Quality consistency evaluation of Kudiezi Injection based on multivariate statistical analysis of the multidimensional chromatographic fingerprint

Journal of Pharmaceutical and Biomedical Analysis 177 (2020) 112868

Contents lists available at ScienceDirect

Journal of Pharmaceutical and Biomedical Analysis journal homepage: www.elsevier.com/locate/jpba

Quality consistency evaluation of Kudiezi Injection based on multivariate statistical analysis of the multidimensional chromatographic fingerprint Hui Wang a,b,1 , Meiling Chen a,b,1 , Jie Li a,b , Ning Chen a,b , Yanxu Chang c,d , Zhiying Dou a,b , Yanjun Zhang a,b , Pengwei Zhuang a,b , Zhen Yang a,b,∗ a

Tianjin Key Laboratory of Chinese medicine Pharmacology, Tianjin University of Traditional Chinese Medicine, Tianjin, 301617, China College of Chinese Materia Medica, Tianjin University of Traditional Chinese Medicine, Tianjin, 301617, China c Tianjin State Key Laboratory of Modern Chinese Medicine, Tianjin University of Traditional Chinese Medicine, Tianjin, 301617, China d Key Laboratory of Pharmacology of Traditional Chinese Medical Formulae, Ministry of Education, Tianjin University of Traditional Chinese Medicine, Tianjin, 301617, China b

a r t i c l e

i n f o

Article history: Received 13 June 2019 Received in revised form 29 August 2019 Accepted 6 September 2019 Available online 8 September 2019 Keywords: HPLC-UV GC–MS HPIEC Information fusion Hierarchical clustering analysis

a b s t r a c t Traditional Chinese Medicine Injection (TCMI) was restricted due to the batch-to-batch variability caused by the variable compositions of botanical raw materials and complexities of the current manufacturing process. To evaluate and control the quality of Kudiezi Injection (KDZI), a comprehensive and practical method based on multidimensional chromatographic fingerprint associated with multivariate statistical analysis was proposed. The multidimensional chromatographic fingerprint was established by integrating three kinds of chromatographic fingerprints, including High Performance Liquid Chromatography-Ultraviolet spectrum (HPLC-UV), Gas Chromatography-Mass Spectrometer (GC–MS) and High performance ion-exchange chromatography (HPIEC), which were used to detect flavones, nucleosides, organic acids, amino acids and saccharides in KDZI. In addition, four main multivariate statistical analyses were compared to assess the batch-to-batch consistency of samples. Results showed that the cosine method, which has been widely used in the quality evaluation of TCM, failed to distinguish the differences among batches based on neither chromatographic peaks’ area nor contents information. t-test and Bayes’ theorem could reveal the content difference among batches, while hierarchical clustering analysis could differentiate KDZI batches, and Luteolin-7-O-␤-D-glucuronopyranoside, Tau, Ser, guanine and allose were the main indicators. In conclusion, multidimensional chromatographic fingerprints could reflect the quality information of KDZI comprehensively and hierarchical clustering analysis was suitable to identify the differences among batches. This could provide an integrated method for consistency evaluation of TCMI, process improvement of TCMI and solving similar problems in TCMI. © 2019 Elsevier B.V. All rights reserved.

1. Introduction Ixeris sonchifolia Hance, named Kudiezi (KDZ), is a traditional Mongolian medicine belonging to the compositae family, which has been recorded in Inner Mongolia Chinese Herbal Medicine since 1972 [1]. It has been used as a food and folk medicine

∗ Corresponding author at: Tianjin Key Laboratory of Chinese medicine Pharmacology, Tianjin University of Traditional Chinese Medicine, Tianjin, 301617, China. E-mail addresses: [email protected] (H. Wang), [email protected] (M. Chen), [email protected] (J. Li), [email protected] (N. Chen), [email protected] (Y. Chang), [email protected] (Z. Dou), [email protected] (Y. Zhang), [email protected] (Z. Yang). 1 These two authors contributed equally to this article. https://doi.org/10.1016/j.jpba.2019.112868 0731-7085/© 2019 Elsevier B.V. All rights reserved.

to treat appendicitis, amygdalitis and alleviate pain in China for thousands of years [2]. KDZ injection (KDZI) is a kind of injection extracted from Ixeris sonchifolia Hance, which was extensively applied in clinical medicine for many years for invigorating blood circulation [6,7], dissipating blood stasis to relieve pain, relaxing of vascular smooth muscle, enhancing of the activities of the fibrinolytic enzymes, decreasing myocardial infarct size and inhibiting thrombosis [5,8,9]. KDZ and KDZI contain flavonoids, amino acids, xylogens, steroids, phenols, adenosines, sesquiterpene lactones, and triterpenoid saponins, etc. Flavonoid glycosides and organic acids, including chlorogenic acid, caffeic acid, ferulic acid, cichoric acid, luteoloside, luteolin-7-glucuronide, are commonly considered as its effective components, and have been used as quality indicators of KDZ and KDZI [3–5].

2

H. Wang, M. Chen, J. Li et al. / Journal of Pharmaceutical and Biomedical Analysis 177 (2020) 112868

Traditional Chinese Medicine Injection (TCMI) has the characteristics of convenient administration and the significant clinical effects; therefore it has played an irreplaceable role in clinic to treat some emergency and severe diseases in recent years [10]. However, due to the complex chemical mixtures in botanical raw materials and inconsistencies involved in current manufacturing processes, variability between different batches has been one of the main problems affecting the effectiveness and safety of TCMI [11]. Therefore, studies on evaluating the batch quality and consistency can be especially indispensable for raw material screening and process improvement of TCMI. Present-day investigations of Kudiezi and KDZI have led to its fingerprint to establish an evaluation standard due to its remarkable active components in clinic applications [12]. Fingerprint is a kind of comprehensive analytical technology, which can fully characterize the type and the content of the chemical compositions in TCM and their preparations. In 2000, fingerprint of TCMI was authorized as an identified and qualified technology of TCMI by National Medical Products Administration (NMPA) [13]. Unidimensional fingerprint, a single fingerprint of some kind of chemical composition by single technique, was focus on modern research due to the rapid and simple determination [14,15]. However, in our previous experiment, quality analysis by unidimensional fingerprint of active flavonoid glycosides and organic acids could not reflect the variability among KDZI batches. Meanwhile, as more kinds of compounds including sugars and amino acids were identified from KDZI, unidimensional fingerprint failed to reflect the complete profile of TCM and usually was not sensitive enough for quantitative analysis [16]. It’s particularly noteworthy that the emergence of multidimensional fingerprints overcomes this limitation and provides a new practical method to evaluate the quality of TCM [12,16]. At present, spectrum and chromatogram are two extensively used methods for multidimensional fingerprints, such as UV, IR, NMR and MS applied in the structural analysis, and TLC, HPLC used in the separation of the compounds [17–20]. In addition, numerous analytical methods including current similarity analysis, clustering analysis and principal component analysis (PCA), were commonly used for statistical analysis of information obtained from fingerprints [21–23]. Similarity analysis based on cosine, t-test or Bayes’ theorem generally focused on calculating similarity indexes among different fingerprints [21,24]. Clustering analysis was a classification and aggregation method for the fingerprints of complex samples [25]. PCA aimed at discrimination between different batches or source of samples [26]. Therefore, it was necessary to develop a proper quality evaluation method for the multidimensional fingerprints combined with multivariate statistical analysis for herbal medicines or their preparations. In this study, a combined use of multidimensional fingerprint and multivariate statistical analysis was explored to assess batchto-batch quality consistency of KDZI. Compared with available HPLC-UV unidimensional fingerprint, multidimensional fingerprint with various compounds obtained by integration of HPLC-UV, GC–MS and IE information could evaluate the quality of samples more comprehensively. Multivariate statistical analyses such as the cosine, t-test, Bayes’ theorem and hierarchical clustering analysis (HCA) based on information fusion were applied to evaluate batch-to-batch quality consistency of KDZI.

2. Materials and methods 2.1. Materials and reagents Ten batches of KDZI (serial numbers are 120421, 120424, 120426, 120428, 120430, 120432, 120434, 120436, 110601, 111011, 111105, 111218, 120109, 120209, 120312, 120438,

120532, and 120,625) were provided by Tonghuahuaxia Co., Ltd. (Jilin, China). Luteolin-7-O-␤-D-glucopyranoside was obtained from Yifang Science and Technology Co., Ltd. (Tianjin, China). Apipenin-7-O-␤-glucuronopyranoside was purchased from Ruifensi Biotechnology Co., Ltd. (Chengdu, China). Luteolin-7-O␤-D-glucuronopyranoside was prepared in laboratory. Caftaric acid, cichoric acid, chlorogenic acid and caffeic acid were all obtained from National Institute for Control of Biological and Pharmaceutical Products of China (Beijing, China). Uridine and cytidine were supplied by SIGMA Co. Guanosine and adenosine was purchased from National Institute for Control of Biological and Pharmaceutical Products of China, and Xi’ensi Biochemical and Technology Co., Ltd. respectively. D-Glucose and D-mannose were purchased from National Institute for Control of Biological and Pharmaceutical Product (Beijing, China). Arabinose and allose were obtained from Yifang Technology Co., Ltd. (Tianjin, China). Inosose were obtained from Chemical Reagent Co., Ltd. (Tianjin China). All of the reference standard substances mentioned above were with purity not less than 98%. Elution reagents were of HPLC grade. Formic acid was purchased from Chemical Reagents Wholesale Company (Tianjin, China). Acetonitrile and methanol were obtained from Sigma-Aldrich. Sodium citrate, Citric acid Sodium chloride, and Ninhydrin of analytical grade were purchased from Jiangtian Chemical Co., Ltd. (Tianjin, China). Pyridine and hexamethyldisilazane (HMDS) were obtained from Tianjin Guangfu Fine Chemical Institute. Acetic anhydride was purchased from Jiangtian Chemical Co., Ltd. and Hydroxylamine hydrochloride was purchased from Guangfu technology development Co., Ltd. (Tianjin, China). Trimethylchlorosilane (TMCS) were obtained from Kelong chemical Reagent Co., Ltd. (Tianjin, China). Ninhydrin was purchased from Jiangtian Chemical Co., Ltd. (Tianjin, China). 2.2. High performance liquid chromatography-ultraviolet spectrum (HPLC-UV) fingerprint of KDZI 2.2.1. Preparation of sample solution Ten batches of KDZIs (1 g raw material/1 mL) were filtered with a 0.45 ␮m acetate filter and stored at 4 ◦ C. Standard substances including guanosine, thymidine, uridine, adenosine, guanine, cytidine, adenine, uracil, chlorogenic acid, caftaric acid, cichoric acid, dihydroxybenzoic acid, caffeic acid, luteolin-7-O␤-D-glucopyranoside, luteolin-7-O-␤-D-glucuronopyranoside and apipenin-7-O-␤-glucuronopyranoside, were accurately weighted and dissolved with double distilled water (electrical conductivity less than 2.2 us/cm) for calibration curves (Table S6). 2.2.2. Chromatographic conditions and methodology validation A Waters e2695 High Performance Liquid Chromatography (HPLC) separations system coupled with a 2998 PDA detector (Milford, Massachusetts, US) and a Symmetry C18 column (4.6 mm × 150 mm, 5 ␮m, Waters, US) were applied in the experiment. The mobile phases were composed of water (A) containing 0.05% (v/v) formic acid and acetonitrile (B) containing 0.05% (v/v) formic acid. The gradient program was set as follows: 0–5 min, 2% B; 5–10 min, 2%–9% B; 10–15 min, 9%–11% B; 15–21 min, 11%–18% B; 21–65 min, 18%–23% B. The injection volume was 20 ␮L with the flow rate maintained at 1.0 ml/min. The column temperature was set at 35 ◦ C and the detector wavelength was 260 nm. The methodology validation was carried out under the optimal determined conditions. The relative standard deviation (RSD) values of the relative retention times (RRTs) and relative peak areas (RPAs) were used to ensure the validity of this method. For the accuracy, the same sample was analyzed for six replicates within one day. The RSD of the RRTs and RPAs of the HPLC-UV fingerprint of KDZI were less than 1.0% and 5.0% respectively, which implied

H. Wang, M. Chen, J. Li et al. / Journal of Pharmaceutical and Biomedical Analysis 177 (2020) 112868

the good accuracy. The repeatability test was evaluated by analyzing six samples from the same batch. The RSD of the RRTs and RPAs were less than 1.0% and 4.3% respectively, which indicated the good reproducibility of the method. The stability was determined by analyzing the same batch. The RSD of the RRTs and RPAs were both less than 1.0%, which showed the sample were stable within 24 h. 2.3. Gas chromatography-mass spectrometer (GC–MS) fingerprint of sugars in KDZI 2.3.1. Sample preparation The KDZI samples were extracted and enriched by filtered with solid phase extraction (HyperSep C18, Thermo). The extract was collected into a round-bottom flask. The pH value of the extract was adjusted to 1 with hydrochloric acid (1 M). The extract was hydrolyzed for 3 h at 70 ◦ C under vacuum condition, and then adjusted to neutral with 10% Sodium hydroxide solution after cooling to room temperature. Five milliliters extract was accurately measured into centrifuge tube and freeze-dried. Allose, inosose, D-mannose, D-glucose and arabinose were prepared as aldoses standard solution and fructose was prepared as ketose standard solution. These stock solutions were serially diluted for calibration curves. The water used for the preparation of KDZI was taken as blank control. 2.3.2. Aldose and ketose derivatization One milliliter of sample solutions and aldoses standard solutions were transferred to centrifuge tube and freeze-dried respectively. Ten milligram of hydroxylamine hydrochloride used as derivatization reagent and 500 ␮L pyridine used as solvent and catalyst were added to mix in the tube. The solutions were placed in the 90 ◦ C water bath to react for 30 min. After cooling to room temperature, 500 ␮L acetic anhydride was added and the mixed solution was kept in the water bath to acetylate for 30 min. After centrifuging at 11,000 rpm for 15 min, the supernatant of derivatization was preserved in refrigerator for analysis. One hundred milligram hydroxylamine hydrochloride and 0.6 ml pyridine were added to accurately measured sample solutions and fructose reference solution. The solutions were kept in the 90 ◦ C water bath for 30 min. Then 0.3 mL HMDS was added and the mixed solutions were kept in the water bath for another half an hour. Derivation of fructose was obtained by silane treatment with TMCS. After centrifuging at 10,000 rpm for 15 min, the supernatant was kept for analysis. 2.3.3. Gas chromatographic conditions and methodoogy validation The GC–MS-QP 2010SE (SHIMADZU, Japan) with the capillary column (Rtx-5 MS, 0.25 nm × 30 m, 0.25 ␮m) were used to analyze the samples. Helium of high purity (99.999%) was used as the carrier gas with the flow rate of 1.0 mL/min. The injection volume was conducted at 1 ␮L. The split ratio was controlled at 30: 1. The oven was initially held at 88 ◦ C for 2 min, and ramped to 180 ◦ C at 20 ◦ C/min, then ramped to 200 ◦ C at 5 ◦ C/min. The final temperature of 280 ◦ C was ramped at 20 ◦ C/min and held for 2 min. The temperature of gasification chamber and gas-mass interface were set at 280 ◦ C and 270 ◦ C respectively. EI were used as ion source with the temperature of 200 ◦ C and the energy of 20 ev. The range of quality was set at 45–550 amu. The method was validated according to the determined conditions and procedures. The RSD values of the RRTs and RPAs in the accuracy tests were less than 0.9% and 0.4% respectively. The sample was stable in 10 h with the RSD values of the RRTs and RPAs were less than 0.04% and 4.0% respectively. The RSD values of the RRTs and RPAs in repeatability experiments were less than 0.04%

3

and 4.0% respectively. The recovery was tested by analyzing the twelve replicate samples of the same concentration, and the RSD of the recovery was less than 5.0%. 2.4. High performance ion-exchange chromatography fingerprint (HPIEC) of amino acids in KDZI 2.4.1. Sample preparation The mixed amino acids standard solution was diluted with double distilled water to 2 nmol/L. The KDZI were evaporated under nitrogen to condense into three times and filtered through 0.22 ␮m membrane filtration prior to injection. Ninhydrin was added to the sample solution as chromogenic agent before flowing into the column. Stock solution was stored in a refrigerator at 4 ◦ C for analysis. 2.4.2. Chromatographic conditions and methodology validation Amino acids in KDZI were analyzed by high performance ion-exchange chromatography. The detection wavelength of V1 channel and V2 were conducted at 570 nm and 440 nm respectively. The elution solvents consisted of sodium citrate, citric acid and sodium chloride were buffers of different pH values (The pH value of B1 , B2 , B3 , B4 solution was 3.3, 3.2, 4.0, 4.9 respectively, and the B5 solution was the sodium hydroxide solution (8%)). The operation was performed under the temperature of 135 ◦ C and the column temperature was controlled at 50 ◦ C. The rate of elution solvents and Ninhydrin solution were 0.4 mL/min and 0.35 mL/min respectively. The accuracy was determined by replicating the sample injection six times a day. The same injection was injected at 0 h, 2 h, 4 h, 6 h, 8 h, 10 h for the stability. The repeatability was tested by analyzing the six replicate injections. Results showed that the RSD values of the RRTs and RPAs to evaluate the accuracy, stability and repeatability were less than 2.8% and 6.0%, 2.7% and 4.4%, 0.18% and 1.0% respectively. All the tests of methodology validation suggested that this method was practical for determination of amino acids in KDZI. 2.5. Data analysis Chemical components in KDZI, including nucleotides, organic acids, flavonoids, saccharides and amino acids, were detected by HPLC-UV fingerprints, GC–MS fingerprints and IE fingerprints respectively. There were large differences among the vector’s absolute value in the three fingerprints. Uniformization method was used to fuse all the chemical information of KDZI to multidimensional fingerprint, all the peak area was uniformized according to the Eq. (1). 

Ai = Ai /



˜ Ai (i = 1n)

(1)



Ai is the peak area after uniformization and Ai is the peak area of the ith peak in a certain fingerprint. 2.5.1. The cosine The cosine was one of the commonly employed analytical methods for calculating similarity of herbal chromatograms suggested by FSDA [31,32]. The calculation was shown in the Eq. (2).

c=

n  i=1

   n  n  2   A¯i 2 Ai A¯i / Ai . i=1

(2)

i=1

Where A¯ i is the uniformized peak area of a certain compound in a certain batch and Ai is the average relative retention times of the ten injections.

4

H. Wang, M. Chen, J. Li et al. / Journal of Pharmaceutical and Biomedical Analysis 177 (2020) 112868

Fig. 1. (A) 15 common compounds identification and (B) Chromatographic fingerprint of 10 batches KDZI in HPLC-UV (1. Cytidine 3. Uracil 4. Uridine 5. Adenosine 6. Guanosine 10. Caftaric acid 11. chlorogenic acid 12. 2,5-dihydroxycinnamic acid 13. Luteolin-7-O-␤-D- glucose-1-2 glucoside 14. Sonchifolactone D 15. Luteoloside 16. Luteolin-7-glucuronide 18. Cichoric acid 19. sonchifolinin 20. Apigenin 7-glucuronide. 2, 7, 8, 9 and 17 were unidentified).

This method was also used for analyzing the similarities among batches after information fusion. Its calculation procedure was in Eq. (3).



s (Bik , Bok ) =

m 

 

Bik .Bok /

k=1

m 



2

Bik .Bok

2

(3)

k=1

Bik is the peak area of a certain common peak of the sample multidimensional fingerprint, and Bok is the peak area of the hypothesis peak from the hypothetical batch of KDZI integrated by the ten batches of KDZI. 2.5.2. T-test T-test is an efficient method to infer the probability of occurrence of differences by t-distribution theory and is widely used for significant evaluation. The calculation was shown in the Eq. (4).

  t = X1 − X2  / S1 2 + S2 2

2.5.4. Hierarchical clustering analysis for KDZI Hierarchical clustering analysis (HCA) is a useful clustering tool that aggregates and sorts the target samples to form a phylogenetic tree, which provides a visual analysis method for evaluating differences in and outside the sample groups [34]. In this study, HCA was used to check if there was a difference among batches of KDZI from the perspective of multiple analytical methods. The result of HCA was displayed as a heatmap generated by Multi Experiment Viewer Software (Version 4.6.0). 3. Results and discussion 3.1. Chromatographic fingerprint and content obtained by HPLC-UV

(4)

X1 and X2 are the average value of the peak area or the compound content of the sample batch and the hypothesis one respectively. S1 and S2 are the standard deviation of the sample and the hypothesis respectively. 2.5.3. Bayes’ theorem Bayes’ theorem method is a versatile probabilistic reasoning algorithm that can be used to predict the similarity of fingerprints [33]. The calculation was shown in the Eq. (5).

  n   √   t = 1/n ri  n  

significant difference, where ␣ represents the confidence level and f means the freedom.

(5)

i=1

r = v1 -v2 v1 and v2 are the fingerprint of the sample and the hypothesis batch of KDZI integrated by the ten batches respectively. Where ri is the i th element of the r and n is the number of the chosen peak. If v1 =v2 , there is no difference between the two fingerprints. If v1 and v2 is close enough, the two fingerprints can be considered of no significant difference. If t > t˛,f , the two fingerprints have no

HPLC-UV has been widely used in the quantitative analysis or qualitative identification due to its high efficiency, sensitivity, and selectivity. In the experiment, a fingerprint of adenosine and flavonoids in KDZI was established by HPLC-UV. The fingerprint was shown in the Fig. 1B. Results analyzed by cosine were almost all greater than 0.99 (Table S3) (Supplementary material), suggesting that the HPLC-UV fingerprints were of little difference. Fifteen common peaks in the fingerprint chromatography were identified by liquid chromatography - mass spectrometry information and comparison with standard substances. (Table S1 (Supplementary material) and Fig. 1A). There were substantial differences of certain compounds in different batches. Chlorogenic acid was selected as an example: the maximum content (0.007 mg/ml) was nearly seven folds higher than the minimum content (0.0011 mg/ml), while the analysis results were all greater than 0.9, indicating that the cosine method could not reflect the differences of the compound contents. 3.2. GC–MS fingerprint of 10 batches of KDZI Due to the complex diversity of carbohydrate composition in TCM, quantitative and qualitative analysis of carbohydrates com-

H. Wang, M. Chen, J. Li et al. / Journal of Pharmaceutical and Biomedical Analysis 177 (2020) 112868

5

Fig. 2. (A) 7 Compounds identification and (B) chromatographic fingerprint of 10 batches KDZI in GC–MS (1. arabinose 2. allose 3. mannose 4. glucose 5. galactose 6. inose 1 7. inose 2).

Fig. 3. (A) Identification of peaks in KDZI compared with (B) reference KDZI by HPIEC (1. Taurine 2.Aspartic acid 3.Threonine 4.Serine 5.Glutamic acid 6.Glycine 7.Alanine 8. Cystine 9. Valine 10. Methionine 11. Isoleucine 12. Leucine 13.Tyrosine 14.Phenylalanine 15.Ornithine 16. Lysine 17.NH3 18. Histidine 19.Tryptophan 12. Arginine 21. Proline).

ponents remains a major challenge [27]. Sensitivity of traditional HPLC-UV or HPLC-ELSD is not sufficient for simultaneous determination of multiple carbohydrates components [28]. A variety of derivatization methods combined with GC–MS technology to determine monosaccharides and polysaccharides in complexes is a rapid and efficient analysis method at the present stage [29]. In our pre-experiment, the content of carbohydrates was higher than 90% of the solid content of KDZI. Therefore, GC–MS was used in the detection of the monosaccharides, the products of polysaccharides

and disaccharides hydrolysis that could be easily detected with volatility and thermos ability after derivatization in the KDZI. The fingerprints of sugars in ten batches of KDZI were shown in Fig. 2B. Compounds identification was shown in Fig. 2A and their content was shown in Table S2 (Supplementary material). The cosine was adopted to analyze the fingerprint similarities data (Table S4 (Supplementary material)). Similarities among 10 batches were almost greater than 0.98. However, there was still some differences in the content of certain compounds in different batches. Arabinose, for

6

H. Wang, M. Chen, J. Li et al. / Journal of Pharmaceutical and Biomedical Analysis 177 (2020) 112868

Fig. 4. Hierarchical clustering analysis heatmap and clustering results.

example, the maximum content is 1.5 fold higher than the minimum content, which suggested that analyzing similarities in this way could not reflect the content difference. 3.3. HPIEC fingerprint of amino acids in KDZI Based on the previous research, content of amino acids was in the fourth place (about 4% of KDZI solid content), ranking only after carbohydrates, HPLC components, and NaCl. IEC is widely used in amino acid separation, derivatization as well as detection, according to various structures, diverse pH values and disparate polarities of amino acids. In the research, to evaluate the KDZI comprehensively, fingerprint of amino acids was obtained by the HPIEC. Fourteen peaks were identified by comparing the RRTs of the sample and reference, as shown in Fig. 3A and Fig. 3B. The cosine was used to analyze the similarities of the fingerprint. Similarities of the fingerprint among the batches were all greater than 0.99, as shown in Table S5 (Supplementary material). However, the content of the same amino acid in different batches varied considerably (Table S2 (Supplementary material)). Take Tau as an example, the maximum content was more than two times the content of the minimum, which showed that the cosine analysis results of HPIEC fingerprint failed to reflect the content difference between batches. According to the above results, three unidimensional fingerprints and cosine analysis couldn’t reflect the content difference and were unable to fulfill the overall evaluation of KDZI, so information fusion was applied for the further analysis. 3.4. Variation analysis base on the cosine, T-test and Bayes’ theorem with information fusion Information fusion simulates the process of managing information comprehensively by human brain, in which the basic idea was integrated multi-information according to certain principles to obtain the reasonable description and overall assessment [30]. In the experiment, information fusion method was applied to uniformize all the peak area obtained from HPLC, GC–MS and HPIEC in order to obtain comprehensive information and overall evaluation of the KDZI. Compared results of cosine with peak area information and fusion information were shown in the Table 1. Compared with unidimensional fingerprint, the similarities among batches after

Table 1 Results analyzed by cosine with peak area information and fusion information of KDZI. Batch number

111105 120312 120,625 120209 120109 120438 110601 111011 120532 111218

Results analyzed by cosine HPLC-UV

GC-MS

IER

fusion information

0.999 0.998 0.997 0.992 0.998 0.997 0.999 0.998 0.996 1.000

1.000 0.998 0.999 1.000 0.996 1.000 1.000 0.997 1.000 0.997

1.000 0.999 0.990 0.999 0.997 0.992 0.992 0.999 0.997 0.981

0.999 0.985 0.997 0.998 0.996 0.996 0.998 0.997 0.999 0.991

Table 2 Results analyzed by Bayes’ theorem and T-test method with the content of chemical components in KDZI. Sample

111218 120532 111011 110601 120438 120109 120209 120,625 120312 111105

with content information Bayes’ theorem

T-test

0.996 0.998 0.999 0.999 0.998 0.904 0.984 0.993 0.978 0.994

0.760 0.880 0.860 0.980 0.940 0.600 0.830 0.990 0.890 0.850

information fusion were almost no less than 0.99 (>0.80), failed to reveal the peak area difference. To find a more proper data analysis approach, other analysis methods including t-test and Bayes’ theorem method were employed to analyze the difference. What’s more, the compound contents were also used as the objective data to analyze the difference shown in the Table 2. These two methods can reveal the content difference among batches. Therefore, it was practical to

H. Wang, M. Chen, J. Li et al. / Journal of Pharmaceutical and Biomedical Analysis 177 (2020) 112868

apply different methods to analyze the differences of fingerprint and content information fusion.

3.5. Hierarchical clustering analysis A hierarchical clustering analysis (HCA) heatmap representation was performed for unsupervised clustering after Pareto scaling with mean-centering based on the contents of 10 batches sample. As shown in Fig. 4. the 10 batches sample had obvious group distinctions; the dendrogram was divided into three main groups. Sample B, D, C, E, H and sample A, I, G, J were clustered into two groups respectively, and sample F was clustered into an independent group. Furthermore, according to the results, Luteolin-7-O-␤-Dglucuronopyranoside (16), Tau (24), Ser (27), guanine (10) and allose (18) were the main indicators of the difference. It was demonstrated that the multidimensional fingerprint can be used for distinguishing different batches of KDZI. Therefore, the hierarchical clustering analysis was more powerful and rational to reveal the difference of batches. Meanwhile, as a result, amino acids, like Tau, Ser, guanine and allose, played an important role in the distinction of different batches, which suggested that for consistency evaluation of TCM, not the active components should be considered, but all the comprehensive and complete information.

7

batches. It would be helpful for optimizing the raw materials and improving the process of TCMI. All the KDZIs in this study were obtained from the same manufacturer, which made the applicability of established method reliable within a relatively limited range. To develop the practicability, KDZIs from different manufacturers and different regions should be adopted for evaluation standard. And more research on multidimensional fingerprint and multivariate statistical analysis of the TCM should be conducted in later study. CRediT authorship contribution statement Hui Wang: Methodology, Writing - original draft. Meiling Chen: Writing - review & editing. Jie Li: Writing - original draft. Ning Chen: Investigation. Yanxu Chang: Writing - review & editing. Zhiying Dou: Validation. Yanjun Zhang: Funding acquisition. Pengwei Zhuang: Project administration. Zhen Yang: Supervision. Declaration of Competing Interest The authors declare that they have no competing interests. Acknowledgements This work was supported by National Natural Science Foundation of China (Grant No. 81703702) and Natural Science Foundation of Tianjin (Grant No. 17JCYBJC28900).

4. Conclusion Appendix A. Supplementary data TCM and its pharmaceutical formulation have been widely used around the world for thousands of years and its complex components make it difficult to evaluate the quality comprehensively. The quality control of Chinese medicine formulation relates to the safety and effectiveness in clinical practice. As a new type of Chinese medicament in recent years, TCMI have encountered a major challenge in quality evaluation. Therefore, a reasonable control of the quality and safety of TCMI is an urgent problem to be solved in the development of TCM. At present, it is one of the commonly used methods to reflect the differences in the quality of TCM by establishing the fingerprints of related compounds [35]. In the study, unidimensional fingerprints of different kinds of components were applied to evaluate the quality of KDZI, but unfortunately each of them alone failed to reflect comprehensive KDZI’s quality and their differences among batches. By applying information fusion based on HPLC-UV, GC–MS and HPIEC technologies, multidimensional fingerprint of KDZI was established for comprehensive quality evaluation. Moreover, similarity-based cosine method, T-test, Bayes’ theorem and hierarchical clustering analysis were applied to analyze the differences after information fusion, which offered all-round perspective to distinguish the KDZI of different batches. Although the cosine failed to identify the differences between sample batches, the differences could still be significant by Bayes’ theorem, T-test and HCA. Analyzing methods including Bayes’ theorem and T-test with content information in KDZI and HCA with fusion information were more practical and more powerful to assess the quality consistency of KDZI. In conclusion, the establishment of multidimensional fingerprint and statistical analysis play an important role in the modernization of TCM development. This could provide an integrated method for consistency evaluation of TCM, which could be extended to production process for discovering the underlying differences among batches. Based on these results, process of TCM could be improved. In our following research, the integrated method would be extended to analyze botanical raw materials and production process for discovering the key process in manufacturing of TCMI which could lead to variability between different

Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.jpba.2019. 112868. References [1] W. Cai, J.Y. Zhang, L.Y. Dong, P.H. Yin, C.G. Wang, J.Q. Lu, H.G. Zhang, Identification of the metabolites of Ixerin Z from Ixeris sonchifolia Hance in rats by HPLC–LTQ-Orbitrap mass spectrometry, J. Pharm. Biomed. Anal. 107 (2015) 290–297. [2] C.G. Wang, C.Y. Liu, M. Wang, Q.T. Ma, Y.Q. Li, T. Wang, B.S. Zhao, UPLC-HRMS-Based plasma metabolomic profiling of novel biomarkers by treatment with KDZI in cerebral ischemia reperfusion rats, Molecules 23 (2018) 1315. [3] X.Z. Feng, S.X. Xu, M. Dong, A new sesquiterpene lactone glucoside from Ixeris sonchifolia, J. Asian Nat. Prod. Res. 3 (2001) 247–251. [4] Y.Q. Lu, Z.M. Zhang, Chemical constituents of Ixeris sonchifolia, J. Shenyang Pharm. Univ. 16 (2007) 214–217. [5] Y. Liu, J.Q. Lu, J.Y. Zhang, Q. Wang, F. Wang, Y.J. Qiao, Y.L. Zhang, Rapid determination of ten polyphenols in Kudiezi injection using ultra-performance liquid chromatography-tandem mass spectrometry in multiple reaction monitoring mode, Anal. Methods 4 (2012) 4230–4236. [6] Y. Gong, Y.L. Zhang, G.D. Hua, The clinical application of kudiezi injection, Medical Recapitulate 20 (2014) 100–102. [7] G. Shen, G. Huang, Analysis on use situation of traditional chinese medicine injections for promoting blood circulation and removing stasis in orthopedic department of our hospital, China Pharm. 23 (2014) 66–68. [8] Y.C. Zhang, F.F. Gan, S.B. Shelar, K.Y. Ng, E.H. Chew, Antioxidant and Nrf2 inducing activities of luteolin, a flavonoid constituent in Ixeris sonchifolia Hance, provide neuroprotective effects against ischemia-induced cellular injury, Food Chem. Toxicol. 59 (2013) 272–280. [9] S. Gao, J. Zhou, Advances in studies on clinical application of Kudiezi Injection, Drugs and Clinic 27 (2012) 180–184. [10] Y.Y. Xie, X. Xiao, J.M. Luo, C. Fu, Q.W. Wang, Y.M. Wang, Q.L. Liang, G.A. Luo, Integrating qualitative and quantitative characterization of traditional Chinese medicine injection by high-performance liquid chromatography with diode array detection and tandem mass spectrometry, J. Sep. Sci. 37 (2014) 1438–1447. [11] H.S. Xiong, L.X. Yu, H.B. Qu, Batch-to-Batch quality consistency evaluation of botanical drug products using multivariate statistical analysis of the chromatographic fingerprint, AAPS PharmSciTech 14 (2013) 802–810. [12] B. Yang, Y. Wang, L.L. Shan, J.T. Zou, Y.Y. Wu, F.F. Yang, Y.N. Zhang, Y.B. Li, Y.J. Zhang, A novel and practical chromatographic “Fingerprint-ROC-SVM” strategy applied to quality analysis of traditional Chinese medicine injections: using KuDieZi injection as a case study, Molecules 22 (2017) 1237.

8

H. Wang, M. Chen, J. Li et al. / Journal of Pharmaceutical and Biomedical Analysis 177 (2020) 112868

[13] State Drug Administration of China, Technical requirements for chromatographic fingerprints of traditional Chinese medicinal injection, Chin. Tradit. Pat. Med. 22 (2000) 671–678. [14] S. Yu, L. Zhu, Z.G. Xiao, J. Shen, J. Li, H.S. Lai, J.J. Li, H.B. Chen, Z.Z. Zhao, T. Yi, Rapid fingerprint analysis of flos carthami by ultra-performance liquid chromatography and similarity evaluation, J. Chromatog. Sci. 54 (2016) 1619–1624. [15] X.D. Zhou, L.Y. Tang, H.W. Wu, G.H. Zhou, T. Wang, Z.Z. Kou, S.X. Li, Z.J. Wang, Chemometric analyses for the characterization of raw and processed seeds of Descurainia sophia (L.) based on HPLC fingerprints, J. Pharm. Biomed. Anal. 111 (2015) 1–6. [16] T. Yi, Y.N. Tang, J.Y. Zhang, Z.Z. Zhao, Z.J. Yang, H.B. Chen, Characterization and determination of six flavonoids in the ethnomedicine “Dragon’s Blood” by UPLC-PAD-MS, Chem. Cent. J. 6 (2012) 116. [17] M.X. Luo, H.J. Luo, P.J. Hu, Y.T. Yang, B. Wu, G.D. Zheng, Evaluation of chemical components in Citri Reticulatae Pericarpium of different cultivars collected from different regions by GC–MS and HPLC, Food Sci. Nutr. 6 (2018) 400–416. [18] Y.D. Xue, L. Zhu, T. Yi, Fingerprint analysis of Resina Draconis by ultra-performance liquid chromatography, Chem. Cent. J. 11 (2017) 67. [19] R.M. Alonso-Salces, K. Heberger, M.V. Holland, J.M. Moreno-Rojas, C. Mariani, G. Bellan, F. Reniero, C. Guillou, Multivariate analysis of NMR fingerprint of the unsaponifiable fraction of virgin olive oils for authentication purposes, Food Chem. 118 (2010) 956–965. [20] C.D. Birk, G. Provensi, G. Gosmann, F.H. Reginatto, E.P. Schenkel, TLC fingerprint of flavonoids and saponins from Passiflora species, J. Liq. Chromatogr. Relat. Technol. 28 (2005) 2285–2291. [21] F. Gan, R.Y. Ye, New approach on similarity analysis of chromatographic fingerprint of herbal medicine, J. Chromatogr. A 1104 (2006) 100–105. [22] K.C. Shum, F. Chen, S.L. Li, J. Wang, P.P.H. But, P.C. Shaw, Authentication of Radix aucklandiae and its substitutes by GC-MS and hierarchical clustering analysis, J. Sep. Sci. 30 (2007) 3233–3239. [23] Y.S. Wu, S.D. Lv, C. Wang, X.M. Gao, J.B. Li, Q.X. Meng, Comparative analysis of volatiles difference of Yunnan sun-dried Pu-erh green tea from different tea mountains: jingmai and Wuliang mountain by chemical fingerprint similarity combined with principal component analysis and cluster analysis, Chem. Cent. J. 10 (2016) 11. [24] G. Qian, S. Sural, Y. Gu, S. Pramanik, Similarity between Euclidean and cosine angle distance for nearest neighbor queries, ACM Symposium on Applied Computing (2004) 1232–1237.

[25] H. Parastar, A. Bazrafshan, Fuzzy C-means clustering for chromatographic fingerprints analysis: a gas chromatography-mass spectrometry case study, J. Chromatogr. A 1438 (2016) 236–243. [26] Y.J. Lu, B.Y. Gao, P. Chen, D. Charles, L.L. Yu, Characterisation of organic and conventional sweet basil leaves using chromatographic and flow-injection mass spectrometric (FIMS) fingerprints combined with principal component analysis, Food Chem. 154 (2014) 262–268. [27] M. Becker, F. Liebner, T. Rosenau, A. Potthast, Ethoximation-silylation approach for mono- and disaccharide analysis and characterization of their identification parameters by GC/MS, Talanta 115 (2013) 642–651. [28] T. Ohra-aho, F.J.B. Gomes, J.L. Colodette, T. Tamminen, Carbohydrate composition in Eucalyptus wood and pulps – comparison between Py-GC/MS and acid hydrolysis, J. Anal. Appl. Pyrolysis 129 (2018) 215–220. [29] A.I. Ruiz-Matute, O. Hernandez-Hernandez, S. Rodriguez-Sanchez, M.L. Sanz, I. Martinez-Castro, Derivatization of carbohydrates for GC and GC-MS analyses, J. Chromatogr. B 879 (2011) 1226–1240. [30] X.N. Xia, H.M. Qu, S.J. Zeng, L.W. Yang, H.M. Liu, W.J. Xu, Investigation on the multiple chromatographic fingerprints of TCM based on fusion of carbonhydrates information, J. Hunan Univ. 35 (2008) 73–76. [31] H. Sun, X. Chen, A.H. Zhang, T. Sakurai, J.Z. Jiang, X.J. Wang, Chromatographic fingerprinting analysis of Zhizhu Wan preparation by high-performance liquid chromatography coupled with photodiode array detector, Pharmacogn. Mag. 10 (2014) 470–476. [32] H. Cai, Z.W. Xu, S.C. Luo, W.W. Zhang, G. Cao, X. Liu, Y.J. Lou, X.Q. Ma, K.M. Qin, B.C. Cai, Study on chemical fingerprinting of crude and processed Atractylodes macroce phalafrom different locations in Zhejiang province by reversed-phase high-performance liquid chromatography coupled with hierarchical cluster analysis, Pharmacogn. Mag. 8 (2012) 300–307. [33] B. Nisius, M. Vogt, J. Bajorath, Development of a fingerprint reduction approach for Bayesian similarity searching based on Kullback-Leibler divergence analysis, J. Chem. Inf. Model. 49 (2009) 1347–1358. [34] H.J. Dong, Z.J. Zhang, J. Yu, Y. Liu, F.G. Xu, Chemical fingerprinting of Andrographis paniculata (Burm. f.) Nees by HPLC and hierarchical clustering analysis, J. Chromatogr. Sci. 47 (2009) 931–935. [35] S. Yao, J.X. Zhang, D.D. Wang, J.J. Hou, W.Z. Yang, J. Da, L.Y. Cai, M. Yang, B.H. Jiang, X. Liu, D.A. Guo, W.Y. Wu, Discriminatory components retracing strategy for monitoring the preparation procedure of chinese patent medicines by fingerprint and chemometric analysis, PLoS One 10 (2015), e0121366.