- Email: [email protected]
Detection of virgin olive oil adulteration using a voltammetric e-tongue
Computers and Electronics in Agriculture 108 (2014) 148–154
Contents lists available at ScienceDirect
Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag
Detection of virgin olive oil adulteration using a voltammetric e-tongue I.M. Apetrei a, C. Apetrei b,⇑ a b
Department of Pharmaceutical Sciences, Faculty of Medicine and Pharmacy, ‘‘Dunarea de Jos’’ University of Galati, Romania Department of Chemistry, Physics and Environment, Faculty of Sciences and Environment, ‘‘Dunarea de Jos’’ University of Galati, 47 Domneasca Street, 800008 Galati, Romania
a r t i c l e
i n f o
Article history: Received 29 March 2014 Received in revised form 12 June 2014 Accepted 2 August 2014
Keywords: Adulteration Virgin olive oil e-Tongue Polyphenol
a b s t r a c t The detection of adulteration in extra virgin olive oils is of great interest in food industry. This article presents the first use of a voltammetric e-tongue for the detection of the adulteration of virgin olive oil. Adulterations of an extra virgin olive oil with different percentages of sunflower oil, soybean oil and corn oil were measured using modified carbon paste based sensors. The square wave voltammetric signals were processed using kernel method. Chemometric methods applied allows discrimination and classification of oils in agreement with botanical origins. Excellent correlations between voltammetric signals and polyphenolic content was obtained by PLS regression. PLS-DA and PLS regression demonstrated the feasibility of detecting adulterations of olive oil with percentages lower than 10% of sunflower, soybean and corn oils. These results indicate that e-tongue can be a useful tool for the detection of olive oil adulteration with seed oils. Ó 2014 Elsevier B.V. All rights reserved.
1. Introduction Extra virgin olive oil (EVOO) is obtained from the olive fruit named Olea europaea. EVOO is extracted by only mechanical procedure without application of refining process. The quality of olive oil ranges from the high quality EVOO to the low quality olive– pomace oil (Tsimidou, 2006). It is one of the primary ingredients of the Mediterranean diet (Psaltopoulou et al., 2004). Different factors such as cultivar, environment and cultural practices determine the quality and uniqueness of specific EVOOs (Tura et al., 2009). The agreeable taste and aroma with the health benefits of EVOOs are important reasons for consumers to consume this product (Amirante et al., 2012). Oil pureness is a very important aspect of the quality edible oil for special reasons of the sensory properties, perceived health values and the confidence of the health foods (Cozzolino, 2012). Because of the high value of EVOO, it can be adulterated with other oils of lower commercial value. The most common adulterants found in EVOO are refined olive oil, seed oils and nut oils (Dourtoglou et al., 2003). The detection of adulteration in EVOOs is a particular concern in the food industry. The adulteration of extra virgin olive oil with other cheaper oils can lead to significant profits for the unscrupulous vendor or raw material supplier. Therefore, continuous caution is required to control the adulteration of EVOOs and to protect the interests of the consumers. ⇑ Corresponding author. Tel.: +40 0236460328; fax: +40 0236461353. E-mail address: [email protected] (C. Apetrei). http://dx.doi.org/10.1016/j.compag.2014.08.002 0168-1699/Ó 2014 Elsevier B.V. All rights reserved.
In order to detect VOO adulteration, a number of chromatographic and spectroscopic methods, including fluorescence (Kunz et al., 2011), near-infrared (NIR) (Özdemir and Öztürk, 2007), Fourier transform infrared spectroscopy (FT-IR), FT-Raman (Heise et al., 2005), nuclear magnetic resonance (NMR) (Xu et al., 2014), mass spectrometry (MR) (Calvano et al., 2012), mid-infrared spectroscopy (MIR) (Gurdeniz and Ozen, 2009) and high-performance liquid chromatography (HPLC) (Lísa et al., 2009) were used. All these methods were applied with some success. However, there are also some shortcomings. Chromatographic, mass and NMR spectroscopy methods need expensive instruments. Fluorescence spectroscopy and chromatography methods require preprocessing of the samples. Furthermore, these methods are not suitable for on-line, in-line or real time analysis. The multivariate data analysis like linear discriminant analysis (LDA), principal component analysis (PCA), partial least square regression (PLS), machine learning (ML), and artificial neural networks (ANN) are applied to further analyze the oil spectroscopy data for oil adulteration detection (Xu et al., 2014; Gurdeniz and Ozen, 2009). Nowadays there is an increasing interest for simple and fast techniques called e-nose and e-tongue for various applications in food industry (Peris and Escuder-Gilabert, 2013). An e-tongue is an device, which consists of an array of chemical sensors and an appropriate pattern recognition method, for recognition (identification, classification, discrimination) of quantitative multicomponent analysis and artificial assessment of taste and flavour of various liquids. In the case of e-tongues, their capability to analyse
149
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
and discriminate a variety beverages such as mineral waters, milks, teas, wines or beers has already been established (Cetó et al., 2014; Yaroshenko et al., 2014; Gutiérrez-Capitán et al., 2013; GhasemiVarnamkhasti et al., 2012; Apetrei et al., 2012). However, few works have been focused to the analysis of olive oils using e-tongues (Cosio et al., 2007; Haddi et al., 2013; Apetrei and Apetrei, 2013). In previous works we have developed different strategies in order to use an voltammetric e-tongue to discriminate oils of different origins and qualities (Apetrei et al., 2007, 2010, 2012). In one method, the olive oil is binder used for preparation of carbon paste-based sensor. The characteristics of the voltammograms observed when the sensor is immersed in different electrolytic solutions reflect the electroactive properties of the oils (Apetrei et al., 2007, 2010). The aim of this work is to evaluate the feasibility of e-tongue (based on modified EO carbon paste-based sensors) to detect the adulteration qualitatively and precisely of EOs with different seed oils based on their tasting fingerprints. Based on our knowledge, it is the first time that a voltammetric e-tongue is employed for detection of adulteration of extra virgin olive oil. Various chemometric methods were performed for discrimination and classification of adulteration levels. 2. Experimental 2.1. Reagent and chemicals Potassium chloride (KCl) was obtained from Sigma–Aldrich. Hydrochloric acid (HCl) was purchased from S.C. Chemical Company S.A. (Romania). Water was purified on a Milli-Q SimplicityÒ Water Purification System (Millipore Corporation) with resistivity of 18.2 MX. Electrolyte solutions were prepared using ultrapure water. Carbon nanopowder, <50 nm particle size (TEM), P99% trace metals basis from Aldrich was used for fabrication of carbon paste electrodes modified with edible oils. 2.2. Samples Four kinds of edible vegetable oil were purchased from local supermarkets, including the extra virgin olive oil, the sunflower oil, the soybean oil and the corn (maize) oil. The oil samples were stored in the fridge at 4 °C until the day of analysis. The adulterant oil is considered the seed oil at different level of concentrations. For e-tongue measurements, mixtures of
seed and extra virgin olive oils, with concentration of seed oil at different levels (2%, 5%, 10%, 20% and 25%) were prepared (Table 1). The mixtures of oils were prepared using an Elmasonic S10H ultrasonic bath. Total phenolic content was determined spectrophotometrically following the Folin–Ciocalteu spectrophotometric method (Herchi et al., 2011). The results are presented in Table 2.
2.3. e-Tongue system For electronic tongue measurements carbon paste electrodes (CPE) modified with edible oils were prepared as previously described (Apetrei et al., 2007, 2010). Voltammetric measurements were carried out in an Biologic Science Instruments SP 150 potentiostat/galvanostat (EC-Lab Express software) using a conventional three-electrode cell. The modified EO carbon paste-based sensors were used as working electrodes. The reference electrode was an Ag/AgCl KCl 3M and the counter electrode was a platinum wire. The electrochemical experiments were performed at a controlled temperature of 25 °C. Three identical EO carbon paste-based sensors were prepared for each oil under study. Each replicate sensor was immersed in one electrolytic solution (0.1 mol L1 HCl or 0.1 mol L1 KCl) and the voltammetric response was registered. Voltammetric measurements were carried out by means of SWV (Square wave voltammetry). SWV were recorded using a frequency of 15 Hz, an amplitude of 0.10 V and a step high of 0.005 V. The SWV curves were registered in the potential range from 0.2 to +1.3 V. Seven measurement replicates were carried out with each sensor (measured after the conditioning step) in both electrolyte solutions.
2.4. Data analysis Several steps were carried out in order to obtain the input matrix for multivariate data analysis. SWV curves were pre-processed using the adaptation of a data reduction technique based on predefined response ‘‘bell shaped-windowing’’ curves called ‘‘kernels’’ (Apetrei et al., 2010). Using this method, ten parameters per each SWV were obtained. Because in this case, the variables are the edible oils from the sensors, the initial input matrix was transposed. In this way, the variables are the electrolyte solutions and the samples are the edible oils from the carbon paste-based sensors.
Table 1 Edible oil samples under study. Sample
Edible oil
Concentration/%
EVOO Sf S C EVOO + Sf EVOO + S EVOO + C
Olive oil Sunflower oil Soybean oil Corn oil Olive oil/sunflower oil Olive oil/soybean oil Olive oil/corn oil
100 100 100 100 98/2 98/2 98/2
95/5 95/5 95/5
90/10 90/10 90/10
80/20 80/20 80/20
75/25 75/25 75/25
Table 2 Total phenolic content of oil samples determined spectrophotometrically. Sample
Total polyphenols content (mg kg1) RSD = 4.5%.
EVOO
275.43
EVOO + Sf
EVOO + S
EVOO + C
98/2
95/5
90/10
80/20
75/25
98/2
95/5
90/10
80/20
75/25
98/2
95/5
90/10
80/20
75/25
271.92
261.63
249.80
220.75
208.23
270.75
260.54
250.18
221.24
207.41
272.43
260.74
250.17
219.94
208.95
150
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
For variable reduction and separation into classes, principal component analysis (PCA) was used. Partial least squares-discriminant analysis (PLS-DA) was used as deterministic classification technique (Gemperline, 2006). For sample groups establishing based on botanical origin analysis of variance (ANOVA) was used. PLS regression was used to establish the correlations between polyphenol contents or concentration of adulterant oil and electrochemical responses (Gemperline, 2006). Both the X-matrix (containing the e-tongue data values) and the Y-matrix (containing the physico-chemical parameter) were pre-processed using a normalisation routine. The optimum number of latent variables (LV) was selected by choosing the first local minimum in the residual Y variance plot. Validation of the predictive models was carried out by full cross-validation method (leave one-out). The multivariate data analysis was performed by using the software Matlab 5.3., Excel 2007 and The Unscrambler 9.1. 3. Results and discussion 3.1. Measurements with the e-tongue Measurements of the e-tongue were carried out using the EO carbon paste-based sensors. Preliminary studies and stabilization of the sensor signals were carried out by cyclic voltammetry (CV). CV was used to study qualitative information about
electrochemical processes, the reversibility of the reactions and stability of the sensor response. The CV experiments were carried out from 0.2 V to +1.3 V with a scan rate of 0.05 V s1. Under these conditions, EVOO carbon paste-based sensors showed a variety of anodic and cathodic peaks. The CV of EVOO carbon paste-based sensors immersed in HCl 0.1 mol L1 showed two anodic peaks at 0.54 V, and 1.05 V and one cathodic peak at 0.04 V (Fig. 1a). However, the resolution of the peaks needed to be improved. A better resolution of the peaks was obtained in forward scan when the SWV technique was used (Fig. 1b). The number and the potential of the peaks remained unchanged but an important increase of the peak currents were observed. Therefore, the use of SWV increases the sensitivity of the method. Furthermore, the SWV curves facilitate data treatment because the curves are pairs current (i)–potential (E). EVOO has in its composition mainly triacylglycerols, polyphenols, tocopherols, pigments, aromatic components and small amount of free fatty acids (Quiles et al., 2006). Polyphenols and tocopherols have antioxidant properties and therefore these compounds can be detected by means of electroanalytic methods (Kim and Kusuda, 1994). In agreement with literature, peaks at 0.54 and at 0.04 V are related to the polyphenolic compounds present in of EVOO (Shiragami et al., 1994). The redox peak found at 1.05 V is related to the presence of tocopherols present in EVOO (Kim and Kusuda, 1994). The electrochemical reaction at the sensor surface is presented in Fig. 2. The dependence of Ep and Ep/2, expressed using Nicholson and Shain equation (Southampton, 1990), can be used to calculate the number of electrons (n) participating in the rate determining stage for an irreversible charge transfer (Eq. (1)):
Ep Ep=2 ¼
Fig. 1. Voltammetric responses of EO carbon paste-based sensors immersed in aqueous HCl 0.1 mol L1. EVOO carbon paste-based sensors: (a) CV; (b) SWV. Sf carbon paste-based sensors: (c) CV; (d) SWV. S carbon paste-based sensors: (e) CV; (f) SWV. C carbon paste-based sensors: (g) CV; (h) SWV.
1:857 R T 47:7 ¼ ðmVÞ anF an
ð1Þ
where Ep is the potential of the peak, Ep/2 is the potential of half peak height, R is the universal gas constant, T is the absolute temperature, F is Faraday constant and a is the charge transfer coefficient. The n value calculated assuming a of 0.5 is 2.01. Therefore, the electrochemical process of tocopherols at the sensor surface involve 2 electrons (Fig. 2). Similar results, number of electrons involved close to 2 were obtained for all EO modified sensors indicating the involvement of the same compound in the redox process. The CV and SWV of Sf carbon paste-based sensors showed one anodic peak associated to tocopherols at ca. 1.06 V (Fig. 1c and d). Sunflower oil contains predominantly linoleic acid in triglyceride form, lecithin, tocopherols, carotenoids and waxes. Sunflower oil is light in taste and appearance. Therefore, the main electroactive compounds from sunflower oil are the tocopherols (Gunstone, 2011). The voltammograms obtained for corn oil and soybean were similar to those obtained for sunflower oil. However, shifts in the peak potentials were observed in Fig. 1e–h (corn oil, Ea = 1.01 V; soybean oil, Ea = 1.00 V). Therefore, the electroactive substances present in corn oil and soybean oil are also the tocopherols. The differences observed in the voltammetric curves are also related to different physico-chemical properties of the oils, i.e. viscosity. Since oils are mixtures of triglycerides viscosity depends on the nature of fatty acids from triglyceride molecules and their properties such as chain length and saturation/unsaturation. In conclusion, the intensity of the observed peaks depend on concentration of tocopherols, their relative concentrations in edible oils and triglyceride profile (Gunstone, 2011).
151
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154 Table 3 ANOVA significance results.
* ** ***
Contrast
Significance
EVOO vs. Sf EVOO vs. S EVOO vs. C Sf vs. C Sf vs. S S vs. C
*** *** *** ** ** *
p < 0.05 (Significant). p < 0.01 (Highly significant). p < 0.0001 (Extremely significant).
Fig. 2. The redox processes of tocopherol at the sensor surface. Table 4 Quantitative data of PLS-DA.
Fig. 3. PCA score plot corresponding to the SWVs data treatment. O – extra virgin olive oil; Sf – sunflower oil; S – soybean oil; C – corn oil.
The voltammetric curves shown in Fig. 1 indicate that the nature and concentration of electroactive compounds found in different edible oils bearing a diversity of electrochemical signals. The voltammetric responses can be used to discriminate edible oils. From the analyse of the voltammetric curves it can be observed that EVOO related curves can be easily distinguished from other oils. Only in the case of EVOO peaks related to polyphenols can be observed.
Type
RMSEC
RMSEP
Extra virgin olive oil Sunflower oil Soybean oil Corn oil
0.066 0.126 0.092 0.162
0.082 0.134 0.103 0.171
graph. Moreover, the method is capable to distinguish among seed oils. Using the electronic tongue, the first principal component is the main responsible of the discrimination of the EVOO from other vegetable oils, whereas second principal component is the main responsible of the discrimination of seeds oils. Electrochemical data were tested for statistical significance using one-way ANOVA routines running under Excel. The factor was botanical origin of edible oils. Values of p < 0.05 were considered statistically significant. In Table 3 are presented the ANOVA significance results. ANOVA was based on honestly significant difference test (Tukey test). This test is based on pairwise comparison among means as presented in Eq. (2):
Mi Mj ts ¼ qffiffiffiffiffiffiffi MSE nh
ð2Þ
Mi – Mj = difference between pair means. MSE = mean square error. nh = the harmonized mean.
3.2. Classification analysis of edible oils In this work different methods to analyse the voltammetric signals related to oil samples were applied. First, PCA, PLS-DA and ANOVA were applied to discriminate and classify the edible oils in agreement with their origin. After that, PCA, PLS-DA and ANOVA were applied to discriminate and classify the VOO adulterated with one adulterant oil. Further, all oil samples, were analyzed by means of PCA and PLS-DA in order to discriminate and classify the original and adulterated VOOs with seed oils in different proportions. PLS regression models were developed and used for determination of concentration limit below that the e-tongue is not able to distinguish the adulterated oil from authentic oil. The discrimination capacity of the e-tongue was evaluated by means of Principal Component Analysis (PCA) of the voltammetric signals. Fig. 3 shows the scores plot of the two first principal components calculated from the parameters extracted from the SWV curves. The first two principal components explain the 73% of the information (PC1 = 46%; PC2 = 27%). The PCA score plot shown in Fig. 3 indicates that the system based on CPE sensors immersed in different electrolytes is able to discriminate among vegetable oils of different origins. Olive oil appeared in the left region of the plot, whereas, the seed oils whose do not contain polyphenols, appear in the opposite side of the
As presented in Table 3 the population means are significantly different. The samples could be classified in groups in agreement with botanical origin of the oil. ANOVA showed significant differences between these groups of oils based on e-tongue results. A supplementary confirmation of the groups shown in PCA was carried out by PLS-DA. Quantitative data of PLS-DA are presented in Table 4. The optimal number of latent variables (5 LVs) was determined by the lowest value of predicted residual error sum of squares. Root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP) have low values indicating the high quality of the model. For all EO samples, when the constructed PLS-DA model was applied, 100% of samples are correctly classified both in calibration and in validation. The validation was performed using leave-one-out fully cross validation. The sensitivity and specificity of the model is 100%. 3.3. Discrimination and classification of adulterated EVOOs In order to determine the capability of the e-tongue to detect adulteration of EVOO, carbon paste-based sensors modified with adulterated oils were constructed. Relative concentrations of oils used in sensors fabrication are presented in Table 1. The SWVs of sensors modified with Sf, EVOO and EVOO/Sf immersed in HCl
152
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
Fig. 4. The SWVs of sensors modified with Sf (——), EVOO (____) and EVOO/Sf (95/5 _._._ ; 90/10 _.._.._; 80/20 ........; 75/25 __ __ __) immersed in HCl 0.1 mol L1.
0.1 mol L1 are presented in Fig. 4. As can be observed, the SWVs of sensors modified with EVOO/Sf present an intermediary behaviour between pure EVOO and Sf oil. The peak corresponding to polyphenols decrease when the concentration of Sf increase. Contrary, the peak corresponding to tocopherol increase when the concentration of Sf increase. The voltammetric responses are related to changing in concentration of electroactive compounds, polyphenols and tocopherols. When Sf is added in EVOO the concentration of polyphenols decrease and the concentration of tocopherol increase. In agreement with literature, Sf oils have higher concentration in tocopherols than EVOO (Gunstone, 2011). Analogous results are obtained in the cases of C and S oils. In the SWVs, the peaks related to polyphenols decrease when the concentration of adulterant oil increase. The peak related to tocopherol remains almost constant in the case of corn oil and slightly decreases in the case of soybean oil. The establishment of correlations between the results obtained by means of e-tongue and the total polyphenolic content is of vital importance for detection of EVOO adulteration. It is well-known that polyphenols are the main contributors to virgin olive oil taste. Furthermore, the main difference between the SWV curves of EVOO modified sensors and EO modified sensors consists in the peak pair related to polyphenols that appears only in the case of EVOO. For these reasons, PLS regression was performed to model the relationships between the voltammetric signals and the total polyphenolic content obtained by means of spectrophotometric method (Herchi et al., 2011). As observed in Fig. 5, a very good correlation was found, with a coefficient of determination (R2) of 0.9888 in calibration and 0.9852 in validation, using 4 latent variables. RMSEC was 2.6773 and RMSEP was 3.0723, demonstrating the quality of the model.
This result indicated that the polyphenolic content is strongly related to the voltammetric responses provided by the electronic tongue. Therefore, the voltammetric e-tongue could detect the adulteration of EVOOs based on modification of polyphenolic content. The results described above suggest that voltammetric e-tongue might be able to detect the presence of other vegetable oils in EVOO. To assess this possibility, the SWV curves of 15 binary mixtures (composition range as presented in Table 1) and 4 pure edible oils were measured in replicates (3 sensors for each oil and binary mixture 2 electrolyte solutions 7 replicates). In the first step of the data treatment, PCA was applied to the processed voltammetric data, and it was found that first three PCs accounting for 82% of the total explained variance successfully separate the mixtures according to composition (Fig. 6). As observed in the Fig. 6, the pure oils are distributed in a similar way such in Fig. 3. The EVOO cluster appears in the opposite side of the graph relative to seed oils. The adulterated oils clusters are situated among pure oils clusters, near to cluster corresponding to EVOO. However, the clusters corresponding to adulterated EVOO with small quantities of seed oil are not totally separated from pure EVOO cluster. PLS-DA and PLS models are necessary in order to quantify the level of adulteration in EVOO. It can be observed that e-tongue is able to distinguish among oils adulterated with different seed oils. The PCA result strongly suggests that these SWV curves contain electroanalytical information that will allow correlation of voltammetric curves features with the concentration of the adulterant oil. For that reason, subsequent supervised analyses using PLS and PLS-DA were used to develop models to quantify the level of adulteration in EVOO. The data were split into a training (or calibration) set and a test set (validation). The training set consisted of 5 replicate SWVs of all oil samples (71.43%), while the test set consisted of 2 replicate SWVs (28.57%). The models were developed for EVOO, pure seed oil, and adulterated samples with corresponding seed oil. Therefore, three models were developed corresponding to EVOO adulterated with sunflower oil, corn oil and soybean oil, respectively. The quantitative PLS-DA models are present in Table 5 in terms of correct and misclassified samples percentages. It was found that the best models (i.e. lowest prediction error in the test set) occurred when six latent variables were used, in all cases. In the case of model of EVOO adulterated with sunflower oil, the sample replicates of 2% Sf/EVOO can be distinguish from pure EVOO in a percentage of 42.86%. The sample replicates of 5% Sf/EVOO can be distinguish from pure EVOO in a percentage of 85.71%. The sample replicates of adulterated oil containing 10% of sunflower oil were correctly classified. In the case of EVOO adulterated with soybean oil the sample replicates are 100% correct classified when the concentration of
Fig. 5. Plots of predicted total polyphenolic content obtained with e-tongue vs. the values of total polyphenolic content obtained by Folin–Ciocalteau method.
153
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
Fig. 7. PLS model for the relationship between actual value and e-tongue predicted value of sunflower oil in EVOO in calibration.
Fig. 6. PCA scores plot corresponding to the SWVs of pure and adulterated oils. O – olive oil; S – soybean oil; Sf – sunflower oil; C – corn oil; Sf2, Sf5, Sf10, Sf20, Sf25 – mixtures of sunflower oil and extra virgin olive oil with concentration of sunflower oil at different levels (2%, 5%, 10%, 20% and 25%); S2, S5, S10, S20, S25 – mixtures of soybean oil and extra virgin olive oil with concentration of soybean oil at different levels (2%, 5%, 10%, 20% and 25%); C2, C5, C10, C20, C25 – mixtures of corn oil and extra virgin olive oil with concentration of corn oil at different levels (2%, 5%, 10%, 20% and 25%).
Table 5 Quantitative data of PLS-DA. Model
Sample
Correct classified sample replicates/%
Misclassified samples replicates/%
Olive oil/sunflower oil
EVOO Sf EVOO + Sf EVOO + Sf EVOO + Sf EVOO + Sf EVOO + Sf
100 100 42.86 85.71 100 100 100
0 0 57.14 14.29 0 0 0
98/2 95/5 90/10 80/20 75/25
100 100 14.29 57.14 100 100 100
0 0 85.71 42.86 0 0 0
98/2 95/5 90/10 80/20 75/25
100 100 28.57 71.43 100 100 100
0 0 71.43 28.57 0 0 0
Olive oil/soybean oil
Olive oil/corn oil
EVOO S EVOO + S EVOO + S EVOO + S EVOO + S EVOO + S EVOO C EVOO + C EVOO + C EVOO + C EVOO + C EVOO + C
98/2 95/5 90/10 80/20 75/25
Fig. 8. Variation of anodic peak potential related to polyphenols in time for EVOOCPE.
3.4. Quantification of Sf, S, and CO in adulterated EVOO samples
soybean oil is 20%, whereas EVOO adulterated with corn oil are 100% correct classified when the concentration of corn oil is 10%. Therefore, the e-tongue is able to detect presence of seed oils in EVOO at levels lower than 10% adulterant oil, level important for practical applications.
Quantification of Sf, S, and CO was carried out with the aid of multivariate calibration. Calibration models developed by means of partial least square (PLS) were used to evaluate the goodness of fit for the relationship between real value (x-axis) and e-tongue predicted value (y-axis) of Sf, S and C oils in EVOO. PLS regression model was calibrated with the training set and cross-validated with the test set. It was found that the best models (i.e. lowest prediction error in the test set) occurred when five latent variables were used. Table 6 compiled the performance of PLS for quantification of Sf, S, and C in EVOO. PLS gives the reasonable R2 either in calibration or in validation (Rc2 = 0.9996; Rp2 = 0.9987) and offers small errors in calibration and validation. Using this model, four LVs are needed to obtained RMSEC value of 0.1670 and RMSEP value of 0.3105. The scatter plot for the relationship between real value and e-tongue predicted value of Sf in EVOO was revealed in Fig. 7. Based on these results, it can be assumed that e-tongue facilitate the detection and quantification of Sf, S and C oils adulterants in EVOO. Furthermore, the models indicate that the composition of the mixture can be successfully predicted in the range of commercial interest, 2–25%. The results are comparable with those reported for other systems
Table 6 The performances of multivariate calibration for analyses of Sf, S and C oils in EVOO. Model
Extra virgin olive oil/sunflower oil Extra virgin olive oil/soybean oil Extra virgin olive oil/corn oil
R2
Equation Calibration
Prediction
Calibration
Prediction
y = 0.9996x + 0.0044 y = 0.9987x + 0.0155 y = 0.9988x + 0.0149
y = 0.9926x + 0.0659 y = 0.9861x + 0.1102 y = 0.9864x + 0.1149
0.9996 0.9987 0.9988
0.9987 0.9953 0.9956
RMSEC
RMSEP
0.1670 0.3107 0.3048
0.3105 0.5962 0.5793
154
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
employed in detection of EVOO adulteration (Gurdeniz and Ozen, 2009; Xu et al., 2014). 3.5. Stability of the sensors CVs of EVOO-CPE immersed in 0.1 M HCl aqueous solution were registered ten consecutive days. CPEs were withdrawing from the solution after each experiment and smoothing the surface with a clean paper filter. Among measurements EVOO-CPE was stored in the dark at 4 °C in dry atmosphere. The variation of anodic peak potential related to polyphenols in time was shown a coefficient of variation of 5% between first and tenth day (Fig. 8). Therefore, the EVOO-CPE presents a good stability since electroactive compounds are included in carbonaceous matrix and the access of oxygen is limited. 4. Conclusions A novel method based on a voltammetric e-tongue to evaluate the adulteration of extra virgin olive oils has been established. The responses obtained using sensors modified with edible oils show peaks related to tocopherols and polyphenols. The voltammetric curves can be used as input variables in chemometrics. It has been shown that e-tongue can successfully discriminate and classify among extra virgin olive, sunflower, corn and soybean oils. PLS-DA models showed that the applied e-tongue is able to classify correctly the adulterated oils when the concentration level of adulterant oil was between 5% and 10%. Moreover, the composition of seed oils and extra virgin olive oil mixtures could be accurately predicted using PLS regression. In conclusion, e-tongue consisting of voltammetric sensors together with appropriate chemometrics, presents itself as a powerful tool for the detection of adulteration of extra virgin olive oil. Application of e-tongue coupled with e-nose to detection of olive oil adulteration could enhance the capability of the system even at lower levels of adulteration. Acknowledgments This work was supported by a Grant of the Romanian National Authority for Scientific Research, CNCS – UEFISCDI, Project number PN-II-ID-PCE-2011-3-0255. References Amirante, P., Clodoveo, M.L., Tamborrino, A., Leone, A., Dugo, G., 2012. Oxygen concentration control during olive oil extraction process: a new system to emphasize the organoleptic and healthy properties of virgin olive oil. Acta Hort. 949, 473–480. Apetrei, I.M., Rodriguez-Mendez, M.L., Apetrei, C., Nevares, I., del Alamo, M., de Saja, J.A., 2012. Monitoring of evolution during red wine aging in oak barrels and alternative method by means of an electronic panel test. Food Res. Int. 45, 244– 249. Apetrei, C., 2012. Novel method based on polypyrrole-modified sensors and emulsions for the evaluation of bitterness in extra virgin olive oils. Food Res. Int. 48, 673–680. Apetrei, C., Apetrei, I.M., Villanueva, S., de Saja, J.A., Gutierrez-Rosales, F., RodriguezMendez, M.L., 2010. Combination of an e-nose, an e-tongue and an e-eye for the characterisation of olive oils with different degree of bitterness. Anal. Chim. Acta 663, 91–97. Apetrei, C., Gutierez, F., Rodríguez-Méndez, M.L., de Saja, J.A., 2007. Novel method based on carbon paste electrodes for the evaluation of bitterness in extra virgin olive oils. Sens. Actuat. B – Chem. 121, 567–575.
Apetrei, I.M., Apetrei, C., 2013. Voltammetric e-tongue for the quantification of total polyphenol content in olive oils. Food Res. Int. 54, 2075–2082. Calvano, C.D., De Ceglie, C., D’Accolti, L., Zambonin, C.G., 2012. MALDI-TOF mass spectrometry detection of extra-virgin olive oil adulteration with hazelnut oil by analysis of phospholipids using an ionic liquid as matrix and extraction solvent. Food Chem. 134, 1192–1198. Cetó, X., Apetrei, C., Del Valle, M., Rodríguez-Méndez, M.L., 2014. Evaluation of red wines antioxidant capacity by means of a voltammetric e-tongue with an optimized sensor array. Electrochim. Acta 120, 180–186. Cosio, M.S., Ballabio, D., Benedetti, S., Gigliotti, C., 2007. Evaluation of different storage conditions of extra virgin olive oils with an innovative recognition tool built by means of electronic nose and electronic tongue. Food Chem. 101, 485– 491. Cozzolino, D., 2012. Recent trends on the use of infrared spectroscopy to trace and authenticate natural and agricultural food products. Appl. Spectrosc. Rev. 47, 518–530. Dourtoglou, V.G., Dourtoglou, Th., Antonopoulos, A., Stefanou, E., Lalas, S., Poulos, C., 2003. Detection of olive oil adulteration using principal component analysis applied on total and Regio FA content. J. Am. Oil Chem. Soc. 80, 203–208. Gemperline, P., 2006. Practical Guide to Chemometrics, Taylor & Francis Group LLC. Ghasemi-Varnamkhasti, M., Rodríguez-Méndez, M.L., Mohtasebi, S.S., Apetrei, C., Lozano, J., Ahmadi, H., Razavi, S.H., de Saja, J.A., 2012. Monitoring the aging of beers using a bioelectronic tongue. Food Contr. 25, 216–224. Gunstone, F.D. (Ed.), 2011. Vegetable Oils in Food Technology: Composition, Properties and Uses. Second ed., Blackwell Publishing Ltd. Gurdeniz, G., Ozen, B., 2009. Detection of adulteration of extra-virgin olive oil by chemometric analysis of mid-infrared spectral data. Food Chem. 116, 519–525. Gutiérrez-Capitán, M., Santiago, J.-L., Vila-Planas, J., Llobera, A., Boso, S., Gago, P., Martínez, M.-C., Jiménez-Jorquera, C., 2013. Classification and characterization of different white grape juices by using a hybrid electronic tongue. J. Agric. Food Chem. 61, 9325–9332. Haddi, Z., Alami, H., El Bari, N., Tounsi, M., Barhoumi, H., Maaref, A., JaffrezicRenault, N., Bouchikhi, B., 2013. Electronic nose and tongue combination for improved classification of Moroccan virgin olive oil profiles. Food Res. Int. 54, 1488–1498. Heise, H.M., Damm, U., Lampen, P., Davies, A.N., Mcintyre, P.S., 2005. Spectral variable selection for partial least squares calibration applied to authentication and quantification of extra virgin olive oils using Fourier transform Raman spectroscopy. Appl. Spectrosc. 59, 1286–1294. Herchi, W., Sakouhi, F., Arráez-Román, D., Segura-Carretero, A., Boukhchina, S., Kallel, H., Fernández-Gutierrez, A., 2011. Changes in the content of phenolic compounds in flaxseed oil during development. J. Am. Oil Chem. Soc. 88, 1135– 1142. Kim, O.S., Kusuda, K., 1994. Electrochemical behaviour of cu-tocopherol in a thin film of Nujol, a model of adipose tissue. Bioelectrochem. Bioenerg. 33, 61–65. Kunz, M.R., Ottaway, J., Kalivas, J.H., Georgiou, C.A., Mousdis, G.A., 2011. Updating a synchronous fluorescence spectroscopic virgin olive oil adulteration calibration to a new geographical region. J. Agric. Food Chem. 59, 1051–1057. Lísa, M., Holcˇapek, M., Bohácˇ, M., 2009. Statistical evaluation of triacylglycerol composition in plant oils based on high-performance liquid chromatography – atmospheric pressure chemical ionization mass spectrometry data. J. Agric. Food Chem. 57, 6888–6898. Özdemir, D., Öztürk, B., 2007. Near infrared spectroscopic determination of olive oil adulteration with sunflower and corn oil. J. Food Drug Anal. 15, 40–47. Peris, M., Escuder-Gilabert, L., 2013. On-line monitoring of food fermentation processes using electronic noses and electronic tongues: a review. Anal. Chim. Acta 804, 29–36. Psaltopoulou, T., Naska, A., Orfanos, P., Trichopoulos, D., Mountokalakis, T., Trichopoulou, A., 2004. Olive oil, the Mediterranean diet, and arterial blood pressure: the Greek European Prospective Investigation into Cancer and Nutrition (EPIC) study. Am. J. Clin. Nutr. 80, 1012–1018. Quiles, J.L., Ramírez-Tortosa, M.C., Yaqoob, P., 2006. Olive Oil and Health, CABI. Shiragami, S., Kim, O.S., Kusuda, K., 1994. Direct survey of redox active compounds in sesame oil by cyclic voltammetry – studies on sesaminol, sesamol and hydroxyhydroquinone in nujol. J. Electroanal. Chem. 379, 315–319. Southampton Electrochemgroup, 1990. Instrumental Methods in Electrochemistry. Ellis Horwood Ltd, New York, pp. 178–227, vol. 380. Tsimidou, M.Z., 2006. Olive oil quality. In: Boskou, D., (Ed.), Olive oil: Chemistry and Technology, second ed. Champaign, Illinois: AOCS Publishing, pp. 93–112. Tura, D., Failla, O., Bassi, D., Pedo, S., Serraiocco, A., 2009. Environmental and seasonal influence on virgin olive (Olea europaea L.) oil volatiles in northern Italy. Sci. Hort. 122, 385–392. Xu, Z., Morris, R.H., Bencsik, M., Newton, M.I., 2014. Detection of virgin olive oil adulteration using low field unilateral NMR. Sensors 14, 2028–2035. Yaroshenko, I., Kirsanov, D., Kartsova, L., Bhattacharyya, N., Sarkar, S., Legin, A., 2014. On the application of simple matrix methods for electronic tongue data processing: case study with black tea samples. Sens. Actuat. B – Chem. 191, 67– 74.
Contents lists available at ScienceDirect
Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag
Detection of virgin olive oil adulteration using a voltammetric e-tongue I.M. Apetrei a, C. Apetrei b,⇑ a b
Department of Pharmaceutical Sciences, Faculty of Medicine and Pharmacy, ‘‘Dunarea de Jos’’ University of Galati, Romania Department of Chemistry, Physics and Environment, Faculty of Sciences and Environment, ‘‘Dunarea de Jos’’ University of Galati, 47 Domneasca Street, 800008 Galati, Romania
a r t i c l e
i n f o
Article history: Received 29 March 2014 Received in revised form 12 June 2014 Accepted 2 August 2014
Keywords: Adulteration Virgin olive oil e-Tongue Polyphenol
a b s t r a c t The detection of adulteration in extra virgin olive oils is of great interest in food industry. This article presents the first use of a voltammetric e-tongue for the detection of the adulteration of virgin olive oil. Adulterations of an extra virgin olive oil with different percentages of sunflower oil, soybean oil and corn oil were measured using modified carbon paste based sensors. The square wave voltammetric signals were processed using kernel method. Chemometric methods applied allows discrimination and classification of oils in agreement with botanical origins. Excellent correlations between voltammetric signals and polyphenolic content was obtained by PLS regression. PLS-DA and PLS regression demonstrated the feasibility of detecting adulterations of olive oil with percentages lower than 10% of sunflower, soybean and corn oils. These results indicate that e-tongue can be a useful tool for the detection of olive oil adulteration with seed oils. Ó 2014 Elsevier B.V. All rights reserved.
1. Introduction Extra virgin olive oil (EVOO) is obtained from the olive fruit named Olea europaea. EVOO is extracted by only mechanical procedure without application of refining process. The quality of olive oil ranges from the high quality EVOO to the low quality olive– pomace oil (Tsimidou, 2006). It is one of the primary ingredients of the Mediterranean diet (Psaltopoulou et al., 2004). Different factors such as cultivar, environment and cultural practices determine the quality and uniqueness of specific EVOOs (Tura et al., 2009). The agreeable taste and aroma with the health benefits of EVOOs are important reasons for consumers to consume this product (Amirante et al., 2012). Oil pureness is a very important aspect of the quality edible oil for special reasons of the sensory properties, perceived health values and the confidence of the health foods (Cozzolino, 2012). Because of the high value of EVOO, it can be adulterated with other oils of lower commercial value. The most common adulterants found in EVOO are refined olive oil, seed oils and nut oils (Dourtoglou et al., 2003). The detection of adulteration in EVOOs is a particular concern in the food industry. The adulteration of extra virgin olive oil with other cheaper oils can lead to significant profits for the unscrupulous vendor or raw material supplier. Therefore, continuous caution is required to control the adulteration of EVOOs and to protect the interests of the consumers. ⇑ Corresponding author. Tel.: +40 0236460328; fax: +40 0236461353. E-mail address: [email protected] (C. Apetrei). http://dx.doi.org/10.1016/j.compag.2014.08.002 0168-1699/Ó 2014 Elsevier B.V. All rights reserved.
In order to detect VOO adulteration, a number of chromatographic and spectroscopic methods, including fluorescence (Kunz et al., 2011), near-infrared (NIR) (Özdemir and Öztürk, 2007), Fourier transform infrared spectroscopy (FT-IR), FT-Raman (Heise et al., 2005), nuclear magnetic resonance (NMR) (Xu et al., 2014), mass spectrometry (MR) (Calvano et al., 2012), mid-infrared spectroscopy (MIR) (Gurdeniz and Ozen, 2009) and high-performance liquid chromatography (HPLC) (Lísa et al., 2009) were used. All these methods were applied with some success. However, there are also some shortcomings. Chromatographic, mass and NMR spectroscopy methods need expensive instruments. Fluorescence spectroscopy and chromatography methods require preprocessing of the samples. Furthermore, these methods are not suitable for on-line, in-line or real time analysis. The multivariate data analysis like linear discriminant analysis (LDA), principal component analysis (PCA), partial least square regression (PLS), machine learning (ML), and artificial neural networks (ANN) are applied to further analyze the oil spectroscopy data for oil adulteration detection (Xu et al., 2014; Gurdeniz and Ozen, 2009). Nowadays there is an increasing interest for simple and fast techniques called e-nose and e-tongue for various applications in food industry (Peris and Escuder-Gilabert, 2013). An e-tongue is an device, which consists of an array of chemical sensors and an appropriate pattern recognition method, for recognition (identification, classification, discrimination) of quantitative multicomponent analysis and artificial assessment of taste and flavour of various liquids. In the case of e-tongues, their capability to analyse
149
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
and discriminate a variety beverages such as mineral waters, milks, teas, wines or beers has already been established (Cetó et al., 2014; Yaroshenko et al., 2014; Gutiérrez-Capitán et al., 2013; GhasemiVarnamkhasti et al., 2012; Apetrei et al., 2012). However, few works have been focused to the analysis of olive oils using e-tongues (Cosio et al., 2007; Haddi et al., 2013; Apetrei and Apetrei, 2013). In previous works we have developed different strategies in order to use an voltammetric e-tongue to discriminate oils of different origins and qualities (Apetrei et al., 2007, 2010, 2012). In one method, the olive oil is binder used for preparation of carbon paste-based sensor. The characteristics of the voltammograms observed when the sensor is immersed in different electrolytic solutions reflect the electroactive properties of the oils (Apetrei et al., 2007, 2010). The aim of this work is to evaluate the feasibility of e-tongue (based on modified EO carbon paste-based sensors) to detect the adulteration qualitatively and precisely of EOs with different seed oils based on their tasting fingerprints. Based on our knowledge, it is the first time that a voltammetric e-tongue is employed for detection of adulteration of extra virgin olive oil. Various chemometric methods were performed for discrimination and classification of adulteration levels. 2. Experimental 2.1. Reagent and chemicals Potassium chloride (KCl) was obtained from Sigma–Aldrich. Hydrochloric acid (HCl) was purchased from S.C. Chemical Company S.A. (Romania). Water was purified on a Milli-Q SimplicityÒ Water Purification System (Millipore Corporation) with resistivity of 18.2 MX. Electrolyte solutions were prepared using ultrapure water. Carbon nanopowder, <50 nm particle size (TEM), P99% trace metals basis from Aldrich was used for fabrication of carbon paste electrodes modified with edible oils. 2.2. Samples Four kinds of edible vegetable oil were purchased from local supermarkets, including the extra virgin olive oil, the sunflower oil, the soybean oil and the corn (maize) oil. The oil samples were stored in the fridge at 4 °C until the day of analysis. The adulterant oil is considered the seed oil at different level of concentrations. For e-tongue measurements, mixtures of
seed and extra virgin olive oils, with concentration of seed oil at different levels (2%, 5%, 10%, 20% and 25%) were prepared (Table 1). The mixtures of oils were prepared using an Elmasonic S10H ultrasonic bath. Total phenolic content was determined spectrophotometrically following the Folin–Ciocalteu spectrophotometric method (Herchi et al., 2011). The results are presented in Table 2.
2.3. e-Tongue system For electronic tongue measurements carbon paste electrodes (CPE) modified with edible oils were prepared as previously described (Apetrei et al., 2007, 2010). Voltammetric measurements were carried out in an Biologic Science Instruments SP 150 potentiostat/galvanostat (EC-Lab Express software) using a conventional three-electrode cell. The modified EO carbon paste-based sensors were used as working electrodes. The reference electrode was an Ag/AgCl KCl 3M and the counter electrode was a platinum wire. The electrochemical experiments were performed at a controlled temperature of 25 °C. Three identical EO carbon paste-based sensors were prepared for each oil under study. Each replicate sensor was immersed in one electrolytic solution (0.1 mol L1 HCl or 0.1 mol L1 KCl) and the voltammetric response was registered. Voltammetric measurements were carried out by means of SWV (Square wave voltammetry). SWV were recorded using a frequency of 15 Hz, an amplitude of 0.10 V and a step high of 0.005 V. The SWV curves were registered in the potential range from 0.2 to +1.3 V. Seven measurement replicates were carried out with each sensor (measured after the conditioning step) in both electrolyte solutions.
2.4. Data analysis Several steps were carried out in order to obtain the input matrix for multivariate data analysis. SWV curves were pre-processed using the adaptation of a data reduction technique based on predefined response ‘‘bell shaped-windowing’’ curves called ‘‘kernels’’ (Apetrei et al., 2010). Using this method, ten parameters per each SWV were obtained. Because in this case, the variables are the edible oils from the sensors, the initial input matrix was transposed. In this way, the variables are the electrolyte solutions and the samples are the edible oils from the carbon paste-based sensors.
Table 1 Edible oil samples under study. Sample
Edible oil
Concentration/%
EVOO Sf S C EVOO + Sf EVOO + S EVOO + C
Olive oil Sunflower oil Soybean oil Corn oil Olive oil/sunflower oil Olive oil/soybean oil Olive oil/corn oil
100 100 100 100 98/2 98/2 98/2
95/5 95/5 95/5
90/10 90/10 90/10
80/20 80/20 80/20
75/25 75/25 75/25
Table 2 Total phenolic content of oil samples determined spectrophotometrically. Sample
Total polyphenols content (mg kg1) RSD = 4.5%.
EVOO
275.43
EVOO + Sf
EVOO + S
EVOO + C
98/2
95/5
90/10
80/20
75/25
98/2
95/5
90/10
80/20
75/25
98/2
95/5
90/10
80/20
75/25
271.92
261.63
249.80
220.75
208.23
270.75
260.54
250.18
221.24
207.41
272.43
260.74
250.17
219.94
208.95
150
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
For variable reduction and separation into classes, principal component analysis (PCA) was used. Partial least squares-discriminant analysis (PLS-DA) was used as deterministic classification technique (Gemperline, 2006). For sample groups establishing based on botanical origin analysis of variance (ANOVA) was used. PLS regression was used to establish the correlations between polyphenol contents or concentration of adulterant oil and electrochemical responses (Gemperline, 2006). Both the X-matrix (containing the e-tongue data values) and the Y-matrix (containing the physico-chemical parameter) were pre-processed using a normalisation routine. The optimum number of latent variables (LV) was selected by choosing the first local minimum in the residual Y variance plot. Validation of the predictive models was carried out by full cross-validation method (leave one-out). The multivariate data analysis was performed by using the software Matlab 5.3., Excel 2007 and The Unscrambler 9.1. 3. Results and discussion 3.1. Measurements with the e-tongue Measurements of the e-tongue were carried out using the EO carbon paste-based sensors. Preliminary studies and stabilization of the sensor signals were carried out by cyclic voltammetry (CV). CV was used to study qualitative information about
electrochemical processes, the reversibility of the reactions and stability of the sensor response. The CV experiments were carried out from 0.2 V to +1.3 V with a scan rate of 0.05 V s1. Under these conditions, EVOO carbon paste-based sensors showed a variety of anodic and cathodic peaks. The CV of EVOO carbon paste-based sensors immersed in HCl 0.1 mol L1 showed two anodic peaks at 0.54 V, and 1.05 V and one cathodic peak at 0.04 V (Fig. 1a). However, the resolution of the peaks needed to be improved. A better resolution of the peaks was obtained in forward scan when the SWV technique was used (Fig. 1b). The number and the potential of the peaks remained unchanged but an important increase of the peak currents were observed. Therefore, the use of SWV increases the sensitivity of the method. Furthermore, the SWV curves facilitate data treatment because the curves are pairs current (i)–potential (E). EVOO has in its composition mainly triacylglycerols, polyphenols, tocopherols, pigments, aromatic components and small amount of free fatty acids (Quiles et al., 2006). Polyphenols and tocopherols have antioxidant properties and therefore these compounds can be detected by means of electroanalytic methods (Kim and Kusuda, 1994). In agreement with literature, peaks at 0.54 and at 0.04 V are related to the polyphenolic compounds present in of EVOO (Shiragami et al., 1994). The redox peak found at 1.05 V is related to the presence of tocopherols present in EVOO (Kim and Kusuda, 1994). The electrochemical reaction at the sensor surface is presented in Fig. 2. The dependence of Ep and Ep/2, expressed using Nicholson and Shain equation (Southampton, 1990), can be used to calculate the number of electrons (n) participating in the rate determining stage for an irreversible charge transfer (Eq. (1)):
Ep Ep=2 ¼
Fig. 1. Voltammetric responses of EO carbon paste-based sensors immersed in aqueous HCl 0.1 mol L1. EVOO carbon paste-based sensors: (a) CV; (b) SWV. Sf carbon paste-based sensors: (c) CV; (d) SWV. S carbon paste-based sensors: (e) CV; (f) SWV. C carbon paste-based sensors: (g) CV; (h) SWV.
1:857 R T 47:7 ¼ ðmVÞ anF an
ð1Þ
where Ep is the potential of the peak, Ep/2 is the potential of half peak height, R is the universal gas constant, T is the absolute temperature, F is Faraday constant and a is the charge transfer coefficient. The n value calculated assuming a of 0.5 is 2.01. Therefore, the electrochemical process of tocopherols at the sensor surface involve 2 electrons (Fig. 2). Similar results, number of electrons involved close to 2 were obtained for all EO modified sensors indicating the involvement of the same compound in the redox process. The CV and SWV of Sf carbon paste-based sensors showed one anodic peak associated to tocopherols at ca. 1.06 V (Fig. 1c and d). Sunflower oil contains predominantly linoleic acid in triglyceride form, lecithin, tocopherols, carotenoids and waxes. Sunflower oil is light in taste and appearance. Therefore, the main electroactive compounds from sunflower oil are the tocopherols (Gunstone, 2011). The voltammograms obtained for corn oil and soybean were similar to those obtained for sunflower oil. However, shifts in the peak potentials were observed in Fig. 1e–h (corn oil, Ea = 1.01 V; soybean oil, Ea = 1.00 V). Therefore, the electroactive substances present in corn oil and soybean oil are also the tocopherols. The differences observed in the voltammetric curves are also related to different physico-chemical properties of the oils, i.e. viscosity. Since oils are mixtures of triglycerides viscosity depends on the nature of fatty acids from triglyceride molecules and their properties such as chain length and saturation/unsaturation. In conclusion, the intensity of the observed peaks depend on concentration of tocopherols, their relative concentrations in edible oils and triglyceride profile (Gunstone, 2011).
151
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154 Table 3 ANOVA significance results.
* ** ***
Contrast
Significance
EVOO vs. Sf EVOO vs. S EVOO vs. C Sf vs. C Sf vs. S S vs. C
*** *** *** ** ** *
p < 0.05 (Significant). p < 0.01 (Highly significant). p < 0.0001 (Extremely significant).
Fig. 2. The redox processes of tocopherol at the sensor surface. Table 4 Quantitative data of PLS-DA.
Fig. 3. PCA score plot corresponding to the SWVs data treatment. O – extra virgin olive oil; Sf – sunflower oil; S – soybean oil; C – corn oil.
The voltammetric curves shown in Fig. 1 indicate that the nature and concentration of electroactive compounds found in different edible oils bearing a diversity of electrochemical signals. The voltammetric responses can be used to discriminate edible oils. From the analyse of the voltammetric curves it can be observed that EVOO related curves can be easily distinguished from other oils. Only in the case of EVOO peaks related to polyphenols can be observed.
Type
RMSEC
RMSEP
Extra virgin olive oil Sunflower oil Soybean oil Corn oil
0.066 0.126 0.092 0.162
0.082 0.134 0.103 0.171
graph. Moreover, the method is capable to distinguish among seed oils. Using the electronic tongue, the first principal component is the main responsible of the discrimination of the EVOO from other vegetable oils, whereas second principal component is the main responsible of the discrimination of seeds oils. Electrochemical data were tested for statistical significance using one-way ANOVA routines running under Excel. The factor was botanical origin of edible oils. Values of p < 0.05 were considered statistically significant. In Table 3 are presented the ANOVA significance results. ANOVA was based on honestly significant difference test (Tukey test). This test is based on pairwise comparison among means as presented in Eq. (2):
Mi Mj ts ¼ qffiffiffiffiffiffiffi MSE nh
ð2Þ
Mi – Mj = difference between pair means. MSE = mean square error. nh = the harmonized mean.
3.2. Classification analysis of edible oils In this work different methods to analyse the voltammetric signals related to oil samples were applied. First, PCA, PLS-DA and ANOVA were applied to discriminate and classify the edible oils in agreement with their origin. After that, PCA, PLS-DA and ANOVA were applied to discriminate and classify the VOO adulterated with one adulterant oil. Further, all oil samples, were analyzed by means of PCA and PLS-DA in order to discriminate and classify the original and adulterated VOOs with seed oils in different proportions. PLS regression models were developed and used for determination of concentration limit below that the e-tongue is not able to distinguish the adulterated oil from authentic oil. The discrimination capacity of the e-tongue was evaluated by means of Principal Component Analysis (PCA) of the voltammetric signals. Fig. 3 shows the scores plot of the two first principal components calculated from the parameters extracted from the SWV curves. The first two principal components explain the 73% of the information (PC1 = 46%; PC2 = 27%). The PCA score plot shown in Fig. 3 indicates that the system based on CPE sensors immersed in different electrolytes is able to discriminate among vegetable oils of different origins. Olive oil appeared in the left region of the plot, whereas, the seed oils whose do not contain polyphenols, appear in the opposite side of the
As presented in Table 3 the population means are significantly different. The samples could be classified in groups in agreement with botanical origin of the oil. ANOVA showed significant differences between these groups of oils based on e-tongue results. A supplementary confirmation of the groups shown in PCA was carried out by PLS-DA. Quantitative data of PLS-DA are presented in Table 4. The optimal number of latent variables (5 LVs) was determined by the lowest value of predicted residual error sum of squares. Root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP) have low values indicating the high quality of the model. For all EO samples, when the constructed PLS-DA model was applied, 100% of samples are correctly classified both in calibration and in validation. The validation was performed using leave-one-out fully cross validation. The sensitivity and specificity of the model is 100%. 3.3. Discrimination and classification of adulterated EVOOs In order to determine the capability of the e-tongue to detect adulteration of EVOO, carbon paste-based sensors modified with adulterated oils were constructed. Relative concentrations of oils used in sensors fabrication are presented in Table 1. The SWVs of sensors modified with Sf, EVOO and EVOO/Sf immersed in HCl
152
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
Fig. 4. The SWVs of sensors modified with Sf (——), EVOO (____) and EVOO/Sf (95/5 _._._ ; 90/10 _.._.._; 80/20 ........; 75/25 __ __ __) immersed in HCl 0.1 mol L1.
0.1 mol L1 are presented in Fig. 4. As can be observed, the SWVs of sensors modified with EVOO/Sf present an intermediary behaviour between pure EVOO and Sf oil. The peak corresponding to polyphenols decrease when the concentration of Sf increase. Contrary, the peak corresponding to tocopherol increase when the concentration of Sf increase. The voltammetric responses are related to changing in concentration of electroactive compounds, polyphenols and tocopherols. When Sf is added in EVOO the concentration of polyphenols decrease and the concentration of tocopherol increase. In agreement with literature, Sf oils have higher concentration in tocopherols than EVOO (Gunstone, 2011). Analogous results are obtained in the cases of C and S oils. In the SWVs, the peaks related to polyphenols decrease when the concentration of adulterant oil increase. The peak related to tocopherol remains almost constant in the case of corn oil and slightly decreases in the case of soybean oil. The establishment of correlations between the results obtained by means of e-tongue and the total polyphenolic content is of vital importance for detection of EVOO adulteration. It is well-known that polyphenols are the main contributors to virgin olive oil taste. Furthermore, the main difference between the SWV curves of EVOO modified sensors and EO modified sensors consists in the peak pair related to polyphenols that appears only in the case of EVOO. For these reasons, PLS regression was performed to model the relationships between the voltammetric signals and the total polyphenolic content obtained by means of spectrophotometric method (Herchi et al., 2011). As observed in Fig. 5, a very good correlation was found, with a coefficient of determination (R2) of 0.9888 in calibration and 0.9852 in validation, using 4 latent variables. RMSEC was 2.6773 and RMSEP was 3.0723, demonstrating the quality of the model.
This result indicated that the polyphenolic content is strongly related to the voltammetric responses provided by the electronic tongue. Therefore, the voltammetric e-tongue could detect the adulteration of EVOOs based on modification of polyphenolic content. The results described above suggest that voltammetric e-tongue might be able to detect the presence of other vegetable oils in EVOO. To assess this possibility, the SWV curves of 15 binary mixtures (composition range as presented in Table 1) and 4 pure edible oils were measured in replicates (3 sensors for each oil and binary mixture 2 electrolyte solutions 7 replicates). In the first step of the data treatment, PCA was applied to the processed voltammetric data, and it was found that first three PCs accounting for 82% of the total explained variance successfully separate the mixtures according to composition (Fig. 6). As observed in the Fig. 6, the pure oils are distributed in a similar way such in Fig. 3. The EVOO cluster appears in the opposite side of the graph relative to seed oils. The adulterated oils clusters are situated among pure oils clusters, near to cluster corresponding to EVOO. However, the clusters corresponding to adulterated EVOO with small quantities of seed oil are not totally separated from pure EVOO cluster. PLS-DA and PLS models are necessary in order to quantify the level of adulteration in EVOO. It can be observed that e-tongue is able to distinguish among oils adulterated with different seed oils. The PCA result strongly suggests that these SWV curves contain electroanalytical information that will allow correlation of voltammetric curves features with the concentration of the adulterant oil. For that reason, subsequent supervised analyses using PLS and PLS-DA were used to develop models to quantify the level of adulteration in EVOO. The data were split into a training (or calibration) set and a test set (validation). The training set consisted of 5 replicate SWVs of all oil samples (71.43%), while the test set consisted of 2 replicate SWVs (28.57%). The models were developed for EVOO, pure seed oil, and adulterated samples with corresponding seed oil. Therefore, three models were developed corresponding to EVOO adulterated with sunflower oil, corn oil and soybean oil, respectively. The quantitative PLS-DA models are present in Table 5 in terms of correct and misclassified samples percentages. It was found that the best models (i.e. lowest prediction error in the test set) occurred when six latent variables were used, in all cases. In the case of model of EVOO adulterated with sunflower oil, the sample replicates of 2% Sf/EVOO can be distinguish from pure EVOO in a percentage of 42.86%. The sample replicates of 5% Sf/EVOO can be distinguish from pure EVOO in a percentage of 85.71%. The sample replicates of adulterated oil containing 10% of sunflower oil were correctly classified. In the case of EVOO adulterated with soybean oil the sample replicates are 100% correct classified when the concentration of
Fig. 5. Plots of predicted total polyphenolic content obtained with e-tongue vs. the values of total polyphenolic content obtained by Folin–Ciocalteau method.
153
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
Fig. 7. PLS model for the relationship between actual value and e-tongue predicted value of sunflower oil in EVOO in calibration.
Fig. 6. PCA scores plot corresponding to the SWVs of pure and adulterated oils. O – olive oil; S – soybean oil; Sf – sunflower oil; C – corn oil; Sf2, Sf5, Sf10, Sf20, Sf25 – mixtures of sunflower oil and extra virgin olive oil with concentration of sunflower oil at different levels (2%, 5%, 10%, 20% and 25%); S2, S5, S10, S20, S25 – mixtures of soybean oil and extra virgin olive oil with concentration of soybean oil at different levels (2%, 5%, 10%, 20% and 25%); C2, C5, C10, C20, C25 – mixtures of corn oil and extra virgin olive oil with concentration of corn oil at different levels (2%, 5%, 10%, 20% and 25%).
Table 5 Quantitative data of PLS-DA. Model
Sample
Correct classified sample replicates/%
Misclassified samples replicates/%
Olive oil/sunflower oil
EVOO Sf EVOO + Sf EVOO + Sf EVOO + Sf EVOO + Sf EVOO + Sf
100 100 42.86 85.71 100 100 100
0 0 57.14 14.29 0 0 0
98/2 95/5 90/10 80/20 75/25
100 100 14.29 57.14 100 100 100
0 0 85.71 42.86 0 0 0
98/2 95/5 90/10 80/20 75/25
100 100 28.57 71.43 100 100 100
0 0 71.43 28.57 0 0 0
Olive oil/soybean oil
Olive oil/corn oil
EVOO S EVOO + S EVOO + S EVOO + S EVOO + S EVOO + S EVOO C EVOO + C EVOO + C EVOO + C EVOO + C EVOO + C
98/2 95/5 90/10 80/20 75/25
Fig. 8. Variation of anodic peak potential related to polyphenols in time for EVOOCPE.
3.4. Quantification of Sf, S, and CO in adulterated EVOO samples
soybean oil is 20%, whereas EVOO adulterated with corn oil are 100% correct classified when the concentration of corn oil is 10%. Therefore, the e-tongue is able to detect presence of seed oils in EVOO at levels lower than 10% adulterant oil, level important for practical applications.
Quantification of Sf, S, and CO was carried out with the aid of multivariate calibration. Calibration models developed by means of partial least square (PLS) were used to evaluate the goodness of fit for the relationship between real value (x-axis) and e-tongue predicted value (y-axis) of Sf, S and C oils in EVOO. PLS regression model was calibrated with the training set and cross-validated with the test set. It was found that the best models (i.e. lowest prediction error in the test set) occurred when five latent variables were used. Table 6 compiled the performance of PLS for quantification of Sf, S, and C in EVOO. PLS gives the reasonable R2 either in calibration or in validation (Rc2 = 0.9996; Rp2 = 0.9987) and offers small errors in calibration and validation. Using this model, four LVs are needed to obtained RMSEC value of 0.1670 and RMSEP value of 0.3105. The scatter plot for the relationship between real value and e-tongue predicted value of Sf in EVOO was revealed in Fig. 7. Based on these results, it can be assumed that e-tongue facilitate the detection and quantification of Sf, S and C oils adulterants in EVOO. Furthermore, the models indicate that the composition of the mixture can be successfully predicted in the range of commercial interest, 2–25%. The results are comparable with those reported for other systems
Table 6 The performances of multivariate calibration for analyses of Sf, S and C oils in EVOO. Model
Extra virgin olive oil/sunflower oil Extra virgin olive oil/soybean oil Extra virgin olive oil/corn oil
R2
Equation Calibration
Prediction
Calibration
Prediction
y = 0.9996x + 0.0044 y = 0.9987x + 0.0155 y = 0.9988x + 0.0149
y = 0.9926x + 0.0659 y = 0.9861x + 0.1102 y = 0.9864x + 0.1149
0.9996 0.9987 0.9988
0.9987 0.9953 0.9956
RMSEC
RMSEP
0.1670 0.3107 0.3048
0.3105 0.5962 0.5793
154
I.M. Apetrei, C. Apetrei / Computers and Electronics in Agriculture 108 (2014) 148–154
employed in detection of EVOO adulteration (Gurdeniz and Ozen, 2009; Xu et al., 2014). 3.5. Stability of the sensors CVs of EVOO-CPE immersed in 0.1 M HCl aqueous solution were registered ten consecutive days. CPEs were withdrawing from the solution after each experiment and smoothing the surface with a clean paper filter. Among measurements EVOO-CPE was stored in the dark at 4 °C in dry atmosphere. The variation of anodic peak potential related to polyphenols in time was shown a coefficient of variation of 5% between first and tenth day (Fig. 8). Therefore, the EVOO-CPE presents a good stability since electroactive compounds are included in carbonaceous matrix and the access of oxygen is limited. 4. Conclusions A novel method based on a voltammetric e-tongue to evaluate the adulteration of extra virgin olive oils has been established. The responses obtained using sensors modified with edible oils show peaks related to tocopherols and polyphenols. The voltammetric curves can be used as input variables in chemometrics. It has been shown that e-tongue can successfully discriminate and classify among extra virgin olive, sunflower, corn and soybean oils. PLS-DA models showed that the applied e-tongue is able to classify correctly the adulterated oils when the concentration level of adulterant oil was between 5% and 10%. Moreover, the composition of seed oils and extra virgin olive oil mixtures could be accurately predicted using PLS regression. In conclusion, e-tongue consisting of voltammetric sensors together with appropriate chemometrics, presents itself as a powerful tool for the detection of adulteration of extra virgin olive oil. Application of e-tongue coupled with e-nose to detection of olive oil adulteration could enhance the capability of the system even at lower levels of adulteration. Acknowledgments This work was supported by a Grant of the Romanian National Authority for Scientific Research, CNCS – UEFISCDI, Project number PN-II-ID-PCE-2011-3-0255. References Amirante, P., Clodoveo, M.L., Tamborrino, A., Leone, A., Dugo, G., 2012. Oxygen concentration control during olive oil extraction process: a new system to emphasize the organoleptic and healthy properties of virgin olive oil. Acta Hort. 949, 473–480. Apetrei, I.M., Rodriguez-Mendez, M.L., Apetrei, C., Nevares, I., del Alamo, M., de Saja, J.A., 2012. Monitoring of evolution during red wine aging in oak barrels and alternative method by means of an electronic panel test. Food Res. Int. 45, 244– 249. Apetrei, C., 2012. Novel method based on polypyrrole-modified sensors and emulsions for the evaluation of bitterness in extra virgin olive oils. Food Res. Int. 48, 673–680. Apetrei, C., Apetrei, I.M., Villanueva, S., de Saja, J.A., Gutierrez-Rosales, F., RodriguezMendez, M.L., 2010. Combination of an e-nose, an e-tongue and an e-eye for the characterisation of olive oils with different degree of bitterness. Anal. Chim. Acta 663, 91–97. Apetrei, C., Gutierez, F., Rodríguez-Méndez, M.L., de Saja, J.A., 2007. Novel method based on carbon paste electrodes for the evaluation of bitterness in extra virgin olive oils. Sens. Actuat. B – Chem. 121, 567–575.
Apetrei, I.M., Apetrei, C., 2013. Voltammetric e-tongue for the quantification of total polyphenol content in olive oils. Food Res. Int. 54, 2075–2082. Calvano, C.D., De Ceglie, C., D’Accolti, L., Zambonin, C.G., 2012. MALDI-TOF mass spectrometry detection of extra-virgin olive oil adulteration with hazelnut oil by analysis of phospholipids using an ionic liquid as matrix and extraction solvent. Food Chem. 134, 1192–1198. Cetó, X., Apetrei, C., Del Valle, M., Rodríguez-Méndez, M.L., 2014. Evaluation of red wines antioxidant capacity by means of a voltammetric e-tongue with an optimized sensor array. Electrochim. Acta 120, 180–186. Cosio, M.S., Ballabio, D., Benedetti, S., Gigliotti, C., 2007. Evaluation of different storage conditions of extra virgin olive oils with an innovative recognition tool built by means of electronic nose and electronic tongue. Food Chem. 101, 485– 491. Cozzolino, D., 2012. Recent trends on the use of infrared spectroscopy to trace and authenticate natural and agricultural food products. Appl. Spectrosc. Rev. 47, 518–530. Dourtoglou, V.G., Dourtoglou, Th., Antonopoulos, A., Stefanou, E., Lalas, S., Poulos, C., 2003. Detection of olive oil adulteration using principal component analysis applied on total and Regio FA content. J. Am. Oil Chem. Soc. 80, 203–208. Gemperline, P., 2006. Practical Guide to Chemometrics, Taylor & Francis Group LLC. Ghasemi-Varnamkhasti, M., Rodríguez-Méndez, M.L., Mohtasebi, S.S., Apetrei, C., Lozano, J., Ahmadi, H., Razavi, S.H., de Saja, J.A., 2012. Monitoring the aging of beers using a bioelectronic tongue. Food Contr. 25, 216–224. Gunstone, F.D. (Ed.), 2011. Vegetable Oils in Food Technology: Composition, Properties and Uses. Second ed., Blackwell Publishing Ltd. Gurdeniz, G., Ozen, B., 2009. Detection of adulteration of extra-virgin olive oil by chemometric analysis of mid-infrared spectral data. Food Chem. 116, 519–525. Gutiérrez-Capitán, M., Santiago, J.-L., Vila-Planas, J., Llobera, A., Boso, S., Gago, P., Martínez, M.-C., Jiménez-Jorquera, C., 2013. Classification and characterization of different white grape juices by using a hybrid electronic tongue. J. Agric. Food Chem. 61, 9325–9332. Haddi, Z., Alami, H., El Bari, N., Tounsi, M., Barhoumi, H., Maaref, A., JaffrezicRenault, N., Bouchikhi, B., 2013. Electronic nose and tongue combination for improved classification of Moroccan virgin olive oil profiles. Food Res. Int. 54, 1488–1498. Heise, H.M., Damm, U., Lampen, P., Davies, A.N., Mcintyre, P.S., 2005. Spectral variable selection for partial least squares calibration applied to authentication and quantification of extra virgin olive oils using Fourier transform Raman spectroscopy. Appl. Spectrosc. 59, 1286–1294. Herchi, W., Sakouhi, F., Arráez-Román, D., Segura-Carretero, A., Boukhchina, S., Kallel, H., Fernández-Gutierrez, A., 2011. Changes in the content of phenolic compounds in flaxseed oil during development. J. Am. Oil Chem. Soc. 88, 1135– 1142. Kim, O.S., Kusuda, K., 1994. Electrochemical behaviour of cu-tocopherol in a thin film of Nujol, a model of adipose tissue. Bioelectrochem. Bioenerg. 33, 61–65. Kunz, M.R., Ottaway, J., Kalivas, J.H., Georgiou, C.A., Mousdis, G.A., 2011. Updating a synchronous fluorescence spectroscopic virgin olive oil adulteration calibration to a new geographical region. J. Agric. Food Chem. 59, 1051–1057. Lísa, M., Holcˇapek, M., Bohácˇ, M., 2009. Statistical evaluation of triacylglycerol composition in plant oils based on high-performance liquid chromatography – atmospheric pressure chemical ionization mass spectrometry data. J. Agric. Food Chem. 57, 6888–6898. Özdemir, D., Öztürk, B., 2007. Near infrared spectroscopic determination of olive oil adulteration with sunflower and corn oil. J. Food Drug Anal. 15, 40–47. Peris, M., Escuder-Gilabert, L., 2013. On-line monitoring of food fermentation processes using electronic noses and electronic tongues: a review. Anal. Chim. Acta 804, 29–36. Psaltopoulou, T., Naska, A., Orfanos, P., Trichopoulos, D., Mountokalakis, T., Trichopoulou, A., 2004. Olive oil, the Mediterranean diet, and arterial blood pressure: the Greek European Prospective Investigation into Cancer and Nutrition (EPIC) study. Am. J. Clin. Nutr. 80, 1012–1018. Quiles, J.L., Ramírez-Tortosa, M.C., Yaqoob, P., 2006. Olive Oil and Health, CABI. Shiragami, S., Kim, O.S., Kusuda, K., 1994. Direct survey of redox active compounds in sesame oil by cyclic voltammetry – studies on sesaminol, sesamol and hydroxyhydroquinone in nujol. J. Electroanal. Chem. 379, 315–319. Southampton Electrochemgroup, 1990. Instrumental Methods in Electrochemistry. Ellis Horwood Ltd, New York, pp. 178–227, vol. 380. Tsimidou, M.Z., 2006. Olive oil quality. In: Boskou, D., (Ed.), Olive oil: Chemistry and Technology, second ed. Champaign, Illinois: AOCS Publishing, pp. 93–112. Tura, D., Failla, O., Bassi, D., Pedo, S., Serraiocco, A., 2009. Environmental and seasonal influence on virgin olive (Olea europaea L.) oil volatiles in northern Italy. Sci. Hort. 122, 385–392. Xu, Z., Morris, R.H., Bencsik, M., Newton, M.I., 2014. Detection of virgin olive oil adulteration using low field unilateral NMR. Sensors 14, 2028–2035. Yaroshenko, I., Kirsanov, D., Kartsova, L., Bhattacharyya, N., Sarkar, S., Legin, A., 2014. On the application of simple matrix methods for electronic tongue data processing: case study with black tea samples. Sens. Actuat. B – Chem. 191, 67– 74.