Food Chemistry 150 (2014) 414–421
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Analytical Methods
On the quality control of traded saffron by means of transmission Fourier-transform mid-infrared (FT-MIR) spectroscopy and chemometrics Stella A. Ordoudi a, Marcelino de los Mozos Pascual b, Maria Z. Tsimidou a,⇑ a b
Laboratory of Food Chemistry and Technology, School of Chemistry, Aristotle University of Thessaloniki (LFCT, AUTH), Thessaloniki 54124, Greece Centro de Investigación Agraria de Albaladejito (CIAA), 16194 Cuenca, Spain
a r t i c l e
i n f o
Article history: Received 14 September 2012 Received in revised form 15 July 2013 Accepted 2 November 2013 Available online 13 November 2013 Keywords: Saffron Storage Quality control FT-IR Principal Component Analysis
a b s t r a c t The present study aimed to extend application of the FT-MIR technique to the quality control of traded saffron that suffers various types of fraud or mislabelling. Spectroscopic data were obtained for samples stored for different periods in the dark. Samples with the highest quality according to ISO 3632 specifications produced a typical spectrum profile (reference set). Principal Component Analysis (PCA) of spectroscopic data for this set along with HPLC-DAD analysis of major apocarotenoids assisted identification of FT-IR bands that carry information about desirable sensory properties that weaken during storage. The band at 1028 cm1, associated with the presence of glucose moieties, along with intensities in the region 1175–1157 cm1, linked with breakage of glycosidic bonds, were the most useful for diagnostic monitoring of storage effects on the evaluation and test set samples. FT-IR was found to be a promising, sensitive and rapid tool in the fight against saffron fraud. Ó 2013 Elsevier Ltd. All rights reserved.
1. Introduction Currently, traded quantities of saffron, the dried red stigmas of Crocus sativus L., are considerably greater than those in previous decades mainly due to exports from non-European producing countries (e.g. Iran, India). Herbs and spices are rather loosely regulated concerning authentication and quality specifications. Nevertheless, wholesalers and distributors of saffron in international transactions apply widely the specifications and methods described in the ISO 3632 trade standard (ISO 2010; ISO 2011). According to the ISO standard, dried stigmas that contain moisture and volatile matter at levels below 12% (w/w) are classified according to three categories (I–III) depending on certain physical and chemical characteristics. Above all, the ‘‘colouring strength’’ of aqueous saffron extracts, expressed as E1% 1cm 440 nm, is considered of utmost importance for the retail price as well as consumer acceptance (Ordoudi & Tsimidou, 2004). In terms of marketing, sample batches with values higher than 200 units belong to category I (highest quality) while those with the values lower than 120 units are sub-standard. In this way, the quality of bulk saffron from different parts of the world is monitored along the trade chain (Rekha, Koul & Ram, 2011) and ‘‘best before’’ dates for packaged saffron are set on the label to guarantee desired properties ⇑ Corresponding author. Tel.: +30 2310 997796; fax: +30 2310 997779. E-mail address: [email protected] (M.Z. Tsimidou). 0308-8146/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.foodchem.2013.11.014
under indicated storage conditions (ISO, 2010). In addition to colour, the taste and aroma of saffron, associated with the presence of picrocrocin and its dehydration product safranal, are expressed 1% as E1% 1cm 257 nm and E1cm 330 nm values, respectively. For example, saffron samples with E1% 1cm 257 nm values less than 40 units and E1% 1cm 330 nm values less than 20 units are considered sub-standard (ISO, 2011). Regarding adulteration issues, the absence of certain exogenous dyes from saffron should be clearly stated in the accompanying certificates of product analysis. Beyond trading standards, professionals in saffron trade chain have developed empirical criteria that allow them to choose ‘‘the best product’’ among those offered for sale. This, however, does not prevent the untrained buyer from being misled, as many problems from mislabelling to fraud have been reported in saffron trade (Ordoudi & Tsimidou, 2011; SAFFRONOMICS COST Action FA1101, 2011–2015; Zalacain, Ordoudi, Blázquez, et al., 2005). To protect fair trade and consumers analytical tools are continuously updated and the potential of new powerful techniques is explored. FT-IR spectroscopy has a strong potential in the analysis and quality control of foods because of its sensitivity, versatility, and speed (e.g. Sun, 2009). Recent advances in the FT-IR instrumentation along with the development of multivariate data analysis has resulted in an increase in applications of the mid-infrared (MIR) spectroscopy to food safety, end-product quality/process control and authenticity studies (e.g. Karoui, Downey, & Blecker, 2010). With regard to band assignment, spectra in the MIR region
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
(4000–400 cm1) are more informative than those in the nearinfrared (NIR, 10,000–4000 cm1) where broad overtone and combination absorption bands occur (e.g. Engelsen & Nørgaard, 1996; Sun, 2009). Depending on the method used (e.g. transmission, diffuse reflectance, attenuated total reflectance) FT-IR spectroscopy minimizes or even eliminates sample preparation. Moreover, chemometric analysis of FT-IR data from complex mixtures, such as most food matrices, leads to graphical representation of differences among samples even if individual component identification through peak assignment is not always straightforward. So far, applications of FT-MIR spectroscopy in the analysis of herbs and spices are extremely limited (e.g. Schulz, Schrader, Quilitzsch, Pfeffer, & Krüger, 2003). One application of this technique along with chemometrics was recently reported by Anastasaki et al. (2010) which described differentiation of saffron samples from four countries (Iran, Spain, Sardinia, Greece); an objective also investigated using FT-NIR some years ago (Zalacain, Ordoudi, Díaz-Plaza, et al., 2005). The authors found that (a) all of the examined samples, which were of the same harvest year, produced the same profile in the region between 2000 and 700 cm1 and (b) discrimination of origin was possible using spectra derived from non-polar extracts of stigmas. Infrared bands in the carbonyl bond (1745, 1670 cm1) and double bond regions (1600 cm1) were thought to be responsible for the differentiation. Variance in these bands was assigned to differences in extract composition, in terms of lipid and safranal content, due to different post-harvest treatments (e.g. drying, storage) in the producing areas. The present study aimed to extend application of FT-MIR to the quality control of saffron. In particular, it was investigated how the typical FT-MIR spectrum of saffron changes as a result of storage under conditions that favour oxidative or hydrolytic decomposition of bound secondary metabolites such as crocins and picrocrocin. The rate of such reactions depends on relative humidity, temperature and light exposure (Alonso, Varon, Salinas, & Navarro, 1993; Raina, Agarwal, Bhatia, & Gaur, 1996; Tsimidou & Biliaderis 1997). Changes in the MIR spectrum of the dried stigmas were thoroughly investigated by overlapping spectra and chemometrics in order to elucidate which of them were of diagnostic value. 2. Materials and methods 2.1. Plant material A total of 52 saffron samples were examined and grouped as ‘‘reference set’’ (n = 29), ‘‘evaluation sets’’ (n = 8) and ‘‘test set’’ (n = 15). Details for the plant material origin, collection year, dehydration process and storage conditions until FT-IR examination are given as Supplementary material (Table S1). 2.2. Reagents and solvents Picrocrocin of 90% chromatographic purity was isolated in our laboratory according to Sánchez, Carmona, del Campo, & Alonso (2009). In particular, 1 g of ground stigmas was de-fatted after extraction with cyclohexane (99.5%, Panreac Quimica S.A.U, Barcelona, Spain) by sporadic agitation for 24 h at room temperature in the dark. After evaporation of the organic solvent under vacuum, the dry residue was mixed with 15 mL of water; the mixture was magnetically stirred for 1 h at room temperature in the dark and centrifuged at 4500 rpm for 10 min. The supernatant was collected and transferred to C18 SPE cartridges (average particle size 50 lm, pore size 67 Angstrom, 500 mg). After washing each cartridge with 10 mL of 2% acetonitrile/water (v/v), picrocrocin was eluted with 15 mL of 10% acetonitrile/water (v/v). The solvent was evaporated under vacuum and the dry residue was examined using HPLC-DAD
415
to calculate the chromatographic purity of the isolated picrocrocin as the percentage of the total peak area at 250 nm. FT-IR grade KBr (>99%) was purchased from Sigma–Aldrich (Steinheim, Germany). HPLC-grade methanol was from Panreac Quimica S.A.U and HPLC-grade acetonitrile was from Chem-Lab NV (Zedelgem, Belgium). Water purification was carried out with the aid of an EASYpureRoDi (Barnstead Int., Dubuque, Iowa, USA) system. Acetic acid glacial (HPLC grade) was from Panreac Quimica S.A.U. 2.3. FT-IR spectroscopy Saffron stigmas were ground according to ISO/TS 3632-2 (ISO, 2011). For FT-IR transmittance measurements, all samples were mixed with KBr at a 1/180 ratio (w/w) and homogenised. This mixture (0.181 g) was then compressed under a pressure of ca. 200 MPa for 1 min to form a thin KBr disc. For each sample, the disc preparation procedure was carried out in triplicate. The total analysis time required per sample was ca. 45 min. FT-IR spectra were obtained using a Shimadzu IRAffinity -1 (Shimadzu Europa GmbH, Duisburg, Germany) spectrometer operating in the region 4000–400 cm1 in the absorbance mode. A total of 64 scans with 4 cm1 resolution were acquired for each spectrum. The spectrum of a clean KBr disc (without saffron) was used for background subtraction. The spectrometer was located in an air-conditioned room (25 °C). The spectra were stored using the software IRsolution (version 1.50) supplied from the same manufacturer. 2.4. Spectral data analysis 2.4.1. Data preprocessing All spectra were smoothed by 15 points using the software function ‘‘smoothing action’’. Then the baseline was corrected using the ‘‘3 point baseline operation’’ facility (set zero at 400, 2000 and 4000 cm1). Finally, the spectra were normalised using the order ‘‘normalise action’’ so that the minimum absorbance was set at Abs = 0 (zero) and the maximum at Abs = 1. The major peaks in each spectrum were identified using the ‘‘peak table’’ function. The ‘‘derivative action’’ facility was also selected to calculate second order derivatives using the Savitzky–Golay method and 11 data points of interval. Inversion of these spectra was achieved using the function ‘‘arithmetic action’’ and multiplying each data point by 1. The data obtained were processed further using the tools in the Microsoft Excel 2010 software. Each spectrum was truncated to 1868 data points. 2.4.2. Principal Component Analysis (PCA) Principal Component Analysis (PCA) transforms a set of variables into a new set of composite variables, the principle components (PCs). PCA of the spectral data was performed with the aid of the SPSS for Windows version 17.0 to reduce the number of variables and explore whether the principal components could be interpreted in terms of physicochemical parameters. General restrictions about sampling adequacy were taken into account (Osborne & Costello, 2005). In detail, basic statistical tests such as the correlation and anti-image correlation matrices, the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy (KMO value > 0.6) and the Bartlett’s Test of sphericity were carried out in order to retain an object-to-variable ratio of 3.0–5.0. Criteria for the identification of ‘principal components’ (PCs) included communalities (>0.8), the Kaiser criterion (all components with eigenvalues greater than 1.0) and the number of variables loaded on each PC (>3). Varimax rotation was selected as the method to produce uncorrelated components and clarify the data structure. Crossloaded variables (with loadings > 0.4 on two PCs) were
416
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
included in the analysis (Osborne & Costello, 2005). Once the most important PCs were identified, the scores on each were calculated for every object (sample) in the ‘‘reference set’’ and the corresponding scatterplots constructed. PCA was repeated for the evaluation set. 2.4.3. Multiple Linear Regression (MLR) analysis MLR analysis was carried out with the aid of SPSS for Windows version 17.0. Regression models between PC scores (dependent variables) and compositional data (independent variables) were constructed using both the enter and the stepwise methods of variable selection.
where Ai: the percentage peak area of the crocin(i) at 440 nm, E1% 1cm 440 nm: colouring strength value, et,c: molar coefficient absorbance value (89,000 for trans- and 63,350 for cis-crocins) (Speranza, Dadà, Manitto, Monti, & Gramatica, 1984), mi: molecular mass of the crocin(i). Repeatability of measurements, expressed as% CV (n = 7) of individual peak areas was satisfactory (below 9.4% for major crocins). Then, chromatographic analysis was carried out in duplicate. The content of dried styles in picrocrocin (% w/w) was estimated using HPLC data at 250 nm from a five-point calibration curve of picrocrocin in water. Repeatability of measurements, expressed as% CV (n = 7) was 2.3%. 3. Results and discussion
2.5. Apocarotenoid analysis in saffron extracts 3.1. Reference set 2.5.1. Extraction of crocins and picrocrocin Methanol–water extracts of ground stigmas were prepared according to the ISO 3632-2 method (ISO, 2010) with slight modifications. Briefly, 0.1 g was mixed with methanol–water (1:1, v/v) in a 200 mL volumetric flask. Crocins and picrocrocin were extracted by rigorous agitation (1000 rpm) for one hour at ambient temperature (25 °C) away from direct sunlight. Prior to analysis, an aliquot from the extract was diluted (1:10) with methanol– water (1:1, v/v) and the corresponding solutions were filtered through RC-55 filter (13 mm i.d., 0.45 lm pore size). 2.5.2. Spectrophotometric examination of saffron extracts The UV–Vis spectra were recorded in the region 200–600 nm using the Shimadzu UV 1601 (Kyoto, Japan) and processed with the aid of UVPC 1601 (Personal Spectroscopy Software, v.3.9, Shimadzu) software facilities. The ISO 3632-1 quality parameters 1% 1% E1% 1cm 440 nm, E1cm 257 nm, E1cm 330 nm were calculated, on dry basis, according to the equation:
E1% 1cm ¼ D 10000=mð100 HÞ
ð1Þ
E1% 1cm :
where the extinction coefficient of a 1% (w/v) solution when path length of the light is 1 cm, D: the absorbance value, m: the mass of the test portion (g) and H: the moisture and volatile content of the sample (% w/w). The latter was determined using the gravimetric method proposed in ISO 3632-2 (ISO, 2010). Since the required sample amount (2.5 g) was not available for all of the samples of the reference set, the H value used for mass correction, 10% w/w, was that of a representative mixture of stigmas from the 29 reference samples. 2.5.3. Chromatographic examination of saffron extracts HPLC-DAD analysis of crocins and picrocrocin in the methanol– water extracts of C. sativus L. was carried out as described by Sánchez, Carmona, Ordoudi, Tsimidou, and Alonso (2008) with some modifications. Briefly, apocarotenoid constituents were separated on a LiChroCART Superspher 100 RP-18 (125 4 mm i.d, 4 lm) end-capped column (Merck, KGaA, Darmstadt, Germany) after injection of a 20 lL aliquot and gradient elution with a mixture of water–acetic acid 1%, v/v, (A) -acetonitrile (B) (20–100% B in 20 min) at a flow rate of 0.5 mL/min. The HPLC system consisted of a pump (Thermo Separation Products, model P4000, San Jose, CA), a Midas autosampler (Spark, Emmen, The Netherlands), a UV 6000LP diode array detector (DAD; Thermo Separation Products) and a Spectra System SCM1000 on-line degasser (Thermo Separation Products). Monitoring was at 250 nm (picrocrocin) and 440 nm (crocins). Due to the lack of pure standards for each crocin, quantification was based on an equation recently applied by Sánchez et al. (2008). The equation was:
% of crocinðiÞ on dry basis ¼ Ai ðE1% 1cm 440nm=et;c Þ ðmi =10Þ
ð2Þ
Quality characteristics of the reference set samples were examined for compliance with the category I according to ISO 3632-1 specifications. The ranges for the calculated E1% 1cm values at 440 nm (219–304), 257 nm (79–104) and 330 mm (23–31) were above the minimum requirements for the category (200, 70 and 20, respectively), indicating that all of the samples were of the highest quality. Frequency distribution of these data showed that ca. 60% of the reference set samples differed among them by ±20 units of colouring strength, ±7 units of E1% 1cm 257 nm and ±2 units of E1% 1cm 330 nm values implying, thus, that the levels of the corresponding secondary metabolites, crocins, picrocrocin and safranal, should be of similar magnitude. 3.1.1. Spectral characteristics Reference set samples typically generated the FT-IR profile shown in Fig. 1. This spectrum was comprised of well-resolved peaks. Nine of them were automatically pinpointed in the characteristic group region (1800–1500 cm1) and less than five in the fingerprint region (1500–800 cm1). The spectra recorded in our work were found to be qualitatively similar to those reported by Anastasaki et al. (2010) using the micro-DRIFTS method of measurement. The longer analysis time needed in the present study because of the use of the KBr discs was counterbalanced by improved resolution of the IR peaks. Different operational modes (transmission vs diffuse reflectance) affect the signal-to-noise ratio in the spectra while other instrumental parameters (e.g. number of scans) as well as spectral data processing parameters (e.g. number of smoothing points) are expected to influence height and position of IR bands (Smith, 2011). For example, a distinct feature of the transmittance spectrum was that intensity ratio values of the band at about 1070 cm1 versus that at 1659 cm1 ranged between 1.2 and 1.8 (Fig. 1). The respective ratio value in the diffuse reflectance spectrum reported by Anastasaki et al. (2010) was close to one, signifying the advantage of the transmission method in quantitative studies. Another characteristic difference between our findings and those reported elsewhere was the position of the various bands in the fingerprint region (1500–800 cm1). These bands were recorded at ca. 20–30 cm1 lower than those reported in published data for diffuse reflectance infrared spectra of saffron nonpolar extracts. Table 1 summarizes the major bands in a typical FT-IR spectrum for saffron and assigns possible identities based on theory (Coates, 2000; Sun, 2009) and relevant reports. Taken that saffron is a complex mixture of chemical compounds, it is difficult to assign all of the bands directly to specific constituents. Nevertheless, since our aim was to identify bands having the strongest estimated correlation with storage effects, we investigated whether specific bands in the infrared spectra can be at least partially, associated with characteristic structural features of major metabolites that are expected to degrade in storage and are
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
417
Fig. 1. Typical FT-IR spectrum profile of a saffron sample belonging to the reference set.
Table 1 Major IR absorption bands and possible assignment in the typical FT-IR spectrum of saffron (Anastasaki et al., 2010; Coates, 2000; Kanou et al., 2005; Nikonenko et al., 2005; Sun, 2009). Band (cm1)*
Assignment
3365–3333 (broad) 2924 (m) and 2857 (m) 1745–1696 (sh) 1659–1653 (s) 1613 (m) 1580–1578 (m) and 1545–1542 (sh) 1454 (w) 1400 (w) and 1375 (w)
Stretching vibration of bonded and non-bonded –OAH groups Asymmetric –CH2–, symmetric –CH3 and –CH2– stretching vibrations –C@O stretching vibrations (e.g. in the –COOR groups of crocetin/–COOH groups of aminoacids) @C–H stretching vibrations (e.g. in the RCH@CHR’ groups of crocetin)/amide I/O–H bending vibrations in water C–C skeletal vibrations C–O vibrations (e.g. in the –COOR groups of crocetin)/amide II/aromatic –C@C stretching vibrations) C–H bending (scissoring) (in CH3 groups)/aromatic –C@C stretching vibrations –OH bending vibrations, –C–O–H in-plane bending vibrations,–CH3 out-of-plane bending vibrations, –CH2– wagging and twisting vibrations C(O)–O stretching vibrations and –OH in plane vibrations/amide III (e.g. in aromatic ethers)
1317 (w), 1294–1292 (w), 1271 (w) and 1227 (s) 1157 (sh) 1075 (s) 1020 (sh) 970–920 (sh) 780–700 (w) *
C–O stretching vibrations (e.g. in C–O–C glycosidic linkages of oligosaccharides or in triacylglycerols) C–1–H bending vibration in sugars C–4–OH (typical for glucose residue of disaccharides) trans = C–H out-of-plane bending cis = C–H out-of-plane bending
m: medium; s: strong; sh: shoulder; w: weak.
responsible for desirable sensory properties (colour, taste and aroma). To our knowledge, there are no published reports providing information about storage effects on other compounds that co-exist in saffron (e.g. proteins, sugars, lipids, etc.). 3.1.2. Selection of infrared bands using PCA To reduce the number of potent variables being at the same time consistent with general restrictions about sampling adequacy in PCA (Osborne & Costello, 2005), we used IR bands in the regions reported by Anastasaki et al. (2010) to be of diagnostic value considering also the information given in Table 1. These were: three bands in the carbonyl and double bond region (1744, 1701, 1655 cm1), eight bands associated with aromatic ring, peptide and glycosidic bonds (1580, 1549, 1456, 1375, 1317, 1271, 1227, 1157 cm1) as well as two bands related to sugars (1053, 1028 cm1) and one associated with the trans configuration in HC = CH moieties of apocarotenoid pigments (968 cm1). The average intensity values for each sample at these frequencies were calculated and introduced to the data matrix consisting of 29 objects (rows) and 14 variables (columns). To carry out Principal Component Analysis and retain an object-to-variable ratio of 3.0–5.0,
the correlation and anti-image correlation matrices as well as the results from Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s Test of sphericity were examined as explained in Section 2.5.2. These basic statistical tests showed that the bands at 1053 and 1028 cm1 were poorly correlated with those above 1300 cm1 and the bands above 1271 cm1 were strongly correlated with each other. On account of the KMO value for the whole set of variables (0.769) further statistical analysis was feasible. However, since KMO values for particular variables in the anti-image correlation matrix were lower than 0.6 the corresponding bands (1744, 1549, 1375 and 1271 and 1053 cm1) were removed leading to improved measures of sampling adequacy for the remaining dataset. A new matrix comprised of 29 objects 9 variables was thus used for PCA. The PCA resulted in high communalities for all variables (0.92–0.99). Two latent variables were found to represent 95% of the total variance (PC1 = 63%, PC2 = 32%). After rotation of the component matrix, absorbance intensities at 1701, 1655, 1580 and 1456 cm1 were found to be loaded on PC1 with positive directions. On the other hand, the corresponding values at 1028 cm1 loaded mainly on PC2 and contributed negatively to the formation
418
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
of PC1. The rest of the variables showed complex structures with moderate positive loadings on both principal components. Using the PC score values a two-dimensional scatterplot was constructed, as shown in Fig. 2. Irrespective of the harvest year, almost all of the reference set samples had score values between ±1.50. They were distributed equally between the (, +) and (+, +) quarters of the PC12 plot and also between the (+, ) and (, ) ones although most of the samples were located in the former two ones. As mentioned above, the observed variance in the scatterplot is actually the graphical representation of differences in the chemical composition of the samples. In Fig. 2, samples lying at the ±edges of the PC1 axis were found to differ by only 10 units of colouring strength while those lying at the ±edges of the PC2 axis differed by 35 units of colouring strength. Based on the frequency distribution data presented in 3.1., these differences could not account for distinct sub-groupings within the cloud of the scattered points in Fig. 2. Obviously, the ISO quality parameters provide quantitative information only about saffron content in apocarotenoids and not other compounds. Taken that high quality saffron is abundant in the former constituents, it is realistic to assume that the projection of the reference set samples onto the PC12 plane did not reflect any sub-grouping due to harvest year. Further observation of Fig. 2 shows samples collected in 2011 had only positive PC2 scores. As discussed earlier, the second PC is primarily formed due to the variance at 1028 cm1, which in turn, has been associated with the presence of bound a- or b-D glucose moieties in aqueous solutions of free disaccharides (Kanou, Nakanishi, Hashimoto, & Kameoka, 2005). Taking into consideration that saffron contains a relatively high amount of gentiobiose (6-O-b-D-glucopyranosyl-D-glucose) and glucose moieties (e.g. esterified to apocarotenoid skeleton), variance in the intensity at 1028 cm1 could also provide information about structural changes (e.g. breakage of glycosidic bonds) of these moieties. To gain insight into the parameters that influence variance at PC2 we examined the reference set composition in major crocins such as the trans- and cis- isomers of crocetin di-b-D-gentiobiosyl ester (t-4-GG and c-4-GG), the trans- crocetin (b-D-gentiobiosyl)-(b-Dglucosyl) ester (t-3-Gg) and picrocrocin (4-(b-D-glucopyranosyl)2,6,6-trimethyl- 1-cyclohexene-1-carboxaldehyde). HPLC-DAD analysis of water–methanol saffron extracts was carried out. On the basis of the estimated % w/w apocarotenoid content, it was found that samples located at the ±edges of the PC2 axis had
different HPLC quantitative profiles. For example, the lowest PC2 score value was obtained for a sample located in the (, ) quarter of the plot (x = 0.72, y = 2.47); it had high t-4-GG (>20.0% w/w), high t-3-Gg (>8.0% w/w), moderate c-4-GG levels (1.0% w/w) and was the richest in picrocrocin (>24.0%). On the other hand, the three samples with the highest PC2 score values (x, y) = (1.15 1.09), (1.05, 1.29) and (0.84, 1.50) were located diagonally opposite in the (+, +) quarter of the plot; they were found to contain moderate t-4-GG (19.5%, 21.0%, 18.5%, respectively) and c-4-GG levels (1.4%, 1.6%, 1.5%, respectively) but relatively low t-3-Gg (6.3%, 6.5%, 5.9%, respectively) and picrocrocin (15.4%, 14.1% and 16.3%, respectively). Summing up it can be deduced that differences in the apocarotenoid content account to a great extent for the observed variance along the PC2 axis. Multiple Linear Regression analysis was carried out to study possible correlations among the HPLC compositional data and the PC score values of the samples. Using the standard method of entering the independent variables to the analysis, two regression equations were produced:
PC1 ¼ 2:827 þ 0:082t 4 GG þ 0:257t 3 Gg þ 0:267c 4 GG 0:495pic ðr ¼ 0:60Þ
ð3Þ
PC2 ¼ 4:924 þ 0:140t 4 GG 0:460t 3 Gg 0:091c 4 GG 0:478pic ðr ¼ 0:72Þ
ð4Þ
The results suggest 35.7% and 52.3% of the variance in PC1 and PC2 values, respectively, was due to the variance in the quantities of all the major apocarotenoids. Using the stepwise regression method the independent variables most effective in predicting the dependent variable were sought, and are described by:
PC1 ¼ 1:885 0:505pic ðr ¼ 0:51Þ
ð5Þ
PC2 ¼ 2:269 0:631pic ðr ¼ 0:63Þ
ð6Þ
PC2 ¼ 5:431 0:525pic 0:344t 3 Gg ðr ¼ 0:71Þ
ð7Þ
Based on the correlation coefficient values in the Eqs. (5) and (6), picrocrocin content alone accounts for the 25.5% and 39.9% of the observed variance, respectively. Structural features of this compound such as the carbonyl bond and methyl groups, double bonds and hydroxyl groups in the glucose moiety, are likely to
Fig. 2. Scatterplot of PC scores for reference set (samples from 2010 and 2011 harvests).
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
419
Fig. 3. FT-IR spectra of test set; (a) five (5) samples from individual producers and (b) ten (10) samples from blends of stigmas obtained from different producers.
produce IR vibration bands contributing to the PCs as described by Eqs. (5) and (6). The negative sign of the correlation coefficients suggests that decrease in picrocrocin content causes increase in the score values. The trans- and cis- isomers of crocetin di-(b-D-gentiobioside) were not shown to be significant predictors. The symmetry of the former molecules could play a role to this result as it reflects the relatively small number of IR active vibration bands (Sun, 2009). On the other hand, the t-3-Gg content was found to improve the regression coefficient value and assist the prediction of PC2 score values (Eq. (7)). 3.1.3. Evaluation sets For evaluation of the PCA findings, a set of six samples was first examined. The samples had been stored for more than a year under conditions that are not likely to bring about significant changes to the quality characteristics of the spice (e.g. reduction in colouring strength is expected to be
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Analytical Methods
On the quality control of traded saffron by means of transmission Fourier-transform mid-infrared (FT-MIR) spectroscopy and chemometrics Stella A. Ordoudi a, Marcelino de los Mozos Pascual b, Maria Z. Tsimidou a,⇑ a b
Laboratory of Food Chemistry and Technology, School of Chemistry, Aristotle University of Thessaloniki (LFCT, AUTH), Thessaloniki 54124, Greece Centro de Investigación Agraria de Albaladejito (CIAA), 16194 Cuenca, Spain
a r t i c l e
i n f o
Article history: Received 14 September 2012 Received in revised form 15 July 2013 Accepted 2 November 2013 Available online 13 November 2013 Keywords: Saffron Storage Quality control FT-IR Principal Component Analysis
a b s t r a c t The present study aimed to extend application of the FT-MIR technique to the quality control of traded saffron that suffers various types of fraud or mislabelling. Spectroscopic data were obtained for samples stored for different periods in the dark. Samples with the highest quality according to ISO 3632 specifications produced a typical spectrum profile (reference set). Principal Component Analysis (PCA) of spectroscopic data for this set along with HPLC-DAD analysis of major apocarotenoids assisted identification of FT-IR bands that carry information about desirable sensory properties that weaken during storage. The band at 1028 cm1, associated with the presence of glucose moieties, along with intensities in the region 1175–1157 cm1, linked with breakage of glycosidic bonds, were the most useful for diagnostic monitoring of storage effects on the evaluation and test set samples. FT-IR was found to be a promising, sensitive and rapid tool in the fight against saffron fraud. Ó 2013 Elsevier Ltd. All rights reserved.
1. Introduction Currently, traded quantities of saffron, the dried red stigmas of Crocus sativus L., are considerably greater than those in previous decades mainly due to exports from non-European producing countries (e.g. Iran, India). Herbs and spices are rather loosely regulated concerning authentication and quality specifications. Nevertheless, wholesalers and distributors of saffron in international transactions apply widely the specifications and methods described in the ISO 3632 trade standard (ISO 2010; ISO 2011). According to the ISO standard, dried stigmas that contain moisture and volatile matter at levels below 12% (w/w) are classified according to three categories (I–III) depending on certain physical and chemical characteristics. Above all, the ‘‘colouring strength’’ of aqueous saffron extracts, expressed as E1% 1cm 440 nm, is considered of utmost importance for the retail price as well as consumer acceptance (Ordoudi & Tsimidou, 2004). In terms of marketing, sample batches with values higher than 200 units belong to category I (highest quality) while those with the values lower than 120 units are sub-standard. In this way, the quality of bulk saffron from different parts of the world is monitored along the trade chain (Rekha, Koul & Ram, 2011) and ‘‘best before’’ dates for packaged saffron are set on the label to guarantee desired properties ⇑ Corresponding author. Tel.: +30 2310 997796; fax: +30 2310 997779. E-mail address: [email protected] (M.Z. Tsimidou). 0308-8146/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.foodchem.2013.11.014
under indicated storage conditions (ISO, 2010). In addition to colour, the taste and aroma of saffron, associated with the presence of picrocrocin and its dehydration product safranal, are expressed 1% as E1% 1cm 257 nm and E1cm 330 nm values, respectively. For example, saffron samples with E1% 1cm 257 nm values less than 40 units and E1% 1cm 330 nm values less than 20 units are considered sub-standard (ISO, 2011). Regarding adulteration issues, the absence of certain exogenous dyes from saffron should be clearly stated in the accompanying certificates of product analysis. Beyond trading standards, professionals in saffron trade chain have developed empirical criteria that allow them to choose ‘‘the best product’’ among those offered for sale. This, however, does not prevent the untrained buyer from being misled, as many problems from mislabelling to fraud have been reported in saffron trade (Ordoudi & Tsimidou, 2011; SAFFRONOMICS COST Action FA1101, 2011–2015; Zalacain, Ordoudi, Blázquez, et al., 2005). To protect fair trade and consumers analytical tools are continuously updated and the potential of new powerful techniques is explored. FT-IR spectroscopy has a strong potential in the analysis and quality control of foods because of its sensitivity, versatility, and speed (e.g. Sun, 2009). Recent advances in the FT-IR instrumentation along with the development of multivariate data analysis has resulted in an increase in applications of the mid-infrared (MIR) spectroscopy to food safety, end-product quality/process control and authenticity studies (e.g. Karoui, Downey, & Blecker, 2010). With regard to band assignment, spectra in the MIR region
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
(4000–400 cm1) are more informative than those in the nearinfrared (NIR, 10,000–4000 cm1) where broad overtone and combination absorption bands occur (e.g. Engelsen & Nørgaard, 1996; Sun, 2009). Depending on the method used (e.g. transmission, diffuse reflectance, attenuated total reflectance) FT-IR spectroscopy minimizes or even eliminates sample preparation. Moreover, chemometric analysis of FT-IR data from complex mixtures, such as most food matrices, leads to graphical representation of differences among samples even if individual component identification through peak assignment is not always straightforward. So far, applications of FT-MIR spectroscopy in the analysis of herbs and spices are extremely limited (e.g. Schulz, Schrader, Quilitzsch, Pfeffer, & Krüger, 2003). One application of this technique along with chemometrics was recently reported by Anastasaki et al. (2010) which described differentiation of saffron samples from four countries (Iran, Spain, Sardinia, Greece); an objective also investigated using FT-NIR some years ago (Zalacain, Ordoudi, Díaz-Plaza, et al., 2005). The authors found that (a) all of the examined samples, which were of the same harvest year, produced the same profile in the region between 2000 and 700 cm1 and (b) discrimination of origin was possible using spectra derived from non-polar extracts of stigmas. Infrared bands in the carbonyl bond (1745, 1670 cm1) and double bond regions (1600 cm1) were thought to be responsible for the differentiation. Variance in these bands was assigned to differences in extract composition, in terms of lipid and safranal content, due to different post-harvest treatments (e.g. drying, storage) in the producing areas. The present study aimed to extend application of FT-MIR to the quality control of saffron. In particular, it was investigated how the typical FT-MIR spectrum of saffron changes as a result of storage under conditions that favour oxidative or hydrolytic decomposition of bound secondary metabolites such as crocins and picrocrocin. The rate of such reactions depends on relative humidity, temperature and light exposure (Alonso, Varon, Salinas, & Navarro, 1993; Raina, Agarwal, Bhatia, & Gaur, 1996; Tsimidou & Biliaderis 1997). Changes in the MIR spectrum of the dried stigmas were thoroughly investigated by overlapping spectra and chemometrics in order to elucidate which of them were of diagnostic value. 2. Materials and methods 2.1. Plant material A total of 52 saffron samples were examined and grouped as ‘‘reference set’’ (n = 29), ‘‘evaluation sets’’ (n = 8) and ‘‘test set’’ (n = 15). Details for the plant material origin, collection year, dehydration process and storage conditions until FT-IR examination are given as Supplementary material (Table S1). 2.2. Reagents and solvents Picrocrocin of 90% chromatographic purity was isolated in our laboratory according to Sánchez, Carmona, del Campo, & Alonso (2009). In particular, 1 g of ground stigmas was de-fatted after extraction with cyclohexane (99.5%, Panreac Quimica S.A.U, Barcelona, Spain) by sporadic agitation for 24 h at room temperature in the dark. After evaporation of the organic solvent under vacuum, the dry residue was mixed with 15 mL of water; the mixture was magnetically stirred for 1 h at room temperature in the dark and centrifuged at 4500 rpm for 10 min. The supernatant was collected and transferred to C18 SPE cartridges (average particle size 50 lm, pore size 67 Angstrom, 500 mg). After washing each cartridge with 10 mL of 2% acetonitrile/water (v/v), picrocrocin was eluted with 15 mL of 10% acetonitrile/water (v/v). The solvent was evaporated under vacuum and the dry residue was examined using HPLC-DAD
415
to calculate the chromatographic purity of the isolated picrocrocin as the percentage of the total peak area at 250 nm. FT-IR grade KBr (>99%) was purchased from Sigma–Aldrich (Steinheim, Germany). HPLC-grade methanol was from Panreac Quimica S.A.U and HPLC-grade acetonitrile was from Chem-Lab NV (Zedelgem, Belgium). Water purification was carried out with the aid of an EASYpureRoDi (Barnstead Int., Dubuque, Iowa, USA) system. Acetic acid glacial (HPLC grade) was from Panreac Quimica S.A.U. 2.3. FT-IR spectroscopy Saffron stigmas were ground according to ISO/TS 3632-2 (ISO, 2011). For FT-IR transmittance measurements, all samples were mixed with KBr at a 1/180 ratio (w/w) and homogenised. This mixture (0.181 g) was then compressed under a pressure of ca. 200 MPa for 1 min to form a thin KBr disc. For each sample, the disc preparation procedure was carried out in triplicate. The total analysis time required per sample was ca. 45 min. FT-IR spectra were obtained using a Shimadzu IRAffinity -1 (Shimadzu Europa GmbH, Duisburg, Germany) spectrometer operating in the region 4000–400 cm1 in the absorbance mode. A total of 64 scans with 4 cm1 resolution were acquired for each spectrum. The spectrum of a clean KBr disc (without saffron) was used for background subtraction. The spectrometer was located in an air-conditioned room (25 °C). The spectra were stored using the software IRsolution (version 1.50) supplied from the same manufacturer. 2.4. Spectral data analysis 2.4.1. Data preprocessing All spectra were smoothed by 15 points using the software function ‘‘smoothing action’’. Then the baseline was corrected using the ‘‘3 point baseline operation’’ facility (set zero at 400, 2000 and 4000 cm1). Finally, the spectra were normalised using the order ‘‘normalise action’’ so that the minimum absorbance was set at Abs = 0 (zero) and the maximum at Abs = 1. The major peaks in each spectrum were identified using the ‘‘peak table’’ function. The ‘‘derivative action’’ facility was also selected to calculate second order derivatives using the Savitzky–Golay method and 11 data points of interval. Inversion of these spectra was achieved using the function ‘‘arithmetic action’’ and multiplying each data point by 1. The data obtained were processed further using the tools in the Microsoft Excel 2010 software. Each spectrum was truncated to 1868 data points. 2.4.2. Principal Component Analysis (PCA) Principal Component Analysis (PCA) transforms a set of variables into a new set of composite variables, the principle components (PCs). PCA of the spectral data was performed with the aid of the SPSS for Windows version 17.0 to reduce the number of variables and explore whether the principal components could be interpreted in terms of physicochemical parameters. General restrictions about sampling adequacy were taken into account (Osborne & Costello, 2005). In detail, basic statistical tests such as the correlation and anti-image correlation matrices, the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy (KMO value > 0.6) and the Bartlett’s Test of sphericity were carried out in order to retain an object-to-variable ratio of 3.0–5.0. Criteria for the identification of ‘principal components’ (PCs) included communalities (>0.8), the Kaiser criterion (all components with eigenvalues greater than 1.0) and the number of variables loaded on each PC (>3). Varimax rotation was selected as the method to produce uncorrelated components and clarify the data structure. Crossloaded variables (with loadings > 0.4 on two PCs) were
416
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
included in the analysis (Osborne & Costello, 2005). Once the most important PCs were identified, the scores on each were calculated for every object (sample) in the ‘‘reference set’’ and the corresponding scatterplots constructed. PCA was repeated for the evaluation set. 2.4.3. Multiple Linear Regression (MLR) analysis MLR analysis was carried out with the aid of SPSS for Windows version 17.0. Regression models between PC scores (dependent variables) and compositional data (independent variables) were constructed using both the enter and the stepwise methods of variable selection.
where Ai: the percentage peak area of the crocin(i) at 440 nm, E1% 1cm 440 nm: colouring strength value, et,c: molar coefficient absorbance value (89,000 for trans- and 63,350 for cis-crocins) (Speranza, Dadà, Manitto, Monti, & Gramatica, 1984), mi: molecular mass of the crocin(i). Repeatability of measurements, expressed as% CV (n = 7) of individual peak areas was satisfactory (below 9.4% for major crocins). Then, chromatographic analysis was carried out in duplicate. The content of dried styles in picrocrocin (% w/w) was estimated using HPLC data at 250 nm from a five-point calibration curve of picrocrocin in water. Repeatability of measurements, expressed as% CV (n = 7) was 2.3%. 3. Results and discussion
2.5. Apocarotenoid analysis in saffron extracts 3.1. Reference set 2.5.1. Extraction of crocins and picrocrocin Methanol–water extracts of ground stigmas were prepared according to the ISO 3632-2 method (ISO, 2010) with slight modifications. Briefly, 0.1 g was mixed with methanol–water (1:1, v/v) in a 200 mL volumetric flask. Crocins and picrocrocin were extracted by rigorous agitation (1000 rpm) for one hour at ambient temperature (25 °C) away from direct sunlight. Prior to analysis, an aliquot from the extract was diluted (1:10) with methanol– water (1:1, v/v) and the corresponding solutions were filtered through RC-55 filter (13 mm i.d., 0.45 lm pore size). 2.5.2. Spectrophotometric examination of saffron extracts The UV–Vis spectra were recorded in the region 200–600 nm using the Shimadzu UV 1601 (Kyoto, Japan) and processed with the aid of UVPC 1601 (Personal Spectroscopy Software, v.3.9, Shimadzu) software facilities. The ISO 3632-1 quality parameters 1% 1% E1% 1cm 440 nm, E1cm 257 nm, E1cm 330 nm were calculated, on dry basis, according to the equation:
E1% 1cm ¼ D 10000=mð100 HÞ
ð1Þ
E1% 1cm :
where the extinction coefficient of a 1% (w/v) solution when path length of the light is 1 cm, D: the absorbance value, m: the mass of the test portion (g) and H: the moisture and volatile content of the sample (% w/w). The latter was determined using the gravimetric method proposed in ISO 3632-2 (ISO, 2010). Since the required sample amount (2.5 g) was not available for all of the samples of the reference set, the H value used for mass correction, 10% w/w, was that of a representative mixture of stigmas from the 29 reference samples. 2.5.3. Chromatographic examination of saffron extracts HPLC-DAD analysis of crocins and picrocrocin in the methanol– water extracts of C. sativus L. was carried out as described by Sánchez, Carmona, Ordoudi, Tsimidou, and Alonso (2008) with some modifications. Briefly, apocarotenoid constituents were separated on a LiChroCART Superspher 100 RP-18 (125 4 mm i.d, 4 lm) end-capped column (Merck, KGaA, Darmstadt, Germany) after injection of a 20 lL aliquot and gradient elution with a mixture of water–acetic acid 1%, v/v, (A) -acetonitrile (B) (20–100% B in 20 min) at a flow rate of 0.5 mL/min. The HPLC system consisted of a pump (Thermo Separation Products, model P4000, San Jose, CA), a Midas autosampler (Spark, Emmen, The Netherlands), a UV 6000LP diode array detector (DAD; Thermo Separation Products) and a Spectra System SCM1000 on-line degasser (Thermo Separation Products). Monitoring was at 250 nm (picrocrocin) and 440 nm (crocins). Due to the lack of pure standards for each crocin, quantification was based on an equation recently applied by Sánchez et al. (2008). The equation was:
% of crocinðiÞ on dry basis ¼ Ai ðE1% 1cm 440nm=et;c Þ ðmi =10Þ
ð2Þ
Quality characteristics of the reference set samples were examined for compliance with the category I according to ISO 3632-1 specifications. The ranges for the calculated E1% 1cm values at 440 nm (219–304), 257 nm (79–104) and 330 mm (23–31) were above the minimum requirements for the category (200, 70 and 20, respectively), indicating that all of the samples were of the highest quality. Frequency distribution of these data showed that ca. 60% of the reference set samples differed among them by ±20 units of colouring strength, ±7 units of E1% 1cm 257 nm and ±2 units of E1% 1cm 330 nm values implying, thus, that the levels of the corresponding secondary metabolites, crocins, picrocrocin and safranal, should be of similar magnitude. 3.1.1. Spectral characteristics Reference set samples typically generated the FT-IR profile shown in Fig. 1. This spectrum was comprised of well-resolved peaks. Nine of them were automatically pinpointed in the characteristic group region (1800–1500 cm1) and less than five in the fingerprint region (1500–800 cm1). The spectra recorded in our work were found to be qualitatively similar to those reported by Anastasaki et al. (2010) using the micro-DRIFTS method of measurement. The longer analysis time needed in the present study because of the use of the KBr discs was counterbalanced by improved resolution of the IR peaks. Different operational modes (transmission vs diffuse reflectance) affect the signal-to-noise ratio in the spectra while other instrumental parameters (e.g. number of scans) as well as spectral data processing parameters (e.g. number of smoothing points) are expected to influence height and position of IR bands (Smith, 2011). For example, a distinct feature of the transmittance spectrum was that intensity ratio values of the band at about 1070 cm1 versus that at 1659 cm1 ranged between 1.2 and 1.8 (Fig. 1). The respective ratio value in the diffuse reflectance spectrum reported by Anastasaki et al. (2010) was close to one, signifying the advantage of the transmission method in quantitative studies. Another characteristic difference between our findings and those reported elsewhere was the position of the various bands in the fingerprint region (1500–800 cm1). These bands were recorded at ca. 20–30 cm1 lower than those reported in published data for diffuse reflectance infrared spectra of saffron nonpolar extracts. Table 1 summarizes the major bands in a typical FT-IR spectrum for saffron and assigns possible identities based on theory (Coates, 2000; Sun, 2009) and relevant reports. Taken that saffron is a complex mixture of chemical compounds, it is difficult to assign all of the bands directly to specific constituents. Nevertheless, since our aim was to identify bands having the strongest estimated correlation with storage effects, we investigated whether specific bands in the infrared spectra can be at least partially, associated with characteristic structural features of major metabolites that are expected to degrade in storage and are
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
417
Fig. 1. Typical FT-IR spectrum profile of a saffron sample belonging to the reference set.
Table 1 Major IR absorption bands and possible assignment in the typical FT-IR spectrum of saffron (Anastasaki et al., 2010; Coates, 2000; Kanou et al., 2005; Nikonenko et al., 2005; Sun, 2009). Band (cm1)*
Assignment
3365–3333 (broad) 2924 (m) and 2857 (m) 1745–1696 (sh) 1659–1653 (s) 1613 (m) 1580–1578 (m) and 1545–1542 (sh) 1454 (w) 1400 (w) and 1375 (w)
Stretching vibration of bonded and non-bonded –OAH groups Asymmetric –CH2–, symmetric –CH3 and –CH2– stretching vibrations –C@O stretching vibrations (e.g. in the –COOR groups of crocetin/–COOH groups of aminoacids) @C–H stretching vibrations (e.g. in the RCH@CHR’ groups of crocetin)/amide I/O–H bending vibrations in water C–C skeletal vibrations C–O vibrations (e.g. in the –COOR groups of crocetin)/amide II/aromatic –C@C stretching vibrations) C–H bending (scissoring) (in CH3 groups)/aromatic –C@C stretching vibrations –OH bending vibrations, –C–O–H in-plane bending vibrations,–CH3 out-of-plane bending vibrations, –CH2– wagging and twisting vibrations C(O)–O stretching vibrations and –OH in plane vibrations/amide III (e.g. in aromatic ethers)
1317 (w), 1294–1292 (w), 1271 (w) and 1227 (s) 1157 (sh) 1075 (s) 1020 (sh) 970–920 (sh) 780–700 (w) *
C–O stretching vibrations (e.g. in C–O–C glycosidic linkages of oligosaccharides or in triacylglycerols) C–1–H bending vibration in sugars C–4–OH (typical for glucose residue of disaccharides) trans = C–H out-of-plane bending cis = C–H out-of-plane bending
m: medium; s: strong; sh: shoulder; w: weak.
responsible for desirable sensory properties (colour, taste and aroma). To our knowledge, there are no published reports providing information about storage effects on other compounds that co-exist in saffron (e.g. proteins, sugars, lipids, etc.). 3.1.2. Selection of infrared bands using PCA To reduce the number of potent variables being at the same time consistent with general restrictions about sampling adequacy in PCA (Osborne & Costello, 2005), we used IR bands in the regions reported by Anastasaki et al. (2010) to be of diagnostic value considering also the information given in Table 1. These were: three bands in the carbonyl and double bond region (1744, 1701, 1655 cm1), eight bands associated with aromatic ring, peptide and glycosidic bonds (1580, 1549, 1456, 1375, 1317, 1271, 1227, 1157 cm1) as well as two bands related to sugars (1053, 1028 cm1) and one associated with the trans configuration in HC = CH moieties of apocarotenoid pigments (968 cm1). The average intensity values for each sample at these frequencies were calculated and introduced to the data matrix consisting of 29 objects (rows) and 14 variables (columns). To carry out Principal Component Analysis and retain an object-to-variable ratio of 3.0–5.0,
the correlation and anti-image correlation matrices as well as the results from Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s Test of sphericity were examined as explained in Section 2.5.2. These basic statistical tests showed that the bands at 1053 and 1028 cm1 were poorly correlated with those above 1300 cm1 and the bands above 1271 cm1 were strongly correlated with each other. On account of the KMO value for the whole set of variables (0.769) further statistical analysis was feasible. However, since KMO values for particular variables in the anti-image correlation matrix were lower than 0.6 the corresponding bands (1744, 1549, 1375 and 1271 and 1053 cm1) were removed leading to improved measures of sampling adequacy for the remaining dataset. A new matrix comprised of 29 objects 9 variables was thus used for PCA. The PCA resulted in high communalities for all variables (0.92–0.99). Two latent variables were found to represent 95% of the total variance (PC1 = 63%, PC2 = 32%). After rotation of the component matrix, absorbance intensities at 1701, 1655, 1580 and 1456 cm1 were found to be loaded on PC1 with positive directions. On the other hand, the corresponding values at 1028 cm1 loaded mainly on PC2 and contributed negatively to the formation
418
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
of PC1. The rest of the variables showed complex structures with moderate positive loadings on both principal components. Using the PC score values a two-dimensional scatterplot was constructed, as shown in Fig. 2. Irrespective of the harvest year, almost all of the reference set samples had score values between ±1.50. They were distributed equally between the (, +) and (+, +) quarters of the PC12 plot and also between the (+, ) and (, ) ones although most of the samples were located in the former two ones. As mentioned above, the observed variance in the scatterplot is actually the graphical representation of differences in the chemical composition of the samples. In Fig. 2, samples lying at the ±edges of the PC1 axis were found to differ by only 10 units of colouring strength while those lying at the ±edges of the PC2 axis differed by 35 units of colouring strength. Based on the frequency distribution data presented in 3.1., these differences could not account for distinct sub-groupings within the cloud of the scattered points in Fig. 2. Obviously, the ISO quality parameters provide quantitative information only about saffron content in apocarotenoids and not other compounds. Taken that high quality saffron is abundant in the former constituents, it is realistic to assume that the projection of the reference set samples onto the PC12 plane did not reflect any sub-grouping due to harvest year. Further observation of Fig. 2 shows samples collected in 2011 had only positive PC2 scores. As discussed earlier, the second PC is primarily formed due to the variance at 1028 cm1, which in turn, has been associated with the presence of bound a- or b-D glucose moieties in aqueous solutions of free disaccharides (Kanou, Nakanishi, Hashimoto, & Kameoka, 2005). Taking into consideration that saffron contains a relatively high amount of gentiobiose (6-O-b-D-glucopyranosyl-D-glucose) and glucose moieties (e.g. esterified to apocarotenoid skeleton), variance in the intensity at 1028 cm1 could also provide information about structural changes (e.g. breakage of glycosidic bonds) of these moieties. To gain insight into the parameters that influence variance at PC2 we examined the reference set composition in major crocins such as the trans- and cis- isomers of crocetin di-b-D-gentiobiosyl ester (t-4-GG and c-4-GG), the trans- crocetin (b-D-gentiobiosyl)-(b-Dglucosyl) ester (t-3-Gg) and picrocrocin (4-(b-D-glucopyranosyl)2,6,6-trimethyl- 1-cyclohexene-1-carboxaldehyde). HPLC-DAD analysis of water–methanol saffron extracts was carried out. On the basis of the estimated % w/w apocarotenoid content, it was found that samples located at the ±edges of the PC2 axis had
different HPLC quantitative profiles. For example, the lowest PC2 score value was obtained for a sample located in the (, ) quarter of the plot (x = 0.72, y = 2.47); it had high t-4-GG (>20.0% w/w), high t-3-Gg (>8.0% w/w), moderate c-4-GG levels (1.0% w/w) and was the richest in picrocrocin (>24.0%). On the other hand, the three samples with the highest PC2 score values (x, y) = (1.15 1.09), (1.05, 1.29) and (0.84, 1.50) were located diagonally opposite in the (+, +) quarter of the plot; they were found to contain moderate t-4-GG (19.5%, 21.0%, 18.5%, respectively) and c-4-GG levels (1.4%, 1.6%, 1.5%, respectively) but relatively low t-3-Gg (6.3%, 6.5%, 5.9%, respectively) and picrocrocin (15.4%, 14.1% and 16.3%, respectively). Summing up it can be deduced that differences in the apocarotenoid content account to a great extent for the observed variance along the PC2 axis. Multiple Linear Regression analysis was carried out to study possible correlations among the HPLC compositional data and the PC score values of the samples. Using the standard method of entering the independent variables to the analysis, two regression equations were produced:
PC1 ¼ 2:827 þ 0:082t 4 GG þ 0:257t 3 Gg þ 0:267c 4 GG 0:495pic ðr ¼ 0:60Þ
ð3Þ
PC2 ¼ 4:924 þ 0:140t 4 GG 0:460t 3 Gg 0:091c 4 GG 0:478pic ðr ¼ 0:72Þ
ð4Þ
The results suggest 35.7% and 52.3% of the variance in PC1 and PC2 values, respectively, was due to the variance in the quantities of all the major apocarotenoids. Using the stepwise regression method the independent variables most effective in predicting the dependent variable were sought, and are described by:
PC1 ¼ 1:885 0:505pic ðr ¼ 0:51Þ
ð5Þ
PC2 ¼ 2:269 0:631pic ðr ¼ 0:63Þ
ð6Þ
PC2 ¼ 5:431 0:525pic 0:344t 3 Gg ðr ¼ 0:71Þ
ð7Þ
Based on the correlation coefficient values in the Eqs. (5) and (6), picrocrocin content alone accounts for the 25.5% and 39.9% of the observed variance, respectively. Structural features of this compound such as the carbonyl bond and methyl groups, double bonds and hydroxyl groups in the glucose moiety, are likely to
Fig. 2. Scatterplot of PC scores for reference set (samples from 2010 and 2011 harvests).
S.A. Ordoudi et al. / Food Chemistry 150 (2014) 414–421
419
Fig. 3. FT-IR spectra of test set; (a) five (5) samples from individual producers and (b) ten (10) samples from blends of stigmas obtained from different producers.
produce IR vibration bands contributing to the PCs as described by Eqs. (5) and (6). The negative sign of the correlation coefficients suggests that decrease in picrocrocin content causes increase in the score values. The trans- and cis- isomers of crocetin di-(b-D-gentiobioside) were not shown to be significant predictors. The symmetry of the former molecules could play a role to this result as it reflects the relatively small number of IR active vibration bands (Sun, 2009). On the other hand, the t-3-Gg content was found to improve the regression coefficient value and assist the prediction of PC2 score values (Eq. (7)). 3.1.3. Evaluation sets For evaluation of the PCA findings, a set of six samples was first examined. The samples had been stored for more than a year under conditions that are not likely to bring about significant changes to the quality characteristics of the spice (e.g. reduction in colouring strength is expected to be
