Analysis of bioactive constituents of saffron using ultrasonic assisted emulsification microextraction combined with high-performance liquid chromatography with diode array detector: a chemometric study

Abstract

In recent years, there has been an increasing interest in the analysis of the major active components of saffron owing to their significant role in various industries, such as food, medicine and perfume. In other words, analysis of the major active components of saffron can give complete information about its chemical composition, which can be used as a reliable index for the quality control of different saffron samples (i.e., natural and commercial). The aim of the present work was to develop a simple, low cost, efficient and comprehensive strategy for the extraction and analysis of the bioactive components of saffron. In this regard, ultrasonic-assisted solvent extraction (UASE) combined with ultrasonic-assisted emulsification microextraction (USAEME) is proposed for the extraction and preconcentration of the bioactive constituents of saffron. The extracted components are then analyzed using reverse-phase high-performance liquid chromatography with diode array detector (RP-HPLC-DAD). The effective parameters on the efficiency of extraction procedure are optimized using multivariate chemometric techniques. As a consequence, the optimum extraction parameters were 79.6 mg of a saffron sample, 1.1 mL of the extraction solvent (water), 62.7 μL of the preconcentration solvent (chloroform) and 18.6 min sonication time. In optimum extraction conditions, the relative standard deviations (RSDs) were below 1.0% (n = 3) for all the components. Moreover, the enrichment factors were higher than 10 for most components. Finally, the developed analytical method is used as a reliable method for the quality control of fifteen commercial saffron samples purchased from different markets. To do this, two multivariate clustering methods, principal component analysis (PCA) and k-means, are used for determining similarities and dissimilarities between standard and commercial saffron samples, according to their HPLC fingerprints. It is concluded that the proposed method is a fast, simple, accurate and unbiased method for analyzing the bioactive components of saffron and fingerprinting commercial saffron samples, obtaining a more complete set of information from data than conventional methods.

---

1. Introduction

Saffron, as the most expensive spice in the world, is the dried stigmas of Crocus sativus L. flowers. Crocus is a genus of the Iridaceae family,¹ and it is widely cultivated in Iran and other countries such as Spain, India, China, Greece, Morocco, Turkey, Italy and Azerbaijan.² Saffron stigmas contain essential compounds, such as vitamins, carbohydrates, minerals and different pigments such as flavonoids and carotenes.³ Saffron is appreciated by consumers as a colorant for foodstuffs, as well as for its aromatic and flavoring properties.^4,5 More recently, there has been an increasing interest in the biological effects of saffron constituents and their possible medical applications, particularly those based on their cytotoxic, anti-carcinogenic and anti-tumour properties.^6,7 The value of saffron is determined by the presence of three main secondary metabolites;⁸ crocin and its derivatives as hydrophilic carotenoids, which are responsible for the characteristic golden yellow color of saffron,^9,10 picrocrocin as a monoterpene glycoside, which is responsible for the bitter taste of saffron,^11,12 and safranal as a monoterpene aldehyde, which is responsible for saffron's aroma. However, from a medicinal point of view, the major biologically active components of saffron are crocin analogues. The amount of these compounds in dried stigma tissues is the most important indicator of the quality of saffron.¹

Different extraction techniques have been proposed for the extraction and preconcentration of volatile and non-volatile components of saffron. Hydrodistillation (HD), vacuum headspace (VHS),¹² supercritical fluid extraction (SFE),^13,14 thermal desorption (TD),¹⁵ extraction with organic solvents,^16,17 solid-phase microextraction (SPME)¹⁸ and ultrasonic-assisted solvent extraction (UASE)¹⁹ have been used for the extraction of the chemical components of saffron. Amongst the above mentioned methods, UASE has some advantages due to the application of ultrasonic waves for extraction. Ultrasound waves pass through a medium by creating compression and expansion, which create bubbles in a liquid and produce a negative pressure. This process produces a phenomenon called cavitation, which indicates the formation, growth and collapse of bubbles. Therefore, a large amount of energy can be produced from the conversion of kinetic energy into thermal energy to heat the contents of the bubble. As a consequence, the use of ultrasound waves in extraction can disrupt biological cell walls, facilitating the release of contents. Thus, an efficient cell disruption and effective mass transfer are cited as two major factors leading to the enhancement of extraction with ultrasonic power. Coupling the extraction process with a preconcentration method, such as ultrasound-assisted emulsification microextraction (USAEME),²⁰ enhances the efficiency of the method. The USAEME method is a fast and simple method with high efficiency, recovery and enrichment factor.^20,21

In recent years, most of the studies on the chemical composition of saffron have been focused on the non-volatile components of saffron^1,3,8 owing to their medical applications. Optimizations of extraction procedures are usually performed using the one-variable-at-a-time (OVAT) approach, which facilitates the interpretation of the obtained results, but interactions between variables are not taken into account.^22,23 Therefore, a false minimum or maximum may be attained, which is not the best analytical response. Experimental design methods (e.g., factorial designs and response surface methodology) have been frequently applied to optimize the extraction procedures.^22,24 In this approach, the main effects of the factors, their interactions and curvatures are estimated. The curvature in the response surface means curvature in the relationship between factors as laid out in the model. These terms are quadratic terms in the developed model. This is one of the greatest advantages of multivariate optimization compared to OVAT optimization. Another advantage is that the number of experiments required is considerably reduced, particularly when there are many factors.^22,24

High-performance liquid chromatography with a diode array detector (HPLC-DAD) is one of the best techniques for the separation and identification of the non-volatile and thermally labile components of saffron. However, HPLC coupled with tandem mass spectrometry (LC-MS/MS) is a better technique for the separation and identification of the bioactive components of saffron but it is more expensive than HPLC-DAD and providing access to it is not as simple as that for HPLC-DAD. Nowadays, there is an increasing interest in the identification of the chemical compositions of complex samples by their chromatographic signals, also known as chromatographic fingerprints. A chromatographic fingerprint is a unique pattern that indicates the presence of chemical components in the analyzed sample. Chromatographic fingerprinting has become one of the most powerful approaches for the quality control of complex natural samples, such as herbal medicines, and represents a comprehensive qualitative approach for the purpose of species authentication, evaluation of quality and ensuring the consistency and stability of the chemical constituents observed by chromatography.^25,26

The aim of the present work was to offer a simple, low cost, efficient and environmentally-friendly technique for the extraction, preconcentration and chromatographic analysis of the bioactive components of saffron with the aid of chemometric techniques. For this purpose, a two-step extraction process consisting of UASE followed by USAEME is proposed. The first step includes the direct extraction of saffron components from solid stigmas into water, as a suitable solvent, accelerated by ultrasound waves. In the second step, the USAEME method is used for the preconcentration of the isolated components. The important parameters of the UASE-USAEME method, including saffron sample, extraction solvent, preconcentration solvent and sonication time, are optimized using response surface methodology (RSM). Finally, the developed method is used for the analysis of commercial saffron samples and obtaining their LC fingerprints to control the quality of different saffron samples. The similarities and dissimilarities among samples are determined using multivariate clustering techniques of principal component analysis (PCA)^27,28 and k-means.²⁹

2. Experimental

2.1. Saffron samples and chemicals

Fresh stigmas of a standard saffron sample were obtained from the area of Qaen in the South Khorasan province of Iran, where they were cultivated. The stigmas were dried at a temperature between 20 and 30 °C in the absence of light for 24 h. The dried stigmas were maintained at 4 °C in the absence of light until their analysis.

HPLC grade methanol, acetonitrile and analytical grade chloroform were purchased from Merck (Darmstadt, Germany). Deionized water was purified by a Milli-Q system from Millipore.

2.2. Extraction procedure

The procedure of extraction and preconcentration of saffron components was as follows: a ground saffron sample was precisely weighed and placed in a round-end test tube containing water before adding extraction solvent. The mixture was subsequently exposed to ultrasonic waves. During this time, the temperature of the ultrasonic bath was maintained at 25 °C. To separate the solid remains of the saffron sample, the mixture was centrifuged. Then, chloroform, as the preconcentration solvent, was added to the upper phase solution and the mixture was sonicated. Accordingly, the solution became cloudy, due to the distribution of tiny drops of chloroform in the sample solution. In this situation, the extraction process was accomplished. Then, the cloudy solution was centrifuged to separate the chloroform as the lower phase. The extraction was performed at ambient temperature and in the absence of direct exposure to light in order to protect the light-sensitive components. Chloroform was dried using a gentle flow of nitrogen, and then methanol was added to it. Finally, 20 μL of extract was analyzed using a RP-HPLC-DAD. In general, 30 experiments were performed with 6 replicates. The relative standard deviations (RSD, %) of the replicates were below 1.0%, which was acceptable.

2.3. Chromatographic conditions and instrumentation

Chromatographic analyses were performed using a Knauer HPLC system, equipped with a Smartline pump 1000 and a diode array absorbance detector (DAD) system. Analyses were carried out on a C18 column (150 mm × 4.6 mm i.d., 5 μm particle diameter, MN, Germany) under the following conditions: flow rate was 1 mL min⁻¹ and injection volume was 20 μL with the mobile phase consisting of water (A) and acetonitrile (B). The gradient elution program was as follows: 20–40% B for 0–3 min, 40–50% B for 3–8 min, 50–50% B for 8–12 min, 50–80% B for 12–15 min. The UV spectra were recorded between 200 and 499 nm. It should be noted that each extract was filtered through a syringe filter (0.22 μm) prior to injection into HPLC-DAD. Moreover, no significant effects of the adsorption of the extracted components of saffron on the filter were observed in this work.

2.4. Software requirements

ChromGate v. 3.3.2 was used for HPLC-DAD data collection, exportation and conversion to ASCII format. The statistical computer package “Design-Expert v. 7.1.3” (Stat-Ease Inc., Minneapolis) was used for designing the experiments, developing the model and optimizing. The PLS Toolbox version 3.5 was used for multivariate cluster analysis. All the calculations were performed in MATLAB v. 8.3 (Mathworks, Natick, MA, USA). The MATLAB codes for baseline correction and elution time shift correction were downloaded from the internet (http://www.models.kvl.dk/algorithms).

3. Results and discussion

3.1. Multivariate optimization of UASE-USAEME procedure

In the first step of the present study, effective parameters on the UASE-USAEME method, including type and volume of extraction solvent, type and volume of preconcentration solvent, amount of sample and sonication time, were optimized using a multivariate strategy. In this regard, the sum of the peak areas of all the detectable constituents in HPLC-DAD was chosen as the response for modeling. At first, the type of extraction and preconcentration solvents were chosen using a univariate strategy.

The extraction solvents for UASE were chosen on the basis of the number of components that can be extracted from saffron and the peak area of these components in the HPLC-DAD chromatogram. In this regard, solvent polarity, molecular weight and viscosity were considered as important solvent parameters in the extraction. Taking these properties into consideration, several solvents such as methanol as a protic solvent (MW = 32.04 g mol⁻¹, d = 0.7918 g cm⁻³, dipole moment = 1.69 D), water (MW = 18.01 g mol⁻¹, d = 0.9999 g cm⁻³, dipole moment = 1.85 D) and acetonitrile as an aprotic solvent (MW = 41.05 g mol⁻¹, d = 0.7860 g cm⁻³, dipole moment = 3.92 D) were tested. Among these solvents, water showed the highest efficiency in terms of peak areas and number of peaks (i.e., 15 peaks). Therefore, it was selected as the extraction solvent.

On the other hand, USAEME requires a high-density preconcentration solvent, which was chosen on the basis of immiscibility in water and solubility of the target analytes. Therefore, 100 μL of aprotic organic solvents, such as ethyl acetate (C₄H₈O₂) (d = 0.897 g cm⁻³), carbon tetrachloride (CCl₄) (d = 1.587 g cm⁻³), chloroform (CHCl₃) (d = 1.489 g cm⁻³) and dichloromethane (CH₂Cl₂) (d = 1.330 g cm⁻³) were individually tested. Inspection of the results showed that chloroform is the most effective solvent in the preconcentration of the analytes (the enrichment factor was higher than 10); therefore, it was chosen as the USAEME solvent in this work.

Fig. 1 depicts the HPLC chromatogram of the extracted saffron components after UASE (a) (60 mg saffron sample, 2.0 mL H₂O as extraction solvent and 20 min sonication time), and after extraction and preconcentration by UASE-USAEME (b) (60 mg saffron sample, 2.0 mL H₂O as extraction solvent, 20 min sonication time and 100 μL chloroform as preconcentration solvent). It should be noted that HPLC analyses were carried out on a C18 column (150 mm × 4.6 mm i.d., 5 μm) under the mentioned conditions in Section 2.3. It is clear that USAEME can enrich most of the extracted components in the UASE step.

Fig. 1 HPLC chromatogram of the extracted saffron components (a) after UASE and (b) after extraction and preconcentration by UASE-USAEME.

After finding proper solvents for the extraction and preconcentration of saffron components, the optimization of other effective parameters on the UASE-USAEME procedure, including extraction solvent volume, preconcentration solvent volume, sonication time and sample amount, was performed using response surface methodology (RSM).

In order to achieve the highest practical method performance and to obtain the conditions that would allow the procedure to generate the best response, a rotatable central composite design (CCD)³⁰ was used. In this study, a rotatable CCD with α = 2.00 was used for the optimization of the effective factors on UASE-USAEME, for the characterization of non-volatile components in saffron. Table 1 shows the effective factors, their abbreviations and their levels for the rotatable CCD. Moreover, Table S1 (ESI†) demonstrates the CCD design matrix and obtained a response for each run. It is important to note that the low and high levels of each factor were determined according to literature data and preliminary studies.^1,3,31

Table 1 Factors, their notations, and their levels in central composite design (CCD)

Factor	Symbol	Level
		−α	−1	0	+1	+α
Preconcentration solvent volume (μL)	A	20	40	60	80	100
Sample amount (mg)	B	10	30	50	70	90
Sonication time (min)	C	5	10	15	20	25
Extraction solvent volume (mL)	D	1	2	3	4	5

---

A step-wise multiple linear regression (MLR) was used to select a suitable response surface model. To evaluate the model and the significance of the effects, the analysis of variance (ANOVA) was used. Table 2 shows the ANOVA table for the CCD design matrix. The F-values indicate that the proposed model is important and the lack of fit is not significant relative to the pure error.

Table 2 Analysis of variance (ANOVA) table of the quadratic response surface model

Source	SS	d.f.	MS	F-Value	Prob > F
Model	2.844 × 10^20	10	2.844 × 10^19	13.12	<0.0001	Significant
A	7.640 × 10^17	1	7.640 × 10^17	0.35	0.5598
B	4.575 × 10^19	1	4.575 × 10^19	21.10	0.0002
C	9.277 × 10^17	1	9.277 × 10^17	0.43	0.5208
D	1.796 × 10^20	1	1.796 × 10^20	82.86	<0.0001
AD	1.189 × 10^19	1	1.189 × 10^19	5.49	0.0302
BD	7.522 × 10^18	1	7.522 × 10^18	3.47	0.0780
A^2	6.821 × 10^18	1	6.821 × 10^18	3.15	0.0921
B^2	1.853 × 10^19	1	1.853 × 10^19	8.55	0.0087
C^2	8.422 × 10^18	1	8.422 × 10^18	3.88	0.0635
D^2	1.872 × 10^19	1	1.872 × 10^19	8.63	0.0084
Residual	4.119 × 10^19	19	2.168 × 10^18
Lack of fit	3.759 × 10^19	14	2.685 × 10^18	3.73	0.0768	Not significant
Pure error	3.596 × 10^18	5	7.192 × 10^17
Corrected total	3.256 × 10^20	29

---

After analyzing the data, a quadratic response surface model based on higher F- and R-values and a lower lack of fit (LOF) to fit the experimental data was selected. This model, which consists of four main effects, a couple of two factor interactions and four curvature effects is shown in eqn (1) in a coded form:

Y = 4.7 × 10⁹ − 1.8 × 10⁸A + 1.4 × 10⁹B + 1.9 × 10⁸C − 2.7 × 10⁹D + 8.6 × 10⁸AD − 6.8 × 10⁸BD + 4.9 × 10⁸A² + 8.2 × 10⁸B² + 5.5 × 10⁸C² + 8.3 × 10⁸D² (1)

The p-value of 0.077 for the LOF indicates that it is not significant relative to the pure experimental error and confirms the validity of the model. Other statistical parameters of the model are shown in Table 3.

Table 3 Statistical parameters for the quadratic model in eqn (1)

Parameter	Value	Parameter	Value
Mean	6.88 × 10^9	RAdj^2	0.807
C.V.%	21.40	RPred^2	0.642
R^2	0.874	Adequate precision	12.27

---

R-square, which is a measure of the amount of variation around the mean explained by the model, was 0.873 for this model. Another important parameter for evaluating the model is the adjusted R-square (R_Adj²). This parameter is considered as a measure of the amount of variation around the mean explained by the model, adjusted for the number of terms in the model. In other words, the R_Adj² decreases as the number of terms in the model increases. In addition, the predicted R-square (R_Pred²), which is a measure of the amount of variation in the new data explained by the model, can be applied for the evaluation of the model. The R_Pred² and the R_Adj² values for the above model were 0.806 and 0.641, respectively. The term “adequate precision” in Table 3 represents the signal-to-noise (S/N) ratio. A ratio greater than 4.0 indicates that the model is adequate.³² For the proposed model, this value is 12.27 and indicates a very good signal-to-noise ratio. All of these statistical parameters show the reliability of the model. After obtaining the desired model and the statistical evaluation of it, confirming the absence of outliers in the data is very important. In this regard, two frequently used methods of leverage and Cook's distance were used.^24,33

Fig. 2(a) shows the leverage plot for the quadratic model obtained from central composite design. It can be seen from this figure that all of the leverage values are lower than 0.75 (the threshold value) and it can be concluded that there are no outliers or unexpected errors in the model. This result has been confirmed with the Cook's distance plot in Fig. 2(b) where all runs are in the confidence interval and there are no outliers in the model.

Fig. 2 (a) Leverage plot and (b) Cook's distance plot for quadratic model obtained from central composite design. Red lines in (a) and (b) show the threshold level which is calculated using statistical tests. An experiment with values greater than the limit values is generally regarded as an outlier in the independent variable space.

Fig. 3(a) and (b) depicts the response surface and contour plot showing the effect of sample amount (B) and extraction solvent volume (D) on the response at fixed values of preconcentration solvent volume (A) and sonication time (C) in their center values (see Table 1). This figure clearly shows that the extraction solvent (water) volume has a negative effect on the response, but the sample amount has a positive effect on the response. The presence of a curvature in the model shows that the interaction between the sample amount and the extraction time is significant.

Fig. 3 (a) 3D response surface and (b) contour plot for extraction solvent volume (D) vs. sample amount (B).

Fig. 4(a) and (b) demonstrates the response surface and contour plot showing the effect of preconcentration solvent volume (A) and extraction solvent volume (D) on the response at the fixed values of the sample amount (B) and sonication time (C) in their center values (see Table 1). This figure also clearly shows that the interaction between preconcentration solvent volume (A) and extraction solvent volume (D) is significant.

Fig. 4 (a) 3D response surface and (b) contour plot for preconcentration solvent volume (A) vs. extraction solvent volume (D).

In general, 30 experiments have been performed using experimental design to determine the optimum conditions for four effective extraction parameters in five levels with corresponding replicates. In the case of the OVAT strategy, minimum 60 experiments (3 replicates for each run) were needed to determine the optimum conditions. However, the interaction effects between factors and their quadratic terms showing the curvature in the response surface cannot be studied.

The validated response surface model was finally optimized using the Nelder–Mead simplex optimization method (also known as variable-size simplex method) to get the optimum values of the effective factors on UASE-USAEME. In this regard, the optimization space of the significant factors in the obtained model was constrained in their initial range (shown in Table 1) and the goal of optimization was obtaining the maximum sum of peak areas. The simplex algorithm found the maximum peak area to be 1.89 × 10¹⁰ for the model in eqn (1) where the optimum extraction parameters were as follows: 79.6 mg of the sample amount, 1.1 mL of water as extraction solvent, 62.7 μL of chloroform as preconcentration solvent, and 18.6 min of sonication time.

Finally, for the evaluation of the developed model and corresponding optimum extraction parameters, the UASE-USAEME procedure and HPLC-DAD analysis were repeated three times (n = 3) at optimum conditions and the experimental response of 1.80 × 10¹⁰ was obtained. The experimental response was in agreement with the one calculated by the model according to the confidence interval in the data, which was in the range from 1.6 × 10¹⁰ to 2.1 × 10¹⁰.

3.2. Characterization of the non-volatile components of saffron

The HPLC-DAD chromatogram of the non-volatile components of saffron under the optimized extraction conditions is shown in Fig. 5.

Fig. 5 The second-order HPLC-DAD chromatogram of extracted saffron constituents in optimum extraction conditions.

As can be seen, a large number of components are extracted from saffron and separated with reasonable chromatographic resolution. Identification of the isolated components from saffron was carried out by comparing their spectral profiles and retention times with those of standards and identified components in the literature.^34–36

The main identified components, their retention times and their maximum absorption wavelengths are presented in Table 4. These main components are safranal, picrocrocin,³⁴ crocetin derivatives¹⁰ such as crocin, crocetin-mono-(β-D-glucosyl)-ester, crocetin-di-(β-D-glucosyl)-ester, and carotenoid derivatives such as kaempferol^35,36 and kaempferol-3,7,40-triglucoside.³⁵

Table 4 Retention time, maximum wavelength of absorption in the UV-Vis spectra and tentative identities of the most important detected constituents in saffron extract

Peak number	Retention time (min)	λmax (nm)	Estimation of chemical species
1	1.3	250	Picrocrocin
2	1.6	260, 320, 380	Kaempferol-3,7,4′-triglucoside
3	3.7	320	Safranal
4	4.4	250, 330, 450	cis-Crocin3
5	5.0	250, 440, 470	trans-Crocin3
6	5.6	255, 425, 450	trans-Crocetin
7	5.9	260, 320, 448	trans-Crocin4
8	6.9	319, 419, 443	cis-Crocetin
9	8.2	260, 353, 450	cis-Crocin4

---

The obtained HPLC-DAD chromatograms of a standard saffron sample can be considered as a reference chromatographic fingerprint for the quality control of different commercial saffron samples. Additionally, the proposed analytical method can be used as an alternative method to the one described in ISO3632 for the quality control of saffron. The proposed method has many advantages over the one described in ISO3632. These advantages are faster extraction (18.6 min instead of 60 min for the ISO method), lower solvent volume (1.1 mL instead of 5.0 mL for the ISO method), lower sample amount (79.6 mg instead of 500 mg for the ISO method), more efficient extraction of components (twenty extracted chemical components for the current method instead of three components for the ISO method) with higher relative concentrations, and a greater number of components in the quality evaluation of saffron.

3.3. Multivariate clustering of commercial saffron samples

To show the potential of our method for saffron quality control, fifteen commercial samples are chosen, extracted and analyzed using the optimized UASE-USAEME-HPLC-DAD method in triplicate.

Fig. 6 shows the overlaid HPLC-DAD fingerprints of fifteen saffron samples from the five different commercial brands listed in Table 5.

Fig. 6 Overlaid HPLC-DAD fingerprints of fifteen saffron samples from five different commercial brands listed in Table 5.

Table 5 The commercial saffron samples and their codes for multivariate cluster analysis

Number	Brand name	Samples code
1	Bahraman	B1(1), B2(2), B3(3)
2	Abbasszadeh	A1(4), A2(5), A3(6)
3	Standard	Std(7)
4	Naffis	N1(8), N2(9), N3(10)
5	Saharkhiz	S1(11), S2(12), S3(13)
6	Golestan	G1(14), G2(15), G3(16)

---

Shifts of elution times for the same chemical components in different samples were an important issue that can be clearly seen from this figure. It should be pointed out that the elution time shifts were different among various runs and for different peaks. In addition, other common chromatographic problems, such as baseline/background contribution, low S/N peaks, noise and peak overlap existed in the chromatographic fingerprints of saffron samples. Therefore, the effects of baseline/background contribution and elution time shifts were corrected using asymmetric least squares (AsLS)³⁷ and correlation optimized warping (COW),³⁸ before cluster analysis.

For multivariate clustering of the chromatographic fingerprints of commercial saffron samples and comparison of their fingerprints with the standard saffron sample, the corrected data matrix was analyzed using PCA. Autoscaling was chosen as a preprocessing step before PCA analysis.

Fig. 7 shows the results of PCA analysis. The PC1–PC3 plot accounted for 51.86% of the explained variance (PC1 = 22.91%, PC2 = 15.81% and PC3 = 13.14%). The scores plot in Fig. 7 shows the samples distribution in 3D space of the first, second and third principal components. The chromatographic fingerprint of the standard saffron sample is shown in red. As can be seen, most of the samples have similar scores to the standard. However, there are some samples with thoroughly different scores on three PCs. PCA can provide information about the similarities and dissimilarities of chromatographic fingerprints of commercial saffron samples compared to the standard one.

Fig. 7 The score plot of PCA analysis. Red circle shows the standard saffron sample.

To have a better discrimination between clear-cut clusters, distance-based clustering methods, such as hierarchical cluster analysis (HCA) and k-means can be used.³⁹

As an example, Fig. 8 shows the cluster analysis results obtained by the k-means method. By selecting a linkage of 1.5 as the threshold in this dendrogram, samples are shown to belong to three clear-cut clusters. The standard saffron sample is highlighted in red in this figure. Similar to the PCA results, the similarities and dissimilarities between standard and commercial saffron samples can be clearly seen using this figure. In other words, samples with similar chemical compositions (chromatographic fingerprints) to the standard saffron sample are placed in the same cluster (green color) and the other samples are placed in two different clusters (red and blue colors).

Fig. 8 Dendrogram obtained by k-means method for standard and commercial saffron samples. Red box demonstrates the standard saffron sample.

In summary, the chemometrics-based strategy used in this work provided a great deal of useful information from the chromatographic fingerprints of saffron. Multivariate optimization of UASE-USAEME-HPLC-DAD was performed, and then the optimized method combined with a multivariate clustering method was used for the quality control of the commercial saffron samples. Additionally, the main chemical components of saffron were tentatively identified.

4. Conclusion

A chemometrics-assisted strategy was proposed for the extraction of the bioactive constituents of saffron using UASE-USAEME combined with HPLC-DAD. Multivariate optimization based on RCCD and MLR was used for the optimization of the effective parameters on the efficiency of the extraction procedure. Good statistical parameters were obtained for the developed model and the values of RSDs and enrichment factors were below 1.0% (n = 3) and higher than 10 for all the extracted components, under the optimum extraction conditions, respectively. All of these results confirmed the reliability of the proposed method. Finally, the developed analytical method was used as a reliable method for the quality control of the commercial saffron samples purchased from different markets. Additionally, the multivariate clustering methods, PCA and k-means, are used for determining similarities and dissimilarities between standard and commercial saffron samples, according to their HPLC fingerprints. Furthermore, the strategy proposed in this work can be used as an alternative method to the one described in ISO3632 for the quality control of saffron. Inspection of the results showed that the strategy proposed in this work is more accurate and unbiased and also extracts a more complete set of information from the data, compared to conventional methods.

Acknowledgements

The authors would like to thank Dr F. Momenbeik from the Department of Chemistry, University of Isfahan for his kind help, comments and discussion in the HPLC-DAD analysis part of this work. Also, H.P. would like to thank the National Elite Foundation of Iran for Dr Ashtiyani's Research grant.

References

1. M. Jalali-Heravi, H. Parastar and H. Ebrahimi-Najafabadi, J. Chromatogr. A, 2009, 1216, 6088–6097.
2. M. Zougagh, B. Simonet, A. Rios and M. Valcarcel, J. Chromatogr. A, 2005, 1085, 293–298.
3. H. Sereshti, R. Heidari and S. Samadi, Food Chem., 2014, 143, 499–505.
4. A. Kyriakoudi, A. Chrysanthou, F. Mantzouridou and M. Z. Tsimidou, Anal. Chim. Acta, 2012, 755, 77–85.
5. M. Zougagh, A. Ríos and M. Valcárcel, Anal. Chim. Acta, 2005, 535, 133–138.
6. J. Escribano, A. Rubio, M. Alvarez-Ortí, A. Molina and J. A. Fernández, J. Agric. Food Chem., 2000, 48, 457–463.
7. C.-Y. Li and T.-S. Wu, J. Nat. Prod., 2002, 65, 1452–1456.
8. M. Jalali-Heravi, H. Parastar and H. Ebrahimi-Najafabadi, Anal. Chim. Acta, 2010, 662, 143–154.
9. R. Kumar, V. Singh, K. Devi, M. Sharma, M. Singh and P. S. Ahuja, Food Rev. Int., 2008, 25, 44–85.
10. P. A. Tarantilis, A. Beljebbar, M. Manfait and M. Polissiou, Spectrochim. Acta, Part A, 1998, 54, 651–657.
11. J. Escribano, G.-L. Alonso, M. Coca-Prados and J.-A. Fernández, Cancer Lett., 1996, 100, 23–30.
12. P. A. Tarantilis and M. G. Polissiou, J. Agric. Food Chem., 1997, 45, 459–462.
13. M. Zougagh, A. Ríos and M. Valcárcel, Anal. Chim. Acta, 2006, 578, 117–121.
14. P. Lozano, D. Delgado, D. Gomez, M. Rubio and J. Iborra, J. Biochem. Biophys. Methods, 2000, 43, 367–378.
15. G. L. Alonso, M. R. Salinas, F. J. Esteban-Infantes and M. A. Sánchez-Fernández, J. Agric. Food Chem., 1996, 44, 185–188.
16. P. A. Tarantilis, M. Polissiou and M. Manfait, J. Chromatogr. A, 1994, 664, 55–61.
17. N. Zarghami and D. Heinz, Phytochemicals, 1971, 10, 2755–2761.
18. M. D'Auria, G. Mauriello and G. L. Rana, Flavour Fragrance J., 2004, 19, 17–23.
19. C. D. Kanakis, D. J. Daferera, P. A. Tarantilis and M. G. Polissiou, J. Agric. Food Chem., 2004, 52, 4515–4521.
20. J. Regueiro, M. Llompart, C. Garcia-Jares, J. C. Garcia-Monteagudo and R. Cela, J. Chromatogr. A, 2008, 1190, 27–38.
21. F. Kamarei, H. Ebrahimzadeh and Y. Yamini, Microchem. J., 2011, 99, 26–33.
22. M. A. Bezerra, R. E. Santelli, E. P. Oliveira, L. S. Villar and L. A. Escaleira, Talanta, 2008, 76, 965–977.
23. D. Baş and İ. H. Boyacı, J. Food Eng., 2007, 78, 836–845.
24. R. Leardi, Anal. Chim. Acta, 2009, 652, 161–172.
25. N. Hakimzadeh, H. Parastar and M. Fattahi, J. Chromatogr. A, 2014, 1326, 63–72.
26. H. Parastar, M. Jalali-Heravi, H. Sereshti and A. Mani-Varnosfaderani, J. Chromatogr. A, 2012, 1251, 176–187.
27. S. Wold, K. Esbensen and P. Geladi, Chemom. Intell. Lab. Syst., 1987, 2, 37–52.
28. R. Bro and A. Smilde, Anal. Methods, 2014, 6, 2812–2831.
29. P. A. Burrough, P. F. van Gaans and R. MacMillan, Fuzzy Set. Syst., 2000, 113, 37–52.
30. D. T. Campbell, J. C. Stanley and N. L. Gage, Experimental and quasi-experimental designs for research, Houghton Mifflin, Boston, 1963.
31. A. Kyriakoudi, A. Chrysanthou, F. Mantzouridou and M. Z. Tsimidou, Anal. Chim. Acta, 2012, 755, 77–85.
32. E. Morgan, Chemometrics: experimental design, Wiley, New York, 1995.
33. D. C. Montgomery, Design and analysis of experiments, John Wiley & Sons, 2008.
34. H. Caballero-Ortega, R. Pereda-Miranda and F. I. Abdullaev, Food Chem., 2007, 100, 1126–1131.
35. M. Carmona, A. M. Sánchez, F. Ferreres, A. Zalacain, F. Tomás-Barberán and G. L. Alonso, Food Chem., 2007, 100, 445–450.
36. M. Ashrafi, S. Bathaie, M. Taghikhani and A. Moosavi-Movahedi, Int. J. Biol. Macromol., 2005, 36, 246–252.
37. P. H. Eilers, Anal. Chem., 2004, 76, 404–411.
38. G. Tomasi, F. van den Berg and C. Andersson, J. Chemom., 2004, 18, 231–241.
39. R. G. Brereton, Multivariate pattern recognition in chemometrics: illustrated by case studies, Elsevier, 1992.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5ra00488h

---