Professional Documents
Culture Documents
1 s2.0 1 Main
1 s2.0 1 Main
Abstract
The freshness of virgin olive oils (VOO) from typical cultivars of Garda regions was evaluated by attenuated total reflectance (ATR) and
Fourier transform infrared (FTIR) spectroscopy, in combination with multivariate analysis. The olive oil freshness decreased during storage mainly
because of oxidation processes. In this research, 91 virgin olive oils were packaged in glass bottles and stored either in the light or in the dark
at room temperature for different periods. The oils were analysed, before and after storage, using both chemical methods and spectroscopic
technique.
Classification strategies investigated were partial least square discriminant analysis (PLS-DA), linear discriminant analysis (LDA), and soft
independent modelling of class analogy (SIMCA).
The results show that ATR-MIR spectroscopy is an interesting technique compared with traditional chemical index in classifying olive oil
samples stored in different conditions. In fact, the FTIR PCA results allowed a better discrimination among fresh and oxidized oils, than sam-
ples separation obtained by PCA applied to chemical data. Moreover, the results obtained by the different classification techniques (PLS-DA,
LDA, SIMCA) evidenced the ability of FTIR spectra to evaluate the olive oil freshness. FTIR spectroscopy results are in agreement with
classical methods. The spectroscopic technique could be applied for the prediction of VOOs freshness giving information related to chemical
modifications. The great advantages of this technique, compared to chemical analysis, are related to rapidity, non-destructive characteristics
and low cost per sample. In conclusion, ATR-MIR represents a reliable, cheap and fast classification tool able to assess the freshness of virgin
olive oils.
© 2007 Elsevier B.V. All rights reserved.
Keywords: Virgin olive oil; Freshness; Oxidation; Storage; Fourier transform infrared (FTIR) spectroscopy; Attenuated total reflection mid infrared (ATR-MIR)
spectroscopy
1. Introduction to preserve the positive attributes of extra virgin oil over time
from production to bottling, up to purchasing.
Extra virgin olive oil shows excellent sensorial and nutritional Nowadays, there are different methods used and/or pro-
properties, due to antioxidant compounds and to an interest- posed for evaluating the oxidative deterioration of olive oil.
ing unsaturated to saturated fatty acids ratio. These properties Among the routine methods, there are the peroxide value (PV),
decrease with time because of oxidation processes affected by which determines the amount of primary oxidation products,
air, heat, light and metals [1]. Oil oxidation, is a very com- UV absorbance at 232 and 270 nm, that measures the forma-
plex phenomena that regards mainly unsaturated fatty acids tion of conjugated dienes and trienes due to the formation of
and modifies oil sensorial characteristics through off-flavours secondary oxidation products and the Rancimat Oxidative Sta-
development and secondary reaction products formation. Con- bility Instrument, determining the oxygen uptake [2]. According
sequently, it is matter of great concern for the olive oil industry to European Commission (EC) [3] and International Olive Oil
Council (IOOC) regulations [4], the extra virgin olive oil oxi-
dation level is assessed by the PV and spectrophotometric
∗ Corresponding author. Tel.: +39 02 5031 9184; fax: +39 02 5031 9191. absorbance, defining the following limits: PV ≤ 20 meq kg−1
E-mail address: ernestina.casiraghi@unimi.it (E. Casiraghi). and K270 ≤ 0.22. The EC legislation also considers the value
0003-2670/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.aca.2007.07.024
N. Sinelli et al. / Analytica Chimica Acta 598 (2007) 128–134 129
compensated with the Atmospheric Compensation command, [18]. The PLS-DA regression used in this study was performed
using Opus software. by using the PLS II model regression of statistical package The
Unscrambler (version 9.6, CAMO, ASA, Norway).
Linear discriminant analysis is a supervised classification
2.4. Data processing technique where the number of categories and the samples
belonging to each category are previously defined. The criterion
Statistical analysis of the chemical parameters was performed of LDA for selection of latent variables maximizes the differ-
by SPSS statistical software (version 13.0 Inc., Chicago, IL, ences between categories and minimizes the variances within
USA) with General Linear Model (GLM) procedure, using categories. The method produces a number of orthogonal linear
Student–Newman–Keuls test. discriminant functions, equal to the number of the categories
To reduce the baseline shift, the spectral data were trans- minus one, that allow the samples to be classified in one or
formed into second derivative (Savitzky–Golay method, number another category [19,20].
of smoothing points = 10). LDA was carried out using SPSS (version 13.0 Inc., Chicago,
Principal component analysis (PCA) was performed as an IL, USA) on the PCA samples scores of the significant compo-
explorative analysis to each data set (chemical data and FTIR nents (PCs), using full cross-validation (leave one out method),
spectra) in order to display the data structure and to reduce the both for chemical and spectral data set.
dimensionality of the spectral data to a small number of com- SIMCA method considers different classes which are mod-
ponents. PCA is a mathematical procedure for resolving sets elled individually by a separate principal component model. The
of data into orthogonal components whose linear combinations number of significant PCs is determinate for each class and as
approximate the original data to any desired degree of accuracy many models as the number of classes are obtained by training.
[17,18]. PCA was performed using The Unscrambler software The test set is used to test all class models [18,21]. SIMCA model
package (version 9.6, CAMO, ASA, Norway). was developed using first four classes and then three classes
To apply classification methods VOO samples were divided (fresh, dark and light) and was performed using the software
into four classes: fresh (class 1), dark 1 year (class 2), dark 2 The Unscrambler (version 9.6, CAMO, ASA, Norway). Class
years (class 3) and light 1 year (class 4). Classification tech- cut-off limits in SIMCA were set at the 5% level.
niques investigated were Partial Least Square Discriminant
Analysis (PLS-DA), linear discriminant analysis (LDA), and
soft independent modelling of class analogy (SIMCA). 3. Results and discussion
PLS-DA is a variant of partial least square (PLS) regression,
in which each sample, in the calibration set, is assigned to a 3.1. Chemical analyses
dummy variable (0,1) as a reference value. The classification of
the olive oils according to VOOs freshness was on the basis of The quality of virgin olive oils was determined by analytical
the 0.5 cut-off value. For each class, olive oils samples were split parameters: acidity, PV, K232 and K270 . Table 1 shows the mean,
into calibration and prediction sample sets, assigning samples to the standard deviation, the minimum and maximum value of
each set on the basis of their position of the class file, i.e. 50% of these analytical parameters. Statistically significant differences
sample was used for calibration and the remainder for prediction among VOO samples were observed for acidity, PV and UV
Table 1
Range of chemical parameters of virgin olive oil samples
Fresh (n = 35) Dark 1 year (n = 16) Dark 2 years (n = 24) Light 1 year (n = 16)
PV, peroxide value; K232 and K270 UV absorbance at 232 and 270 nm; S.D., standard deviation; min, the minimum value, max, the maximum value.
*** Significant difference (p < 0.001).
N. Sinelli et al. / Analytica Chimica Acta 598 (2007) 128–134 131
Fig. 3. Score plot of the first two PCs obtained by PCA applied to the second
Fig. 1. Biplot of the first two PCs of olive oil samples using chemical data. derivative of virgin olive oils spectra.
132 N. Sinelli et al. / Analytica Chimica Acta 598 (2007) 128–134
Table 2
PLS-DA classification results of virgin olive oil samples using FTIR spectral data (class 1 = fresh; class 2 = dark 1 year; class 3 = dark 2 years; class 4 = light 1 year)
Class Calibration Prediction
No. samples Total % samples False negative False positive No. samples correctly % samples correctly False negative False positive
correctly classified correctly classified classified classified
Table 3
LDA classification results of virgin olive oil samples using FTIR spectral data (class 1 = fresh; class 2 = dark 1 year; class 3 = dark 2 years; class 4 = light 1 year)
Classes 1 2 3 4 Total
Fitting 1 35 (100%) 0 0 0 35
2 0 14 (87%) 2 6 16
3 0 2 22 (92%) 0 24
4 0 0 0 16 (100%) 16
Cross-validation 1 35 (100%) 0 0 9 35
2 0 12 (75%) 4 6 16
3 0 2 22 (92%) 0 24
4 0 0 0 16 (100%) 16
Table 4
SIMCA classification results using four classes of virgin olive oil samples (class 1 = fresh; class 2 = dark 1 year; class 3 = dark 2 years; class 4 = light 1 year)
Classes PCs No. of samples correctly classified % Samples correctly classified False negative False positive
1 4 18/18 (100%) 0 0
2 6 5/8 (62%) 3 7
78%
3 6 5/12 (42%) 7 3
4 6 8/8 (100%) 0 0
Table 5
SIMCA classification results using three classes of virgin olive oil samples (class 1 = fresh; class 2 = dark 1 and 2 years; class 3 = light 1 year)
Classes PCs No. of samples correctly classified % Samples correctly classified False negative False positive
1 4 18/18 (100%) 0 0
2 6 20/20 (100%) 100% 0 0
3 6 8/8 (100%) 0 0
ing only three classes, one of them including all olive oils stored olive oil samples. In addition, the same spectra can be used to
in darkness (both for 1 and 2 years). The improved results are obtain quantitative information on several parameters used for
shown in Table 5. The new total rate of samples correctly classi- routine quality control of oils.
fied was 100%. In fact, the fresh olive oils, the olive oils oxidized A higher number of samples might be necessary to develop
in the dark for 1 and 2 years, and in the light were correctly a more robust method to be used commercially as an indus-
identified. trial application. More samples are also needed to improve the
precision of the classification models developed.
Anyway, the work points out the great advantages of using
4. Conclusions
ATR-MIR spectroscopy instead of chemical analysis. Advan-
tages are related to rapidity, non-destructive characteristics and
These results demonstrated that ATR-MIR spectroscopy is
low cost per sample of MIR spectroscopy as well as need-
able to classify virgin olive oil samples stored in different con-
ing neither particular sample preparation nor specially qualified
ditions and to evaluate the freshness of olive oils (i.e. their
laboratory personnel.
oxidation). In fact, the FTIR PCA results allowed a better
In conclusion, ATR-MIR could represent a reliable, cheap
discrimination among fresh and oxidized oils, than samples sep-
and fast classification tool, able to verify the freshness of virgin
aration obtained by PCA applied to chemical data. Moreover,
olive oils, not requiring chemical analyses for discrimination
FTIR spectroscopy gave sufficient information to describe the
among freshness classes.
different storage conditions and storage periods. The ability of
FTIR spectra to discriminate among VOOs stored in the light
and in the dark was also demonstrated by the different classifi- References
cation technique. It was also interesting to note the absence of
[1] A. Bendini, L. Cerretani, S. Vecchi, A. Carrasco-Pancorbo, G. Lercker, J.
any false positive in PLS-DA and SIMCA models, using three Agric. Food Chem. 54 (2002) 4880.
classes. [2] F. Shahidi, U.N. Wanasundara, In: Fereidoon Shahidi (Ed.) Natural Antiox-
Therefore, the spectroscopic technique could be applied for idant, Chemistry, Health Effect and Applications, 1996.
the prediction of olive oil freshness providing information com- [3] E.C. Off. J. Eur. Communities, Regulation 2568/1991, 11 July 1991.
parable to chemical modifications, as FTIR spectroscopy results [4] COI/T.15/NC no. 2/REV. 10, 2001.
[5] V. Papadimitriou, T.G. Sotiroudis, A. Xenakis, N. Sofikiti, V. Stavyian-
are in agreement with those obtained by classical methods noudaki, N.A. Chaniotakis, Anal. Chim. Acta 573 (2006) 453.
results. The success of this technique depends on the concen- [6] T. Oshima, A. Hopia, B. German, E.N. Frankel, Lipids 31 (1996) 1091.
tration and ratio between the major chemical components of the [7] P. Rovellini, N. Cortesi, E. Fedeli, Riv. Ital. Sostanze Grasse 75 (1998) 57.
134 N. Sinelli et al. / Analytica Chimica Acta 598 (2007) 128–134
[8] M.T. Morales, R. Aparicio, J.J. Rios, J. Chromatogr. A 668 (1994) 455. [17] K.R. Beebe, R.J. Pell, M.B. Seasholtz, Chemometrics, A Practical Guide,
[9] F. Caponio, M.T. Bilancia, A. Pasqualone, E. Sikorska, T. Gomes, Eur. John Wiley and Sons, New York, NY, 1998.
Food Res. Technol. (2005) 92. [18] T. Naes, T. Isaksson, T. Fearn, A.M.C. Davies, A User-friendly Guide to
[10] D.A. Skoog, F.J. Holler, T.A. Nieman, Principles of Instrumental Analysis, Multivariate Calibration and Classification, NIR Publication, Chichester,
fifth ed., Saunders College Pub., Philadelphia, 1998. UK, 2000.
[11] N.B. Colthup, L.H. Daly, S.E. Wiberly, Introduction to Infrared and Raman [19] M. Meloun, J. Militky, M. Forina, Chemometrics for Analytical Chemistry,
Spectroscopy, third ed., Accademic Press, Inc., New York, NY, 1990. Ellis Horwood, New York, NY, 1992.
[12] M. Khalique Ahmed, J.K. Daun, R. Przybylski, J. Food Compos., Anal. 18 [20] M. Blanco, J. Pages, Anal. Chem. 463 (2002) 295.
(2005) 359. [21] Q. Chen, J. Zhao, H. Zhang, X. Wang, Anal. Chem. 572 (2006) 77.
[13] H. Yang, J. Irudayaraj, M.M. Paradkar, Food Chem. 93 (2005) 25. [22] E.C. Off. J. Eur. Communities, Regulation 1513/2001, 23 July 2001.
[14] M. Safar, D. Bertrand, P. Robert, M.F. Devaux, C. Genot, J. Am. Oil Chem. [23] M.D. Guillen, N. Cabo, J. Agric. Food Chem. 46 (1998) 1788.
Soc. 71 (1994) 371. [24] J. Coates, in: R.A. Meyers (Ed.), Encyclopaedia of Analytical Chemistry,
[15] D.B. Dahlberg, S.M. Lee, S.J. Wenger, J.A. Vargo, Appl. Spectrosc. 51 John Wiley & Sons Ltd., Chichester, 2000, p. 10815.
(1997) 1118. [25] J. Workman, Handbook of Organic Compounds, Academic Press, London,
[16] Y.W. Lai, E.K. Kemsley, R.H. Wilson, J. Agric. Food Chem. 42 (1994) 2001, p. 209.
1154.