Professional Documents
Culture Documents
Abstract: An evaluation has been made of the potential of near-infrared (NIR) technologies in the assessment of essential oil com-
ponents and in the identification of individual essential oils. The results showed that cross-validation models are able to predict
accurately almost all of the components of essential oils. In different cinnamon (Cinnamomum zeylanicum) and clove (Syzygium
aromaticum) essential oils, which showed a similar composition, 23 components (representing 97.8 – 99.9% of the oil) were accu-
rately predicted, as well as 20 components (93.0 – 99.1%) in Cinnamomum camphora (ravintsara), 32 components (92.3 – 98.1%) in
Ravensara aromatica (ravensara), and 26 components (96.6 – 98.4%) in Lippia multiflora. For almost all of the components, the
modelled and reference values obtained by GC-FID were highly correlated (r 2 ≥ 0.985) and exhibited a low variance (less than 5%).
The model was also able to discriminate between the ravintsara and ravensara essential oils. It was shown that two commercial
oils labelled as R. aromatica were actually ravintsara (C. camphora), revealing the misidentification of these essential oils in the
marketplace. The study demonstrates the application of NIR technology as a quality control tool for the rapid identification of
individual essential oils, for product authentication, and for the detection of adulteration. Copyright © 2006 John Wiley & Sons,
Ltd.
Keywords: near-infrared spectroscopy; GC-FID; quality control; volatile oils; Cinnamomum zeylanicum; Cinnamomum camphora;
Lippia multiflora; Ravensara aromatica; Syzygium aromaticum.
Copyright © 2006 John Wiley & Sons, Ltd. Phytochem. Anal. 17: 121–128 (2006)
DOI: 10.1002.pca
122 H. R. JULIANI ET AL.
ranging from 50 to 60% as the main component, with Wild plants of Lippia multiflora were collected at full
minor amounts of sabinene (10–12%) and α-terpineol flowering from Ghana (Nyankpala, Sari, Atebubu, Aman-
(7%; Behra et al., 2001; Juliani et al., 2005a). tin and Kobre). Voucher specimen (Nyankpala GC
Ravensara aromatica Sonn. (Lauraceae, syn 47801, Sari GC 47807, Atebubu GC 47803, Amantin
Ravensara anisata; ravensara) is a well-known en- GC 47812 and Kobre GC 47808) were deposited
demic aromatic tree from Madagascar and is locally at Ghana Herbarium (GC). The essential oils were
known as ‘Havozo’, which means ‘aromatic tree’ in the extracted in Ghana (by A.Y. Mensah and C. Quansah,
Malagasy language. All the aerial parts contain essen- Knust Kumasi, Ghana) by steam distillation from
tial oils. The leaf oils are dominated by limonene the dried leaves and inflorescences. Pure eugenol and
(14–23%) and sabinene (10–16%) with minor amounts 1,8-cineole as standards were obtained from Sigma
of methyl chavicol (2–12%). The reddish bark is very (St. Louis, MO, USA).
aromatic and contains volatile oils that are rich in
methyl chavicol (90–95%), whilst the fruits also contain Reference analyses. Samples were analysed using an
essential oils (Behra et al., 2001). Agilent (Palo Alto, CA, USA) 6890 GC system coupled to
The essential oils of ravensara and ravintsara are Agilent 5973 series FID and mass selective detectors,
frequently misidentified in the marketplace (Juliani and fitted with an HP5 column (30 m × 0.25 mm i.d.;
et al., 2005a), through a confusion that appears 0.25 µm). Samples were injected using an Agilent 7683
to have originated years ago (Behra et al., 2001). The series autosampler with an inlet temperature of 200°C.
botanical name of the genus Ravensara came from The column temperature programme was: 60°C for
latinisation of the Malagasy word ‘ravintsara’ (meaning 1 min, increased by 4°C/min to 200°C, and held for
‘good leaves’), which refers to the introduced species 15 min. The carrier gas was helium at a flow rate of
C. camphora (Rasonaivo and de la Gorce, 1998). 1 mL/min, and the FID temperature was 220°C. In-
Lippia multiflora Moldenke (Verbenaceae) is another dividual identifications were made by matching spectra
important aromatic plant from West Africa that is with those from MS libraries (Wiley 275.L), and
widely used as a tea and also has many medicinal identities were confirmed by retention indices (RI;
properties. The essential oils extracted from the leaves Adams, 1995). The relative percentages of each
have long been used locally as spices, condiments, component, listed in order of elution from the HP5
flavours in foods and drinks and as cosmetics, and (DB5 equivalent) column, were calculated using the
these oils have a great potential on the international FID detector by peak integration without calculating
marketplace (Juliani et al., 2005b). the detector response factor. The relative percentages
The lack of quality assurance and control of African of those components that were absent or present in
natural materials using assays designed to ensure the only trace amounts (less than 0.01%) were assumed to
authenticity of botanical products is the major obstacle be zero.
to international trade. The present study evaluates the
potential application of NIR technology in the assess- Near-infrared spectroscopy. Essential oils (1 mL) were
ment of essential oil quality, and in the identification placed in quartz cuvettes (path length 5 cm) and
and adulteration of individual essential oils of a range scanned 10 times in the range 300–2500 nm using a
of selected African essential oils. portable NIR spectrometer (LabSpec Pro; Analytical
Spectral Devices, Boulder, CO, USA). Each individual
scan was the result of an average of 25 scans auto-
EXPERIMENTAL matically made by the equipment. Data processing
and development of appropriate chemometrics methods
Materials. Cinnamon (C. zeylanicum) leaf oil and clove were carried out using the program GRAMS 386
(S. aromaticum) bud oils were obtained from the (Galactic Industries Corporation, Salem, NH, USA).
Malagasy companies EPAM (Fianarantsoa), Phael Flor Calibration accuracy was defined in terms of the
(PF; Antananarivo) and Parapharma (PP; Antananarivo). multiple coefficient of determination (r 2), that measures
For comparison purposes, essential oils of these the linearity between the modelled and reference val-
plants obtained from US companies were also include ues (GC-FID), and the variance (V), that measures the
in this study. Ravintsara (C. camphora) and ravensara overall error between modelled and reference values,
(R. aromatica) oils were obtained from the Malagasy expressed as the percentage of standard error of cross-
companies Biosave (Antananarivo), EPAM, Label Con- validation (SECV) in relation to the average of each
servation Biodiversity and Development (Antananarivo), essential oil component (V = SECV*100/mean value).
and the National Center for Application of Phar- Values ranging from 0 to 5% are considered excellent
maceutical Research (Antananarivo). Two ravensara and acceptable for any NIR application, whilst values
(R. aromatica) oils derived from Madagascar were >5% are considered unsuitable for NIR analysis
obtained in the US marketplace for comparison (Williams, 2001). The numbers of factors were set to
purposes. minimize the variance of the calibration.
Copyright © 2006 John Wiley & Sons, Ltd. Phytochem. Anal. 17: 121–128 (2006)
DOI: 10.1002.pca
ESSENTIAL OIL DETERMINATION BY NIR SPECTROSCOPY 123
RESULTS AND DISCUSSION Correlation statistics showed that the model was
able to predict with high accuracy the relative percent-
Essential oils of cinnamon leaf and clove buds age of all 23 essential oil components constituting 98–
99.9% of the total oil (Table 1). The coefficient of
Six cinnamon (C. zeylanicum) leaf, five clove (S. determination of the cross-validation model for each of
aromaticum) bud, and one clove leaf essential oil the components was very high (r 2 ≥ 0.993), while the
samples, together with pure eugenol (99.9%), were variance between the model and reference values for
used to build a cross-validation model (Table 1). The eugenol, eugenyl acetate and (E)-caryophyllene was low
compositions of the cinnamon and clove oils were simi- (<1.75%). For the majority of the minor components,
lar, being dominated by eugenol (59 – 82 and 69–81%, the variance was less than 4.6%. However, for com-
respectively) with minor and varying amounts of ponents with relative percentages of less than 0.18%,
eugenyl acetate (2–11 and 1–15%, respectively) and (E)- such as furfural, α-thujene and sabinene, the variation
caryophyllene (2– 4 and 5–10%, respectively). The cin- was much higher (≥7%; Table 1).
namon leaf oils exhibited a more complex mixture of
volatile constituents containing significant amounts of
monoterpenes (4– 7%) that were almost absent from Ravintsara and ravensara essential oils
the clove oils (0– 0.22%). The cinnamon leaf oils also
contained higher levels of safrole and benzyl benzoate Ten essential oils of ravintsara (C. camphora) and 20
(1–11 and 5%, respectively) than the clove oils (0– 0.1 components were used to build a cross-validation
and 0– 0.7%). model. The major components of the essential oil,
Table 1 Percentage compositions of cinnamon leaf (Cinnamomum zeylanicum), clove leaf and clove bud (Syzygium aromaticum)
essential oils by GC and the corresponding NIR correlation statistics
a
Relative percentage composition derived from GC-FID.
b
Overall error between modelled and reference values calculated from (SECV*100/mean value) for each component.
c
The value 0.0 implies that the component was absent or present in trace amounts of less than 0.01%.
d
The range was extended by including pure eugenol (99.9% by FID) in the model.
e
Pure eugenol was removed from the model.
Copyright © 2006 John Wiley & Sons, Ltd. Phytochem. Anal. 17: 121–128 (2006)
DOI: 10.1002.pca
124 H. R. JULIANI ET AL.
Table 2 Percentage composition of ravintsara (Cinnamomum camphora) essential oils by GC-FID and the corresponding NIR
correlation statistics
a
Relative percentage composition derived from GC-FID.
b
Overall error between modelled and reference values calculated from (SECV*100/mean value) for each component.
namely, 1,8-cineole (58%), sabinene (11%) and α- with a high coefficient of determination (r2 > 0.992) and
terpineol (8%), were accurately predicted by the with a low variance (2% and less; Table 3).
cross-validation model with a high coefficient of deter-
mination (r2 ≥ 0.993) and the variance between the
modelled and the reference values was less than 1% Essential oil of L. multiflora
(Table 2).
The remaining components also showed a high co- Five essential oils of L. multiflora plus the GC-FID data
efficient of determination (r 2 ≥ 0.985; Table 2). The of 26 components were used to build a cross-validation
components with relative percentages ranging from 1.9 model (Table 4). The major components p-cymene
to 4.6% (α-pinene, β-pinene, limonene and an (14%), γ-terpinene (10%), thymol (37%), carvacrol (5%)
unknown) showed a higher variance (1.3–1.7%). The and thymyl acetate (14%) were accurately predicted
minor components ranging from 0.4 to 1.2% showed a with a high coefficient of determination (r 2 ≥ 0.985) and
variance of 2.3% or less. The components with relative a low variance (V < 2%). The remaining 21 minor com-
percentages less than 0.3% (i.e. camphene, sabinene ponents were also predicted with a high coefficient of
hydrate, unknown, and terpinolene) showed a very correlation (r 2 ≥ 0.978) and with variance ranging from
high variance (3–16%; Table 2). 1 to 4% (Table 4).
Five essential oils samples of ravensara (R. The results showed that the cross-validation
aromatica) and 32 components were used to build a models were able to predict accurately nearly all of the
cross-validation model (Table 3). The major compo- components in the essential oils. For almost all of the
nents of the essential oil, namely, sabinene (13%), 20–32 compounds present in the different essential oil
limonene (17%) and methyl eugenol (11%), were samples, the modelled and reference values obtained
predicted with a very high coefficient of determination by GC-FID were highly correlated (r 2 ≥ 0.985) and
(r2 = 0.999), and the variance between modelled and exhibited a low variance (V ≤ 5%). In previous
reference values was less than 1% (Table 3). The reports using NIR models for the analysis of isolated
remaining 29 minor components were also predicted essential oils (Schulz et al., 1999; Steuer et al., 2001;
Copyright © 2006 John Wiley & Sons, Ltd. Phytochem. Anal. 17: 121–128 (2006)
DOI: 10.1002.pca
ESSENTIAL OIL DETERMINATION BY NIR SPECTROSCOPY 125
Table 3 Percentage composition of ravensara (Ravensara aromatica) essential oils by GC-FID and the corresponding NIR correla-
tion statistics
a
Relative percentage composition derived from GC-FID.
b
Overall error between modelled and reference values calculated from (SECV*100/mean value) for each component.
Wilson et al., 2001) only 10 components or fewer were components of the oils, menthone, isomenthone and
predicted, whereas in the present study two to three menthol, showed a high coefficient of determination
times the number of components were included since and low variance (Schulz et al., 1999); however, the
the essential oils were far more complex. variance of the minor components (limonene, 1,8-
The reference values obtained by GC-FID for six cineole, menthofuran, menthyl acetate, pulegone and
components of seven Citrus oils were predicted by NIR piperitone) was much higher, ranging from 8 to 25%.
(Steuer et al., 2001). The main components, limonene In a series of Eucalyptus oils, the content of the main
and γ-terpinene, and the minor ones, myrcene and β- constituent 1,8-cineole (71.5 – 85%), measured by the
pinene, were predicted with a high coefficient of deter- freezing point method (Wilson et al., 2001) was ac-
mination (r 2 = 0.99) and with a low variance, whilst the curately predicted using NIR. By extending the range of
remaining two components, α-pinene and sabinene, 1,8-cineole content in the NIR assay by up to 99%
exhibited a higher variance (10% or higher). Through (with pure 1,8-cineole), a more accurate calibration
NIR and GC-FID, it was possible to predict 10 com- was obtained (Wilson et al., 2001). Results from the
ponents in different mint essential oils. The main present study showed that the extension of the range
Copyright © 2006 John Wiley & Sons, Ltd. Phytochem. Anal. 17: 121–128 (2006)
DOI: 10.1002.pca
126 H. R. JULIANI ET AL.
Table 4 Percentage composition of Lippia multiflora essential oils by GC-FID and the corresponding NIR correlation statistics
a
Relative percentage composition derived from GC-FID.
b
Overall error between modelled and reference values calculated from (SECV*100/mean value) for each component.
of eugenol concentration by adding pure standard did ing that the R. aromatica samples from the US com-
not improve the correlation but slightly increased the panies were actually from C. camphora. We were able
variation from 0.63 to 1% (Table 1). to confirm this misidentification on the basis of the
traditional organoleptic, physicochemical and chemical
properties of these oils (Juliani et al., 2005a), thus
A case study validating the qualitative conclusion reached by NIR
technology. Other essential oils, such as citrus and
In order to verify the identity of the essential oil of mint, were also discriminated on the basis of their indi-
ravensara (R. aromatica), two discrimination models vidual spectral properties (Steuer et al., 2001; Schulz
were built with essential oils of ravintsara (C. et al., 1999).
camphora) and ravensara. Authentic ravensara and We then prepared an adulterated essential oil by
ravintsara oils were analysed using both discrimination mixing pure 1,8-cineole with authentic ravensara (R.
models and the samples passed the appropriate tests. aromatica) oil in order to mimic the relative percentage
Two commercial samples (obtained from companies A of 1,8-cineole found in ravintsara (C. camphora) oil.
and B) both labelled as being oil of R. aromatica were When this adulterated product was analysed according
obtained in the US market place. These oils were to the NIR models, it failed to pass either the ravintsara
scanned using the NIR equipment. Neither sample or the ravensara discrimination model even though the
passed the ravensara discrimination model, whilst oil contained 1,8-cineole in an amount similar to that
both passed the ravintsara (C. camphora) model, show- present in true ravintsara oil.
Copyright © 2006 John Wiley & Sons, Ltd. Phytochem. Anal. 17: 121–128 (2006)
DOI: 10.1002.pca
ESSENTIAL OIL DETERMINATION BY NIR SPECTROSCOPY 127
Table 5 Actual and predicted values of the components of ravintsara (Cinnamomum camphora) essential oils obtained from US
companies A and B and incorrectly labelled Ravensara aromatica essential oils
a
Difference between actual and predicted values expressed as the percentage of the actual value.
In order to predict the composition of essential oils cross-validation models could be further improved by
in the two commercial samples incorrectly labelled as analysing more samples so that the entire variability of
R. aromatica, a calibration model was built using the product could be used to predict more accurately
ravintsara oil and the result of the prediction was the minor components.
compared with the actual value from the GC-FID data Although a small number of essential oils (five to 13
(Table 5). The model was able to predict with high samples of each type) were used to build the models,
accuracy the actual value for the major component our results showed that accurate models could be built
1,8-cineole, viz. actual values of 59.6 and 57.7 and and that the data show the great potential of NIR
predicted values of 59.5 and 58.7, respectively, for the technology in the accurate analysis of almost all the
products from companies A and B. The differences components in essential oils. This technology has
between the predicted and actual values (expressed as the ability to identify adulterated essential oils and to
percentages of the actual values) for α- and β-pinene, distinguish essential oils of different chemotypes and
sabinene and α-terpineol were also low (<8%) (Table 5). species. Nowadays, GC analysis is very rapid (through
However, for those minor components present in the the application of Fast and Ultra-Fast GC) and may
range from 0 to 3%, the differences were much higher provide competitive and comparable results in 1–3 min.
(8 –220%). However, the described NIR procedure is even more
Whilst the calibration model was able to predict rapid since, once the models have been created, it re-
almost all of the components in the two mislabelled quires only a few seconds to analyse the individual
essential oils, only the major compounds (1,8-cineole, oil. For this reason, NIR is ideal for routine and quality
sabinene, α-terpineol, and α- and β-pinene) were accu- control analysis.
rately predicted in relation to the reference values. By Coupled with GC, as the reference technique, NIR
international standards, only the major components technology is a very fast and reliable method. With the
are analysed and required to be between specific added value of NIR to identify individual essential oils
ranges in order to meet a defined grade and standard. and the possible detection of adulterations, NIR
Most international standards for essential oils, for ex- spectroscopy becomes a very promising and powerful
ample, have a maximum of three to seven constituents technique in the field of research and quality control of
defined by their minimum–maximum ranges. The essential oils.
Copyright © 2006 John Wiley & Sons, Ltd. Phytochem. Anal. 17: 121–128 (2006)
DOI: 10.1002.pca
128 H. R. JULIANI ET AL.
Copyright © 2006 John Wiley & Sons, Ltd. Phytochem. Anal. 17: 121–128 (2006)
DOI: 10.1002.pca