Characterization of aniseed-flavoured spirit drinks by headspace solid-phase microextraction gas chromatography–mass spectrometry and chemometrics

Talanta 72 (2007) 506–511

Characterization of aniseed-flavoured spirit drinks by headspace solid-phase microextraction gas chromatography–mass spectrometry and chemometrics J.M. Jurado a , O. Ballesteros b , A. Alc´azar a , F. Pablos a,∗ , M.J. Mart´ın a , J.L. V´ılchez b , A. Naval´on b a

Department of Analytical Chemistry, Faculty of Chemistry, University of Sevilla, C/Profesor Garc´ıa Gonz´alez, 1, E-41012 Sevilla, Spain b Research Group of Analytical Chemistry and Life Sciences, Department of Analytical Chemistry, University of Granada, Avda. Fuentenueva, s/n, E-18071 Granada, Spain Received 7 June 2006; received in revised form 25 October 2006; accepted 8 November 2006 Available online 8 December 2006

Abstract The volatile composition of aniseed-flavoured spirit drinks was studied by headspace solid-phase microextraction (HS-SPME) coupled with gas chromatography–mass spectrometry (GC–MS). The influence of the time, temperature, volume of sample and ionic strength on the extraction were considered. Several aniseed-flavoured spirit drinks, such as pastis, sambuca, anis and raki were analyzed. The major compounds found were trans-anethole (41.22–98%), cis-anethole (0.77–18.65%) and estragole (0.1–17.96%). ␥-Himachalene (0–28.07%) and ␣-himachalene (0–4.8%) were also present in anis and raki beverages. The compounds determined were used as chemical descriptors to differentiate the four types of beverages considered. Principal component analysis (PCA), linear discriminant analysis (LDA) and artificial neural networks (ANN) were used as chemometric tools for characterization purposes. © 2006 Elsevier B.V. All rights reserved. Keywords: Beverages analysis; Aniseed-flavoured spirit drinks; Solid-phase microextraction; Gas chromatography–mass spectrometry; Chemometrics; Principal component analysis; Linear discriminant analysis; Artificial neural networks

1. Introduction Aniseed-flavoured spirit drinks are produced by flavouring and distilling ethyl alcohol of agricultural origin with natural extracts of star anise (Illicium verum), anise (Pimpinella anisum), fennel (Foeniculum vulgare), or any other plant which contains the same principal aromatic constituent [1]. The main flavouring constituent is anethole, but other components like estragole, ␣- and ␥-himachalene and terpene hydrocarbons [2–5] are also present. Aniseed-flavoured spirit drinks are produced in several countries like France, Greece, Italy, Spain and Turkey, with different names such as pastis, ouzo, sambuca, anis and raki, respectively. The analysis of the volatile compounds in alcoholic beverages is usually carried out by gas chromatography (GC) though

∗

Corresponding author. Tel.: +34 95 4557173; fax: +34 95 4557168. E-mail address: [email protected] (F. Pablos).

0039-9140/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.talanta.2006.11.008

requires enrichment as a basis for identification and quantification. Thus, liquid–liquid extraction, static and dynamic headspace (HS) and solid-phase extraction have been commonly used for the analysis of spirits [6–11]. Solid-phase microextraction (SPME) is an alternative to these techniques that is easy to use, provides high sensitivity, good reproducibility and low cost. This technique eliminates use of organic solvents and substantially shortens the time of analysis. [12,13]. It has been successfully applied in studies for the characterization of wines and spirits [8,14–21]. The characterization of samples, using their chemical composition as input data, is of great interest for the identification of the geographical origin and authenticity of food products. Assessment of food samples origin has been mostly conducted through multivariate analysis in combination with pattern recognition techniques [22–24]. Multivariate analysis methods for data visualization, dimensionality reduction and sample classification include principal component analysis (PCA) and linear discriminant analysis (LDA), respectively. Artificial neural net-

J.M. Jurado et al. / Talanta 72 (2007) 506–511

works (ANN) are also powerful tools, especially when other statistical techniques are not able to predict complicated phenomena [25]. In this work, a HS-SPME/GC–MS method is used to determine the volatile components of aniseed-flavoured spirit drinks. These components have been used as input variables to apply pattern recognition techniques to differentiate several types of these beverages. 2. Experimental 2.1. Reagents and materials All reagents were analytical grade unless specified otherwise. Water was purified with a Milli-Q plus system (Millipore, Bedford, USA). Ethanol and sodium chloride were supplied from Panreac (Barcelona, Spain). Menthol (Merck, Darmstadt, Germany) was used as internal standard. A standard solution of 1000 mg L−1 was prepared in ethanol. A manual fiber holder for SPME was purchased form Supelco (Bellefonte, PA, USA). Four types of fibers were used: 100 ␮m polydimethylsiloxane (PDMS), 65 ␮m polydimethylsiloxane–divinylbenzene (PDMS/DVB), 75 ␮m carboxen–polydimethylsiloxane (CAR/PDMS) and 50 ␮m divinylbenzene– carboxen–polydimethylsiloxane (DVB/CAR/PDMS). These fibers were obtained from the same manufacturer and were conditioned under their specifications before use. Samples (N = 76) were obtained from local stores, belonging to four different types of aniseed-flavoured spirit drinks: anis (N = 46), pastis (N = 12), sambuca (N = 6) and raki (N = 12). All samples were contained in glass bottles and stored at 4 ◦ C until analysis. 2.2. HS-SPME procedure The influence of the variables, extraction temperature and time, volume of sample and salt concentration, was studied using a central composite experimental design. The extraction temperature was varied from 25 to 45 ◦ C, the extraction time between 10 and 50 min, the amount of NaCl from 0 to 0.5 g and the volume of sample percentage from 10% to 90%, for a total volume of 10 mL. Each variable had five levels and the experiments were divided in three blocks. A central point at 35 ◦ C, 30 min, 60% of sample volume and, 0.3 g of NaCl was done by duplicate in each of the three blocks considered. The HS-SPME conditions were established according to the results obtained in this study. Then, each sample was prepared taking a volume of 6 mL of aniseed drink, 0.18 g of NaCl and 1 mL of menthol solution into a 10-mL standard flask, adding water till the mark. The content of the standard flask was transferred to a 20-mL vial containing a magnetic stirring bar. The vial was hermetically sealed with PTFE faced silicone septum and placed in a thermostated block at 34 ◦ C on a magnetic stirrer (Agimatic-N, Selecta, Spain). The CAR/PDMS fiber was exposed to the headspace of the sample for 35 min. During this time, the sample was stirred at a constant speed of 300 rpm. After sampling, the fiber was removed

507

from the sample vial and inserted into the injection port of the GC, where thermal desorption of the analytes was carried out at 270 ◦ C during 2 min. Quantitative analysis was performed by using the internal normalization method. 2.3. Apparatus Chromatographic analysis was performed using an Agilent 6890 gas chromatograph interfaced to an Agilent 5973 Network mass spectrometer (Agilent Technologies, Little Falls, DE, USA). A capillary column ZB-5 (30 m × 0.25 mm i.d.; 0.25 ␮m film thickness) from Phenomenex (Jasco, Madrid, Spain) was used. Oven temperature program started at 40 ◦ C, heated at 5 ◦ C/min up to 150 ◦ C and held for 5 min. Helium (purity 99.999%) was used as carrier gas at a flow rate of 1 mL min−1 . A split ratio 10:1 was fitted. The injection was made in splitless mode for 2 min using a 0.75 mm liner at a temperature of 270 ◦ C, a SPME inlet guide and pre-drilled Thermogreen LB-2 septa from Supelco (Bellefonte, PA, USA). Transfer line temperature was 270 ◦ C. All mass spectra were acquired in electron impact mode (EI) at 70 eV using full scan with a scan range of 50–400 amu at a rate of 2.5 scans/s. Data acquisition and integration were carried out with the ChemStation chromatography software. The identification of the peaks was achieved through mass spectrometry by comparing mass spectra of the unknown peaks with those stored in the Wiley GC–MS library. 2.4. Multivariate analysis The ratio of the analyte area/internal standard area of the compounds was used as input data. A data matrix whose rows are the cases and columns the variables was prepared and used in the chemometric calculations. Statistica 7.0 software package (StatSoft Inc., 2004) was used for the statistical data analysis. 3. Results and discussion 3.1. Optimization of HS-SPME conditions To develop a suitable HS-SPME method optimization of several variables such as SPME fiber selection, extraction time and temperature, volume of sample and salt concentration is required. SPME is a process which depends on the equilibrium process involving partitioning of the analytes from the sample into the stationary phase. The type of coating was the first variable considered. Four different coatings were compared, 100 ␮m PDMS, 65 ␮m PDMS/DVB, 75 ␮m CAR/PDMS and 50 ␮m DVB/CAR/PDMS. All extractions were carried out using an extraction time of 30 min at 30 ◦ C. Higher recoveries and abundances for most of the target compounds were obtained with the fiber CAR/PDMS. The fiber maintained its performance for >100 extractions with between-day precision below 10%. To optimize extraction time and temperature, volume of sample and salt concentration a central composite experimental design was carried out. Values and intervals of the considered variables were included in Section 2.2. Response obtained for

508

J.M. Jurado et al. / Talanta 72 (2007) 506–511

trans- and cis-anethole, estragole, ␣- and ␥-himachalene was considered. Taking into account the ANOVA results obtained from the experimental design, the influence of the variables on the response for trans-, cis-anethole and estragole was not significant. In the case of ␣- and ␥-himachalene a significant influence of the variables was observed and maximum responses were obtained at 35 min, 34 ◦ C, 60% of volume of sample and 0.18 g of NaCl. Consequently, these were the selected extraction conditions.

detected. ␣- and ␤-himachalene were present in anis and raki samples and traces of ␥-himachalene and p-anisaldehyde were detected in pastis. Taking into account these results, differences in the composition of the volatile fraction of the aniseed-flavoured spirit drinks types can be assumed, and the detected compounds were used chemical descriptors to characterize the samples.

3.2. Analysis of volatile composition of aniseed-flavoured spirit drinks

PCA-based display methods allow a detailed study of the data trends. The three first PCs were calculated explaining up to 72.6% of the total variance. PC1 explains 39.2%, PC2 accounts for 24.8% and PC3 for 8.6%. Fig. 2 shows the scores plot of the samples in the three-dimensional space constructed with the three first PC. As can be seen, pastis, anis and raki samples appear almost grouped, and no clear separation can be observed. Otherwise, sambuca samples are grouped at the negative scores of PC1. Considering the factor loadings of the variables the most contributing ones were decane, limonene, ␣-terpinene, p-vinylanisole, dodecane, anisylacetone, (E)-5-tetradecene, ntetradecane, (EXO)-␣-bergamotene, ␤-caryophyllene, (ENDO)␣-bergamotene, ␣-humulene, ledene, bicyclogermacrene, ␦cadinene, Z-3-hexadecene, Z-7-hexadecene, hexadecane, transanethole, ␦-elemene, ␣-longipinene, ␣-ylangene, ␤-elemene,

Quantitation of the compounds present in the volatile fraction of the aniseed drinks was performed on the basis of their peak areas. All samples were analyzed three times. The ranges of percentages of the compounds in each type of aniseed drink are included in Table 1. Chromatograms of samples of pastis, sambuca, raki and anis are depicted in Fig. 1. More than 40 compounds were detected including propenyphenols, monoterpenes, sesquiterpenes and hydrocarbons. Propenylphenols like trans-anethole (41.22–98%), cis-anethole (0.77–18.65%) and estragole (0.10–17.96%) were the main components. Monoterpenes and sesquiterpenes like ␣- and ␤-pinene, limonene, ␤-bisabolene, zingiberene, ␣-, ␤- and ␥-himachalene were also

3.3. Characterisation of aniseed-flavoured spirit drinks

Fig. 1. Chromatograms corresponding to (a) anis, (b) pastis, (c) sambuca and (d) raki samples. See chromatographic conditions in Section 2.2. (1) ␣-Pinene; (2) ␤-pinene; (3) n-decane; (4) p-cymene; (5) limonene; (6) ␣-terpinene; (7) p-vinylanisole; (8) (Z)-2-dodecene; (9) estragole; (10) dodecane; (11) p-n-propilanisole; (12) cis-anethole; (13) p-anisaldehyde; (14) trans-anethole; (15) ␦-elemene; (16) ␣-longipinene; (17) ␣-ylangene; (18) ␣-copaene; (19) ␣-cubebene; (20) anisylacetone; (21) ␤-bourbonene; (22) ␤-elemene; (23) (E)-5-tetradecene; (24) tetradecane; (25) (EXO)-␣-bergamotene; (26) ␤-caryophyllene; (27) (ENDO)-␣-bergamotene; (28) ␣-himachalene; (29) ␣-humulene; (30) trans-␤-farnesene; (31) ␥-himachalene; (32) Ar-curcumene; (33) ledene; (34) bicyclogermacrene; (35) zingiberene; (36) ␤-himachalene; (37) ␤-bisabolene; (38) ␦-cadinene; (39) (Z)-3-hexadecene; (40) (Z)-7-hexadecene; (41) hexadecane.

J.M. Jurado et al. / Talanta 72 (2007) 506–511

509

Table 1 Composition (%) of volatile fraction of aniseed-flavoured spirit drinks Peak numbera

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41

Compound

␣-Pinene ␤-Pinene n-Decane p-Cymene Limonene ␣-Terpinene p-Vinylanisole (Z)-2-Dodecene Estragole Dodecane p-n-Propilanisole cis-Anethole p-Anisaldehyde trans-Anethole ␦-Elemene ␣-Longipinene ␣-Ylangene ␣-Copaene ␣-Cubebene Anisylacetone ␤-Bourbonene ␤-Elemene (E)-5-Tetradecene Tetradecane (EXO)-␣-Bergamotene ␤-Caryophyllene (ENDO)-␣-Bergamotene ␤-Himachalene ␤-Humulene trans-␤-Farnesene ␥-Himachalene Ar-Curcumene Ledene Bicyclogermacrene Zingiberene ␤-Himachalene ␤-Bisabolene ␦-Cadinene (Z)-3-Hexadecene (Z)-7-Hexadecene Hexadecane

Retention indexb

769 888 1000 1033 1040 1074 1160 1194 1198 1200 1209 1312 1322 1343 1363 1371 1382 1386 1387 1390 1390 1393 1396 1400 1413 1418 1435 1450 1456 1460 1482 1486 1493 1498 1501 1504 1515 1525 1563 1595 1600

Samples Anis

Pastis

Sambuca

Raki

0.00–0.26 0.00–0.15 – t 0.00–0.39 – – – 0.27–17.96 – 0.00–0.82 1.68–18.65 – 41.22–98.00 0.00–2.65 0.00–1.08 0.00–1.41 0.00–0.54 – – – 0.00–1.16 – – t t 0.00–1.17 0.00–4.80 – 0.00–1.68 0.00–28.07 0.00–2.02 – – 0.00–11.09 0.00–2.52 0.00–2.46 t – – –

– – – – – – 0.00–0.27 0.00–0.36 0.10–1.37 – – 0.77–10.36 0.28–8.86 79.22–97.52 0.00–0.16 – – – 0.00–0.11 0.08–0.41 – 0.00–0.18 0.00–0.66 t 0.00–0.54 0.00–2.97 0.00–2.09 – 0.00–0.37 0.00–0.30 t – – – – – – – t 0.00–0.36 –

– – 0.94–1.44 – 0.30–0.66 0.22–0.38 0.18–0.23 0.36–0.51 1.54–2.01 3.96–6.86 – 6.57–10.12 – 66.16–73.02 – – – 0.50–0.68 – 0.17–0.56 – 0.00–0.13 0.50–0.98 0.46–1.29 0.56–0.73 2.08–2.77 2.83–3.81 – 0.38–0.50 0.23–0.31 – – 0.26–0.32 0.11–0.15 – – 0.19–0.24 0.27–0.34 0.12–0.26 0.26–0.41 0.44–0.87

– – – t t – – 0.00–0.48 0.97–8.34 t – 7.50–11.39 – 56.78–89.95 0.00–0.84 t 0.00–0.44 – – – 0.00–0.11 0.00–0.26 0.00–0.85 0.06–0.30 – – – 0.00–1.88 – 0.00–0.34 0.00–18.25 – – – 0.00–0.63 0.00–1.32 0.00–0.49 t 0.00–0.24 0.00–0.59 –

t, traces (<0.06%). a See chromatograms of Fig. 1. For experimental conditions see Section 2. b Relative to C –C n-alkanes determined using ZB-5 capillary column. 7 16

␣-himachalene, trans-␤-farnesene, ␥-himachalene, zingiberene, ␤-himachalene, ␤-bisabolene, ␣-cubebene, ␣-pinene, ␤-pinene, ␤-bourbonene. Considering that highly correlated variables provide similar information a Pearson correlation study was carried out to find out variables that were strongly correlated. From this study, the variables limonene, anisylacetone, (E)-5-tetradecene, n-tetradecane, (ENDO)-␣-bergamotene, ␦cadinene, trans-anethole, trans-␤-farnesene, ␥-himachalene, ␤-pinene, ␣-cubebene and ␤-bourbonene were considered as the most significant ones. In order to obtain suitable classification rules forward LDA was carried out by using the above mentioned variables. Three

discriminant functions (DF) were obtained. Fig. 3 shows the distribution of the samples in the plane generated by the two first DF. As can be seen, a good separation of sambuca class was achieved. For pastis, raki and anis classes some tendencies can be seen, though there is not a clear separation. The recognition ability, according to the a posteriori probabilities was of 100% for sambuca and anis, 83.3% for pastis and 91.7% for raki. The leave one out method [26] was used as cross-validation procedure to evaluate the classification performance. The prediction ability of the constructed model was 100% for sambuca and anis, 91.7% for raki and 83.3% for pastis. According to LDA results the most discriminant variables were ␦-cadinene, anisylacetone, ␤-bourbonene, E-5-tetradedecene, ␣-cubebene, n-tetradecane,

510

J.M. Jurado et al. / Talanta 72 (2007) 506–511

applying the sensitivity and specificity parameters [28,29] to the test set. Both parameters were 100% indicating that MLP models the class distribution better than LDA. The possible cause of this behaviour can be the intrinsically non-linear nature of the class distribution. 4. Conclusions HS-SPME/GC–MS has been used to analyze the volatile fraction of aniseed-flavoured spirit drinks. CAR/PDMS SPME fiber was used and the extraction conditions were optimized using a central composite experimental design. Pastis, sambuca, raki and anis samples were analyzed. Propenylphenols, monoterpenes, sesquiterpenes and hydrocarbons were present. Propenylphenols such as trans-anethole, cis-anethole and estragole were the main components. Other components like monoterpenes, sesquiterpenes and hydrocarbons were also present. Using a chemometric approach, the volatile fraction provides a suitable method to differentiate aniseed beverages, the best results being obtained with MLP. Fig. 2. Scores plot of the aniseed-flavoured spirit drinks in the three-dimensional space of the firsts PCs: (P) pastis; (S) sambuca; (R) raki; (A) anis.

limonene, (ENDO)-␣-bergamotene, ␥-himachalene and transanethole. Considering that the linear model did not provide a complete solution to the classification problem, non-linear approach such as ANN was used. Using the most discriminant variables extracted from LDA as inputs and the class of each case as output, a multilayer perceptron ANN (MLP) [27] was used to solve the classification problem. The data set was divided in three subsets, training (N = 38), verification (N = 19) and test (N = 19) sets, respectively. A three layer MLP was applied. The architecture was 10 neurons in the input layer, 13 in the hidden and four in the output one. The network was trained by back propagation during 1000 epochs with a learning rate and momentum of 0.16 and 0.52, respectively. The classification procedure was validated by

Fig. 3. Scatter plot of the aniseed-flavoured spirit drinks in the plane of the two first discriminant functions.

References [1] Council Regulation (EC) No. 1576/89 of 29 May 1989, Official Journal L160, 12/06/1989, P. 0001. [2] V.M. Rodrigues, P.T.V. Rosa, M.O.M. Marques, A.J. Petenate, M.A.A. Meireles, J. Agric. Food Chem. 51 (2003) 1518. [3] A. Besharati-Seidani, A. Jabbari, Y. Yamini, Anal. Chim. Acta 530 (2005) 155. [4] P.M. Santos, A.C. Figueiredo, M.M. Oliveira, J.G. Barroso, L.G. Pedro, S.G. Deans, A.K.M. Younus, J.J. Scheffer, Phytochemistry 48 (1998) 455. [5] N. Tabanca, B. Demirci, T. Ozek, N. Kirimer, K.H. Can Baser, E. Bedir, I.A. Khan, D.E. Wedge, J. Chromatogr. A 1117 (2006) 194. [6] O. Lablanquie, G. Snakkers, R. Cantagrel, G. Ferrari, Anal. Chim. Acta 458 (2002) 191. [7] H. Guichard, S. Lemesle, J. Ledauphin, D. Barillier, B. Picocheid, J. Agric. Food Chem. 51 (2003) 424. [8] S.E. Ebeler, M.B. Terrien, C.E. Butzke, J. Sci. Food Agric. 80 (2000) 625. [9] T. Fern´andez-Garc´ıa, M.E. Mart´ın, A. Casp, Z. Lebens. Unter. Forsch. 206 (1998) 414. [10] V. Andrea, N. Nadia, R.M. Teresa, A. Andrea, J. Agric. Food Chem. 51 (2003) 4978. [11] S.C. Di´eguez, M.L.G. de la Pe˜na, E.F. G´omez, J. Agric. Food Chem. 51 (2003) 7385. [12] H. Lord, J. Pawliszyn, J. Chromatogr. A 885 (2000) 153. [13] C. Dietz, J. Sanz, C. Camara, J. Chromatogr. A 1103 (2006) 183. [14] R.F. Alves, A.M.D. Nascimento, J.M.F. Nogueira, Anal. Chim. Acta 546 (2005) 11. [15] V.A. Watts, C.E. Butzke, R.B. Boulton, J. Agric. Food Chem. 51 (2003) 7738. [16] W. Wardencki, P. Sowinski, J. Curylo, J. Chromatogr. A 984 (2003) 89. [17] B. Vallejo-C´ordoba, A.F. Gonz´alez-Cordova, M.C. Estrada-Montoya, J. Agric. Food Chem. 52 (2004) 5567. [18] J. Pino, M.P. Mart´ı, M. Mestres, J. P´erez, O. Busto, J. Guasch, J. Chromatogr. A 954 (2002) 51. [19] G. Fitzgerald, K.J. James, K. MacNamara, M.A. Stack, J. Chromatogr. A 896 (2000) 351. [20] C. Da Porto, L. Pizzale, M. Bravin, L.S. Conte, Flavour Frag. J. 18 (2003) 66. [21] E.A. Nonato, F. Carazza, F.C. Silva, C.R. Carvalho, Z.L. Cardenal, J. Agric. Food Chem. 49 (2001) 3533.

J.M. Jurado et al. / Talanta 72 (2007) 506–511 [22] M.S. Valdenebro, M. Le´on-Camacho, F. Pablos, A.G. Gonz´alez, M.J. Martin, Analyst 124 (1999) 999. [23] D. Gonz´alez-Arjona, V. Gonz´alez-Gallero, F. Pablos, A.G. Gonz´alez, Anal. Chim. Acta 381 (1999) 257. [24] J.M. Jurado, A. Alc´azar, F. Pablos, M.J. Mart´ın, A.G. Gonz´alez, Talanta 66 (2005) 1350. [25] J. Zupan, J. Gasteiger, Neural Networks in Chemistry and Drug Design, Wiley–VCH, Weinhein, 1999.

511

¨ [26] R. Henrion, G. Henrion, Uberwatchte Klassifikation in Multivariate Dateanalysen, Springer-Verlag, Berlin, 1995. [27] C. Bishop, Neural Networks for Pattern Recognition, University Press, Oxford, 1995. [28] M. Forina, C. Armanino, R. Leardi, G. Drava, J. Chemometr. 5 (1991) 435. [29] D. Gonz´alez-Arjona, G. L´opez-P´erez, A.G. Gonz´alez, Chemometr. Intell. Lab. Syst. 57 (2001) 133.