Impaginato 329 Adv. Hort. Sci., 2017 31(4): 329-337 DOI: 10.13128/ahs-22155 EVOO or not EVOO? A new precise and simple analytical tool to discriminate extra virgin olive oils C. Taiti 1 (*), E. Marone 2 1 Dipartimento di Scienze delle Produzioni Agroalimentari e dell’Ambiente, Università degli Studi di Firenze, Viale delle Idee, 30, 50019 Sesto Fiorentino (FI), Italy. 2 Bioscienze e Tecnologie Agro-Alimentari e Ambientali, Università di Teramo, Via R. Balzarini, 1, 64100 Teramo, Italy. Key words: flavors and off-flavors, Panel Test, partial least square-discriminant analysis (PLS-DA), PTR-ToF-MS, volatile organic compounds (VOCs). Abstract: International Olive Oil Council (IOOC) states chemical and organolep- tic parameters to classify the commercial grade of olive oil. Finding tools or analytical procedures able to support the organoleptic evaluation would be helpful to streamline and facilitate the commercial classification. The aim of the present study was to evaluate a new tool and validate a procedure that allows a fast and non-invasive volatile compounds detection system, able to assign each sample to its right trade category. Moreover, we tried to test the capabili- ty of PTR-ToF-MS in grading olive oils according to their fruity intensity levels. A total of 273 olive oil samples collected from Argentina (21), Chile (10), Italy (191), Morocco (17), Tunisia (4) and EU (30) were analyzed and classified through: (1) Panel Test and (2) PTR-ToF-MS analysis. On the whole PTR-ToF-MS data EVOO and Not EVOO as resulted by Panel Test were clustered by PCA in two main groups and correctly classified by PLS-DA model, confirming the high confidence level (95%) in utilizing analytical spectral data for helping Panel Test and able to easy monitoring the quality formation in the oils, by a fast and cheap control from harvest until the store. The eight protonated masses detected as VIP by the model may be linked to negative olfactory notes. Finally, PCA applied on the volatile profile of 122 classified EVOO highlighted a shift of the samples distribution following the trend of the fruity intensity as assessed by the panelists. In conclusion, this trial confirmed the availability of a new, precise and simple analytical tool as the PTR-ToF-MS, which coupled with an appropriate multivariate data analysis, allows to classify EVOO according to their trade category and fruity intensity. 1. Introduction The virgin olive oil is the only vegetable oil consumed without any refinement and characterized by a peculiar synthesis between taste and aroma. The importance of extra virgin olive oil (EVOO) is due to its high content of oleic acid and phenolic compounds, which act as natural (*) Corresponding author: cosimo.taiti@unifi.it Citation: TAITI C., MARONE E., 2017 - EVOO or not EVOO? A new precise and simple analytical tool to dis- criminate virgin olive oils. - Adv. Hort. Sci., 31(4): 329-337 Copyright: © 2017 Taiti C., Marone E. This is an open access, peer reviewed article published by Firenze University Press (http://www.fupress.net/index.php/ahs/) and distribuited under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Competing Interests: The authors declare no competing interests. Received for publication 31 October 2017 Accepted for publication 28 November 2017 AHS Advances in Horticultural Science Invited paper Adv. Hort. Sci., 2017 31(4): 329-337 330 antioxidants (Bendini et al., 2007). Its composition makes it not only a food and dressing, but also a product able to protect the human organism from some dysfunctions and pathologies (Marone and Fiorino, 2012). As reported by Aued-Pimentel et al. (2013), EVOO has unique characteristics compared to other vegetable oils, such as exceptional sensory and nutritional attributes, therefore worldwide the olive oils are the most valuable ones with a price (normally 3-5 times) higher than other edible oils (Zou et al., 2009). As a consequence, in the last years, some adulterations of EVOO with olive oils of lower quality, or with oils of different botanical origin (Catharino et al., 2005; Vlachos et al., 2006) have been found. As defined by the International Olive Oil Council (IOOC), olive oil is split in trade categories of different quality and commercial value. Because the high commercial value, and the relatively low availability against a high consumption, some traders and bottlers are prompted to sell as EVOO inferior olive oils that does not reflect the parameters established by the IOOC. According with the IOOC rules, the trade class attri- bution depends not only on chemically defined para- meters (i.e. free acidity and peroxides index) but also on a sensory evaluation (SE) that assesses off-flavors and fruity presence and intensity. Therefore the EVOO is the only traditional food that must be tested through a Panel Test. Taste and aroma are deter- mined by the presence and the amount of peculiar volatile organic compounds (VOCs), giving to the product unique appealing proprieties. On average the olive oil contains more than 100 volatile com- pounds belonging to different chemical categories (Guadarrama et al., 2000). VOCs emission by the olive fruit and/or olive oil is mostly related to oxida- tive reactions (i.e. due to injuries during the fruits crushing and malaxation processes). VOCs develop according to distinctive biosynthetic pathways and, among these, the ‘‘LipOXygenase (LOX) cascade’’ determines the enzymatic splitting of polyunsaturat- ed fatty acids (linoleic and linolenic) with the ‘‘con- trolled’’ production of aldehydes, ketones, alcohols, carboxylic acids, esters and other VOCs (Angerosa et al., 2004; Kalua et al., 2007). The importance of the SE for the olive oils, is due both to its ability to identify the positive attributes and also to evaluate the defects (Peri and Rastelli, 1994). Indeed, the volatile compounds can be used: (a) to discriminate EVOO and virgin olive oil (VOO); and (b) as quality parameters, being the VOCs responsible especially for the green notes and fruity of high-quality EVOO oils (Gomez-Rico et al., 2006). While the chemical parameters are easily evaluat- ed through chemical analyses, the flavor and off-fla- vors are assessed with more subjectivity through the sensory analyses. The SE by the Panel Test is based on strict and laborious rules, and needs trained peo- ples; therefore, as currently planned, the Panel Test is time consuming and very expensive. Thus, while the chemical analyses guarantees objectivity, repeatability and speediness, the sensory analysis does not allow this result. Indeed, as reported by Marone et al. (2017) the SE presents some disadvan- tages such as: (1) subjectivity of the analysis which could influence the overall evaluation; (2) the need of a large number of trained panelist (8-12) to allow the statistical validation of the results; (3) a limited number of samples evaluable by each panelist a day. Moreover, the results are difficult to generalize, because a lack of a common standard shared in the world, neither easy exploitable in any situations, nor to apply at any step of chain of olive oil making before sale (i.e. processing, storage). Consequently, there is no doubt that the detection of each type of olive oil manipulation needs to be addressed to ensure a correct trade classification, quality and con- sumer price. Currently, the most common used analytical tech- niques to detect VOCs emitted by olive oil are both chromatographic and spectrophotometric methods, as the dynamic headspace gas chromatograph (DHS- GC) (Procida et al., 2016), electronic nose and elec- tronic tongue (Aparicio et al., 2000; Cosio et al., 2007) and the Proton Transfer Reaction-Time of Flight-Mass Spectrometer (PTR-ToF-MS) (Aprea et al., 2006; Marone et al., 2017). The PTR-ToF-MS shows a high resolution coupled to a rapid screening power of samples, it is easy to handle and does not need any sample manipulation (Blake et al., 2009; Taiti et al., 2017). Moreover, this tool is applicable to any step of the olive oil produc- tion, from the processing to the market, including the product storage (Marone et al., 2017). Furthermore, as the PTR-ToF-MS could work at temperature near those of the tasting, it should give as output a bulk of VOCs at least similar to those perceived by panelists or consumers. A first attempt to directly link spectral data from PTR-ToF-MS as protonated masses to the olfactory sensations perceived by the panelists, to distinguish EVOO from Not EVOO, and consequently correctly classify the virgin olive oils in their trade category, was recently carried out by Marone et al. (2017). In this cited work, although employing a low number of samples, it was possible to build up, in a Taiti and Marone - EVOO or not EVOO? 331 statistically meaningful way, a color codified card highlighting some specific VOCs that seem to charac- terize the off flavors as perceived by the panelist. Starting from this result, the aim of the current work was to develop and to test a fast analytical method that combines efficiency, accuracy and reliability for a rapid screening and quantification of volatile com- pounds in olive oil samples. This analysis method should be helpful to: (1) detect the main defective odors and distinguish the olive oils trade categories; (2) understand if there is an accurate and precise cor- relation between the judgment provided by the Panel Test and the analysis of the volatile component by PTR-ToF-MS; (3) evaluate different quality and types of EVOO using the positive attributes (i.e. fruity and green notes). 2. Materials and Methods Oil sampling Analyses were carried out during 3 years of sur- veys (from 2015 to 2017) on the whole spectra of 273 olive oil samples, produced from 2012 to 2017 (Table 1). The olive oils came from three different continent (Africa, South America, and Europe); most samples were obtained from producers or supermar- kets, both blend or monocultivar stocks; in this last case, the most of the olive oil samples were obtained at the olive mill. To enhance and enlarge the samples set variability, EVOO from supermarkets labeled as origin were acquired together with “aged” samples (certainly processed two or more years before to be analyzed). For each sample, two filled dark bottles of 250 ml were collected and quickly sent to the storage refrigerated room (17°C) until the organoleptic and VOCs analyses were carried out. Finally, for some samples, the VOCs and SE analyses were repeated during the three years of analysis. Panel test After the spectrometric determinations all sam- ples were submitted to a Panel Test. All panels were organized according to the official E.U. olive oil sen- sory analysis Regulation (n. 2568/91 and its succes- sive modifications) (Table 1). Each taster on the panel shall enter the intensity of the negative and positive attributes on the 10-cm scale in the profile sheet. The oil is graded by the Panel Leader in line with the median of the defects and the median for the fruity attribute. According to the reference ranges, an olive oil is graded as extra virgin if the median of the defects is 0 and the median of the fruity attribute is above 0. In the present work, all the samples that did not result EVOO were classified as Not EVOO, with- out any further distinction. Volatile compounds detection Measurements were performed with a chemical ionization mass spectrometer (PTR-MS) equipped with a Time-of-Flight (ToF) analyzer (PTR-ToF 8000 model, Ionicon Analytik, Innsbruck, Austria) in its standard mode and using H3O+ as ions for the chemi- cal ionization. PTR-ToF-MS has some advantages compared to the other traditional electron ionization such as: reduced fragmentation which eases com- pound identification and guarantees high sensitivity with a very high time resolution and no need of sam- ple treatments (Taiti et al., 2017). Previously Blake et al. (2009) provided a complete and detailed descrip- tion of the PTR-MS technology. All the instrumental parameters used during the measurement were set as follow: a constant drift voltage of 600 V and a pressure of 2.20 ± 0.02 mbar were maintained in the reaction chamber and the instrument operated at a standard E/N value (electric field strength/gas num- ber density) of 138 Td (1 Td = 10-17 cm2 V-1 s-1). Each sample was prepared on the basis of the following protocol: 15 g of oil (T 25°C) were introduced in apposite glass jar (750 ml), afterwards were fluxed with clean air (Zero air generator, Peak scientific) for 120 seconds and subsequently were hermetically sealed and incubated for 60 seconds at 25°C inside an incubator. Each jar was provided with inlet and outlet Teflon pipes, which connect the glass jar to the PTR-ToF-MS system and to the zero-air generator, respectively. Two replicates of each sample were analyzed and the order of samples was randomized. Besides, at the beginning of the experiment and always after three oils sample an identical empty jar was analyzed for background subtraction. Headspace concentrations of each oil sample were subsequently averaged over the two replicates and used for fur- ther statistical analysis. The range of mass spectra was recorded in the range of 20-210 m/z at 1 spec- trum per 1 second, and the mass calibration was based on m/z = 21.022 (H3O+), m/z = 59.049 (C3H7O+) and m/z = 137.132 (C10H17 +); the calibration was made before starting each files and, subsequently, all files were recalibrated off-line. Data were recorded with the TofDaq software (Tofwerk AG, Switzerland) and all spectra were acquired and analyzed using a procedure previously reported by Taiti et al., 2017. Data were expressed in ppbv following a procedure Adv. Hort. Sci., 2017 31(4): 329-337 332 described by Lindinger and Jordan (1998). Then, the data obtained were filtered by eliminating peaks that were lower than a threshold of 0.50 ppbv and elimi- nating all signals relative to ions hard to quantify pre- cisely. After filtration, data have been sent to the sta- tistical analysis (Fig. 1). Provenience zone Cultivar/blend Number of samples Get from: producer (1), supermarkets (2) Processing campaign (A), or acquisition (B) year PTR-ToF-MS analysis year EVOO (0)/Not EVOO (1) Argentina Arbosana 4 1 2017 A 2017 1 (4) Argentina Blend 4 2 2012 B 2016 1 (4) Argentina Blend 4 1 2013 A 2016 1 (4) Argentina Blend 4 1 2014 A 2016 1 (4) Argentina Coratina 2 1 2015 A 2016 1 (2) Argentina Coratina 1 1 2017 A 2017 0 (1) Argentina Koroneiki 2 1 2017 A 2017 0 (2) Chile Arbequina 2 1 2012 A 2017 1 (2) Chile Arbosana 4 1 2012 A 2017 1 (4) Chile Arbosana 1 1 2013 A 2017 1 (1) Chile Koroneiki 1 1 2012 A 2017 1 (1) Chile Koroneiki 2 1 2013 A 2017 1 (2) Italy Arbequina 4 1 2012/13 A 2016 1 (4) Italy Arbequina 3 1 2013/14 A 2016 1 (3) Italy Arbequina 7 1 2015/16 A 2016 0 (7) Italy Arbequina 5 1 2016/17 A 2017 0 (5) Italy Arbosana 3 1 2015/16 A 2016 0 (3) Italy Arbosana 2 1 2016/17 A 2017 1 (2) Italy Blend 2 1 2012/13 A 2017 1 (2) Italy Blend 11 1 2015/16 A 2015 0(11) Italy Blend 12 1 2015/16 A 2016 0(12) Italy Blend 14 1 2016/17 A 2017 0(10) 1(4) Italy Blend 17 2 2017 B 2017 0(13) 1(4) Italy Carolea 5 1 2013/14 A 2015 1(5) Italy Frantoio 8 1 2013/14 A 2015 1(8) Italy Frantoio 2 1 2015/16 A 2015 0(2) Italy Gentile di Chieti 9 1 2015/16 A 2015 0(9) Italy Gentile di Chieti 4 1 2015/16 A 2016 0(4) Italy Intosso 7 1 2015/16 A 2015 0(7) Italy Intosso 4 1 2015/16 A 2016 0(4) Italy Itrana 7 1 2015/16 A 2015 0(7) Italy Itrana 4 1 2015/16 A 2016 0(4) Italy Koroneiki 3 1 2015/16 A 2016 0(3) Italy Koroneiki 4 1 2016/17 A 2017 0(4) Italy Leccino 5 1 2013/14 A 2015 1(5) Italy Maurino sel. Vittoria 3 1 2015/16 A 2016 0(3) Italy Maurino sel. Vittoria 4 1 2016/17 A 2017 0(4) Italy Oliana 5 1 2016/17 A 2017 1(5) Italy Olivastra seggianese 18 1 2015/16 A 2016 0(3) 1(15) Italy Peranzana 5 1 2015/16 A 2015 0(5) Italy Peranzana 4 1 2015/16 A 2016 0(4) Italy Sikitita 5 1 2015/16 A 2016 0(5) Italy Sikitita 5 1 2016/16 A 2017 1(5) Morocco Arbequina 2 1 2012/13 A 2017 1(2) Morocco Picholine maroccaine 3 1 2014/15 A 2016 1(3) Morocco Picholine maroccaine 3 1 2014/15 A 2017 1(3) Morocco Picholine maroccaine 3 1 2015/16 A 2016 1(3) Morocco Picholine maroccaine 6 1 2015/16 A 2017 1(6) Tunisia Koroneiki 4 1 2012/13 A 2017 1(4) U.E. Blend 5 2 2016 B 2016 1(5) U.E. Blend 25 2 2017 B 2017 1(25) Table 1 - Description of 273 olive oil samples in relation to provenience zone, cultivar, acquisition from producer or supermarkets, pro- cessing campaign or getting year, PTR-ToF-MS analysis year, and Panel Test judgement (EVOO/Not EVOO) Taiti and Marone - EVOO or not EVOO? 333 Multivariate data analysis A principal component analysis (PCA, unsuper- vised method) was applied to the spectral data of 273 olive oil samples, submitted to a logarithmic transformation and mean centering as pre-process- ing. Computations were performed by PLS-Toolbox v. 8.0.2 (Eigenvector Research Inc., West Eaglerock Drive, Wenatchee, WA) for MATLAB® R2015b (Mathworks Inc., Natick, MA, USA). A multivariate partial least squares-discriminant analysis (PLS-DA, supervised method) was applied on the spectra of the 273 olive oil samples, to develop a model for dif- ferentiating EVOO from Not EVOO. As pre-processing data, they were submitted to a logarithmic transfor- mation and auto-scaling. The training set (85% of the samples) allowed to select the optimal number of latent variables (LVs) throughout the calibration and cross validation phases. The training and validation subsets were obtained by the Euclidean distances based on the algorithm of Kennard and Stone (1968). The test set (prediction) consisted of 15% of the sam- ples previously removed from the dataset. As cross validation procedure, Venetian blind with 10 splits and 1 sample per split was chosen. The performances of the model were evaluated by the number of cor- rect assignments and the root-mean-squared error of cross-validation (RMSECV), and prediction (RMSEP). The optimal number of LVs resulted associated to the minimum error and misclassification rate of the cali- bration dataset. The reliability of the model was test- ed by confusion matrices. The threshold to assign a sample to a class was chosen based on the Bayes the- orem, minimizing the number of false positives and false negatives. Variable Importance in Projection (VIP) scores (p=0.01) were also calculated. A random permutation of the class labels (permutation test) was also performed (500 iterations), so to generate nonsense datasets for comparison with the true model, to evaluate the probability that the model is significantly different from one casually built up under the same conditions. PLS-DA analysis was per- formed by PLS-Toolbox v. 8.0.2 (Eigenvector Research Inc., West Eaglerock Drive, Wenatchee, WA) for MATLAB® R2015b (Mathworks Inc., Natick, MA, USA). A PCA was then applied to the spectral (PTR-ToF-MS) data of the 122 samples resulting EVOO based on the Panel Test, previously submitted to a logarithmic transformation and auto-scaling. 3. Results and Discussion EVOO or Not EVOO VOCs emission by olive oil is characterized by the presence of different compounds belonging mainly to alcohols, esters, aldehydes, ketones, terpenes and hydrocarbons. C6 molecules are the main volatile compounds derived from polyunsaturated fatty acids Fig. 1 - Schematic representation of oil samples analyses and classification using the PTR-ToF-MS. This technique allows rapid and non- destructive VOCs detection throughout the entire food-to-fork chain (e.g. oils) without any sample pretreatment. All data acquired by PTR-ToF-MS were used to obtain analytical information regarding the quality of product and for trade categories, varieties and geographical origin classification applying different multivariate analyses. Adv. Hort. Sci., 2017 31(4): 329-337 334 through the LipOXygenase pathway (Cecchi and Alfei, 2013), generally characterized by low molecular weight. These compounds easily come in contact with the olfactory cells and help to create flavor and sometimes off-flavor. According to Marone et al. (2017), there is the possibility to directly relate the volatile profile obtained by PTR-ToF-MS to distin- guish EVOO from Not EVOO, and, as a consequence, to correctly classify the virgin olive oils in their trade category. To confirm the preliminary results obtained by Marone et al. (2017), and validate the new proce- dure and methodology, we used a huge number of samples that were collected and analyzed in different years. In the present work, to define their trade cate- gory, 273 olive oil samples were submitted to the SE, that classified 151 samples as Not EVOO, and 122 as EVOO. By analyzing each oil sample (Table 1), 63 volatile compounds were detected within a mass range of m/z = 20-210 (data not shown). PCA applied to the whole dataset (ppbv) allowed to get a first general overview of the data distribution. Two main groups of EVOO and Not EVOO were clearly highlight- ed (Fig. 2) in the bidimensional space of the first two components, despite the great variability present in the original data set, due to the great diversification in the olive oil samples. This variability is also evi- denced by the need to consider the first 7 compo- nents to justify 90.17% of the total variance (respec- tively: 60.26%, 12.56%, 5.17%, 3.81%, 3.39%, 2.78, and 2.20%). The data ordination clearly highlights that the VOCs spectra provided by Not EVOO samples were well distinguishable from those of the EVOO, with a few partially overlapping zones in the upper right and bottom left quadrants. This behaviour indi- cates a different spectral distribution between fla- vors and off-flavors, confirming the same result obtained by the SE. Subsequently, a partial least squares discriminant analysis (PLS-DA) approach was applied to determine the trade category of the olive oil samples. A seven-component PLS-DA model, eval- uated by its performances indicators (Table 2), result- ed robust to discriminate the Not EVOO from the EVOO samples in the model/validation data set, and in the independent test set. The optimal number of latent variables (LVs), associated to the minimum error rate and the minimum number of not assigned samples, resulted in 7 (Table 2). The permutation test Table 2 - PLS-DA statistics for each Y-Block (class 1 = Not EVOO; class 0 = EVOO) related to 273 olive oil samples. Sensitivity (SE); Specificity (SP); Class error, RMSEC, RMSECV, and RMSEP for Calibration (Cal), Cross Validation (CV), and Prediction (Pred), respectively. Confusion matrices for Calibration, Cross Validation, and Prediction Statistics LVs SE (Cal) SP (Cal) SE (CV) SP (CV) Class. error (Cal) Class. error (CV) Class. error (Pred) RMSEC RMSE CV RMSE P Not EVOO 7 0.972 0.968 0.954 0.960 0.029 0.043 0.042 0.206 0.261 0.226 EVOO 0.968 0.972 0.960 0.954 Confusion matrices Classes Matthew's correlation coefficient1-Not EVOO 0-EVOO Calibration results P re d ic te d a s 1-Not EVOO 105 4 0.940 0-EVOO 3 121 Cross validation results 1-Not EVOO 103 5 0.914 0-EVOO 5 120 Prediction results 1-Not EVOO 16 2 0.903 0-EVOO 0 22 Fig. 2 - PCA ordination of 273 olive oil samples. Green = Not EVOO, red = EVOO. Taiti and Marone - EVOO or not EVOO? 335 PLS-DA model also allowed to evidence the significant (>1.5) VIP scores, indicating the role of the selected protonated masses to differentiate the two classes (Fig. 3). VIP scores reported in figure 3 confirm the results of our preliminary work (Marone et al., 2017). In particular, the masses m/z = 47.050 (Tentatively identified (TI) as: ethanol), m/z = 61.030 (TI: acetic acid), m/z = 75.040 (TI: propanoic acid) and m/z = 89.060 (TI: butanoic acid) resulted as factors able to distinguish EVOO from Not EVOO. Indeed, ethanol and acetic acid are generally considered as com- pounds deriving from microbial alterations due to a long time of olive storage before processing (Morales et al., 2000) and therefore represent a known defect. Likewise propanoic acid and butanoic acid are both considered defective compounds, that can be linked to fermentation processes in olive fruits as a long time of storage (Angerosa et al., 1996) or related to the sugar fermentation (Morales et al., 2013). Overall, for all the samples evaluated, it is inter- esting to note that the procedure applied in this study to discriminate EVOO from Not EVOO is not affected by factors such as: year under analysis, har- vesting year, variety and geographical origin. In fact the model resulted significant at 95% confidence level only considering as classification (variability) factor the EVOO/Not EVOO distinction. Classification of different EVOO fruity intensity EVOO are currently also labeled according to the fruity intensity perceptions (robust, medium, deli- cate), based on the IOOC regulation (COI/T.20/Doc. No 15/Rev. 8, 2015). Thus, we tried to evaluate the different EVOO fruity intensity using the dataset pro- vided only by the VOCs profile of the 122 EVOO sam- ples (according to the Panel Test) (Table 1) through a indicated that the model is significant at 95% confi- dence level. In fact, the probability of model insignifi- cance vs. permuted samples resulted 0.0 based on the Wilcoxon and Sign Test, both in Self-Prediction and Cross-Validated, and 0.005 by the Rand t-test. The model successfully classified 96.9% of samples into their trade category based on the Panel Test results in fitting, 95.5% in cross validation (internal validation) and 95% in prediction (external valida- tion). That is, as reported in the confusion matrices (Table 2), in the calibration, on a total of 233 sam- ples, 226 were correctly classified, while 3 resulted false positive (predicted as EVOO from the Panel Test, but classified as Not EVOO by the spectrome- ter) and 4 false negative (predicted as Not EVOO from the Panel Test, but classified as EVOO by the spectrometer). In the cross validation, 223 samples were correctly classified, while 5 resulted false posi- tive, and 5 false negative. In the prediction results, on 40 samples, 38 were correctly assigned to their right class, while only 2 resulted false positive. The occurring of false positive (3 samples in prediction, judged as EVOO by the Panel Test, and classified as Not EVOO by the spectrometer), can be related to the fact that all compounds (including off-flavors) are only perceived by the human olfactory when they exceed their specific threshold values (Morales et al., 2013). Thus, we can assume that, below this thresh- old values, the presence of a given compound linked to a defect is not perceived by the human olfactory, but is inexorably detected by the spectrometer. On the other hand, only a few borderline olive oils judged as Not EVOO by the panelists but classified as EVOO by the tool (false negative) were detected. A scores plot of the first two components of the PLS-DA model for all oil samples is shown in figure 3. The Fig. 3 - Score plot (LV1, LV2) of the PLS-DA model. Green = Not EVOO, red = EVOO; VIP scores > 1.5. Adv. Hort. Sci., 2017 31(4): 329-337 336 PCA analysis. The first two components explained about 63% of the total variability, and the derived scatterplot (Fig. 4) showed three groups of samples that are rather well separated. It is interesting to note that linking the fruity score assigned by the Panel Test to each sample, the three groups distrib- uted in the chart according to their fruity intensity (Fig. 4), even if the samples grouped by the tool at the bottom of the figure (Fig. 4) show VOCs profiles relatively close, while the fruity intensity scores attributed to the same samples by the panelists ranged from 4 to 7. According to this chemometric approach, the subjectivity of the Panel Test becomes evident at intermediate values of fruity intensity. PCA analysis also underlines two outliers group. In the first one, labeled with “M”, the three samples belonging to cv. Maurino sel. Vittoria harvested in year 2016 are found; this can be linked to peculiar flavor notes characterizing this Tuscan clone, that showed the highest amount of terpene compounds compared to all other samples (data not shown). The second one, represented by a few samples separated from the central bulk and shifting to the right part, labeled with “S” (Fig. 4), is formed by samples with particular flavor notes (data not shown). These sam- ples, belonging to the cv. Sikitita, as reported by García-Gonzalez et al. (2010), are in fact character- ized by typical aromas. Associating the fruity score assigned by the Panel Test to each sample, it is remarkable as the entire aromatic profile detected by the PTR-ToF-MS seems to be linked to changes in the amount of masses within the spectra rather than to the presence of specific compounds in the human olfactory percep- tion of the fruity intensity. 4. Conclusions The chemometric classification model proposed in this trial and based on the VOCs fingerprint acquired by the PTR-ToF-MS allows to distinguishing olive oil samples of different trade category. In particular, it was demonstrated that: (1) the entire volatile profile can be useful to classify oils belonging to different commercial categories (as the Panel Test), (2) the dif- ferent qualities and types of EVOO can be split by using the fruity intensity. The accuracy of classifica- tion proposed is very high and it is more efficient than that obtained by other authors using different tools. Indeed, this tool does not require any sample pre-treatment and allows identifying compounds with low molecular weight (i.e methanol, ethanol, etc.) compared to other ones. Given our results and the emerging need of the olive oil sector that requires the developmental ana- lytical tools to support or integrate the Panel Test, this work opens the way for the use of PTR-ToF-MS coupled with an appropriate multivariate analysis, as a quick and cheap tool with high confidence level and Fig. 4 - PCA ordination of 122 extra virgin olive oil samples. The objects key color indicates fruity inten- sity scores as evaluated by the Panel test increasing from blue (fruity = 1) to yellow (fruity = 8.5). Black circled samples indicate: Maurino sel. Vittoria (M) and Sikitita (S). Taiti and Marone - EVOO or not EVOO? 337 comparable to the Panel Test, for the olive oil quality identification. References ANGEROSA F., LANZA B., MARSILIO V., 1996 - Biogenesis of ‘‘fusty’’ defect in virgin olive oils. - Grasas Aceites, 47: 142-150. ANGEROSA F., SERVILI M., SELVAGGINI R., TATICCHI A., ESPOSTO S., MONTEDORO G.F., 2004 - Review: Volatile compounds in virgin olive oil: occurrence and their relation- ship with the quality. - J. Chromatogr. A, 1054(1-2): 17-31. APARICIO R., ROCHA S.M., DELGADILLO I., MORALES M.T., 2000 - Detection of rancid defect in virgin olive oil by the electronic nose. - J. Agric. Food Chem., 48(3): 853-860. APREA E., BIASIOLI F., SANI G., CANTINI C., MARK T.D., GASPERI F., 2006 - Proton transfer reaction - mass spectrometry (ptr-ms) headspace analysis for rapid detection of oxidative alteration of olive oil. - J. Agric. Food Chem., 54(20): 7635-7640. AUED-PIMENTEL S., SILVA S.A.D., TAKEMOTO E., CANO C.B., 2013 - Stigmastadiene and specific extinction (270 nm) to evaluate the presence of refined oils in virgin olive oil commercialized in Brazil. - Food Science and Technology (Campinas), 33(3): 479-484. BENDINI A., CERRETANI L., CARRASCO-PANCORBO A., GOMEZ-CARAVACA A.M., SEGURA-CARRETERO A., FER- NANDEZ-GUTIERREZ A., LERCKER G., 2007 - Phenolic molecules in virgin olive oils: a survey of their sensory properties, health effects, antioxidant activity and ana- lytical methods. An overview of the last decade. - Molecules, 12(8): 1679-1719. BLAKE R.S., MONKS P.S., ELLIS A.M., 2009 - Proton-transfer reaction mass spectrometry - Chem. Rev., 109(3): 861-896. CATHARINO R.R., HADDAD R., CABRINI L.G., CUNHA I.B., SAWAYA A.C., EBERLIN M.N., 2005 - Characterization of vegetable oils by electrospray ionization mass spec- trometry fingerprinting: classification, quality, adulter- ation, and aging. - Analytical Chemistry, 77(22): 7429- 7433. CECCHI T., ALFEI B., 2013 - Volatile profiles of Italian mono- varietal extra virgin olive oils via HS-SPME-GC-MS: Newly identified compounds, flavors molecular markers, and terpenic profile. - Food Chem., 141(3): 2025-2035. 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(2): 485-491. GARCÍA-GONZÁLEZ D.L., TENA N., APARICIO R., 2010 - Quality characterization of the new virgin olive oil var. Sikitita by phenols and volatile compounds. - J. Agric. Food Chem., 58(14): 8357-8364. GOMEZ-RICO A., SALVADOR M.D., LA GRECA M., FREGA- PANE G, 2006 - Phenolic and volatile compounds of extra virgin olive oil (Olea europaea L. cv Cornicabra) with regard to fruit ripening and irrigation manage- ment. - J. Agric. Food Chem., 54(19): 7130-7136. GUADARRAMA A., RODRIGUEZ-MENDEZ M.L., DE SAIA J.A., RIOS J.L., OLIAS J.M., 2000 - Array of sensors based on conducting polymers for the quality control of the aroma of the virgin olive oil. - Sensors and Actuators B: Chemical, 69(3): 276-282. INTERNATIONAL OLIVE OIL COUNCIL, 2015 - Sensory analysis of olive oil. Method for the organoleptic assessment of virgin olive oil. - COI, T.20, Doc. No 15, Rev. 8, pp. 20. KALUA C.M., ALLEN M.S., BEDGOOD Jr. D.R., BISHOP A.G., PRENZLER P.D., ROBARDS K., 2007 - Olive oil volatile compounds, flavor development and quality: A critical review. - Food Chem., 100: 273-286. KENNARD R.W., STONE A., 1968 - Computer aided design of experiments. - Technometrics, 11: 137-148. LINDINGER W., JORDAN A., 1998 - Proton-transfer-reac- tion mass spectrometry (PTR-MS): On-line monitoring of volatile organic compounds at pptv levels . - Chemical Society Reviews, 27: 347-375. MARONE E., FIORINO P., 2012 - Oleiculture in progress. - Adv. Hort. Sci., 26(3-4): 163-175. MARONE E., MASI E., TAITI C., PANDOLFI C., BAZIHIZINA N., AZZARELLO E., FIORINO P., MANCUSO S., 2017 - Sensory, spectrometric (PTR-ToF-MS) and chemometric analyses to distinguish extra virgin from virgin olive oils. - J. Food Sci. Technol., 54(6):1368-1376. MORALES M.T., APARICIO-RUIZ R., APARICIO R., 2013 - Chromatographic methodologies: compounds for olive oil odor issues, pp. 261-310. - In: APARICIO R., and J. HARWOOD (eds.) Handbook of olive oil: analysis and properties. 2 edition. Springer, New York, USA, pp. 772. MORALES M.T., LUNA G., APARICIO R., 2000 - Sensory and chemical evaluation of winey-vinegary defect in virgin olive oils. - Eur. Food Res. Technol., 211: 222-228. PERI C., RASTELLI C., 1994 - Implications for the future and recommendations for modifications to current regula- tions concerning virgin olive oil. - Grasas Aceites, 1: 60-61. PROCIDA G., CICHELLI A., LAGAZIO C., CONTE L.S., 2016 - Relationships between volatile compounds and sensory characteristics in virgin olive oil by analytical and chemometric approaches. - J. Sci. Food Agric., 96(1): 311-318. TAITI C., COSTA C., GUIDI NISSIM W., BIBBIANI S., AZZARELLO E., MASI E., PANDOLFI C., PALLOTTINO F., MENESATTI P., MANCUSO S., 2017 - Assessing VOC emission by different wood cores using the PTR-ToF-MS technology. - Wood Sci. Technol., 51(2): 273-295. VLACHOS N., SKOPELITIS Y., PSAROUDAKI M., KOSTANTI- NIDOU V., CHATZILAZAROU A., TEGOU E., 2006 - Applications of Fourier transform-infrared spectroscopy to edible oils. - Analytica Chimica Acta, 573: 459-465. ZOU M.Q., ZHANG X.F., QI X.H., MA H.L., DONG Y., LIU C.W., GUO X., WANG H., 2009 - Rapid authentication of olive oil adulteration by Raman spectrometry. - J. Agric. Food Chem., 57(14): 6001-6006.