Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 148 (2015) 131–137

Contents lists available at ScienceDirect

Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy journal homepage: www.elsevier.com/locate/saa

Review Article

Determination of geographical origin of alcoholic beverages using ultraviolet, visible and infrared spectroscopy: A review Veronika Urícˇková ⇑, Jana Sádecká Institute of Analytical Chemistry, Faculty of Chemical and Food Technology, Slovak University of Technology, Radlinského 9, 812 37 Bratislava, Slovak Republic

h i g h l i g h t s

g r a p h i c a l a b s t r a c t

 Geographical classification of wines

and distilled drinks.  UV, Vis, NIR and MIR spectroscopic

techniques were used.  PCA–LDA, SVM, SIMCA and PLS-DA

was applied.  Average correct classification higher

than 82%.

a r t i c l e

i n f o

Article history: Received 12 January 2015 Received in revised form 10 March 2015 Accepted 27 March 2015 Available online 3 April 2015 Keywords: Beverages Ultraviolet and visible spectroscopy Infrared spectroscopy Pattern recognition methods Geographical origin

a b s t r a c t The identification of the geographical origin of beverages is one of the most important issues in food chemistry. Spectroscopic methods provide a relative rapid and low cost alternative to traditional chemical composition or sensory analyses. This paper reviews the current state of development of ultraviolet (UV), visible (Vis), near infrared (NIR) and mid infrared (MIR) spectroscopic techniques combined with pattern recognition methods for determining geographical origin of both wines and distilled drinks. UV, Vis, and NIR spectra contain broad band(s) with weak spectral features limiting their discrimination ability. Despite this expected shortcoming, each of the three spectroscopic ranges (NIR, Vis/NIR and UV/Vis/NIR) provides average correct classification higher than 82%. Although average correct classification is similar for NIR and MIR regions, in some instances MIR data processing improves prediction. Advantage of using MIR is that MIR peaks are better defined and more easily assigned than NIR bands. In general, success in a classification depends on both spectral range and pattern recognition methods. The main problem still remains the construction of databanks needed for all of these methods. Ó 2015 Elsevier B.V. All rights reserved.

Contents Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Geographical origin of wines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . UV/Vis/IR spectra of wines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Classification of wines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Geographical origin of distilled drinks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Brandy and Cognac . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

⇑ Corresponding author. E-mail address: [email protected] (V. Urícˇková). http://dx.doi.org/10.1016/j.saa.2015.03.111 1386-1425/Ó 2015 Elsevier B.V. All rights reserved.

132 132 132 134 136 136

V. Urícˇková, J. Sádecká / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 148 (2015) 131–137

132

Scotch whisky . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Other distilled drinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Compliance with ethics requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Introduction The European Union Regulation establishes the rules for protecting designations of origin and geographical indications for foodstuffs. A protected designation of origin (PDO) indicates that the product must be both produced and processed within the defined geographic area. To attain protected geographical indication (PGI), the product must be produced or processed in the geographical area (either or both). Therefore, the PGI allows greater flexibility than the PDO. In many cases, the EU PDO/PGI system works parallel with the system used in the specified country. A powerful method for determining the geographical origin of food products is pattern recognition methods of the data provided by analytical instruments. Chromatographic methods are relatively expensive, time-consuming and require highly skilled operators. Common approach is multi-elemental analysis followed by stable-isotope ratio methods. The best results come from combining the two techniques; however, implementation of this strategy is quite difficult in routine analysis. Recently, attention has focused on the development of non-invasive and non-destructive instrumental techniques such as ultraviolet (UV), visible (Vis), near infrared (NIR) and mid infrared (MIR) spectroscopy [1–8]. Large amounts of spectral data, containing useful analytical information, noise, variabilities, uncertainties and unrecognized features, are usually obtained from spectroscopic instruments. Thus, pattern recognition methods are required to extract as much relevant information from spectral data as possible. Pattern-recognition methods are subdivided into supervised and non-supervised methods. Non-supervised methods do not require any a priori knowledge about the group structure in the data, but instead produces the grouping, i.e. clustering, itself. This type of analysis is often very useful at an early stage of an investigation and can be performed with simple visual techniques, such as hierarchical cluster analysis (HCA) or principal component analysis (PCA) [2]. When employing HCA, the original data are separated into a few general groups, each of which is further divided into still smaller groups until finally the individual objects themselves remain. The results are presented in the form of dendograms to facilitate the visualization of sample relationships [2]. PCA is usually the first step in spectroscopic data exploration. The aims of performing a PCA are two-such as fold. Firstly, PCA reduces the dimensions of the spectral dataset by explaining a large part of the variance using synthetic factors, called principal components (PCs). Therefore, the whole range of wavelengths can be compressed into the first few PCs. Secondly, PCA performed on spectral data makes it possible to draw similarity maps of the samples and to get spectral patterns. Classification of objects is done by constructing similarity maps of the samples, using PCs chosen by the researcher. The spectral patterns corresponding to the PCs provide information about the characteristic peaks which are the most discriminating for the samples observed on the similarity maps [9]. With supervised methods, the number of groups is known in advance and representative samples of each group are available. This information is used to develop a suitable discrimination rule or discriminant function with which new, unknown samples can be assigned to one of the groups. The commonly used supervised methods are linear discriminant analysis (LDA), support

136 136 136 136 137 137

vector machine (SVM), soft independent modeling of class analogy (SIMCA) and partial least squares discriminant analysis (PLS-DA). LDA is concerned with determining the so-called discriminant functions as linear combinations of the descriptors which best separate the groups according to minimization of the ratio of within group and between-group sum of squares. LDA requires that the number of variables (wavelengths) must be smaller than the number of samples in each group. Consequently, large spectral datasets with few samples cannot be analyzed using LDA. Combining LDA with a PCA overcomes this problem [10]. SVM maps the sample data with specific kernel functions to a higher dimensional feature space to linearize the boundary and generate the optimal separating hyperplane. There are number of kernels that can be used in SVM models. These include linear, polynomial, radial basis function and sigmoid. SVM is still effective in cases where the number of variables is greater than the number of samples. If the number of variables is much greater, the method is likely to give poor performances [11]. In PLS-DA, the standard PLS algorithm can be used and group labels can be given for the dependent vector. In the two-group case, usually the values of the dependent variable are given 1 for one group and 0 or 1 for the other group and a PLS1 algorithm is used. In case of more than 2 groups, dummy variables are defined and a PLS2 algorithm is used. PLS-DA is well suited to deal with a much larger number of variables than samples and with correlated variables [12]. SIMCA builds a distinct confidence region around each class after applying PCA. New measurements are projected in each PCs space that describes a certain class to evaluate whether they belong to it or not. The classification of a sample in one or several classes, or in none of them, is possible with SIMCA, while discriminant methods only permits to classify a sample in a unique class. Another advantage of SIMCA is that there is no restriction on the number of variables [13]. This review covers UV, Vis, NIR and MIR spectroscopy applications for determining geographical origin of alcoholic beverages, namely wines and distilled drinks. Table 1 provides a summary of the application of spectroscopic and pattern recognition methods done country by country, while Table 2 shows a prediction results for models developed using different spectral regions, regardless of geographical origin. Regarding supervised methods (Tables 1 and 2), LDA was based on the first PCs, SIMCA was used on PCA classes, while PLS-DA and SVM were applied over the original spectra. Table 3 provides the most important variables or regions related with the differences in the data sets obtained by PCA. Geographical origin of wines UV/Vis/IR spectra of wines The UV/Vis absorption spectra of wine samples are similar in profile, but differ in the absorption at the measurement wavelengths. The highest absorbances are observed at 202 and 230 nm. These are associated with carboxyl groups of organic acids. Moreover, the UV/Vis spectra contain information regarding phenolic compounds, e.g., benzoic acids (235–305 nm),

V. Urícˇková, J. Sádecká / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 148 (2015) 131–137

133

Table 1 Application of UV/Vis/IR and pattern recognition methods to predict geographical origin of wines and distilled drinks. Geographical origina

Spectral regionb

Pattern recognition methods

Most discriminating variablesb,c

Quality of classificationb,c,d

Refs.

Argentina; 35 Sauvignon blanc wines from regions Río Negro (10 samples), Mendoza (20) and San Juan (5) Argentina; 80 Torrontés wines from regions Mendoza, San Juan, Salta and Rio Negro Australia; 98 Shiraz wines from four South Australian regions McLaren Vale (29), Barossa Valley (19), Clare Valley (17) and Coonawarra (23), and one region of Western Australia (10)

UV/Vis (200–500 nm)

PCA, PCA– LDA, PLSDA PCA, PCA– LDA, PLSDA PCA, PCA– LDA, SIMCA

300–350 nm (pH = 10.2)

PCA–LDA and PLS-DA: 100% classification

[30]

300–350 nm (pH = 10.2)

PCA–LDA and PLS-DA: 100% classification

[31]

MIR (910–1488 cm 1), ethanol, glycerol, glucose, fructose, organic acids, aldehydes

[25]

China; 38 rice wines from regions Shaoxing (30) and Jiashan (8) China; 54 rice wines from regions Fujian (11), non-Shaoxing (20) and Shaoxing (23) France; 338 wines from areas Gaillac, Beaujolais and Touraine Spain; white wines from the DO La Mancha, zones Quintanar de la Orden (A), Fuente de Pedro Naharro (B) and Mota del Cuervo (C) Spain; red wines from the DO La Mancha, zones Quintanar de la Orden and Villacanas Spain; 82 white from the DO La Mancha (15), Madrid (20), Penedés (12), Rioja (19) and Valdepenˇas (16)

NIR (800–2500 nm)

PCA–LDA: UV (46%), Vis (62%), NIR (60%) and MIR (73%); PCA–LDA using MIR: Barossa Valley (63%), Clare Valley (65%), McLaren Vale (52%), Coonawarra (90%), Western Australia (100%), overall 73%; SIMCA using MIR: Barossa Valley (90%), Clare Valley (89%), McLaren Vale (100%), Coonawarra (87%), Western Australia (100%), overall 95% PLS-DA: 100% PCA–LDA and PLS-DA: 100%

[20]

800–1800 cm 1, phenolic compounds A and B: UV, hydroxycinnamic acids; B and C: Vis 620–640 nm, phenolics 500 and 560 nm, anthocyanins

PLS-DA: Gaillac (71%), Beaujolais (90%), Touraine (97%), average 85% SIMCA: average 90%

[27]

SIMCA: average 80%

[32]

SVM: La Mancha (67%), Madrid (50%), ˇ as Penedés (100%), Rioja (100%), Valdepen (71%), average 78%; SIMCA: average 44%; PLS-DA: average 63% SVM: La Mancha (67%), Madrid (67%), ˇ as Penedés (100%), Rioja (75%), Valdepen (71%), Ribera Del Duero (71%), Toro (100%), Somontano (78%), average 79%; SIMCA: average 42%; PLS-DA: average 49% SIMCA: overall 78%

[33]

PCA–LDA: UV/Vis/NIR (86%), Vis/NIR (82%), Vis (81%), NIR (81%), UV(68%), UV/ Vis(67%); SVM: UV/Vis/NIR (84%), Vis/NIR (82%), Vis (83%), NIR (85%), UV(70%), UV/ Vis(83%); PLS-DA: New Zealand (80%), Australia (53%), average 67% PLS-DA: New Zealand, NIR (80%), MIR (93%), NIR/MIR (93%); Australia, NIR (73%), MIR (86%), NIR/MIR (93%); average NIR (76%), MIR (90%), NIR/MIR (93%)

[21]

UV/Vis (200–500 nm)

UV/Vis/NIR (400– 2500 nm), MIR (400– 4000 cm 1)

MIR (800– 1800 cm 1) UV/Vis (300–800 nm)

PCA, PLSDA PCA, PCA– LDA, PLSDA PCA, PLSDA PCA, SIMCA

UV/Vis (300–800 nm)

PCA, SIMCA

UV/Vis (200–800 nm)

SVM, SIMCA, PLSDA

240, 260, 280, 310, 340, 360 and 390 nm

Spain; 153 red wines from the DO La Mancha (9), Madrid (20), Penedés (14), Rioja (20), Valdepenˇas (23), Ribera Del Duero (31), Toro (15) and Somontano (21) Spain; 60 red and 60 white wines, origin was known for 40 white wines, the DO La Mancha, zones Quintanar de la Orden (15) and Fuente de Pedro Naharro (25) Spain; 33 wines from the DO Rías Baixas, zones Condado (10), Salnés (10), Rosal (8) and Ribeira do Ulla (5)

UV/Vis (200–800 nm)

SVM, SIMCA, PLSDA

290, 330, 360, 490, 540, 610, 650, 760 and 800 nm

PCA, SIMCA

1150–1300 and 2300– 2400 cm 1

UV/Vis/NIR (190– 2500 nm)

PCA, PCA– LDA, SIMCA, SVM

UV/Vis/NIR (190–2500 nm), Vis/NIR (400–2500 nm), Vis (400–780 nm)

Australia (34) and New Zealand (30); Sauvignon Blanc wines Australia (34) and New Zealand (30); Sauvignon Blanc wines

Vis (400–750 nm)

PCA, PLSDA PCA, PLSDA

–

Australia (36) and Spain (27); Tempranillo red wine; Australia: South Australia (17), New South Wales (2), Victoria (13) and Western Australia (4); Spain: Rioja (11), La Mancha (2), Ribera del Duero (10) and Toro (4) Australia (10), New Zealand (5) and France and Germany (8); 50 Riesling white wines

Vis/NIR (400– 2500 nm)

PCA, PLSDA, PCA– LDA

Vis/NIR (400– 2500 nm)

PCA, PLSDA, PCA– LDA

Phenolics, organic acids, sugars

France (60 red wines) and Germany (60 red wines)

Fluorescence (Ex: 250–350 nm, Em: 376 nm), Em: 275– 450 nm, Ex: 261 nm) NIR/MIR (926– 5012 cm 1), (10800–

PCA, PCAFDA

Excitation spectra, 250– 350 nm, (Em: 376 nm)

PCA

3700–4900 cm

France, Spain and South Africa; brandies and other distilled drinks, 34 samples:

NIR (800–2500 nm)

MIR (400–4000 cm

1

)

NIR (750–2500 nm), MIR (400–4000 cm 1)

1300–1650 nm, water, ethanol, sugars Water, ethanol, sugars

NIR, alcohol, sugars, aromatic; MIR (900– 500 cm 1), ethanol, glycerol, sugars, aromatic groups, organic acids, aldehydes Alcohol, sugars, phenolics, organic acids, wine pigments

1

[19]

[32]

[33]

[28]

[22] [22]

PLS-DA: Australia (100%), Spain (86%), average 93%; PCA–LDA: Australia (72%), Spain (85%), average 78%

[23]

PLS-DA: Australia (97%), New Zealand (80%), Europe (France, Germany) (70%), average 83%; PCA–LDA: Australia (86%), New Zealand (67%), France (67%), Germany (87%), average 77% The excitation spectra allowed a good discrimination

[24]

Cognacs (100%) from the other samples, Armagnacs (100%) from the other samples

[36]

[18]

(continued on next page)

V. Urícˇková, J. Sádecká / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 148 (2015) 131–137

134 Table 1 (continued) Geographical origina

Spectral regionb

Spanish brandies (12), French brandies (5), Cognacs (9), Armagnacs (4) and South African brandies (4) France; Cognac and non-Cognac groups, 151 samples: Cognacs (51), brandies, (24) whiskies (10), Armagnacs (8), rums (9), bourbons, (4) and counterfeit products (45) Scotland; 18 Scotch whiskies, 15 singlemalts from highlands (Highland and Speyside) (9) and islands (Islay, Mull and Arran) (6). Three blends with a variety of malts.

1995 nm)

Belgian (18) and Slovak (16) juniperflavored spirit drinks

Pattern recognition methods

Most discriminating variablesb,c

Quality of classificationb,c,d

Refs.

MIR of raw product, dry extract and dry phenolic extract (400–4000 cm 1)

PCA, PLSDA

Concatenated spectra from phenolic and dry extract (850–1800 cm 1)

Cognac and non-Cognac groups, PLS-DA: dry extract (87%), phenolic dry extract (96%), phenolic and dry extract (96%)

[37]

UV/Vis (290– 600 nm), NIR 1200– 1880 nm), fluorescence (Em: 450–700 nm, Ex: 404 nm) Synchronous fluorescence (220– 700 nm, Dk = 10– 100 nm)

PCA, PCA– LDA

Joint-data matrix: PC1, PC2 (UV/Vis), PC1, PC2, PC3 (NIR), PC1 (fluorescence)

PCA–LDA: 100%

[40]

PCA, HCA, PCA–LDA

250–350 nm collected at wavelength interval 20 nm

PCA–LDA: 99%

[42]

a

Number of samples is given in parentheses; DO, designation of origin. UV, ultraviolet; Vis, visible; NIR, near infrared; MIR, mid infrared; Ex, excitation wavelength; Em, emission wavelength; Dk, difference between emission and excitation wavelength. c PCA, principal component analysis; LDA, linear discriminant analysis; PLS-DA, partial least squares regression discriminant analysis; SIMCA, soft independent modeling of class analogy; SVM, support vector machines; FDA, factor discriminant analysis; HCA, hierarchical cluster analysis. d Correct classification (%) is given in parentheses. b

Table 2 Classification results for models developed using different spectral regions in wine analysis. Spectral region

Correct classification (%)a PCA–LDA

PLS-DA

SIMCA

SVM

Average (%)

UV Vis UV/Vis

46, 68 62, 81 100, 100, 67 60, 100, 81 73 – 82, 78, 77 86 77

– 67 100, 100, 63, 49 76, 100, 100

– – 90, 80, 44, 42 –

70 83 78, 79, 83 85

61 73 77

90, 85 93 93, 83 – 84

95, 78 – – – 71

– – 82 84 80

84 93 82 85 –

NIR MIR NIR/MIR Vis/NIR UV/Vis/NIR Average (%)

86

a PCA–LDA, principal component analysis linear discriminant analysis; PLS-DA, partial least squares regression discriminant analysis; SIMCA, soft independent modeling of class analogy; SVM, support vector machines.

hydroxycinnamic acids (227–245 nm and 310–332 nm), flavonols (250–270 nm and 350–390 nm), anthocyanins (267–275 nm and 475–545 nm) and catechins (280 nm). The difference between red and white wine is observed in the range 400–650 nm, where only red wine exhibits an absorbance peak associated with anthocyanins [14]. The geographical origin and methodology of winemaking strongly influences the phenolic composition of the final product and consequently UV/Vis spectra [15,16]. Several important wine components are fluorescent compounds most of which are polyphenols. All wines fluoresce in the emission region 300– 400 nm, when they are excited at wavelengths below 290 nm (e.g., benzoic acids, anthocyanins and flavanols). Another emission region is found between 350 and 450 nm, when the samples are excited at wavelengths longer than 290 nm (hydroxycinnamic acids and stilbenes) [17,18]. In general, no two wines show exactly the same IR absorption; therefore a unique fingerprint IR spectrum can be obtained for each wine. However, the IR absorption bands of the main constituents of wine (water and ethanol) may overlap the characteristic IR vibrations. In the NIR region (750–2500 nm, 12820–4000 cm 1), spectra

show various overlapping bands which are difficult to assign to specific chemical groups. These bands are the result of the first and second overtones, as well as of the combinations of two or more fundamental vibrations that occur in the MIR region. The highest absorbances are observed at about 1950 nm, which are related to O–H combinations (water and ethanol), O–H first overtones (water and ethanol) and C–H stretch first overtones. Additionally, absorptions are observed at 1450 nm related to O– H second overtones of both water and ethanol, in the 2200– 2300 nm region related to the C–H combinations, C–H overtones and O–H stretch overtones associated with ethanol, sugars, and phenolic compounds, and in the 1700–1800 nm region associated with sugars [19–24]. In the MIR region (2500–25000 nm, 4000– 400 cm 1), spectra contain well-defined peaks which can be assigned to specific chemical groups. These peaks are the result of the fundamental stretching, bending, and rotating vibrations of the sample molecules. Thus, the broad peak found between 3000 and 3500 cm 1 is mainly due to the stretching vibration of O H bond of water, peaks between 670 and 900 cm 1 can be attributed to aromatic C H out of plane (750–1000 cm 1) and in plane bending (950–1220 cm 1). Although the whole MIR spectral range is usually recorded for each sample, only the so-called fingerprint region (900–1600 cm 1) is sometimes considered. The peak found between 1400 and 1500 cm 1 is due to the deformation vibration of the carbon–carbon bonds in the phenolic groups, MIR absorption within the range of 950 and 1200 cm 1 is due to the presence of sugar functional groups, and peaks at 1618 and 1407 cm 1 corresponds to carboxylic acid, ester, or carbonyl groups [22,25–29]. Classification of wines No obvious differences in the UV, Vis and IR spectra between wine samples from different geographical origins are observed. Therefore, pattern recognition methods are necessary to extract the most relevant information from the data (Table 1). Regarding UV/Vis region, the relevant wavelengths for white wines fall within the range of 240–400 nm, relating to esters of hydroxycinnamic acids [30–33]. This region can be used to discriminate between white wines from different Argentinean regions [30,31], different zones [32] within the Spanish LaMancha DO, or one particular

V. Urícˇková, J. Sádecká / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 148 (2015) 131–137

135

Table 3 The results obtained by principal component (PC) analysis. Geographical origina

PC

The highest eigenvectors

Australia or Australia and New Zealand; wine

1 1 1

1045–1080 cm 1268 cm 1 1419–1454 cm

China; rice wine

1 1 2 2 3

China, rice wine

Australia and New Zealand; wine

Australia and Spain; wine

Australia, New Zealand, France and Germany; wine

a b

The features associated with the highest eigenvectorsb

Refs.

C–OH of ethanol, glycerol and sugars Aromatic groups of phenolics CO@O, C@C, C–H2, C–H3, organic acids, aldehydes

[25,22]

1860 nm 2276 nm 2247 nm 1390, 1490–1760 nm 2120–2300 nm

Comb. of stretch and def. of O–H group in water CH2 groups of ethanol C@C and C–H of ethanol and sugar O–H stretch of water and NH2 groups of N compounds

[20]

1 1 1 1 1 2

1450 nm 1884 nm 2064 nm 2336 nm 2370 nm 1410, 1884 nm

Comb. of stretch and def. of O–H group in water O–H stretch and C@O second overtone comb. Comb. of stretch and def. of O–H group Comb. of stretch and def. of C–H group C–H stretch O–H overtones, water

[19]

1

2180–2300 nm

[22]

2

1700, 2200–2300 nm

C–H comb. and O–H stretch overtones, alcohol content, sugars, aromatic rings and organic acids Ethanol and aromatic groups

1

2180–2300 nm

[23]

2 3

400–700 nm 450–700, 2200– 2300 nm

C–H comb. and O–H stretch overtones, alcohols, sugars, phenolics and organic acids Wine pigments Wine pigments, ethanol, and phenolics

1 1 1 2

C–H comb. and O–H stretch overtones O–H first overtones C–H first overtones

[24]

1

1

3

2250–2350 nm 1400–1460 nm 1660–1760 nm 1400–1460, 2250– 2350 nm 410–540 nm

France; Cognac and non-Cognac (dry extract)

1 1 2 3

850–1400 cm 1 1500–1650 cm 1 1500–1800 cm 1 850–1250 cm 1

Alcohol vibrations of carbohydrates Aromatic ring stretch Aromatic ring stretch, aromatic comb. bands Alcohol vibrations of carbohydrates

[37]

France; Cognac and non-Cognac (dry phen. extr.)

1 2

850–1700 cm 1 1500–1800 cm 1

Aromatic fingerprint and C–O valence vibrations Aromatic ring stretch, aromatic comb. bands

[37]

Scotland; whisky (UV/Vis)

1 2

UV band 300–400 nm

Shoulder at 350 nm

Scotland; whisky (NIR)

1 2 3

1450 nm, 1850 nm 1400–1500 nm 1450 nm

First overtone of the O–H stretch Shape of the 1450 nm peak Height of the 1450 nm peak

[40]

Scotland; whisky (FL)

1

520 nm

Peak height at 520 nm

[40]

[40]

Phen. extr., phenolic extract; FL, fluorescence. Comb., combination; def., deformation.

[33] Spanish DO from others. The best results (100% classification) were obtained for Argentinean wines [30,31] adjusted to pH 10.2. In contrast to the discrimination power of the UV region for white wines, the range of relevant variables (240–800 nm) for red wines also includes the visible wavelengths reflecting the presence of anthocyanins and/or other phenolic compounds [32,33]. For example, the region between 500 and 560 nm enables the highest discrimination of red wines from different zones within the LaMancha DO [32]. Fluorescence spectra mainly allow the classification of wines according to variety, typicality and manufactures. The ability of fluorescence spectra to differentiate between red wines produced in France and Germany has been also investigated. The emission spectra were characterized by a maximum at 376 nm and a shoulder at 315 nm and the excitation spectra showed two peaks located at about 260 and 320 nm. A classification of the French and German wine samples was observed using the excitation spectra [18]. Regarding NIR region, wavelength range of 1300–1650 nm, associated with a combination of stretch and deformation of the O–H group in water, gives the best result for Chinese rice wines of the two geographical origins, with 100% classification and root mean square error of prediction (RMSEP) = 0.259. The results for

800–2500 nm and 1100–2500 nm regions are a little worse (100% classification, RMSEP = 0.239) [19]. The wavelengths of the highest eigenvectors for the first two PCs of PCA performed on the full spectral region, 800–2500 nm, are shown in Table 3. To discriminate Chinese rice wine from the three different origins, the range of 1880–2100 and 2330–2500 nm was excluded from PCA [20]. Thus, different wavelengths of the highest eigenvectors were observed (Table 3) [19,20]. The relevant wavelengths for grape wines fall within the range of 2200–2300 nm, relating to alcohols, sugars, phenolics and organic acids, and providing the highest eigenvectors in the first PCs of PCA [22]. Two other regions, 1440–1460 and 1660–1760 nm, are also partially significant (Table 3) [24]. In contrast to 100% classification of rice wines [19,20], 60–85% classification is observed for grape wine [21,22,25]. NIR spectral data are often complemented with measurements in the UV/Vis range [21,23,24]. Success in a classification depends on both spectral range and pattern recognition methods (Table 1). The UV, Vis and NIR spectra, and their combinations, can classify zones from DO Rías Baixas [21]. The best total classification (86%) was obtained in the UV/Vis/NIR range using PCA–LDA. The SIMCA classification was 100% for Condado wines using the UV/ Vis/NIR, NIR or Vis/NIR ranges, for Salnés and Ribeira de Ulla wines

136

V. Urícˇková, J. Sádecká / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 148 (2015) 131–137

using the UV/Vis range, and for Rosal wines in the UV range. The PCA–LDA classification was 100% for Rosal wines using the Vis/NIR range. The SVM classification was 100% for Condado using the UV/Vis/NIR and UV/Vis ranges, and for Salnés using the Vis range [21]. Vis/NIR spectra allow discrimination between wines from different countries (Australia and Spain, or Australia, New Zealand, France and Germany); in both cases PLS-DA provides better results than PCA–LDA [23,24]. Regarding MIR region, the relevant wavenumbers for wines fall within the range of 900–1500 cm 1. This region can be used to discriminate between wines from Australian regions [25], or from Australia and New Zealand [22]. Table 1 shows the results of PCA–LDA obtained for the classification of Australian wines using UV, Vis, NIR, and MIR spectra [25]. The best PCA–LDA results were obtained using the MIR spectra (73%). Overall, the best classification results were obtained using SIMCA and the MIR spectra (95%). Considering Australian and New Zealand wines, the MIR region provides better classification than NIR or Vis; however, the concatenation of NIR and MIR improves prediction for the Australian wines [22]. Small spectral differences in the two wavenumber regions, 1150–1300 and 2300–2400 cm 1, and SIMCA allow the classification of Spanish wines from two zones within LaMancha DO (73%) [28]. Geographical origin of distilled drinks Brandy and Cognac The UV absorbance band (280 nm) of brandies/Cognacs results from the total contribution of phenolic oak wood compounds (coniferyl, sinapic, syringic aldehydes, gallic, syringic and vanilic acids and vanilin) and of caramel color components (furans), whereas compounds that give a color to drinks are detected at 420 nm [8]. The fluorescence spectra of brandies are characterized by the main bands centered at an excitation/emission wavelength of 400/495 nm. These bands arise from the caramel, and from coumarins, tannins and other fluorescent compounds originating from wooden casks [34]. Studies of brandy/Cognac using IR are scarce. Ethanol, water and sugars have a high response in the IR spectra. In the NIR region, absorptions are observed at 1450, 1690, 1790, 1950 nm and between 2200 and 2400 nm associated with ethanol, water, sugars, and phenolic compounds [35]. Preliminary results show that Cognacs can be clearly (100%) differentiated from the other samples (Armagnacs and brandies from different countries) by the 3700–4900 cm 1 region [36]. The same occurs with the Armagnacs, although there are three Spanish brandies presenting great similarities with Armagnacs. The remaining samples present groupings related to their origins, although with higher variability, in comparison with this observed for the Cognacs and Armagnacs [36]. Ethanol dominates the MIR spectra of raw products (peaks at 1085, 1045 and 878 cm 1) and masks the absorption bands of other compounds. Thus, the MIR range is not useful to discriminate raw products with different geographical origin. The MIR spectra of dry extracts and polyphenolic dry extracts provide additional information and allow good discrimination between Cognac and non-Cognac drinks [37]. The wavelengths of the highest eigenvectors for the first PCs of PCA performed on the spectral region 850–1800 cm 1 are shown in Table 3.

UV/Vis spectra of specific brands of Scotch whisky were found to produce consistent absorbance ranges which allowed the development of the portable instrument that can be used to confirm the authenticity of Scotch whisky samples. An algorithm for drawing conclusions was based on the sum of squared scores for absorbance results at 10 pre-selected wavelengths (220, 230, 240, 250, 260, 280, 300, 320, 340 and 360 nm) [38]. The instrument can also be used with rums, brandies, Irish whisky and Bourbon [39]. The fluorescence spectra of whisky shows the main bands centered at an excitation/emission wavelength pair of 404/520 nm associated with caramel, coumarins, tannins and other fluorescent compounds [40]. The combination of absorption (UV/Vis, NIR) and fluorescence spectroscopic data demonstrated the possibility of grouping single-malt whiskies according to their geographic area of production [40]. First, PCA was applied to each data-block; Table 3 shows the results. Next a 18  6 (samples  PCA scores) joint-data matrix containing PC1 and PC2 scores from UV/Vis data, PC1, PC2 and PC3 scores from NIR data and PC1 scores from fluorescence data was created. Then, LDA was applied to this matrix and 100% classification of single-malt types according to two macro-areas of production, highlands and islands, was achieved. Other distilled drinks Twenty-six homemade samples of fruit distillates (plum, apple and pear brandies) originating from different areas of RomaniaTransylvania (Maramure, Cluj, Bistrita-Nasaud, Alba and Bihor Counties) have been analyzed using UV/Vis spectrometry. The significant differences between plum, apple and pear brandies and between the plum brandies from different counties were observed [41]. Our preliminary study has demonstrated the feasibility of synchronous fluorescence spectroscopy for differentiating 16 Slovak from 18 Belgian juniper-flavored spirit drinks using synchronous fluorescence spectra (250–350 nm) collected at wavelength interval 20 nm. After performing backward PCA–LDA, a classification function was obtained for individual analyzed drinks containing five variables (excitation wavelengths): 259, 278, 300, 324 and 335 nm, which provided 99% correct predictions for Belgian and Slovak drinks [42]. Conclusions Although there are numerous reports on the use of UV, Vis and IR spectroscopy in food analysis, until now not many articles were found in the literature on the use of the methods to determine geographical origin of beverage samples. UV, Vis, and NIR spectra contain broad band(s) with weak spectral features limiting their discrimination ability. Despite this expected shortcoming, each of the three spectroscopic ranges (NIR, Vis/NIR and UV/Vis/NIR) provides average correct classification higher than 82%. Although average correct classification is similar for NIR and MIR regions, in some instances MIR data processing improves prediction. Advantage of using MIR is that MIR peaks are better defined and more easily assigned than NIR bands. In general, success in a classification depends on both spectral range and pattern recognition methods. Thus, each issue requires a specific investigation to develop accurate and robust discriminant model. Spectroscopic methods provide a relative rapid and low cost alternative to traditional chemical composition or sensory analyses. However, still remains the main problem the construction of representative databanks needed for all of these methods.

Scotch whisky Compliance with ethics requirements A typical UV/Vis spectrum of whisky shows the highest absorbance at 200 nm and shoulders at 280, 390, 420 and 460 nm resulted from polyphenolic and caramel color components. The

This article does not contain any studies with human or animal subjects.

V. Urícˇková, J. Sádecká / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 148 (2015) 131–137

Acknowledgements This research was supported by the Scientific Grant Agency of the Ministry of Education of Slovak Republic and the Slovak Academy of Sciences VEGA No. 1/0051/13. This publication was supported by the Competence Center for SMART Technologies for Electronics and Informatics Systems and Services, ITMS 26240220072, funded by the Research & Development Operational Programme from the ERDF. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18]

D.M.A.M. Luykx, S.M. van Ruth, Food Chem. 107 (2008) 897–911. M. Kozak, C.H. Scaman, J. Sci. Food Agric. 88 (2008) 1115–1127. A. Tsakiris, S. Kallithraka, Y. Kourkoutas, J. Sci. Food Agric. 94 (2014) 404–414. E. Capuano, R. Boerrigter-Eenling, G. van der Veer, S.M. van Ruth, J. Sci. Food Agric. 93 (2013) 12–28. J. Saurina, Trends Anal. Chem. 29 (2010) 234–245. D. Cozzolino, W. Cynkar, N. Shah, P. Smith, Anal. Bioanal. Chem. 401 (2011) 1475–1484. S.A. Drivelos, C.A. Georgiou, Trends Anal. Chem. 40 (2012) 38–51. S.A. Savchuk, V.N. Vlasov, S.A. Appolonova, V.N. Arbuzov, A.N. Vedenin, A.B. Mezinov, B.R. Grigor’yan, J. Anal. Chem. 56 (2001) 214–231. M.J. Adams, Chemometrics in Analytical Spectroscopy, The Royal Society of Chemistry, Letchworth, 1995. J. Yang, J.Y. Yang, Pattern Recognit. 36 (2003) 563–566. V. Vapnik, The Nature of Statistical Learning Theory, second ed., SpringerVerlag, New York, 1995. 2000. R. De Maesschalck, A. Candolfi, D.L. Massart, S. Heuerding, Chemom. Intell. Lab. Syst. 47 (1999) 65–77. J.A. Westerhuis, H.C.J. Hoefsloot, S. Smit, D.J. Vis, A.K. Smilde, E.J.J. van Velzen, J.P.M. van Duijnhoven, F.A. van Dorsten, Metabolomics 4 (2008) 81–89. A. Versari, G.P. Parpinello, L. Laghi, Spectroscopy 27 (2012) 36–47. M. Kumšta, P. Pavloušek, J. Kupsa, Food Sci. Biotechnol. 21 (2012) 1593–1601. A. Pena-Neira, T. Hernández, C. García-Vallejo, I. Estrella, J.A. Suarez, Eur. Food Res. Technol. 210 (2000) 445–448. D. Airado-Rodríguez, I. Durán-Merás, T. Galeano-Díaz, J.P. Wold, J. Food Comp. Anal. 24 (2011) 257–264. E. Dufour, A. Letort, A. Laguet, A. Lebecque, J.N. Serra, Anal. Chim. Acta 563 (2006) 292–299.

137

[19] H. Yu, Y. Zhou, X. Fu, L. Xie, Y. Ying, Eur. Food Res. Technol. 225 (2007) 313– 320. [20] F. Shen, D.T. Yang, Y.B. Ying, B.B. Li, Y.F. Zheng, T. Jiang, Food Bioprocess. Technol. 5 (2012) 786–795. [21] M.J. Martelo-Vidal, F. Dominguez-Agis, M. Vazquez, Aust. J. Grape Wine Res. 19 (2013) 62–67. [22] D. Cozzolino, N. Shah, W. Cynkar, P. Smith, Food Chem. 126 (2011) 673–678. [23] L. Liu, D. Cozzolino, W.U. Cynkar, M. Gishen, C.B. Colby, J. Agric. Food Chem. 54 (2006) 6754–6759. [24] L. Liu, D. Cozzolino, W.U. Cynkar, R.G. Dambergs, L. Janik, B.K. O’Neill, C.B. Colby, M. Gishen, Food Chem. 106 (2008) 781–786. [25] R. Riovanto, W.U. Cynkar, P. Berzaghi, D. Cozzolino, J. Agric. Food Chem. 59 (2011) 10356–10360. [26] D. Picque, T. Cattenoz, C. Trelea, C. Cuinier, G. Corrieu, Bull. OIV 75 (2002) 861– 862. [27] D. Picque, T. Cattenoz, G. Corrieu, J.L. Berger, Sci. Aliment 25 (2005) 207–220. [28] M. Urbano, M.D. Luque de Castro, P.M. Pérez, M.A. Gómez-Nieto, Anal. Bioanal. Chem. 381 (2005) 953–963. [29] S. Agatonovic-Kustrin, D.W. Morton, A.P.M. Yusof, Mod. Chem. Appl. 1 (2013) 110, http://dx.doi.org/10.4172/2329-6798.1000110. [30] S.M. Azcarate, M.A. Cantarelli, R.G. Pellerano, E.J. Marchevsky, J.M. Camiña, J. Food Sci. 78 (2013) 432–436. [31] S.M. Azcarate, M.A. Cantarelli, E.J. Marchevsky, J.M. Camiña, J. Food Res. 2 (2013) 48–56. [32] M. Urbano, M.D. Luque de Castro, P.M. Perez, J. Garcia-Olmo, M.A. GomezNieto, Food Chem. 97 (2006) 166–175. [33] F.J. Acevedo, J. Jiménez, S. Maldonado, E. Domínguez, A. Narváez, J. Agric. Food Chem. 55 (2007) 6842–6849. [34] J. Sádecká, J. Tóthová, P. Májek, Food Chem. 117 (2009) 491–498. [35] D. Markechová, P. Májek, A. Kleinová, J. Sádecká, Anal. Methods 6 (2014) 379– 386. [36] M. Palma, C.G. Barroso, Talanta 58 (2002) 265–271. [37] D. Picque, P. Lieben, G. Corrieu, R. Cantagrel, O. Lablanquie, G. Snakkers, J. Agric. Food Chem. 54 (2006) 5220–5226. [38] W.M. MacKenzie, R.I. Aylott, Analyst 129 (2004) 607–612. [39] . [40] A.G. Mignani, L. Ciaccheri, B. Gordillo, A.A. Mencaglia, M.L. González-Miret, F.J. Heredia, B. Culshaw, Sens. Actuators B 171–72 (2012) 458–462. [41] T.E. Rusu (Coldea), C. Socaciu, F. Fetea, F. Ranga, R. Parlog, Bull. UASVM Agric. 68 (2011) 518–528. [42] J. Sádecká, P. Májek, Lˇ. Píš, Luminescence 25 (2010) 224–225.