F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 1179
FOOD COMPOSITION AND ADDITIVES
C
ARSTEN
F
AUHL
and R
EINER
W
ITTKOWSKI
Bundesinstitut für Risikobewertung (BfR), PO Box 1033, D-14195 Berlin, Germany
J
ANICE
L
OFTHOUSE
, S
IMON
H
IRD
, and P
AUL
B
RERETON
Central Science Laboratory (CSL), Sand Hutton, York YO41 1LZ, United Kingdom
G
IUSEPPE
V
ERSINI
Istituto Agrario di San Michele all’Adige, Centro Sperimentale, Laboratorio di Analisi e Ricerche, I-38010 San Michele all’Adige (Tn), Italy
M
ICHELE
L
EES
Eurofins Scientific Analytics, Rue Pierre Adolphe Bobierre, BP 42301, 44323 Nantes Cedex 3, France
C
LAUDE
G
UILLOU
European Commission, Joint Research Centre, BEVABS Laboratory, I-21020 Ispra (Va), Italy
Collaborators: D. Anderson; L. Bolognini; B. Brakowiecka-Sassy; N. Christoph; R. Di Stefano; O. Endres;
K. Haase-Aschoff; K. Habersaat; P. Höfken; S. Moser; J. Noser; D. Renzi; D. Sielaff; E. Vaudano; G. Versini; A. Welter
The aim of the present study was to provide the official wine control authorities with an internationally validated method for the determination of 3-methoxy-1,2-propanediol
(3-MPD) and cyclic diglycerols (CycDs)—both of which are recognized as impurities of technical glycerol—in different types of wine. Because glycerol gives a sweet flavor to wine and contributes to its full-body taste, an economic incentive is to add glycerol to a wine to mask its poor quality. Furthermore, it is known that glycerol, depending on whether it is produced from triglycerides or petrochemicals, may contain considerable amounts of 3-MPD in the first case or
CycDs in the second. However, because these compounds are not natural wine components, it is possible to detect glycerol added to wine illegally by determining the above-mentioned by-products.
To this end, one of the published methods was adopted, modified, and tested in a collaborative study. The method is based on gas chromatographic/mass spectrometric analysis of diethyl ether extracts after salting out with potassium carbonate. The interlaboratory study for the determination of 3-MPD and CycDs in wine was performed in 11 laboratories in 4 countries. Wine samples were prepared and sent to participants as
5 blind duplicate test materials and 1 single test material. The concentrations covered ranges of
Received September 29, 2003. Accepted by SG March 10, 2004.
Corresponding author's e-mail: c.fauhl@bfr.bund.de.
0.1–0.8 mg/L for 3-MPD and 0.5–1.5 mg/L for
CycDs. The precision of the method was within the range predicted by the Horwitz equation. HORRAT values obtained for 3-MPD ranged from 0.8 to 1.7, and those obtained for CycDs ranged from 0.9 to
1.3. Average recoveries were 104 and 109%, respectively.
G lycerol is present in wine as a natural by-product of the fermentation process. During fermentation, about
92% of the sugar molecules undergoes alcoholic fermentation to produce ethanol; the remaining 8% undergoes glycero-pyruvic fermentation to yield glycerol.
Glycerol is believed to be responsible for the mouth-feel characteristics that are often indicative of high-quality wines.
It has a sweet taste, but only in very high concentrations
(>20 g/L) does glycerol contribute to the perceived viscosity (1). However, because it is established as an important quality-determining constituent that contributes to the sugar-free extract, glycerol may be deliberately added to mask a wine of poor quality. Because such a practice is prohibited by European oenological legislation, the addition of glycerol to wine is an increasingly prevalent fraudulent practice.
The simplest method for detecting illegal addition of glycerol to wines is based on determination of the glycerol and ethanol content and comparison of these values. Typically for an authentic wine, the glycerol content will be in the range of
6–10% of the ethanol concentration. However, because of the high variability of the natural glycerol:ethanol ratio and the natural variability of all other wine constituents, which are

1180 F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 classically used to assess glycerol addition, such as gluconic acid—a co-indicator of a Botrytis cinerea infection on grapes—or 2,3-butanediol, the usefulness of these methods is limited. It is feasible that glycerol could be added to a wine with a low natural glycerol content and still remain within the range for authentic samples.
More recently, adulteration with industrial glycerol has been detected by determining trace amounts of by-products formed during glycerol production (2).
Industrial glycerol is mainly produced either by fat cleavage or synthetically from a propylene feedstock. It is known that glycerol produced by transesterification of plant and animal triglycerides, using methanol, contains considerable amounts of 3-methoxy-1,2-propanediol
(3-MPD). The synthesis of glycerol from petrochemicals leads to impurities of cyclic diglycerols (CycDs). As both types of compounds do not occur naturally in wine, it is possible to determine the illegal addition of glycerol to wine by quantitation of these impurities by gas chromatography/mass spectrometry (GS/MS). Their chemical structures are shown in Figure 1.
The first GC/MS determination of the by-products in wine included the chromatography of the analytes on a polar phase. Extraction was performed with K
2
CO
3
to salt out the analytes, with ethanol and cyclohexane as solvents (2). A second method included a derivatization step and gas chromatographic separation on a nonpolar phase, with chloroform as the extraction solvent for the salting-out procedure (3). Another method used the chloroform extraction but with a polar phase separation for the determination of CycDs (4). Quantitation of the glycerol addition is difficult because of the variation in by-product content of glycerols obtained from different producers. In addition, advances in purification techniques have resulted in the availability of high-purity glycerols that do not contain 3-MPD or CycDs in detectable quantities.
However, the quantitation of by-products is still a very useful tool for the authentication of wine because positive determination of these impurities, independent of their quantity, clearly indicates a fraudulent wine-making practice in a much more direct way than the classical assessment. Official wine control requires internationally accepted and validated methods, and therefore one of the published methods was adopted, modified, and tested in an interlaboratory study. The
Bundesinstitut für Risikobewertung (BfR) and Central
Science Laboratory were the coordinating partners of this study. In this paper, we present the optimized method and report the results of the interlaboratory study. It is expected that the method will be presented to the
European Commission for adoption as an official method of analysis for wine.
Figure 1. Structures of 3-MPD and CycDs.
Interlaboratory Study
Pretrial Studies
Substantial testing of the method was performed prior to the interlaboratory study. The method was first assessed in-house by the BFR, Germany, and then tested externally by
2 peer laboratories (Central Science Laboratory, UK, and
Istituto Agrario di San Michele all’Adige, Italy). The peer laboratories tested the method and checked for agreement of results. Finally, the potential participants were allowed to familiarize themselves with the method in a pretrial study. For the later study, each of the 13 participants was sent a detailed protocol as well as 4 test materials (1 pair of blind duplicates and 1 set of split-level samples) to analyze. Participants were asked to apply the method and provide results with comments.
All 13 laboratories returned results.
The results and chromatograms obtained from the participants were scrutinized by the coordinating laboratory to assess whether the analytical systems were appropriate and met the specified quality assurance criteria. For both analytes, acceptable HORRAT values (1.1 for 3-MPD and 1.6 for
CycDs) were obtained, suggesting the method was suitable forward to the main trial. Minor amendments to the written

F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 1181 method were made at this stage and circulated to the participants before commencement of the main trial. It was necessary to allow more variety in the chromatographic and detection systems. In addition, more quality assurance was introduced, which enabled control of any interference of the matrix with the analytes and/or the internal standard. This was achieved by running a complete blank sample as part of the matrix calibration.
Sample Scheme
In November 2002, each participating laboratory was sent
11 wine samples, an introduction sheet, a method protocol, a result sheet, and an acknowledgment sheet. Participants were asked to determine each test material once and to submit copies of chromatograms in addition to their results, which were to be reported in mg/L for both analytes.
The test materials consisted of 5 sets of blind duplicates and 1 additional single test material. Dry white wines, dry red wines, and a sweet red wine were used as test materials. They were spiked with glycerol and/or pure analytes. Because the concentration of the by-products in commercial glycerol varies significantly (2), it is not possible to define a typical expectation range for 3-MPD and CycDs resulting from a fraudulent addition to wines. However, the method can be applied successfully, in particular, if small quantities of glycerol (10–25% wine glycerol) are added to wine—quantities that could hardly be detected by classical wine analysis so far. Therefore, it was decided to consider rather low concentrations of impurities in wine resulting from moderate additions of glycerol containing low levels of contaminants. Consequently, low concentration levels were chosen within a range of 0.1–0.8 mg/L for 3-MPD and a range of 0.5–1.5 mg/L for CycDs. The glycerol was either synthetic or resulted from cleavage of fat or maize oil. Surprisingly, the maize oil glycerol was found to contain no 3-MPD, so those samples spiked with maize oil glycerol also had 3-MPD and
CycDs added. Selected samples were fortified directly with standards for calculation of trueness. Specifications of the test materials are shown in Table 1.
The samples were subjected to homogeneity testing (5).
The analyte concentration shown in Table 1 was determined during the homogeneity testing.
Sample Preparation
The white wine samples were authentic samples that had been produced for the European Union (EU) wine databank; therefore they should contain no detectable 3-MPD or
CycDs. The red wine samples were purchased at a local retail market. All samples were checked to ensure they were free of the analytes before further preparation of the test samples.
For the samples spiked with glycerol (Samples A, C, D, F, and G), appropriate amounts of glycerol were weighed and transferred to 3 L wine. The concentration of by-products in the glycerol used for fortification had been determined before addition. The maize glycerol used for the preparation of Samples A and G was found to be free of analytes.
Samples (A, B, and G) were fortified by adding standard ethanol solutions of 3-MPD and CycDs to 3 L wine samples.
After mixing thoroughly, 50 mL portions were prepared in screw-cap bottles (50 mL) and stored in the dark at room temperature until transmittal. All test materials were labeled with their assigned sample code numbers by the organizing laboratory.
Homogeneity
Homogeneity of all samples was demonstrated by using internationally approved procedures (5). Either 5 or 10 randomly selected bottles were analyzed in duplicate. Homogeneity was assessed by using 1-way analysis of variance (ANOVA) and by comparison with the Horwitz standard deviation (6). According to the “International Harmonized Protocol for Proficiency
Testing” (5), the between-bottle standard deviation should be not
Table 1. Sample scheme (main trial)
Code Test material Sample type
A
B
C
D
F
1, 9
8
4, 7
2, 6
3, 11
White wine (spiked with pure analytes)
Red wine (spiked with pure analytes)
White wine
Red wine
Sweet red wine
G 5, 10 White wine (spiked with pure analytes) a
As determined by homogeneity testing.
5.2
7.0
6.7
7.0
6.4
6.3
Wine glycerol
Concn, g/L
Added glycerol
1.27 (maize glycerol)
Not added
1.5 (fat-cleavage glycerol)
0.5 (synthetic glycerol)
0.7 (synthetic)
0.6 (fat-cleavage glycerol)
1.57 (maize glycerol)
3-MPD
Concn, mg/L
CycDs
0.30
(spiked)
0.12
(spiked)
0.21
a
—
0.38
a
0.80
(spiked)
1.50
(spiked)
0.53
(spiked)
—
1.03
a
0.65
a
0.5
(spiked)

1182 F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 more than one-third of the Horwitz standard deviation for the actual concentration level, calculated as the arithmetic mean of the 10 or 20 determinations. All samples were considered homogeneous according to both criteria.
METHOD
Scope and Field of Application
The described method is suitable for the determination of
3-MPD and CycDs ( cis -, trans -2,6-bis (hydroxymethyl)
-1,4-dioxane; trans -2,5-bis(hydroxymethyl)-1,4-dioxane; cis -, trans2-hydroxymethyl-6-hydroxy-1,4-dioxepane) in white, red, sweet, and dry wines.
Principle
The analytes and the internal standard are salted out by addition of K
2
CO
3
and extracted by using diethyl ether.
Extracts are analyzed directly by GC/MS on a polar column.
Detection is then performed in the selected-ion monitoring
(SIM) mode.
Apparatus
( a ) Analytical balance.
—0.0001 g readability.
( b ) Laboratory centrifuge.
— ³ 4000 rpm/min.
( c ) Gas chromatograph .—With mass spectrometric detector and split-splitless injector.
( d ) Diverse quantitative pipets and volumetric flasks.
( e ) Pasteur pipets.
( f ) Centrifugation vials.
—40 mL.
( g ) GC vials.
—1.5–2.0 mL.
( h ) Thermostat.
( i ) Shaking machine.
Chemicals
( a ) K
2
CO
3
, p.A. analytical grade.
—Merck (Darmstadt,
Germany), or equivalent.
( b ) Diethyl ether .—Uvasol, for spectroscopy (Merck), or equivalent.
( c ) Molecular sieve.
—2 mm diam., 0.5 nm pore size
(Merck), or equivalent.
( d ) Absolute ethanol.
—Merck, or equivalent.
Standards
( a ) Cyclic diglycerol mixture (6 components).
—Solvay
Alkali GmbH, 89.3% (Solvay Alkali GmbH no longer provides the standard mixture; solutions of the mixture may be obtained from BfR, Germany); cis -, trans -2,6-bis(hydroxymethyl)-1,4-dioxane; cis -,
Table 2. Pipetting scheme of matrix calibration
Matrix calibration level
Blank
Concn of analyte in wine m g/L
0 mg/L
0
ML0
ML1
ML2
ML3
ML4
ML5
Analyte
CycDs
IS
3-MPD
CycDs
IS
3-MPD
CycDs
IS
IS
3-MPD
CycDs
IS
3-MPD
CycDs
IS
3-MPD
3-MPD
CycDs
IS
3-MPD
CycDs
Spike, m L
100
50
20
100
50
100
25
100
100
30
100
200
40
—
—
100
100
—
—
—
100
Standard solution added Volume of wine, mL
10
S1
S1
S2
S1
S1
S1
S1
S1
S1
S0
S1
S1
S0
S1
S1
S0
10
10
10
10
10
10
1000
2000
1000
1000
3000
1000
2000
4000
1000
100
500
1000
250
1000
1000
500
1.00
2.00
1.00
1.00
3.00
1.00
2.00
4.00
1.00
0.10
0.50
1.00
0.25
1.00
1.00
0.50

F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 1183 trans -2,5-bis(hydroxymethyl)-1,4-dioxane; trans -2-hydroxymethyl-6-hydroxy-1,4-dioxepane.
cis -,
( b ) 3-MPD.
—Aldrich Cat. No. 26,040-1, 98%; or equivalent.
( c ) .—1,4-Butanediol-1,1,2,2,3,3,4,4d
8
(Aldrich Cat. No. 26,56-5, 98%; or equivalent).
Preparation of Standard Solutions
( a ) S0 stock solutions.
—Accurately weigh 10.0 ± 0.05 mg of each standard substance (11.2 mg is weighed for the
CycDs, corresponding to 89.3% purity) and transfer to 10 mL volumetric flask (1 for each). Add exactly 10 mL ethanol and mix thoroughly. The concentration of this solution is
1000 ng/ m L.
( b ) S1 working solutions.
—Volumetrically transfer
1000 L S0 stock solution to 10 mL volumetric flask, dilute the contents to volume with ethanol, thoroughly stopper the flask, and invert flask to mix. The concentration of this solution is 100 ng/ m L.
( c ) S2 working solutions.
—Volumetrically transfer contents to volume with ethanol, thoroughly stopper the flask, and invert flask to mix. The concentration of this solution is
10 ng/ m L.
( d ) Overview of required standard solutions.
—CycDs mixture (6 components): S0, 1000 ng/ m L; S1, 100 ng/ m L.
3-MPD : S0, 1000 ng/ m L; S1, 100 ng/ m L; S2, 100 ng/ m L. 1,4
Butanediol-1,1,2,2,3,3,4,4d
8
(IS): S0, 1000 ng/ m L; S1,
100 ng/ m L.
Preparation of the Matrix Calibration Curve
Matrix-matched calibration solutions are prepared, according to Table 2, in an uncontaminated wine. It is necessary to analyze this wine first to check that it is not contaminated with 3-MPD or CycDs. If the concentrations of the analytes in the sample are outside the range of the calibration curve, additional levels must be prepared. To ensure that the IS does not interfere with any wine components, a blank should be included.
Extraction
Add 100 m L internal standard solution (S1) to 10 mL wine in a suitable centrifugation vial, e.g., 40 mL. (This corresponds to a butanediold
8
concentration of 1 mg/L.)
Carefully add 10 g K
2
CO
3
and mix. Take care during this addition because heat is produced as a result of the evolution of CO
2
. After cooling the solution to approximately 20°C in a water bath, add 1 mL diethyl ether. Shake the mixture for
5 min, using a vertical shaking machine. Centrifuge the vials at 4000 rpm for 5 min. For better removal of the organic phase, the extract can be partially transferred to a vial with a smaller diameter. Using a Pasteur pipet, transfer the upper organic phase, composed of diethyl ether and ethanol, to a GC vial. Add approximately 120 mg molecular sieve to the vial.
Close the vial, and let stand for ³ 2 h, shaking well from time to time. Transfer the clear supernatant to a second GC vial for the
GC/MS analysis.
GC/MS Analysis
Typical parameters for the GC/MS analysis are provided.
Alternative systems may be used, if they provide similar chromatographic performance and adequate sensitivity. The chromatographic system must be able to separate the IS from phenylethanol, a potential interference.
Typical GC conditions: gas chromatograph, HP 5890, or equivalent; DB-Wax (J&W) 60 m, 0.32 mm id, 0.25 m m film thickness, 2 m retention gap, same dimensions or equivalent; carrier gas, H
2
; flow, column head pressure of 60 kPa; temperature program: 90°C for 2 min, increase at 10°C/min to
165°C, hold for 6 min, increase at 4°C/min to 250°C, hold for
5 min; injection: 250°C, splitless, 2 m L, 90 s.
Typical MS conditions: mass spectrometer, Finnigan SSQ
710, or equivalent; transfer line, 280°C; source, 150°C; MS detection: Window 1: 0 1–25 min: 14.3 min, 3-MPD: m/z 75, m/z 61; 16.7 min, IS: m/z 78, m/z 61.
Acquisition time for each mass was 250 m s (dwell time).
Monitor at m/z 91 to assess the separation of the IS from phenylethanol, which also produces a fragment at m/z 78.
Window 2: 25–40 min: 32–34.5 min, CycDs: m/z 57, m/z 117.
Acquisition time for each mass was 250 m s (dwell time).
It has been observed that the analysis may degrade chromatographic performance. In particular, the injection of the high-boiling CycDs mixture is suspected to cause irreversible damage. Injections of reference standard solutions should be avoided; analysis should be restricted to salted-out solutions with low analyte concentrations. In addition, a 1–2 m retention gap is recommended in order to protect the analytical column. Nevertheless, the analytical column has to be considered consumable and must be replaced quite regularly, although ³ 400 injections are possible without significant loss in resolution.
Identification
Record the relative retention time of each analyte (compared with the retention time of the IS). Calculate the mean relative retention time of the analytes in the calibration standards. The relative retention time of the analyte should be the same as that of the standard within a margin of ± 0.5%. As a confirmation criterion, an ion ratio can be calculated for each analyte from
SIM. These ratios are 117/57 for CycDs, 75/61 for 3-MPD, and
78/61 for the IS. The ratio should be within ± 20% of that which is found for the spiked sample. Confirmation by full-scan acquisition can also be used.
Quantification
The quantification is done by using a matrix calibration curve prepared as described in the appropriate section. The analyte/IS area ratios of the indicated mass ratios are plotted versus the concentration of the analyte and correlated by linear regression. Quantification of the CycDs is achieved by summing the peak areas of all 6 peaks and calculating the total content to allow for other distributions of the 6 characteristic
CycDs than those in the standard. The following m/z traces are

1184 F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 used for quantification: 3-MPD: m/z 75; IS: m/z 78; and
CycDs: m/z 117.
Results and Discussion
Interlaboratory Trial
3-MPD .—Ten of the 13 successful pretrial laboratories submitted results for 3-MPD, which were subjected to statistical analysis (7). Laboratory 6 did not provide results for
3-MPD because of separation problems with the wine matrix.
Laboratory 12 appeared to have a systematic error in the determination of 3-MPD, because its results for the main trial samples were consistently higher than those from the other laboratories, with a high y -axis intercept on the calibration graph. The results of this laboratory were therefore identified as technical outliers and were excluded completely from further statistical evaluation. All of the other laboratories produced acceptable calibration graphs. Two of the samples
(Sample D, test materials 2 and 6) were found to be below the limit of detection (LOD) by all of the laboratories.
The concentration of 3-MPD in each test material, as reported by the participating laboratories, and details of results identified as outliers are given in Table 3.
CycDs .—Eleven of the 13 successful pretrial laboratories submitted analytical results for CycDs. All laboratories produced acceptable calibration graphs. Results for 2 of the samples were found to be below the LOD by all laboratories except 8 and 12. Test material C was spiked solely with glycerol derived from fat cleavage. The presence of low concentrations of CycDs in this test material, as reported by
Laboratory 12, is likely to be a false positive (this level was below the LOD for 5 of the laboratories). The concentration of
CycDs in each test material, as reported by the participating laboratories, and details of results identified as outliers are given in Table 4.
Statistical Analysis
Statistical analysis was carried out according to the
“Protocol for the Design, Conduct and Interpretation of
Method Performance Studies” (7) by using a blind duplicate model. Outliers were determined by Cochran and single and double Grubbs tests by using a significance level of P = 0.025.
The Cochran test was carried out to control the variation between the blind duplicate determinations within one laboratory in comparison with the intralaboratory performance of the others. Afterwards, 2 Grubbs tests were performed, single and paired, to detect outlying laboratory means. Outliers are indicated with a footnote b for 3-MPD in
Table 3 and for CycDs in Table 4. Only Cochran outliers were detected in this study. For 3-MPD, the maximum number of outliers identified was 2 laboratories for 1 sample, giving still acceptable data ranging from 7 (1 sample) to 9 laboratories.
Regarding the CycDs, the maximum number of outliers identified was 2 laboratories for 1 sample, giving acceptable data ranging from 9 to 11 laboratories.
Repeatability and reproducibility.
—Calculations for repeatability (r) and reproducibility (R) as defined by that protocol (7) were performed on those results remaining after removal of outliers (Tables 5 and 6).
HORRAT value (Ho).—When a new method is assessed, there is often no validated reference or statutory method with which to compare precision criteria; thus, it is useful to compare the precision data obtained in a collaborative trial with “predicted” levels of precision. These “predicted” levels are calculated from the Horwitz equation (6):
RSD
R
= 2
( .
LogC )
Table 3. Laboratory results obtained for the concentration of 3-MPD (mg/L)
Test material No. (sample code)
Lab 1(A) 9(A) 8(B) 4(C) 7(C) 2(D) 6(D) 3(F) 11(F) 5(G)
2
3
4
5
7
8
9
10
(12) c
13
0.26
0.40
0.28
0.30
0.27
0.43
b
0.11
b
0.30
(1.34)
0.33
0.28
0.40
0.26
0.28
0.29
0.21
b
0.19
b
0.28
(1.85)
0.32
0.12
0.20
0.12
0.17
0.15
0.23
0.08
0.12
(1.94)
0.12
0.25
0.30
0.24
0.27
0.21
0.37
0.15
0.24
(2.05)
0.25
0.24
0.30
0.23
0.23
0.24
0.32
0.17
0.23
(2.05)
0.25
<LOD
<LOD
<LOD
< LOD
<LOD
<LOD
<LOD
<LOD
(1.48)
<LOD a
<LOD
<LOD
<LOD
< LOD
<LOD
<LOD
<LOD
<LOD
(1.84)
<LOD
0.37
0.60
0.67
0.43
0.50
0.42
0.37
0.47
b
(3.10)
0.50
0.32
0.60
0.66
0.47
0.50
0.50
0.37
1.34
b
(1.56)
0.46
a LOD = Limit of detection for the respective laboratory.
b
Outlying result as determined by the Cochran test at the P < 0.025 level; not used in calculation of statistical parameters.
c Noncompliant because of the interference of 3-MPD.
0.70
1.1
0.98
0.66
0.74
0.47
0.55
0.78
b
( 1.86)
0.80
10(G)
0.68
1.1
0.85
0.53
0.71
0.43
0.57
1.54
b
(1.34)
0.79

F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 1185
Table 4. Laboratory results obtained for the concentration of CycDs (mg/L)
Test material No. (sample code)
Lab 1(A) 9(A) 8(B) 4(C) 7(C) 2(D) 6(D) 3(F) 11(F) 5(G)
6
7
8
9
2
3
4
5
10
12
13
1.35
2.00
1.48
1.64
1.43
b
1.54
1.54
6.16
1.54
1.51
1.64
b
1.34
1.90
1.42
1.67
2.35
b
1.68
1.02
1.32
1.47
1.51
1.58
b
0.48
0.70
0.50
0.77
0.597
0.558
0.61
0.39
0.54
0.85
0.52
<LOD a
<LOD
<LOD
<LOD
<LOD
<LOD
<LOD
<LOD
<LOD
0.23
b
<LOD
<LOD
<LOD
<LOD
<LOD
<LOD
<LOD
0.44
b
<LOD
<LOD
0.14
b
<LOD
0.70
0.90
0.66
0.89
0.526
0.658
1.04
2.24
b
0.91
0.90
0.71
0.90
0.90
0.68
0.86
0.667
0.668
0.86
0.61
b
0.90
0.95
0.70
0.87
1.20
0.84
1.15
0.969
0.951
0.99
0.69
0.93
b
1.11
b
0.95
0.85
1.20
0.89
1.01
1.01
1.10
0.86
0.76
3.06
b
0.40
b
1.04
a b
LOD = Limit of detection for the respective laboratory.
Outlying results as determined by the Cochran test at the P < 0.025 level; not used in calculation of statistical parameters.
0.51
0.70
0.63
0.62
0.756
0.627
0.36
0.46
0.57
b
0.48
0.60
10(G)
0.53
0.80
0.59
0.56
0.600
0.594
0.41
0.45
1.49
b
0.37
0.59
Table 5. Results of 3-MPD a
Parameter Sample A, white wine Sample B, red wine b
Sample C, white wine
Sample F, sweet red wine Sample G, white wine
Mean, mg/L
Spike, mg/L
Recovery, % n nc
No. of outliers r n1 s r
RSD r
, %
Ho r
R s
R
RSD
R
, %
Ho
R
0.30
0.30
100
10
1
2
7
0.03
0.01
3.20
0.30
0.13
0.05
15.50
0.80
0.145
0.12
121
10 b
1 b
0
9 b
—
—
—
—
0.13
0.05
32.67
1.53
0.25
—
—
10
1
0
9
0.05
0.02
7.20
0.60
0.15
0.05
21.20
1.10
0.48
—
—
10
1
1
8
0.08
0.03
5.80
0.50
0.31
0.11
22.70
1.30
0.05
6.57
0.59
0.59
0.21
28.91
1.72
0.73
0.80
91
10
1
1
8
0.13
a b
Mean = Arithmetic mean of the data used in the statistical analysis; n = total number of sets of data submitted; nc = number of results
(laboratories) excluded because of noncompliance; No. of outliers = number of results (laboratories) excluded because of determination as outliers by either the Cochran or the Grubbs test; n1 = number of results (laboratories) used in statistical analysis; s r
= standard deviation of the repeatability; RSD
S r
); Ho r r
= relative standard deviation of the repeatability (s r
= HORRAT value for repeatability, which is the observed RSD r
´ 100/mean); r = repeatability (within-laboratory variation) (2.8
divided by the RSD r
value estimated from the Horwitz equation by
´ using the assumption r = 0.66R; R = reproducibility (between-laboratories variation 2.8 ´ s
R
); s
R
= standard deviation of the reproducibility;
RSD
R
RSD
R
= relative standard deviation of the reproducibility (s
R
value divided by the RSD
R
´ 100/mean); Ho
value calculated from the Horwitz equation.
R
= HORRAT value for reproducibility, which is the observed
Single test sample; n, nc, and n1 are single results.

1186 F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004
Table 6. Results for cyclic dyglycerols a
Parameter Sample A, white wine Sample B, red wine b
Sample D, red wine
0.80
Mean, mg/L
Spike, mg/L
Recovery, % n nc
No. of outliers r n1 s r
RSD r
, %
Ho r
R s
R
RSD
R
, %
Ho
R
1.55
1.50
103
11
0
2
9
0.37
0.13
8.50
0.90
0.61
0.22
14.00
0.90
a b
See footnote a in Table 5 for definitions.
Single test sample; n and n1 are single results.
0.593
0.53
113
11 b
0
0
11 b
—
—
—
—
0.379
0.135
22.827
1.319
11
0
1
10
0.19
0.07
8.60
0.80
0.39
0.13
17.30
1.00
11
0
2
9
0.18
0.07
6.70
0.60
0.41
0.15
15.20
0.90
Sample F, sweet red wine
0.96
Sample G, white wine
0.56
0.50
112
11
0
1
10
0.15
0.05
9.30
0.80
0.34
0.12
21.50
1.20
where C = measured concentration of analyte expressed as a decimal (e.g., 1 mg/L = 0.000001) and RSD
R
= reproducibility standard deviation.
The HORRAT (8) value gives a comparison of the actual precision measured with the precision predicted by the
Horwitz equation for a method measuring at that particular concentration of analyte. It is calculated as follows
Ho
R
=
The HORRAT value is an established criterion for the acceptability of an analytical method during its assessment (9). Values around 1 indicate a precision as it could be predicted. A value of <2 still indicates an acceptable method at the low level of concentration. Ho r
is also calculated and used to assess intralaboratory precision, by using the approximation RSD r
(Horwitz) = 0.66 RSD
R
(Horwitz). This assumes the approximation r = 0.66 R. The HORRAT values calculated from the results of this trial are given in Tables 5 and 6.
Collaborators’ Comments
Because the requirements for the chromatographic and detection system were left open in the method protocol, details about the systems used are summarized here. All laboratories used a type of polyethylene glycol column, with various dimensions (30–60 m length, 0.25–0.32 mm diameter, and
0.15–0.5 m m film thickness). In all but one case, the injection type was splitless, and the injection volume varied between
0.5 and 2 m L. For MS, all laboratories employed SIM rather than scanning techniques. No obvious relationship was found between the GC/MS conditions employed and performance.
Few laboratories reported deviations from the protocol.
Two laboratories, 6 and 12, reported difficulties with the chromatographic separation of 3-MPD from other wine constituents, one laboratory, 6, could not separate peaks 2 and
3 of the CycDs. Better chromatographic performance was achieved by 1 laboratory, 10, without using a precolumn.
Precision Characteristics and Interpretation of
Results
3-MPD.
—The precision data for 3-MPD in all samples are summarized in Table 5. Sample B was a single test material, which therefore allowed only the calculation of the reproducibility. RSD r
values ranged from 3 to 7%. RSD
R values ranged from 16 to 33%.
The HORRAT values (Ho
R
) calculated for the test materials were acceptable (0.8–1.7, average 1.3), and results were similar to those obtained for the pretrial samples (Ho
R was 1.1 for the pretrial samples).
Samples A, B, and G were spiked directly with a standard to enable the calculation of trueness. For 3-MPD, the average recovery was 104%, which fulfills the typical criteria for sufficient recovery. For the concentration levels considered, the recovery should be >70 and <110%.
CycDs.
—The precision data for CycDs in all samples are summarized in Table 6. RSD r
values ranged from 7 to 9%.
RSD
R
values ranged from 14 to 23%. Because Sample B was

F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 1187
Figure 2. Representative chromatograms obtained for a wine (ML2) spiked with 3-MPD at 0.25 mg/L (* = phenylethanol).
a single test material, only reproducibility data were calculated.
HORRAT values (0.9–1.3 for Ho
R
) were calculated for all sets of blind duplicate samples. On average, a HORRAT of 1.1 was obtained for the samples investigated, indicating an acceptable precision. The method for the determination of
CycDs performed better than in the pretrial, based on the
HORRAT values (Ho
R
was 1.6 for the pretrial samples). The higher Horrat value in the pretrial may be explained in part by the fact that there were no outliers.
For CycDs, the average recovery was 109%, calculated with the results for Samples A, B, and G. Also for CycDs, the recovery is within the relevant range of >70 and <110% for the concentration level considered.
The chromatographic separation is a crucial point of the method. Separation of 3-MPD and the IS from the matrix should be ensured by the system as stated in the method’s protocol. For the determination of 3-MPD, one laboratory, 6, did not provide results because of problems with separation of the analyte from coextractives. Another laboratory, 12, provided a calibration graph with a high y -intercept, indicating a separation problem. Results were deemed not valid (“noncompliant”) and were therefore not included in the subsequent statistical analysis. Most of the other laboratories
Figure 3. Representative chromatogram obtained for a wine (ML2) spiked with CycDs at 1.0 mg/L.

1188 F AUHL ET AL.
: J OURNAL OF AOAC I NTERNATIONAL V OL . 87, N O . 5, 2004 provided chromatograms, which showed that generally
3-MPD was not detected in the zero-point calibrant (ML0).
Figure 2 shows a typical chromatogram of a wine that contains
3-MPD at 0.25 mg/L. The IS elutes shortly before the phenyl ethanol, and the chromatograms obtained for 3-MPD must be observed carefully for interference.
Chromatographic separation of all 6 CycDs was not achieved by all participants; one laboratory, 5, reported an overlap of peaks 2 and 3. Quantification by taking the area sum of all 6 CycDs makes the method more robust in terms of resolution. Figure 3 shows the typical chromatographic pattern of the 6 CycDs as they are present in the standard.
Conclusions
The results of this study have shown the successful validation of methods for the determination of 3-MPD and
CycDs in wine. The precision was within the range predicted by the Horwitz equation, and the calculated recovery was acceptable.
Acknowledgments
We thank the European Commission for their financial support and express our appreciation to the following collaborators for their participation in the study:
D. Anderson, CSL, York, UK
L. Bolognini and D. Renzi, Unione Italiana Vini, Verona,
Italy
B. Brakowiecka-Sassy, BfR, Berlin, Germany
N. Christoph, BLGL, Würzburg, Germany
R. Di Stefano and E. Vaudano, Istituto Sperimentale per l’Enologia, Asti, Italy
O. Endres and A. Welter, LUA, Speyer, Germany
K. Hasse-Aschoff, Labor Dr. Haase-Aschoff, Bad
Kreuznach, Germany
K. Habersaat and P. Höfken, CLUA, Münster, Germany
J. Noser, Kantonales Laboratorium, Füllinsdorf,
Switzerland
D. Sielaff, LUA, Koblenz, Germany
G. Versini and S. Moser, ISMAA, S. Michele all Adige,
Italy
References
(1) Noble, A.C., & Bursick, G.F. (1984) Am. J. Enol. Vitic.
35 ,
110–112
(2) Lampe, U., Kreisel, A., Burkhard, A., Bebiolka, H.,
Brzezina, T., & Dunkel, K. (1997) Dtsch. Lebensm. Rundsch.
93 , 103–110
(3) Otteneder, H., Zimmer, M., & Schaab, J. (1999) Dtsch.
Lebensm. Rundsch.
95 , 172–175
(4) Bononi, M., Favale, C., Lubian, E., & Tateo, F. (2001) J. Int.
Sci. Vigne Vin.
35 , 225–229
(5) Thompson, M., & Wood, R. (1993) J. AOAC Int.
76 ,
926–940
(6) Horwitz, W. (1982) Anal. Chem.
54 , 67A–76A
(7) Horwitz, W. (1995) Pure Appl. Chem.
67 , 331–343
(8) Peeler, J.T., Horwitz, W., & Albert, R. (1989) J. Assoc. Off.
Anal. Chem.
72 , 784–806
(9) Pocklington, W.D. (1990) Pure Appl. Chem.
62 , 149–162