Determination of volatile constituents in spirits using headspace trap technology

Journal of Chromatography A, 1145 (2007) 204–209

Determination of volatile constituents in spirits using headspace trap technology Katja Schulz a , Jan Dreßler a , Eva-Maria Sohnius b , Dirk W. Lachenmeier b,∗ a

b

Institut f¨ur Rechtsmedizin, Technische Universit¨at Dresden, Fetscherstr. 74, D-01307 Dresden, Germany Chemisches und Veterin¨aruntersuchungsamt (CVUA) Karlsruhe, Weissenburger Str. 3, D-76187 Karlsruhe, Germany Received 15 November 2006; received in revised form 16 January 2007; accepted 19 January 2007 Available online 27 January 2007

Abstract The use of headspace adsorbent traps in combination with gas chromatography was evaluated for the determination of volatile constituents (i.e. higher alcohols and other congeners of alcoholic fermentation) in spirits. The headspace trap technology comprises an enhanced static headspace system that allows enrichment and focusing of analytes on adsorbent traps prior to gas chromatographic separation. Extraction yields 35–55 times higher than those seen with static headspace sampling were achieved. An excellent agreement of analysis results in comparison to the European reference procedure was found (R > 0.98, p < 0.0001). The fully automated headspace trap procedure requires only minimal sample preparation and is easy to apply. © 2007 Elsevier B.V. All rights reserved. Keywords: Headspace adsorbent trap (HS-trap); Dynamic extraction; Thermal desorption; Gas chromatography (GC); Volatile compounds; Spirits

1. Introduction The European reference methods for the analysis of volatile substances prescribe the direct injection of the spirit drink, or appropriately diluted spirit drink into a gas chromatographic (GC) system [1]. This method is not suitable for spirits with contents of total dry extract (e.g. polyphenols in aged spirits or sugar in liqueurs) because the extract contaminates the chromatographic system. Furthermore, the injection of large amounts of water may result in broad peaks and disturbance of FID signal stability [2]. Traditionally, distillation was used in such cases to obtain chromatographiable solutions. Over the last decade, several headspace (HS) extraction and enrichment techniques, including solid-phase micro-extraction (SPME) [2–5] and solidphase dynamic extraction (SPDE) [6], have been developed for the efficient analysis of spirits. In contrast to distillation, these modern extraction techniques are fully automatable, and therefore faster and more efficient. In this study, the so-called headspace trap technique patented by Tipler and Mazza [7] that was recently commercialized is

applied for the first time in beverage analysis. In contrast to SPME and SPDE, which apply small fibres or coated capillaries with sorbent volumes of 0.94 or 5.99 mm3 , respectively [8], the headspace traps used in this study are tubes packed with a solid sorbent with a significantly greater volume of 160 mm3 . The principle of the method is shown in Fig. 1. In the first step, the traps are loaded by pressurizing the sample vials and allowing the pressure to decay through the cooled adsorbent trap (Fig. 1A). They are then dried by passing carrier gas through the trap to remove moisture from the sample (Fig. 1B). Finally, the analytes are thermally desorbed and transported by the carrier gas into the GC column for separation (Fig. 1C). The first applications of the headspace trap included the analysis of air pollutants [9] and the determination of environmental volatile organic compounds and fuel oxygenates [10,11], and it was shown that the headspace trap offers improved detection limits [12]. 2. Experimental 2.1. Reagents and standards

∗

Corresponding author. Tel.: +49 721 9265434; fax: +49 721 9265539. E-mail address: [email protected] (D.W. Lachenmeier).

0021-9673/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.chroma.2007.01.082

All reagents were of analytical grade unless otherwise specified. Water (aqua ad injectabilia) was obtained from

K. Schulz et al. / J. Chromatogr. A 1145 (2007) 204–209

205

many). PTFE/butyl septa and crimp caps were obtained from Perkin-Elmer (Rodgau-J¨ugesheim, Germany). The commercial aqueous congener alcohol standard with following concentrations was used as standard 1a: methanol (16 mg/l), 1-propanol (2 mg/l), 2-butanone (2 mg/l), 2-butanol (2 mg/l), isobutanol (2 mg/l), 1-butanol (2 mg/l), 2-methyl-1butanol (2 mg/l) and 3-methyl-1-butanol (2 mg/l). A standard solution of ethyl acetate with 20 mg/l was prepared in water and used as standard 1b. Standard solutions 1a and b were used for classical static headspace without trap enrichment. A standard solution of ethyl lactate (20 mg/l), 1-hexanol (2 mg/l), benzaldehyde (2 mg/l), and ethyl octanoate (2 mg/l) was used as standard 2 and prepared in water with ethanol (20%) as solubiliser. Standard solution 2 was used for static headspace with trap enrichment. All standard solutions were stored in dark bottles at 4 ◦ C until use for calibration. Tert-butanol (180 ␮l of 1 mg/l) was used as internal standard. The calibration was carried out with three data points and two replications for all standard solutions. For standard solution 1 and calibration level 3, 200 ␮l of standard 1a and 20 ␮l of standard 1b were applied. Calibration levels 2 and 1 were prepared by dilution of standards 1a and b 1:1 (level 2) and 1:3 (level 1) with aqua ad injectabilia each. For standard solution 2 and calibration level 3, 200 ␮l of standard 2 were applied. Further calibration levels were prepared by dilution of standard 2 with aqua ad injectabilia each 1:1 (level 2) and 1:3 (level 1). The total volume of each calibration level was 400 ␮l. 2.2. Sample preparation

Fig. 1. Principle of sample preparation with headspace (HS) trap. The loading is accomplished by pressurizing the sample vials and allowing the pressure to decay through the cooled adsorbent trap (A). A drying step removes moisture from the sample (B). After thermal desorption, the analytes are transported by the carrier gas into the GC column for separation (C).

Samples of alcoholic beverages (11 German fruit spirits and 19 Mexican tequilas, alcoholic strengths between 34 and 54 vol%) were diluted with water 1:100 (for static headspace with trap enrichment) and 1:20 (for classical static headspace without trap enrichment). The 200 ␮l of each dilution were placed in headspace vials. In the same way, 200 ␮l of each level were placed in headspace vials to prepare the calibration samples. Diluted beverage samples as well as calibration levels were spiked with internal standard tert-butanol (180 ␮l), 20 ␮l aqua ad injectabilia (to reach a total volume of 400 ␮l) and Na2 SO4 (0.4 g) and immediately sealed with PTFE/butyl septa. Headspace analyses took place within 48 h. To exclude carry-over, a blank vial was analysed between each sample. 2.3. Apparatus and software

Fresenius Kabi AG (Bad Homburg, Germany). Methanol, 1-propanol, 2-butanone, 2-butanol, iso-butanol, 1-butanol, 2methyl-1-butanol and 3-methyl-1-butanol were purchased as congener alcohol aqueous standard medidrug BGSTM level 3 from Medichem (Steinenbronn, Germany). Ethyl acetate was supplied by Merck (Darmstadt, Germany). Ethyl lactate, 1hexanol, benzaldehyde and ethyl octanoate were obtained from Sigma–Aldrich (Taufkirchen, Germany). Tert-butanol and namyl alcohol, which were used as internal standard, were obtained from Sigma–Aldrich (Taufkirchen, Germany), Na2 SO4 was purchased from Merck (Darmstadt, Germany). The 22 ml headspace-vials were supplied by Macherey-Nagel (D¨uren, Ger-

The headspace analysis was performed with the Perkin-Elmer TurboMatrix HS-110 trap automatic headspace sampler with trap enrichment and flame ionization detector (Perkin-Elmer, Shelton, USA). A capillary column Rtx 1701 (60 m × 0.530 mm I.D.; 1.5 ␮m film thickness) with phenylcyanopropyl phase from Restek was used. Data acquisition and integration were carried out with TotalChrom (Version 6.2.1) software. The investigation of the standard solution 1 was carried out in classical static headspace mode with an injection volume of 300 ␮l, without trap enrichment. The headspace conditions were: sample temperature at 80 ◦ C, needle temperature at 90 ◦ C,

206

K. Schulz et al. / J. Chromatogr. A 1145 (2007) 204–209

transfer line temperature at 95 ◦ C, thermostatting time at 20 min, pressurization time at 3 min, injection time at 0.15 min and column pressure at 90 kPa. The investigation of the standard solution 2 was carried out with trap enrichment. The headspace conditions were the same as above with the following modifications: trap low temperature at 45 ◦ C, trap high temperature at 375 ◦ C, dry purge time at 7.5 min (nitrogen), trap hold time of 10 min, desorbtion time of 10 min, decay time of 1.2 min, vial pressure at 180 kPa and desorb pressure at 100 kPa. One trap cycle was used. Trap material was Air ToxicTM (Carbotrap and Carbosieve sandwich trap). The injection port of the GC was set at 220 ◦ C and the samples were automatically injected in the splitless mode. The detector was set at 250 ◦ C. The GC oven temperature program applied for classical static headspace and calibration standard solution 1 was as follows: the initial oven temperature was set at 37 ◦ C, held for 6 min, then increased to 100 ◦ C via ramp of 10 ◦ C/min, held for 5 min and again increased to 200 ◦ C via ramp of 20 ◦ C/min, and maintained at 200 ◦ C for 10 min. The GC oven temperature program applied for static headspace with trap enrichment and calibration standard solution 2 was as follows: the initial oven temperature was set at 40 ◦ C, held for 5 min, then increased to 220 ◦ C via ramp of 5 ◦ C/min and maintained at 220 ◦ C for 4 min. The carrier gas was high-purity nitrogen (99.999%). The flow rate was regulated about the column pressure of 90 kPa. 2.4. Reference procedure For benchmarking purposes, all samples were analyzed for methanol, 1-propanol, 1-butanol, 2-butanol, isobutanol, 2-/3methyl-1-butanol and ethyl acetate according to the Community reference methods for the analysis of spirits [1]. The range of substances prescribed in the reference methods was extended with ethyl lactate, benzaldehyde and 1-hexanol as they might be of relevance for evaluation of certain products like fruit spirits. In divergence from the reference methods, the amyl alcohols (2-/3methyl-1-butanol) were calculated as sum of both isomers and the aldehydes (ethanal and acetal) were not quantitatively determined. The GC system used for analysis was a Trace 2000 gas chromatograph (Thermo Electron Scientific Instrument Division, Dreieich, Germany). Data acquisition and analysis were performed using the Chromeleon Chromatography Information Management System (Dionex, Idstein, Germany). Substances were simultaneously separated on two fused silica capillary columns, which were attached to the inlet using an inlet splitter (Graphpack 2 M dual column injector adapter, Gerstel Analytical Equipment, M¨ulheim/Ruhr, Germany). This dual column injector adapter permits the simultaneous connection of two fused silica capillary columns with different polarities to one injector and thus the parallel production of two chromatograms. Thanks to the design of the adapter, no split distortions due to contact of the capillary column with the inside wall of the inlet liner can occur. Column 1: CP-WAX 52CB, 60 m × 0.32 mm I.D., film thickness 0.5 ␮m, Varian Deutschland GmbH, Darmstadt, Germany; column 2: CP-SIL 5CB, 60 m × 0.32 mm I.D., film thickness 5 ␮m, Varian Deutschland

GmbH, Darmstadt, Germany. Temperature program: 40 ◦ C hold for 15 min, 4 ◦ C/min up to 200 ◦ C, hold for 10 min, 15 ◦ C/min up to 230 ◦ C, hold for 10 min. The temperature for the injection port was set at 260 ◦ C. After addition of an internal standard (n-amyl alcohol), the spirits were directly injected using split injection mode (2 ␮l, 1:5) and helium with a constant flow rate of 6.5 ml/min as carrier gas. For quantification, peak area ratios of the analytes to the internal standard were calculated as a function of the concentration of the substances. The average of both columns was calculated as result. 2.5. Statistics All data were evaluated using Origin V7.5 (Originlab, Northampton, USA). Statistical significance was assumed at below the 0.05 probability level. Pearson’s test was used to evaluate the significance of linearity of calibration curves as well as of the correlation between the HS-trap and reference procedure. 3. Results and discussion The evaluated HS-trap system is an enhanced static headspace system. The autosampler was advanced by addition of the trap (fused silica capillaries filled with adsorbent material) and additional gas circuits that allow the transport of sample headspace through the adsorbent and, after subsequent thermal desorption, to the GC column. By this process, the volatile analytes are pre-concentrated and focused prior to GC separation, so that a splitless transfer is possible. The loading of the adsorbent trap was achieved in a relatively simple way, as pressurization of the sample vial is used to vent the sample vapour through the trap. No expensive coolants or additional valves are required. During preliminary tests, it was determined that the Air ToxicTM phase material, which is also the standard material provided by the manufacturer, was best suited for the analytes under investigation. At the beginning of method development, two different sorbents were evaluated on the HS-trap: Air ToxicTM and TenaxTM . TenaxTM did not work well. The recoveries (peak areas) were significantly lower (two to three times) for the majority of investigated substances. The best recovery of the volatile compounds was found with Air ToxicTM , which was used for all further experiments. The optimisation of the headspace trap parameters was separately accomplished for the standard solutions 1 (a and b) and 2. Because of the large concentrations differences between the major compounds (e.g. methanol) and the minor ones (e.g. ethyl octanoate), a simultaneous optimization was not possible. In each case, sample temperature, enrichment time (decay time), enrichment temperature (trap low temperature), desorption temperature (trap high temperature), desorption time, dry purge time, number of trap cycles, split/splitless injection and salt addition were optimized. For the standard solution 1, the following optimal parameters were determined: sample temperature at 80 ◦ C, decay time at 1.2 min, trap low temperature at 45 ◦ C, trap high temperature at 375 ◦ C, dry purge time at 7.5 min (nitrogen), one trap cycle, splitless injection, and addition of salt (Na2 SO4 without water). Further optimal headspace trap parameters are: needle

K. Schulz et al. / J. Chromatogr. A 1145 (2007) 204–209

207

Table 1 Results of method validation according to a German standard procedure (DIN 32645) comparing static headspace with headspace trap enrichment. Static headspace

Methanol 1-Propanol 2-Butanol Isobutanol 1-Butanol 2-/3-Methyl-1butanol Ethyl acetate Ethyl lactate 1-Hexanol Benzaldehyde Ethyl octanoate

Headspace trap

Linear range [␮g/l]

R2

100–16,000 30–10,000 30–10,000 30–10,000 30–10,000 100–20,000

0.999 0.999 0.999 0.999 0.999 0.999

150–10,000 300–25,000 150–10,000 100–10,000 600–10,000

0.999 0.999 0.999 0.999 0.997

LODa

LOQa

Precisionb

Precisionb

[␮g/l]

[␮g/l]

intraday [%]

interday [%]

574 66.0 114 88.0 129 161 202 645 350 411 1761

860 98.0 170 132 193 240 302 964 524 615 2617

Linear range [␮g/l]

R2

LODa [␮g/l]

LOQa [␮g/l]

Precisionb intraday [%]

Precisionb interday [%]

2.3 2.0 2.4 4.6 2.7 3.6

4.7 3.1 4.8 3.2 3.1 4.2

30–16,000 10–4,000 4–4,000 4–4,000 10–4,000 10–8,000

0.999 0.998 0.999 0.999 0.999 0.999

110 20.0 8.0 7.0 22.0 28.0

165 30.0 11.0 11.0 33.0 41.0

4.4 2.0 1.7 4.9 1.9 4.1

7.1 3.3 2.4 5.0 2.8 4.9

3.3 7.6 7.0 4.3 2.0

5.6 8.1 12.6 5.9 4.7

10–10,000 20–25,000 75–10,000 150–10,000 150–10,000

0.999 0.999 0.990 0.999 0.997

32.0 149 1072 256 1119

48.0 223 1601 383 1672

3.6 7.9 9.0 2.2 3.6

5.2 8.4 12.9 4.9 5.7

a

Limit of detection and quantitation were determined by establishing a specific calibration curve from samples containing the analyte in the range of LOQ. The limits were calculated from the residual standard deviation of the regression line. b Precisions are expressed as RSD [%], intraday (n = 7), interday (n = 5).

temperature at 90 ◦ C, transferline temperature at 95 ◦ C, trap hold time at 10 min, desorption time at 10 min, thermostatting time at 20 min, pressurization time at 3 min, column pressure at 90 kPa, vial pressure at 180 kPa and desorption pressure at 100 kPa. On this basis, the statistical data of the analytes were determined according to DIN 32645 (Table 1). With the help of the trap enrichment, better statistical data could clearly be determined. For the standard solution 2, the following optimal parameters were determined: sample temperature at 70 ◦ C, decay time at 1.0 min, trap low temperature at 55 ◦ C, and dry purge time at

1.0 min (nitrogen). All other optimal headspace trap parameters are the same as described above. Chromatograms of all standard substances acquired with and without trap enrichment are shown in Fig. 2. The advantage of the trap enrichment is obvious as significantly higher peak areas were found by using the HS-trap. Using only one trap extraction cycle, the peaks were 33–55 times higher than they were in the static headspace mode. For the purpose of spirit analysis, this one cycle provided adequate sensitivity. If a higher sensitivity were required, e.g. in the case of blood analysis of drinkers of

Fig. 2. Comparison between chromatograms of a standard sample (1-propanol, 2-butanol, 2-butanone, isobutanol, 1-butanol, 3-methyl-1-butanol, 2-methyl-1-butanol, ethyl acetate, ethyl lactate, benzaldehyde and ethyl octanoate each 2 mg/l; methanol 16 mg/l, acetone 10 mg/l and tert-butanol (internal standard) 1 mg/l) acquired using static headspace (lower chromatogram) and headspace trap (upper chromatogram). (1 = methanol, 2 = ethanol, 3 = acetone, 4 = internal standard tert-butanol, 5 = 1-propanol, 6 = ethyl acetate, 7 = 2-butanone, 8 = 2-butanol, 9 = isobutanol, 10 = 1-butanol, 11 = 2- and 3-methyl-1-butanol, 12 = ethyl lactate, 13 = benzaldehyde, 14 = ethyl octanoate).

208

K. Schulz et al. / J. Chromatogr. A 1145 (2007) 204–209

alcoholic beverages to substantiate claims of drinking, up to four trap-enrichment cycles (so-called pulsed headspace extraction and trap) can be used to achieve even lower detection limits. It was found that the trap enrichment is necessary only for the minor compounds like ethyl lactate, 1-hexanol, benzaldehyde, and ethyl octanoate, whereas the major volatiles can be adequately analyzed in the static headspace mode (as can be derived from the validation data in Table 1). A great advantage of this method over SPME is therefore the fact that the trap system can be easily used with or without enrichment (i.e. in normal static mode or with trap) leading to a great dynamic linear range (e.g. 30–16,000 ␮g/l for methanol). This change of measuring mode can be done using software control, whereas for SPME autosamplers, modifications of the hardware are required to switch between SPME and headspace mode. Headspace micro-extraction techniques (especially those based on absorption on polydimethylsiloxane phase materials) have proven that the analysis of alcoholic beverages is restricted because the excess of ethanol leads to competitive binding on the coating and to a decrease in the response of the target analytes [3,13]. Using the HS-trap relying on absorption into a bulk phase, this effect was not detected, and all analytes could be reliably determined. In the case of alcoholic beverages, the different extraction principle, combined with a higher volume of sorbent and the higher surfaces inside the packed column, therefore present an additional advantage over micro-extraction techniques. Water management in the system is a critical issue and demands careful consideration. Even if headspace extraction is used, Griffith et al. reported that it is still necessary to reduce the amount of water transferred through the trap to the analytical column, because some degradation of chromatography will occur at this level [9]. However, too high trap temperatures might lead to analyte losses so that a compromise between maximization of

analyte enrichment and minimization of water retention must be found. In the case of spirit analysis, dry purging at a temperature above ambient, but well below the desorption temperature, was found to vent water successfully from the system and minimize analyte loss before the sample was transferred to the analytical column. The applicability of HS-trap in the analysis of real samples was tested with different spirits. In Fig. 3, the chromatogram of a typical German fruit spirit is shown. Due to the high sensitivity of the method, the samples can be diluted with water, which might also suppress possible matrix and co-solvent effects. Calibration of the method was performed using standard solutions 1 and 2 in the concentration range of 1.0–4000 ␮g/l (methanol 4.0–16,000 ␮g/l and ethyl lactate, 1-hexanol, benzaldehyde and ethyl octanoate 1.0–25,000 ␮g/l). Statistical data for the calibration functions, calculated according to a German standard procedure (DIN 32645), are summarized in Table 1. Limits of detection (LOD) for the compounds of standard solution 1 in the range of 7.0–149 ␮g/l with trap enrichment and significantly better than classical static headspace (LOD: 66.0–570 ␮g/l). Limits of detection for the compounds of standard solution 2 were in the range of 149–1119 ␮g/l with trap enrichment and were also significantly better than classical static headspace (LOD: 350–1761 ␮g/l). This proves the excellent sensitivity of the method. The correlation coefficients of the calibration curve emphasize a good linearity in the investigated concentration range. The relative standard deviations (RSD), both intraday and interday, show good reproducibility of the method. Finally, 30 authentic spirit samples submitted to the CVUA Karlsruhe in the context of official food control were analyzed by both the new headspace procedure and the European reference method. The correlation between both methods was excellent (p < 0.0001 for all compounds), with correlation coefficients higher than 0.98 (Table 2). Only 1-hexanol showed a lower cor-

Fig. 3. Comparison between chromatograms of an authentic sample (German fruit spirit) acquired using static headspace (lower chromatogram) and headspace trap (upper chromatogram). (1 = ethyl lactate, 2 = 1-hexanol, 3 = benzaldehyde, 4 = ethyl octanoate).

K. Schulz et al. / J. Chromatogr. A 1145 (2007) 204–209 Table 2 Comparison between the headspace sampling and the reference procedure (direct injection) Analyte

R (n = 30)

Methanol 1-Propanol 1-Butanol 2-Butanol Isobutanol 2-/3-Methyl-1-butanol Ethyl acetate 1-Hexanola Ethyl lactatea Benzaldehydea

0.996 0.999 0.981 0.999 0.994 0.990 0.979 0.888 0.988 0.983

a

Substances determined with headspace trap enrichment.

relation, due to the fact that the HS-trap was able to detect it in cases that were below the limit of detection of the reference procedure. (No correlation could be calculated for ethyl octanoate because it was not detected in the majority of samples using both procedures.) The main advantage of HS-trap in analysis of alcoholic beverages lies in the application to such minor analytes or products with only low concentrations of volatile compounds (e.g. vodka). 4. Conclusions Headspace-trap technology effectively enhances static headspace techniques because it is easy to apply and automate. The selection of adsorbent material allows for optimal adjustment to specific substances or substance classes. In contrast to established micro-extraction techniques, the higher amount of sorbent material and the larger surfaces in the packed HS-traps achieve an optimized extraction and show a higher durability,

209

as no parts have to be moved at any time during the procedure (in contrast to fragile SPME fibres, which must be moved during sampling). Condensed water or solvents can be removed using the dry purge function, which leads to reduced chromatographic background. A very simple sample preparation is required, which corresponds to the sample preparation required by static headspace sampling. With trap enrichment, significantly greater extraction yields are obtained. The research thus shows that HS-trap can be very successfully used for the determination of volatile compounds in spirits. References [1] European Commission, Off. J. Eur. Commun. L333 (2000) 20. [2] E.A. Nonato, F. Carazza, F.C. Silva, C.R. Carvalho, Z. de L. Cardeal, J. Agric. Food Chem. 49 (2001) 3533. [3] D.W. Lachenmeier, U. Nerlich, T. Kuballa, J. Chromatogr. A 1108 (2006) 116. [4] K. Lachenmeier, F. Mußhoff, B. Madea, E.M. Sohnius, W. Frank, D.W. Lachenmeier, Deut. Lebensm.-Rundsch. 101 (2005) 187. [5] N. Sch¨afer, D.W. Lachenmeier, Deut. Lebensm.-Rundsch. 101 (2005) 534. [6] M.A. Jochmann, M.P. Kmiecik, T.C. Schmidt, J. Chromatogr. A 1115 (2006) 208. [7] A. Tipler, C. Mazza, US Patent Application No. US2006/0099718 A1 (2006). [8] F. Musshoff, D.W. Lachenmeier, L. Kroener, B. Madea, J. Chromatogr. A 958 (2002) 231. [9] H. Griffith, L. Marotta, A. Tipler, Z. Grossner, Symposium on Air Quality Measurement Methods and Technology 2004 Proceedings, Air & Waste Management Association Pittsburgh, PA, USA, 2004, p. 267. [10] H. Grecsek, LC–GC Suppl. (2005) 47. [11] H. Grecsek, LC–GC Suppl. (2005) 45. [12] H. Helms, U. Felix, LaborPraxis 28 (2004) 46. [13] J.-P. Dufour, R. Wierda, P. Delbecq, M. Leus, P. Silcock, Proceedings of the 10th Weurman Flavor Research Symposium Editions Tec & Doc, Paris, France, 2003, p. 584.