitable for simultaneously quantitative determination of fourteen compounds in P. cuspidatum. 18

3.6. Quantitative application of the methodology to 14 P. cuspidatum samples The developed ASE coupled with HPLC-UV method was subsequently applied to quality evaluation of fourteen batches of the P. cuspidatum samples. The contents of the 15 compounds from different samples were summarized in Table 4. A large variation was found in the contents of 15 marker compounds in the fourteen batches of samples, which represented the quality instability of the crude drug. 3.7. Comparison between SA and HCA & LDA To reflect quality fluctuations between batches, the following parameter was applied, which was employed in other studies [31]:

P 

ci ci

where ci denotes the measured concentration of a given compound, while ci denotes the average concentration of fourteen batches of P. cuspidatum. The closer P value is to 1, the better consistency between batches is. In general, values in the range of 0.75-1.25 are considered acceptable. The results are shown in the box chart in Fig. 4, where six outliers of P values are observed, namely epicatechin (peaks 7), polydatin

(peak

15),

epicatechin

gallate

(peak

16),

R6GG

(peak

18),

torachrysone-8-O-β-D-glucoside (peak 25), physcion-8-O-β-D-glucoside (peak 28). These outliers were the import factors that resulted in quality fluctuations of fourteen P. cuspidatum samples. In order to discriminate the differences in classifiers obtained by SA and HCA & 19

LDA, the quantification of the six outliers from fourteen P. cuspidatum samples were analyzed. From the details of Table 4, we found that the contents of the six outliers in sample 6 were much higher than other samples, especially for polydatin, the main pharmaceutical compound in P. cuspidatum. Meanwhile, the contents of the six outliers from sample 2, 3, 4 and 14 were at a slightly lower concentration. The results is consistent with that of SA. It could deduce that SA classifies samples through the mean values (the mean chromatogram), and the herbs containing relatively low or high amounts of main compounds were regarded as samples from different origins. On the other hand, samples 2, 3, 4, 9, 11 containing relatively small amounts of the six compounds overall were grouped in cluster B by HCA and other samples with relatively high amounts of the six outliers were in cluster A. It agreed with the theory of HCA that the two most similar objects or clusters are merged into a new cluster until all objects belong to only one cluster. Conclusively, due to the different theories of SA and HCA, variances existed in the classifier of origins of herbs. The principle of parameter P was consistent with that of principal component analysis (PCA). Both the outliers from the parameter P and the loadings from PCA could show the effect of compounds on quality of herbs. The outliers and loadings with large dispersion degree mean that the compounds had a large influence on the discrimination of the samples’ quality. Compared to PCA, the parameter P

was

easily calculated and understood which can avoid the usage of professional statistical software. 4. Conclusions 20

In this study, a comprehensive strategy using chromatographic profiles combined with reliable chemometric methods, including SA and HCA & LDA was successfully performed and applied to the quality analysis of P. cuspidatum. To compare the differences obtained in classifiers of samples calculated by SA and HCA & LDA and evaluate the quality of fourteen batches of P. cuspidatum samples, a total of fifteen quantitative markers in P. cuspidatum were first selected, which were considered to be important for the quality evaluation of P. cuspidatum, including thirteen common peaks selected by SA and ten discriminate peaks by HCA & LDA. Secondly, effective and confirmatory ASE and HPLC-UV methods were successfully established to the simultaneous determination of fifteen characterized components in fourteen batches of P. cuspidatum samples. Lastly, the quantitative results were analysed to reflect quality fluctuations of P. cuspidatum samples. It was concluded that SA classifies samples through the mean values and HCA & LDA classifies similar objects. In all, the study could provide a valuable reference for the utilization of chemometric methods and quality control of herbal medicines. Acknowledgments This work was supported by the National Natural Science Foundation, P.R. China (Grant No. 81273473 and 81573584) and the Major Project of National Science and Technology (2014ZX09J14106-06C).

21

References [1] F.Y. Gao, Y.S. Hu, G. Fang, G. Yang, Z.H. Xu, L.L. Dou, Z. Chen, G.R. Fan, Recent developments in the field of the determination of constituents of TCMs in body fluids of animals and human, J. Pharm. Biomed. Anal. 87 (2014) 241-260. [2] N. Hakimzadeh, H. Parastar, M. Fattahi, Combination of multivariate curve resolution and multivariate

classification

techniques

for

comprehensive

high-performance

liquid

chromatography-diode array absorbance detection fingerprints analysis of Salvia reuterana extracts, J. Chromatogr. A 1326 (2014) 63-72. [3] G. Qian, S.Y. Leung, G. Lu, K.S. Leung, Differentiation of rhizoma et radix Polygoni Cuspidati from closely related herbs by HPLC fingerprinting, Chem. Pharm. Bull. 54 (2006) 1179-1186. [4] T. Yi, H. Zhang, Z.W. Cai, Analysis of Rhizoma Polygoni Cuspidati by HPLC and HPLC-ESI/MS, Phytochem. Anal. 18 (2007) 387-392. [5] G. Qian, S.Y. Leung, G. Lu, K.S. Leung, Optimization and validation of a chromatographic method for the simultaneous quantification of six bioactive compounds in Rhizoma et Radix Polygoni Cuspidati, J. Pharm. Pharmacol. 60 (2008) 107-113. [6] J.R. Lucio-Gutiérrez, J. Coello, S. Maspoch, Enhanced chromatographic fingerprinting of herb materials by multi-wavelength selection and chemometrics, Anal. Chim. Acta 710 (2012) 40-49. [7] F.Y. Gao, Y.S. Hu, X.L. Ye, J. Li, Z. Chen, G.R. Fan, Optimal extraction and fingerprint analysis of Cnidii fructus by accelerated solvent extraction and high performance liquid 22

chromatographic analysis with photodiode array and mass spectrometry detections, Food Chem. 141 (2013) 1962-1971. [8] G.E. Morlock, P. Ristivojevic, E.S. Chernetsova, Combined multivariate data analysis of high-performance thin-layer chromatography fingerprints and direct analysis in real time mass spectra for profiling of natural products like propolis, J. Chromatogr. A 1328 (2014) 104-112. [9] Y. Han, J. Wen, T.T. Zhou, G.R. Fan, Chemical fingerprinting of Gardenia jasminoides Ellis by HPLC-DAD-ESI-MS combined with chemometrics methods, Food Chem. 188 (2015) 648-657. [10] L. Cuadros-Rodríguez, C. Ruiz-Samblás, L. Valverde-Som, E. Pérez-Castaño, A. González-Casado, Chromatographic fingerprinting: An innovative approach for food 'identitation' and food authentication – A tutorial, Anal. Chim. Acta 909 (2016) 9-23. [11] M. Goodarzi, P.J. Russell, Y. Vander Heyden, Similarity analyses of chromatographic herbal fingerprints: a review, Anal. Chim. Acta 804 (2013) 16-28. [12] A. Giacomino, O. Abollino, M. Malandrino, E. Mentasti, The role of chemometrics in single and sequential extraction assays: A Review. Part II. Cluster analysis, multiple linear regression, mixture resolution, experimental design and other techniques, Anal. Chim. Acta 688 (2011) 122-139. [13] L. Peng, Y.Z. Wang, H.B. Zhu, Q.M. Chen, Fingerprint profile of active components for Artemisia selengensis Turcz by HPLC–PAD combined with chemometrics, Food Chem. 125 (2011) 1064-1071. [14] S. Yudthavorasit, K. Wongravee, N. Leepipatpiboon, Characteristic fingerprint based on 23

gingerol derivative analysis for discrimination of ginger (Zingiber officinale) according to geographical origin using HPLC-DAD combined with chemometrics, Food Chem. 158 (2014) 101-111. [15] B.C. Vastano, Y. Chen, N. Zhu, C.T. Ho, Z. Zhou, R.T. Rosen, Isolation and identification of stilbenes in two varieties of Polygonum cuspidatum, J. Agric. Food Chem. 48 (2000) 253-256. [16] Y.W. Lin, F.J. Yang, C.L. Chen, W.T. Lee, R.S. Chen, Free radical scavenging activity and antiproliferative potential of Polygonum cuspidatum root extracts, J. Nat. Med. 64 (2010) 146-152. [17] H.W. Lin, M.X. Sun, Y.H. Wang, L.M. Yang, Y.R. Yang, N. Huang, L.J. Xuan, Y.M. Xu, D.L. Bai, Y.T. Zheng, K. Xiao, Anti-HIV activities of the compounds isolated from Polygonum cuspidatum and Polygonum multiflorum, Planta Med. 76 (2010) 889-892. [18] J.A. Shin, J.H. Shim, J.G. Jeon, K.H. Choi, E.S. Choi, N.P. Cho, S.D. Cho, Apoptotic effect of Polygonum Cuspidatum in oral cancer cells through the regulation of specificity protein 1, Oral Dis. 17 (2011) 162-170. [19] X.B. Wu, X.Q. Luo, S.Y. Gu, J.H. Xu, The effects of Polygonum cuspidatum extract on wound healing in rats, J. Ethnopharmacol. 141 (2012) 934-937. [20] B. Hu, H.M. An, K.P. Shen, H.Y. Song, S. Deng, Polygonum cuspidatum extract induces anoikis in hepatocarcinoma cells associated with generation of reactive oxygen species and downregulation of focal adhesion kinase, Evid. Based Complement. Alternat. Med. 2012 (2012) 607-675. [21] A. Kirino, Y. Takasuka, A. Nishi, S. Kawabe, H. Yamashita, M. Kimoto, H. Ito, H. Tsuji, 24

Analysis and functionality of major polyphenolic components of Polygonum cuspidatum (itadori), J. Nutr. Sci. Vitaminol. 58 (2012) 278-286. [22] J.H. Han, W. Koh, H.J. Lee, H.J. Lee, E.O. Lee, S.J. Lee, J.H. Khil, J.T. Kim, S.J. Jeong, S.H. Kim, Analgesic and anti-inflammatory effects of ethyl acetate fraction of Polygonum cuspidatum in experimental animals, Immunopharmacol. Immunotoxicol. 34 (2012) 191-195. [23] C.C. Lee, Y.T. Chen, C.C. Chiu, W.T. Liao, Y.C. Liu, H.M. David Wang, Polygonum cuspidatum extracts as bioactive antioxidation, anti-tyrosinase, immune stimulation and anticancer agents, J. Biosci. Bioeng. 119 (2015) 464-469. [24] J. Dong, H. Wang, L. Wan, Y. Hashi, S. Chen, Identification and determination of major constituents in Polygonum cuspidatum Sieb. et Zucc. by high performance liquid chromatography/electrospray ionization-ion trap-time-of-flight mass spectrometry, Se pu 27 (2009) 425-430. [25] M. Jang, L. Cai, G.O. Udeani, K.V. Slowing, C.F. Thomas, C.W. Beecher, H.H. Fong, N.R. Farnsworth, A.D. Kinghorn, R.G. Mehta, R.C. Moon, J.M. Pezzuto, Cancer chemopreventive activity of resveratrol, a natural product derived from grapes, Science 275 (1997) 218-220. [26] A.A. Bertelli, F. Ferrara, G. Diana, A. Fulgenzi, M. Corsi, W. Ponti, M.E. Ferrero, A. Bertelli, Resveratrol, a natural stilbene in grapes and wine, enhances intraphagocytosis in human promonocytes: a co-factor in antiinflammatory and anticancer chemopreventive activity, Int. J. Tissue React. 21 (1999) 93-104. [27] L.M. Hung, J.K. Chen, S.S. Huang, R.S. Lee, M.J. Su, Cardioprotective effect of resveratrol, a natural antioxidant derived from grapes, Cardiovasc. Res. 47 (2000) 549-555. 25

[28] H. Zhang, C.H. Yu, Y.P. Jiang, C. Peng, K. He, J.Y. Tang, H.L. Xin, Protective effects of polydatin from Polygonum cuspidatum against carbon tetrachloride-induced liver injury in mice, PloS one, 7 (2012) e46574. [29] H. Jayasuriya, N.M. Koonchanok, R.L. Geahlen, J.L. McLaughlin, C.J. Chang, Emodin, a protein tyrosine kinase inhibitor from Polygonum cuspidatum, J. Nat. Prod. 55 (1992) 696-698. [30] F.Y. Gao, T.T. Zhou, Y.S. Hu, L.A. Lan, Y.V. Heyden, J. Crommen, G.C. Lu, G.C. Fan, Cyclodextrin-based ultrasonic-assisted microwave extraction and HPLC-PDA-ESI-ITMSn separation and identification of hydrophilic and hydrophobic components of Polygonum cuspidatum: A green, rapid and effective process, Ind. Crops. Prod. 80 (2016) 59-69. [31] X.S. Liu, Z.Z. Wu, K. Yang, H.Y. Ding, Y.J. Wu, Quantitative analysis combined with chromatographic fingerprint for comprehensive evaluation of Danhong injection using HPLC-DAD, J. Pharm. Biomed. Anal. 76 (2013) 70-74.

26

Figure captions Figure 1 HPLC profiles of fourteen batches of P. cuspidatum samples extracted by ASE (A) and a reference chromatogram constructed by Similarity Evaluation System for Chromatographic Fingerprint of TCMs (B).

27

Figure 2 The hierarchical cluster dendrogram of fourteen batches of P. cuspidatum samples obtained by HCA.

28

Figure 3 HPLC chromatograms of mixed standard solution (A), P. cuspidatum extract (B) and P. cuspidatum extract spiked with internal standard (C).

29

Figure 4 Box chart of fifteen compounds from fourteen batches of P. cuspidatum samples.

30

Table 1 Sources of fourteen batches of P. cuspidatum samples. Sample no.

Sources

1

Anhui I

2

Anhui II

3

Fujian

4

Hubei I

5

Hubei II

6

Hubei III

7

Jiangsu

8

Jiangxi

9

Ningxia

10

Hebei

11

Sichuan

12

Yunan I

13

Yunan II

14

Zhejiang

31

Table 2 The similarities (correlation coefficients) of fourteen batches of P. cuspidatum samples. correlation coefficients Sample no. average

median

1

0.982

0.973

2

0.902

0.90

3

0.885

0.90

4

0.943

0.945

5

0.964

0.956

6

0.936

0.924

7

0.983

0.989

8

0.994

0.995

9

0.992

0.993

10

0.961

0.954

11

0.963

0.965

12

0.977

0.972

13

0.992

0.993

14

0.942

0.944

Note: The correlation coefficients were calculated with the average and median data.

32

Table 3 Regression equation, linear range, determination coefficient (R2), limit of quantitation (LOQ), intra-day and inter-day precisions, reproducibility, stability and recovery of fourteen compounds. Lin e

LO

ran

Q

Pe ak

Compounds

no

Regressio n equation

.

ge

R2

(μg

(μg

/ml

/ml

)

) 5

0.9 Y=0.0046

10-

X-0.0011

120

Y=0.0057

2-4

X-0.0063

0

gallic acid

99

7

Inter

Reprod

Stab

Recover

-day

-day

ucibility

ility

y (n=3)

prec

prec

(n=6)

(n=

M

R

ision

ision

RSD

6)

ea

S

(n=6

(n=3

(%)

RS

n

D

)

)

D

(%

(

RSD

RSD

(%)

)

%

(%)

(%)

2.67

2.84

) 1.63

2.25

10

97

2.

.4

66

97

2.

.7

83

10

1.

1.

71

6 0.9

epicatechin

Intra

99

1.83

1.32

1.71

2.77

2

6 15

0.9 Y=0.0054

60-

X-0.0143

560

epicatechin

Y=0.007X

2-1

gallate

-0.0109

00

polydatin

99

16

0.97

1.20

1.55

1.00

60

9

2

0.9 99

2.27

2.75

2.36

2.27

2

95

2.

.9

18

99

2.

.8

00

10

2.

2.

55

6 18

0.9 Y=0.0055

5-1

X-0.019

50

emodin-1-O-β-D-

Y=0.0023

1-1

glucoside

X-0.0012

00

R6GG

99

22

1.26

1.32

2.52

1.50

5

8 0.9 99

1.10

1.06

2.88

1.90

1

8 24

2

0.9 Y=0.0102

10-

X-0.0474

250

torachrysone-8-

Y=0.0137

1-1

O-β-D-glucoside

X-0.0318

50

resveratrol

99

25

1.79

1.88

2.00

1.27

10

97

2.

.5

08

92

1.

.6

32

91

2.

7 0.9 99

1.15

1.34

1.68

0.91

1

6 26

emodin-8-O-β-D-

Y=0.0062

8-5

0.9

8

0.89

1.25

2.83

0.71

33

glucoside

X-0.0381

0

99

.5

57

10

2.

5.

14

8 28

0.9 physcion-8-O-β-

Y=0.0051

8-5

D-glucoside

X+0.0121

0

99

1.21

1.33

2.74

1.11

8

6 33

6

0.9 2-methoxystypan

Y=0.0061

2-7

drone

X-0.0048

5

Y=0.009X

10-

+0.0513

500

99

34

1.64

1.11

2.85

2

97

2.

.1

11

93

1.

.7

98

97

2.

.4

13

96

2.

.7

75

7 0.9

emodin

1.89

99

1.71

1.62

2.52

1.60

10

9 35

0.9 Y=0.0107

5-4

X-0.0216

0

Y=0.0025

8-2

X-0.0229

00

torachrysone

99

38

0.43

1.74

1.34

5

9 0.9

physcion

0.55

99

1.04 8

1.34

2.01

1.78

5

34

Table 4 Extraction contents (mg/g) of fifteen compounds in fourteen batches of the P. cuspidatum samples. Sample

gallic

no.

acid

epicatechin

torachrysone-8-O-

emodin-8-O-

physcion-8-O-

β-D-glucoside

β-D-glucoside

β-D-glucoside

1

1.2

ND

11.1

3.7

1.6

10.8

2

1.2

ND

1.1

3.5

0.3

3

1.9

2.2

3.2

6.8

4

0.6

1.7

2.3

1.8

1.0

2.2

26.1

1.8

1.2

ND

17.8

0.5

1.2

ND

16.9

9

1.2

ND

10

1.1

11

1.3

12

epicatechin

polydatin

emodin-1-OR2GG

R6GG

2-methoxystypandrone

emodin

torachrysone

physcion

0.6

0.6

2.0

3.9

2.8

1.3

5.1

0.2

4.8

8.8

ND

0.7

1.5

3.5

0.9

1.4

10.6

0.2

8.3

ND

6.2

0.6

0.7

1.4

9.3

2.5

1.1

9.8

0.4

9.7

1.6

ND

8.0

ND

4.4

0.8

7.7

3.1

2.6

8.8

0.2

10.1

5

0.9

0.4

19.4

4.4

3.1

2.3

12.0

3.5

1.8

4.6

0.2

6.0

6

1.3

0.6

5.2

7.2

2.9

4.2

17.4

4.1

2.3

5.4

0.3

5.5

7

1.9

0.8

1.7

2.8

2.0

1.1

10.2

2.1

1.3

8.3

0.3

7.0

8

1.1

0.7

3.5

4.0

3.4

1.1

9.9

1.7

2.0

7.3

0.3

5.4

11.8

0.8

0.7

2.2

2.9

3.8

1.0

8.5

2.1

1.4

7.3

0.2

6.0

0.7

21.8

1.8

0.9

4.4

3.7

4.7

1.6

12.6

3.6

1.5

4.6

0.2

6.7

ND

10.5

0.6

0.7

2.7

1.9

5.4

0.7

6.7

1.8

1.3

7.9

0.2

8.2

0.9

ND

19.0

1.9

0.7

3.0

6.8

3.3

2.0

13.0

2.3

1.5

5.7

0.2

5.4

13

1.7

0.7

14.1

0.9

0.7

2.0

3.3

2.3

1.1

8.9

2.1

1.3

6.2

0.3

5.9

14

1.8

ND

14.9

0.7

0.6

1.7

2.9

4.0

0.8

7.8

1.7

1.1

8.2

0.2

5.5

gallate

β-D-glucoside

resveratrol

ND, not detected

35