Lebensm.-Wiss. u.-Technol.

, 35, 99–103 (2002)

Authentication of Olive Oil Adulterated with

Vegetable Oils Using Fourier Transform
Infrared Spectroscopy
A. Tay, R. K. Singh*, S. S. Krishnan and J. P. Gore

Mid Infrared Sensing Diagnostics and Control Consortium, Purdue University, Food Science
A. Tay, R. K. Singh: Formerly with Purdue University, Department of Food Science, W. Lafayette,
IN 47907 (U.S.A.)
A. Tay: The Ohio State University (U.S.A.)
R. K. Singh: University of Georgia, Department of Food Science & Technology, Athens, GA 30602-7610 (U.S.A.)
S. S. Krishnan, J. P. Gore: Purdue University, School of Mechanical Engineering, W. Lafayette, IN 47907 (U.S.A.)
(Received July 18, 2001; accepted August 31, 2001)

Determination of the authenticity of extra virgin olive oils has become more important in recent years following some infamous
adulteration and contamination scandals. The study focused on application of Fourier transform infrared spectroscopy to identify the
adulteration of olive oils. Single-bounce attenuated total reflectance measurements were made on pure olive oil and olive oil samples
adulterated with varying concentrations of sunflower oil (20–100 mL vegetable oil/L of olive oil). Discriminant analysis using 12
principal components was able to classify the samples as pure and adulterated olive oils based on their spectra. A partial least squares
model was developed and used to verify the concentrations of the adulterant. Furthermore, the discriminant analysis method was used
to classify olive oil samples as distinct from other vegetable oils based on their infrared spectra.

r 2002 Elsevier Science Ltd

Keywords: Fourier transform infrared (FTIR); olive oil; adulteration; discriminant analysis; partial least squares (PLS)

Introduction treatment, esterification, or refining. The composition of

the oils is based on the fatty acids present in the tri-
Adulteration of food products, involving the replace- acylglycerols and their location on the glycerol back-
ment of high-cost ingredients with lower grade and bone. This composition varies not only with the type of
cheaper substitutes can be very attractive and lucrative oil and extraction method but also with geographical
for a food manufacturer or raw material supplier. origin and meteorological effects during the growth and
Adulteration of food products is not only a major harvest of the olives. This variation can be used for oil
economic fraud but can also have major health authentication and the identification of adulteration.
implications for consumers. In the 1980s, more than Various physical and chemical tests have been used to
400 deaths and 20,000 casualties occurred from the establish the authenticity of olive oil and to detect the
disease known as ‘Spanish toxic syndrome,’ caused by level of adulterants in it. UV spectroscopy based on
the consumption of adulterated oil (Jimeno, 1982). There- 208–210 and 310–320 nm has been widely used to detect
fore, the detection of adulteration is of importance. the adulteration of virgin with the refined olive oil
The olive oil production is a big business subject to (Passaloglou-Emmanouillidou, 1990). A second deriva-
serious attempts at fraudulent marketing of low-quality tive spectrometry method was reported to be able to
or adulterated oils. There are various methods for detect the adulteration at a level of 6% (Kapoulas and
extracting the oil that yield different quality grades. Andrikopulos, 1987). Analysis of fatty acid profile
Extra virgin olive oil is obtained from the olive Olea after methylation using gas chromatography (GC)
europaea sativa H. by purely mechanical means, and the was reported for quantification of seed oils in olive
lower grade oils are obtained by solvent extraction, heat oil (Kapoulas and Passaloglou-Emmanouillidou, 1981;
Morchio et al., 1989; Lanzon et al., 1989). HPLC
*To whom correspondence should be addressed. analysis of the fatty acid and triglycerides composition
E-mail: [email protected] was also performed for the detection of adulteration of
0023-6438/02/020099+05 $35.00/0 doi:10.1006/fstl.2001.0864
r 2002 Elsevier Science Ltd All articles available online at http://www.idealibrary.com on

99
lwt/vol. 35 (2002) No. 1

olive oil (Sanchis and Rodriguez, 1991). Nuclear Discriminant and partial least squares analysis
magnetic resonance (NMR) analysis (Sacchi et al., Chemometric analysis including discriminant and PLS
1990; Mavromoustakos et al., 1997; Sachhi et al., analysis was carried out using TQ analyst (Thermo-
1997) and a spectrofluorometric method (Marini et al., Nicolet) software package. The spectral regions where
1990) were also reported for detecting the adulteration the variations were observed were chosen for developing
of olive oil. the discriminant analysis and the PLS model. The first
The conventional method of analysing these oils is 12 factors were used for discriminant analysis model and
destructive and time-consuming, involving the hydro- the first five factors for the PLS model, which gave
lysis and methylation of the resulting fatty acids. This 99.9% of spectral information.
approach, however, loses any information associated
with the location of the fatty acids on the original
glycerol backbone. This concern has led to studies using Results and Discussion
the underivatized oils by 13 C-NMR spectroscopy.
Further studies aimed at identifying minor components The typical spectra of extra virgin olive oil and
as an aid to locating and quantifying adulteration have sunflower oil are shown in Fig. 1. The prominent peaks
also been carried out. However, publication of a new due to C–H stretching mode in the wavenumber region
method relying on the selection of specific trace of 2800–3100 cm 1, C=O stretching in the region of
components as indicators immediately gives adultera- 1700–1800 cm 1 and C–O–C stretching and C–H bend-
tors the method of avoiding detection. Work has also ing in the region of 900–1400 cm 1 can be easily
successfully demonstrated the use of pyrolysis mass observed. The entire range of spectra looks almost
spectrometry combined with artificial neural networks similar for both the oils unless one observes very closely.
and supercritical fluid extraction/gas chromatography to This is due to the similar chemical composition of the
assess virgin olive oil adulteration by high concentra- selected oils. Most of the spectral information used for
tions of various oils (Morales et al., 1998). Authentica- the discriminant analysis is contained in the wavenum-
tion of food constituents is a major challenge in food ber regions of 3100–2800 cm 1 and 1800–900 cm 1. The
analysis. Generally, complex, time-consuming and major result of the method developed for the discrimi-
tedious chemical treatment of the samples is required nant analysis relies on the exploitation of these small
for analysis. In this study, Fourier transform infrared- changes in the regions of interest. Figure 2 shows that
attenuated total reflectance (FTIR–ATR) spectroscopy there are certain variations between the spectra of extra
was used as a rapid, nondestructive, authenticity- virgin olive oil and sunflower oil in the fingerprint
measuring tool for edible oils, particularly for extra wavenumber region of 800–1200 cm 1.
virgin olive oils. The samples were classified into two groups: adulterated
and pure extra virgin olive oils. Discriminant analysis
was applied to both classes in the wavenumber regions
Materials and Methods of 3100–2800 cm 1 and 1800–900 cm 1. Figure 3 shows
a Cooman plot for the classification of pure extra
Samples virgin oils and adulterated olive oils with addition of
Thirty-two extra virgin olive oil and seven vegetable oil 20–100 mL/L sunflower oil to extra virgin olive oils
(Canola, walnut, sunflower, soybean, peanut, corn, using 12 principal components for the classification. The
sesame) samples were purchased from local grocery x-axis shows the mahalanobis distance to the olive oil
stores. The olive oil and sunflower oil samples were class while the y-axis shows the distance to the
mixed to obtain the standard or trained sets of 32 pure
and 17 adulterated samples. The amount of sunflower
oil in olive oil ranged from 20 to 100 mL/L. The samples
containing sunflower oils were marked as adulterated
while olive oil samples were marked as pure olive oil and
Arbitrary absorbance unit

classified using the FTIR spectra. In the second part of

the experiment, 32 olive oil samples and seven vegetable
oil samples were compared and classified.

Instrumentation and spectral acquisition

A ThermoNicolet Nexus 670 instrument equipped with
a Mercury Cadmium Telluride A (MCTA) detector and
KBr optics was used to measure the spectra of oil
samples. The oil samples with different and known olive
oil and sunflower oil composition were placed on a
4000 3000 2000 1000
single bounce ZnSe ATR crystal. About 128 scans were
Wavenumber (cm−1)
taken and averaged and measurements were taken at a
resolution of 2 cm 1. The background was collected Fig. 1 The typical spectra of extra virgin olive oil and
before every sample was measured. sunflower oil: F, sunflower oil; —, extra virgin olive oil

100
 lwt/vol. 35 (2002) No. 1

Table 1 Eigen analysis for the classification of pure

extra virgin olive oils and adulterated olive oils
Principal Full spectrum Analysis region
component contribution contribution
1 61.8 91.2
2 79.1 95.4
3 87.3 97.6
4 91.7 98.3
5 95.0 98.7
6 96.8 99.1
7 97.5 99.3
8 98.0 99.5
9 98.4 99.7
Fig. 2 The finger print spectra of extra virgin olive oil and
sunflower oil

3
Distance to adulteration

0 Fig. 4 The spectral variation between olive oil samples and

0 1 2 3 4 5 6
vegetable oil samples in the finger print region
Distance to olive oils

Fig. 3 Cooman plot for the classification of pure extra virgin

oils and adulterated olive oils: (&) extra virgin olive oils; (~) 6
adulterated samples
5
Distance to other oils

adulterated oil class. The discriminant analysis model 4

classified 100% of all samples accurately. The model
was able to classify up to 20 mL/L adulteration level of 3
sunflower oil. Table 1 shows the eigen analysis for 12
principal components (PC) and their analysis contribu- 2
tion. It can be seen that with the first 6 PC’s it is possible
to extract about 99% of the desired information. This 1
shows that there is sufficient variability in the method.
Generally, for discriminant analysis the number of 0
0 5 10 15 20 25
independent sources of spectral variation should be at
Distance to olive oil
least the same as the number of components or classes.
The mahalanobis distance is a very useful way of Fig. 5 Cooman plot of extra virgin olive oils and other
determining the similarity or dissimilarity of a set of vegetable oils: (&) extra virgin olive oil; (~) other oils
values from an unknown sample to a set of values
measured from a collection of known samples. The
actual mathematics of the mahalanobis distance calcula- the spectrum of the unknown sample matches (or does
tion has been known for some time. In fact, this method not match) the original training spectra.
has been applied successfully for spectral discrimination In the second part of this study, the pure extra virgin
in a number of cases. One of the main reasons the olive oil samples were compared with different vegetable
mahalanobis distance method was used in this study is oils such as soybean, peanut, walnut, sesame, canola,
that it is very sensitive to inter-variable changes in the corn, sunflower, and cottonseed oil. All vegetable oils
training data. In addition, since the mahalanobis were classified into one group and compared with the
distance is measured in terms of standard deviations olive oil samples. Discriminant analysis was applied to
from the mean of the training samples, the reported classify the samples into extra virgin olive oil and
matching values give a statistical measure of how well vegetable oil. The spectral variation between olive oil

101
lwt/vol. 35 (2002) No. 1

Table 2 Eigen analysis for the classification of pure 5

extra virgin olive oils and vegetable oils
Principal Full spectrum Analysis region 4
component contribution contribution
1 64.7 97.6 3

RMSECV
2 96.7 99.1
3 98.6 99.6
4 99.1 99.7 2
5 99.4 99.8
6 99.5 99.9
7 99.7 99.9 1
8 99.7 99.9
9 99.8 99.9
0
0 1 2 3 4 5 6 7 8
PC factors

120 Fig. 7 Root mean square error of cross validation

y = 0.9962x + 0.2273 (a) (RMSECV) vs. PC factors
r2 = 0.9962
100
Calculated concentration (mL/L)

80
components (PC) and their analysis contribution for
discriminant analysis. It was possible to extract 97.6%
of information from the first PC. The mahalanobis
60
distance from the Cooman plot revealed that walnut oil
was the farthest and peanut oil was the closest vegetable
40 oil to extra virgin olive oil samples.

20
Partial Least Squares Analysis
0
0 20 40 60 80 100 120 In this part of the study, the results from discriminant
Actual concentration (mL/L)
analysis were confirmed using the PLS algorithm. Olive
oil samples contaminated with 20–100 mL/L of sun-
120 flower oil were used for quantification purposes. The
y = 0.9651x + 2.8064 (b) spectral region used for discriminant analysis was also
r2 = 0.9735
100 used for PLS model. Figure 6a shows the concentration
Calculated concentration (mL/L)

values obtained from the PLS model vs. the actual

80
concentration of soybean oil in olive oil samples. The R2
value for the PLS model was 0.996. The cross validation
shows a reasonable R2 value of 0.974. The cross
60
validation was done by removing one standard at a
time. The result obtained from cross validation is shown
40 in Figure 6b. Confirmation and validation of the analysis
region used for developing the PLS model were done by
20 calculating the predicted residual error sum of squares
(PRESS) values for different PC factors. Figure 7 shows
0 that there is significant contribution from the first 2 PC
0 20 40 60 80 100 120 factors. This confirms that the spectral region used
Actual concentration (mL/L) for developing the PLS model for quantification of
Fig. 6 Concentration values for adulteration obtained from sunflower oil shows significant correlation with the
the PLS model vs. the actual concentration of sunflower oil (a); concentration.
cross validation of the PLS model by removing one standard at
a time (b)
Conclusions

samples and vegetable oil samples is shown in Fig. 4. The presence of sunflower oil (as adulterant) in extra
There is a clear separation of curves for various oils near virgin olive oil can be detected through FTIR spectro-
wavenumber 1400 cm 1. scopy using the ATR sampling method. Multivariate
Figure 5 shows the Cooman plot of extra virgin olive oils classification methods like discriminant analysis can be
and other vegetable oils. The model successfully used to classify pure vs. adulterated olive oil samples
classified olive oil samples and other vegetable oils. successfully down to an adulteration level of 20 mL of
Table 2 shows the eigen analysis for 10 principal sunflower oil in 1 L of extra virgin olive oil. This is much

102
 lwt/vol. 35 (2002) No. 1

lower than the adulteration level at which such MAVROMOUSTAKOS, T., ZERVOU, M., THEDOROPOULOU, E.,
adulteration becomes economically significant. Multi- PANAGIOTOPOULOS, D., BONAS, G., DAY, M. AND HELMIS,
variate quantification based on PLS successfully quan- A. 13C-NMR analysis of the triacylglycerol composition of
Greek virgin olive oils. Magnetic Resonance in Chemistry,
tified the composition of sunflower oil, i.e. the degree of 35, S3-S7 (1997)
adulteration in the olive oil. Finally, discriminant MORALES, M. T., BERRY, A. J., MCINTYRE, P. S. AND APARICIO, R.
analysis was used to successfully classify extra virgin Tentative analysis of virgin olive oil aroma by supercritical
olive oil and other vegetable oils. fluid extraction-high-resolution gas chromatography–mass
spectrometry. Journal of Chromatography A, 819, 267–275
(1998)
MORCHIO, G., DI BELLO, A., MARIANI, C. AND FEDELLI, E.
References Detection of refined oils in virgin olive oil. Rivista Italiana
Sostanze Grasse, 66, 251–257 (1989)
JIMENO, S. A. The Spanish toxic symptoms. Trends in PASSALOGLOU-EMMANOUILLIDOU, S. A. A comparative study of
Analytical Chemistry, 1, 4–6 (1982) UV spectrophotometric methods for the detection olive oil
KAPOULAS, V. M. AND ANDRIKOPULOS, N. K. Detection of adulteration by refine oils. Zeitschrift für Lebensmittel-
virgin olive oil adulteration with refined oils by second Untersuchung und-Forschung, 191, 132–134 (1990)
derivative spectrophotometry. Food Chemistry, 23, 183–192 SACHHI, R., ADDEO, F., GIUDICIANNI, I. AND PAOLILLO, L.
(1987) Application of Carbon-13 NMR to the determination of
KAPOULAS, V. M. AND PASSALOGLOU-EMMANOUILLIDOU, S. A. mono- and diglycerides and free fatty acids in virgin and
Detection of adulteration of olive oil with seed oils by a refined olive oil. Rivista Italiana Sostanze Grasse, 67,
combination of column and gas chromatography. Journal of 245–252 (1990)
the American Oil Chemists’ Society, 58, 694–697 (1981) SACHHI, R., ADDEO, F. AND PAOLILLO, L. 1H and 13C- NMR of
LANZON, A., CERT, A. AND ALBI, T. Detection of refined olive virgin olive oil. An overview. Magnetic Resonance in
oil in virgin olive. Grasas Aceites, 40, 385–388 (1989) Chemistry, 35, S133–S145 (1997)
MARINI, D., GRASSI, L., BALESTERI, F. AND PASCUCCI, E. SANCHIS, R. J. AND RODRIGUEZ, S. J. Rapid HPLC procedure
Spectrophotofluorometric analysis of olive oil. Rivista for the detection of adulteration of olive oil by seed oils.
Italiana Sostanze Grasse, 67, 95–99 (1990) Alimentaria(Madrid), 28, 27–29 (1991)

103