Protecting the authentic extra virgin olive oil

Extra virgin olive oil (EVOO) is the only vegetable oil that is consumed as it is—freshly extracted from the fruit. Thanks to its balanced taste and flavored aroma, EVOO is capable of enhancing the most popular gastronomic recipes and is thus considered the chef’s gold. EVOO also offers highly beneficial health effects, thanks to both its high content of monounsaturated fatty acids, vitamins, and polyphenols—the antioxidant substances.

EVOO extraction, being carried out by mechanical means only, preserves the contents of antioxidant compounds. Some of these molecules are not contained in other vegetable oils (e.g. polyphenols), or are removed by refining, while those that are contained in every vegetable oil (tocopherols) are partially removed by refining. Seed oils usually undergo to refining both because of inadequate characteristics to direct human consumption and with the aim to fit standards.

Among all vegetable oils, EVOO contains the highest levels of antioxidants and the highest amount of monounsaturated fatty acids. It is a “healing fat”, since it controls the “bad” LDL cholesterol levels while raising the “good” HDL ones [1, 2]. Studies have shown that people who consumed 25 ml—about two tablespoons—of EVOO daily for 1 week presented less oxidation of LDL cholesterol and higher levels of antioxidant compounds in the blood [3], and, more generally, that EVOO is a panacea of the entire cardiovascular system [4].

Because of the time-consuming agronomical practices used for EVOO production, and the low production efficiency—frequently lower than 20% as oil yield—the price of EVOO is high, especially compared with that of other vegetable oils. Consequently, EVOO is prone to adulteration with cheaper oils in order to increase profits. In addition to an economic burden, EVOO adulteration is detrimental if consumers react by buying other cooking fats or dressings, thinking that EVOO cannot be trusted. The negative implications on consumer confidence are even worse than the economic ones. Lastly, EVOO protection measures also imply the product area conservation, as far as landscape, tourism, and job preservation are concerned.

While the European Commission regulations are indicating the characteristics of olive oil types, and are suggesting the methods of analysis [5], a lot of research is currently carried out, in order to experiment innovative techniques for authenticating extra virgin olive oils and predicting potential frauds.

Numerous methodologies exist for EVOO authentication, both for adulteration detection and quantification. Chromatographic, nuclear magnetic resonance and thermal techniques are frequently used, as well as dielectric spectroscopy or electronic noses [6–15]. They are mostly suitable for laboratory use, since the instrumentation is cumbersome and some treatments of the analyzed sample are required. Optical spectroscopy is also frequently used and sometimes preferred because it allows a rapid and non-destructive analysis and requires minimum or no sample preparation. Infrared, mid-infrared, and fluorescence spectroscopic techniques have been proposed, as well as absorption spectroscopy in the ultraviolet, visible, or near-infrared spectral ranges [16–24]. These optical techniques are usually combined with chemometric methods for spectroscopic data processing, thus providing an excellent EVOO authentication [25–28]. Currently, the intrinsic optical and mechanical characteristics of optical fibers, together with the wide availability of bright LEDs and miniaturized spectrometers, further enhance the potentials of absorption spectroscopy in the visible and near-infrared spectral ranges, and make possible the implementation of compact instrumentation. However, none of the absorption spectroscopy techniques experimented so far takes into account the intrinsic turbidity of the olive oil, which can considerably impair absorption measurements because of the unavoidable attenuation due to the scattering produced by suspended particles.

Indeed, although the intrinsic turbidity of the oil can be regarded as a peculiar characteristic, it has an unstable and non-reproducible influence on absorption measurements because of its time-dependent nature. In fact, suspended particles created during production of the olive oil usually settle down in a non-reversible way, because they tend to aggregate at the bottom of the container, creating a sort of sludge. Sample filtering is not only a time-consuming procedure, but is also an action that alters the composition of the sample. In fact, turbidity is also due to the presence of water, and water removal causes a serious loss of water-soluble compounds—such as polyphenols—that are responsible for the unusual character and authenticity of olive oil.

This paper shows an absorption spectroscopy experiment, carried out in the wide 400–1,700 nm spectral range by means of optical fiber technology, for predicting the adulteration of authentic EVOO produced in the Italian region of Tuscany caused by lower-grade olive oil. EVOO produced in Tuscany is highly appreciated by consumers thanks to its mild fruity aroma with a touch of artichoke, thistle, and grass. It is expensive, being produced at the early ripening of olives. EVOO is one of the most important and popular agricultural products of the region. Actually, one of the main actions of the Regional Board of Tuscany is the protection of EVOO, which is also carried through the early detection of commercial frauds.

Olive-pomace oil (OPO), refined olive-pomace oil (ROPO), refined olive oil (ROO), and deodorized olive oil (DOO) were considered as adulterants. While the detection of EVOO adulteration caused by ROPO, and ROO has been previously achieved by means of absorption spectroscopy [29], we innovatively tested the adulteration caused by DOO, an emerging adulterant, the detection of which is hard to achieve by means of conventional techniques. For the first time, to the best of our knowledge, the spectral fingerprints of authentic and adulterated EVOOs were obtained by means of diffuse-light absorption spectroscopy, which provided intrinsically scattering-insensitive measurements. Then, a customized multivariate processing of spectroscopic data was applied for a straightforward prediction of adulterant concentration and identification of adulterant type.

In order to evaluate the effectiveness of optical spectroscopy as compared to other standard methods, a gas chromatographic analysis of fatty acid composition was also carried out. The results obtained by means of an optical analysis were comparable with those provided by the standard technique, thus demonstrating the validity of an optical approach.

Diffuse-light absorption spectroscopy: the principle and set-up

Absorption spectroscopy in the visible and near-infrared spectral regions is one of the most popular measuring methods of conventional analytic chemistry [30, 31]. More recently, absorption spectra measured in wide spectral ranges, like in the visible and near-infrared, have been considered as signatures or fingerprints from which to predict quality indicators of the product by means of multivariate data processing or other chemometric methods [32]. An added value to absorption spectroscopy is the use of optical fiber technology, since it allows real time and in situ measurements.

When carrying out absorption spectroscopy in liquids, the measurements are found to be influenced by dispersed scattering particles that cause turbidity. Scattering dependence is the main drawback of absorption spectroscopy, especially when particles settle down, or change their position in time. For this reason, liquids must be filtered prior to measurements, with a consequent loss of time and increased costs.

Diffuse-light absorption spectroscopy, that is, spectroscopy carried out by means of an integrating cavity, has been proposed in the literature as an effective method for overcoming scattering-dependence problems in process control [33] and biological applications [34], as well as for more general quantitative spectrophotometry [35].

Diffuse-light absorption spectroscopy makes use of an integrating cavity, typically an integrating sphere which contains the sample under test. The source and the detector are butt-coupled to the sphere. Almost all the light impinging on the sphere surface is diffusely reflected, and the detector can be placed anywhere in the sphere in order to gather the average flux [36–39]. By inserting an absorbing medium in the cavity, a reduction of the radiance in the sphere occurs. The reduction is related only to the absorption of the sample and to its volume, and is independent of scattering, which does not change the average radiance in the cavity. This technique has been recently proposed for effective gas analyses [40, 41].

Efficient diffuse-light measurements need bright sources. A conventional deuterium/halogen lamp is enough, provided that it is butt-coupled to the integrating sphere. However, when optical fibers are needed for a better geometrical versatility of the measuring system, conventional lamps provide poor and insufficient light intensity.

In order to overcome insufficient lighting, we used a compact, high-brightness supercontinuum fiber optic source, the revolutionary advent of which has recently changed the perspectives of optical spectroscopy [42, 43]. This innovative source is made of a holey optical fiber, typically a photonic crystal fiber, which is pumped by a high-power nano- or femtosecond laser. The bright light generated by the holey fiber over a wide spectral range was guided to the input port of the integrating sphere. Another port of the sphere accommodated an optical fiber coupled to a spectrometer, so as to achieve an efficient set-up for diffuse-light absorption spectroscopy, as shown in Fig. 1. The detected light-power, P, is described by Equation 1, as:

$$ P\quad = \quad \frac{{R\;{P_0}\;{A_d}}}{S}\;\frac{1}{{1 - \frac{R}{S}\;\left( {\,S - {A_s} - \alpha \;V} \right)}} $$
(1)

where P 0 is the source power; α is the sample absorption coefficient; V is the sample volume; A d is the detector fiber area; A s is the source fiber area; R is the cavity power reflectivity; S is the cavity surface area.

Fig. 1
figure 1

Set-up for diffuse-light absorption spectroscopy by means of optical fiber technology

Full size image

Commercially available components were used for the practical implementation of the experimental set-up [44]. The Fianium-SC400 fiber optic supercontinuum source was used for illumination: it emits 4 W throughout the entire 415–1,800 nm spectral range. The Instrument System-Spectro 320 fiber optic spectrometer was used as detector, and scanned the wide 400–1,700 nm spectral range with a resolution of 1.37 nm. The Labsphere LMS100 cavity was used as a diffusing sphere, the ports of which were equipped by means of fiber optic connectors for coupling to both the source and the detector. The olive oil sample under test was contained in a glass vial having a volume of 32 cm3. This set-up was previously used for lubricant oil analysis—it allowed for a successful spectral fingerprinting of the lubricant oil and for predicting functional parameters and wear indicators [45].

The collection of authentic extra virgin olive oils and adulterants

Authentic EVOOs were four different types of oils collected in Tuscany, which were produced according to local traditions around the area of Grosseto. The lower-grade olive oils, OPO, ROO, ROPO, and DOO, were provided by the Università di Udine. Table 1 summarizes the codes used for identifying the various oil types.

Table 1 Codes of experimented authentic EVOOs and adulterants
Full size table

Four series of EVOO-adulterant mixtures were prepared. Each of them was made of five samples in which the adulterant fraction was 0.05, 0.25, 0.50, 0.75, and 0.95 of total weight, respectively. They were used for calibration procedures. For each series, three replica mixtures of EVOOs with 0.25, 0.50, 0.75 in weight of adulterants were also prepared. These provided a validation set for the regression model. The entire collection of measured oils consisted of 136 samples, 88 for calibration and 48 for validation, respectively.

For every sample, an empty vial was used to measure P 0, which was replaced by the vial containing the sample under test to measure P. The diffuse-light absorbance, A, hereinafter referred to absorbance only, was calculated as:

$$ A\quad = \quad \frac{{{P_0}}}{P}\quad - \quad 1 $$
(2)

for obtaining a quantity which is linearly related to α, also having A = 0 when there is no attenuation, as in conventional absorption spectroscopy.

Figure 2-top shows the measured diffuse-light absorption spectra of all authentic EVOOs and adulterants. Their mixtures show intermediate spectra: as an example, Fig. 2-bottom shows the O1 EVOO adulterated by means of two different adulterants, OPO and DOO, respectively, providing very similar (O1F1 mix) or highly different (O1F4 mix) spectral signatures. A multivariate processing of the spectroscopic data allowed for predicting the fraction and the type of adulterant. All data processing was carried out in MATLAB® code, by means of customized programs.

Fig. 2
figure 2

Diffuse-light absorption spectra: entire collection of authentic EVOOs (a), adulterants (b), O1F1 mixtures giving similar spectra (c), and O1F4 mixtures giving very different spectral signatures (d)

Full size image

Predicting the fraction of adulterant

As a first processing, the spectra were smoothed by means of Savitsky–Golay algorithm, employing a second degree polynomial and a smoothing window of 15 points (30 nm). Then, the prediction of the adulterant fraction in the mixtures was achieved by using a multivariate analysis method called partial least squares regression (PLS) [46]. This method is used when the predictor matrix has many collinear variables and the usual multiple linear regression cannot be applied. PLS looks for a limited number of PLS “factors” (PF) which are linear combinations of the original predictors. These new variables are mutually orthogonal (thus uncorrelated) and have the maximum possible covariance with the target variable, among all possible combinations of the original predictors. The idea is that each PF should be linked to a different source of data variance, with the first PF being the most linked to the target variable.

The estimation of the optimal number of factors needed to fit the data is a critical issue of PLS. In our experiment, theoretical considerations suggested that only one factor should be needed, because the only “physical” cause for the variability of absorbance was the adulterant fraction. However, due to the presence of saturated absorption peaks, the relation between absorbance and adulterant fraction was not strictly linear (and PLS is a linear method), and one additional PF was sometimes needed to compensate for non linearity. This was particularly evident for F4, which showed the strongest absorption among all the analyzed oils. The optimal number of factors was assessed by testing each PLS model on the validation set and by choosing that minimized the root mean square error of prediction (RMSEP). Two other parameters were evaluated in order to assess the goodness of the fit: the root mean square error of calibration (RMSEC) and the determination coefficient (R 2). RMSEC is, like RMSEP, an estimation of the “expected” prediction error, but is evaluated on the calibration set. R 2 is, instead, the squared correlation coefficient between predicted and reference values, for the calibration set; thus the fit is as better as this value is closer to 1.

Table 2 summarizes the values of these parameters for each EVOO-adulterant mixture, together with the chosen number of PF (# PF). Note that all mixtures involving F4 needed two PFs for achieving the best fit. Indeed, all values of R 2 are very good. The best prediction is obtained for O3 EVOO adulterated by means of ROO (F2), showing R 2 = 0.997 and RMSEP = 0.02. The worst prediction, which is still very good, is obtained for O3 EVOO adulterated by means of ROPO (F5), showing R 2 = 0.932 and RMSEP = 0.1. Figure 3 shows the linear regression graphics of these two cases.

Table 2 Summary of parameters for predicting the fraction of adulterant by means of optical spectroscopy
Full size table
Fig. 3
figure 3

Linear regression between true and predicted fraction of adulterant: the best (O3F2) and the worst (O2F2) cases (circles calibration points, squares validation points)

Full size image

Discriminating the type of adulterant

The previous section showed how it is possible to predict the fraction of adulterant when the adulterant type is known a priori. However, in practice, the type of adulterant is usually unknown. Therefore, we investigated how to discriminate among the different types of adulterants by means of multivariate calibration and classification methods [47].

The principal component analysis (PCA) was firstly used for data exploration. For each EVOO, the spectra of pure adulterants and of calibration mixtures were considered, thus taking into account 24 samples. Figure 4 shows the results of PCA processing obtained for O1 EVOO; similar results were obtained for the other EVOOs. Figure 4-left shows the PCA score plot, and Fig. 4-right the relative loadings. The score plot highlights that DOO (the F4 adulterant) can be easily distinguished along the PC1 axis. In fact, as shown by the loadings, PC1 is linked to the average absorbance in the 500–900 nm range, where DOO absorbance is higher and broader than any other adulterant.

Fig. 4
figure 4

Results of PCA processing obtained for O1 calibration EVOO—left PCA score plot; right relative loadings

Full size image

However, for a better discrimination of the other three adulterants, a more dedicated method such as the linear discriminant analysis (LDA) was needed. Since overfitting is likely with LDA direct processing of large variable sets, like spectra, a two-step PCA + LDA model was considered. For each EVOO, calibration and validation sets were created. The calibration set was made by the spectra of pure OPO, ROO, and ROPO adulterants, and of the relative calibration mixtures (total 18 samples), while the validation set was made by the validation mixtures of the same adulterants (total 9 samples). PCA was applied to the calibration set, showing that two PCs only were sufficient to obtain explained variances of 96% or higher in any case. LDA was then performed on the PCA score matrix, obtaining two discrimination functions. Finally, the decision boundaries separating the three classes of adulterants were calculated. Figure 5 shows the results of PCA + LDA processing for discriminating OPO, ROO, and ROPO (F1, F2, and F5) in O1 EVOO; similar figures were obtained for the other EVOOs. Empty and filled dots represent the calibration and validation samples, respectively. Figure 5-left shows the discriminating map: labels indicating the adulterant percentage in the mixture are added, while adulterants are simply identified by their code. Figure 5-right shows the discriminating map also including the discriminating boundaries.

Fig. 5
figure 5

Results of PCA + LDA processing for discriminating OPO, ROO, and ROPO (F1, F2, and F5) in O1 EVOO (empty and filled dots represent the calibration and validation samples, respectively)—left discriminating map with labels indicating the adulterant percentage in the mixture; right discriminating map with boundaries

Full size image

As expected, the best discrimination among the adulterants is achieved with high adulterant concentrations, and the dots converge towards a point where the pure EVOO should be. Note that the spectrum of authentic EVOO was not considered in the LDA processing, for not introducing a fourth class populated by a single element, which contrasts with LDA principles.

Table 3 summarizes the errors of classification for each authentic EVOO, for both the calibration and the validation set. The errors for the calibration set are higher because this set includes the samples with 0.05 adulterant fraction, which are the most difficult to separate.

Table 3 Errors of OPO, ROO, and ROPO classification for each authentic EVOO, for both the calibration and the validation set
Full size table

The discrimination of OPO, ROO, and ROPO adulterants in O1, O2, and O4 EVOOs is very good, while it worsen for O3. Since O3, unlike the other EVOOs, has an absorbance sensibly higher than all adulterants (except DOO, which is not considered here), the main effect of O3 spiking, whichever adulterant is used, is a general lowering of absorbance, which masks more subtle changes in the spectrum shape.

Comparison of optical spectroscopy analysis with capillary gas chromatography

In order to evaluate the effectiveness of optical spectroscopy with respect to other standard methods, a capillary gas chromatographic analysis of fatty acid composition was carried out. In fact, the determination of fatty acids proved to be a valid technique for the classification of vegetable oils and for the detection of adulteration [48, 49].

The fatty acid composition of the entire collection of olive oils was determined by using the HRGC MEGA 2 series Fisons Instruments. The fatty acid composition was determined as the corresponding methyl esters, which were prepared by means of an alkaline treatment carried out by mixing 0.05 g of oil dissolved in 2 ml of n-hexane with 1 ml of 2 N potassium hydroxide in methanol [50, 51]. A fused silica capillary column (50 m in length, 0.25 mm i.d.), coated by CPSil-88 (0.25 μm film thickness, Varian Palo Alto, CA) was utilized. Table 4 summarizes the fatty acid composition of authentic EVOOs and adulterants. An ANOVA test [52] was carried out to detect significant differences in fatty acid content between authentic EVOOs and adulterants. The following four fatty acids: linoleic, arachidic, behenic, and lignoceric showed substantial differences at a significance level of 5%.

Table 4 Summary of fatty acid composition for authentic EVOOs and adulterant oils
Full size table

In order to predict the fraction of adulterant, a PLS regression algorithm was used, which employed the fatty acid composition as the predictor pattern. For every EVOO-adulterant mixture, the best fit was achieved by using a single PF. The values of fit statistics (RMSEC, RMSEP, R 2) are summarized in Table 5. As expected, the gas chromatographic analysis showed an accurate prediction of the adulterant fraction. However, the results obtained by means of optical spectroscopy, without any sample treatment, are comparable with those obtained using the standard method.

Table 5 Summary of parameters for predicting the fraction of adulterant by means of gas chromatography
Full size table

Perspectives

Diffuse-light absorption spectroscopy performed in the 400–1,700 nm spectral range, combined with a multivariate processing of spectroscopic data, have demonstrated the capability of predicting the adulteration and concentration of diverse lower-grade olive oils which are frequently used as adulterants of authentic EVOOs produced in Tuscany, a centrally located Italian region. Being scattering-independent, this technique can be used for EVOO analysis during the entire shelflife of the product.

The obtained results are encouraging, especially because of the chemical similarities of authentic EVOOs and the lower-grade olive oils considered as adulterants. To the best of our knowledge, we have demonstrated for the first time that optical spectroscopy can be successfully used to quantify the fraction of DOO in authentic EVOO. It is an innovative result, especially in view of the growing use of DOO as adulterant. We are currently considering other types of adulterants, as well as authentic EVOOs from other regions for achieving a wider data base.

Verifying the authenticity of EVOOs is just one of the many other potential applications that diffuse-light absorption spectroscopy has, especially in combination with a suitable processing of the spectroscopic data. Other types of expensive foodstuffs can be authenticated, such as bio-juices, honeys, alcoholic beverages, as well as many other liquids, the most promising of which can be dietary supplements based on herbs and natural cosmetics.