Using the liquid-chromatographic-fingerprint of sterols fraction to discriminate virgin olive from other edible oils

Journal of Chromatography A, 1380 (2015) 64–70

Contents lists available at ScienceDirect

Journal of Chromatography A journal homepage: www.elsevier.com/locate/chroma

Using the liquid-chromatographic-fingerprint of sterols fraction to discriminate virgin olive from other edible oils ˜ M. Sánchez-Vinas, ˜ M.G. Bagur-González ∗ , E. Pérez-Castano, D. Gázquez-Evangelista Department of Analytical Chemistry, Faculty of Sciences, University of Granada, Avda. Fuentenueva s/n, 18071 Granada, Spain

a r t i c l e

i n f o

Article history: Received 17 June 2014 Received in revised form 6 December 2014 Accepted 18 December 2014 Available online 29 December 2014 Keywords: Sterols fraction LC fingerprint Edible oils Savitsky–Golay filtering Unsupervised and supervised pattern recognition methods

a b s t r a c t A method to discriminate virgin olive oil from other edible vegetable oils such as, sunflower, pomace olive, rapeseed, canola, corn and soybean, applying chemometric techniques to the liquid chromatographic representative fingerprint of sterols fraction, is proposed. After a pre-treatment of the LC chromatogram data – including baseline correction, smoothing signal and mean centering – different unsupervised and supervised pattern recognition procedures, such as principal component analysis (PCA), hierarchical cluster analysis (HCA), and partial least squares-discriminant analysis (PLSDA), have been applied. From the information obtained from PCA and HCA, two groups can be clearly distinguished (virgin olive and the rest of vegetable oils tested) which have been used to discriminate between two defined classes by means of a PLSDA model. Five latent variables (LVs) explained 76.88% of X-block variance and 95.47% of the defined classes block (-block) variance. A root mean square error for calibration and cross validation of 0.10 and 0.22 respectively, confirmed these results and a root mean square error for prediction of 0.15 evidences that the classification model proposed presents an adequate prediction capability. The contingency table also shows the good performance of the model, proving the capability of the LC-R-FpM, to discriminate virgin olive from other vegetable edible oils. © 2014 Elsevier B.V. All rights reserved.

1. Introduction According to several authors [1–3], a fingerprint can be defined as a characteristic profile reflecting the complex chemical composition of an analyzed sample and can be obtained by spectroscopic, chromatographic or electrophoretic techniques. The chromatographic methods are able to characterize the chemical composition of samples using the chromatographic signal as a fingerprint which describes a sample as unique (as a human fingerprint) [4]. In this sense, the comparison among chromatographic fingerprints (FINGERPRINTING) [5–7] can be used to uncover or explain the variability caused by differences in the chemical composition of samples, being useful for quality assessment and authentication of them [8–11]. Fingerprinting appears first in authentication of foods in the Project TRACE (FP6-2003-FOOD-2-A): Tracing Food Commodities in Europe [12], in which food analysis by fingerprinting techniques is described as part of Work Package 2 in analytical tools group. These techniques describe a variety of analytical techniques which

∗ Corresponding author. Tel.: +34 958 243 327; fax: +34 958 243 328. E-mail address: [email protected] (M.G. Bagur-González). http://dx.doi.org/10.1016/j.chroma.2014.12.052 0021-9673/© 2014 Elsevier B.V. All rights reserved.

can measure the composition of some foods in a non-selective way. The information obtained, defined as instrumental fingerprint (InF), is a signal (i.e. a spectrum or a chromatogram which are function of the chemical composition and have different specificity grade) provided and registered by an analytical instrument that requires a mathematical treatment, normally a chemometric approach including projection, clustering, modelling techniques, etc. [13–15] with the objective of characterizing the food. In the last years the olive oil has won in popularity not only by its quality, but also for the potential benefits for health derived from its consumption. The interest in its chemical composition has increased due to different reasons among them, to assure its quality and origin, to guarantee the fulfilment of the current regulations and to detect possible adulterations or frauds. The olive oil can be defined as a matrix of great complexity and diversity and some of the methods that are in use for its analysis have been adopted and regulated by official institutions as the International Olive Oil Council (IOOC) or the Codex Alimentarius Commission [16,17]. The European Union (EU) [18,19] has published two regulations about the commercialization of olive oil and olive oil mixed with other vegetable edible oils. In the labelling of these products, the presence of olive oil must be indicated when its percentage is higher than 50%. Therefore, it would be interesting to have methods that

M.G. Bagur-González et al. / J. Chromatogr. A 1380 (2015) 64–70

allow establishing the presence of olive oil based on easily obtained instrumental fingerprints that, not being necessary to quantify, allow discriminating between this type of oil and other edible oils. The oil fraction which permits to differentiate a type of oil from other is known as unsaponifiable fraction. In the olive oils it covers the 1–2% of its total content and it is rich in minor metabolites such as, triterpene dialcohols, phenolic compounds, tocopherols, hydrocarbons, pigments, terpenic acids, mono- and diacylglycerols, etc., being sterols the major proportion of this fraction and the most important olive oil minor compounds in authentication purposes [20]. In relation with these analytes, most of the chromatographic studies, as far as we concerned, have been made using the area/height of peaks (peak profiling) or concentration data (compositional profiling) rather than using raw data (InF). The most well-known chromatographic fingerprinting is focused on the utilization of well resolved InF [21–29]. In this paper, from poorly resolved chromatograms of unsaponifiable of different edible vegetable oils, the zone in which the sterols fraction appears, has been used as raw data to obtain the samples fingerprint matrix. A chemometric approach on the “representative fingerprint matrix” (R-FpM) obtained after a pretreatment of the chromatograms (base line correction, signal smoothing, and mean centering), using different unsupervised and supervised pattern recognition techniques, has been applied to discriminate virgin olive oil from the other oils studied.

2. Materials and methods 2.1. Chemicals and reagents To do the saponification of the oils, a 2 M ethanolic solution of potassium hydroxide Panreac (Barcelona, Spain), was prepared by dissolving 130 g in 200 mL of distilled water and then, made up to 1 L with ethanol (96% (v/v), Panreac). This solution once prepared, must be kept in a well-stoppered dark glass bottle. Other reagents, as diethyl ether (98% purity), anhydrous sodium sulphate (99.5% purity) and acetone (99.5% purity), were purchased from Panreac and all of them were of analytical grade. HPLC-grade solvents (hexane and tert-butylmethyl ether (TBME)) were from Merck (Darmstadt, Germany).

65

2.4. Analytical procedures 2.4.1. Sample preparation 5 g of oil sample were saponified with 50 mL of a 2 M ethanolic potassium hydroxide solution by refluxing, at approximately, 80 ◦ C, with constant shaking until the solution was clarified (ca. 1 h). The unsaponifiable fraction was extracted with ethyl ether, the extract dried with anhydrous sodium sulphate, and the solvent was evaporated to dryness in a rotary evaporator at 30 ◦ C. The resulting residue was dissolved in 1 mL of mobile phase (80:20 v/v, n-hexane:TBME). Then, 20 ␮L of this solution were injected into the HPLC system. 2.4.2. HPLC-UV analysis The chromatogram of the unsaponifiable fraction was obtained by HPLC [30], using a Lichrospher 100 CN column (250 mm × 4.0 mm i.d., 5 ␮m) with a Lichrospher guard column (10 mm × 4.0 mm i.d., 5 ␮m) from Merck, a mobile phase constituted by hexane:TBME, 80:20 (v/v), a flow rate of 0.8 mL min−1 and a detection wavelength of 208 nm [31,32]. The chromatograms were exported to csv format from ChemStation software and imported to MATLAB version 7.8.0347 R2009a (Mathworks Inc., Natick, MA, USA) to handle chromatographic data matrices. 2.5. Construction of the LC-fingerprints In a previous work [30], it could be observed that the LCchromatogram of the unsaponifiable fraction of an edible vegetable oil was a bad resolved chromatogram, morphologically dependent of the type of oil analyzed. The last registered peak, assigned to the sterols fraction (that comprised the zone of the chromatogram between retention times 7.5 and 9.5 min), could be considered as a specific fingerprint of the sterols fraction as well as the edible oil. Fig. 1 shows the procedure used for the construction of the oil fingerprint based on its sterols fraction. It can be seen that for each chromatogram, a data matrix IM“j (retention time × absorbance), which initially had a dimension of 6916 × 2 variables, was obtained. Once the matrices were combined and transposed, all the TR rows were eliminated. For each sample, the variables related with the sterols fraction (2001), were selected, generating a fingerprint matrix (FpM) with a dimension of 51 × 2001. 2.6. Chemometric approach

2.2. Samples Fifty-one trademark edible vegetable oils, from different origins, i.e. Spain, Mexico, France and USA, were purchased in local markets or gourmet shops. The following oils (code, number of samples studied) were analyzed: Virgin olive (VOO, 25), pomace olive (POO, 4), sunflower (SFO, 11), rapeseed (RO, 3), canola (CanO, 2), soybean (SyO, 3) and corn (CoO, 3).

The chemometric study was made using principal component analysis (PCA), as exploratory data analysis [13,14,26,33,34], combined with hierarchical cluster analysis (HCA) [35–37] to confirm the results obtained previously. Finally, a partial least squaresdiscriminant analysis (PLSDA) [38–42] was used to discriminate virgin olive oil from other edible oils. All chemometric treatment was performed using PLS Toolbox Version 7.0.3. (Eigenvector Research, Inc., West Eaglerock Drive, Wenatchee, WA).

2.3. Instrumentation 3. Results and discussion The analysis was performed with a HPLC 1050 Series Chromatograph equipped with an UV–visible variable wavelength detector (VWD) Agilent Technologies (Palo Alto, CA, USA), and a Reodhyne (Reodhyne, Inc., Cotati, CA, USA) 7125 loop injector provided with a 20 ␮L sample loop. The main characteristics of the VWD used were: wavelength range, 190–600 nm; wavelength accuracy, ±2 nm; wavelength reproducibility, ±0.3 nm; band width, 6.5 nm; response time, 1 s. The software used for acquisition and handling of the chromatographic data was an Agilent ChemStation (Rev. A.08.03).

3.1. Obtaining the representative fingerprint matrix (R-FpM) To obtain the representative fingerprint matrix (R-FpM), a tree steps pre-processing of the FpM (51 × 2001) was carried out (see Fig. 1) using a baseline correction, a signal smoothing and finally a mean centering. (i) In order to separate the analytical signal of interest from the signals due to other factors, a baseline correction was made.

66

M.G. Bagur-González et al. / J. Chromatogr. A 1380 (2015) 64–70

Fig. 1. General procedure used for the construction of the representative-fingerprint matrix (R-FpM) of edible oils sterols fraction.

The algorithm applied fitted the variables to a first order polynomial function, using the average response obtained in regions where no analytes peaks eluted. Once this polynomial was determined, the fitted line was subtracted from each row of the FpM [43]. (ii) As the sterols fingerprints had some instrumental noise, to reduce it, a smoothing using the Savitzky–Golay algorithm [44] was applied. Smoothing assumes that variables closely related, i.e. located in adjacent columns, contain similar information. This fact can be used to reduce noise without significant losses of the signal of interest, averaging variables contained in a determined region called “window”. In this case, it was made selecting for each reduced chromatogram a “window” of 31 points, which were fitted to a second order polynomial function. (iii) The MC pre-processing was performed subtracting, point by point, the average data vector of all the rows in smoothed FpM from each row vector in the data set.

3.2. Unsupervised pattern recognition methods applied to the sterols fraction fingerprint In order to obtain the most useful information from the R-FpM and to evaluate the grouping of different oils using the defined sterols fraction fingerprints, the next analyses were made.

3.2.1. Principal component analysis On the basis of eigenvalues >1 criterion, from the singular value decomposition (SVD) algorithm [45] applied to the auto-scaled data, four principal components were extracted explaining about 78% of the total variance. Thus, in some way, all of the edible oils profiles represented by the 2001 points of each representative fingerprint, in a reduced space defined by four sets of calculated PCs,

are explained. Fig. 2 shows the scatter plot for the first and second principal components. According to PC1 two main groups of samples can be observed: (i) Group I, which contains all the virgin olive oils samples, is characterized by negative scores for this variable and (ii) Group II, which contains the rest of the oil samples analyzed (sunflower, rapeseed, pomace olive, corn and soybean) is, in general terms, characterized by positive or negative closely to zero scores for PC1. The PC1 loadings are shown in Fig. 3 where, it can be observed that the most significant part of the R-FpM is the one comprised between the variables 1000 and 1500 because is the zone where the differences among the signals are the highest. From a chemical point of view it can be explained considering: (a) the sterols fraction in the virgin olive oils is always lesser than in the others, and (b) for the rest of edible oils, the sterols content depends on the type of oil.

3.2.2. Hierarchical cluster analysis The Dendrogram obtained from the R-FpM, using the Euclidean distance of the scores obtained from PCA as similarity criterion and Ward method as agglomerative rule, is represented in Fig. 4. Using a Dlinkage = 0.66 Dmax as criterion of selection of the number of groups, it can be seen that two groups or clusters are obtained: Cluster I (virgin olive samples) and Cluster II (rest of edible oils analyzed). This last one is characterized by the biggest Euclidean distance (high significance clustering). The clusters obtained are in complete agreement with the groups previously defined by PCA. Comparing both figures, it can be observed that in relation to Group 2 (Cluster II), it would be also possible to consider two subgroups, the first constituted for sunflower oils, and the second for the rest (pomace olive, soy bean, corn and canola/colza oils). From a chemical point of view, the two clusters can be also justified considering the sterols content, which is lower in virgin olive

M.G. Bagur-González et al. / J. Chromatogr. A 1380 (2015) 64–70

Fig. 2. Scatter plot for PC1 and PC2 obtained when PCA is applied to representative-fingerprint matrix (R-FpM). Legend: VOO ( (

) and SYO (

67

), CanO/RO ( ), SFO (

). Samples coded as in Section 2.2.

Fig. 3. (a) Pre-processed chromatograms of the analyzed edible oils. (b) Loading plot of PC1.

), POO (

), CoO

68

M.G. Bagur-González et al. / J. Chromatogr. A 1380 (2015) 64–70

and therefore the band of the sterol fraction is always smaller in this type of oil. In our opinion, these facts justify the use of this kind of representative fingerprints to differentiate the virgin olive oil from the rest. 3.3. Supervised pattern recognition methods applied to the sterols fraction fingerprint According to the results obtained by PCA and HCA analyses, a PLSDA was carried out in order to discriminate between the two main classes assigned, i.e., Class 1:virgin olive and Class 2: other edible oils. From the R-FpM (51 × 2001), two X“I -blocks were designed, one for training and validation set (modelling step) and other for testing the model (prediction step) The samples considered in each X“I block were selected in a randomized way. In the Xmodelling block, 19 rows correspond to virgin olive oils, whereas the other rows correspond to 19 representative fingerprints of the other edible oils (sunflower 9, pomace olive 3, rapeseed, soybean and corn oil 2, respectively, and finally, canola 1). In the Xprediction block, 6 rows correspond to virgin olive oils and 7 correspond to the other edible oils. Thus the X blocks were: Fig. 4. Dendrogram of analyzed oils clustered according to its representativefingerprint matrix (R-FpM).

Table 1 Figures of merit for the PLSDA discrimination model.

(1) Calibration BiasCalibration RMSECbias R2

Class 1a

Class 2b

∼ =0 (2 × 10−16 ) 0.106 0.955

∼ =0 (−1 × 10−16 ) 0.106 0.955

(2) Crossvalidation (venetian blind, W-split = 6) −0.019 BiasCrossvalidation RMSECVbias 0.229 0.797 R2 (3) Prediction BiasPrediction RMSEPbias R2 a b

0.019 0.229 0.797

−0.019 0.152 0.912

−0.019 0.152 0.912

Class 1: virgin olive. Class 2: other edible oils.

- An Xmodelling block (38 × 2001), for the fitting and cross validation of the discrimination model. Venetian blinds with a W-split of 6 were used for the internal cross-validation of the fitted model - An Xprediction block (13 × 2001), used to validate the prediction capability of the discriminant model previously validated. The Y-blockmodelling (38 × 2001) was designed as a dummy or integer matrix with two columns: Class 1, virgin olive; Class 2, other edible oils. For each sample, the columns were fulfilled with ones (for the sample class) and zeroes (for the other). The model was fitted selecting five LVs which explained 76.88% of X-block and 95.47% of Y-block variances. In order to properly evaluate the discriminant capability of the model according to da Silva et al. [46,47], an estimation of the bias associated to each step was made. These values were used to obtain the accurate values of the root mean squares errors for calibration (RMSEC), cross-validation (RMSCV) and prediction (RMSPred.), respectively. Table 1 shows the main figures of merits for the discrimination model obtained, being the most significant, for calibration, the root-mean-squared error ((RMSECbias ) over 0.11) and the R2 > 0.95 and for prediction, the root-mean-squared error ((RMSEPbias ) over 0.15) and the R2 > 0.91.

Table 2 2 × 2 Contingency table for the PLSDA model based on representative LC fingerprints of edible oils. Class 1a (virgin olive)

Class 2a (other edible oils)

(1) Calibration Class 1 (virgin olive)b Class 2 (other edible oils)b

19 0

0 19 Number of samples in the Xmodelling matrix: 38 Overall % CC: 100%

(2) Cross-validation Class 1 (virgin olive) Class 2 (other edible oils)

18 1

1 18 Number of samples in the Xmodelling matrix: 38 Overall % CC: 92.31%

(3) Prediction Class 1 (virgin olive) Class 2 (other edible oils)

6 0

0 7 Number of samples in the Xprediction matrix: 13 Overall % CC: 100%

a b

True classes. Predicted classes.

Probability (%) 100 0

94.74 5.26

100 0

M.G. Bagur-González et al. / J. Chromatogr. A 1380 (2015) 64–70

The 2 × 2 contingency table (Table 2) also shows the good performance of the model, proving the capability of the LC-R-FpM, to discriminate virgin olive from other vegetable edible oils. This table shows according to Brereton [34] the probabilities (expressed as percentage) for each defined class, calculated from the likelihood ratios. Finally, although the values of probability for each class as well as the percentage of overall correctly classified samples (overall % CC) in the cross validation step, are lesser than the ones obtained for the calibration and prediction steps, the fact that the model can classify correctly external samples (not used to establish the model) confirms that an adequate pre-processing of poorly resolved chromatograms, together with a selection of the most representative zone, can be used to discriminate olive oil. An increase in the number of samples, including mixtures of virgin olive with other edible oils, could be adequate to quantify virgin olive oil following the requirements established by the European Community rules. 4. Conclusions

[10]

[11]

[12] [13]

[14]

[15] [16] [17] [18] [19]

From the unsaponifiable fraction chromatograms of some types of common edible oils (virgin and pomace olive, sunflower, soybean, canola, rapeseed, and corn), a representative LC-fingerprint has been obtained, using the zone corresponding to the sterols fraction. The treatment of the R-FpM with different chemometric tools has permitted to establish a discrimination-prediction model capable to distinguish virgin olive from the rest of edible analyzed oils. The whole chemometric study, establishes that this type of “representative fingerprint”, obtained from a poorly resolved chromatogram, is capable of discriminating virgin olive oils from the rest, not being necessary neither the chromatographic separation nor an ulterior quantification of the main phytosterols associated to this type of samples. Acknowledgements The authors are grateful to the Consejería de Educación y Ciencia de la Junta de Andalucía for financial assistance (Research Group FQM 0232).

[20]

[21]

[22]

[23]

[24]

[25]

[26]

[27]

[28]

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.chroma.2014.12.052.

[29]

[30]

References [1] C. Tistaert, B. Dejaegher, Y.V. Heyden, Chromatographic separation techniques and data handling methods for herbal fingerprints: a review, Anal. Chim. Acta 690 (2011) 148–161. [2] E. Deconinck, P.Y. Sacre´ı, P. Courselle, J.O. De Beer, Chromatography in the detection and characterization of illegal pharmaceutical preparations, J. Chromatogr. Sci. 51 (2013) 791–806. [3] A.D. McNaught, A. Wilkinson, IUPAC Compendium of Chemical Terminology, second ed., Royal Society of Chemistry, Cambridge, UK, 1997. [4] M. Daszykowskia, Y. Vander Heydenb, C. Bouconc, B. Walczaka, Automated alignment of one-dimensional chromatographic fingerprints, J. Chromatogr. A 1217 (2010) 6127–6133. [5] P. Daisa, E. Hatzakisb, Quality assessment and authentication of virgin olive oil by NMR spectroscopy: a critical review, Anal. Chim. Acta 765 (2013) 1–27. [6] O. Fiehn, Combining genomics, metabolome analysis, and biochemical modelling to understand metabolic networks, Comp. Funct. Genomics 2 (2001) 155–168. [7] D.I. Ellis, W.B. Dunn, J.L. Griffin, J.W. Allwood, R. Goodacre, Metabolic fingerprinting as a diagnostic tool, Pharmacogenomics 8 (9) (2007) 1243–1266. [8] J. Trygg, E. Holmes, T. Lundstedt, Chemometrics in metabonomics, J. Proteome Res. 6 (2007) 469–479. [9] N. Hakimzadeha, H. Parastara, M. Fattahi, Combination of multivariate curve resolution and multivariate classification techniques for comprehensive high-performance liquid chromatography-diode array absorbance detection

[31] [32]

[33] [34] [35] [36]

[37]

[38] [39]

69

fingerprints analysis of Salvia reuterana extracts, J. Chromatogr. A 1326 (2014) 63–72. H. Parastar, M. Jalali-Heravi, H. Sereshti, A. Mani-Varnosfaderani, Chromatographic fingerprint analysis of secondary metabolites in citrus fruits peels using gas chromatography–mass spectrometry combined with advanced chemometric methods, J. Chromatogr. A 1251 (2012) 176–187. J.M. Amigo, M.J. Popielarz, R.M. Callejón, M.L. Morales, A.M. Troncoso, M.A. Petersen, T.B. Toldam-Andersen, Comprehensive analysis of chromatographic data by using PARAFAC2 and principal components analysis, J. Chromatogr. A 1217 (2010) 4422–4429. P. Brereton, J. Krommer, TRACE: Tracing the Origin of Food (Contract no: 006942), www.trace.eu.org (available in October 2014). D.L. Massart, B.G.M. Vandeginste, L.M.C. Buydens, S. De Jong, P.J. Lewi, J. Smeyers-Verbeke (Eds.), Handbook of Chemometrics and Qualimetric, Part A, Elsevier, Amsterdam, 1997. B.G.M. Vandeginste, D.L. Massart, L.M.C. Buydens, S. De Jong, P.J. Lewi, J. Smeyers-Verbeke (Eds.), Handbook of Chemometrics and Qualimetric, Part B, Elsevier, Amsterdam, 1998. L.A. Berrueta, R.M. Alonso-Salces, K. Héberger, Supervised pattern recognition in food analysis, J. Chromatogr. A 1158 (2007) 196–214. http://www.internationaloliveoil.org/estaticos/view/224-testing-methods (available in November 2014). www.codexalimentarius.org/ (available in November 2014). Reglamento (UE) N◦ 1169/2011 del Parlamento Europeo y del Consejo de 25 de octubre de 2011. Reglamento de Ejecución (UE) N◦ 29/2012 de la comisión de 13 de enero de 2012. R. Aparicio, M.T. Morales, R. Aparicio-Ruiz, N. Tena, D.L. García-González, Authenticity of olive oil: Mapping and comparing official methods and promising alternatives, Food Res. Int. 54 (2013) 2025–2038. ˜ M. Martínez, Sterol and erythrodiol + uvaol J. Sánchez, E. Osorio, A.M. Montano, content of virgin olive oils from cultivars of Extremadura (Spain), Food Chem. 87 (2004) 225–230. ˜ M.V. Ruíz-Méndez, A. Garrido-Fernández, Sterols, A. López-López, A. Montano, fatty alcohols, and triterpenic alcohols in commercial table olives, J. Am. Oil Chem. Soc. 85 (2008) 253–262. A. López-López, F. Rodríguez-Gómez, M.V. Ruiz-Méndez, A. Cortes-Delgado, A. Garrido-Fernández, Sterols, fatty alcohol and triterpenic alcohol changes during ripe table olive processing, Food Chem. 117 (2009) 127–134. C. Crews, P. Hough, J. Godward, P. Brereton, M. Lees, S. Guiet, W. Winkelmann, Study of the main constituents of some authentic walnut oils, J. Agric. Food Chem. 53 (2005) 4853–4860. C. Mariani, G. Bellan, E. Lestini, R. Aparicio, The detection of the presence of hazelnut oil in olive oil by free and esterified sterols, Eur. Food Res. Technol. 223 (2006) 655–661. R. Bucci, A.D. Magrí, A.L. Magrí, D. Marini, F. Marini, Chemical authentication of extra virgin olive oil varieties by supervised chemometric procedures, J. Agric. Food Chem. 50 (2002) 413–418. Y. Ouni, G. Flamini, N.B. Youssef, M. Guerfel, M. Zarrouk, Sterolic composition and triacylglycerols of Oueslati virgin olive oil: comparison among different geographic areas, Int. J. Food Sci. Technol. 46 (2011) 1747–1754. J. Parcerisa, I. Casals, J. Boatella, R. Codony, M. Rafecas, Analysis of olive and hazelnut oil mixtures by high-performance liquid chromatography–atmospheric pressure chemical ionization mass spectrometry of triacylglycerols and gas–liquid chromatography of non-saponifiable compounds (tocopherols and sterols), J. Chromatogr. A 881 (2000) 149–158. K. Nagy, D. Bongiorno, G. Avellone, P. Agozzino, L. Ceraulo, K. Vékey, High performance liquid chromatography–mass spectrometry based chemometric characterization of olive oils, J. Chromatogr. A 1078 (2005) 90–97. ˜ ˜ D. Gázquez-Evangelista, E. Pérez-Castano, M. Sánchez-Vinas, M.G. BagurGonzález, Using offline HPLC-GC-FID 4-desmethylsterols concentration profiles, combined with chemometric tools, to discriminate different vegetable oils, Food Anal. Methods 7 (2014) 912–925. S.L. Abidi, Chromatographic analysis of plant sterols in foods and vegetable oils, J. Chromatogr. A 935 (2001) 173–201. M. Careri, L. Elviri, A. Mangia, Liquid chromatography–UV determination and liquid chromatography–atmospheric pressure chemical ionization mass spectrometric characterization of sitosterol and stigmasterol in soybean oil, J. Chromatogr. A 935 (2001) 249–257. S. Wold, K. Esbensen, P. Geladi, Principal component analysis, Chemom. Intell. Lab. Syst. 2 (1987) 37–52. R.G. Brereton, Chemometric for Pattern Recognition, John Willey & Sons, Chichester, UK, 2009. D.L. Massart, L. Kaufman, The Interpretation of Analytical Data by the Use of Cluster Analysis, John Wiley, New York, 1983. K.I. Poulli, G.A. Mousdis, C.A. Georgiou, Classification of edible and lampante virgin olive oil based on synchronous fluorescence and total luminescence spectroscopy, Anal. Chim. Acta 542 (2005) 151–156. M. De Luca, W. Terouzi, G. Ioele, F. Kzaiber, A. Oussama, F. Oliverio, R. Tauler, G. Ragno, Derivative FTIR spectroscopy for cluster analysis and classification of morocco olive oils, Food Chem. 124 (2011) 1113–1118. D. Ballabio, V. Consonni, Classification tools in chemistry. Part 1: linear models. PLS-DA, Anal. Methods 5 (2013) 3790–3798. E. Anderssen, K. Dyrstad, F. Westad, H. Martens, Reducing over-optimism in variable selection by cross-model validation, Chemom. Intell. Lab. Syst. 84 (2006) 69–74.

70

M.G. Bagur-González et al. / J. Chromatogr. A 1380 (2015) 64–70

[40] M. Forina, P. Oliveri, S. Lanteri, M. Casale, Class-modeling techniques, classic and new, for old and new problems, Chemom. Intell. Lab. Syst. 93 (2008) 132–148. [41] T. Naes, C. Irgens, H. Martens, Comparison of linear statistical methods for calibration of NIR instruments, Appl. Stat. 35 (2) (1986) 195–206. [42] M. Forina, M. Casale, P. Oliveri, Application of chemometrics to food chemistry, in: S.D. Brown, R. Tauler, B. Walczak (Eds.), Comprehensive Chemometrics: Chemical and Biochemical Data Analysis. Vol. 4: Applications, Elsevier, Amsterdam, 2009, pp. 75–128. [43] J.J. Harynuk, A.P. de la Mata, N.A. Sinkov, Application of chemometrics to the interpretation of analytical separations data, in: K. Varmuza (Ed.), Chemometrics in Practical Applications, InTech, 2012, pp. 305–326. [44] A. Savitzky, M.J.E. Golay, Smoothing and differentiation of data by simplified least squares procedures, Anal. Chem. 36 (1964) 1627–1639.

[45] M.E. Wall, A. Rechtsteiner, L.M. Rocha, Singular value decomposition and principal component analysis, in: D.P. Berrar, W. Dubitzky, M. Granzow (Eds.), A Practical Approach to Microarray Data Analysis, Springer, New York, 2003, pp. 91–109. [46] V.A.G. da Silva, M. Talhavini, J.J. Zacca, B.R. Tridade, J.W.B. Braga, Discrimination of black pen inks on writing documents using visible reflectance spectroscopy and PLS-DA, J. Braz. Chem. Soc. 25 (9) (2014) 1552–1564. [47] V.A.G. da Silva, M. Talhavini, I.C.F. Peixoto, J.J. Zacca, A.O. Maldaner, J.W.B. Braga, Non-destructive identification of different types and brands of blue pen inks in cursive handwriting by visible spectroscopy and PLS-DA for forensic analysis, Microchem. J. 116 (2014) 235–243.