- Email: [email protected]
Usefulness of the direct coupling headspace–mass spectrometry for sensory quality characterization of virgin olive oil samples
Analytica Chimica Acta 583 (2007) 411–417
Usefulness of the direct coupling headspace–mass spectrometry for sensory quality characterization of virgin olive oil samples Silvia L´opez-Feria b , Soledad C´ardenas a , Jos´e Antonio Garc´ıa-Mesa b , Miguel Valc´arcel a,∗ a
Department of Analytical Chemistry, Marie Curie Building (Annex), Campus de Rabanales, University of C´ordoba, E-14071 C´ordoba, Spain b CIFA Venta del Llano, IFAPA, Ctra. Bail´ en-Motril km 18.5 Mengibar, E-23620 Ja´en, Spain Received 14 July 2006; received in revised form 28 September 2006; accepted 9 October 2006 Available online 26 October 2006
Abstract The applicability of the headspace coupled to mass spectrometry for evaluation of the sensory quality of virgin olive oil samples is presented. The volatiles of the oil are directly transferred from the sample vial to the detector without chromatographic separation. The mass spectrum obtained can be considered as a fingerprint of the oil sample and can be used for classification purposes. After a training step with samples previously qualified following the official method, a classification model was created using the supervised technique soft independent modeling of class analogy (SIMCA). Eight negative (rancid, winey-vinegary, muddy sediment, hay-wood, vegetable water, earthy, fusty and musty-humidity) and three principal positive attributes (fruity, bitter and pungent) have been included in this study. With them, a classification model consisting of two main classes (extra- and lampante-virgin olive oil) was constructed. In addition, the unsupervised technique cluster analysis permited the discrimination among oils with different negative attributes. The proposed procedure has been applied to the classification of commercial samples (as extra- or lampante-virgin olive oils) and the results were compared with those provided by the expert’s panel with acceptable correlation. © 2006 Elsevier B.V. All rights reserved. Keywords: Headspace; Mass spectrometry; Olive oil sensory quality; Direct analysis; Chemometric techniques
1. Introduction Virgin olive oil, extracted by purely mechanical procedures from the fruit of the olive tree, Olea europaea L., is one of the basic components of the Mediterranean diet. It can be classified as edible (extra-virgin and virgin) and non-edible (lampante) virgin olive oils [1]. Its largely demonstrated benefits for human health constitute an added value of this food, which clearly affects the price it can reach in the market. Virgin olive oil can be marketed under a variety of denominations, which can refer to the variety of olive used for its production or its sensorial attributes. Flavors and aroma of the oils are generated by a number of volatile constituents which are present at very low concentrations, which in most cases affects the quality of the virgin olive oil and therefore, its price [2]. The sensory evaluation of these samples is usually carried out by an experts’ panel which allows the characterization of the olive oil within a sin-
∗
Corresponding author. Tel.: +34 957 218 616; fax: +34 957 218 616. E-mail address: [email protected] (M. Valc´arcel).
0003-2670/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2006.10.027
gle analysis. The attributes detected can be either positives or negatives, being the latter consequence of a bad collection process, olive/oil storage or mixture with other oils of lower quality [3]. However, this method is lengthy and expensive, whose final results depends on the panelists’ training and the specific vocabulary used in sensory analysis. Several alternatives oriented at overcoming these disadvantages have been recently published. Most of them involve the direct analysis of the samples by means of dynamic headspace (HS) and mass spectrometry (MS), with or without chromatographic separation. The applicability of the direct coupling of a HS module to a mass spectrometer in food analysis has been recently reviewed [4]. The main advantage of this configuration is that the sample can be directly analyzed without any pretreatment. The mass spectrum obtained provides a characteristic fingerprint of the volatile compounds of the sample. The subsequent chemometric treatment of the spectral information permits sample characterization once the instrument has been properly trained. In the field of olive oil analysis, the HS–MS coupling has been proposed for the determination of contaminants [5,6], adulteration with other edible oils [7,8] and classification purposes.
412
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
Concerning this last topic, HS–MS has been applied to the classification of the three main types of olive oils [9], the characterization of monovarietal olive oils [10] and the discrimination of different geographical origin olive oils [11]. The inclusion of a gas chromatograph (GC) within this configuration permits the measurement of retention times and possible identification of the compounds present in the sample which are responsible for the property being modeled [12–14]. The evaluation of olive oil sensory defects (winey-vinegary, musty-humidity, fusty and rancid) has been afforded by using both HS–GC-FID and MS and HS–GC-olfactometry [15] and metal-oxide sensors [16]. Moreover, a HS–GC–MS method has been proposed to assign the chromatographic peaks obtained to the defects detected by a panel test. In this case, only four defects (muddy sediment, fusty, winey-vinegary and musty-humidity) were included [17]. Finally, an HS–GC method allowed the identification of the volatile compounds responsible for the pleasant sensory perception as well as the undesirable flavor from Lianolia variety [18]. In this paper, the direct use of the HS–MS for the rapid determination of the quality of virgin olive oil samples, based on eight defects (rancid, winey-vinegary, muddy sediment, hay-wood, vegetable water, earthy, fusty and mustyhumidity) and three principal positive attributes (fruity, bitter and pungent), is proposed as an alternative to the experts’ panel. The classification model, which was used to discriminate among extra- and lampante-virgin olive oil, was constructed by using 13 defective and 9 high quality monovarietal virgin olive oil samples, previously qualified by the corresponding experts’ panel. Based on the previous experience of our research group [5], it was concluded that SIMCA is more adequate for the objective of this work and therefore, it was applied to the training set including a cross-validation step. The optimized model was used for the classification of a new set of virgin olive oil samples and the results compared with those of the panel. Acceptable results were obtained in all cases as more than 80% of the lampante virgin olive oil samples were correctly classified.
2.2. Apparatus All experiments were carried out by using a ChemSensor 4440 system (Gerstel, M¨ulhein an der Ruhr, Germany) consisting of two modules: a MPS2 headspace autosampler (Gerstel, M¨ulhein an der Ruhr, Germany) and a Hewlett-Packard HP5973 mass spectrometry detector (Palo Alto, CA, USA). The former was a 32-space autosampler for headspace vials including a robotic arm and an oven for sample heating/headspace generation. An automated injector fitted with a 2.5 mL gastight HS-syringe was used for the introduction of 2.5 mL of the homogenized headspace of the vial into the injector (200 ◦ C) of a chromatographic oven. Helium (5.0 grade purity, Air Liquide, Seville, Spain), regulated by a digital pressure and flow controller, was used as carrier gas (1.4 mL min−1 ) for driving the analytes to the detector through the transfer line (5 m × 0.25 mm i.d. fused silica capillary) maintained at 170 ◦ C. The quadrupole mass spectrometer detector was operated in full scan mode, with a scanned mass range from m/z 41 to 180; electron impact ionization (70 eV) was used for analyte fragmentation. The MS source and quadrupole temperatures were maintained at 230 and 150 ◦ C, respectively. Total ion current chromatograms were acquired and processed using G1701BA standalone data analysis software (Agilent Technologies) on a Pentium 4 computer that also controlled the whole system. Ten-milliliter clear glass round bottom headspace vials with 20-mm polytetrafluoroethylene/silicone septum caps and a magnetic crimp cap (Supelco, Madrid, Spain) were also employed. 2.3. Procedure
2.1. Samples
The instrumental configuration used is schematically depicted in Fig. 1. Aliquots of 5.0 g of each oil sample were introduced in 10 mL headspace vials and sealed hermetically. The robotic arm took each vial from the tray and placed it into the oven; samples were then heated at 70 ◦ C for 30 min, under controlled stirring to enrich and equilibrate the gaseous phase in the volatile compounds of the sample. After that, a 2.5 mL quantity of headspace material was introduced into the injector, maintained at 200 ◦ C, from where a helium stream carried the volatiles directly to the mass detector through the transfer
Three hundred fifty seven samples (including replicates) from 22 different virgin olive oils were analyzed. These samples were divided as follows: 141 samples of extra-virgin olive oil (EVOO) of different Spanish varieties (Arbequina, Manzanilla, Picual, Picudo, Pico lim´on, Vidue˜na, Cornicabra and Hojiblanca) and 216 samples of lampante-virgin olive oil (LVOO) with several sensory defects (rancid, winey-vinegary, muddy sediment, hay-wood, vegetable water, earthy, fusty and mustyhumidity). From these samples, 274 were used for building the classification model and 83 were the prediction set. Finally, 10 different new EVOO and 50 LVOO samples were used as validation samples and analyzed using the proposed classification model. All samples were obtained from market (EVOO) and olive oil industry (LVOO) and were analyzed by the expert’s panel.
Fig. 1. Schematic diagram of the HS–MS coupling used for the direct analysis of the volatile fraction of the olive oil samples. MS, mass spectrometer.
2. Experimental
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
413
Fig. 2. (A) Global volatile profile of an EVOO analyzed by HS–MS. (B) Mass spectrum associated to the global volatile profile obtained in the chromatogram after the analysis.
line, heated at 170 ◦ C. Despite the presence in the configuration of a gas chromatograph, there is no chromatographic separation under the selected instrumental conditions. Therefore, all the volatiles reach the mass spectrometer at the same time providing a broad signal of ca. 3 min duration, as it is shown in Fig. 2. The mass spectrum associated with this global signal of the volatiles profile is used for classification purposes.
according to the European Union Commission. Four physicochemical parameters were used to evaluate the quality of virgin olive oil samples according to method recommended by the regulation (CE) no. 756/2002, namely: acidity (expressed as percentage of oleic acid), UV extinction coefficients at λ = 232 and 270 nm (called K232 and K270 , respectively) and peroxides value (expressed as mequiv. O2 kg−1 ).
2.4. Data analysis
3. Results and discussion
All chemometric analyses were performed by means of the statistical software package “Pirouette 3.11”, developed by Infometrix Inc. (Bothell, WA, USA). The data matrix was established for 140 columns, corresponding to m/z 41–180 scanned by detector, and 274 rows, corresponding to the different oil samples (108 extra-virgin olive oil samples and 166 lampantevirgin olive oil samples). Two unsupervised techniques (Cluster Analysis, CA and Principal Components Analysis, PCA) were employed in order to provide a quality check of the data and identify variables with high information content. Later, one supervised technique (soft independent modeling of class analogy, SIMCA) was used to obtain an adequate classification procedure [20]. 2.5. Sensorial and physicochemical analyses Sensorial evaluation of the virgin olive oils was performed by analytical panel test using COI/T20/Doc15Rev.1 method
3.1. Conventional methods: sensorial and physicochemical analyses Table 1 shows the extra-virgin olive oils and lampante-virgin olive oils of different varieties employed to build classification model as well as their majority positive and negative attributes, respectively, which have been carried by experts’ panel. The EVOO samples included in this study represent a part of the wide flavour’s variability that can be found in different virgin olive oils, as can be seen in the fourth column in Table 1. A score equal to or higher 5 was given for the negative attributes when there were only one in the sample assessed and it was equal to or higher than 2.5 in multidefective samples. On the other hand, for positive attributes, the average score given by the panelists were 3.5 or slightly lower. The mean and standard deviation of each index analyzed for each physicochemical parameter are indicated in Table 2. The olive oil variety Pico Lim´on ana-
414
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
Table 1 Sensorial evaluation of the virgin olive oil samples by the expert’s panel Type (variety)
COI sensory attributesa
Other sensory attributesb
1
EVOO (Arbequina)
Fruity, bitter, pungent
2
EVOO (Manzanilla)
Fruity, bitter, pungent
3
EVOO (Picual)
Fruity, bitter, pungent
4 5 6
EVOO (Picudo) EVOO (Arbequina) EVOO (Vidue˜na)
Fruity, bitter, pungent Fruity, bitter, pungent Fruity, bitter, pungent
7
EVOO (Hojiblanca)
Fruity, bitter, pungent
EVOO (Cornicabra) EVOO (Pico lim´on) LVOO (Empeltre) LVOO (Arbequina catalana) LVOO (Picual) LVOO (Picual) LVOOc LVOO (Picual) LVOO (Picual) LVOO (Picual) LVOO (Picual) LVOOc LVOOc LVOOc LVOOc
Fruity, bitter, pungent Fruity, bitter, pungent Fruity, bitter, pungent, earthy Fruity, winey-vinegary, rancid Vegetable water Winey-vinegary, musty Muddy sediment, winey-vinegary Muddy sediment, musty, winey-vinegary Hay-wood, musty, winey-vinegary, fruity, bitter, pungent Vegetable water, fusty Winey-vinegary, earthy, fruity, bitter, pungent Earthy, fruity, bitter, pungent Rancid Fusty Musty
Macedoine, apple, green, almond, sweet, tomato, banana, chlorophyll Macedoine, apple, green, almond, sweet, tomato, banana, chlorophyll, astringent Apple, green, almond, sweet, tomato, chlorophyll, astringent, fig tree, nettle Macedoine, green, almond, sweet, astringent, banana Macedoine, apple, green, sweet Apple, green, almond, sweet, tomato, astringent, fig tree, nettle, banana Macedoine, apple, green, almond, sweet, tomato, banana, fig tree – Green, almond, tomato, astringent, banana – – – – – – – – – – – – –
Sample
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 a b c
Established by the expert’s panel; in bold, the highest score attribute. Other attributes assessed by the expert’s panel in addition to the officially established ones. Data not facilitated by the supplier.
Table 2 Mean and standard deviation of the analytical indices in each category Sample
Acidity
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
0.53 0.47 0.14 0.19 0.31 0.21 0.32 0.33 0.21 0.25 0.36 1.13 1.59 1.43 2.14 0.53 1.47 1.12 0.45 1.29 1.91 4.40
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
K232 0.03 0.01 0.02 0.01 0.01 0.01 0.01 0.02 0.01 0.01 0.01 0.02 0.05 0.05 0.04 0.03 0.05 0.01 0.01 0.01 0.01 0.04
1.55 1.73 1.58 1.79 2.11 2.03 1.69 1.95 2.06 1.91 1.91 1.86 1.81 3.05 1.76 1.63 1.61 1.62 1.49 9.95 1.95 2.20
K270 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.02 0.02 0.03 0.20 0.10 0.10 0.06 0.02 0.07 0.02 0.06 0.06 0.07 0.19 0.08 0.03 0.05 0.01 0.01 0.60 0.02 0.09
0.137 0.174 0.121 0.136 0.185 0.176 0.128 0.115 0.260 0.144 0.198 0.182 0.215 0.405 0.156 0.127 0.144 0.131 0.120 2.815 0.317 0.304
Peroxide value ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.005 0.001 0.005 0.003 0.001 0.013 0.001 0.014 0.007 0.011 0.009 0.009 0.007 0.003 0.013 0.003 0.003 0.004 0.001 0.020 0.007 0.015
7 11 8 8 12 14 11 6 12 7 6 12 13 41 7 14 7 8 7 147 16 9
±1 ±2 ±±1 ±1 ±2 ±2 ±2 ±1 ±2 ±1 ±1 ±2 ±2 ±3 ±1 ±2 ±1 ±1 ±1 ±5 ±2 ±1
The limits imposed by law for each parameter for the EVOO are: acidity ≤ 0.8; K232 ≤ 2.50; K270 ≤ 0.20 (after cleaning up through alumina, K270 ≤ 0.10) and peroxide value ≤ 20.
lyzed (sample 9) shows a high absorbance value which results in UV extinction coefficients over the highest limit established for EVOO by the current legislation. However, after purification with alumina, Pico Lim´on samples fulfilled the limits established by the norm and therefore, this variety was considered as EVOO. 3.2. Selection of instrumental variables Chemical and instrumental variables involved in the generation of the headspace of the sample were studied in order to achieve the best separation among the different classes [5]. The value for each individual variable was selected by means of the maximum interclass distance parameter between extravirgin and lampante-virgin olive oil classes. In all cases, the complete volatile profile (m/z 41–180) was used as analytical signal. The sample amount was the first variable studied taking into account its marked influence in the headspace volume, thus affecting the sensitivity of the measurement. This variable was studied within the interval 3.0–7.0 g being 5.0 g selected to study the instrumental parameters of the headspace generation, namely: oven temperature and equilibration time. These variables were optimized in the range 50–90 ◦ C and 20–60 min, respectively, being optimum 70 ◦ C and 30 min.
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
The repeatability of the system was investigated by analysing 11 independent aliquots of the same EVOO of Cornicabra variety. The relative standard deviations of m/z 41, 42, 45, 46 and 48 were used to express this property due to the higher abundance of these fragments in the samples (m/z 44 was not selected as its main contribution comes from atmospheric CO2 ). The average RSD value was found to be ca. 4%.
415
3.3.1. Unsupervised techniques The primary purpose of CA is to present data in a manner which emphasizes natural groupings. The presentation of the results in the form of a dendrogram facilitates the visual recognition of such categories. In this paper, a hierarchical agglomerative procedure, with centroid linkage was used to locate the virgin olive oil samples on basis of Euclidean distance. Others linkage methods (as single, complete, group average among others) were employed with similar results. CA was applied to the autoscaled data as this pretreatment can magnify the importance of small variable (m/z fragment) allowing better discrimination among negative attributes. This technique also allows the detection of potential outliers which, after confirmation by PCA, were not considered for model building. Fig. 3 shows the results obtained for the samples with the eight neg-
ative attributes assayed. As can be seen in the figure, there is a clear separation among the clusters established for each attribute. It should also be mentioned that the dataset used in this study consisted of LVOOs with a single negative attribute or, when it was not feasible, samples with two or three defects, being one of them of higher intensity, which is consistent with the high similarity obtained in some cases between classes. This is corroborated by the results of the expert’s panel assessment as in these cases the scores were ca. 2.5 for all the negative attributes detected in the samples. For example, class 1 (haywood, panel’s score 2.5) shows also negative attribute 2 (wineyvinegary) with similar intensity (2). Finally, it has also to be considered that, for class eight, “earthy” was the sole attribute detected which is more related to the taste than to the aroma of the oil. PCA provides a way to reduce the dimensionality of the data, retaining the maximum variability present in the original data and eliminating possible dependence between variables (in our case, scanned m/z ratios) and therefore overcoming noise and similar information. For this reason, it becomes a powerful visualization tool. PCA was performed on the pre-processed data by means of mean-center and normalization [20]; the different data pre-processing is attributed to the different composition of the data matrix used for CA and PCA, as in this case both negative and positive attributes have been considered. The scores plot of the principal components corresponding to the virgin olive oil samples analyzed is presented in Fig. 4. The EVOO samples appear in a narrow region of the space and thus, the limits of the class can be clearly defined. On the other hand, the wider distribution of the LVOO can be attributed to the large number of defects studied and the different compounds responsible for them which provide a rather different volatile profile.
Fig. 3. Dendrogram of CA for LVOO with different negative attributes. S, similarity index: 1, hay-wood; 2, winey-vinegary; 3, rancid; 4, fusty; 5, muddy sediment; 6, musty-humidity; 7, vegetable water; 8, earthy.
Fig. 4. A 3D scores plot of PCA for EVOO and LVOO. The percentage of explained variance by each PC was 99.2, 0.5 and 0.2%, respectively. For details, see text.
3.3. Chemometrics The aim of this work was the discrimination between virgin olive oils with positive and negative organoleptic attributes, by using different chemometric tools, namely CA, PCA and SIMCA. The unsupervised techniques permit determination of internal structure or clustering of the data and the supervised technique provides an adequate classification procedure.
416
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
Table 3 Classification and prediction matrix for SIMCA and specificity and sensitivity for each class of virgin olive oil Classesa
SIMCA 1
Classification 1 (108) 108 2 (166) 0
2
No match
0 166
0 0
Specificity (%)
Sensitivity (%)
100 100
100 100
Prediction set 1 1 (33) 11 2 (50) 0
0 45
22 5
100 100
33 90
Prediction set 2 1 (10) 3 2 (50) 0
0 39
7 6
100 100
30 78
a 1 corresponds to EVOO samples and 2, LVOO samples. In parenthesis, the number of virgin olive oil samples analyzed.
After the data analysis by unsupervised techniques it can be concluded that the HS–MS configuration can distinguish between extra-virgin and defective olive oil samples, even among the selected negative attributes included within the LVOO group. 3.3.2. Soft independent modeling of class analogy (SIMCA) This is a class-modeling technique which constructs an independent PCA model for each class. The classification rule for a given class is that an object is assigned to this class if it is situated inside the boundaries of the corresponding class-box in a pattern space and an outlier if it falls outside. The same LVOO and EVOO samples used for PCA were employed to construct the SIMCA model. SIMCA classification model was built by using mean-centered, normalized data pre-treatments and 10 and 22 principal components for each class, respectively which provides a percentage of explained variance of 99.99%. A cross-validation step was included and several validation methods were assayed with similar results. The option adopted was finally leave-out one sample a time. This model was applied to an external data set composed of 83 samples not employed for the model building. It was checked that a lower number of principal components made the prediction worse, because it classified LVOO as EVOO, and vice versa. The results obtained in terms of classification and prediction abilities are summarised in Table 3. As can be seen, there was no misclassification within the prediction set; however, the percentage of unmatched samples was very high as regards the extra-virgin olive oil samples. It can be ascribed to the fact that it is rather difficult to include all the flavour variability in the model as it depends on a large number of factors (geographical origin, olive variety among others). These samples should be analyzed following the official method, that is, the expert’s panel. However, the reliability of the proposed classification model is very high as none of the defective samples were considered under the EVOO category, and vice versa. Finally, a total of 10 EVOO and 50 LVOO new set of samples provided by the same supplier were analyzed with the Chemsensor system according to the optimized procedure in a range of
140 variables (m/z 41–180). Results of prediction slightly lower than that those achieved for the validation model were obtained for the LVOO (ca. 80%). However, the prediction ability of the EVOO remained constant. No samples misclassification was obtained in any case. 4. Conclusions The research work presented in this paper has been the rapid identification of defective olive oil samples to prevent their distribution and commercialization. Moreover, the developed methodology can be used to detect fraudulent practices in the industry, mainly the adulteration of extra-virgin olive oils with low quality edible oils in order to increase their economical benefits. The use of an instrumental method for sample classification is more reliable than the “experts” panel if the resulting model is correctly trained. In any case, it can reduce the number of samples to be analyzed by the official methods, being used as a vanguard analytical strategy [19]. Future research efforts will be aimed at the classification of virgin olive oil samples with positive attribute on the basis of their protected designation of origin, geographical origin and olive variety, among others. Acknowledgements This work was supported by grant CTQ2004-01220 of the DGI of the Spanish Ministry of Science and Technology. SLF wants to thank the IFAPA for the finantial support through a CICyE (Junta de Andaluc´ıa) fellowship. The authors wish to acknowledge to “Oleo-Cata Xauen” and especially to Antonia Fern´andez for the supply of the virgin olive oil samples and the sensory evaluation of those. References [1] Manual del Aceite de Oliva, A. Madrid Vicente, 2003. [2] R. Aparicio, M.T. Morales, M.V. Alonso, J. Am. Oil Chem. Soc. 73 (1996) 1253. [3] COI/T.20/Doc. no. 15/Rev. 1, Method Organoleptic Assessment of Virgin Olive Oil, International Olive Oil Council, 1996. [4] C. P´er`es, F. Bernaud, J. Berdagu´e, Trends Anal. Chem. 22 (2003) 858. [5] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, J. Am. Oil Chem. Soc. 80 (2003) 613. [6] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, Anal. Chim. Acta 526 (2004) 77. [7] I.M. Lorenzo, J.L. P´erez Pav´on, M.E. Fern´andez Laespada, C. Garc´ıa Pinto, B. Moreno Cordero, J. Chromatogr. A 942 (2002) 221. [8] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, J. Chromatogr. A 1074 (2005) 215. [9] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, J. Am. Oil Chem. Soc. 79 (2002) 1103. [10] I.M. Lorenzo, J.L. P´erez Pav´on, M.E. Fern´andez Laespada, C. Garc´ıa Pinto, B. Moreno Cordero, L.R. Henriques, M.F. P´er`es, M.P. Sim˜oes, P.S. Lopes, Anal. Bioanal. Chem. 374 (2002) 1205. [11] C. Cerrato Oliveros, R. Boggia, M. Casale, C. Armanino, M. Forina, J. Chromatogr. A 1076 (2005) 7. [12] R. Aparicio, R. Aparicio Ru´ız, J. Chromatogr. A 881 (2000) 93. [13] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, J. Chromatogr. A 1052 (2004) 137. [14] G. Luna, M.T. Morales, R. Aparicio, Food Chem. 98 (2006) 243.
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417 [15] M.T. Morales, G. Luna, R. Aparicio, Food Chem. 91 (2005) 293. [16] D.L. Garc´ıa Gonz´alez, R. Aparicio, Eur. Food Res. Technol. 215 (2002) 118. [17] G. Procida, A. Giomo, A. Cichelli, L.S. Conte, J. Sci. Food Agric. 85 (2005) 2175.
417
[18] N.T. Skizas, M. Tasioula-Margari, M.M. Komaitis, Food Flavors Chem. 274 (2001) 248. [19] M. Valc´arcel, S. C´ardenas, Trends Anal. Chem. 24 (2005) 67. [20] Multivariate Data Analysis, Ver 3.11, Infometrix Inc., Woodinville, WA, 2003.
Usefulness of the direct coupling headspace–mass spectrometry for sensory quality characterization of virgin olive oil samples Silvia L´opez-Feria b , Soledad C´ardenas a , Jos´e Antonio Garc´ıa-Mesa b , Miguel Valc´arcel a,∗ a
Department of Analytical Chemistry, Marie Curie Building (Annex), Campus de Rabanales, University of C´ordoba, E-14071 C´ordoba, Spain b CIFA Venta del Llano, IFAPA, Ctra. Bail´ en-Motril km 18.5 Mengibar, E-23620 Ja´en, Spain Received 14 July 2006; received in revised form 28 September 2006; accepted 9 October 2006 Available online 26 October 2006
Abstract The applicability of the headspace coupled to mass spectrometry for evaluation of the sensory quality of virgin olive oil samples is presented. The volatiles of the oil are directly transferred from the sample vial to the detector without chromatographic separation. The mass spectrum obtained can be considered as a fingerprint of the oil sample and can be used for classification purposes. After a training step with samples previously qualified following the official method, a classification model was created using the supervised technique soft independent modeling of class analogy (SIMCA). Eight negative (rancid, winey-vinegary, muddy sediment, hay-wood, vegetable water, earthy, fusty and musty-humidity) and three principal positive attributes (fruity, bitter and pungent) have been included in this study. With them, a classification model consisting of two main classes (extra- and lampante-virgin olive oil) was constructed. In addition, the unsupervised technique cluster analysis permited the discrimination among oils with different negative attributes. The proposed procedure has been applied to the classification of commercial samples (as extra- or lampante-virgin olive oils) and the results were compared with those provided by the expert’s panel with acceptable correlation. © 2006 Elsevier B.V. All rights reserved. Keywords: Headspace; Mass spectrometry; Olive oil sensory quality; Direct analysis; Chemometric techniques
1. Introduction Virgin olive oil, extracted by purely mechanical procedures from the fruit of the olive tree, Olea europaea L., is one of the basic components of the Mediterranean diet. It can be classified as edible (extra-virgin and virgin) and non-edible (lampante) virgin olive oils [1]. Its largely demonstrated benefits for human health constitute an added value of this food, which clearly affects the price it can reach in the market. Virgin olive oil can be marketed under a variety of denominations, which can refer to the variety of olive used for its production or its sensorial attributes. Flavors and aroma of the oils are generated by a number of volatile constituents which are present at very low concentrations, which in most cases affects the quality of the virgin olive oil and therefore, its price [2]. The sensory evaluation of these samples is usually carried out by an experts’ panel which allows the characterization of the olive oil within a sin-
∗
Corresponding author. Tel.: +34 957 218 616; fax: +34 957 218 616. E-mail address: [email protected] (M. Valc´arcel).
0003-2670/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2006.10.027
gle analysis. The attributes detected can be either positives or negatives, being the latter consequence of a bad collection process, olive/oil storage or mixture with other oils of lower quality [3]. However, this method is lengthy and expensive, whose final results depends on the panelists’ training and the specific vocabulary used in sensory analysis. Several alternatives oriented at overcoming these disadvantages have been recently published. Most of them involve the direct analysis of the samples by means of dynamic headspace (HS) and mass spectrometry (MS), with or without chromatographic separation. The applicability of the direct coupling of a HS module to a mass spectrometer in food analysis has been recently reviewed [4]. The main advantage of this configuration is that the sample can be directly analyzed without any pretreatment. The mass spectrum obtained provides a characteristic fingerprint of the volatile compounds of the sample. The subsequent chemometric treatment of the spectral information permits sample characterization once the instrument has been properly trained. In the field of olive oil analysis, the HS–MS coupling has been proposed for the determination of contaminants [5,6], adulteration with other edible oils [7,8] and classification purposes.
412
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
Concerning this last topic, HS–MS has been applied to the classification of the three main types of olive oils [9], the characterization of monovarietal olive oils [10] and the discrimination of different geographical origin olive oils [11]. The inclusion of a gas chromatograph (GC) within this configuration permits the measurement of retention times and possible identification of the compounds present in the sample which are responsible for the property being modeled [12–14]. The evaluation of olive oil sensory defects (winey-vinegary, musty-humidity, fusty and rancid) has been afforded by using both HS–GC-FID and MS and HS–GC-olfactometry [15] and metal-oxide sensors [16]. Moreover, a HS–GC–MS method has been proposed to assign the chromatographic peaks obtained to the defects detected by a panel test. In this case, only four defects (muddy sediment, fusty, winey-vinegary and musty-humidity) were included [17]. Finally, an HS–GC method allowed the identification of the volatile compounds responsible for the pleasant sensory perception as well as the undesirable flavor from Lianolia variety [18]. In this paper, the direct use of the HS–MS for the rapid determination of the quality of virgin olive oil samples, based on eight defects (rancid, winey-vinegary, muddy sediment, hay-wood, vegetable water, earthy, fusty and mustyhumidity) and three principal positive attributes (fruity, bitter and pungent), is proposed as an alternative to the experts’ panel. The classification model, which was used to discriminate among extra- and lampante-virgin olive oil, was constructed by using 13 defective and 9 high quality monovarietal virgin olive oil samples, previously qualified by the corresponding experts’ panel. Based on the previous experience of our research group [5], it was concluded that SIMCA is more adequate for the objective of this work and therefore, it was applied to the training set including a cross-validation step. The optimized model was used for the classification of a new set of virgin olive oil samples and the results compared with those of the panel. Acceptable results were obtained in all cases as more than 80% of the lampante virgin olive oil samples were correctly classified.
2.2. Apparatus All experiments were carried out by using a ChemSensor 4440 system (Gerstel, M¨ulhein an der Ruhr, Germany) consisting of two modules: a MPS2 headspace autosampler (Gerstel, M¨ulhein an der Ruhr, Germany) and a Hewlett-Packard HP5973 mass spectrometry detector (Palo Alto, CA, USA). The former was a 32-space autosampler for headspace vials including a robotic arm and an oven for sample heating/headspace generation. An automated injector fitted with a 2.5 mL gastight HS-syringe was used for the introduction of 2.5 mL of the homogenized headspace of the vial into the injector (200 ◦ C) of a chromatographic oven. Helium (5.0 grade purity, Air Liquide, Seville, Spain), regulated by a digital pressure and flow controller, was used as carrier gas (1.4 mL min−1 ) for driving the analytes to the detector through the transfer line (5 m × 0.25 mm i.d. fused silica capillary) maintained at 170 ◦ C. The quadrupole mass spectrometer detector was operated in full scan mode, with a scanned mass range from m/z 41 to 180; electron impact ionization (70 eV) was used for analyte fragmentation. The MS source and quadrupole temperatures were maintained at 230 and 150 ◦ C, respectively. Total ion current chromatograms were acquired and processed using G1701BA standalone data analysis software (Agilent Technologies) on a Pentium 4 computer that also controlled the whole system. Ten-milliliter clear glass round bottom headspace vials with 20-mm polytetrafluoroethylene/silicone septum caps and a magnetic crimp cap (Supelco, Madrid, Spain) were also employed. 2.3. Procedure
2.1. Samples
The instrumental configuration used is schematically depicted in Fig. 1. Aliquots of 5.0 g of each oil sample were introduced in 10 mL headspace vials and sealed hermetically. The robotic arm took each vial from the tray and placed it into the oven; samples were then heated at 70 ◦ C for 30 min, under controlled stirring to enrich and equilibrate the gaseous phase in the volatile compounds of the sample. After that, a 2.5 mL quantity of headspace material was introduced into the injector, maintained at 200 ◦ C, from where a helium stream carried the volatiles directly to the mass detector through the transfer
Three hundred fifty seven samples (including replicates) from 22 different virgin olive oils were analyzed. These samples were divided as follows: 141 samples of extra-virgin olive oil (EVOO) of different Spanish varieties (Arbequina, Manzanilla, Picual, Picudo, Pico lim´on, Vidue˜na, Cornicabra and Hojiblanca) and 216 samples of lampante-virgin olive oil (LVOO) with several sensory defects (rancid, winey-vinegary, muddy sediment, hay-wood, vegetable water, earthy, fusty and mustyhumidity). From these samples, 274 were used for building the classification model and 83 were the prediction set. Finally, 10 different new EVOO and 50 LVOO samples were used as validation samples and analyzed using the proposed classification model. All samples were obtained from market (EVOO) and olive oil industry (LVOO) and were analyzed by the expert’s panel.
Fig. 1. Schematic diagram of the HS–MS coupling used for the direct analysis of the volatile fraction of the olive oil samples. MS, mass spectrometer.
2. Experimental
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
413
Fig. 2. (A) Global volatile profile of an EVOO analyzed by HS–MS. (B) Mass spectrum associated to the global volatile profile obtained in the chromatogram after the analysis.
line, heated at 170 ◦ C. Despite the presence in the configuration of a gas chromatograph, there is no chromatographic separation under the selected instrumental conditions. Therefore, all the volatiles reach the mass spectrometer at the same time providing a broad signal of ca. 3 min duration, as it is shown in Fig. 2. The mass spectrum associated with this global signal of the volatiles profile is used for classification purposes.
according to the European Union Commission. Four physicochemical parameters were used to evaluate the quality of virgin olive oil samples according to method recommended by the regulation (CE) no. 756/2002, namely: acidity (expressed as percentage of oleic acid), UV extinction coefficients at λ = 232 and 270 nm (called K232 and K270 , respectively) and peroxides value (expressed as mequiv. O2 kg−1 ).
2.4. Data analysis
3. Results and discussion
All chemometric analyses were performed by means of the statistical software package “Pirouette 3.11”, developed by Infometrix Inc. (Bothell, WA, USA). The data matrix was established for 140 columns, corresponding to m/z 41–180 scanned by detector, and 274 rows, corresponding to the different oil samples (108 extra-virgin olive oil samples and 166 lampantevirgin olive oil samples). Two unsupervised techniques (Cluster Analysis, CA and Principal Components Analysis, PCA) were employed in order to provide a quality check of the data and identify variables with high information content. Later, one supervised technique (soft independent modeling of class analogy, SIMCA) was used to obtain an adequate classification procedure [20]. 2.5. Sensorial and physicochemical analyses Sensorial evaluation of the virgin olive oils was performed by analytical panel test using COI/T20/Doc15Rev.1 method
3.1. Conventional methods: sensorial and physicochemical analyses Table 1 shows the extra-virgin olive oils and lampante-virgin olive oils of different varieties employed to build classification model as well as their majority positive and negative attributes, respectively, which have been carried by experts’ panel. The EVOO samples included in this study represent a part of the wide flavour’s variability that can be found in different virgin olive oils, as can be seen in the fourth column in Table 1. A score equal to or higher 5 was given for the negative attributes when there were only one in the sample assessed and it was equal to or higher than 2.5 in multidefective samples. On the other hand, for positive attributes, the average score given by the panelists were 3.5 or slightly lower. The mean and standard deviation of each index analyzed for each physicochemical parameter are indicated in Table 2. The olive oil variety Pico Lim´on ana-
414
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
Table 1 Sensorial evaluation of the virgin olive oil samples by the expert’s panel Type (variety)
COI sensory attributesa
Other sensory attributesb
1
EVOO (Arbequina)
Fruity, bitter, pungent
2
EVOO (Manzanilla)
Fruity, bitter, pungent
3
EVOO (Picual)
Fruity, bitter, pungent
4 5 6
EVOO (Picudo) EVOO (Arbequina) EVOO (Vidue˜na)
Fruity, bitter, pungent Fruity, bitter, pungent Fruity, bitter, pungent
7
EVOO (Hojiblanca)
Fruity, bitter, pungent
EVOO (Cornicabra) EVOO (Pico lim´on) LVOO (Empeltre) LVOO (Arbequina catalana) LVOO (Picual) LVOO (Picual) LVOOc LVOO (Picual) LVOO (Picual) LVOO (Picual) LVOO (Picual) LVOOc LVOOc LVOOc LVOOc
Fruity, bitter, pungent Fruity, bitter, pungent Fruity, bitter, pungent, earthy Fruity, winey-vinegary, rancid Vegetable water Winey-vinegary, musty Muddy sediment, winey-vinegary Muddy sediment, musty, winey-vinegary Hay-wood, musty, winey-vinegary, fruity, bitter, pungent Vegetable water, fusty Winey-vinegary, earthy, fruity, bitter, pungent Earthy, fruity, bitter, pungent Rancid Fusty Musty
Macedoine, apple, green, almond, sweet, tomato, banana, chlorophyll Macedoine, apple, green, almond, sweet, tomato, banana, chlorophyll, astringent Apple, green, almond, sweet, tomato, chlorophyll, astringent, fig tree, nettle Macedoine, green, almond, sweet, astringent, banana Macedoine, apple, green, sweet Apple, green, almond, sweet, tomato, astringent, fig tree, nettle, banana Macedoine, apple, green, almond, sweet, tomato, banana, fig tree – Green, almond, tomato, astringent, banana – – – – – – – – – – – – –
Sample
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 a b c
Established by the expert’s panel; in bold, the highest score attribute. Other attributes assessed by the expert’s panel in addition to the officially established ones. Data not facilitated by the supplier.
Table 2 Mean and standard deviation of the analytical indices in each category Sample
Acidity
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
0.53 0.47 0.14 0.19 0.31 0.21 0.32 0.33 0.21 0.25 0.36 1.13 1.59 1.43 2.14 0.53 1.47 1.12 0.45 1.29 1.91 4.40
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
K232 0.03 0.01 0.02 0.01 0.01 0.01 0.01 0.02 0.01 0.01 0.01 0.02 0.05 0.05 0.04 0.03 0.05 0.01 0.01 0.01 0.01 0.04
1.55 1.73 1.58 1.79 2.11 2.03 1.69 1.95 2.06 1.91 1.91 1.86 1.81 3.05 1.76 1.63 1.61 1.62 1.49 9.95 1.95 2.20
K270 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.02 0.02 0.03 0.20 0.10 0.10 0.06 0.02 0.07 0.02 0.06 0.06 0.07 0.19 0.08 0.03 0.05 0.01 0.01 0.60 0.02 0.09
0.137 0.174 0.121 0.136 0.185 0.176 0.128 0.115 0.260 0.144 0.198 0.182 0.215 0.405 0.156 0.127 0.144 0.131 0.120 2.815 0.317 0.304
Peroxide value ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.005 0.001 0.005 0.003 0.001 0.013 0.001 0.014 0.007 0.011 0.009 0.009 0.007 0.003 0.013 0.003 0.003 0.004 0.001 0.020 0.007 0.015
7 11 8 8 12 14 11 6 12 7 6 12 13 41 7 14 7 8 7 147 16 9
±1 ±2 ±±1 ±1 ±2 ±2 ±2 ±1 ±2 ±1 ±1 ±2 ±2 ±3 ±1 ±2 ±1 ±1 ±1 ±5 ±2 ±1
The limits imposed by law for each parameter for the EVOO are: acidity ≤ 0.8; K232 ≤ 2.50; K270 ≤ 0.20 (after cleaning up through alumina, K270 ≤ 0.10) and peroxide value ≤ 20.
lyzed (sample 9) shows a high absorbance value which results in UV extinction coefficients over the highest limit established for EVOO by the current legislation. However, after purification with alumina, Pico Lim´on samples fulfilled the limits established by the norm and therefore, this variety was considered as EVOO. 3.2. Selection of instrumental variables Chemical and instrumental variables involved in the generation of the headspace of the sample were studied in order to achieve the best separation among the different classes [5]. The value for each individual variable was selected by means of the maximum interclass distance parameter between extravirgin and lampante-virgin olive oil classes. In all cases, the complete volatile profile (m/z 41–180) was used as analytical signal. The sample amount was the first variable studied taking into account its marked influence in the headspace volume, thus affecting the sensitivity of the measurement. This variable was studied within the interval 3.0–7.0 g being 5.0 g selected to study the instrumental parameters of the headspace generation, namely: oven temperature and equilibration time. These variables were optimized in the range 50–90 ◦ C and 20–60 min, respectively, being optimum 70 ◦ C and 30 min.
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
The repeatability of the system was investigated by analysing 11 independent aliquots of the same EVOO of Cornicabra variety. The relative standard deviations of m/z 41, 42, 45, 46 and 48 were used to express this property due to the higher abundance of these fragments in the samples (m/z 44 was not selected as its main contribution comes from atmospheric CO2 ). The average RSD value was found to be ca. 4%.
415
3.3.1. Unsupervised techniques The primary purpose of CA is to present data in a manner which emphasizes natural groupings. The presentation of the results in the form of a dendrogram facilitates the visual recognition of such categories. In this paper, a hierarchical agglomerative procedure, with centroid linkage was used to locate the virgin olive oil samples on basis of Euclidean distance. Others linkage methods (as single, complete, group average among others) were employed with similar results. CA was applied to the autoscaled data as this pretreatment can magnify the importance of small variable (m/z fragment) allowing better discrimination among negative attributes. This technique also allows the detection of potential outliers which, after confirmation by PCA, were not considered for model building. Fig. 3 shows the results obtained for the samples with the eight neg-
ative attributes assayed. As can be seen in the figure, there is a clear separation among the clusters established for each attribute. It should also be mentioned that the dataset used in this study consisted of LVOOs with a single negative attribute or, when it was not feasible, samples with two or three defects, being one of them of higher intensity, which is consistent with the high similarity obtained in some cases between classes. This is corroborated by the results of the expert’s panel assessment as in these cases the scores were ca. 2.5 for all the negative attributes detected in the samples. For example, class 1 (haywood, panel’s score 2.5) shows also negative attribute 2 (wineyvinegary) with similar intensity (2). Finally, it has also to be considered that, for class eight, “earthy” was the sole attribute detected which is more related to the taste than to the aroma of the oil. PCA provides a way to reduce the dimensionality of the data, retaining the maximum variability present in the original data and eliminating possible dependence between variables (in our case, scanned m/z ratios) and therefore overcoming noise and similar information. For this reason, it becomes a powerful visualization tool. PCA was performed on the pre-processed data by means of mean-center and normalization [20]; the different data pre-processing is attributed to the different composition of the data matrix used for CA and PCA, as in this case both negative and positive attributes have been considered. The scores plot of the principal components corresponding to the virgin olive oil samples analyzed is presented in Fig. 4. The EVOO samples appear in a narrow region of the space and thus, the limits of the class can be clearly defined. On the other hand, the wider distribution of the LVOO can be attributed to the large number of defects studied and the different compounds responsible for them which provide a rather different volatile profile.
Fig. 3. Dendrogram of CA for LVOO with different negative attributes. S, similarity index: 1, hay-wood; 2, winey-vinegary; 3, rancid; 4, fusty; 5, muddy sediment; 6, musty-humidity; 7, vegetable water; 8, earthy.
Fig. 4. A 3D scores plot of PCA for EVOO and LVOO. The percentage of explained variance by each PC was 99.2, 0.5 and 0.2%, respectively. For details, see text.
3.3. Chemometrics The aim of this work was the discrimination between virgin olive oils with positive and negative organoleptic attributes, by using different chemometric tools, namely CA, PCA and SIMCA. The unsupervised techniques permit determination of internal structure or clustering of the data and the supervised technique provides an adequate classification procedure.
416
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417
Table 3 Classification and prediction matrix for SIMCA and specificity and sensitivity for each class of virgin olive oil Classesa
SIMCA 1
Classification 1 (108) 108 2 (166) 0
2
No match
0 166
0 0
Specificity (%)
Sensitivity (%)
100 100
100 100
Prediction set 1 1 (33) 11 2 (50) 0
0 45
22 5
100 100
33 90
Prediction set 2 1 (10) 3 2 (50) 0
0 39
7 6
100 100
30 78
a 1 corresponds to EVOO samples and 2, LVOO samples. In parenthesis, the number of virgin olive oil samples analyzed.
After the data analysis by unsupervised techniques it can be concluded that the HS–MS configuration can distinguish between extra-virgin and defective olive oil samples, even among the selected negative attributes included within the LVOO group. 3.3.2. Soft independent modeling of class analogy (SIMCA) This is a class-modeling technique which constructs an independent PCA model for each class. The classification rule for a given class is that an object is assigned to this class if it is situated inside the boundaries of the corresponding class-box in a pattern space and an outlier if it falls outside. The same LVOO and EVOO samples used for PCA were employed to construct the SIMCA model. SIMCA classification model was built by using mean-centered, normalized data pre-treatments and 10 and 22 principal components for each class, respectively which provides a percentage of explained variance of 99.99%. A cross-validation step was included and several validation methods were assayed with similar results. The option adopted was finally leave-out one sample a time. This model was applied to an external data set composed of 83 samples not employed for the model building. It was checked that a lower number of principal components made the prediction worse, because it classified LVOO as EVOO, and vice versa. The results obtained in terms of classification and prediction abilities are summarised in Table 3. As can be seen, there was no misclassification within the prediction set; however, the percentage of unmatched samples was very high as regards the extra-virgin olive oil samples. It can be ascribed to the fact that it is rather difficult to include all the flavour variability in the model as it depends on a large number of factors (geographical origin, olive variety among others). These samples should be analyzed following the official method, that is, the expert’s panel. However, the reliability of the proposed classification model is very high as none of the defective samples were considered under the EVOO category, and vice versa. Finally, a total of 10 EVOO and 50 LVOO new set of samples provided by the same supplier were analyzed with the Chemsensor system according to the optimized procedure in a range of
140 variables (m/z 41–180). Results of prediction slightly lower than that those achieved for the validation model were obtained for the LVOO (ca. 80%). However, the prediction ability of the EVOO remained constant. No samples misclassification was obtained in any case. 4. Conclusions The research work presented in this paper has been the rapid identification of defective olive oil samples to prevent their distribution and commercialization. Moreover, the developed methodology can be used to detect fraudulent practices in the industry, mainly the adulteration of extra-virgin olive oils with low quality edible oils in order to increase their economical benefits. The use of an instrumental method for sample classification is more reliable than the “experts” panel if the resulting model is correctly trained. In any case, it can reduce the number of samples to be analyzed by the official methods, being used as a vanguard analytical strategy [19]. Future research efforts will be aimed at the classification of virgin olive oil samples with positive attribute on the basis of their protected designation of origin, geographical origin and olive variety, among others. Acknowledgements This work was supported by grant CTQ2004-01220 of the DGI of the Spanish Ministry of Science and Technology. SLF wants to thank the IFAPA for the finantial support through a CICyE (Junta de Andaluc´ıa) fellowship. The authors wish to acknowledge to “Oleo-Cata Xauen” and especially to Antonia Fern´andez for the supply of the virgin olive oil samples and the sensory evaluation of those. References [1] Manual del Aceite de Oliva, A. Madrid Vicente, 2003. [2] R. Aparicio, M.T. Morales, M.V. Alonso, J. Am. Oil Chem. Soc. 73 (1996) 1253. [3] COI/T.20/Doc. no. 15/Rev. 1, Method Organoleptic Assessment of Virgin Olive Oil, International Olive Oil Council, 1996. [4] C. P´er`es, F. Bernaud, J. Berdagu´e, Trends Anal. Chem. 22 (2003) 858. [5] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, J. Am. Oil Chem. Soc. 80 (2003) 613. [6] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, Anal. Chim. Acta 526 (2004) 77. [7] I.M. Lorenzo, J.L. P´erez Pav´on, M.E. Fern´andez Laespada, C. Garc´ıa Pinto, B. Moreno Cordero, J. Chromatogr. A 942 (2002) 221. [8] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, J. Chromatogr. A 1074 (2005) 215. [9] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, J. Am. Oil Chem. Soc. 79 (2002) 1103. [10] I.M. Lorenzo, J.L. P´erez Pav´on, M.E. Fern´andez Laespada, C. Garc´ıa Pinto, B. Moreno Cordero, L.R. Henriques, M.F. P´er`es, M.P. Sim˜oes, P.S. Lopes, Anal. Bioanal. Chem. 374 (2002) 1205. [11] C. Cerrato Oliveros, R. Boggia, M. Casale, C. Armanino, M. Forina, J. Chromatogr. A 1076 (2005) 7. [12] R. Aparicio, R. Aparicio Ru´ız, J. Chromatogr. A 881 (2000) 93. [13] F. Pe˜na, S. C´ardenas, M. Gallego, M. Valc´arcel, J. Chromatogr. A 1052 (2004) 137. [14] G. Luna, M.T. Morales, R. Aparicio, Food Chem. 98 (2006) 243.
S. L´opez-Feria et al. / Analytica Chimica Acta 583 (2007) 411–417 [15] M.T. Morales, G. Luna, R. Aparicio, Food Chem. 91 (2005) 293. [16] D.L. Garc´ıa Gonz´alez, R. Aparicio, Eur. Food Res. Technol. 215 (2002) 118. [17] G. Procida, A. Giomo, A. Cichelli, L.S. Conte, J. Sci. Food Agric. 85 (2005) 2175.
417
[18] N.T. Skizas, M. Tasioula-Margari, M.M. Komaitis, Food Flavors Chem. 274 (2001) 248. [19] M. Valc´arcel, S. C´ardenas, Trends Anal. Chem. 24 (2005) 67. [20] Multivariate Data Analysis, Ver 3.11, Infometrix Inc., Woodinville, WA, 2003.