Traceability monitoring of Greek extra virgin olive oil by Differential Scanning Calorimetry

Traceability monitoring of Greek extra virgin olive oil by Differential Scanning Calorimetry

Thermochimica Acta 576 (2014) 9–17 Contents lists available at ScienceDirect Thermochimica Acta journal homepage: www.elsevier.com/locate/tca Trace...

1003KB Sizes 0 Downloads 17 Views

Thermochimica Acta 576 (2014) 9–17

Contents lists available at ScienceDirect

Thermochimica Acta journal homepage: www.elsevier.com/locate/tca

Traceability monitoring of Greek extra virgin olive oil by Differential Scanning Calorimetry S.E. Chatziantoniou, D.J. Triantafillou ∗ , P.D. Karayannakidis, E. Diamantopoulos Department of Supply Chain Management and Logistics, Alexander Technological Educational Institute (A.T.E.I.) of Thessaloniki, Branch of Katerini, Kanelopoulou 2, 60100 Katerini, Greece

a r t i c l e

i n f o

Article history: Received 31 March 2013 Received in revised form 3 November 2013 Accepted 15 November 2013 Available online 24 November 2013 Keywords: DSC Extra virgin olive oil Traceability Cultivar Geographical origin

a b s t r a c t In the present study, Differential Scanning Calorimetry (DSC) was used as the sole analytical tool for the determination of the varietal and geographical origin of Greek extra virgin olive oil (EVOO). EVOO samples were of four different monovarietal categories and originated from four geographical regions of Greece. Thermal properties of EVOOs were studied by obtaining the heating and cooling profiles of samples. Linear discriminant analysis (LDA) of combined results, obtained from heating and cooling profiles, showed that the DSC protocols studied can be effectively used for the determination of varietal and geographical origin of Greek EVOOs, correctly classifying 97.3% and 88.5% of samples, respectively. Furthermore, EVOOs of Koroneiki cultivar from two different geographical regions of Greece were correctly classified by 90.5% in their assigned groups of geographical origin. The applied DSC was shown to have great potential of being used as a tool for traceability monitoring of Greek EVOOs. © 2013 Elsevier B.V. All rights reserved.

1. Introduction Olive cultivation is widespread throughout the Mediterranean region and it is important for the rural economy, local heritage and the environment, whereas olive oil represents the main source of dietary fats in Mediterranean countries [1]. In particular, extra virgin olive oil (EVOO) is the highest quality product among olive oils, as it is obtained from olive fruits using only mechanical processing steps or other physical means under conditions that do not lead to oil alteration. EVOO is highly appreciated for its natural flavor and aroma, as well as for its health and nutritional properties [2]. Traceability of EVOO is of growing interest among producers, since it leads to supply chain optimization, increase of producers’ competitiveness and prevention of mislabeling of geographical origin and olive varieties of products, so as to consequently assure correct information to the consumers [3]. Furthermore, EC regulation on marketing standards for olive oil enables producers to market their extra virgin and virgin olive oils on the basis of geographic origin [4]. The complete field to fork traceability of olive oil involves the characterization of the oils obtained from the main cultivars in each production zone, as the chemical composition of EVOO is well known to be influenced by genotype, different agronomic, environmental and technological factors [5].

∗ Corresponding author. Tel.: +30 23510 20940; fax: +30 23510 47860. E-mail address: [email protected] (D.J. Triantafillou). 0040-6031/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.tca.2013.11.014

Greece is the third largest producer of virgin olive oils in the world, with about 353,000 tons in 2010, around 80% of which are extra virgin olive oils, thus making Greece one of the world’s largest EVOO producing countries [6]. In the case of Greek EVOO, the implementation of a reliable traceability system is very important, given the nutritional and economic value of this product for the Greek population. Among the parameters that should be recorded in such a system are those that verify the varietal and geographical origin of the product. Such data, as well as other parameters of a traceability system should be collected by means of analytical or other methods, which must be simple and rapid, while the results obtained should be accurate, sensitive and reproducible. Various attempts to confirm the varietal and geographical origin of edible oils have been reported, using different analytical techniques or combinations of techniques, such as gas chromatography (GC) [2,5], gas chromatography–mass spectrometry (GC–MS) [7], ultraviolet–visible spectroscopy (UV–vis) [8], near–infrared spectroscopy (NIR) [9], Fourier transform infrared spectroscopy (FTIR) [10], nuclear magnetic resonance spectroscopy (NMR) [1], high–pressure liquid chromatography (HPLC) [11], inductively coupled plasma mass spectrometry (ICP–MS) [12] and others. A number of the above mentioned techniques are expensive, require analytical expertise, are time consuming and have a high environmental impact. The development of additional analytical techniques as supporting tools for currently used methods would improve EVOO traceability monitoring. In this respect, Differential Scanning Calorimetry (DSC) could be a promising alternative. This technique, which allows the

10

S.E. Chatziantoniou et al. / Thermochimica Acta 576 (2014) 9–17

physical changes that occur upon heating/cooling of a sample to be determined, possesses the advantages of being accurate, repeatable, relatively quick and simple to carry out, with minimal sample preparation, involving no chemical treatments [13]. It has been mostly used in studies of oxidation processes and deterioration of oils [14–16]. Moreover, it has been applied for the characterization of oils [14,17,18], providing a reproducible method for their identification. A data bank with the calorimetric “fingerprints” of the main edible oils has been created [19] and a classification of peanut oils according to variety or geographical origin has been attempted [20]. The effects of heating and cooling rate on the melting and crystallization properties have been studied with DSC and the applicability of calorimetric methods in oil quality control discussed [17]. A good relationship has been reported between thermal properties and components of monovarietal EVOO samples derived from both cooling and heating profiles [21–23]. DSC application upon cooling and heating also appeared very promising in discriminating among oil samples from olives of different commercial categories, cultivars and/or harvesting periods [21,23–26]. There is limited number of studies reporting application of DSC aiming at discriminating monovarietal EVOOs according to cultivar or geographical origin. Chiavaro and co-workers studied the thermal properties of three Sicilian monovarietal EVOOs during cooling and heating [21,22], which were related to their chemical composition (triacylglycerols, diacylglycerols, total and free fatty acids, oxidation status). They reported differences in various thermal parameters among samples of different cultivars after deconvolution of DSC cooling and heating thermograms. These differences were explained by the observed differences in chemical composition of EVOOs. Kotti and co-authors studied the thermal properties of two Tunisian monovarietal EVOOs by DSC, aiming at discriminating olive varieties grown in different geographical regions of Tunisia [27]. These researchers reported an effective discrimination of geographical provenience by evaluating onset and offset temperatures of crystallization observed during cooling of samples. Moreover, all thermal parameters obtained by analyzing the heating thermograms significantly differed among EVOOs produced by the same olive varieties but grown in different geographical regions. A recent study revealed that DSC results in combination with results from GC and isotope analysis are effective in the evaluation of EVOO’s geographical provenience and botanical differentiation [28]. To the authors’ best knowledge, no literature data are present where DSC is employed as the sole technique for the geographical and botanical discrimination of EVOO samples. The objective of the present study was the investigation of the applicability of DSC for the determination of the varietal and geographical origin of Greek EVOO. Heating and cooling profiles of EVOOs were studied in order to identify, by means of linear discriminant analysis, which thermal parameters provide efficient sample differentiation and grouping. Fatty acid contents were also determined, aiming to further characterize Greek EVOOs.

Table 1 Geographical origin, cultivar and number of analyzed Greek EVOO samples. Cultivar

Koroneiki Lianolia Adramitiani Thasitiki

Number of samples per geographical origin Crete

Peloponnesus

8

6

Ionian Sea

N. Aegean Sea

5 4 4

The olives were harvested by hand or collected in nets at their optimum maturity, during the crop period of 2010–2011, processed by cold-pressed extraction (at 27 ◦ C maximum temperature) and belonged to four important Greek cultivars (Koroneiki, Lianolia, Adramitiani and Thasitiki), all of which are representative of the main olive varieties in Greece. The botanical origin and quality grade of all samples were guaranteed by the Cooperatives Unions, which implement ISO 22005:2007 as a traceability system. The acquired samples were of known cultivar, geographical origin, method and date of production and were shipped in amber glass bottles, without headspace. Broad aspects of traceability were taken into consideration, as samples from different cultivars and regions of Greece were selected. Specifically, samples of Koroneiki cultivar originated from Peloponnesus and Crete (Southwestern and Southern Greece, respectively), of Lianolia from Kerkyra in Ionian Islands (Western Greece) and of Adramitiani and Thasitiki from the islands of N. Aegean Sea (Northeastern Greece), Lesvos and Thassos, respectively. Samples from these specific geographical regions were selected on the premise that the chosen cultivars under analyses are mainly grown in these regions. A total of 27 samples were collected and stored in the dark, at room temperature until the time of analysis, which was no more than 6 months after production. 2.2. Methods

2.1. Materials

2.2.1. Free acidity and fatty acid composition Free acidity, expressed as percentage of oleic acid, was measured using the analytical method described in European Commission Regulation EEC 2568/91 and later amendments [29]. Contents of fatty acids (FA) were determined after preparation of fatty acid methyl esters (FAMEs) via transesterification in accordance to EEC 2568/91 [29]. FAMEs were analyzed by capillary gas chromatography (GC). 2 ␮l of sample were injected into the gas chromatograph, a type GC-17A, Ver. 3 (Shimadzu Corporation, Kyoto, Japan), equipped with an FID detector and 50 m fused silica capillary column, type CP-Sil 88 (Chrompack, Raritan, NJ, USA). The oven temperature was programmed initially at 150 ◦ C, raised to 170 ◦ C at 5 ◦ C/min, kept for 10 min and finally raised to 220 ◦ C at 5 ◦ C/min and kept for 25 min. Total time of analysis was 49 min. Injector and detector temperatures were set at 250 ◦ C. Helium was used as a carrier gas (0.1 kPa/min, constant flow). Identification of FAMEs was possible using appropriate analytical standards of FAMEs (Sigma, St. Louis, MO, USA). All determinations were performed in triplicate and the mean values are reported. Results are expressed as percentage of total fatty acids detected.

All reagents were of analytical grade (Merck, Darmstadt, Germany). Monovarietal EVOO samples included in this study were kindly provided by olive oil Cooperatives Unions from four different geographical regions of Greece (Crete, Peloponnesus, Ionian and Northern Aegean Islands) belonging to different local olive cultivars (Table 1). These Cooperatives Unions are established in central areas and collect and process olives from specific cultivars that originate exclusively from their locale.

2.2.2. Differential Scanning Calorimetry EVOO samples (∼50 ml) were pre-treated according to a method employed by Ferrari and co-workers [25], with slight modifications. Samples were gently stirred with a magnetic stirrer, while being heated at 50 ◦ C for 5 min in a water bath, in order to erase thermal history and improve reproducibility. Quantities (6–10 mg) of samples to be analyzed per protocol described below were accurately weighed (0.01 mg) into aluminum pans, covers were hermetically sealed into place and analyzed with a DSC Q100 (TA

2. Materials and methods

S.E. Chatziantoniou et al. / Thermochimica Acta 576 (2014) 9–17

Instruments, New Castle, DE, USA), equipped with a liquid nitrogen cooling system (LNCS). Dry nitrogen (99.999% purity) was used as purge gas at a flow rate of 50 ml/min. Indium (melting temperature 156.60 ◦ C, H = 28.45 J/g) and n-dodecane (melting temperature 9.65 ◦ C, H = 216.73 J/g) were used to calibrate the instrument and an empty pan was used as reference. Prior to analysis of samples, the baseline was obtained and subtracted from all heat flow curves. The two developed protocols of analysis, cooling and heating, involved cooling and heating of EVOO samples, respectively, at a scanning rate of 10 ◦ C/min. The cooling protocol, from which cooling profiles were obtained, involved equilibration at 50 ◦ C for 2 min, then cooling to −40 ◦ C and subsequent holding at −40 ◦ C for 10 min. The heating protocol, from which heating profiles were obtained, involved equilibration at −50 ◦ C for 10 min and then heating to 40 ◦ C. Three replicates were analyzed per sample for each protocol. The resulting thermograms were analyzed by the Universal Analysis software (Version 4.2E, TA Instruments, New Castle, DE, USA). Overlapping transitions of heating thermograms were deconvoluted into individual constituent peaks using OriginPro software (Version 8.0724, OriginLab Corporation, Northampton, MA, USA), in order to closely monitor heating profiles and potentially differentiate samples. Obtained thermal parameters were onset (Ton ), offset (Toff ) temperatures, which represent intersections of baseline and tangents at each transition, as well as peak temperatures (Tp ), in ◦ C. Range of transitions (Range) was calculated as temperature difference between onset and offset temperatures, in ◦ C. In order to differentiate thermal parameters obtained via the heating or cooling protocol, the distinction (h) or (c), respectively, was added to the thermal parameter abbreviations. Additional thermal parameters regarding specific events or individual transitions derived from deconvolution of heating curves were recorded, which are explicated in Section 3.3. 2.3. Statistical analysis Statistical analysis was conducted using the SPSS 16.0 statistical software package (SPSS Inc., Chicago, IL, USA). One-way analysis of variance (ANOVA) was employed on free acidity and FA content, as well as all twelve (12) thermal parameters of this study, which are explicated in Section 3.3, aiming to reveal potential significant effects among the different cultivars and geographical regions studied. Whenever significant effects were detected (p < 0.05) the Student–Newman–Keuls (SNK) multiple range test was applied. At a second step, the method of linear discriminant analysis (LDA) was applied in order to build predictive models of cultivar or origin group membership based on the values of an appropriate set of independent variables (parameters obtained by GC and DSC). The significance level was 5%. LDA could allow for classification of unknown samples after establishing a model with samples of known cultivar or geographical origin. The technique involves finding a linear combination of independent variables (predictors) – the discriminant function – that creates the maximum difference between group membership in the categorical dependent variable (cultivar in the first case and geographical origin in the second case). Care was taken in identifying outliers, by evaluating box plots, as outliers could have an effect on the outcome of a multivariate data analysis. Estimation of the probability for correct prediction was based on group size, since the number of samples of each cultivar or origin were different. The stepwise selection model was chosen in LDA based on Wilks’ lambda criterion (Wilks’ , the ratio of within-groups sums of squares to the total sums of squares). Values of Wilks’  approaching zero are obtained with well resolved categories, whereas overlapped categories make Wilks’  to approach one. In this algorithm, a predictor (variable) is only selected when the reduction of the Wilks’  criterion produced after its inclusion in

11

the model exceeds Fentry , which is the entrance threshold of a test of comparison of variances or F-test. Chosen criteria were probability values of Fentry and Fremoval equal to 0.05 and 0.10 [30]. Furthermore, cross-validation of the models, by the leave-oneout method, was used in order to ensure that models with the best fit generally include only a subset of independent variables that are deemed truly informative. The optimum model dimensionality was identified as that with the maximum classification success rate by cross-validation. 3. Results and discussion 3.1. Free acidity and FA content Table 2 presents the results of free acidity and fatty acid (FA) content of the analyzed Greek EVOOs, which were grouped according to cultivar (four groups) and geographical origin (four groups), so as to assess differences in samples of the same olive variety and/or geographical provenience. In particular, samples grouped as originating from Crete and Peloponnesus all belonged to Koroneiki cultivar. Samples grouped as of Ionian origin were of Lianolia cultivar. EVOOs assigned to the N. Aegean origin group belonged to both Adramitiani and Thasitiki cultivars. All EVOO samples exhibited free acidity values not exceeding 0.8%, which verified their belonging to the extra virgin oil category. EVOOs of Lianolia cultivar, originating from Kerkyra in the Ionian Islands, had a slightly increased free acidity (0.6%), compared to all other samples (0.2–0.3%). Regarding the FA composition of Greek EVOOs studied, the major FA component was the monounsaturated FA (MUFA) oleic acid (C18:1), representing over 63% of the total FAs detected. Palmitic acid (C16:0) was the predominant saturated FA (SFA) and linoleic acid (C18:2) was the major polyunsaturated fatty acid. Several FAs were also detected in lower concentrations, such as stearic (C18:0), linolenic (C18:3), palmitoleic (C16:1), arachidic (C20:0), gadoleic (C20:1), lignoceric (C24:0), behenic (C22:0) and margaric (C17:0). EVOOs presented similar FA compositions, with slight differences amongst groups (p < 0.05). Similar results have been reported by other studies on FA composition of Greek EVOOs [2]. Samples of Koroneiki and Thasitiki cultivar had a higher content of oleic acid than samples of Adramitiani and Lianolia. Koroneiki produced EVOOs with the lowest content of polyunsaturated FAs (PUFA), mainly due to the content of linoleic acid. Cretan EVOOs had a higher SFA content. No statistical differences (p > 0.05) were observed between samples of Koroneiki cultivar from Crete and Peloponnesus, suggesting that the FA composition of this cultivar was not greatly influenced by environmental factors. Samples from islands of the N. Aegean Sea presented the lowest content of SFA, due to the lower amount of these FAs in samples of Thasitiki cultivar. Ionian EVOOs (Lianolia cultivar) had an increased content of margaric acid (C17:0) and the highest PUFA content. These samples also presented lower MUFA than samples of Koroneiki or Thasitiki cultivar. 3.2. DSC protocol development The liquid ↔ solid phase transitions of oils have been particularly studied by DSC, as they are affected by molecular composition changes [31,32]. Therefore, the study of crystallization and melting by DSC are promising methods in order to assess oil nature, quality and origin. Both heating and cooling thermograms are strongly dependent on the temperature-scanning rate and the thermal history of the sample [17,26]. In order to effectively apply calorimetry to EVOO traceability, the following two requirements must be fulfilled: (i) homogeneity of liquid samples during melting, as well as reproducible and complete solidification of samples during

12

S.E. Chatziantoniou et al. / Thermochimica Acta 576 (2014) 9–17

Table 2 Free acidity (%) and fatty acid content of Greek EVOO samples, as grouped according to cultivar and geographical origin.* Grouping

Cultivar

Geographical origin

Lianolia Number of samples Free acidity (oleic acid %) FA (%) C16:0** C16:1 C17:0 C18:0 C18:1 C18:2 C18:3 C20:0 C20:1 C22:0 C24:0 SFA MUFA (%) PUFA (%)

Adramitiani

5 0.56 ± 0.15a

14.45 1.10 0.15 1.98 63.98 15.93 1.37 0.43 0.40 0.15 0.15 17.30 65.48 17.30

± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.71a 0.16a 0.06a 0.10a 0.16b 0.74a 0.26a 0.05a 0.08a 0.06a 0.06a 0.61a 1.11b 0.54a

4 0.28 ± 0.14b

15.00 1.10 0.08 2.2 63.5 15.60 1.30 0.35 0.55 0.10 0.15 17.88 65.15 16.90

± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.14a 0.28a 0.11b 0.14a 0.28b 0.14a 0.14a 0.07a 0.07a 0.14a 0.07a 0.04a 0.07b 0.20a

Thasitiki

Koroneiki

4 0.32 ± 0.06b

12.00 ± 1.00 ± 0b 2.10 ± 66.58 ± 15.68 ± 1.15 ± 0.40 ± 0.58 ± 0.03 ± 0.48 ± 15.00 ± 68.15 ± 16.83 ±

0.63a 0.12a 0.26a 0.81a 0.54a 0.17a 0.30a 0.17a 0.02a 0.13a 0.67b 0.60a 0.66a

14 0.23 ± 0.09b

15.09 1.10 0.01 2.28 66.79 11.99 1.21 0.58 0.49 0.16 0.34 18.43 68.37 13.21

± ± ± ± ± ± ± ± ± ± ± ± ± ±

1.93a 0.22a 0.01b 0.20a 1.55a 2.12b 0.14a 0.06a 0.09a 0.10a 0.29a 1.76a 1.58a 2.21b

Ionian

N. Aegean

5 0.56 ± 0.15x

14.45 1.10 0.15 1.98 63.97 15.93 1.37 0.43 0.40 0.15 0.15 17.30 65.48 17.30

± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.71xy 0.16x 0.06x 0.10y 1.16y 0.74x 0.26x 0.05xy 0.08x 0.06x 0.06x 0.61xy 1.11y 0.54x

8 0.28 ± 0.11y

13.00 1.03 0.03 2.13 65.55 15.65 1.20 0.38 0.57 0.05 0.37 15.95 67.15 16.85

± ± ± ± ± ± ± ± ± ± ± ± ± ±

1.63y 0.16x 0.02y 0.22xy 1.71xy 0.42x 0.17x 0.21y 0.14x 0.08x 0.20x 1.57y 1.62xy 0.51x

Peloponnesus 6 0.21 ± 0.10y

14.08 ± 1.12 ± 0y 2.23 ± 66.72 ± 12.93 ± 1.23 ± 0.56 ± 0.50 ± 0.15 ± 0.50 ± 17.50 ± 68.33 ± 14.17 ±

2.51xy 0.26x 0.21xy 0.38x 2.52y 0.12x 0.25x 0.13x 0.10x 0.33x 2.10xy 0.52x 2.59y

Crete 8 0.25 ± 0.09y

15.84 1.09 0.01 2.31 66.84 11.29 1.20 0.58 0.48 0.16 0.23 19.13 68.40 12.49

± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.97x 0.22x 0.01y 0.20x 2.08x 1.59y 0.17x 0.23x 0.05x 0.11x 0.21x 1.14x 2.11x 1.71y

* Data are expressed as means ± standard deviations of triplicate determinations. Different superscripts across rows of cultivar grouping (a, b) or geographical origin grouping (x, y) denote statistically significant differences (p < 0.05). ** C16:0, palmitic acid (hexadecanoic acid); C16:1, palmitoleic acid (cis-9-hexadecenoic acid); C17:0, margaric acid (heptadecanoic acid); C18:1, oleic acid (cis-9octadecenoic acid); C18:2, linoleic acid (cis,cis-9,12-octadecadienoic acid); C18:3, linolenic acid (cis,cis,cis-9,12,15-octadecatrienoic acid); C20:0, arachidic acid (eicosanoic acid); C20:1, gadoleic acid (cis-11-eicosenoic acid); C22:0, behenic acid (docosanoic acid); C24:0, lignoceric acid (tetracosanoic acid); SFA, saturated FA; MUFA, monounsaturated FA; PUFA, polyunsaturated FA.

cooling and (ii) the implementation of fast and convenient measuring protocols. In this direction, the pre-treatment of samples by stirring in a water bath (at 50 ◦ C) for 5 min, which allowed sample homogenization, was shown to produce coincident thermograms for both measuring protocols, thus highly reproducible results. The observed improved reproducibility was also reported by Ferrari and co-workers [25], as well as Tan and Che Man [17]. Reproducibility was evaluated by peak temperatures of the major event during heating and of the first exothermic event during cooling (coefficient of variation <10%). Regarding the cooling protocol, 50 ◦ C was chosen as the maximum temperature of the protocol, aiming to provide homogenous samples, also chosen by other studies [17,25,31], while other researchers used a higher initial temperature, at 60 ◦ C for erasing of crystallization memory [18]. This heating has been shown to dissolve and homogenize any waxy material present in the sample, which can act as nuclei that would accelerate the formation of crystals during cooling [31]. Samples were initially equilibrated at 50 ◦ C for 2 min and then cooled down to −40 ◦ C at a scanning rate of 10 ◦ C/min. This lowest temperature was chosen after an observed completion of the solidification process of EVOOs under analyses, where crystallization peaks effectively ended and heat flow returned to null. A similar protocol was employed for the analysis of edible oils, aiming to characterize the authenticity of olive oil [26], followed by a heating ramp back to 50 ◦ C. Analogous combined cooling, isothermal crystallization and subsequent heating protocols have been employed for the study of thermal properties of EVOO, involving cooling down to −30 ◦ C [25], or −80 and −100 ◦ C [14]. Regarding the heating protocol suggested here, the initial isothermal at −50 ◦ C for 10 min was shown to solidify all samples completely, as it is longer than the time required for the complete solidification of EVOOs [25], as also observed for the Greek EVOOs under analysis. Samples were then heated to 40 ◦ C at a scanning rate of 10 ◦ C/min. Up to this maximum temperature of 40 ◦ C all transitions have been completed and heat flow returned to null for all analyzed samples. Therefore, the two chosen protocols, cooling and heating, were shown to produce complete Greek EVOO’s thermal profiles. Similar protocols were also employed by other

researchers [18,21–23,28] studying the thermal properties of olive oil, using a scanning rate of 2 ◦ C/min. In the present study, total run times during the application of heating and cooling protocols were 19 and 21 min, respectively, which are, to our best knowledge, the fastest protocols reported in the literature for obtaining characteristic and informative thermal profiles of EVOO. The choice of the two individual protocols was based on the premise that thermal properties of EVOOs had to be studied individually during heating and cooling, so as to elucidate whether application of one or the other protocol would eventually be considered as redundant. 3.3. Thermograms and transitions Regarding the cooling protocol, a temperature–heat flow is shown in Fig. 1a and the corresponding time–heat flow curve of a typical Greek EVOO sample is shown in Fig. 1b. All samples exhibited multiple transitions during cooling from 50 to −40 ◦ C, as shown in Fig. 1b, where two well discernible exothermic peaks were observed, marked as peak 1 and peak 2. From cooling thermograms, regarding the first exothermic event occurring at higher temperatures (peak 1 in Fig. 1b), the recorded parameters (in ◦ C) were peak temperature [Tp-1 (c)], onset temperature [Ton-1 (c)], offset temperature [Toff-1 (c)] and range of transition [Range1 (c)]. For Greek EVOOs this event was shown to peak [Tp-1 (c)] from a minimum of −27.6 ◦ C to a maximum of −17.1 ◦ C. This transition has been associated, following deconvolution of cooling curves, to the crystallization of more saturated triacylglycerols (TAG) fractions [17,22], specifically disaturated triacylglyceroles (DSTAG) [21]. From cooling thermograms, an additional parameter was recorded, referring to the time (in minutes) that the second and largest exothermic event peaked (peak 2 in Fig. 1b), coded as Timep-2 (c). This event was shown to take place between 10.2 and 11.9 min. The parameter Timep-2 (c) was recorded from time–heat flow curves since the particular cooling protocol cannot give information regarding the peak temperature of the second exothermic event, because it takes place during the isothermal crystallization at −40 ◦ C. This event has been attributed to the crystallization of highly unsaturated TAG, in particular triolein (OOO) [18,22]. The area of this lower-temperature exotherm, obtained via deconvolution, has been found to be statistically

S.E. Chatziantoniou et al. / Thermochimica Acta 576 (2014) 9–17

13

Fig. 1. (a) Heating and cooling protocols applied to Greek EVOOs and the resulting typical temperature–heat flow curves, (b) cooling protocol applied to Greek EVOOs and resulting typical heat flow curve against time of analysis.

correlated with the amount of triunsaturated triacylglycerols present in Sicilian EVOOs [21]. Regarding the heating protocol, during which heating profiles of Greek EVOOs were obtained as temperature–heat flow curves (Fig. 1a), all samples exhibited multiple transitions. Heating thermograms exhibited a minor exothermic peak, occurring in the −38 to −15 ◦ C temperature range, which is attributed to the transition/rearrangement of TAG polymorphic crystals into more stable forms. In particular, part of the metastable ˇ crystals rearrange into the more thermodynamically stable ˇ form, as shown by Xray diffraction (XRD) [18]. The exothermic event is followed by multiple endothermic events, which occur during further heating. These multiple events can be explained by the presence of mixed glyceride groups with different melting points under the specific experimental conditions [23] or the melting–re-crystallization of the original fat crystals, known as polymorphism. The major event, which peaks at lower temperatures, has been related to the melting of different polymorphic crystal forms of more unsaturated TAG fractions, while the minor event at higher temperatures was attributed to the melting of more saturated TAG fractions [14]. Barba and co-workers demonstrated that this minor event at higher temperatures involves the melting of two crystal structures, identified as ˇa and ˇb [18]. Chiavaro and co-workers also identified a clear exothermic event followed by four [23] or five [27] overlapping peaks during heating of Italian EVOOs by DSC, after deconvolution of the complex melting transition. Four overlapping peaks were also identified by Kotti and co-workers during melting of Tunisian EVOOs [27]. However, Greek EVOO samples of a specific geographical origin studied here, as discussed below, were shown to reveal only three overlapping peaks. Deconvolution of multiple overlapping peaks during application of the heating protocol produced fitted curves comprising of multipeaks. Deconvoluted peaks during heating of samples of different Greek cultivars are presented in Fig. 2, where temperature–heat flow curves depict the experimental data, as well as the distinguished and integrated events. The constituent peaks of the complex transitions during heating were established using the 1st derivative method. Fitting of peaks was controlled using asymmetric double Gaussian and asymmetric double sigmoid functions (a tolerance of 0.05) and were evaluated for an increased goodness of fit (R2 > 0.99). The minor exothermic event is presented as peak 0 in Fig. 2. It is shown that minor endothermic events overlap with the major melting transition, which is distinguished as peak 2 in Fig. 2a and as peak 3 in Fig. 2b. The recorded parameters from the heating thermograms referring to the above mentioned largest endothermic event were peak temperature [Tp-major (h)], onset temperature [Ton-major (h)], offset temperature [Toff-major (h)] and range of major transition [Rangemajor (h)], in ◦ C. This major melting transition was

Fig. 2. Temperature–heat flow curves depicting the deconvoluted and integrated events during heating of Greek EVOO samples, typical of different cultivars: (a) Adramitiani or Thasitiki and (b) Koroneiki or Lianolia (, experimental data, ----constituent peaks). fitted curve,

shown to peak [Tp-major (h)] at the −12.2 to −3.6 ◦ C temperature range and was studied in depth since it is referring to the main fraction that affects the thermal properties of EVOOs. Moreover, it was closely studied in order to simplify follow up analysis of heating thermograms, in the case that only those parameters related to this specific major endothermic event provide satisfactory classification according to varietal or geographical origin, thus possibly eliminating the need for deconvolution of adjacent multiple overlapping peaks. An additional parameter was recorded, referring to the full range of transition during heating and coded as Range(h), in ◦ C. The variety of minor events overlapping with the largest melting transition are presented as peaks 1 and 3 in Fig. 2a and as peaks 1, 2 and 4 in Fig. 2b. As seen from Fig. 2, a major difference was observed in the sum of peaks [Sum of peaks(h)] of the various melting transitions of samples of Adramitiani and Thasitiki cultivar, all of which originated from the N. Aegean Islands, specifically Lesvos and Thassos, respectively, in relation to all other samples. In temperature–heat flow curves of some samples of the above

14

S.E. Chatziantoniou et al. / Thermochimica Acta 576 (2014) 9–17

mentioned cultivars, following the first exothermic event, three overlapping peaks were detected (Fig. 2a), while samples from all other cultivars revealed four overlapping peaks during heating (Fig. 2b), even at first–sight analysis, which appeared sharper and more distinct. Deconvolution of the multiple transitions verified the above findings (R2 > 0.99). Specifically, the melting temperature of the final minor endothermic event, Tp-final (h) (◦ C), for samples of N. Aegean peaked at 2.4 ± 1.7 ◦ C and differed (p < 0.05) among samples of all other origins, which were in the range of 4.2–5.5 ◦ C. This concludes that differences are evident in lipid fractions of samples of N. Aegean Islands in relation to those from other geographical regions of Greece. Chiavaro and co-workers [23] showed that the minor endotherm peaking at higher temperatures during heating of Italian EVOOs was in the range of 6.0–8.0 ◦ C, slightly higher than those determined for Greek EVOOs by the present study. The above mentioned researchers discussed that this event showed different peak profiles among samples, being more evident and wider, with a maximum skewed toward higher temperature, or sharper and more symmetric or still hardly distinguishable. These differences had also been observed for EVOOs of different lipid composition in relation to other commercial categories. Also, minor differences in this final peaks’ profile during heating of olive oils were observed for samples of different cultivars from Tunisia [27]. 3.4. Classification Greek EVOO samples were assigned to four groups according to cultivar and into another four groups according to geographical region of origin, as mentioned above. Regarding DSC results, the variables obtained by the heating protocol, were Ton-major (h), Tp-major (h), Toff-major (h), Rangemajor (h), all referring to the major endothermic event, Tp-final (h) corresponding to the final endothermic event, Range(h), which indicates the full range of transition and Sum of peaks(h), which is the number of endothermic peaks overlapping during the melting transitions. The variables obtained by the cooling protocol were Ton-1 (c), Tp-1 (c), Toff-1 (c), Range1 (c), all referring to the first exothermic event occurring at higher temperatures and Timep-2 (c), which corresponds to the time in which the second and largest crystallization transition occurs. One-way analysis of variance indicated that each of the thermal parameters studied partially differentiated samples according to cultivar and geographical origin (p < 0.001). Results of statistical analysis for specific thermal parameters, which were shown to produce better EVOO differentiation by cultivar or geographical origin, are shown in Fig. 3a and b, respectively. Results shown in Fig. 3a reveal those three out of four olive cultivars that can be differentiated using any of the thermal parameters Tp-major (h), Range(h), Tp-1 (c) and Timep-2 (c). Cultivars such as Koroneiki and Lianolia are well distinguished by all thermal parameters shown. SNK’s multiple range test showed that the parameters Tp-major (h), Range(h) and Timep-2 (c) grouped Adramitiani and Thasitiki cultivars as not significantly different, which was anticipated since samples of these two cultivars were from adjacent islands of the N. Aegean Sea. In contrast, peak temperatures of the first exothermic event during cooling [Tp-1 (c)] effectively distinguished samples of Adramitiani and Thasitiki, but grouped Adramitiani and Koroneiki as not statistically different. Results shown in Fig. 3b reveal that three out of four groups of EVOO samples from different geographical regions of Greece can be correctly differentiated using any of the thermal parameters Tp-major (h), Range(h), Tp-1 (c) and Timep-2 (c). Specifically, all of the above parameters differentiated samples from N. Aegean and Ionian Islands, but did not discriminate amongst groups from Crete and Peloponnesus, which was anticipated, since these two geographical regions consisted only of samples of Koroneiki cultivar.

Table 3 Factor structure matrix of discriminant functions of the LDA model, regarding classification of EVOOs according to cultivar. Predictors

f1

f2

f3

Toff-major (h) Tp-1 (c) Range(h) Timep-2 (c)

0.698 −0.666 −0.205 −0.159

−0.110 −0.076 0.697 0.656

0.213 −0.010 −0.600 0.570

Abbreviations: Toff-major (h), offset temperature of the major endotherm during heating. Tp-1 (c), peak temperature of first exotherm during cooling. Range(h), full range of transition during heating. Timep-2 (c), time in which the second exotherm occurs during cooling.

In order to improve Greek EVOO group differentiation by cultivar and geographical origin, all parameters were evaluated with the help of LDA, so as to reveal which of these parameters, singly or in combination, would produce a significant EVOO differentiation and correct classification in their assigned groups. LDA explicitly attempted to model the differences between the classes of data, thus providing Greek EVOO classification according to varietal or geographical origin. Regarding DSC results, as reported above, significant mean differences (p < 0.001) were observed for all the predictors on these dependent variables. Generally, the thermal parameters, from application of both protocols, with strong prediction capabilities for grouping Greek EVOOs according to cultivar were, in decreasing order, Toff-major (h), Range(h), Rangemajor (h), Sum of peaks(h), Tp-final (h), Timep-2 (c), Tp-major (h) and Tp-1 (c). In this respect, thermal parameters obtained either solely from heating [Toff-major (h) and Rangemajor (h)] or solely from cooling profiles [Timep-2 (c)], were shown to correctly classify 92.0 and 80.0% of samples, respectively, as members of their assigned cultivars. Furthermore, the combination of four independent variables from both heating and cooling protocols, namely Toff-major (h), Range(h) with Timep-2 (c) and also Tp-1 (c), improved EVOO classification to 97.3% of samples (Wilks’  = 0.01, p < 0.001). The above parameters chosen as predictors suggest that the effort for deconvolution of thermograms during heating could be avoided in such a case, therefore facilitating data management. Three discriminant functions were calculated, with a combined 2 (12) = 321.566 (p < 0.001). After removal of the first function, there was still highly significant discriminating power (2 (6) = 150.708, p < 0.001). Discriminant functions 1 and 2 (f1 and f2) accounted for 67.9 and 28.2% of the between group variance, respectively. Fig. 4 presents a score plot of the two canonical discriminant functions of the LDA model, constructed for EVOOs’ discrimination according to cultivar (Fig. 4a) or geographical origin (Fig. 4b). Group centroids are the mean values of the discriminant scores for a given category of the dependent variable, which in Fig. 4a is cultivar. The first discriminant function, f1, maximally separates Lianolia and Koroneiki cultivars, while the second function, f2, maximally separates Lianolia from Thasitiki combined with Adramitiani. The factor structure matrix of correlations between predictor variables and discriminant functions, as shown in Table 3, suggests that the primary variables in distinguishing Lianolia from Koroneiki, were Toff-major (h) and Timep-2 (c). In addition, Range(h) and Tp-1 (c) were the primary factors of discrimination between Lianolia and Thasitiki combined with Adramitiani, whereas Toff-major (h) and Timep-2 (c) only marginally loaded on f2 (correlation coefficients, rho, −0.110 and −0.076, respectively). Specifically, 16.7% of cases of Thasitiki cultivar appeared not to be correctly classified by the above analysis, thus grouped as members of Adramitiani cultivar, while all others were effectively grouped (100%) in their assigned cultivars. In this direction, as seen from Fig. 3a, Thasitiki cultivar can be effectively differentiated from all other cultivars, by the thermal parameter Tp-1 (c)

S.E. Chatziantoniou et al. / Thermochimica Acta 576 (2014) 9–17

15

Fig. 3. Error chart plots of various thermal parameters of Greek EVOOs (means ± 95% confidence intervals based on the pooled variance of ANOVA) obtained from the heating protocol: Tp-major (h) (◦ C), peak temperature of major melting transition, Range(h) (◦ C), full range of transitions, or the cooling protocol: Tp-1 (c) (◦ C), crystallization temperature of first exothermic event and Timep-2 (c) (min), time in which the second exothermic event peaked, as affected by (a) cultivar, (b) geographical origin.

(−24.9 ± 2.5 ◦ C) from cooling profiles, since it is clearly different (p < 0.001) from all other groups (ranging from −20.1 ± 0.7 ◦ C to −15.7 ± 1.0 ◦ C), possibly due to the observed lower content of SFA of those samples. Therefore, a combined overview of the above thermal parameter during cooling and results of LDA would correctly classify all Greek EVOO cultivars under analysis. Furthermore, a more detailed analysis of results revealed that the combination of three variables, namely Toff-major (h),

Range(h) and Rangemajor (h), in decreasing order of discriminant capabilities, obtained solely from heating profiles effectively differentiated (p < 0.001) 100.0% of the remainder three cultivars, namely Koroneiki, Lianolia and Adramitiani. In addition, samples from adjacent islands of N. Aegean Sea, such as Lesvos and Thassos, which were of Adramitiani and Thasitiki cultivar, respectively, were 95.2% correctly classified (p < 0.001) using three variables, namely Ton-1 (c), Range1 (c) and Timep-2 (c), obtained exclusively from

Fig. 4. Score plots of two canonical discriminant functions (a) of cultivar classification, using variables Range(h), Tp-1 (c), Toff-major (h) and Timep-2 (c), (b) of geographical origin classification, using variables Range(h), Toff-major (h), Timep-2 (c), Rangemajor (h) and Tp-1 (c).

16

S.E. Chatziantoniou et al. / Thermochimica Acta 576 (2014) 9–17

Table 4 Factor structure matrix of the LDA model, regarding classification of EVOOs according to geographical origin. Predictors

f1

f2

Toff-major (h) Tp-1 (c) Rmajor (h) Range(h) Timep-2 (c)

−0.621 0.526 −0.055 0.467 0.319

0.358 −0.443 −0.277 0.311 0.228

Table 5 Classification of Greek EVOOs of Koroneiki cultivar originating from Crete and Peloponnesus. Observed origin

Crete Peloponnesus Overall percentage

Predicted origin Crete

Peloponnesus

Percentage (%)

23 3

1 15

95.8 83.3 90.5

Abbreviations: Toff-major (h), offset temperature of the major endotherm during heating. Tp-1 (c), peak temperature of first exotherm during cooling. Rangemajor (h), range of major endotherm during heating. Range(h), full range of transition during heating. Timep-2 (c), time in which the second exotherm occurs during cooling.

cooling profiles. The above findings show that according to traceability plans, some groups could be discriminated by either protocol, heating or cooling, while effective classification of all samples to the assigned cultivar groups studied here would be possible by the combined results of the aforementioned techniques. Regarding the differentiation of Greek EVOOs according to geographical origin, in general, the variables with strong prediction capabilities were, in decreasing order, Ton-1 (c), Tp-1 (c), Toff-major (h), Tp-final (h), Rangemajor (h), Sum of peaks(h), Range(h) and Timep-2 (c). Best classification was achieved by the subset of the variables, Range(h), Toff-major (h), Timep-2 (c), Rangemajor (h) and Tp-1 (c) (in decreasing order of discriminant capability), where 88.5% of EVOOs were correctly classified (Wilks’  = 0.01, p < 0.001). Three discriminant functions were calculated, with a combined 2 (15) = 333.032, p < 0.001. The third function did not have a significant discriminant power 2 (1) = 2.952, p = 0.399, while f1 and f2 accounted for 53.0% and 46.8% of the between group variance, respectively. The first discriminant function, f1, maximally separates Ionian origin from the other three groups, while f2 maximally separates groups of N. Aegean from those of Peloponnesus and Crete (Fig. 4b). The factor structure matrix of correlations between predictor variables and discriminant functions, as seen in Table 4, suggests that the variable Toff-major (h) distinguished Ionian origin from the others, while Timep-2 (c) was the factor of discrimination of samples of N. Aegean from those of Peloponnesus and Crete. Samples originating from Ionian and N. Aegean Islands were 100% correctly classified to their assigned groups. However, 25.0% of cases from Crete and 38.9% from Peloponnesus could not be classified to their assigned groups using the above mentioned parameters. This is evident in results shown in Fig. 4b, presenting a score plot of the first and second canonical discriminant functions of the LDA model, constructed for EVOO’s discrimination according to geographical origin, where centroids of Crete and Peloponnesus groups are close together, since all of them belong to Koroneiki cultivar. From all EVOO samples analyzed in the present study, the only groups of samples originating from more than one geographical region were those of Koroneiki cultivar, being the major EVOO producing cultivar in Greece. Stepwise discriminant analysis suggested a model with the variables Tp-final (h) and Timep-2 (c) as predictors, producing a 90.5% overall correct classification. Because only a separation between two origins was required, only one discriminant function (f1) was approximated. Table 5 presents the classification results of the above mentioned analysis regarding Greek EVOOs of Koroneiki cultivar, which originated from Crete and Peloponnesus. The primary variable in distinguishing Crete from Peloponnesus was Tp-final (h) (rho = 0.729), followed by Timep-2 (c) (rho = −0.563). Fig. 5 shows the values of the subset of data related only to samples of Koroneiki cultivar from Crete and Peloponnesus, displaying distributions by the above chosen variables, Tp-final (h) and Timep-2 (c).

Fig. 5. Scatterplot of values of EVOOs of Koroneiki cultivar by variables Tp-final (h) and Timep-2 (c).

From this figure, a clear grouping of EVOOs of Koroneiki cultivar according to geographical origin is presented. GC results of this study were also evaluated so as to determine whether FA composition could discriminate Greek EVOOs, using the same grouping as mentioned for DSC results. FA contents alone or in combination with thermal parameters from DSC did not prove to be effective in discriminating EVOOs in relation to cultivar or geographical origin, producing very low classification scores. Longobardi and co-workers reported results on the chemical characterization of samples of virgin olive oils originating from Ionian Islands, which included FA contents, TAGS, chemical indices, chlorophyll and phenol contents. They employed LDA on multiple parameters and achieved a predicted classification of 95.3% of samples [2]. A recent paper on discrimination of Mediterranean EVOOs according to geographical origin and cultivar was published by Chiavaro and co-workers [28]. A multidisciplinary approach was used, involving deconvoluted DSC results, during cooling down to −80 ◦ C, in combination with results from GC and isotope analysis and correct classification of Italian EVOOs, by means of linear discriminant analysis of combined results, was reported. LDA revealed that variables referring to the first exothermic event during cooling were shown to be better predictors of Italian EVOOs’ cultivar and geographical origin, in relation to other thermal parameters derived from application of the cooling protocol. In particular, LDA of combined results from DSC, GC and isotope analysis, produced correct classification of 94% of Italian EVOOs according to Italian region of origin (a total of 9 chosen variables) and of 100% of Italian EVOOs according to cultivar (a total of 10 variables). This multidisciplinary approach also proved to be efficient in discriminating EVOOs by varietal and geographical origin. The techniques proposed in the present study revealed promising results, comparable to those produced using multidisciplinary approaches. The present prediction models were produced by using techniques of increased simplicity and time efficiency. The findings of this research could be further strengthened by creating larger databases of Greek EVOOs, incorporating monovarietal samples of additional cultivars and from different geographical regions of Greece.

S.E. Chatziantoniou et al. / Thermochimica Acta 576 (2014) 9–17

4. Conclusions Heating and cooling heat flow curves produced by DSC analysis of Greek monovarietal EVOO samples could discriminate samples according to cultivar and origin in a fast and simple way, suitable for oil industry and market traceability systems. Our observations, which are considered as results of an exploratory study with preliminary results, strengthen the premise that the sole use of DSC provides an efficient, fast, accurate and environmentally friendly method for characterizing EVOO samples, with the help of multivariate analysis. In particular, the present study revealed that an accurate method for rapid screening of differences among Greek monovarietal EVOO samples was developed using DSC during cooling, by application of the proposed cooling protocol. Additional and valuable information obtained by application of the heating protocol was shown to correctly classify the majority of samples under analyses, according to cultivar and geographical origin. These two proposed protocols of analysis were shown to be ideal for studying the thermal properties of EVOO samples originating from Greece. Furthermore, DSC can be a valuable technique, suitable for obtaining a specific sample’s thermal “fingerprint”. This thermal “fingerprint” can be accurately used in conformity tests by EVOO manufacturers, standardization centers and other members of EVOO handling supply chains in Greece or as an authentication procedure for EVOOs of a particular region (Protected Designation of Origin or Protected Geographical Indication). The above could be extremely useful provided that a significant data bank is organized for each geographical region or cultivar of interest. Acknowledgement Financial support from the Research Committee Fund of Alexander Technological Educational Institute (project No. 80152/2010, Research Committee of A.T.E.I. of Thessaloniki) is gratefully acknowledged. References [1] F. Longobardi, A. Ventrella, C. Napoli, E. Humpfer, B. Schütz, H. Schäfer, M.G. Kontominas, A. Sacco, Classification of olive oils according to geographical origin by using 1 H NMR fingerprinting combined with multivariate analysis, Food Chem. 130 (2012) 177–183. [2] F. Longobardi, A. Ventrella, G. Casiello, D. Sacco, M. Tasioula-Margari, A.K. Kiritsakis, M.G. Kontominas, Characterisation of the geographical origin of Western Greek virgin olive oils based on instrumental and multivariate statistical analysis, Food Chem. 133 (2012) 169–175. [3] A. Regattieri, M. Gamberi, R. Manzini, Traceability of food products: general framework and experimental evidence, J. Food Eng. 81 (2007) 347–356. [4] European Community, Commission Regulation 1019/2002 of 13 June 2002 on marketing standards for olive oil, Off. J. Eur. Commun. L 155 (2002). [5] T. Cajka, K. Riddellova, E. Klimankova, M. Cerna, F. Pudil, J. Hajslova, Traceability of olive oil based on volatiles pattern and multivariate analysis, Food Chem. 121 (2010) 282–289. [6] Virgin Olive Oil Database, Food and Agriculture Organization Statistics Division (FAOSTAT), Rome, Italy, 2012. [7] S. López-Feria, S. Cárdenas, J.A. García-Mesa, M. Valcárcel, Classification of extra virgin olive oils according to the protected designation of origin, olive variety and geographical origin, Talanta 75 (4) (2008) 937–943. [8] M. Casale, C. Armanino, C. Casolino, M. Forina, Combining information from headspace mass spectrometry and visible spectroscopy in the classification of the Ligurian olive oils, Anal. Chim. Acta 589 (2007) 89–95.

17

[9] O. Galtier, N. Dupuy, Y. Le Dréau, D. Ollivier, C. Pinatel, J. Kister, J. Artaud, Geographic origins and compositions of virgin olive oils determined by chemometric analysis of NIR spectra, Anal. Chim. Acta 595 (2007) 136–144. [10] S. Hennessy, G. Downey, C.P. O’Donell, Confirmation of food origin claims by Fourier transform infrared spectroscopy and chemometrics: extra virgin olive oil from Liguria, J. Agric. Food Chem. 57 (5) (2009) 1735–1741. [11] E. Stefanoudaki, F. Kotsifaki, A. Koutsaftakis, The potential of HPLC triglyceride profiles for the classification of Cretan olive oils, Food Chem. 60 (3) (1997) 425–432. [12] C. Benincasa, J. Lewis, P. Enzo, G. Sindona, A. Tagarelli, Determination of trace element in Italian virgin olive oils and their characterization according to geographical origin by statistical analysis, Anal. Chim. Acta 585 (2007) 366–370. [13] L. Cerretani, R.M. Maggio, C. Barnaba, T.G. Toschi, E. Chiavaro, Application of partial least square regression to differential scanning calorimetry data for fatty acid quantitation in olive oil, Food Chem. 127 (2011) 1899–1904. [14] C.P. Tan, Y.B. Che Man, Differential scanning calorimetric analysis of edible oils: comparison of thermal properties and chemical composition, J. Am. Oil Chem. Soc. 77 (2) (2000) 143–155. [15] S. Besbes, C. Becker, C. Deroanne, G. Lognay, N. Drira, H. Attia, Heating effects on some quality characteristics of date seed oil, Food Chem. 91 (2005) 469–476. [16] M. Jansen, J. Birch, Composition and stability of olive oil following partial crystallization, Food Res. Int. 42 (2009) 826–831. [17] C.P. Tan, Y.B. Che Man, Comparative differential scanning calorimetric analysis of vegetable oils. I. Effects of heating rate variation, Phytochem. Anal. 13 (2002) 129–141. [18] L. Barba, G. Arrighetti, S. Calligaris, Crystallization and melting properties of extra virgin olive oil studied by synchrotron XRD and DSC, Eur. J. Lipid Sci. Technol. 115 (2013) 322–329. [19] S.M. Dyszel, S.K. Baish, Characterization of tropical oils by DSC, Thermochim. Acta 212 (1992) 39–49. [20] S.M. Dyszel, Characterization of peanuts by DSC for country of origin, Thermochim. Acta 226 (1993) 265–274. [21] E. Chiavaro, E. Vittadini, M.T. Rodriguez-Estrada, L. Cerretani, M. Bonoli, A. Bendini, G. Lercker, Monovarietal extra virgin olive oils: correlation between thermal properties and chemical composition, J. Agric. Food Chem. 55 (26) (2007) 10779–10786. [22] E. Chiavaro, E. Vittadini, M.T. Rodriguez-Estrada, L. Cerretani, M. Bonoli, A. Bendini, Monovarietal extra virgin olive oils. Correlation between thermal properties and chemical composition: heating thermograms, J. Agric. Food Chem. 56 (2008) 496–501. [23] E. Chiavaro, M.T. Rodriguez-Estrada, C. Barnaba, E. Vittadini, L. Cerretani, A. Bendini, Differential scanning calorimetry: a potential tool for discrimination of olive oil commercial categories, Anal. Chim. Acta 625 (2008) 215–226. [24] M. Angiuli, C. Ferrari, L. Lepori, E. Matteoli, G. Salvetti, E. Tombari, A. Banti, N. Minnaja, On testing quality and traceability of virgin olive oil by calorimetry, J. Therm. Anal. Calorim. 84 (1) (2006) 105–112. [25] C. Ferrari, M. Angiuli, E. Tombari, M.C. Righetti, E. Matteoli, G. Salvetti, Promoting calorimetry for olive oil authentication, Thermochim. Acta 459 (2007) 58–63. [26] M. Angiuli, G.C. Bussolino, C. Ferrari, E. Matteoli, M.C. Righetti, G. Salvetti, E. Tombari, Calorimetry for fast authentication of edible oils, Int. J. Thermophys. 30 (2009) 1014–1024. [27] F. Kotti, E. Chiavaro, L. Cerretani, C. Barnaba, M. Gargouri, A. Bendini, Chemical and thermal characterization of Tunisian extra virgin olive oil from Chetoui and Chemlali cultivars and different geographical origin, Eur. Food Res. Technol. 228 (2009) 735–742. [28] E. Chiavaro, L. Cerretani, A. Di Matteo, C. Barnaba, A. Bendini, P. Iacumin, Application of a multidisciplinary approach for the evaluation of traceability of extra virgin olive oil, Eur. J. Lipid Sci. Technol. 113 (2011) 1509–1519. [29] Commission Regulation (EEC) no. 2568/91 of 11 July 1991 on the characteristics of olive oil and olive-residue oil and on the relevant methods of analysis, Official J. L248 (1991) 1–83. [30] G.J. McLachlan, Discriminant Analysis and Statistical Pattern Recognition, John Wiley & Sons, Inc, Hoboken, NJ, USA, 2005. [31] A. Adhvaryu, S.Z. Erhan, J.M. Perez, Wax appearance temperatures of vegetable oils determined by differential scanning calorimetry: effect of triacylglycerol structure and its modification, Thermochim. Acta 395 (2003) 191–200. [32] H. Vaikousi, A. Lazaridou, C.G. Biliaderis, J. Zawistowski, Phase transitions, solubility, and crystallization kinetics of phytosterols and phytosterol-oil blends, J. Agric. Food Chem. 55 (2007) 1790–1798.