Determination of water content in olive oil by 31P NMR spectroscopy

Article abstract of DOI:10.1021/jf073227n

A method for moisture determination in olive oil using ³¹P NMR spectroscopy is developed. This method is based on the replacement of the hydrogen atoms of water molecules with the tagging agents 2-chloro-4,4,5,5- tetramethyl-1,3,2-dioxaphospholane and diphenylphosphinic chloride. Both reagents were successful in determining moisture in olive oil. However, only the second reagent provided a clean and instantaneous reaction under mild condition with no side reactions as observed with the first reagent. A study comparison was made to assess the agreement between the present analytical NMR method and the well-established methods of Karl Fischer titration.

Full text of DOI:10.1021/jf073227n

1866 J. Agric. Food Chem. 2008, 56, 1866–1872
31
Determination of Water Content in Olive Oil by P
NMR Spectroscopy
EMMANUEL HATZAKIS AND PHOTIS DAIS*
NMR Laboratory, Department of Chemistry, University of Crete, P.O. Box 2208, Voutes Campus,
710 03 Heraklion, Crete, Greece
A method for moisture determination in olive oil using 31P NMR spectroscopy is developed. This
method is based on the replacement of the hydrogen atoms of water molecules with the tagging
agents 2-chloro-4,4,5,5-tetramethyl-1,3,2-dioxaphospholane and diphenylphosphinic chloride. Both
reagents were successful in determining moisture in olive oil. However, only the second reagent
provided a clean and instantaneous reaction under mild condition with no side reactions as observed
with the first reagent. A study comparison was made to assess the agreement between the present
analytical NMR method and the well-established methods of Karl Fischer titration.
KEYWORDS: Moisture; olive oil, 31P NMR spectroscopy; Karl Fischer titration
INTRODUCTION
hydroxyl and carboxyl groups with the phosphitylating agent
2-chloro-4,4,5,5-tetramethyl-1,3,2-dioxaphospholane (compound
1 in reaction Scheme 1) (11, 12). In these studies (11, 12), the
31P chemical shifts of the various signals of the phosphitylated
minor constituents were reported relative to a signal at δ 132.20,
which has been assigned to the respective anhydride depicted
as compound 2 in reaction Scheme 1.
We thought that the same reaction could be used for the in
situ determination of water content in olive oil. This methodol-
ogy has been used earlier (13) for the moisture determination
in coals after extracting the water with a suitable solvent. In
this paper we report the results of our attempt to perform
moisture determination of several olive oil samples by employ-
ing 31P NMR spectroscopy. Also reported is the successful
moisture determination by using an alternative phosphorus
reagent, diphenylphosphinic chloride (3), which appears to be
superior to 1 (13) for reasons to be discussed later. This
methodology for moisture measurements in olive oil will be
compared with the conventional method of the Karl Fischer (KF)
titration.
Water content has long been recognized as an important factor
determining the quality of olive oil. Small quantities of water
in olive oil are responsible for the creation and persistence of
the suspended and dispersed material that constitutes the so-
called “veiling” of extra virgin olive oil (EVOO) (1, 2).
Although there are conflicting experimental results in the
literature (1–5) concerning the oxidative stability of veiled
EVOO relative to filtered ones, the former is not attractive to
the consumer mainly due to its appearance.
Several methods for moisture determination in lipids including
olive fruits and oils have been elaborated in the past relying on
Karl Fischer titration (6), IR and Raman spectroscopy (7, 8),
low-resolution time-domain NMR spectroscopy (9), dry-oven
method with continuous weighing (10), and azeotropic distil-
lation with an immiscible organic solvent such as toluene or
xylene (10). These approaches are designed as rapid commercial
methods to monitor moisture with reasonable accuracy. In
particular, the dry-oven technique is accompanied by several
advantages such as easy performance speed and good reproduc-
ibility. Nevertheless, measurements of this type include the
volatile compounds of olive oil in addition to moisture. The
possibility of olive oil oxidation constitutes another disadvantage
of heating, because this process is usually conducted in open
air. Distillation methods show difficulties associated with low
precision of the measuring device and the codistillation of water-
soluble components leading to erroneous results.
MATERIALS AND METHODS
Olive Oil Samples and Reagents. Fourteen extra virgin olive oil
(EVOO) samples (years of harvesting 2004–2005) and one sample of
olive-pomace oil (sample 15) were used in the present study. The
EVOOs obtained from Zakynthos (samples 1–3), Messinia (samples
4–6), Lakonia (sample 7), and Heraklion (samples 8–10) were extracted
from the olive variety Koroneiki, whereas those from Lesvos originated
from two different olive varieties, that is, Andramitini (samples 11–13)
and Kolovi (sample 14).
In our laboratory, we have utilized intensively high-resolution
31P NMR spectroscopy for the qualitative and quantitative
determination of minor constituents (mono- and diacylglycerols,
polyphenols, total free sterols, free fatty acids, glycerol, D-
glucose, and maslinic acid) of olive oil after derivatization of
Protonated solvents and reagents for synthesis (reagent or analytical
grade), deuterated chloroform, and pyridine were purchased from
Sigma-Aldrich (Athens, Greece). The Karl Fischer reagents (HYDRA-
NAL-composite 5 and HYDRANAL methanol dry) were purchased
from Hydranal (Riedel-de-Haan, Seelze, Germany).
* Author to whom correspondence should be addressed (telephone
+302810545037; fax +302810545001; e-mail dais@chemistry.uoc.gr).
10.1021/jf073227n CCC: $40.75  2008 American Chemical Society
Published on Web 02/27/2008

Determination of Water Content in Olive Oil
J. Agric. Food Chem., Vol. 56, No. 6, 2008 1867
Scheme 1
Table 1. Calibration of Karl Fischer Titration: Comparison of Titration
Results with Known Weights of Water
volume ratio (2:1), in which 22.3 mg/mL of Cr(acac)3 and 0.2 g of the
internal standard methyltriphenylphosphonium iodide were added. The
stock solution was protected from moisture with 5A molecular sieves.
The required volume of the stock solution (0.4 mL) and the reagent 3
(100 µL) were placed in a 5 mm NMR tube, and the mixture was left
for about 5 min at room temperature to allow reagent 3 to react with
the moisture of the medium. The 31P NMR spectrum was then recorded
to determine the background moisture content. After the determination
of the basal moisture, 100 mg of olive oil was added; the reaction
mixture was left to react for about 15 min at room temperature, and
upon completion of the reaction, the solution was used to obtain the
31P NMR spectra.
weighed (% w/w)
measured (% w/w)
0.300
0.500
0.700
1.000
1.300
0.419
0.575
0.684
1.017
1.453
R ) 0.984 ( 0.041
slope ) 0.960 ( 0.040
intercept ) -0.047 ( 0.034
Tests of Reagents 1 and 3 with Known Water Concentrations.
Reagents 1 and 3 were added in excess (150 µL each) in 400 mL of
the standard solutions prepared as described above. After determination
of the basal moisture of the mixture of solvents and the reagent, water
Preparation of Compounds. The derivatizing reagent 1 was
synthesized from pinacol and phosphorus trichloride according to the
published procedure (14), with a slight modification by us (11) to
increase the yield of the reaction. The phosphorus reagent 3 was
prepared from dibutylphosphite and the Grignard reagent phenyl
magnesium chloride as described in the literature (15); the yield was
65% against 75% for the literature value. The internal standard
methyltriphenylphosphonium iodide (4) was synthesized from triph-
enylphosphine and methyl iodide following the literature procedure (16);
the product yield was 55% (literature value, 65%). Compound 2-chloro-
4,4,5,5-tetramethyldioxaphospholane-2-oxide (5) was obtained upon
photo-oxidation of reagent 1 with singlet oxygen as follows: In a test
tube containing 5 mL of toluene solvent and 100 µL of 1, a preweighed
amount of the photosensitizer tetraphenylporphyrin (TPP) was added.
The concentration of the photosensitizer was 10-4 M. The test tube
was placed on an ice bath and irradiated by a Cermax 300 W xenon
lamp. Oxidation was carried out by bubbling pure oxygen through the
solution for 4 h under continuous stirring.
in different portions in 5 mL of dry pyridine was injected, and the 31
P
NMR spectrum was obtained after each addition. Pyridine was dried
by distillation over barium oxide before use.
NMR Experiments. All NMR experiments were conducted on a
Bruker AMX500 spectrometer operating at 500.1 and 202.2 MHz for
proton and phosphorus-31 nuclei, respectively, at 26 ( 1 °C.
One-dimensional 13P NMR spectra were recorded by employing the
inverse gated decoupling technique to suppress NOE effects. Typical
spectral parameters for quantitative studies using reagent 1 were as
follows: 90°; pulse width, 12.5 µs; sweep width, 55 kHz; relaxation
delay, 25 s; memory size, 32K. To ensure quantitative spectra, the
magnitude of the relaxation delay adopted was >5 times the relaxation
time (T1 ) 4.6 s) of the phosphitylated cyclohexanol; 32 transients
were accumulated for each spectrum.
The spectral parameters for obtaining the 31P NMR spectra by
employing reagent 3 were modified somewhat, because the relaxation
time measured for the internal standard methyltriphenylphosphonium
iodide (T1 ) 0.1 s) was >1 order smaller than that of the phosphitylated
cyclohexanol. Typical spectral parameters for quantitative studies using
reagent 3 were as follows: 90°; pulse width, 12.5 µs; sweep width, 3
kHz; relaxation delay, 0.5 s; memory size, 32K; 32 transients were
accumulated for each spectrum. For all FIDs, line broadening of 1 Hz
was applied, and drift correction was performed prior to Fourier
transform. Polynomial fourth-order baseline correction was performed
before integration. For control experiments, each FID was processed
three times for each spectrum, and the average value was taken for the
calculation of the moisture content.
One-dimensional 1H NMR spectra were acquired with the following
acquisition parameters: time domain, 64K; 90°; pulse width, 9.3 µs;
spectral width, 12 ppm; relaxation delay, 2 s; 32 scans and 8 dummy
scans were accumulated. Baseline correction was performed carefully
by applying a polynomial fourth-order function to achieve a quantitative
evaluation of all signals of interest. The spectra were acquired without
spinning the NMR tube to avoid artificial signals, such as spinning
side bands of the first or higher order.
IR Experiments. The IR spectra were recorded on a FTIR Spectrum
BX (PerkinElmer). The preparation of the sample for the IR measure-
ments was as follows: 100 µL of reagent 1 was added in 1 mL of the
mixture pyridine/chloroform (1.6:1 v/v). After completion of the
reaction of the reagent with background moisture, 200 µL of the mixture
was transferred to a NaCl cell of 0.8 mm path length. The solvents
were removed by a stream of nitrogen gas, and the IR spectrum was
recorded (range of 400–4000 cm-1) at room temperature with 20 scans
and a resolution of 1 cm-1. The IR spectrum of the mixture was
recorded again after the addition of 5 mg of water.
The purity of the various reaction compounds mentioned above was
1
checked by H and 31P NMR spectroscopy.
Karl Fischer Titration. KF titrations were carried out with a
TURBO2 blending titrator (Thermo Orion) applying the one-component
technique using the titrant HYDRANAL-composite 5 and the solvent
HYDRANAL methanol dry. Titrations were carried out at room
temperature using 800 mg of olive oil. Samples were introduced into
the titration cell by means of a 5 mL disposable syringe without needle.
Calibration of this method was performed by titrating known weights
of water. The good correlation between the KF results and the weights
of water is reflected in the correlation coefficient and the slope of the
linear regression shown in Table 1.
Sample Preparation for Moisture Determination in Olive Oil
with Reagents 1 and 3. A stock solution was prepared by dissolving
0.6 mg of chromium acetylacetonate, Cr(acac)3 (0.165 µM), and 13.5
mg of the internal standard cyclohexanol (13.47 mM) in 10 mL of a
mixture of pyridine and CDCl3 solvents (1.6:1.0 volume ratio) and
protected from moisture with 5A molecular sieves. The required volume
of the stock solution (0.4 mL) and the reagent 1 (100 µL) were placed
in a 5 mm NMR tube, and the mixture was left for about 5 min at
room temperature to allow reagent 1 to react with the moisture of the
medium. The 31P NMR spectrum was then recorded to determine the
background moisture content. After the determination of the basal
moisture, 100 mg of olive oil was added; the reaction mixture
was left to react for about 15 min at room temperature, and upon
completion of the reaction the solution was used to obtain the 31P NMR
spectra.
The same solvent mixture of pyridine and CDCl3 (10 mL) was used
for the water derivatization with reagent 3, with a slightly different

1868 J. Agric. Food Chem., Vol. 56, No. 6, 2008
Hatzakis and Dais
Table 2. Concentration (Percent w/w) Changes of the Reaction Products
of Reagent 1a with Water upon Addition of Known Quantities of Water
water added (mg)
2
5
6
0.0
0.1
0.5
0.7
1.0
1.0
1.0
0.028
0.034
0.046
0.057
0.081
0.099
0.105
0.009
0.010
0.010
0.009
0.010
0.009
0.010
0.003
0.004
0.008
0.012
0.022
0.034
0.048
a In excess; see Material and Methods.
Figure 1. 202.3 MHz 31P NMR spectrum of a phosphitylated sample of
olive oil with reagent 1 in pyridine/chloroform solution. The left-hand-side
inset shows the region where the signals of the phosphitylated diacyg-
lycerols, total free sterols, appear. The right-hand-side inset illustrates
the signals of the reaction products of reagent 1 with the water content
in olive oil.
RESULTS AND DISCUSSION
31P NMR Methodology. Reaction of the Phosphorus Re-
agents 1 with Water. In a previous paper (11), we reported the
usefulness of reagent 1 in the quantitative analysis of olive oil,
as well as its application to the quality control and authentication
of EVOO. A number of minor compounds bearing hydroxyl
and carboxyl groups were identified. The inset in Figure 1
shows the well-documented 31P NMR spectrum (11, 12, 17) of
an EVOO sample in the region where the signals of the
phosphitylated compound with the reagent 1 monoacylglycerols,
the two diacylglycerol isomers, total free sterols, and free fatty
acids are observed. Expansion of this spectrum to low frequen-
cies reveals three more signals. The signal at low magnetic field
strength (δ 132.20) has been assigned already to the phosphi-
tylation product 2 of water with reagent 1 according to reaction
1, in which the phosphorus nucleus is in oxidation state +3.
This signal has been used as a reference signal for the 31P
chemical shifts of the various phosphitylated compounds in the
corresponding 31P NMR spectra (11). The 31P chemical shifts
of the other two signals at δ 15.99 and 15.87 indicate the
formation of two additional products, assigned tentatively to
compounds 2-chloro-4,4,5,5-tetramethyl-1,3,2-dioxaphospholane-
2-hydride (6) and 2-chloro-4,4,5,5-tetramethyl-1,3,2-dioxaphos-
pholane-2-oxide (5), respectively, in which the phosphorus
nucleus is in oxidation state +5. It is known (18) that the
phosphorus nucleus in an oxidation state of +3 resonates at
higher frequencies than the same nucleus in an oxidation state
+5. The relative proportion of compounds 2, 5, and 6 was
dependent on the water content.
To examine the influence of water content on the formation
of the three products, the concentration of the reagent 1 was
kept in excess while small amounts of water were added to
pyridine/chloroform solutions (1.6:1 volume ratio). As shown
in Table 2, the amount of compound 5 remained practically
constant, whereas the amounts of both compounds 2 and 6
increased continually, with compound 2 being the predominant
entity in solution.
Identification of Compounds 5 and 6. By recording a coupled
31P NMR spectrum (not shown), we were able to confirm
immediately that one of the two byproducts at low frequency
had a hydrogen attached directly to the phosphorus nucleus,
inasmuch only the signal at δ 15.99 was split into a doublet
with a characteristic one-bond P-H coupling constant of 710.13
Hz (19). Further support for this conclusion was derived from
Figure 2. 500 MHz 1H NMR spectrum of the reaction products of
reagent 1 with water. The inset is an expansion of the region where the
signals of the methyl protons of the various phosphitylated compounds
appear.
1
the 500 MHz H NMR spectrum of the mixture of the three
products shown in Figure 2. The doublet at δ 7.25 (JH,P
)
710.18 Hz) and the two strong singlets at δ 1.37 and 1.47 with
relative intensities of 1:6:6 confirmed the formation of a P-H
bond and the presence of two pairs of nonequivalent methyl
groups. The geminal methyl groups at positions C-4 and C-5
of the phospholane ring are diastereoscopic and mutually
equivalent by symmetry due to the presence of a plane of
symmetry bisecting the CsC bond of the five-membered
phospholane ring and passing though the HsPdO bond. The
geometry of 6 was confirmed by performing DFT quantum
mechanical calculations. The phosphorus atom in an oxidation
state of +5 has a tetrahedral arrangement surrounded by four
atoms: three oxygens and one hydrogen. Therefore, the signal
at δ 15.99 in the 31P NMR spectrum has been definitely assigned
to the phosphorus nucleus of compound 6. Moreover, to prove
that the proton nucleus in compound 6 originated from water,
the reaction 1 of reagent 1 was carried out in D2O. The 31P
NMR spectrum of the reaction product 6 showed a triplet (JP,D
) 106.81 Hz) of relative intensities 1:1:1 (not shown) indicative
of the presence of a P-D bond (19).
A careful scrutiny of the spectrum in Figure 2 reveals the
presence of five more singlets of much lower intensity (indicated
by numbers) than those observed for compound 6. The two pairs
of signals, each with relative intensity 1:1, were ascribed to the
nonequivalent geminal methyl protons of compound 2 and to
those of the unknown compound 5. The broad singlet at δ 1.35
was assigned to the four methyl groups of reagent 1. The lack
of observation of two separate signals for each nonequivalent
pair of geminal methyl groups for 1 was attributed to solvent
effects (mixture of pyridine/chloroform), because the 1H NMR
spectrum of 1 in pure chloroform-d solutions showed clearly
two signals for the methyl protons (not shown). It should be
noted that no coupling between the phosphorus nuclei with

Determination of Water Content in Olive Oil
J. Agric. Food Chem., Vol. 56, No. 6, 2008 1869
at 501 cm-1 in the IR spectrum of the reaction mixture with
reagent 1 (Figure 3A) was correctly attributed to bond P(V)-Cl
of compound 5. The IR spectrum in Figure 3A shows a few
more characteristic bands of compounds 1, 2, 4, and 5. A third
and more convincing piece of evidence supporting the structure
of 5 was its synthesis by photo-oxidation of reagent 1 with
reactive singlet oxygen. Photo-oxidation of 1 produced a product
that had the same 1H and 31P chemical shifts in pyridine/
chloroform solution as compound 5.
Mechanism of Reaction 1. The reaction pathway, according
to the aforementioned transformations upon addition of water
in the mixture of pyridine/chloroform (1.6:1 volume ratio)
containing reagent 1, can be described by the reaction in Scheme
2. In pyridine/chloroform solution, the chlorine atom of reagent
1 is replaced by a hydroxide ion by a simple nucleophilic
substitution affording the intermediate 2-hydroxy-4,4,5,5-tet-
ramethyldioxaphospholane (compound 9). The short-lived in-
termediate cannot be detected by NMR and either isomerizes
to compound 6 via a self-redox reaction or reacts with a second
molecule of reagent 1 to produce compound 2. At the same
time, in both reaction steps, the released HCl gas is captured
by the pyridine base. Compound 5 is not included in the above
reaction scheme, because it does not constitute a byproduct of
the reaction of water with reagent 1; it is rather produced by
direct photo-oxidation of the phosphorus reagent 1. This
conclusion is supported by the IR spectrum of the reaction
products of reagent 1 with an excess of water as seen in Figure
3B. The spectrum shows that the intensity of the band due to
the P(V)-Cl bond of compound 5 at 501 cm-1 does not change
appreciably, whereas the consumption of 1 is reflected in the
complete disappearance of the band of bond P(III)-Cl at 578
cm-1. The constancy of the concentration of compound 5
regardless of the amount of the added water in the reaction
advocates for this conclusion (Table 2). Therefore, the calcula-
tion of water content in test samples and olive oil samples was
based only on the integration of the signals of compounds 2
and 6 in the 31P NMR spectra.
Figure 3. (A) IR spectrum of the reaction products of reagent 1 with
water; (B) the same IR spectrum, but after the addition of an excess of
water.
methyl protons of compounds 1, 2, 5, and 6 was observed in
1
their coupled H and 31P NMR spectra.
Compound 5 has a structure rather similar to that of
compound 6, bearing instead a directly attached chlorine atom
to the phosphorus nucleus. Its minimum energy geometry
calculated by DFT was similar to that of compound 6. The
phosphorus atom has a tetrahedral arrangement surrounded by
three oxygens and one chlorine atom. There are several reasons
favoring this structure: first, the 2-fold increase in the linewidth
(∼8 Hz) of the signal at δ 15.87 in the 31P NMR spectrum
relative to those measured for the other two signals (∼4 Hz).
This observation may be attributed to the presence of the
quadrupolar chlorine nucleus in compound 5. Second, the IR
spectrum of the reaction products of reagent 1 with water,
depicted in Figure 3A, shows a band at 501 cm-1 ascribed to
the bond formed by the pentavalent (V) phosphorus atom with
the chorine atom, P(V)-Cl. It is known (20) that the P-Cl bond
of trivalent (III) and pentavalent (V) phosphorus atoms absorbs
in the IR spectrum between 500 and 650 cm-1 and that bond
P(III)-Cl absorbs at longer wavelengths than bond P(V)-Cl.
Moreover, methyl substitution at the phospholane ring shifts
the IR band of the P-Cl bond to shorter wavelengths. For
instance, bond P(III)-Cl of the compound 2-chloro-1,3,2-
dioxaphospholane (7) absorbs at 599 cm-1 (21). This absorption
band shifts to shorter wavelengths by 21 cm-1 for compound 1
(578 cm-1), which is the tetramethyl derivative of 7 (22).
Accordingly, the absorption band of bond P(V)-Cl of com-
pound 5 is expected to appear at shorter wavelengths by about
20 cm-1 relative to that of bond P(V)-Cl of compound
2-chloro-1,3,2-dioxaphospholane-2-oxide (8). The bond P(V)-Cl
of compound 8 absorbs at 520 cm-1 (23); therefore, the band
Reaction of the Phosphorus Reagent 3 with Water. The next
choice for the olive oil moisture determination was the less
reactive diphenylphosphinic chloride 3 used by Verkade and
co-workers (13, 24) for moisture determination in coals and by
Dadey et al. (25) for the quantification of total phenol content
in various coal-derived materials. 31P NMR spectra of the
reaction of 3 with water in pyridine/chloroform solutions
(reaction Scheme 3) showed that the anhydride 10 was produced
without the formation of any other byproduct. Such a spectrum
is shown in Figure 4; the sole reaction product, 10, gives a
signal at δ 28.05.
Scheme 2

1870 J. Agric. Food Chem., Vol. 56, No. 6, 2008
Hatzakis and Dais
Table 4. Repeatabilitya and Reproducibilitya of the 31P NMR Methodology
by Using Reagents 1 and 3
calcd water content (% w/w)
reagent 1
reagent 3
A/A
repeatability
reproducibility
repeatability
reproducibility
1
2
3
4
0.203
0.205
0.198
0.202
0.195
0.201
0.003
1.493
0.203
0.194
0.196
0.204
0.197
0.199
0.005
2.513
0.202
0.199
0.201
0.203
0.196
0.200
0.003
1.500
0.201
0.205
0.196
0.193
0.203
0.200
0.005
2.500
5
av
SD
%CV
a Calculated using solutions containing 0.2% w/w of water.
Figure 4. 202.3 MHz 31P NMR spectrum of a phosphitylated sample of
olive oil with reagent 3 in pyridine/chloroform solution. Methyltriphenylphos-
phonium iodide is the internal standard.
containing 0.2% w/w water, whereas the reproducibility was
estimated by performing measurements on different days
(interday experiments) on five different water samples (0.2%
w/w) and using the same experimental protocol for each
measurement. Statistical analysis of the data in Table 4 shows
that the performance of both reagents for moisture determination
is equally successful.
Scheme 3
Moisture in Olive Oils. For a more rigorous comparison test
with our 31P NMR results, the moisture contents of a number
of EVOO samples and one olive-pomace oil sample were
determined by employing the KF titration. The results of the
Table 3. 31P NMR Spectroscopic Moisture Determination of Known
Percentage Content of Water with Reagents 1 and 3
moisture determinations by KF titration and by the present 31
P
water calcd (%)
water calcd (%)
water added
NMR methodology using both reagents 1 and 3 are collected
in Table 5. Linear regression of the data in Table 5 provides
very good correlation coefficients (0.980 for reagent 1 and 0.993
with reagent 3) and slopes close to unity (0.998 for reagent 1
and 1.006 with reagent 3). However, a more rigorous treatment
of the data in Table 5 should take into consideration the errors
of the various experimental methods. These errors are reflected
in the variances of the residuals (observed values - predicted
values) (26). Because these variances are usually unknown, we
used the loss function in eq 1, which appears to be more
appropriate because it reflects the proportionality of the inverse
of the variances of the residuals to the values of the independent
variables (26).
with reagent 1
with reagent 3
0.594
1.000
2.000
3.000
4.000
5.000
9.000
0.486
1.026
2.016
2.665
4.320
5.130
8.880
0.559
1.026
2.080
2.808
3.888
4.986
9.140
R ) 0.999 ( 0.005
slope ) 0.006 ( 0.034
intercept ) 0.988 ( 0.095
R ) 0.999 ( 0.003
slope ) -0.048 ( 0.017
intercept ) 1.020 ( 0.002
The reactivity of reagent 3 against the hydroxyl groups of
organic materials was found to be inferior compared to that of
reagent 1. Therefore, the use of the internal standard cyclohex-
anol for the quantification of compound 10 was avoided, because
the complete phosphitylation of cyclohexanol with 3 lasted about
an hour, lengthening thus considerably the duration of the
analysis. After several attempts with other phosphorus com-
pounds, we chose methyltriphenylphosphonium iodide (4) that
gave a well-separated signal at δ 22.00 (Figure 4).
1
KF2
loss function ) (obsd - pred)2 ×
(1)
The KF titration is considered to be an independent variable.
The new regression data given in Table 5 indicate that inclusion
of errors in regression analysis weakens somewhat the correla-
tion between the two experimental analytical techniques. It
should be noted at this stage that regression as a statistical
approach for comparison studies appears to be inappropriate
for several reasons (27, 28). For instance, linear regression
correlates simply the data of two methods, but does not prove
the existence of any agreement between methods.
An alternative approach to the use of linear regression and
correlation was the difference or bias plot recommended by
Bland and Altman (29). On the abscissa they used the mean
value of the methods to be compared, and on the ordinate they
plotted the calculated difference between measurements by the
two methods. They further estimated the mean and standard
deviation (SD) of differences and displayed horizontal lines for
the mean and for the mean ( 2SD. The two horizontal lines
corresponding to the mean ( 2SD constitute the limits of
agreement, which represent the 95% confidence interval for
individual differences between the field and reference method.
Validation of the 31P NMR Methodology. The capability
of reagents 1 and 3 for moisture determination was tested by
injecting known amounts in the reaction medium for each
reagent as described under Materials and Methods. After the
addition of each new portion of water, a 31P NMR spectrum
was recorded. Comparison of the known amount of water and
that calculated from the 31P NMR spectra is shown in Table 3.
Good correlation was obtained between the 31P NMR method
by using both reagents and the known weight of water. Linear
regression of the data in Table 3 resulted in good correlation
coefficients (R) and a regression coefficient close to unity. The
repeatability and reproducibility of our 31P NMR method for
moisture determination using reagents 1 and 3 were also
examined (Table 4). The repeatability for both reagents was
calculated by recording five consecutive spectra on the same
day (intraday experiments), and using the same solution

Determination of Water Content in Olive Oil
J. Agric. Food Chem., Vol. 56, No. 6, 2008 1871
Table 5. Moisture in Olive Oils Determined by 31P NMR Spectroscopy Using Reagents 1 and 3 and by Karl Fischer Titration
origin
NMR (reagent 1)
NMR (reagent 3)
Karl
Fischer
Zakynthos
Zakynthos
Zakynthos
Messinia
Messinia
Messinia
Lakonia
Heraklion
Heraklion
Heraklion
Lesvos
0.274
0.230
0.252
0.750
0.228
0.264
0.287
0.272
0.202
0.247
0.572
0.191
0.310
0.257
0.372
0.309
0.291
0.288
0.771
0.259
0.271
0.267
0.249
0.267
0.213
0.552
0.270
0.335
0.300
0.397
0.284
0.277
0.266
0.769
0.259
0.260
0.287
0.245
0.243
0.204
0.603
0.253
0.329
0.289
0.419
Lesvos
Lesvos
Lesvos
Herakliona
R
0.980 ( 0.055
0.998 ( 0.056
-0.015 ( 0.020
0.993 ( 0.033
1.006 ( 0.033
0.008 ( 0.012
slope
intercept
Rb
slopeb
interceptb
0.977 ( 0.023
0.992 ( 0.021
0.056 ( 0.005
0.980 ( 0.020
1.090 ( 0.023
-0.028 ( 0.008
a Olive-pomace oil sample. b Obtained by using weighted least-squares (see text).
In summary, this plot allowed the assessment of how the
differences differ systematically from zero (bias) and how much
the difference varies (error). Figure 5 illustrates the bias plot
comparing the 31P NMR methodology with the KF conventional
analytical method for moisture determination. Figure 5A shows
the plot of the difference [31P NMR(R1) - KF] versus the
average values of the two methods, [31P NMR(R1) + KF]/2,
using phosphorus reagent 1 (R1), whereas Figure 5B illustrates
the difference [31P NMR(R3) - KF] versus [31P NMR(R3) +
KF] utilizing reagent 3 (R3). As shown in Figure 5, the mean
differences of the two methods for the measurements of moisture
were slightly different from zero (-0.015 for reagent 1 and
0.010 for reagent 3), indicating that there is at most a negligible
systematic difference between measurements of moisture in olive
oil. In Figure 5A, which illustrates the distribution of the data
points in the bias plot of moisture determined by using reagent
1, all measurements are located within the upper (+0.0445) and
lower (-0.0749) limits of agreement. Apart from 1 measurement
of moisture using reagent 3 (Figure 5B), which is close to the
upper limit of agreement (+0.0459), the remaining 14 measure-
ments are within the limits of agreement (lower limit, -0.0254)
in the respective bias plot. Moreover, no any obvious relation-
ship between the difference and the average are observed in
either bias plot; that is, the differences do not vary systematically
over the range of measurements.
Because results in this study furnish only the sample statistics,
it is necessary for the generalization of the results to other
populations to report confidence intervals (CIs). Within a
specified probability (95%), CIs show a range of values based
on the observed data within which the population value lies
(30).
2
/
√
95% CI for mean bias ) d ( t × SD
n
(2)
(3)
(4)
Figure 5. Difference (bias) plots of olive oil samples measured by 31P
NMR spectroscopy and Karl Fischer titration against the average of
measurements with the mean difference (solid lines) and limits of
agreement (dotted lines) by using (A) phosphorus reagent 1 and (B)
phosphorus reagent 3.
2
/
√
95% CI for upper limit ) (d + 2SD) ( t × 3SD
n
n
2
/
√
95% CI for lower limit ) (d - 2SD) ( t × 3SD
The (95% CIs calculated from the formulas reported in ref 28
were found to be -0.03173-0.000133 and 0.00039-0.020143
for the mean differences [31P NMR(R1) - KF] and [31P
NMR(R3) - KF], respectively. The (95% CIs for the upper
andlowerlimitsofagreementwereestimatedtobe0.0159-0.0732

1872 J. Agric. Food Chem., Vol. 56, No. 6, 2008
Hatzakis and Dais
(12) Dais, P.; Spyros, A. 31P NMR spectroscopy in the quality control
and authentication of extra virgin olive oil. A review of recent
progress. Magn. Reson. Chem. 2007, 45, 367–377.
(13) Wroblewski, A. E.; Reinartz, K.; Verkade, J. G. Moisture
determination of argone premium coal extracts by 31P NMR
spectroscopy. Energy Fuels 1991, 5, 786–791.
(14) Zwierzak, A. cyclic organophosphorus compounds. I. Synthesis
and infrared spectral studies of cyclic hydrogen phosphites and
thiophosphites. Can. J. Chem. 1967, 45, 2501–2512.
(15) Crofts, P. C.; Downie, I. M.; Williamson, K. The reaction of
pyrophosphoryl chloride with Grignard reagents. J. Chem. Soc.
1964, 1240–1244.
(16) Lightner, D. A.; Crist, B. V.; Kalyanam, N.; May, L. M.; Jackman,
D. E. The octant rule. 151 Antioctant effects: synthesis and circular
dichroism of 2-exo- and 2-endo-alkylbicyclo[2.2.1]heptan-7-ones
and -bicycle[ 3.2.]octan-8-ones. J. Org. Chem. 1985, 50, 3867–
3878.
(17) Fronimaki, P.; Spyros, A.; Christophoridou, C.; Dais, P. Deter-
mination of the diglyceride content in Greek virgin olive oils and
some commercial olive oils by employing 31P NMR spectroscopy.
J. Agric. Food Chem. 2002, 50, 2207–2213.
and -0.0463 to 0.1036 for the NMR methodology using reagent
1, respectively, whereas those for the NMR methodology using
reagent 3 were found to be 0.0288-0.0630 and -0.0083 to
0.0425, respectively. In eqs 2-4 d is the mean value, SD2 is
the variance of the difference, and t is the critical value for the
5% two-sided test drawn from tables of t distribution with n -
1 degrees of freedom (df), where n is the sample size.
In summary, this study proposes a new methodology for
moisture determination in olive oil based on 31P NMR spec-
troscopy. This methodology can be extended to other edible
oils. Two phosphorus reagents were used, both reacting with
water molecules quantitatively under mild conditions. Measure-
ments with phosphorus reagent 1 require two integrations, and
therefore they are somewhat less precise than those performed
with reagent 3; the latter gives a sole product with water and
thus a single resonance in the corresponding 31P NMR spectrum.
Nevertheless, the use of reagent 1 allows the simultaneous
quantification of several olive oil constituents in a single
experiment.
(18) Gorenstein, D. Phosphorus-31 chemical shifts and spin-spin
coupling constants: principles and empirical observations. In
Phosphorus-31 NMR; Gorenstein, D., Ed.; Academic Press:
Orlando, FL, 1984; Chapter 1, pp 7–36.
(19) Gorenstein, D. Phosphorus-31 spin-spin coupling constants:
principles and applications. In Phosphorus-31 NMR; Gorenstein,
D., Ed.; Academic Press: Orlando, FL, 1984; Chapter 2, pp 37–
53.
Supporting Information Available: Coupled 31P NMR spec-
tra of the reaction product 6 of reagent 1 with H2O and D2O,
respectively; 1H NMR signals of the nonequivalent methyl
protons of reagent 1 in a mixture of deuterated pyridine/
chloroform solvents, and in chloroform-d solvent. This material
is available free of charge via the Internet at http://pubs.acs.org.
(20) Minkwitz, R.; Dzyk, M. Trichlorohydrophosphoniumhexafluoro
metallate-synthesis, spectroscopic characterization and crystal
structure of Cl3POH+AsF6-. Eur. J. Inorg. Chem. 2002, 564–
568.
(21) IR spectrum of compound 7. Aldrich catalog no. 391220 (CAS
Registry No. 822-39-9).
LITERATURE CITED
(1) Lercker, G.; Frega, N.; Bocci, F.; Servidio, G. “Veiled” extra-
virgin olive oils: dispersion response related to olive quality. J. Am.
Oil Chem. Soc. 1994, 71, 657–658.
(2) Frega, N.; Mozzon, M.; Lercker, G. Effects of free fatty acids on
oxidative stability of vegetable oils. J. Am. Oil Chem. Soc. 1999,
76, 325–329.
(3) Ambrosone, L.; Angelico, G.; Cinelli, G.; Di Lorenzo, V.; Ceglie,
A. The role of water in the oxidation process of extra virgin olive
oil. J. Am. Oil Chem. Soc. 2002, 79, 577–581.
(4) Tsimidou, M. Z.; Georgiou, A.; Koidis, A.; Boskou, D. Loss of
stability of “veiled” (cloudy) virgin olive oils in storage. Food
Chem. 2005, 93, 377–383.
(5) Fregapane, G.; Lavelli, V.; Leon, S.; Kapuralin, J.; Salvador, D. M.
Effect of filtration on virgin olive oil stability during storage. Eur.
J. Sci. Technol. 2006, 108, 134–142.
(6) AOAC Official Method 984.20. Moisture in oils and fats. Karl
Fischer method. Official Methods of Analysis of AOAC Interna-
tional. 16th ed.; AOAC: Gaithersburg, MD,1998; Vol. I. Chapter
41, pp 41–42.
(7) Muik, B.; Lendl, B.; Molina-Diaz, A.; Ayora-Canada, M. J. Fourier
transform Raman spectrometry for the quantitative analysis of oil
content and humidity in olives. Appl. Spectrosc. 2003, 57, 233–
237.
(8) Jimenez, A.; Izquierdo, E.; Rodriguez, F.; Duenas, J. I.; Tortosa,
C. Determination of fat and moisture in olives by near-infrared
reflectance spectroscopy. Grasas Aceites 2000, 51, 311–315.
(9) Todt, H.; Guthausen, G.; Burk, W.; Schmalbein, D.; Kamlowski,
A. Water/moisture and fat analysis by time-domain NMR. Food
Chem. 2006, 96, 436–440.
(22) IR spectrum of compound 1. Aldrich catalog no. 447536 (CAS
Registry No. 14812-59-0).
(23) IR spectrum of compound 8. Spectral Database of Organic
Compounds (SDBS No. 19749).
(24) Verkade, J. G. 31P NMR spectroscopy in the anlysis of coal
materials. In Phosphorus-31 NMR Spectral Properties in Com-
pound Characterization and Structural Analysis; Quin, L. D.,
Verkade, J. G., Eds.; Wiley-VCH: New York, 1994; Chapter 28,
pp 373–383.
(25) Dadey, E. J.; Smith, S. L.; Davis, B. H. Determination of hydroxyl
group concentration in coal liquids by 31P NMR. Energy Fuels
1988, 2, 326–332.
(26) Neter, J.; Wasserman, W.; Kutner, M. H. Applied Linear Statistical
Models. Regression Analysis of Variance and Experimental
Designs; Irwin: Homewood, IL, 1985.
(27) Pollock, M. A.; Jefferson, S. G.; Kane, J. W.; Mackinnon, G.;
Winnard, C. B. Method comparisonsa different approach. Ann.
Clin. Biochem. 1992, 29, 556–560.
(28) Hollis, S. Analysis of method comparison studies. Ann. Clin.
Biochem. 1996, 33, 1–4.
(29) Bland, J. M.; Altman, D. G. Statistical methods for assessing
agreement between two methods of clinical measurements. Lancet
1986, 8, 307–310.
(30) Gardner, M. J.; Altman, D. G. Estimating with confidence. Br.
Med. J. 1988, 296, 1210–1211.
Received for review November 5, 2007. Revised manuscript received
January 9, 2008. Accepted January 20, 2008. We thank the Greek
General Secretariat of Research and Technology and the European
Union for financial support through Program 4.2.6.2B EPAN. Also,
we are indebted to the program Regional Pole of Innovation for Value
Creation funded by the administration of the Region of Crete and the
European Union.
(10) Pierce, M. M. Water. Structures, properties, and determination.
In Food Science, Food Technology and Nutrition; Trugo, L.,
Finglas, P. M., Eds.; Elsevier Science: Amsterdam, The Nether-
lands, 2003; pp 6083–6086.
(11) Dais, P.; Spyros, A. Application of 31P NMR spectroscopy in food
analysis. Quantitative determination of the mono- and diglyceride
composition of olive oils. J. Agric. Food Chem. 2000, 48, 802–
805.
JF073227N