Multivariate Statistical Analysis Uncovers Spectrum–Effect Relationship between HPLC Fingerprints and Antioxidant Activity of Saffron

Crocus sativus L. is commonly used as functional food and medicinal herb in traditional Chinese medicine. In this study, the spectrum–effect relationship was established betweenHPLC fingerprints and in vitro antioxidant activity of saffron to improve the quality evaluation method of saffron. 0e fingerprints of 21 batches of saffron collected from different regions were assessed, and the data were further analyzed by chemometric methods, including similarity analysis, hierarchical clustering analysis, principal component analysis, and orthogonal partial least squares discriminant analysis. 0e spectrum–effect relationship between fingerprints and antioxidant effect of saffron was analyzed by grey relational analysis and partial least square methods to figure out the antioxidant component of saffron. 0irteen common peaks of 21 batches of saffron were included in the analysis, and peak 3 (picrocrocin), peak 7 (crocin I), and peak 10 (crocin II) were identified as the main active components responsible for antioxidant efficacy. Besides, a multi-index quality control method was developed for simultaneous determination of these three antioxidant components in saffron. Taken together, this study provided new strategies for the quality control and the development of new bioactive products of saffron in the future.


Introduction
Saffron is an expensive spice derived from the stigma of the Crocus sativus L., which has been mainly cultivated in Iran, Greece, Morocco, India, Spain, and Italy. About 70,000 Crocus sativus L. flowers are required to produce 1 kg dry saffron, which is the main reason for its high cost. In addition to being used as food and dye, saffron has many therapeutic properties. Previous studies have demonstrated that saffron has the biological activities of cardiovascular protection, liver protection, antidepression, anticancer, and anti-inflammatory and can be potentially used as a hypoglycemia agent and immune enhancer [1][2][3][4][5][6][7]. e scarcity of resources and high cost of saffron lead to the frequent occurrence of saffron adulteration in the market, such as plant-derived materials like Zea mays L. (stigma), Chrysanthemum morifolium Ramat. (stigma), Carthamus tinctorius L. (stigma), corn silk, dyed corn stigma, turmeric, gardenia, rubia, calendula and artificial colorants like tartrazine, amaranth, sunset yellow, orange II, and new coccine [8][9][10][11][12][13]. e extracts of gardenia were added to saffron commonly because of the pigments in the extracts were similar to crocetin esters (crocins) present in saffron and thus could be concealed to a greater degree in the saffron [14,15].
Adulteration of saffron in the market brings about attention of the quality control of saffron. e international standard ISO 3632-2011 for grading saffron reports a standard UV-Vis spectrophotometric method, which tests the strength of aroma, color, and flavour of saffron by determining the concentrations of safranal, crocin, and picrocrocin [16]. However, recent studies indicated that there was no correlation between the content of safranal and the UV absorbance value at 330 nm (the maximum UV absorption wavelength of safranal) according to the international standard ISO 2631-2011 [17]. Additionally, the standard UV-Vis method of ISO was used for grading saffron and may not reveal saffron adulteration with amounts up to 20% (w/w) of safflower, turmeric, or calendula [17,18]. It can thus be seen that the grading of saffron depend on ISO method is not credible and cannot adequately distinguish between genuine and adulterated saffron.
Many published studies have focused on the quality standard of saffron, and various analytical techniques were applied to the quality control of saffron, such as high-performance liquid chromatography (HPLC), gas chromatography (GC), near-infrared spectroscopy (NIRS), ultra-highperformance liquid chromatography coupled to tandem mass spectrometry (UHPLC-MS/MS), and electronic nose (E-Nose) [19][20][21][22][23]. For the authentication of saffron, different analytical methods, including Raman spectroscopy, optical nanosensor, gas chromatography with mass spectrometry detection (GC-MS), microchip electrophoresis (MCE), headspace flash gas chromatography with flame ionization detection (HS-GC-FID), and nuclear magnetic resonance (NMR) spectroscopy, were assessed [16,[24][25][26][27]. ese techniques were conducive to the determination of different components of saffron, especially crocin. However, few studies focused on the correlation between quality evaluation and biological activities of saffron.
Spectrum-effect relationships have been widely used to screen the active compounds of TCMs by combining chromatographic fingerprint of TCMs with their biological activity. Chromatographic fingerprints of traditional Chinese medicines (TCMs) contain a large number of information and could express the chemical characteristics of samples integrally [28,29]. In recent years, spectrum-effect relationships were often being used to assess the quality control of TCMs [30]. For example, chlorogenic acid and 3,4-dicaffeoylquinic acid from Lonicerae Japonicae Flos and Lonicerae Flos were selected as the major antibacterial components by spectrum-effect relationships [31]. Menthone, isomenthone, pulegone, piperitone, and β-caryophyllene were identified as the dominant constituents responsible for the antioxidant and anti-inflammatory activities of S. tenuifolia essential oil [32]. In Chinese Pharmacopoeia (2020 edition), the contents of picrocrocin were newly used as the quality control standard for saffron, coupled with the content of crocins I and II. However, the relationship between these components and bioactivity requires further investigation to improve the quality control standard of saffron.
In this study, HPLC was used to establish the fingerprints of 21 batches of saffron. en, similarity analysis (SA), hierarchical clustering analysis (HCA), principal component analysis (PCA), and orthogonal partial least squares discriminant analysis (OPLS-DA) were applied to distinguish differences among the 21 batches of saffron. Subsequently, the antioxidant activity was evaluated by 1,1-diphenyl-2picrylhydrazyl (DPPH) radical-scavenging assay and hydroxyl (•OH) radical-scavenging assay. e spectrumeffect relationship between HPLC fingerprints and antioxidant activities were elucidated by grey relational analysis (GRA) and partial least square (PLS) analysis. e potential active compounds of saffron were discovered. Finally, a quantitative method for the determination of the potential active compounds, crocin I, crocin II, and picrocrocin, was developed.

Materials and Reagents.
Twenty-one batches of saffron were collected from different regions, as shown in Table 1.
ese were obtained directly from the producers and packers with a guarantee of their origin and freedom from fraud. All samples were authenticated as plant of Crocus sativus L., and voucher specimens were deposited in the herbarium of Zhejiang University of Technology.

Preparation of Samples and Reference Substance Solutions.
All the saffron samples used in this study were dried samples. e sample drying method was in accordance with the low temperature drying (<60°C) stipulated in technical regulation for production of saffron crocus (Crocus sativus L.) (NO. DB 33/T 530-2014, 2014 Version). e sample solutions were prepared according to the method of ISO 3632-2011. 50 mg of grounded saffron was accurately weighed and quantitatively transferred to a 50 mL volumetric flask and diluted to scale with 40 mL 50% ethanol. Ultrasonic was carried out at 50°C for 10 min and kept away from light. en, 50% ethanol was added to a constant volume and tightened. e volumetric flask was shaken evenly. e sample solution was filtered through 0.45-μm filter membrane before use.
All reference substances were weighed to obtain 1 mg/ mL picrocrocin, crocin I, and crocin II stock solutions. An appropriate amount of stock solution was taken and diluted to an appropriate concentration with 50% ethanol. e solution was filtered using 0.45-μm filter membrane before sample injection.

Optimization of HPLC Conditions.
e preparation methods of the sample were first studied according to the stirring method in ISO-3632-2011 and the ultrasonic extraction method in our previous study [19,33]. e results showed that there was no significant difference between the two extraction methods. Considering the stability of sample solution and the convenience of experiment, the ultrasonic extraction method was selected to prepare sample solution. To obtain excellent HPLC fingerprint, various parameters including extraction solvent (water, 25%, 50%, 75%, and 100% of methanol, 25%, 50%, 75%, 100% of ethanol, v/v), extraction time (10 min, 20 min, 30 min, 40 min, 50 min, 60 min), light (on light or away from light), and extraction solid-liquid ratio (1 : 0.5, 1 : 1, 1 : 5, 1 : 20, w/v) were optimized. e results showed that 50% of ethanol and 1 : 1 of solid-liquid ratio provided better extraction efficiency and chromatographic separation ( Figure 1). Furthermore, dual wavelengths (257 nm and 440 nm) have been proved to provide more detection peaks, stronger UV absorption, and better peak shape, so these two wavelengths were selected as the detection wavelengths in the HPLC analysis of fingerprint.

HPLC Fingerprint Analysis and Similarity Analysis.
Recently, fingerprints combined with multivariate statistical analysis have been used to classify and discriminate different TCMs sources successfully [34]. In the current study, the spectrum of sample S4 was used as the reference spectrum, and the representative HPLC fingerprints of 21 batches of saffron were established as shown in Figures 2(a) and 2(b). e generated representative fingerprints were showed in Figures 2(c) and 2(d). e RT and PA information of 13 common peaks in the fingerprint were extracted, and the RT and characteristic absorption wavelength of the standard reference substances were compared. e following results were obtained: peak 3 was picrocrocin, peak 7 was crocin I,  peak 10 was crocin II, peak 12 was crocin III, and peak 13 was cis-crocin I. In this study, the similarity calculation was carried out by Similarity Evaluation System for Chromatographic Fingerprint of Traditional Chinese Medicine software (Version 2012) issued by Chinese Pharmacopoeia Committee. Time window width was set as 0.1, average mode was used, and then, all the samples had multipoint correction and automatic match to generate a representative fingerprint that represented the characteristic mode. e similarity was calculated by comparing chromatograms of saffron with the representative fingerprints [35,36]. e results showed that the similarity of 21 3  200  190  180  170  160  150  140  130  120  110  100  90  80  70  60  50  40  30  20 10 0   0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 (c)  560  540  520  500  480  460  440  420  400  380  360  340  320  300  280  260  240  220  200  180  160  140  120  100  80  60  40 20 0 (d) 0.998, and 1, respectively. e high similarity (≥0.962) indicated that the quality of saffron was generally stable and the established fingerprint method could be used for the identification and quality control of saffron.

Method Validation for the HPLC Fingerprint Analysis.
e method validation results of the HPLC fingerprint showed that the relative standard deviations (RSDs) of precision, repeatability, and stability of the retention time (RT) and the peak area (PA) met the prescribed requirements. e variations in the RT of the characteristic peaks were less than 0.5%, and the variations in the PA are less than 3.0% (n � 6).

Method Validation of Quantitative Analysis.
e method of quantitative analysis was validated in terms of linearity of calibration curves, precision, stability, repeatability, and recovery. Taking the sample concentration as the abscissa (x, mg/L) and PA as the ordinate (y), the linear regression analysis was carried out. e regression equation, R 2 , linear range, and standard errors were shown in Table 2.
e results indicated that the linear relationship obtained for each target compound was reliable, and the obtained calibration curves were suitable for HPLC analysis. e precision, stability, and repeatability were assessed by the PA of P3, P7, and P10, respectively. e RSDs of precision of crocin I, crocin II, and picrocrocin were 1.11%, 0.10%, and 0.10%, respectively, the RSDs of stability were 1.31%, 0.47%, and 0.21%, respectively, and the RSDs of repeatability were 1.60%, 1.26%, and 0.21%, respectively. Recovery was measured by the standard addition method. e sample (S4) was added with high, medium, and low levels of a mixed standard solution of the three compounds in triplicate. e average recovery rates of crocin I;, crocin II; , and picrocrocin were 99.58%, 98.18%, and 100.04%, respectively. All the results of the method validation tests demonstrated that the proposed method was reliable and valid.

HCA.
In this study, SPSS statistical software (Version 24.0) was used to perform HCA on the fingerprint of 21 batches of saffron, using the square Euclidean distance as the interval and using the intragroup linkage method, as shown in Figure 3. According to the results of HCA, 21 batches of saffron could be divided into two categories: S5, S6, S8, S9, S10, S11, S19, and S20 were in one category, while S1-S4, S7, S12-S18, and S21 were in the other category. In order to evaluate the difference between the two categories of saffron, PCA and OPLS-DA were used to the further analysis.

PCA and OPLS-DA.
PCA is a multivariate statistical method which converts multiple variables into a few unrelated comprehensive variables. e purpose of PCA is to remove overlapping information among numerous of information by dimensionality reduction [37]. e PA of 13 common peaks was inputted as the variables. en, the PCA scoring map was obtained by SIMICA-P software, which divided saffron into two categories: S2, S3, S5, S6, S8, S9, S10, S11, S19, and S20 were assigned to one category, while S1, S4, S7, S12-S18, and S21 were in another category (Figure 4(a)). SPSS was used for PCA analysis to obtain principal components (PCs). According to the principle of screening PCs in PCA, the first three PCs were identified with eigenvalue λ > 1, of which the cumulative contribution rate was 77.515% (Table 3). ereby, the three PCs can be used for the evaluation of saffron quality [30,38]. Among the three PCs, the cumulative contribution rate of the PC1 was 47.932%, which represented the highest influence to the quality control of saffron [39].
is information primarily originated from P2, P3, and P6-P11, while P4, P9, P12, and P13 showed the high loading values to the PC2. P2, P5, and P13 showed the high loading value to the PC3 (Table 4). ese results showed that the quality difference of saffron was affected by various components, but not by a single one. e predictive ability parameter Q 2 was 0.326. Based on previous study, the range of Q 2 from 0.3 to 0.4 indicated that the model had poor predictive power, which might be caused by the within-class divergence of individual groups [40]. It was with these considerations in mind that the OPLS-DA was further performed to extend a regression of the PCA because OPLS-DA had better discriminant ability for the samples with larger within-class divergence than PCA [41].
In OPLS-DA, the corresponding model was obtained by using the PA of 13 common peaks of saffron as input variable. e results were shown in Figures 4(b) and 4(c). In the OPLS-DA model, the cumulative explanatory ability parameters R 2 X and R 2 Y were 0.944 and 0.781, the predictive ability parameters Q 2 were 0.726, and R 2 and Q 2 were all greater than 0.5, which indicated that the model had certain stability and reliability and could be used to evaluate and distinguish different batches of saffron [42]. According to the OPLS-DA score, saffron from different origins were divided into two categories, which was similar to the results of HCA and PCA analysis. Statistical significance markers were selected according to the VIP predicted value in the model. Within the confidence interval of 0.95, VIP >1.0 was selected as the  -4 -2 S1 S4 S13 S16 S21 S12 S14 S7 S18 S15 S17 S19 S3 S2 S20 S10 S11 S8 S5
Permutation test, a computer-based resampling method for remodeling and predicting, was widely used in the computation of variable importance and confidence intervals [43,44]. It could be considered that the model had not been overfitted when the Y-axis intercept of R 2 and Q 2 for the established OPLS-DA models was less than 0.3 and 0.05, respectively [45][46][47]. After 200 permutations, the intercepts of R 2 and Q 2 were 0.143 and -0.429, respectively, which indicated that the established OPLS-DA model was reliable and not overfitted ( Figure 5).
In Iranian pharmacopoeia and European pharmacopoeia, the quality control of saffron does not include quantitative analysis of chemical composition, while Japanese pharmacopoeia and Korean pharmacopoeia list the sum of crocin I and crocin II for the quality control of saffron for only qualitative analysis. Many literatures have shown that the crocins play a critical role in quality control of saffron. Additionally, a variety of components, including crocins, picrocrocin, crocetin, and safranal, show unique pharmacological activity [10,48,49]. To sum up, the picrocrocin, crocin I, and crocin II were identified as the quality control of saffron in this study.

Quantitative Analysis of 21 Batches of Saffron.
e proposed HPLC method was successfully used for simultaneous determination of crocin I, crocin II, and picrocrocin. As shown in Figure 6(a), the highest content of crocin I was from Henan (204.97 mg/g), the lowest origin was from Zhejiang (60.06 mg/g), and the median origin was from Xizang (146.80 mg/g). e highest content of crocin II was from Henan (74.11 mg/g) and the lowest origin was from Anhui (20.10 mg/g), and the median origin was from Zhejiang (44.65 mg/g). e highest content of picrocrocin was from Shanghai (187.53 mg/g), the lowest origin was from Zhejiang (57.98 mg/g), and the median origin was from Zhejiang (110.02 mg/g).
Based on the contents of the three major components, the 21 batches of saffron could be divided into two categories by HCA (Figure 6(b)): category 1 included two batches from Henan, one batch from Shanghai, three batches from Iran, one batch from Zhejiang, three batches from Xizang, and one batch from Anhui; category 2 included three batches from Zhejiang, two batches from Iran, four batches from Anhui, and one batch from Dubai.
As the results of quantitative analysis show, the contents of three major components of four batches of saffron from Zhejiang province were significantly different from each other. However, one batch of saffron from Zhejiang province was very similar to those from Tibet and Iran et al., indicating that the differences between samples had little correlation with origin. To sum up, origin was not the key factor   e difference on quality may be caused by cultivation methods, nutrition and quality of bulb, growth environment, climate, and other factors.

Antioxidant Activities of 21 Batches of Saffron.
Oxidation is an important process for the energy productive of many biological organs [50]. Studies have shown that cancer, inflammation, blood diseases, and other diseases are closely related to oxidative free radicals, and excessive free radicals in the human body will lead to aging [51][52][53]. Commonly, antioxidant capacity can be assessed by in vivo and in vitro methods [54]. e in vitro methods were widely used for their advantages of simple operation, stable results, and short period. Multiple in vitro antioxidant methods have been reported in previous studies, including DPPH method, •OH radical method, 2,2′-azinobis(3-ethylbenzothiazoline-6-sulfonic) acid ammonium salt (ABTS) method, and ferric reducing antioxidant potential assay (FRAP) method [55]. Because many studies focused on the pharmacological of saffron extracts (ethanolic and aqueous extracts) and saffron components (crocin and safranal), there were few literatures pertaining to antioxidant activity of saffron in vitro [56,57]. In this study, DPPH method and •OH radical method were used to evaluate the antioxidant capacity of saffron. e antioxidant results in Figure 6 revealed that 21 batches of saffron exhibited a concentration dependence relationship with DPPH and •OH radical-scavenging activity.
As shown in Figure 7(c), two curves of IC 50 values of DPPH and •OH scavenging activity represented the same trend, which showed the results of antioxidant are credible.

PLS of Antioxidant Spectrum-Effect
Relationship. e PA of 13 common peaks of 21 batches of saffron was inputted as the independent variable, while the IC50 of DPPH and •OH radical-scavenging capacity were inputted as the dependent variable, and the partial least squares method was used to analyze the variables. e correlation coefficient and variable projection importance value (VIP) were obtained to evaluate the correlation between 13 chromatographic peaks and drug efficacy and their contribution to drug efficacy. It is generally believed that when VIP>1, the independent variable has significant importance on explaining the dependent variable. In this study, the VIP value of each chromatographic peak of DPPH radical-scavenging capacity is ranked as follows: (Figures 8(a) and 8(b)). e VIP value of each chromatographic peak of •OH radical-scavenging capacity is ranked as follows: (Figures 8(c) and 8(d)). It can be seen that the VIP values of P7 (crocin I), P10 (crocin II), and P3 (picrocrocin) were higher than 1, indicating that these compounds could be the core components of antioxidant activity in saffron.
Limited relevant research showed that polysaccharide and ethanol extracts of saffron from seven different productions had remarkable antioxidant activities [58]. Further research showed that both stigmas and flowers had antioxidant capacity, due to the apocarotenoids, flavonoids, and flavonols presented in the stigmas and the flavonoids, antocyanins, and tannins abundant in the remaining portions of C. sativus flower. However, the exact mechanisms and compounds responsible for the antioxidant activity of saffron were still not clear. Based on the results in this study, the grey correlation  degree and OPLS-DA of antioxidant spectrum-effect relationship showed crocin I, crocin II, and picrocrocin had large contribution coefficients, closely related to antioxidant ability. Many studies have shown that water-soluble carotenoids show excellent antioxidant potential, which was related to the conjugated large π bond in the molecular structure of carotenoids and easily react with free radicals to form harmless products or scavenge-free radicals by destroying free radical chains. Crocins ( Figure 9) are a group of water soluble-carotenoids, produced by zeaxanthin through enzymatic cleavage [59]. Two of the most abundant crocin were crocin I (trans-crocetin di-(β-d-gentiobiosyl) ester) and crocin II (crocetin-(β-d-glucosyl)-(β-gentibiosyl) ester) [60]. It can be inferred that the antioxidant capacity of these two components is closely related to their chemical structures and higher content in saffron [58]. In this study, picrocrocin was demonstrated to be directly related to antioxidant activity for the first time. Crocin I, crocin II, and picrocrocin accounted for a large proportion of contents in saffron and presented the good antioxidant ability, which made saffron a good candidate for an antioxidant food. Previous studies have focused on mainly the analysis of its chemical composition or only bioactivities, limiting the development and utilization of saffron as a commercial medicine [61,62]. In this study, the correlation between chemical components and pharmacological activity of saffron were analyzed by spectrum-effect relationship. Crocin I, crocin II, and picrocrocin were identified to be closely related to antioxidant capacity, which were conducive to the establishment of a more reliable and scientific quality standard for saffron. However, the drawback of the current study was that only 21 batches of samples were used in this study, which was not conducive to establishing a firm correlation. In order to confirm the findings of this research, a larger sample size should be considered in the future.

Conclusion
e spectrum-effect relationships of HPLC fingerprints and scavenging capacity for DPPH and •OH were established to analyze the active components of saffron.
e spectrumeffect relationships on the basis of grey correlation degree and OPLS-DA analysis revealed that crocin I, crocin II, and picrocrocin were the main components contributing to the antioxidant activities of saffron and these compounds had synergistic antioxidant effects. rough this study, the main antioxidant components of saffron were further determined, which could provide clue for establishing reliable and reasonable quality standards for saffron and its products.

Data Availability
e data used to support the findings of this study are included within the article.

Disclosure
Ya You and Zijin Xu are co-first authors.

Conflicts of Interest
e authors have no conflicts of interest to declare.

Authors' Contributions
Ya You, Zijin Xu, Suhong Chen, and Ping Wang contributed equally to this work. Ping Wang, Suhong Chen, and Yifeng Cao conceived of and designed the experiments; Ya You, Zijin Xu, Qingrou Zhong, and Lin Zhu performed the experiments; YaYou, Yi Tao, Susu Lin, and Qiaoqiao Li analyzed the data; Ya You wrote the paper.

12
Journal of Chemistry