**3. Results and discussion**

#### **3.1 <sup>1</sup> H-NMR spectra of virgin olive oil**

1H-NMR spectra of VOOs were recorded (Fig. 1). Olive oil is mainly made up of triglycerides, differing in their substitution patterns in terms of length, degree and kind of unsaturation of the acyl groups (Harwood & Aparicio, 2000). The chemical shifts of their 1H signals are well known (Mannina & Segre, 2002; Sacco et al., 2000). However, the 1H signals of the minor oil components, such as mono- and di-glycerides, sterols, tocopherols, aliphatic alcohols,

were used to perform the processing of the spectra. The region of the NMR spectra studied was 0-7 ppm for the geographical origin determination of VOOs, and 0-10 ppm in the VOO stability study. The data tables generated with the spectra of all samples, excluding the eight buckets in the reference region 4.10-4.26 ppm, were submitted to multivariate data analysis.

The data matrices, consisting of the 1H-NMR buckets (variables) arranged in columns and VOO samples in rows, were firstly analyzed by univariate procedures (ANOVA, Fisher index and Box-Whisker plots), and afterwards, by the following multivariate techniques, already described in bibliography (Berrueta et al., 2007): unsupervised ones as principal component analysis (PCA); and supervised as linear discriminant analysis (LDA) and partial least squares discriminant analysis (PLS-DA). Statistic and chemometric data analysis were performed by means of the statistical software packages Statistica 6.1 (StatSoft Inc., Tulsa, OK, USA, 1984-2004), The Unscrambler 9.1 (Camo Process AS, Oslo, Norway, 1986- 2004) and SIMCA-P 11.0 (Umetrics AB, Umea, Sweden, 1992-2005). Strategies used for variable selection in LDA and selection of the optimum number of PLS components in PLS-

For the geographical characterization of VOOs, the supervised techniques were applied to the autoscaled (or standardised) or Pareto-scaled data matrix of the VOO profiles following these steps: (*i*) the data set was divided into a training-test set and an external data set; (*ii*) the training-test set was subsequently divided into a training set and a test set several times in order to perform cross-validation; (*iii*) the training-test set was used for the optimization of parameters characteristic of each multivariate technique by cross-validation, for instance for variable selection in LDA or the number of PLS components in PLS-DA; (*iv*) a final mathematical model was built using all the samples of the training-test set and the optimized parameters; (v) this model was validated using an independent test set of samples (external data set), i.e. performing an external validation. During the parameter optimization step, the models were validated by 3-fold cross-validation (3-fold CV) or leave-one-out cross-validation (LOO). The reliability of the classification models achieved in the cross-validation was studied in terms of recognition ability (percentage of the samples in the training set correctly classified during the modeling step) and prediction ability (percentage of the samples in the test set correctly classified by using the models developed in the training step). The reliability of the final model was evaluated in terms of classification ability (percentage of the samples of the training-test set correctly classified by using the optimized model) and the prediction ability in the external validation (percentage of the samples of the external set correctly classified by

1H-NMR spectra of VOOs were recorded (Fig. 1). Olive oil is mainly made up of triglycerides, differing in their substitution patterns in terms of length, degree and kind of unsaturation of the acyl groups (Harwood & Aparicio, 2000). The chemical shifts of their 1H signals are well known (Mannina & Segre, 2002; Sacco et al., 2000). However, the 1H signals of the minor oil components, such as mono- and di-glycerides, sterols, tocopherols, aliphatic alcohols,

DA are described elsewhere (Rosa M. Alonso-Salces et al., 2010b).

using the optimized model) (Berrueta et al., 2007).

**H-NMR spectra of virgin olive oil** 

**3. Results and discussion** 

**3.1 <sup>1</sup>**

**2.3 Data analysis** 

hydrocarbons, fatty acids, pigments and phenolic compounds (Harwood & Aparicio, 2000), are only observed by 1H-NMR when their signals do not overlap with those of the main components and their concentrations are high enough to be detected (R. M. Alonso-Salces et al., 2010a; Rosa M. Alonso-Salces et al., 2010b; D'Imperio et al., 2007; Guillen & Ruiz, 2001; Mannina et al., 2003; Sacchi et al., 1996). Table 1 gathers the common 1H-NMR signals of the major and some minor compounds together with their chemical shifts and their assignments to protons of the different functional groups. Several signals of minor compounds were found in 1H-NMR spectra recorded because they were not overlapped by those of the triglyceryl protons: cycloartenol at 0.318 ppm and 0.543 ppm, -sitosterol at 0.669 ppm, stigmasterol at 0.687 ppm, squalene at 1.662 ppm, sn-1,2 diglyceryl group protons at 3.71 ppm and 5.10 ppm, and three unknown terpenes at 4.571 ppm, 4.648 ppm and 4.699 ppm.

Fig. 1. 1H-NMR spectra of a VOO (signal numbering, see Table 1).

#### **3.2 Geographical origin of virgin olive oil**

The large dataset of VOOs was studied regarding the situations that the antifraud authorities and regulatory bodies face. The PDO "*Riviera Ligure*", some Italian regions and the main countries that produce VOOs were used as examples to prove the potential of the tools to detect the mislabeling of non-PDO oils as PDO VOOs, or the mislabeling of the provenance of VOOs at the regional or national level. With this purpose in mind, several multivariate data analysis techniques, datasets, types of data scaling and cross-validation were evaluated to attain the best classification models for each case study.

After removing 28 extreme samples, the dataset (935 x 342) was analyzed by PCA. The four first principal components, accounting for 63% of total system variability (TSV), showed that samples were distributed in a compact cluster. However, some overlapping sub-clusters due to the harvest year were observed in the score plot of the samples in the space defined by PC2 (13% TSV), PC3 (11% TSV) and PC4 (7% TSV). Taking into account that 70% of the samples were Italian and the rest from countries in the Mediterranean region, seasonal aspects seem to affect all samples in the same way, independently of their geographical origin. Therefore, in the modeling for the authentication of agricultural food products, it is important to include chemical data of several harvests in order to obtain general classification models that allow for the seasonal variability.

Quality Assessment of Olive Oil by <sup>1</sup>

**3.2.1 PDO virgin olive oils** 

H-NMR Fingerprinting 193

Under the PDO of *Riviera Ligure*, only extra virgin olive oils produced in Liguria (Italy) that fulfill the PDO requirements related to olive varieties, farming practices, oil extraction procedures, bottling and labeling (Dossier Number: IT/PDO/0017/1540, Official Journal L22 24.01.1997) can be marketed. The 1H-NMR dataset of VOOs from different geographical origins and PDOs was studied to create a classification model that differentiates between

Univariate data analysis (ANOVA, Fisher index and Box-Whisker plots) disclosed that any single variable could distinguish between Ligurian (belonging to the PDO *Riviera Ligure*) and non-Ligurian (not belonging to the PDO) samples. So, it was necessary to apply supervised pattern recognition methods to build classification models that can distinguish VOOs of this PDO from the rest. Several multivariate approaches (LDA and PLS-DA) were tested using balanced or unbalanced data sets, different cross-validation methods (LOO and 3-fold CV), different data scaling techniques (auto-scaling and Pareto-scaling) to find the best approach for the authenticity and traceability of PDO olive oils (Rosa M. Alonso-Salces et al., 2010b; Rosa M. Alonso-Salces et al., 2011b). Table 2 summarized the results of the best classification models achieved. Both supervised pattern recognition techniques performed better if using a balanced training-test set than an unbalanced data set (Rosa M. Alonso-Salces et al., 2010b; Rosa M. Alonso-Salces et al., 2011b). PLS-DA outperformed LDA. LDA achieved classifications of around 85% of hits for both categories. PLS-DA provided a model with 5 PLS components and the boundary at 0.540, that achieved slightly better results for the Liguria class (prediction ability in the cross-validation, 86-88%; classification ability of the final model, 92%; and prediction ability of the final model in the external validation, 88%) than for the non-Liguria VOOs (86-87%, 90% and 86% respectively). These results, together with the facts that in the cross-validation the recognition ability was higher but close to the prediction ability and the classification ability of the final model was also higher but close to prediction ability in the external validation, disclosed that the model achieved was feasible and not random, as well as

Regarding the most important NMR variables on the classification models provided by these pattern recognition techniques, the variables selected in LDA were among the variables that presented the highest weighted regression coefficients (Esbensen et al., 2002) in the PLS-DA models (the larger the regression coefficient, the higher the influence of the variable on the PLS model). Thus, both pattern recognition techniques arrived at consistent results, and provided information about the most important features for the characterization of PDO *Riviera Ligure* VOOs. In this sense, the variables selected for the LDA model were five NMR buckets centered at the following chemical shifts: 6.61 ppm; 5.11 or 5.09 ppm; 4.57 ppm; 4.05 ppm; and 0.33 ppm. These buckets correspond to signals of the following VOO components: phenolic compounds and unsaturated alcohols, which present characteristic resonances in the spectral region 6-7.5 ppm (Owen et al., 2000) and 4.5-5 ppm respectively; sn-1,2-diglycerides (5.09-5.11 ppm) and sn-1,3-diglycerides (4.05 ppm), due to their CH glycerol protons; and cycloartenol (0.33 ppm), to the methylene proton of its cyclopropanoic

In the PLS-DA models, the variables that presented the highest weighted regression coefficients were: 6.85-6.83 ppm, 6.75 ppm, 6.67 ppm, 6.59 ppm, and 6.23 ppm belonged to

VOOs belonging to the PDO *Riviera Ligure* and those not belonging to this PDO.

being well-represented by the samples in the dataset.

ring (Sacchi et al., 1996).


a Signal multiplicity: s, singlet; d, doublet; t, triplet; m, multiplet.

Table 1. Chemical shift assignments of 1H-NMR signals of the main components in VOOs.

#### **3.2.1 PDO virgin olive oils**

192 Olive Oil – Constituents, Quality, Health Properties and Bioconversions

**shift (ppm) Multiplicitya Functional group Attribution** 

1 0.318 d -C*H*2- (cyclopropanic ring) cycloartenol

3 0.543 d -C*H*2- (cyclopropanic ring) cycloartenol 4 0.669 s -C*H*3 (C18-steroid group) -sitosterol 5 0.687 s -C*H*3 (C18-steroid group) stigmasterol

0.87 ppm)

0.87 ppm)

8 0.960 t -C*H*3 (acyl group) linolenic (or -3)

11 1.243 -(C*H*2)n- (acyl group) saturated (palmitic,

13 1.288 -(C*H*2)n- (acyl group) linoleic and linolenic

7 0.866 t -C*H*3 (acyl group) saturated, oleic (or -9)

and linoleic (or -6)

stearic)

6 0.740 t -C*H*3 (13C satellite of signal at

9 0.987 t -C*H*3 (13C satellite of signal at

14 1.51-1.65 -OCO-CH2-C*H*2- (acyl group)

16 1.96-2.07 -C*H*2-CH=CH- (acyl group) 17 2.26-2.32 m -OCO-C*H*2- (acyl group) 18 2.72-2.82 =CH-C*H*2-CH= (acyl group)

12 1.256 -(C*H*2)n- (acyl group) oleic

15 1.662 s -C*H*<sup>3</sup> squalene

19 2.754 t =CH-C*H*2-CH= (acyl group) linoleic 20 2.789 t =CH-C*H*2-CH= (acyl group) linolenic

21 3.69-3.73 d -C*H*2OH (glyceryl group) *sn* 1,2-diglycerides

26 5.05-5.15 m >C*H*OCOR (glyceryl group) *sn* 1,2-diglycerides

Table 1. Chemical shift assignments of 1H-NMR signals of the main components in VOOs.

27 5.22-5.28 m >C*H*OCOR (glyceryl group) triglycerides

28 5.28-5.38 m -C*H*=C*H*- (acyl group) a Signal multiplicity: s, singlet; d, doublet; t, triplet; m, multiplet.

22 4.09-4.32 -C*H*2OCOR (glyceryl group) triglycerides 23 4.571 d terpene 24 4.648 s terpene 25 4.699 s terpene

10 1.19-1.37 -(C*H*2)n- (acyl group)

**# Chemical** 

2 0.527 s

Under the PDO of *Riviera Ligure*, only extra virgin olive oils produced in Liguria (Italy) that fulfill the PDO requirements related to olive varieties, farming practices, oil extraction procedures, bottling and labeling (Dossier Number: IT/PDO/0017/1540, Official Journal L22 24.01.1997) can be marketed. The 1H-NMR dataset of VOOs from different geographical origins and PDOs was studied to create a classification model that differentiates between VOOs belonging to the PDO *Riviera Ligure* and those not belonging to this PDO.

Univariate data analysis (ANOVA, Fisher index and Box-Whisker plots) disclosed that any single variable could distinguish between Ligurian (belonging to the PDO *Riviera Ligure*) and non-Ligurian (not belonging to the PDO) samples. So, it was necessary to apply supervised pattern recognition methods to build classification models that can distinguish VOOs of this PDO from the rest. Several multivariate approaches (LDA and PLS-DA) were tested using balanced or unbalanced data sets, different cross-validation methods (LOO and 3-fold CV), different data scaling techniques (auto-scaling and Pareto-scaling) to find the best approach for the authenticity and traceability of PDO olive oils (Rosa M. Alonso-Salces et al., 2010b; Rosa M. Alonso-Salces et al., 2011b). Table 2 summarized the results of the best classification models achieved. Both supervised pattern recognition techniques performed better if using a balanced training-test set than an unbalanced data set (Rosa M. Alonso-Salces et al., 2010b; Rosa M. Alonso-Salces et al., 2011b). PLS-DA outperformed LDA. LDA achieved classifications of around 85% of hits for both categories. PLS-DA provided a model with 5 PLS components and the boundary at 0.540, that achieved slightly better results for the Liguria class (prediction ability in the cross-validation, 86-88%; classification ability of the final model, 92%; and prediction ability of the final model in the external validation, 88%) than for the non-Liguria VOOs (86-87%, 90% and 86% respectively). These results, together with the facts that in the cross-validation the recognition ability was higher but close to the prediction ability and the classification ability of the final model was also higher but close to prediction ability in the external validation, disclosed that the model achieved was feasible and not random, as well as being well-represented by the samples in the dataset.

Regarding the most important NMR variables on the classification models provided by these pattern recognition techniques, the variables selected in LDA were among the variables that presented the highest weighted regression coefficients (Esbensen et al., 2002) in the PLS-DA models (the larger the regression coefficient, the higher the influence of the variable on the PLS model). Thus, both pattern recognition techniques arrived at consistent results, and provided information about the most important features for the characterization of PDO *Riviera Ligure* VOOs. In this sense, the variables selected for the LDA model were five NMR buckets centered at the following chemical shifts: 6.61 ppm; 5.11 or 5.09 ppm; 4.57 ppm; 4.05 ppm; and 0.33 ppm. These buckets correspond to signals of the following VOO components: phenolic compounds and unsaturated alcohols, which present characteristic resonances in the spectral region 6-7.5 ppm (Owen et al., 2000) and 4.5-5 ppm respectively; sn-1,2-diglycerides (5.09-5.11 ppm) and sn-1,3-diglycerides (4.05 ppm), due to their CH glycerol protons; and cycloartenol (0.33 ppm), to the methylene proton of its cyclopropanoic ring (Sacchi et al., 1996).

In the PLS-DA models, the variables that presented the highest weighted regression coefficients were: 6.85-6.83 ppm, 6.75 ppm, 6.67 ppm, 6.59 ppm, and 6.23 ppm belonged to

Quality Assessment of Olive Oil by <sup>1</sup>

**Binary modelb** *<sup>N</sup>prior* 

Class codes: "Region", 1; "non-Region, 0.

non-Calabria: 3 PLS components, boundary at 0.445.

H-NMR Fingerprinting 195

**Classification** *<sup>N</sup>prior* 

*prob* **% Prediction** 

**Origin Cross-validation Model External Validation** 

Umbria *35 0.45* 71.4 82.9 *12 0.014* 50.0 Non-Umbria *43 0.55* 74.4 79.1 *845 0.986* 74.8 Sicily *54 0.47* 92.6 98.1 *24 0.029* 87.5 Non-Sicily *62 0.53* 85.5 88.7 *795 0.971* 85.8 Puglia *47 0.42* 68.1 72.3 *22 0.027* 81.8 Non-Puglia *64 0.58* 62.5 71.9 *802 0.973* 65.1 Lazio *40 0.49* 80.0 97.5 *19 0.022* 73.7 Non-Lazio *41 0.51* 68.3 90.2 *835 0.978* 69.3 Garda *36 0.46* 72.2 91.7 *13 0.015* 69.2 Non-Garda *43 0.54* 74.4 90.7 *843 0.985* 80.1 Campania *21 0.43* 71.4 81.0 *7 0.008* 57.1 Non-Campania *28 0.57* 64.3 78.6 *879 0.992* 62.9 Calabria *17 0.38* 70.6 94.1 *5 0.006* 60.0 Non-Calabria *28 0.62* 85.7 96.4 *885 0.994* 79.9

a See abbreviations: Table 2; Models obtained by PLS-DA using autoscaling, LOO and The Unscrambler;

b Binary models: Umbria *vs.* non-Umbria: 2 PLS components, boundary at 0.525; Sicily *vs.* non-Sicily: 3 PLS components, boundary at 0.460; Puglia *vs.* non-Puglia: 2 PLS components, boundary at 0.4435; Lazio *vs.* non-Lazio: 4 PLS components, boundary at 0.515; Garda *vs.* non-Garda: 3 PLS components, boundary at 0.555; Campania *vs.* non-Campania: 2 PLS components, boundary at 0.430; Calabria *vs.*

Table 3. Classification results obtained by supervised pattern recognition techniques for the

The model obtained to authenticate VOOs from Sicily recognized 98% of the Sicilian oils and 89% of the non-Sicilian ones and managed to correctly predict in the cross-validation step 93% and 86% of Sicilian and non-Sicilian oils respectively. Since this model achieved similar predictions in the external validation (higher than 85% of hits for both categories) to those in the modeling step, it can be considered stable and robust. In contrast, the models created for other regions such as Lazio, Garda and Calabria, were not so satisfactory: although the classification abilities were close to 90% of correct hits or even higher, the prediction abilities in the cross-validation were from 10 to 24% lower, which meant that the classification results were very dependent on the samples included in the training set in the modeling step. This also occurred for Umbria and Campania, but the models achieved about 80% of correct classification for the training set, and predictions on the test set were more than 10% lower, except for the oils belonging to the non-Umbria category (5% less). The external validation

authentication of VOO from certain Italian regions using 1H-NMR spectral data.a

*prob* **% Prediction %** 

signals of phenolic compounds; 5.15-5.07 ppm were due to the CH glycerol protons of *sn*-1,2-diglycerides; 4.99 ppm to unsaturated alcohols; 4.71 ppm, 4.65 ppm and 4.57 ppm, to terpenes; 2.79 ppm, to diallylic proton of linolenic acyl group; 1.29 ppm, to methylene proton of linoleic and linolenic acyl group; and 0.33 ppm, to cycloartenol.


a Abbreviations: N, number of samples; prior prob, prior probability; Lig, Liguria; Non-Lig, Non-Liguria; LDA, linear discriminant analysis; PLS-DA, partial least square discriminant analysis; Class codes: Liguria, 1; non-Liguria, 0.

b Statistica.; c The Unscrambler.; d SIMCA-P.

Table 2. Classification results obtained by supervised pattern recognition techniques for the authentication of VOO of the PDO *Riviera Ligure* using 1H-NMR spectral data (balanced data set).a

#### **3.2.2 Virgin olive oils from different regions**

The large sample set of VOOs available was also studied from the point of view of the authentication of VOOs at the regional level, in particular, VOOs produced in certain Italian regions. The regions selected were those best represented in the dataset: Umbria (which is also a registered PDO: PDO *Umbria*), Sicily (6 PDOs: *Monte Etna*, *Val di Mazara*, *Valli Trapanesi*, *Valle del Belice*, *Valdemone* and *Monti Iblei*), Puglia (4 PDOs: *Terra d'Otranto*, *Collina di Brindisi*, *Dauno* and *Terra di Bari*), Lazio (3 PDOs: *Tuscia*, *Canino* and *Sabina*), Garda (3 PDOs: *Garda*, *Laghi Lombardi* and *Veneto Valpolicella, Veneto Euganei e Berici, Veneto del Grappa*), Campania (3 PDOs: *Peninsola Sorrentina*, *Colline Salernitane* and *Cilento*) and Calabria (3 PDOs: *Lametia*, *Alto Crotonese* and *Bruzio*). The binary classification models created for these regions were developed using an auto-scaled balanced training-test set by PLS-DA and LOO cross-validation (Rosa M. Alonso-Salces et al., 2010b). The final models were also evaluated by external validation. The results are summarized in Table 3.

signals of phenolic compounds; 5.15-5.07 ppm were due to the CH glycerol protons of *sn*-1,2-diglycerides; 4.99 ppm to unsaturated alcohols; 4.71 ppm, 4.65 ppm and 4.57 ppm, to terpenes; 2.79 ppm, to diallylic proton of linolenic acyl group; 1.29 ppm, to methylene

a Abbreviations: N, number of samples; prior prob, prior probability; Lig, Liguria; Non-Lig, Non-Liguria; LDA, linear discriminant analysis; PLS-DA, partial least square discriminant analysis; Class

authentication of VOO of the PDO *Riviera Ligure* using 1H-NMR spectral data

Table 2. Classification results obtained by supervised pattern recognition techniques for the

The large sample set of VOOs available was also studied from the point of view of the authentication of VOOs at the regional level, in particular, VOOs produced in certain Italian regions. The regions selected were those best represented in the dataset: Umbria (which is also a registered PDO: PDO *Umbria*), Sicily (6 PDOs: *Monte Etna*, *Val di Mazara*, *Valli Trapanesi*, *Valle del Belice*, *Valdemone* and *Monti Iblei*), Puglia (4 PDOs: *Terra d'Otranto*, *Collina di Brindisi*, *Dauno* and *Terra di Bari*), Lazio (3 PDOs: *Tuscia*, *Canino* and *Sabina*), Garda (3 PDOs: *Garda*, *Laghi Lombardi* and *Veneto Valpolicella, Veneto Euganei e Berici, Veneto del Grappa*), Campania (3 PDOs: *Peninsola Sorrentina*, *Colline Salernitane* and *Cilento*) and Calabria (3 PDOs: *Lametia*, *Alto Crotonese* and *Bruzio*). The binary classification models created for these regions were developed using an auto-scaled balanced training-test set by PLS-DA and LOO cross-validation (Rosa M. Alonso-Salces et al., 2010b). The final models

were also evaluated by external validation. The results are summarized in Table 3.

codes: Liguria, 1; non-Liguria, 0.

(balanced data set).a

b Statistica.; c The Unscrambler.; d SIMCA-P.

**3.2.2 Virgin olive oils from different regions** 

proton of linoleic and linolenic acyl group; and 0.33 ppm, to cycloartenol.


a See abbreviations: Table 2; Models obtained by PLS-DA using autoscaling, LOO and The Unscrambler; Class codes: "Region", 1; "non-Region, 0.

b Binary models: Umbria *vs.* non-Umbria: 2 PLS components, boundary at 0.525; Sicily *vs.* non-Sicily: 3 PLS components, boundary at 0.460; Puglia *vs.* non-Puglia: 2 PLS components, boundary at 0.4435; Lazio *vs.* non-Lazio: 4 PLS components, boundary at 0.515; Garda *vs.* non-Garda: 3 PLS components, boundary at 0.555; Campania *vs.* non-Campania: 2 PLS components, boundary at 0.430; Calabria *vs.* non-Calabria: 3 PLS components, boundary at 0.445.

Table 3. Classification results obtained by supervised pattern recognition techniques for the authentication of VOO from certain Italian regions using 1H-NMR spectral data.a

The model obtained to authenticate VOOs from Sicily recognized 98% of the Sicilian oils and 89% of the non-Sicilian ones and managed to correctly predict in the cross-validation step 93% and 86% of Sicilian and non-Sicilian oils respectively. Since this model achieved similar predictions in the external validation (higher than 85% of hits for both categories) to those in the modeling step, it can be considered stable and robust. In contrast, the models created for other regions such as Lazio, Garda and Calabria, were not so satisfactory: although the classification abilities were close to 90% of correct hits or even higher, the prediction abilities in the cross-validation were from 10 to 24% lower, which meant that the classification results were very dependent on the samples included in the training set in the modeling step. This also occurred for Umbria and Campania, but the models achieved about 80% of correct classification for the training set, and predictions on the test set were more than 10% lower, except for the oils belonging to the non-Umbria category (5% less). The external validation

Quality Assessment of Olive Oil by <sup>1</sup>

dependent on the samples included in the training set.

**Binary modelb** *<sup>N</sup>prior* 

Class codes: "Country", 1; "non-Country, 0.

using 1H-NMR spectral data.a

evermore apparent.

H-NMR Fingerprinting 197

were actually produced in another, are actual events that the antifraud authorities have to deal with regularly. The need for chemical approaches to detect these fraudulent activities is

The 1H-NMR data of the VOOs from the main olive oil producing countries, i.e. Spain, Italy and Greece, were analyzed by multivariate techniques with the purpose of creating classification models to distinguish the geographical origin of VOOs from these three countries (Rosa M. Alonso-Salces et al., 2010b). PLS-DA, using LOO cross-validation, was applied to the autoscaled data to provide binary classification models (country *vs.* noncountry), which were also evaluated by external validation (Table 4). The model 'Greece *vs.* non-Greece' distinguishes Greek VOOs from all the rest of the VOOs; it classified properly more than 97% of the samples of both categories, Greece and non-Greece, and predicted correctly more than 90% of the samples in the test set of the cross-validation, as well as in the external validation. The binary models for Italy and Spain presented classification abilities of 89% for the Italian oils and the Spanish oils, 84% for the non-Italy category and 85% for the non-Spain category. The prediction abilities in the cross-validation for the model for Spain were ca. 80% of hits for both classes; whereas the predictions in the external validation were considerably different, for the Spanish VOOs it was overoptimistic (92%), and for the non-Spanish VOOs it was considerably low (67%). In the model for Spain, the variability of the non-Spain category was under-represented in the training-test sets. As a result, this model did not provide good predictions for this category in the external set. The model for Italy provided prediction abilities in the cross-validation of ca. 76% for both classes; and in external validation, close to this value. So, these predictions were substantially lower than the recognition ability of the model, indicating that the model was

**Origin Cross-validation Model External Validation** 

a See abbreviations: Table 2; Models obtained by PLS-DA using autoscaling, LOO and The Unscrambler;

b Binary models: Italy *vs* non-Italy: 4 PLS components, boundary at 0.4020; Spain *vs* non-Spain: 3 PLS components, boundary at 0.3563; Greece *vs* non-Greece: 5 PLS components, boundary at 0.4725.

Table 4. Classification results obtained by supervised pattern recognition techniques for the authentication of VOO from the main producing countries, i.e. Italy, Spain and Greece,

Italy *72 0.35* 75.0 88.9 *568 0.78* 75.7 Non-Italy *135 0.65* 77.0 84.4 *160 0.22* 71.9 Spain *71 0.34* 78.9 88.7 *70 0.10* 92.9 Non-Spain *136 0.66* 80.9 85.3 *658 0.90* 67.2 Greece *64 0.31* 92.2 98.4 *31 0.04* 96.8 Non-Greece *143 0.69* 93.7 97.9 *697 0.96* 90.0

*prob* **% Prediction % Classification** *<sup>N</sup>prior* 

*prob* **% Prediction** 

of some models (only 50% of Umbria, 57% of Campania, and 60% of Calabria VOOs were correctly predicted) confirmed that the classes were not well represented in the modeling step. Puglian VOOs, as well as non-Garda VOOs, were much better predicted in the external data set (82% of hits) than in the cross-validation (68% and 72% of hits respectively). This was probably due to the way samples were divided into the training-test set and the external set: the PCA scores of all the VOOs were regarded to select samples from the whole cloud of points including the borders. This procedure assured that the training-test set was representative of all the samples (at least of the 3 harvests studied), however the predictions on the external set could be overoptimistic.

Regarding the most influential variables, i.e. those with the highest weighted regression coefficients, on the binary PLS-DA models achieved for each region are the following. The signals due to cycloartenol (0.31-0.33 ppm) and *sn*-1,2-diglycerides (5.07-5.15-ppm) were important for all models except for Garda, as well as the resonances in the phenolic region at 6.73-6.79 ppm, which only did not influence the model for Sicily. The acyl group methylene protons of saturated fatty acids (1.23 ppm), 13C satellite of signal at 4.09-4.32 ppm ( methylene protons of the glyceryl group of triglycerides) at 3.97 ppm and the signal at 5.57 ppm were important specifically for the Umbria model; the signals at 0.53 ppm and 0.79 ppm, for the Sicily model; the methylic proton of the C18-steroid group of -sitosterol (0.67 ppm) and the terpene signal at 4.57-4.59 ppm, for the Puglia model; the signal of the cycloartenol at 0.55 ppm, 13C satellite of signal at 2.26-2.32 ppm (-methylene protons of the acyl group) at 2.15 ppm, the glycerol proton of *sn*-1,2-diglycerides (3.71 ppm) and signals at 6.19 ppm and 6.15 ppm in the phenolic region, for the Lazio model; signals in the region 1.35-1.43 ppm, 2.35-2.39 ppm, and 4.33-4.35 ppm, the -methylene protons of the acyl group (2.29 ppm and 2.33 ppm), the signal at 3.75 ppm, the -methylene protons of the glyceryl group of triglycerides (4.27 ppm) and the signal at 6.15 ppm in the phenolic region for the Campania model; and the signal at 5.93 ppm for the Calabria model. The glyceryl protons of *sn*-1,3-diglycerides (4.05-4.07 ppm) and triglycerides (5.25 ppm, 5.29 ppm) were influent for the models of Umbria, Lazio, Umbria and Campania respectively; signals in the phenolic region at 6.25-6.29 ppm for the models of Puglia and Calabria; signals in the phenolic region at 6.63-6.65 ppm and 6.69-6.71 ppm for the models of Umbria and Campania ; signals in the phenolic region at 6.45-6.47 ppm for the models of Umbria and Garda.

These results disclosed that 1H-NMR spectra of VOOs contained information related to the region of provenance of the oil, nonetheless further studies should be carried out with a considerably larger sample set for each region, and even for each of their PDOs, in order to guarantee the detection of counterfeit VOOs. In this regard, Sicily, which is an island at the southernmost part of Italy, produces an olive oil which is markedly influenced by pedoclimatic factors, in accordance with its geographical position. It is therefore coherent that the VOO produced on this island presents a characteristic chemical composition that allows one to distinguish it from all other VOO coming from different geographical regions.

#### **3.2.3 Virgin olive oils from the main producing countries: Spain, Italy and Greece**

The adulteration of VOOs from a certain country with VOOs produced in another country at a lower cost, or the false labeling of the VOOs as coming from a certain country when they

of some models (only 50% of Umbria, 57% of Campania, and 60% of Calabria VOOs were correctly predicted) confirmed that the classes were not well represented in the modeling step. Puglian VOOs, as well as non-Garda VOOs, were much better predicted in the external data set (82% of hits) than in the cross-validation (68% and 72% of hits respectively). This was probably due to the way samples were divided into the training-test set and the external set: the PCA scores of all the VOOs were regarded to select samples from the whole cloud of points including the borders. This procedure assured that the training-test set was representative of all the samples (at least of the 3 harvests studied), however the predictions

Regarding the most influential variables, i.e. those with the highest weighted regression coefficients, on the binary PLS-DA models achieved for each region are the following. The signals due to cycloartenol (0.31-0.33 ppm) and *sn*-1,2-diglycerides (5.07-5.15-ppm) were important for all models except for Garda, as well as the resonances in the phenolic region at 6.73-6.79 ppm, which only did not influence the model for Sicily. The acyl group methylene protons of saturated fatty acids (1.23 ppm), 13C satellite of signal at 4.09-4.32 ppm ( methylene protons of the glyceryl group of triglycerides) at 3.97 ppm and the signal at 5.57 ppm were important specifically for the Umbria model; the signals at 0.53 ppm and 0.79 ppm, for the Sicily model; the methylic proton of the C18-steroid group of -sitosterol (0.67 ppm) and the terpene signal at 4.57-4.59 ppm, for the Puglia model; the signal of the cycloartenol at 0.55 ppm, 13C satellite of signal at 2.26-2.32 ppm (-methylene protons of the acyl group) at 2.15 ppm, the glycerol proton of *sn*-1,2-diglycerides (3.71 ppm) and signals at 6.19 ppm and 6.15 ppm in the phenolic region, for the Lazio model; signals in the region 1.35-1.43 ppm, 2.35-2.39 ppm, and 4.33-4.35 ppm, the -methylene protons of the acyl group (2.29 ppm and 2.33 ppm), the signal at 3.75 ppm, the -methylene protons of the glyceryl group of triglycerides (4.27 ppm) and the signal at 6.15 ppm in the phenolic region for the Campania model; and the signal at 5.93 ppm for the Calabria model. The glyceryl protons of *sn*-1,3-diglycerides (4.05-4.07 ppm) and triglycerides (5.25 ppm, 5.29 ppm) were influent for the models of Umbria, Lazio, Umbria and Campania respectively; signals in the phenolic region at 6.25-6.29 ppm for the models of Puglia and Calabria; signals in the phenolic region at 6.63-6.65 ppm and 6.69-6.71 ppm for the models of Umbria and Campania ; signals in the

phenolic region at 6.45-6.47 ppm for the models of Umbria and Garda.

These results disclosed that 1H-NMR spectra of VOOs contained information related to the region of provenance of the oil, nonetheless further studies should be carried out with a considerably larger sample set for each region, and even for each of their PDOs, in order to guarantee the detection of counterfeit VOOs. In this regard, Sicily, which is an island at the southernmost part of Italy, produces an olive oil which is markedly influenced by pedoclimatic factors, in accordance with its geographical position. It is therefore coherent that the VOO produced on this island presents a characteristic chemical composition that allows one to distinguish it from all other VOO coming from different geographical regions.

**3.2.3 Virgin olive oils from the main producing countries: Spain, Italy and Greece** 

The adulteration of VOOs from a certain country with VOOs produced in another country at a lower cost, or the false labeling of the VOOs as coming from a certain country when they

on the external set could be overoptimistic.

were actually produced in another, are actual events that the antifraud authorities have to deal with regularly. The need for chemical approaches to detect these fraudulent activities is evermore apparent.

The 1H-NMR data of the VOOs from the main olive oil producing countries, i.e. Spain, Italy and Greece, were analyzed by multivariate techniques with the purpose of creating classification models to distinguish the geographical origin of VOOs from these three countries (Rosa M. Alonso-Salces et al., 2010b). PLS-DA, using LOO cross-validation, was applied to the autoscaled data to provide binary classification models (country *vs.* noncountry), which were also evaluated by external validation (Table 4). The model 'Greece *vs.* non-Greece' distinguishes Greek VOOs from all the rest of the VOOs; it classified properly more than 97% of the samples of both categories, Greece and non-Greece, and predicted correctly more than 90% of the samples in the test set of the cross-validation, as well as in the external validation. The binary models for Italy and Spain presented classification abilities of 89% for the Italian oils and the Spanish oils, 84% for the non-Italy category and 85% for the non-Spain category. The prediction abilities in the cross-validation for the model for Spain were ca. 80% of hits for both classes; whereas the predictions in the external validation were considerably different, for the Spanish VOOs it was overoptimistic (92%), and for the non-Spanish VOOs it was considerably low (67%). In the model for Spain, the variability of the non-Spain category was under-represented in the training-test sets. As a result, this model did not provide good predictions for this category in the external set. The model for Italy provided prediction abilities in the cross-validation of ca. 76% for both classes; and in external validation, close to this value. So, these predictions were substantially lower than the recognition ability of the model, indicating that the model was dependent on the samples included in the training set.


a See abbreviations: Table 2; Models obtained by PLS-DA using autoscaling, LOO and The Unscrambler; Class codes: "Country", 1; "non-Country, 0.

b Binary models: Italy *vs* non-Italy: 4 PLS components, boundary at 0.4020; Spain *vs* non-Spain: 3 PLS components, boundary at 0.3563; Greece *vs* non-Greece: 5 PLS components, boundary at 0.4725.

Table 4. Classification results obtained by supervised pattern recognition techniques for the authentication of VOO from the main producing countries, i.e. Italy, Spain and Greece, using 1H-NMR spectral data.a

Quality Assessment of Olive Oil by <sup>1</sup>

altering its organoleptic properties.

43

39





0.0

PC2 (12% of

total variability)

0.5

1.0

1.5

2.0

2.5

41

30

40

37 36

23

29

26

28

34 33

17

14

22

25

19

24


Fig. 2. PCA score plot of the samples used to study the stability of VOO on the space defined

by the two first principal components. Samples are numbered according to the time

16

18

21

<sup>12</sup> <sup>11</sup>

13

9

6

8 7

1

5 4

0

2

0-12 months 13-24 months 25-36 months 37-44 months

3

10

15

35

27

(months) that they had been at r.t. in the dark before analysis.

31

H-NMR Fingerprinting 199

The most influential features, i.e. buckets of 1H-NMR spectra, on PC1 are those with the highest loadings in absolute value, and are shown in Table 5. Some of these chemical shifts correspond to 1H-NMR signals of compounds involved in the hydrolytic and oxidative degradation of VOO. During the oxidation process, hydroperoxides (primary oxidation compounds) are produced (Guillen & Ruiz, 2001, 2006) , which may degrade into secondary oxidation products such as aldehydes, ketones, lactones, alcohols, acids, etc. The oxidation of edible oils is a matter of major concern also from a safety point of view because some oxidation products such as aldehydes are toxic (Guillen & Ruiz, 2001, 2006). Furthermore, several saturated and unsaturated aldehydes have been found to be responsible for rancid sensory defect in VOO (Morales et al., 2005), as well as for off-odours (Kalua et al., 2007),

The most influential variables, i.e. those with the highest weighted regression coefficients, on the binary PLS-DA models obtained for each country are due to signals in the phenolic regions at 6.45-6.47 ppm and 6.83-6.85 ppm, which were important for the three models (Rosa M. Alonso-Salces et al., 2010b). In contrast, the model for Spain was particularly influenced by the methylic proton of the C18-steroid group of -sitosterol (0.67 ppm), the methylene protons of the acyl group (1.59 ppm, 1.67 ppm), the allylic protons of the acyl group (1.99-2.07 ppm), the diallylic protons of the acyl group of linoleic (2.73-2.77 ppm) and linolenic (2.77-2.81 ppm), the glycerol proton of *sn*-1,2-diglycerides (3.71 ppm), *sn*-1,3 diglycerides (4.05-4.07 ppm) and triglycerides (5.25 ppm, 5.29 ppm), the olefinic protons of the acyl groups (5.37 ppm), the signals in the phenolic region at 6.37 ppm, 6.61 ppm and 6.71 ppm, and the signals at 0.53 ppm, 1.75-1.77 ppm, 2.35 ppm. Among the most important variables, those who only affected the model for Greece were the methylic proton of the linolenic acyl group (0.97 ppm), and the terpene signal at 4.55-4.57 ppm, and the signals at 0.77 ppm and 3.81 ppm. The resonances of cycloartenol (0.31-0.33 ppm and 0.55 ppm) and phenolic compounds at 6.23 ppm and 6.27 ppm were important for the models of Italy and Greece.

These results show that 1H-NMR fingerprinting of VOOs can be a useful tool to assure authenticity and traceability of VOOs at the national level. From this study, a stable model was achieved to distinguish Greek VOOs from oils from other countries. However, for Italian and Spanish VOOs further studies should be performed with a larger balanced data set, in which all categories will be well represented, to obtain robust models. In the present data set, Spain was clearly under-represented, being the main producer (50% of EU production of olive oil); and Italy, even though it was quite well-represented, the number of samples were very unbalanced regarding the other countries, and so few Italian samples were used in the modeling step. The classification results might therefore be very dependent on the samples in the training-test set.

#### **3.3 Stability of virgin olive oil**

Regarding the importance of oil stability on its quality and nutritional properties, the stability of VOO was studied over a period of 43 months in the dark at room temperature by 1H-NMR. The high stability of VOO is mainly due to its relatively low degree of fatty acid unsaturation and to the antioxidant activity of some of the unsaponifiable components. For instance, the oxidative susceptibility of olive oil is related to the antioxidant activity of tocopherol, which also showed a synergistic effect in association with some phenolic compounds with significant activity (Deiana et al., 2002).

In contrast with previous studies on the oxidative stability of edible oils, which were performed at high temperatures (Frankel, 2010); we studied VOO stability at r.t. (Alonso-Salces et al., 2011a). The 1H-NMR spectra of 40 VOO samples in the spectral region 0-10 ppm (492 buckets) were analyzed by PCA. The samples represented in the PCA score plot of the first two principal components (Fig. 2) are distributed in the direction of the first principal component (PC1) in clusters, partially overlapped, according to the length of time they had been at r.t. So, PC1, accounting for 15% TSV, contains information related to the change and evolution of the chemical composition of VOO during storage in the dark at r.t., and hence, about the degradation of olive oil under these conditions.

The most influential variables, i.e. those with the highest weighted regression coefficients, on the binary PLS-DA models obtained for each country are due to signals in the phenolic regions at 6.45-6.47 ppm and 6.83-6.85 ppm, which were important for the three models (Rosa M. Alonso-Salces et al., 2010b). In contrast, the model for Spain was particularly influenced by the methylic proton of the C18-steroid group of -sitosterol (0.67 ppm), the methylene protons of the acyl group (1.59 ppm, 1.67 ppm), the allylic protons of the acyl group (1.99-2.07 ppm), the diallylic protons of the acyl group of linoleic (2.73-2.77 ppm) and linolenic (2.77-2.81 ppm), the glycerol proton of *sn*-1,2-diglycerides (3.71 ppm), *sn*-1,3 diglycerides (4.05-4.07 ppm) and triglycerides (5.25 ppm, 5.29 ppm), the olefinic protons of the acyl groups (5.37 ppm), the signals in the phenolic region at 6.37 ppm, 6.61 ppm and 6.71 ppm, and the signals at 0.53 ppm, 1.75-1.77 ppm, 2.35 ppm. Among the most important variables, those who only affected the model for Greece were the methylic proton of the linolenic acyl group (0.97 ppm), and the terpene signal at 4.55-4.57 ppm, and the signals at 0.77 ppm and 3.81 ppm. The resonances of cycloartenol (0.31-0.33 ppm and 0.55 ppm) and phenolic compounds at 6.23 ppm and 6.27 ppm were important for the models of Italy and

These results show that 1H-NMR fingerprinting of VOOs can be a useful tool to assure authenticity and traceability of VOOs at the national level. From this study, a stable model was achieved to distinguish Greek VOOs from oils from other countries. However, for Italian and Spanish VOOs further studies should be performed with a larger balanced data set, in which all categories will be well represented, to obtain robust models. In the present data set, Spain was clearly under-represented, being the main producer (50% of EU production of olive oil); and Italy, even though it was quite well-represented, the number of samples were very unbalanced regarding the other countries, and so few Italian samples were used in the modeling step. The classification results might therefore be very dependent

Regarding the importance of oil stability on its quality and nutritional properties, the stability of VOO was studied over a period of 43 months in the dark at room temperature by 1H-NMR. The high stability of VOO is mainly due to its relatively low degree of fatty acid unsaturation and to the antioxidant activity of some of the unsaponifiable components. For instance, the oxidative susceptibility of olive oil is related to the antioxidant activity of tocopherol, which also showed a synergistic effect in association with some phenolic

In contrast with previous studies on the oxidative stability of edible oils, which were performed at high temperatures (Frankel, 2010); we studied VOO stability at r.t. (Alonso-Salces et al., 2011a). The 1H-NMR spectra of 40 VOO samples in the spectral region 0-10 ppm (492 buckets) were analyzed by PCA. The samples represented in the PCA score plot of the first two principal components (Fig. 2) are distributed in the direction of the first principal component (PC1) in clusters, partially overlapped, according to the length of time they had been at r.t. So, PC1, accounting for 15% TSV, contains information related to the change and evolution of the chemical composition of VOO during storage in the dark at r.t., and hence,

Greece.

on the samples in the training-test set.

compounds with significant activity (Deiana et al., 2002).

about the degradation of olive oil under these conditions.

**3.3 Stability of virgin olive oil** 

The most influential features, i.e. buckets of 1H-NMR spectra, on PC1 are those with the highest loadings in absolute value, and are shown in Table 5. Some of these chemical shifts correspond to 1H-NMR signals of compounds involved in the hydrolytic and oxidative degradation of VOO. During the oxidation process, hydroperoxides (primary oxidation compounds) are produced (Guillen & Ruiz, 2001, 2006) , which may degrade into secondary oxidation products such as aldehydes, ketones, lactones, alcohols, acids, etc. The oxidation of edible oils is a matter of major concern also from a safety point of view because some oxidation products such as aldehydes are toxic (Guillen & Ruiz, 2001, 2006). Furthermore, several saturated and unsaturated aldehydes have been found to be responsible for rancid sensory defect in VOO (Morales et al., 2005), as well as for off-odours (Kalua et al., 2007), altering its organoleptic properties.

Fig. 2. PCA score plot of the samples used to study the stability of VOO on the space defined by the two first principal components. Samples are numbered according to the time (months) that they had been at r.t. in the dark before analysis.

Quality Assessment of Olive Oil by <sup>1</sup>

**4. Conclusion** 

crossvalidation and external validation.

H-NMR Fingerprinting 201

The presence of hydroperoxides in the samples which had been stored at r.t. and protected from light for more than one year was confirmed by the 1H-NMR signals at 8.09-8.19 ppm due to hydroperoxide protons; 6.51-6.57 ppm, 5.95-6.01 ppm and 5.55-5.57 ppm due to protons of the conjugated diene systems; and 4.35-4.37 ppm due to the methine proton of the hydroperoxide group, as observed by other authors (Guillen & Ruiz, 2001). All these signals presented very small intensities in comparison with characteristic VOO signals, indicating that the oxidative degradation was taking place at a very low rate and yield. This was also supported by the fact that characteristic resonances of aldehydes (9.3-9.9 ppm), the main secondary oxidation products, were not detected in the VOO over the 3 and half years of storage at r.t., so the secondary oxidation process had not yet occurred. These facts agree with the recognized high oxidative stability of VOO. Some minor signals at 6.97 ppm and 6.75 ppm were assigned to phenolic compounds (Owen et al., 2003). The decrease or disappearance, respectively, of these signals during storage at r.t. was in agreement with the role that these

substances play as antioxidants during the oxidative degradation process of VOO.

During hydrolytic degradation of olive oil, triglycerides hydrolyze thereby increasing the content of free fatty acids and consequently, the acidity of the oil, which means deterioration in the oil quality. Several resonances indicated the occurrence of hydrolytic degradation. In this sense, slight changes in the intensity of the tryglyceride signals at 5.25 ppm, 4.45 ppm and 4.27 ppm and the -methylene protons of the acyl group (13C satellite of the signal at 2.26-2.32 ppm) at 2.15-2.21 ppm were observed. Moreover, a slight decrease in the intensity of the signals at 2.75-2.79 ppm of the diallylic protons and at 2.03 ppm of the allylic protons of linoleic and linolenic acyl groups, and at 1.25-1.29 ppm of the methylene proton signal of oleic, linoleic and linolenic acyl groups, during storage at r.t., revealed that tryglycerides were degrading. The increase in the intensity of the signal at 4.05-4.09 ppm, assigned to the proton of the glyceryl group of *sn*-1,3-diglycerides, was indicative of the loss of quality and freshness of the VOO (Guillen & Ruiz, 2001). Young, good quality olive oils contain mainly native *sn*-1,2-diacylglyceride and only small amounts of *sn*-1,3-diacylglyceride. The increase in the latter was observed after one year of storage at r.t., which seems to be caused by intermolecular transposition and/or lipolytic phenomena (Sacchi et al., 1996). Moreover, in the samples stored for longer than 18 months, a broad signal also appeared in the region of saturated alcohols at 3.85-3.89 ppm, which can arise from lipolysis (Sacchi et al., 1996).

1H-NMR fingerprinting of olive oil is a valuable analytical tool for the traceability of VOOs

For authentication purposes, 1H-NMR fingerprints of VOOs analyzed by supervised pattern recognition techniques allow the determination of their geographical origin at the national, regional and/or PDO level. PLS-DA afforded the best model to distinguish the PDO *Riviera Ligure* VOOs: 88% of the Ligurian and 86% of non-Ligurian oils were correctly predicted in the external validation. At the regional level, a stable and robust PLS-DA model was obtained to authenticate VOOs from Sicily, predicting well the origin of more than 85% of the samples in the external sample set. At the national level, Greek and non-Greek VOOs were properly classified by PLS DA: >90% of the oils were correctly predicted in the

from different points of view, i.e. food authentication and food quality.


a Signal multiplicity: s, singlet; d, doublet; t, triplet; q, quadruplet; m, multiplet.

Table 5. Stability of VOO: Loadings of the most influential variables on the first principal component, and chemical shift assignments of the 1H-NMR signals.

broad signal -OO*H* (hydroperoxide group) hydroperoxides

system)conjugated hydroperoxides

conjugated diene system) hydroperoxides

conjugated diene system) hydroperoxides

4.09-4.32 ppm, glyceryl group) triglycerides

q >C*H*-OH (glyceryl group) *sn* 1,3-diglycerides

broad signal saturated alcohols

hydroperoxide group) hydroperoxides

**(ppm) Loadings Multiplicitya Functional group Attribution** 

6.97 0.766 s -Ph-*H* (phenolic ring) phenolic compounds 6.75 0.811 d -Ph-*H* (phenolic ring) phenolic compounds

<sup>t</sup>-C*H*=C*H*-C*H*=C*H*- (*cis, trans* diene

>C*H*-OOH (methine proton of

<sup>t</sup>-C*H*=C*H*-C*H*=C*H*- (*cis, trans*

5.25 0.880 m >C*H*OCOR (glyceryl group) triglycerides

4.27 0.770 m -C*H*2OCOR (glyceryl group) triglycerides

3.59 -0.708 broad signal saturated alcohols

2.77 0.839 =CH-C*H*2-CH= (acyl group) linolenic and linoleic

2.03 0.792 -C*H*2-CH=CH- (acyl group) linoleic and linolenic

Table 5. Stability of VOO: Loadings of the most influential variables on the first principal

<sup>m</sup>-OCO-C*H*2- (13C satellite of signal at 2.26-2.32 ppm, acyl group)

0.852 -(C*H*2)n- (acyl group) linoleic and linolenic

2.79 0.924 t =CH-C*H*2-CH= (acyl group) linolenic

2.75 0.870 t =CH-C*H*2-CH= (acyl group) linoleic

1.25 0.784 -(C*H*2)n- (acyl group) oleic

a Signal multiplicity: s, singlet; d, doublet; t, triplet; q, quadruplet; m, multiplet.

component, and chemical shift assignments of the 1H-NMR signals.


4.45 0.726 m -C*H*2OCOR (13C satellite of signal at

**Bucket** 

8.19 8.17 8.15 8.13 8.11 8.09

6.57 6.55 6.53 6.51

6.01 5.99 5.97 5.95

5.57 5.55

4.37 4.35

4.09 4.07 4.05

3.89 3.87 3.85

2.21 2.19 2.17 2.15

1.29 1.27 -0.807 -0.860 -0.823 -0.820 -0.749 -0.716








0.819

The presence of hydroperoxides in the samples which had been stored at r.t. and protected from light for more than one year was confirmed by the 1H-NMR signals at 8.09-8.19 ppm due to hydroperoxide protons; 6.51-6.57 ppm, 5.95-6.01 ppm and 5.55-5.57 ppm due to protons of the conjugated diene systems; and 4.35-4.37 ppm due to the methine proton of the hydroperoxide group, as observed by other authors (Guillen & Ruiz, 2001). All these signals presented very small intensities in comparison with characteristic VOO signals, indicating that the oxidative degradation was taking place at a very low rate and yield. This was also supported by the fact that characteristic resonances of aldehydes (9.3-9.9 ppm), the main secondary oxidation products, were not detected in the VOO over the 3 and half years of storage at r.t., so the secondary oxidation process had not yet occurred. These facts agree with the recognized high oxidative stability of VOO. Some minor signals at 6.97 ppm and 6.75 ppm were assigned to phenolic compounds (Owen et al., 2003). The decrease or disappearance, respectively, of these signals during storage at r.t. was in agreement with the role that these substances play as antioxidants during the oxidative degradation process of VOO.

During hydrolytic degradation of olive oil, triglycerides hydrolyze thereby increasing the content of free fatty acids and consequently, the acidity of the oil, which means deterioration in the oil quality. Several resonances indicated the occurrence of hydrolytic degradation. In this sense, slight changes in the intensity of the tryglyceride signals at 5.25 ppm, 4.45 ppm and 4.27 ppm and the -methylene protons of the acyl group (13C satellite of the signal at 2.26-2.32 ppm) at 2.15-2.21 ppm were observed. Moreover, a slight decrease in the intensity of the signals at 2.75-2.79 ppm of the diallylic protons and at 2.03 ppm of the allylic protons of linoleic and linolenic acyl groups, and at 1.25-1.29 ppm of the methylene proton signal of oleic, linoleic and linolenic acyl groups, during storage at r.t., revealed that tryglycerides were degrading. The increase in the intensity of the signal at 4.05-4.09 ppm, assigned to the proton of the glyceryl group of *sn*-1,3-diglycerides, was indicative of the loss of quality and freshness of the VOO (Guillen & Ruiz, 2001). Young, good quality olive oils contain mainly native *sn*-1,2-diacylglyceride and only small amounts of *sn*-1,3-diacylglyceride. The increase in the latter was observed after one year of storage at r.t., which seems to be caused by intermolecular transposition and/or lipolytic phenomena (Sacchi et al., 1996). Moreover, in the samples stored for longer than 18 months, a broad signal also appeared in the region of saturated alcohols at 3.85-3.89 ppm, which can arise from lipolysis (Sacchi et al., 1996).
