- Email: [email protected]
Quantification of Coffea arabica and Coffea canephora var. robusta concentration in blends by means of synchronous fluorescence and UV-Vis spectroscopies
Talanta 172 (2017) 215–220
Contents lists available at ScienceDirect
Talanta journal homepage: www.elsevier.com/locate/talanta
Quantification of Coffea arabica and Coffea canephora var. robusta concentration in blends by means of synchronous fluorescence and UV-Vis spectroscopies
MARK
⁎
A. Dankowskaa, , A. Domagałab, W. Kowalewskic a b c
Department of Food Commodity Science, Poznań University of Economics and Business, al. Niepodległości 10, 61-875 Poznań, Poland Department of Econometrics, Poznań University of Economics and Business, al. Niepodległości 10, 61-875 Poznań, Poland Department of Geoinformation, Adam Mickiewicz University, B. Krygowskiego 10, 61-680 Poznań, Poland
A R T I C L E I N F O
A BS T RAC T
Keywords: Food fraud Coffee authenticity Fluorescence spectroscopy UV-Vis spectroscopy Data fusion Multivariate data analysis
The potential of fluorescence, UV-Vis spectroscopies as well as the low- and mid-level data fusion of both spectroscopies for the quantification of concentrations of roasted Coffea arabica and Coffea canephora var. robusta in coffee blends was investigated. Principal component analysis was used to reduce data multidimensionality. To calculate the level of undeclared addition, multiple linear regression (PCA-MLR) models were used with lowest root mean square error of calibration (RMSEC) of 3.6% and root mean square error of cross-validation (RMSECV) of 7.9%. LDA analysis was applied to fluorescence intensities and UV spectra of Coffea arabica, canephora samples, and their mixtures in order to examine classification ability. The best performance of PCA-LDA analysis was observed for data fusion of UV and fluorescence intensity measurements at wavelength interval of 60 nm. LDA showed that data fusion can achieve over 96% of correct classifications (sensitivity) in the test set and 100% of correct classifications in the training set, with low-level data fusion. The corresponding results for individual spectroscopies ranged from 90% (UV-Vis spectroscopy) to 77% (synchronous fluorescence) in the test set, and from 93% to 97% in the training set. The results demonstrate that fluorescence, UV, and visible spectroscopies complement each other, giving a complementary effect for the quantification of roasted Coffea arabica and Coffea canephora var. robusta concentration in blends.
1. Introduction Coffee is one of the most popular drinks and is consumed for its refreshing, stimulating taste and health benefits. Most of the commercially available coffees are beans of Coffea arabica and Coffea canephora var. robusta or their blends. On account of the high price of Coffea arabica as compared to Coffea canephora var. robusta, undeclared additions of cheaper species or varieties of coffee is practiced for purposes of economy, which makes coffee fraud a real issue. Maintaining high production standards can enhance fair competition in the industry as well as satisfaction and health of consumers. Therefore, there is need for analytical techniques to control authenticity of coffee samples. Various instrumental methods have been proposed to establish coffee authenticity and detect the level of adulteration. The techniques used most widely are spectroscopic and
chromatographic methods: UV-Vis [1,2], NIR [3–5], GC [6,7] and HPLC [8]. Numerous investigations have proven the ability of fluorescence and UV-Vis spectroscopies to detect food adulteration [1,2,9– 16]. Synchronous fluorescence and UV-Vis spectroscopies are quick to use and dispense with sample preparation steps very often limited to dilution or filtration; therefore, these techniques are simpler, less expensive, and quicker than other widely used techniques. Although few compounds are fluorophores, even small amounts of the compounds can exhibit fluorescence, which makes fluorescence spectroscopy a high sensitive and selective tool. Analyzing synchronous fluorescence spectra, obtained by simultaneous scanning of monochromators with constant wavelength intervals between excitation and emission light, can increase selectivity even further; this procedure also leads to a simplification of spectra and a reduction in spectral overlap [17]. Moreover, fluorescence spectroscopy and UV-Vis spectroscopies
Abbreviations: SF, Synchronous fluorescence; UV-Vis, Ultraviolet and visible spectroscopy; PCA, Principal Component Analysis; MLR, Multiple Linear Regression; LDA, Linear Discriminant Analysis; R, Coefficient of correlation; Adjusted R2, adjusted multivariate coefficient of determination; RSMEC, root mean square error of calibration; RSMEV, root mean square error of validation; CE_CS, Classification error in the calibration set of all 115 samples; MCE_LOO, Crossvalidation Leave-one-out Classification Error; MCE_8020, Crossvalidation 80 – 20% Classification Error ⁎ Corresponding author. E-mail address: [email protected] (A. Dankowska). http://dx.doi.org/10.1016/j.talanta.2017.05.036 Received 12 January 2017; Received in revised form 9 May 2017; Accepted 12 May 2017 Available online 17 May 2017 0039-9140/ © 2017 Elsevier B.V. All rights reserved.
Talanta 172 (2017) 215–220
A. Dankowska et al.
analysis was chosen on the basis of the Kaiser criterion [18] (since PCs with eigenvalues higher than one provide more information than average single variable); it is one of the most popular criteria used to select significant PCs. Samples were classified into one of three classes: Coffea arabica (A), Coffea robusta (R) or mixture of two (M). The PCAMLR (PCR) analysis was performed using Statistica 12.0 (StatSoft Inc., place City Tulsa, country-region US) while the PCA-LDA analysis was carried out using R, version 3.2.5 (2016-04-14), software environment for statistical computing.
are complementary, even though the latter one is less sensitive and its spectra are more overlapped compared to fluorescence spectroscopy. So far data fusion of fluorescence spectroscopy and UV-Vis spectroscopies for food authentication has not received sufficient attention though. The aim of this research was to investigate the application of synchronous fluorescence spectroscopy and UV-Vis spectroscopies for the quantification of Coffea canephora var. robusta addition to Coffea arabica, as well as the complementarity of data obtained with the use of both spectroscopy techniques.
2.4. Validation methods
2. Materials and methods
PCA-MLR prediction models were validated using leave-one-out cross-validation. Root mean square error of calibration (RSMEC) and root mean square error of cross-validation (RSMEV) were calculated. PCA–LDA classification models were validated using cross-validation of two types: leave-one-out cross-validation (with mean classification error of the leave-one-out cross-validation calculated, MCE_LOO) and v-fold cross-validation with 1000 repeats. In the latter, 115 items out of all samples (genuine coffee samples and mixtures with 40%, 50% and 60% level of adulteration) were 1000 times randomly split into two subsets: training or calibration set (80% of all 115 samples) and test set (20%), assuming that the content of both sets does not repeat. Each time LDA was performed using the training set to estimate the parameters of discriminant functions, and classification error was then calculated on the basis of the test set. The process was repeated 1000 times and then mean classification error of cross-validation 80-20 (MCE_8020) was calculated. There was no call to make all possible analyses with 80–20% split of all 115 samples (the number of all analyses would amount to 9.097·1023) as simulations conducted show that MCE_8020 stabilizes with the number of repeats approaching 1000. Classification error of LDA with all 115 samples used as calibration set (i.e. without a division into a training and a test set) were also calculated (CE_CS).
2.1. Chemical reagents and samples Roasted (various degrees of roast) coffee samples were purchased at a local shop in Poznan, Poland. There were in total 33 coffee samples (25 Coffea arabica and 8 Coffee canephora var. robusta samples) from different countries: India (6), Brazil (5), Guatemala (4), Columbia (3), Ethiopia (3), Indonesia(2), Vietnam (2), Nicaragua (1), Kenya (1), Honduras (1), Costa Rica (1), Tanzania (1), Peru (1), Mexico (1), El Salvador (1). All the coffees were ground before the preparation of experimental mixtures. The models of adulterated coffee were constructed by mixing ground Coffea arabica and Coffea canephora var. robusta at levels ranging from 0% to 100% with 10% intervals on the dry basis (w/w). Four series of experimental mixtures were prepared. Proportions of coffee species in the prepared mixtures reflect possible market practices and the commercial samples. Six-per-cent (w/v) water (95 °C) extracts of coffee were prepared, cooled, filtered and diluted 1:120 (v:v) with distilled water. In total, 147 synchronous fluorescence and UV-Vis spectra were measured. All analyses were carried out in triplicate for each genuine coffee sample or in duplicate for the mixtures of coffee samples. All the reagents used in the experiment were of analytical grade. 2.2. Synchronous fluorescence spectra measurement
3. Results and discussion
Fluorescence spectra were obtained by using Spectrofluorometer Thermo Scientific Lumina with Xenon lamp as a source of excitation. Excitation and emission slit widths were of 10 and 5 nm, respectively. Acquisition interval and integration time were maintained at 1 nm and 0.1 s, respectively. Right-angle geometry was used for coffee samples diluted in redistilled water (1% v/v) in a 10 mm fused quartz cuvette. Synchronous fluorescence spectra were acquired by simultaneously scanning the excitation and emission monochromator at excitation wavelengths ranging from 240 to 700 nm with constant wavelength distances (Δλ) of 60 and 80 nm. Fluorescence intensities were plotted as a function of the excitation wavelength. Synchronous fluorescence spectra were collected in triplicate for each genuine coffee sample or in duplicate for the mixtures of coffee samples.
3.1. Synchronous fluorescence spectra of coffee samples Synchronous fluorescence intensities acquired for Coffea arabica and Coffea robusta samples and their mixtures as a function of excitation wavelength are shown in Fig. 1. Arabica and Robusta coffees exhibit differences in fluorescence spectra caused by the different contents of tocochromanols, polyphenols, fatty acids, and chlorophylls [7,19]. A simplification and an amplification of synchronous fluorescence spectra of Arabica and Robusta coffees using different wavelength intervals (60 and 80 nm) are shown in Fig. 1. 3.2. UV-Vis spectra of coffee samples The spectra presented in Fig. 2 indicate the potential of UV-Vis spectrophotometry for discrimination between Coffea arabica, Coffea robusta and their mixture samples. Fig. 2 presents the UV-Vis absorption spectra of aqueous extracts of the genuine Arabica and Robusta coffee samples and their mixtures in the range 190–700 nm. The intensity of UV-Vis spectra of different coffee species depends, among other things, on the contents of caffeine, chlorogenic acids, and trigonelline molecules. Caffeine shows maximum absorption around 275 nm, so the spectrum in this region is strongly related to chromophore absorption of caffeine, but impact of other compounds, e.g. other methylxanthines, on this spectrum cannot be excluded [20]. It is worth noting that chlorogenic acids and trigonelline are decomposed at roasting [21]; thus the presence of bands around 290 and 320 nm suggests that these compounds were not entirely decomposed in the medium roasting process [1].
2.3. Statistical analysis Measurements of spectra are usually followed by a chemometric analysis of data. In the low-level data fusion, individual spectra obtained from the SF and UV-Vis spectroscopies were grouped into a new matrix composed of 147 measurements and 592 variables. As LDA cannot be performed for such large numbers of variables, PCA was employed instead. PCA was then followed by MLR and LDA. In the mid-level data fusion, PCA was first carried out on the SF and UV-Vis spectra separately, and then MLR and LDA were applied to the combination of the PC scores. Principal components (PCs) are characterized by a decreasing variance (which is a measure of their linear information capacity), so that the first principal component explains the highest percentage of total observable variable variance. The number of principal components included in the MLR and LDA 216
Talanta 172 (2017) 215–220
A. Dankowska et al.
Fig. 1. Synchronous fluorescence spectra of Coffea arabica, Coffea robusta, and their mixtures (diluted 1:120 v/v in water) (A - Coffea arabica, R - Coffea robusta, M - mixtures of Coffea arabica and Coffea robusta).
3.4. Synchronous fluorescence and UV-Vis intensities versus addition of adulterant (PCA-MLR) Multiple linear regression analysis was applied to principal components (PCs) obtained for synchronous and UV-Vis spectroscopy measurements, as well as for low-level data fusion of measurements obtained with the use of both spectroscopies (Table 2). The number of principal components (PCs) incorporated into multiple linear regression analysis was established according to the Kaiser criterion [18] (Table 1). PCA-MLR (PCR) models for synchronous fluorescence as well as for data fusion of synchronous fluorescence and UV-Vis intensities were built separately for data acquired at two wavelength intervals (60 and 80 nm). For all PCA-MLR models R coefficients of at least 0.90 were obtained. Higher adjusted R2 coefficients were obtained for UV-Vis spectroscopy than for synchronous florescence spectroscopy. This suggests that UV-Vis spectroscopy, as analyzed individually, is more useful for quantification of Coffea arabica and Coffea robusta concentration in blends than synchronous fluorescence spectroscopy. Higher adjusted R2 coefficients of at least 0.95 were obtained for data fusion analysis than for individual spectroscopies. The lowest RMSEC and RMSEV for PCA-MLR models built with data obtained by applying individual spectroscopy methods equaled 5.3% and 21.9% for synchronous fluorescence spectroscopy, and 8.2% and 8.9% for UV-Vis spectroscopy. Application of UV-Vis spectroscopy produced a significantly lower error of prediction as to the level of addition of Coffea robusta to Coffea arabica than for synchronous fluorescence spectroscopy. The RMSEC and RMSEV values calculated for data fusion models (SF+UV+Vis) were lower than for individual spectroscopy methods. The lowest RMSEC and RMSEV values (3.6% and 7.9%) were obtained for the mid-level data fusion PCA-MLR model calculated for fluorescence intensities measured at Δλ=60 nm.
Fig. 2. UV-Vis spectra of Coffea arabica, Coffea robusta, and their mixtures (diluted 1:120 v/v in water) (A - Coffea arabica, R - Coffea robusta, M - mixtures of Coffea arabica and Coffea robusta).
3.3. Data dimensionality reduction with principal component analysis Principal component analysis (PCA) was employed for exploratory spectral analysis and subsequently MLR and LDA were performed. In the low-level data fusion, all data of synchronous fluorescence and UVVis spectroscopies were grouped into new matrices (separately for fluorescence data obtained at Δλ=60 and 80 nm), and then PCA was performed (Fig. 3). The number of principal components (PCs) taken for further analysis and percentage of explained variance are presented in Table 1. All selected PCs cumulatively account for at least 96% of the total variance for each spectroscopy. It is worth noting that in the case of UV-Vis spectral data the number of principal components which explained 99.16% of the variance equaled 6, while in the case of synchronous fluorescence spectra, both for low- and mid-level data fusion matrices, the number of principal components taken for further analysis exceeded 30 which indicates dispersion of information.
3.5. Linear discriminant analysis of synchronous fluorescence and UV-Vis spectra (PCA-LDA) Linear discriminant analysis was applied to principal components obtained previously by PCA. In the case of LDA, fluorescence and UVVis intensities of Coffea arabica and Coffea robusta, and their mixtures containing 40%, 50%, and 60% of adulterant were analyzed. LDA was carried out separately for synchronous fluorescence data acquired at each wavelength interval (60, 80 nm), UV-Vis intensities, as well as low- and mid-level data fusion models. Plots of the first two discriminant functions (DF1*DF2) show classification performance of the PCA217
Talanta 172 (2017) 215–220
A. Dankowska et al.
Table 1 Number of principal components chosen for further data analysis according to Kaiser criterion. Procedure/ Statistical parameter
Wavelength interval [nm]
Number of principal components (Kaiser criterion)
Cumulative proportion of variance explained
SF
60 80 –
32 32 6
96,10% 96,09% 99,16%
Low-level Data Fusion (SF +UV-Vis)
60 80
37 37
97,98% 97,95%
Mid-level Data Fusion (SF +UV-Vis)
60
38 (32 + 6)
80
38 (32 + 6)
96,10% 99,16% 96,09% 99,16%
UV-Vis
(SF), (UV-Vis) (SF), (UV-Vis)
SF – Synchronous fluorescence spectroscopy, UV-Vis – Ultraviolet and visible spectroscopy. Table 2 Statistical characteristics of multiple linear regression models calculated for individual procedures (SF and UV-Vis) and low-level and mid-level data fusion (SF+UV-Vis). Procedure/ statistical parameter
Wavelength interval [nm]
R
Adjusted R2
RMSEC
RMSECV
SF
60 80 –
0.90 0.93 0.97
0.71 0.80 0.93
5.3 5.3 8.2
21.9 22.0 8.9
Low-level Data Fusion (SF +UV-Vis)
60 80
0.99 0.99
0.98 0.95
5.4 5.3
10.1 11.3
Mid-level Data Fusion (SF +UV-Vis)
60 80
0.99 0.99
0.98 0.95
3.6 3.8
7.9 8.1
UV-Vis
SF – Synchronous fluorescence spectroscopy. UV-Vis – Ultraviolet and visible spectroscopy. R – Coefficient of correlation. Adjusted R2 – adjusted multivariate coefficient of determination. RSMEC – Root mean square error of calibration. RSMEV – Root mean square error of cross-validation.
Perhaps Robusta samples contain some characteristic factor, e.g., 16O-methylcafestol, typical for Coffea robusta and absent in Coffea arabica. The presence of this factor in the sample would explain why the mixture is closer to Coffea Robusta in the plots of the first two discriminant functions (DF1*DF2) [22]. The results presented in Fig. 4 were confirmed by a comparison of LDA results for individual procedures and for data fusion as shown in Table 3, where error classification rates in Coffea arabica, Coffea canephora var. Robusta, and their mixtures for PCA-LDA models for individual spectroscopies and data fusion are represented. The best discrimination ability of PCA-LDA among the models obtained for individual spectroscopies was observed for the UV-Vis spectroscopy with classification errors CE_CS, MCE_LOO, MCE_8020 of 6.09%, 10.43%, and 9.13%, respectively. Compared with individual observations, LDA in data fusion offers better results. It is worth noting that lower classification errors were obtained for models with fluorescence intensities obtained at wavelength intervals of 60 nm than 80 nm, and for low-level data fusion than for mid-level data fusion. Comparing LDA results between individual spectroscopies as well as data fusion of both spectroscopies, it was also found that the best classification results among all models were obtained for the low-level data fusion model with synchronous fluorescence measurements
Fig. 3. First two PCs loading plots of PCA obtained for (a) synchronous fluorescence intensities acquired at Δλ=60 nm, (b) UV-Vis, (c) low-level data fusion of synchronous fluorescence (Δλ=60 nm) and UV-Vis spectra (A - Coffea arabica, R - Coffea robusta, M mixtures of Coffea arabica and Coffea robusta).
LDA of measured samples in three clusters: Coffea arabica, Coffea robusta, and their mixtures, acquired for synchronous fluorescence at Δλ=60 nm, UV-Vis spectroscopy, and low- and mid-level data fusion of both methods (Fig. 4). Fluorescence intensities at wavelength intervals of 60 and 80 nm did not allow for good classification performance because the clusters were scattered and overlapped. It is interesting to note that for the models acquired for fluorescence, UV-Vis spectroscopy and the low- and mid-level data fusion of both methods, the group formed by mixtures of Coffea arabica and Coffea canephora var. robusta is closer to the group of Coffea canephora var. Robusta alone.
218
Talanta 172 (2017) 215–220
A. Dankowska et al.
Fig. 4. PCA-LDA plots of (a) synchronous fluorescence intensities acquired at Δλ=60 nm, (b) UV-Vis, (c) low-level data fusion of synchronous fluorescence (Δλ=60 nm) and UV-Vis spectra, and (d) mid-level data fusion of synchronous fluorescence (Δλ=60 nm) and UV-Vis spectra Coffea arabica (A), Coffea robusta (R), and their mixtures (M).
acquired at wavelength intervals of 60 nm and classification errors of 0.0% (CE_CS), 2.61% (MCE_LOO), and 3.21% (MCE_8020).
Table 3 Error classification for PCA-LDA in Coffea arabica and Coffea canephora var robusta and their mixture [%] calculated for individual procedures (SF, UV), low-level and midlevel data fusion (SF+UV-Vis). Procedure/ Statistical parameter
Wavelength interval [nm]
CE_CS classification error [%]
MCE_LOO crossvalidation leave-one-out classification error [%]
MCE_8020 crossvalidation 80 – 20 classification error [%]
SF
60 80 –
6.09 2.61 6.09
20.00 20.87 10.43
22.37 22.87 9.13
Low-level Data Fusion (SF+UV)
60 80
0.00 0.00
2.61 6.09
3.21 6.53
Mid-level Data Fusion (SF+UV)
60 80
0.00 0.00
2.61 6.96
4.84 7.93
UV
4. Conclusion The study has shown that water extracts of Coffea arabica and Coffea canefora var. robusta exhibit significant differences in their synchronous fluorescence and UV-Vis spectra patterns. A comparison of MLR and LDA results between individual spectroscopies and lowand mid-level data fusion of the two spectroscopies leads us to conclude that UV-Vis spectroscopy offers complementary information to fluorescence spectroscopy. The best prediction ability of MLR models was obtained for the mid-level data fusion model of UV-Vis and fluorescence intensities at 60 nm wavelength interval with RMSEC and RMSEV of 3.6% and 7.9%, respectively. Compared to individual spectroscopies, data fusion showed better discrimination ability with the highest classification accuracy over 96.0% obtained for the lowlevel LDA model with fluorescence intensities at 60 nm wavelength interval. This recommends to combine synchronous fluorescence and UV-Vis spectroscopies, along with chemometric analysis, for quantification of roasted Coffea arabica and Coffea canephora var. robusta concentration in blends. The findings may contribute to a better control of coffee fraud and provide a technical boost of consumer interest against profit-driven practices on the food market.
CE_CS – Classification error in calibration set of all 115 samples, MCE_LOO – Crossvalidation leave-one-out classification error, MCE_8020 - Cross-validation 80 – 20% classification error.
219
Talanta 172 (2017) 215–220
A. Dankowska et al.
[10] A. Dankowska, M. Małecka, W. Kowalewski, Application of synchronous fluorescence spectroscopy with multivariate data analysis for determination of butter adulteration, Int. J. Food Sci. Technol. 49 (2014) 2628–2634. [11] A. Dankowska, M. Małecka, W. Kowalewski, Detection of plant oil addition to cheese by synchronous fluorescence spectroscopy, Dairy Sci. Technol. 95 (2015) 413–424. [12] D. Markechová, P. Májek, J. Sádecká, Fluorescence spectroscopy and multivariate methods for the determination of brandy adulteration with mixed wine spirit, Food Chem. 159 (2014) 193–199. [13] Q. Chen, S. Qi, H. Li, X. Han, Q. Ouyang, J. Zhao, Determination of rice syrup adulterant concentration in honey using three-dimensional fluorescence spectra and multivariate calibrations, Spectrochim. Acta A 131 (2014) 177–182. [14] C.V. Di Anibal, M.S. Rodríguez, L. Albertengo, Synchronous fluorescence and multivariate classification analysis as a screening tool for determining Sudan I dye in culinary spices, Food Control. 56 (2015) 18–23. [15] M. Ntakatsane, X.M. Liu, P. Zhou, Rapid detection of milk fat adulteration with vegetable oil by fluorescence spectroscopy, J. Dairy Sc. 96 (2013) 2130–2136. [16] I. Sergiel, P. Pohl, M. Biesaga, A. Mironczyk, Suitability of three-dimensional synchronous fluorescence spectroscopy for fingerprint analysis of honey samples with reference to their phenolic profiles, Food Chem. 145 (2014) 319–326. [17] D. Patra, A.K. Mishra, Recent developments in multi-component synchronous fluorescence scan analysis, Trends Anal. Chem. 21 (2002) 787–798. [18] ΗF. Kaiser, The application of electronic computers to factors analysis, Educ. Psychol. Meas. (1960) 141–151. [19] R.C. Alvesa, S. Casal, M.R. Alves, M.B. Oliveira, Discrimination between arabica and robusta coffee species on the basis of their tocopherol profiles, Food Chem. 114 (2009) 295–299. [20] L. López-Martínez, P.L. López-de-Alba, R. García-Campos, L.M.D. León-Rodríguez, Simultaneous determination of methylxanthines in coffees and teas by UV–Vis spectrophotometry and partial least squares, Anal. Chim. Acta 493 (2003) 83–94. [21] A. Illy, R. Viani, Expresso Coffee: The Chemistry of Quality, Academic, San Diego, 1995. [22] L. Servillo, A. Giovane, R. Casale, D. Cautela, N. Onofrio, M.L. Balestrieri, D. Castaldo, Homostachydrine (pipecolic acid betaine) as authentication marker of roasted blends of Coffea arabica and Coffea canephora (Robusta) beans, Food Chem. 205 (2016), 2016, pp. 52–57.
Conflict of interest statement The authors declare no conflict of interest. References [1] U.T.C.P. Souto, M.J.C. Pontes, E.C. Silva, R.K.H. Galvão, M.C.U. Araújo, F.A.C. Sanches, F.A.S. Cunha, M.S.R. Oliveira, UV-Vis spectrometric classification of coffees by SPA-LDA, Food Chem. 119 (2010) 368–371. [2] U.T.C.P. Souto, M.F. Barbosa, H.V. Dantas, A.S. Pontes, W. Lyra, P.H.G.D. Diniz, M.C.U. Araújo, E.C. Silva, Identification of adulteration in ground roasted coffees using UV-Vis spectroscopy and SPA-LDA, Food Sci. Technol. 63 (2015) 1037–1041. [3] C. Pizzaro, I. Esteban-Díez, J.M. González-Sáiz, Mixture resolution according to the percentage of robusta variety in order to detect adulteration in roasted coffee by near infrared spectroscopy, Anal. Chim. Acta 585 (2007) 266–276. [4] J.R. Santos, M. Lopo, A. Rangel, J.A. Lopes, Exploiting near infrared spectroscopy as an analytical tool for on-line monitoring of acidity during coffee roasting, Food Control. 60 (2016) 408–415. [5] R. Calvini, A. Ulrici, J.M. Amigo, Practical comparison of sparse methods for classification of Arabica and Robusta coffee species using near infrared hyperspectral imaging, Chem. Intell. Lab. Syst. 146 (2015) 503–511. [6] F. Carrera, M. León-Camacho, F. Pablos, A.G. González, Authentication of green coffee varieties according to their sterolic profile, Anal. Chim. Acta 370 (1998) 131–139. [7] M.J. Martin, F. Pablos, A.G. González, M.S. Valdenebro, M.L. Carnacho, Fatty acid profiles as discriminant parameters for coffee varieties differentiation, Talanta 54 (2001) 291–297. [8] E.D. Pauli, F. Barbieri, P.S. Garcia, T.B. Madeira, V.R. Acquaro Junior, I.S. Scarminio, C.A.P. Camara, S.L. Nixdorf, Detection of ground roasted coffee adulteration with roasted soybean and wheat, Food Res. Int. 61 (2014) 112–119. [9] A. Dankowska, M. Małecka, W. Kowalewski, Discrimination of edible olive oils by means of synchronous fluorescence spectroscopy with multivariate data analysis, GyA 64 (2013) 425–431.
220
Contents lists available at ScienceDirect
Talanta journal homepage: www.elsevier.com/locate/talanta
Quantification of Coffea arabica and Coffea canephora var. robusta concentration in blends by means of synchronous fluorescence and UV-Vis spectroscopies
MARK
⁎
A. Dankowskaa, , A. Domagałab, W. Kowalewskic a b c
Department of Food Commodity Science, Poznań University of Economics and Business, al. Niepodległości 10, 61-875 Poznań, Poland Department of Econometrics, Poznań University of Economics and Business, al. Niepodległości 10, 61-875 Poznań, Poland Department of Geoinformation, Adam Mickiewicz University, B. Krygowskiego 10, 61-680 Poznań, Poland
A R T I C L E I N F O
A BS T RAC T
Keywords: Food fraud Coffee authenticity Fluorescence spectroscopy UV-Vis spectroscopy Data fusion Multivariate data analysis
The potential of fluorescence, UV-Vis spectroscopies as well as the low- and mid-level data fusion of both spectroscopies for the quantification of concentrations of roasted Coffea arabica and Coffea canephora var. robusta in coffee blends was investigated. Principal component analysis was used to reduce data multidimensionality. To calculate the level of undeclared addition, multiple linear regression (PCA-MLR) models were used with lowest root mean square error of calibration (RMSEC) of 3.6% and root mean square error of cross-validation (RMSECV) of 7.9%. LDA analysis was applied to fluorescence intensities and UV spectra of Coffea arabica, canephora samples, and their mixtures in order to examine classification ability. The best performance of PCA-LDA analysis was observed for data fusion of UV and fluorescence intensity measurements at wavelength interval of 60 nm. LDA showed that data fusion can achieve over 96% of correct classifications (sensitivity) in the test set and 100% of correct classifications in the training set, with low-level data fusion. The corresponding results for individual spectroscopies ranged from 90% (UV-Vis spectroscopy) to 77% (synchronous fluorescence) in the test set, and from 93% to 97% in the training set. The results demonstrate that fluorescence, UV, and visible spectroscopies complement each other, giving a complementary effect for the quantification of roasted Coffea arabica and Coffea canephora var. robusta concentration in blends.
1. Introduction Coffee is one of the most popular drinks and is consumed for its refreshing, stimulating taste and health benefits. Most of the commercially available coffees are beans of Coffea arabica and Coffea canephora var. robusta or their blends. On account of the high price of Coffea arabica as compared to Coffea canephora var. robusta, undeclared additions of cheaper species or varieties of coffee is practiced for purposes of economy, which makes coffee fraud a real issue. Maintaining high production standards can enhance fair competition in the industry as well as satisfaction and health of consumers. Therefore, there is need for analytical techniques to control authenticity of coffee samples. Various instrumental methods have been proposed to establish coffee authenticity and detect the level of adulteration. The techniques used most widely are spectroscopic and
chromatographic methods: UV-Vis [1,2], NIR [3–5], GC [6,7] and HPLC [8]. Numerous investigations have proven the ability of fluorescence and UV-Vis spectroscopies to detect food adulteration [1,2,9– 16]. Synchronous fluorescence and UV-Vis spectroscopies are quick to use and dispense with sample preparation steps very often limited to dilution or filtration; therefore, these techniques are simpler, less expensive, and quicker than other widely used techniques. Although few compounds are fluorophores, even small amounts of the compounds can exhibit fluorescence, which makes fluorescence spectroscopy a high sensitive and selective tool. Analyzing synchronous fluorescence spectra, obtained by simultaneous scanning of monochromators with constant wavelength intervals between excitation and emission light, can increase selectivity even further; this procedure also leads to a simplification of spectra and a reduction in spectral overlap [17]. Moreover, fluorescence spectroscopy and UV-Vis spectroscopies
Abbreviations: SF, Synchronous fluorescence; UV-Vis, Ultraviolet and visible spectroscopy; PCA, Principal Component Analysis; MLR, Multiple Linear Regression; LDA, Linear Discriminant Analysis; R, Coefficient of correlation; Adjusted R2, adjusted multivariate coefficient of determination; RSMEC, root mean square error of calibration; RSMEV, root mean square error of validation; CE_CS, Classification error in the calibration set of all 115 samples; MCE_LOO, Crossvalidation Leave-one-out Classification Error; MCE_8020, Crossvalidation 80 – 20% Classification Error ⁎ Corresponding author. E-mail address: [email protected] (A. Dankowska). http://dx.doi.org/10.1016/j.talanta.2017.05.036 Received 12 January 2017; Received in revised form 9 May 2017; Accepted 12 May 2017 Available online 17 May 2017 0039-9140/ © 2017 Elsevier B.V. All rights reserved.
Talanta 172 (2017) 215–220
A. Dankowska et al.
analysis was chosen on the basis of the Kaiser criterion [18] (since PCs with eigenvalues higher than one provide more information than average single variable); it is one of the most popular criteria used to select significant PCs. Samples were classified into one of three classes: Coffea arabica (A), Coffea robusta (R) or mixture of two (M). The PCAMLR (PCR) analysis was performed using Statistica 12.0 (StatSoft Inc., place City Tulsa, country-region US) while the PCA-LDA analysis was carried out using R, version 3.2.5 (2016-04-14), software environment for statistical computing.
are complementary, even though the latter one is less sensitive and its spectra are more overlapped compared to fluorescence spectroscopy. So far data fusion of fluorescence spectroscopy and UV-Vis spectroscopies for food authentication has not received sufficient attention though. The aim of this research was to investigate the application of synchronous fluorescence spectroscopy and UV-Vis spectroscopies for the quantification of Coffea canephora var. robusta addition to Coffea arabica, as well as the complementarity of data obtained with the use of both spectroscopy techniques.
2.4. Validation methods
2. Materials and methods
PCA-MLR prediction models were validated using leave-one-out cross-validation. Root mean square error of calibration (RSMEC) and root mean square error of cross-validation (RSMEV) were calculated. PCA–LDA classification models were validated using cross-validation of two types: leave-one-out cross-validation (with mean classification error of the leave-one-out cross-validation calculated, MCE_LOO) and v-fold cross-validation with 1000 repeats. In the latter, 115 items out of all samples (genuine coffee samples and mixtures with 40%, 50% and 60% level of adulteration) were 1000 times randomly split into two subsets: training or calibration set (80% of all 115 samples) and test set (20%), assuming that the content of both sets does not repeat. Each time LDA was performed using the training set to estimate the parameters of discriminant functions, and classification error was then calculated on the basis of the test set. The process was repeated 1000 times and then mean classification error of cross-validation 80-20 (MCE_8020) was calculated. There was no call to make all possible analyses with 80–20% split of all 115 samples (the number of all analyses would amount to 9.097·1023) as simulations conducted show that MCE_8020 stabilizes with the number of repeats approaching 1000. Classification error of LDA with all 115 samples used as calibration set (i.e. without a division into a training and a test set) were also calculated (CE_CS).
2.1. Chemical reagents and samples Roasted (various degrees of roast) coffee samples were purchased at a local shop in Poznan, Poland. There were in total 33 coffee samples (25 Coffea arabica and 8 Coffee canephora var. robusta samples) from different countries: India (6), Brazil (5), Guatemala (4), Columbia (3), Ethiopia (3), Indonesia(2), Vietnam (2), Nicaragua (1), Kenya (1), Honduras (1), Costa Rica (1), Tanzania (1), Peru (1), Mexico (1), El Salvador (1). All the coffees were ground before the preparation of experimental mixtures. The models of adulterated coffee were constructed by mixing ground Coffea arabica and Coffea canephora var. robusta at levels ranging from 0% to 100% with 10% intervals on the dry basis (w/w). Four series of experimental mixtures were prepared. Proportions of coffee species in the prepared mixtures reflect possible market practices and the commercial samples. Six-per-cent (w/v) water (95 °C) extracts of coffee were prepared, cooled, filtered and diluted 1:120 (v:v) with distilled water. In total, 147 synchronous fluorescence and UV-Vis spectra were measured. All analyses were carried out in triplicate for each genuine coffee sample or in duplicate for the mixtures of coffee samples. All the reagents used in the experiment were of analytical grade. 2.2. Synchronous fluorescence spectra measurement
3. Results and discussion
Fluorescence spectra were obtained by using Spectrofluorometer Thermo Scientific Lumina with Xenon lamp as a source of excitation. Excitation and emission slit widths were of 10 and 5 nm, respectively. Acquisition interval and integration time were maintained at 1 nm and 0.1 s, respectively. Right-angle geometry was used for coffee samples diluted in redistilled water (1% v/v) in a 10 mm fused quartz cuvette. Synchronous fluorescence spectra were acquired by simultaneously scanning the excitation and emission monochromator at excitation wavelengths ranging from 240 to 700 nm with constant wavelength distances (Δλ) of 60 and 80 nm. Fluorescence intensities were plotted as a function of the excitation wavelength. Synchronous fluorescence spectra were collected in triplicate for each genuine coffee sample or in duplicate for the mixtures of coffee samples.
3.1. Synchronous fluorescence spectra of coffee samples Synchronous fluorescence intensities acquired for Coffea arabica and Coffea robusta samples and their mixtures as a function of excitation wavelength are shown in Fig. 1. Arabica and Robusta coffees exhibit differences in fluorescence spectra caused by the different contents of tocochromanols, polyphenols, fatty acids, and chlorophylls [7,19]. A simplification and an amplification of synchronous fluorescence spectra of Arabica and Robusta coffees using different wavelength intervals (60 and 80 nm) are shown in Fig. 1. 3.2. UV-Vis spectra of coffee samples The spectra presented in Fig. 2 indicate the potential of UV-Vis spectrophotometry for discrimination between Coffea arabica, Coffea robusta and their mixture samples. Fig. 2 presents the UV-Vis absorption spectra of aqueous extracts of the genuine Arabica and Robusta coffee samples and their mixtures in the range 190–700 nm. The intensity of UV-Vis spectra of different coffee species depends, among other things, on the contents of caffeine, chlorogenic acids, and trigonelline molecules. Caffeine shows maximum absorption around 275 nm, so the spectrum in this region is strongly related to chromophore absorption of caffeine, but impact of other compounds, e.g. other methylxanthines, on this spectrum cannot be excluded [20]. It is worth noting that chlorogenic acids and trigonelline are decomposed at roasting [21]; thus the presence of bands around 290 and 320 nm suggests that these compounds were not entirely decomposed in the medium roasting process [1].
2.3. Statistical analysis Measurements of spectra are usually followed by a chemometric analysis of data. In the low-level data fusion, individual spectra obtained from the SF and UV-Vis spectroscopies were grouped into a new matrix composed of 147 measurements and 592 variables. As LDA cannot be performed for such large numbers of variables, PCA was employed instead. PCA was then followed by MLR and LDA. In the mid-level data fusion, PCA was first carried out on the SF and UV-Vis spectra separately, and then MLR and LDA were applied to the combination of the PC scores. Principal components (PCs) are characterized by a decreasing variance (which is a measure of their linear information capacity), so that the first principal component explains the highest percentage of total observable variable variance. The number of principal components included in the MLR and LDA 216
Talanta 172 (2017) 215–220
A. Dankowska et al.
Fig. 1. Synchronous fluorescence spectra of Coffea arabica, Coffea robusta, and their mixtures (diluted 1:120 v/v in water) (A - Coffea arabica, R - Coffea robusta, M - mixtures of Coffea arabica and Coffea robusta).
3.4. Synchronous fluorescence and UV-Vis intensities versus addition of adulterant (PCA-MLR) Multiple linear regression analysis was applied to principal components (PCs) obtained for synchronous and UV-Vis spectroscopy measurements, as well as for low-level data fusion of measurements obtained with the use of both spectroscopies (Table 2). The number of principal components (PCs) incorporated into multiple linear regression analysis was established according to the Kaiser criterion [18] (Table 1). PCA-MLR (PCR) models for synchronous fluorescence as well as for data fusion of synchronous fluorescence and UV-Vis intensities were built separately for data acquired at two wavelength intervals (60 and 80 nm). For all PCA-MLR models R coefficients of at least 0.90 were obtained. Higher adjusted R2 coefficients were obtained for UV-Vis spectroscopy than for synchronous florescence spectroscopy. This suggests that UV-Vis spectroscopy, as analyzed individually, is more useful for quantification of Coffea arabica and Coffea robusta concentration in blends than synchronous fluorescence spectroscopy. Higher adjusted R2 coefficients of at least 0.95 were obtained for data fusion analysis than for individual spectroscopies. The lowest RMSEC and RMSEV for PCA-MLR models built with data obtained by applying individual spectroscopy methods equaled 5.3% and 21.9% for synchronous fluorescence spectroscopy, and 8.2% and 8.9% for UV-Vis spectroscopy. Application of UV-Vis spectroscopy produced a significantly lower error of prediction as to the level of addition of Coffea robusta to Coffea arabica than for synchronous fluorescence spectroscopy. The RMSEC and RMSEV values calculated for data fusion models (SF+UV+Vis) were lower than for individual spectroscopy methods. The lowest RMSEC and RMSEV values (3.6% and 7.9%) were obtained for the mid-level data fusion PCA-MLR model calculated for fluorescence intensities measured at Δλ=60 nm.
Fig. 2. UV-Vis spectra of Coffea arabica, Coffea robusta, and their mixtures (diluted 1:120 v/v in water) (A - Coffea arabica, R - Coffea robusta, M - mixtures of Coffea arabica and Coffea robusta).
3.3. Data dimensionality reduction with principal component analysis Principal component analysis (PCA) was employed for exploratory spectral analysis and subsequently MLR and LDA were performed. In the low-level data fusion, all data of synchronous fluorescence and UVVis spectroscopies were grouped into new matrices (separately for fluorescence data obtained at Δλ=60 and 80 nm), and then PCA was performed (Fig. 3). The number of principal components (PCs) taken for further analysis and percentage of explained variance are presented in Table 1. All selected PCs cumulatively account for at least 96% of the total variance for each spectroscopy. It is worth noting that in the case of UV-Vis spectral data the number of principal components which explained 99.16% of the variance equaled 6, while in the case of synchronous fluorescence spectra, both for low- and mid-level data fusion matrices, the number of principal components taken for further analysis exceeded 30 which indicates dispersion of information.
3.5. Linear discriminant analysis of synchronous fluorescence and UV-Vis spectra (PCA-LDA) Linear discriminant analysis was applied to principal components obtained previously by PCA. In the case of LDA, fluorescence and UVVis intensities of Coffea arabica and Coffea robusta, and their mixtures containing 40%, 50%, and 60% of adulterant were analyzed. LDA was carried out separately for synchronous fluorescence data acquired at each wavelength interval (60, 80 nm), UV-Vis intensities, as well as low- and mid-level data fusion models. Plots of the first two discriminant functions (DF1*DF2) show classification performance of the PCA217
Talanta 172 (2017) 215–220
A. Dankowska et al.
Table 1 Number of principal components chosen for further data analysis according to Kaiser criterion. Procedure/ Statistical parameter
Wavelength interval [nm]
Number of principal components (Kaiser criterion)
Cumulative proportion of variance explained
SF
60 80 –
32 32 6
96,10% 96,09% 99,16%
Low-level Data Fusion (SF +UV-Vis)
60 80
37 37
97,98% 97,95%
Mid-level Data Fusion (SF +UV-Vis)
60
38 (32 + 6)
80
38 (32 + 6)
96,10% 99,16% 96,09% 99,16%
UV-Vis
(SF), (UV-Vis) (SF), (UV-Vis)
SF – Synchronous fluorescence spectroscopy, UV-Vis – Ultraviolet and visible spectroscopy. Table 2 Statistical characteristics of multiple linear regression models calculated for individual procedures (SF and UV-Vis) and low-level and mid-level data fusion (SF+UV-Vis). Procedure/ statistical parameter
Wavelength interval [nm]
R
Adjusted R2
RMSEC
RMSECV
SF
60 80 –
0.90 0.93 0.97
0.71 0.80 0.93
5.3 5.3 8.2
21.9 22.0 8.9
Low-level Data Fusion (SF +UV-Vis)
60 80
0.99 0.99
0.98 0.95
5.4 5.3
10.1 11.3
Mid-level Data Fusion (SF +UV-Vis)
60 80
0.99 0.99
0.98 0.95
3.6 3.8
7.9 8.1
UV-Vis
SF – Synchronous fluorescence spectroscopy. UV-Vis – Ultraviolet and visible spectroscopy. R – Coefficient of correlation. Adjusted R2 – adjusted multivariate coefficient of determination. RSMEC – Root mean square error of calibration. RSMEV – Root mean square error of cross-validation.
Perhaps Robusta samples contain some characteristic factor, e.g., 16O-methylcafestol, typical for Coffea robusta and absent in Coffea arabica. The presence of this factor in the sample would explain why the mixture is closer to Coffea Robusta in the plots of the first two discriminant functions (DF1*DF2) [22]. The results presented in Fig. 4 were confirmed by a comparison of LDA results for individual procedures and for data fusion as shown in Table 3, where error classification rates in Coffea arabica, Coffea canephora var. Robusta, and their mixtures for PCA-LDA models for individual spectroscopies and data fusion are represented. The best discrimination ability of PCA-LDA among the models obtained for individual spectroscopies was observed for the UV-Vis spectroscopy with classification errors CE_CS, MCE_LOO, MCE_8020 of 6.09%, 10.43%, and 9.13%, respectively. Compared with individual observations, LDA in data fusion offers better results. It is worth noting that lower classification errors were obtained for models with fluorescence intensities obtained at wavelength intervals of 60 nm than 80 nm, and for low-level data fusion than for mid-level data fusion. Comparing LDA results between individual spectroscopies as well as data fusion of both spectroscopies, it was also found that the best classification results among all models were obtained for the low-level data fusion model with synchronous fluorescence measurements
Fig. 3. First two PCs loading plots of PCA obtained for (a) synchronous fluorescence intensities acquired at Δλ=60 nm, (b) UV-Vis, (c) low-level data fusion of synchronous fluorescence (Δλ=60 nm) and UV-Vis spectra (A - Coffea arabica, R - Coffea robusta, M mixtures of Coffea arabica and Coffea robusta).
LDA of measured samples in three clusters: Coffea arabica, Coffea robusta, and their mixtures, acquired for synchronous fluorescence at Δλ=60 nm, UV-Vis spectroscopy, and low- and mid-level data fusion of both methods (Fig. 4). Fluorescence intensities at wavelength intervals of 60 and 80 nm did not allow for good classification performance because the clusters were scattered and overlapped. It is interesting to note that for the models acquired for fluorescence, UV-Vis spectroscopy and the low- and mid-level data fusion of both methods, the group formed by mixtures of Coffea arabica and Coffea canephora var. robusta is closer to the group of Coffea canephora var. Robusta alone.
218
Talanta 172 (2017) 215–220
A. Dankowska et al.
Fig. 4. PCA-LDA plots of (a) synchronous fluorescence intensities acquired at Δλ=60 nm, (b) UV-Vis, (c) low-level data fusion of synchronous fluorescence (Δλ=60 nm) and UV-Vis spectra, and (d) mid-level data fusion of synchronous fluorescence (Δλ=60 nm) and UV-Vis spectra Coffea arabica (A), Coffea robusta (R), and their mixtures (M).
acquired at wavelength intervals of 60 nm and classification errors of 0.0% (CE_CS), 2.61% (MCE_LOO), and 3.21% (MCE_8020).
Table 3 Error classification for PCA-LDA in Coffea arabica and Coffea canephora var robusta and their mixture [%] calculated for individual procedures (SF, UV), low-level and midlevel data fusion (SF+UV-Vis). Procedure/ Statistical parameter
Wavelength interval [nm]
CE_CS classification error [%]
MCE_LOO crossvalidation leave-one-out classification error [%]
MCE_8020 crossvalidation 80 – 20 classification error [%]
SF
60 80 –
6.09 2.61 6.09
20.00 20.87 10.43
22.37 22.87 9.13
Low-level Data Fusion (SF+UV)
60 80
0.00 0.00
2.61 6.09
3.21 6.53
Mid-level Data Fusion (SF+UV)
60 80
0.00 0.00
2.61 6.96
4.84 7.93
UV
4. Conclusion The study has shown that water extracts of Coffea arabica and Coffea canefora var. robusta exhibit significant differences in their synchronous fluorescence and UV-Vis spectra patterns. A comparison of MLR and LDA results between individual spectroscopies and lowand mid-level data fusion of the two spectroscopies leads us to conclude that UV-Vis spectroscopy offers complementary information to fluorescence spectroscopy. The best prediction ability of MLR models was obtained for the mid-level data fusion model of UV-Vis and fluorescence intensities at 60 nm wavelength interval with RMSEC and RMSEV of 3.6% and 7.9%, respectively. Compared to individual spectroscopies, data fusion showed better discrimination ability with the highest classification accuracy over 96.0% obtained for the lowlevel LDA model with fluorescence intensities at 60 nm wavelength interval. This recommends to combine synchronous fluorescence and UV-Vis spectroscopies, along with chemometric analysis, for quantification of roasted Coffea arabica and Coffea canephora var. robusta concentration in blends. The findings may contribute to a better control of coffee fraud and provide a technical boost of consumer interest against profit-driven practices on the food market.
CE_CS – Classification error in calibration set of all 115 samples, MCE_LOO – Crossvalidation leave-one-out classification error, MCE_8020 - Cross-validation 80 – 20% classification error.
219
Talanta 172 (2017) 215–220
A. Dankowska et al.
[10] A. Dankowska, M. Małecka, W. Kowalewski, Application of synchronous fluorescence spectroscopy with multivariate data analysis for determination of butter adulteration, Int. J. Food Sci. Technol. 49 (2014) 2628–2634. [11] A. Dankowska, M. Małecka, W. Kowalewski, Detection of plant oil addition to cheese by synchronous fluorescence spectroscopy, Dairy Sci. Technol. 95 (2015) 413–424. [12] D. Markechová, P. Májek, J. Sádecká, Fluorescence spectroscopy and multivariate methods for the determination of brandy adulteration with mixed wine spirit, Food Chem. 159 (2014) 193–199. [13] Q. Chen, S. Qi, H. Li, X. Han, Q. Ouyang, J. Zhao, Determination of rice syrup adulterant concentration in honey using three-dimensional fluorescence spectra and multivariate calibrations, Spectrochim. Acta A 131 (2014) 177–182. [14] C.V. Di Anibal, M.S. Rodríguez, L. Albertengo, Synchronous fluorescence and multivariate classification analysis as a screening tool for determining Sudan I dye in culinary spices, Food Control. 56 (2015) 18–23. [15] M. Ntakatsane, X.M. Liu, P. Zhou, Rapid detection of milk fat adulteration with vegetable oil by fluorescence spectroscopy, J. Dairy Sc. 96 (2013) 2130–2136. [16] I. Sergiel, P. Pohl, M. Biesaga, A. Mironczyk, Suitability of three-dimensional synchronous fluorescence spectroscopy for fingerprint analysis of honey samples with reference to their phenolic profiles, Food Chem. 145 (2014) 319–326. [17] D. Patra, A.K. Mishra, Recent developments in multi-component synchronous fluorescence scan analysis, Trends Anal. Chem. 21 (2002) 787–798. [18] ΗF. Kaiser, The application of electronic computers to factors analysis, Educ. Psychol. Meas. (1960) 141–151. [19] R.C. Alvesa, S. Casal, M.R. Alves, M.B. Oliveira, Discrimination between arabica and robusta coffee species on the basis of their tocopherol profiles, Food Chem. 114 (2009) 295–299. [20] L. López-Martínez, P.L. López-de-Alba, R. García-Campos, L.M.D. León-Rodríguez, Simultaneous determination of methylxanthines in coffees and teas by UV–Vis spectrophotometry and partial least squares, Anal. Chim. Acta 493 (2003) 83–94. [21] A. Illy, R. Viani, Expresso Coffee: The Chemistry of Quality, Academic, San Diego, 1995. [22] L. Servillo, A. Giovane, R. Casale, D. Cautela, N. Onofrio, M.L. Balestrieri, D. Castaldo, Homostachydrine (pipecolic acid betaine) as authentication marker of roasted blends of Coffea arabica and Coffea canephora (Robusta) beans, Food Chem. 205 (2016), 2016, pp. 52–57.
Conflict of interest statement The authors declare no conflict of interest. References [1] U.T.C.P. Souto, M.J.C. Pontes, E.C. Silva, R.K.H. Galvão, M.C.U. Araújo, F.A.C. Sanches, F.A.S. Cunha, M.S.R. Oliveira, UV-Vis spectrometric classification of coffees by SPA-LDA, Food Chem. 119 (2010) 368–371. [2] U.T.C.P. Souto, M.F. Barbosa, H.V. Dantas, A.S. Pontes, W. Lyra, P.H.G.D. Diniz, M.C.U. Araújo, E.C. Silva, Identification of adulteration in ground roasted coffees using UV-Vis spectroscopy and SPA-LDA, Food Sci. Technol. 63 (2015) 1037–1041. [3] C. Pizzaro, I. Esteban-Díez, J.M. González-Sáiz, Mixture resolution according to the percentage of robusta variety in order to detect adulteration in roasted coffee by near infrared spectroscopy, Anal. Chim. Acta 585 (2007) 266–276. [4] J.R. Santos, M. Lopo, A. Rangel, J.A. Lopes, Exploiting near infrared spectroscopy as an analytical tool for on-line monitoring of acidity during coffee roasting, Food Control. 60 (2016) 408–415. [5] R. Calvini, A. Ulrici, J.M. Amigo, Practical comparison of sparse methods for classification of Arabica and Robusta coffee species using near infrared hyperspectral imaging, Chem. Intell. Lab. Syst. 146 (2015) 503–511. [6] F. Carrera, M. León-Camacho, F. Pablos, A.G. González, Authentication of green coffee varieties according to their sterolic profile, Anal. Chim. Acta 370 (1998) 131–139. [7] M.J. Martin, F. Pablos, A.G. González, M.S. Valdenebro, M.L. Carnacho, Fatty acid profiles as discriminant parameters for coffee varieties differentiation, Talanta 54 (2001) 291–297. [8] E.D. Pauli, F. Barbieri, P.S. Garcia, T.B. Madeira, V.R. Acquaro Junior, I.S. Scarminio, C.A.P. Camara, S.L. Nixdorf, Detection of ground roasted coffee adulteration with roasted soybean and wheat, Food Res. Int. 61 (2014) 112–119. [9] A. Dankowska, M. Małecka, W. Kowalewski, Discrimination of edible olive oils by means of synchronous fluorescence spectroscopy with multivariate data analysis, GyA 64 (2013) 425–431.
220