- Email: [email protected]
Raman spectroscopy for the differentiation of Arabic coffee genotypes
Food Chemistry 288 (2019) 262–267
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Raman spectroscopy for the differentiation of Arabic coffee genotypes a
b
c
Luisa Pereira Figueiredo , Flávio Meira Borém , Mariana Ramos Almeida , ⁎ Luiz Fernando Cappa de Oliveirad, Ana Paula de Carvalho Alvesb, Cláudia Mendes dos Santosb, , b Paula Almeida Rios
T
a
Departamento de Ciência dos Alimentos, Universidade Federal de Lavras, P.O. Box: 3037, Lavras, MG 37200-000, Brazil Departamento de Engenharia, Universidade Federal de Lavras, P.O. Box: 3037, Lavras, MG 37200-000, Brazil c Departamento de Química, Universidade Federal de Minas Gerais, Av. Antônio Carlos, n° 6627, Belo Horizonte, MG 31270-901, Brazil d Departamento de Química, Universidade Federal de Juiz de Fora, Rua José Lourenço Kelmer, s/n – Campus Universitário, Juiz de Fora, MG 36036-900, Brazil b
ARTICLE INFO
ABSTRACT
Keywords: Coffea Arabica L. Chemometric analyses Kahweol Fatty acids PCA PLS-DA
The objective of this study was to evaluate the ability of Raman spectroscopy to identify the genotype of green coffee beans. Four genotypes of Arabic coffee: one Mundo Novo line (G1) and three Bourbon lines (G2, G3, and G4). The harvest was selected using a wet processing method. Raman spectra of the samples were obtained using a FT-Raman RFS/100 spectrometer in the spectral range of 3500–400 cm−1. The data were treated using chemometric unsupervised classification tools and supervised analysis. Using the unsupervised analysis (PCA), the apparent tendency of agglomeration between samples G1 and G3 was verified. These differences were present in the spectral bands that are characteristic of fatty acids and kahweol. Based on this information, a classification model to discriminate (PLS-DA) the Mundo Novo and Bourbon samples was utilized. Raman spectroscopy allowed the building of an adequate model to differentiate between coffee genotypes.
1. Introduction Coffee has great worldwide economic importance. It is estimated that between 75 and 125 million people are connected to the coffee sector and that the global coffee consumption is approximately 155 million 60-kg bags per year (International Coffee Organization, 2017). Despite the market for commodity coffee representing the majority of the coffee transactions worldwide, the specialty coffees segment has been prominent in the international market. The coffee market grows approximately 2% each year, while the specialty coffee segment advances between 10% and 15% annually (Associação Brasileira De Cafés Especiais, 2017). In general, any cultivar of Coffea Arabica has the potential to produce high-quality coffees. However, the Bourbon cultivar is highlighted because of its intrinsic attributes and globally recognized sensorial characteristics, and it is used for the production of specialty coffees in many regions of the world. Coffee quality is expressed in many ways as a function of the crop location and is directly influenced by environmental and technological aspects (Alves et al., 2017). Through the interaction between the distinct Bourbon genotypes and different environments, as well as the phenotypic and agronomic variability, we verified that there are groups of Bourbon genotypes that are more
adapted for the production of quality coffees for each environment (Figueiredo et al., 2013). Based on the higher valuation of specialty coffees and the increasing demand for this product because of its genetic characteristics, we justify the development of techniques that aim at discriminating coffees as a function of these parameters to associate them to a sensorial profile and for detecting contaminations or frauds. Genetic factors also influence the chemical composition of coffee, which affects the quality of the final product. Thus, the quality and acceptability of coffee are directly related to the chemical composition of the beans (Figueiredo et al., 2018). Therefore, the intrinsic quality of the Bourbon cultivar is associated with the production of distinguished coffees that justifies studies aimed at understanding the relationship between these intrinsic chemical factors and the production of specialty coffees. Many studies have determined the chemical composition of coffee beans using chromatographic techniques, which have aimed at identifying their relation with the sensorial quality of coffee (Bertrand et al., 2012; Borém et al., 2016; Borém, Isquierdo, & Taveira, 2013; Figueiredo et al., 2015; Ribeiro et al., 2016). However, when considering particular techniques for determining specific groups of compounds, they are often inefficient for relating and discriminating coffees, environments, and types of processing.
Corresponding author. E-mail addresses: [email protected] (L.P. Figueiredo), [email protected] (L.F.C.d. Oliveira), [email protected] (C.M.d. Santos), ",0,0,2 >[email protected] (P.A. Rios).
⁎
https://doi.org/10.1016/j.foodchem.2019.02.093 Received 8 October 2018; Received in revised form 31 January 2019; Accepted 21 February 2019 Available online 28 February 2019 0308-8146/ © 2019 Elsevier Ltd. All rights reserved.
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al.
However, undirected methods and the use of spectroscopic techniques, such as near and mid-infrared spectroscopy (NIR and MIR), nuclear magnetic resonance (NMR), and Raman spectroscopy, have been employed to analyze and distinguish coffee genotypes and provide information on their sensorial quality (Arana et al., 2015; Cagliani, Pellegrino, Giugno, & Consonni, 2013; Marquetti, Link, Lemes, dos Scholz, & S., Valderrama, P., & Bona, E. , 2016; Monakhova et al., 2015; Rubayiza & Meurens, 2005; Wermelinger, D’Ambrosio, Klopprogge, & Yeretzian, 2011). The employment of Raman spectroscopy in the analysis of different foods has been recently demonstrated in the literature (Boyaci et al., 2015). Compared to traditional techniques, this method has the advantages of demanding minimum to no sample manipulation, the water is not an interfering substance, and it can provide information on the concentration, structure, and interaction of the molecules (Boyaci et al., 2015; El-Abassy, Donfack, & Materny, 2011; Wermelinger et al., 2011). Raman spectroscopy was employed on dairy products to identify and quantify adulterations in cream cheese (Oliveira, De Souza Callegaro, Stephani, Almeida, & De Oliveira, 2016) and powdered milk samples (Almeida, Oliveira, Stephani, & Oliveira, 2012). Technological aspects of the powdered milk were evaluated using Raman spectroscopy (Torres et al., 2017), which was also used to identify the chemical changes in whey protein concentrate (WPC) samples (Stephani et al., 2017). The use of Raman spectroscopy in coffee samples has been reported in the literature for the quick discrimination between Arabic and Robusta species (Dias & Yeretzian, 2016; El-Abassy et al., 2011; Rubayiza & Meurens, 2005; Wermelinger et al., 2011). We found no studies using this technique to discriminate between coffee genotypes. The objective of this study was to employ Raman spectroscopy to characterize green coffee beans of genotypes Bourbon and Novo Mundo. Chemometric tools were used to build a classification model. The differences in the chemical composition of green beans between the genotypes were evaluated and related to the sensorial quality of the roasted coffee for two consecutive harvests.
nitrogen was added to facilitate grinding and to avoid oxidation of the samples. The samples were then conditioned in falcon tubes and stored in a freezer at −80 °C until analysis. 2.4. Roasting and sensory evaluation All procedures were performed according to the protocol described by the Specialty Coffee Association (SCA) (Lingle, 2011). In total, 100 g of each coffee sample was roasted in a Probat TP2 (Curitiba, Brazil) for no longer than 24 h before tasting. The roasting was terminated when the coffee samples attained the desired roasting, which was visually determined using a system of color classification that employed standardized discs (SCA/Agtron Roast Color Classification System; reference color number 65 for milled beans and 55 for whole beans). The temperature and time of roasting were monitored by a thermometer and cronometer, respectively, with a time range of roasting between 8 and 12 min. Samples were weighed to obtain a predetermined ratio of 8.25 ± 0.25 g per 150 ml of water and then milled in a Mahlkönig Guatemala (Hamburg, Germany). Ten sensory attributes were evaluated by a panel of trained judges and scored on a scale of 10 points according to SCA (Lingle, 2011). The sensory attributes included aroma, uniformity, absence of injuries, sweetness, flavor, acidity, body, balance, completion and overall impression. The final sensory grade was generated from the sum of all of the evaluated attributes. For each evaluation, five cups of coffee representing each genotype were evaluated with one session of sensory analysis for each repetition and a total of three repetitions. Each environment was evaluated separately, and the results of the sensory analyses were scored on a scale representing the quality level in intervals of 0.25 points. 2.5. Raman measurements The Raman spectra of the coffee samples were obtained using a FTRaman RFS/100 spectrometer (Bruker) using an excitation Nd:YAG laser with λ = 1064 nm. The spectra were recorded in the spectral range of 3500 to 50 cm−1 at a spectral resolution of 4 cm−1. Optimized analyses were obtained with a laser power of 50 mW and 256 accumulations. The Raman spectra were obtained in duplicate to evaluate the intensity and position of the observed bands.
2. Material and methods 2.1. Samples We evaluated four genotypes of Arabic coffee (Coffea arabica L.) harvested during two agricultural years (2010 and 2011). The genotypes evaluated were Mundo Novo IAC 502/9 (G1), Yellow Bourbon IAC J9 (G2), Yellow Bourbon that originated from São Sebastião do Paraíso/SSP (G3), and Yellow Bourbon that originated from Carmo de Minas/CM (G4).
2.6. Sensory analysis The sensory attribute data were initially submitted to analysis of variance (ANOVA), and when significant differences were detected by the F test, the Scott-Knott test was applied at a significance level of 5% using the SISVAR software (Ferreira, 2011).
2.2. Coffee harvesting and processing The coffee fruits were selectively harvested and chosen if completely ripe to ensure their uniformity, integrity and high quality. Then, the coffee fruits were peeled to obtain the pulped coffee. Drying was conducted via patio drying immediately after processing according to the method of Borém, Reinato, and Isquierdo (2013) until coffee beans were at a moisture content of 11% (wet basis, w. b.).
2.7. Data treatment The baseline of the Raman spectra was corrected using the concave rubber band function of the OPUS 6 software from Bruker Optik GmbH (Ettlingen, Germany); this function removes thermal background. After correcting the baseline, the Raman spectra were organized in the form of a matrix in which the lines represented the samples and the columns represented the Raman intensities for each wavenumber. The spectral range employed was from 1800 to 650 cm−1. The data were imported into the MATLAB environment and chemometric tools, including principal component analysis (PCA) for exploratory analysis and partial least square discriminant analysis (PLS-DA) for classification, were used. Before the chemometric analysis, the Raman spectra were normalized to unit vector length to reduce systematic variations and mean centering. For the exploratory data analysis (PCA), the number of principal components was selected based on the percentage of variance explained for each component. Regarding the supervised analysis employing the PLS-DA, a vector y was built with numbers 1 and 0, where class 1 refers
2.3. Sample preparation After drying, the samples were packed in paper bags, covered with plastics bags, identified and stored in chambers at a controlled temperature of 18 °C for 60 days. Then, the samples were hulled, and the defects were removed to standardize the samples and minimize interferences that were unrelated to the genetic material. Chemical analysis and roasting were performed for beans that retained on screen 16 and higher (16, 17 and 18/64 in.). For the chemical analysis, the green coffee beans were ground for approximately 1 min in an 11A primary grinder (IKA, Brazil), and liquid 263
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al.
to the Mundo Novo coffees and class 0 refers to the Bourbon coffees. The set of data was divided into 70% for a training set (42 samples) and 30% for the test set (21 samples). The division of the samples to the training and test sets was performed using the Kennard-Stone algorithm (Kennard & Stone, 1969). In the training set, the model parameters were calculated, and we defined the threshold based on Bayesian statistics. The number of latent variables (LV) for the PLS-DA model was selected by cross validation using a random subset, and the choice was based on the smallest cross-validation classification error. The test set was used to validate the model. To evaluate the performance of the classification model and its prediction capacity, figures of merit, such as sensitivity, selectivity, classification efficiency, and Matthew coefficient of correlation, were calculated (Mendes, Porto, Almeida, Fantini, & Sena, 2019) To include the chemical characterization in the developed PLS-DA model, the most relevant variables for the model were calculated by VIP (variable importance in projection) scores and associated with spectral information. The VIP scores show the importance of the variables for the prediction ability of the model, and when the VIP value is greater than one, the variable is considered to be important for the model (Chong & Jun, 2005).
The mean Raman spectra of the four coffee genotypes (G1, G2, G3, and G4) are shown in Fig. 1. The region between 1800 and 800 cm−1 renders information on the composition of coffee beans. This spectral profile of the green coffee beans is dominated by the intense bands of chlorogenic acid observed at 1606, 1637, and 1680 cm−1. They are attributed to a coupled form of C]C and C]O bonds present in the structure of the chlorogenic acid (Eravuchira et al., 2012; Keidel, Von Stetten, Rodrigues, Máguas, & Hildebrandt, 2010). At 1657 cm−1, we verified a band corresponding to the double C bond (C]C), and three bands at 1440, 1302, and 1266 cm−1 corresponded to the deformation vibrations of the CH3, CH2, and CH groups, respectively (Silveira, Silveira, Villaverde, Pacheco, & Pasqualucci, 2010). Two weak Raman bands at 1567 and 1478 cm−1 were observed in the coffee bean spectra. These bands are characteristic of Arabica coffees and are due to kahweol content (Keidel et al., 2010). In the work of Rubayiza and Meurens (2005) Raman spectra of cafestol and
kahweol, which are diterpenes that are characteristic of lipids, were compared. The kahweol Raman spectrum has two bands at 1567 and 1478 cm−1, whereas the cafestol spectrum presents a dispersion band at 1500 cm−1. Arabica coffees show Raman bands at 1567 and 1478 cm−1, which are absent from Robusta coffees. These bands were employed to discriminate the Arabica from the Robusta coffees (Keidel et al., 2010). In this work, all the samples were of Arabica coffee, and kahweol was present in the compositions; however, the amount of this diterpene can vary between genotypes. We also observed bands at 2900 cm−1, which represent the symmetric and asymmetric stretching of the CeH bond and can be attributed to the presence of fatty acids in the beans. These bands can also be used as markers for the presence of these compounds in the coffee beans (El-Abassy et al., 2011) The Raman spectra presented very similar spectral profiles, making the discrimination between genotypes via impossible visual analysis. Thus, multivariate analysis was employed using the PCA as the first exploratory analysis stage. Principal component analysis was applied to evaluate the differences between the genotypes based on the Raman spectra. Fig. 2 shows the scores plot for the two first principal components, PC1 and PC2, which represent 59.78% of the total spectral variation in which PC1 explained 36.21% and PC2 explained 23.57% of the data variance. The low variance explained by PC1 and PC2 can be attributed to the considerable variability of the samples. The G1 coffee samples (Mundo Novo genotype) were concentrated on the positive side of the PC1, whereas the G3 coffee samples (Bourbon genotype) were agglomerated on the negative side of PC1. At the center of the scores plot, we observe a higher concentration of the G4 (Bourbon genotype) coffee samples distributed over all the space of the components. For the G2 (Bourbon genotype) coffees, we found no clear separation between samples. The tendency of a more precise separation occurred between the G1, which was the Mundo Novo genotype, and G3, which was the Bourbon/Origin SSP genotype. The loading of PC1 (Fig. 3) shows the relation between the principal component and the spectral regions, which revealed which variables are responsible for the differences observed between samples. A variable with a higher positive or negative weight has considerable importance for the component in question. The bands responsible for the separation that occurs in the PC1 are 1657, 1568, 1479, 1441, 1304, and 1265 cm−1 and refer to the samples on the positive side of the graphic, whereas the band at 1614 cm−1 refers to the samples on the negative side. Despite the Raman spectra spectral range being between
Fig. 1. Raman spectra of the Arabica coffee beans: G1 = Mundo Novo IAC502/ 9, G2 = Yellow Bourbon IAC J9, G3 = Yellow Bourbon/Origin SSP, G4 = Yellow Bourbon/Origin CM.
Fig. 2. Scores plot of PC1 × PC2 of the genotype samples: Mundo Novo IAC502/9 (G1) (▾); Yellow Bourbon IAC J9 (G2) (●); Yellow Bourbon/Origin SSP (G3) (■); and Yellow Bourbon/Origin CM (G4) (+).
3. Results and discussion 3.1. Raman spectroscopy
264
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al.
Fig. 3. Loadings of the PC1.
Fig. 4. Results of the PLS-DA model for the training and test samples.
1800 and 650 cm−1, the loadings of the PC1 (Fig. 3) revealed that the distinction between the coffees is restricted to the spectral region between 1200 and 1600 cm−1. Thus, the discrimination between the analyzed coffees, even upon consideration of the complete Raman spectrum, can be described by the general change in the bands characteristic of fatty acids, chlorogenic acids, and kahweol. As shown in Fig. 3, we observed the presence of a band at 1657 cm−1 attributed to the double carbon bond of cyclohexane (Rubayiza & Meurens, 2005) and to the presence of chlorogenic acids (Eravuchira et al., 2012), which were also observed in the region of 1614 cm−1. The bands at 1568 and 1479 cm−1 are related to the presence of kahweol and can be used as markers for kahweol in the coffee grain structure (Rubayiza & Meurens, 2005; Wermelinger et al., 2011). The bands at 1441 and 1304 cm−1 are attributed to the deformation vibrations of CH3, CH2, and CH bands (Rubayiza & Meurens, 2005) and the presence of fatty acids (Ozaki, Cho, Ikegaya, Muraishi, & Kawauchi, 1992; Silveira et al., 2010). At 1265 cm−1, we observed a band that was attributed to the deformation vibrations of CH3, CH2, and CH bonds (Rubayiza & Meurens, 2005), the presence of fatty acids (Ozaki et al., 1992; Silveira et al., 2010) and the double carbon bond (Olsen, Rukke, Flåtten, & Isaksson, 2007). In a study, Figueiredo et al. (2015) used chromatographic techniques to determine the content of fatty acids in the samples of green coffee of the four genotypes, and they observed that most of the fatty acids do not have differences between the studied genotypes, except of linoleic acid (C18: 2), which caused G1 and G4 to differ from G2 and G3 and arachidic acid (C20: 0); this also caused G4 to differ from the others. Our Raman data are consistent with the results obtained by Figueiredo et al. (2015) in the fatty acid analysis (Fig. 2). Based on this information, we built a classification model to discriminate (PLS-DA) the samples of the Mundo Novo genotype (G1) from the Bourbon genotype (G2, G3, and G4). The Bourbon coffees present higher quality and, generally, higher sensory analysis scores. The discrimination between the genotypes is essential given that the Bourbon coffees have a higher aggregated value because they service distinguished markets. Thus, a methodology to discriminate the genotypes is essential to avoid frauds. A partial least square discriminant analysis was performed to classify the Mundo Novo (G1) and Bourbon (G2, G3, and G4) samples. The result is presented in Fig. 4. The number of latent variables chosen for the model by crossvalidation was five, which represents 74.81% of the variance explained for matrix X and 95.28% in matrix Y. The threshold calculated for separating the classes was 0.6934. Below this value, the samples were classified as Bourbon coffees, and above, they were classified as Mundo Novo coffees. We observed in Fig. 4 that all samples (training and test
sets) were correctly classified, resulting in no false positive or false negative classifications. These results corroborate the good accuracy of the model with a sensitivity and selectivity of 100%, an efficiency rate of 100% and a Matthews correlation coefficient estimated at 1, which represents a perfect classification. Thus, the model built using the Raman spectroscopy data is suitable for classifying the genotype of new samples. To verify which regions of the spectra differentiate the samples that allow discrimination between genotypes, we calculated the VIP (variable importance in projection) score values (Fig. 5). The VIP scores show the importance of the variables for the prediction ability of the model. The VIP scores of the highest intensities are responsible for the separation. Analysis of the spectrum shows that the bands at 1567 and 1479 cm−1, which are related to kahweol, presented a higher contribution for differentiating between genotypes, followed by the bands at 1502 and 1442 cm−1, which are related to the cafestol compound and lipids, respectively (Rubayiza & Meurens, 2005; Wermelinger et al., 2011). Deformation forms of the torsion type of the CH2 group were at 1302 cm−1 (El-Abassy et al., 2011). The kahweol Raman bands are important for discrimination between Mundo Novo and Bourbon genotypes.
Fig. 5. VIP scores for the PLS-DA model showing the Raman bands responsible for the separation of the coffee genotypes. 265
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al.
Table 1 Final sensorial score of the four coffee genotypes. Coffee genotype
Final Sensorial Score
G1 G2 G3 G4
80.38 81.61 81.76 79.87
± ± ± ±
Associação Brasileira De Cafés Especiais (2017). Café especial: produção no Brasil deve ultrapassar Colômbia em menos de dois anos. Retrieved August 15, 2017, from http://siteantigo.bsca.com.br/noticia.php?id=453%3E. Bertrand, B., Boulanger, R., Dussert, S., Ribeyre, F., Berthiot, L., Descroix, F., & Joet, T. (2012). Climatic factors directly impact the volatile organic compound fingerprint in green Arabica coffee bean as well as coffee beverage quality. Food Chemistry, 135(4), 2575–2583. https://doi.org/10.1016/j.foodchem.2012.06.060. Borém, F. M., Figueiredo, L. P., Ribeiro, F. C., Taveira, J. H. S., Giomo, G. S., & Salva, T. J. G. (2016). The relationship between organic acids, sucrose and the quality of specialty coffees. African Journal of Agricultural Research, 11(8), 709–717. https://doi. org/10.5897/AJAR2015.10569. Borém, F. M., Isquierdo, E. P., & Taveira, J. H. S. (2013). Coffee processing. Borém, F. M Handbook of coffee post-harvest technology (pp. 49–68). . Borém, F. M., Reinato, C. H. R., & Isquierdo, E. P. (2013). Coffee drying. Borém, F. M Handbook of coffee Post-Harvest Technology (pp. 91–109). . Boyaci, I. H., Temiz, H. T., Geniş, H. E., Acar soykut, Yazgan, N. N., Bozkurt, B., ... Şeker, F. C. (2015). Dispersive and FT-Raman spectroscopic methods in food analysis. RSC Advances, 5(70), 56606–56624. https://doi.org/10.1039/C4RA12463D. Cagliani, L. R., Pellegrino, G., Giugno, G., & Consonni, R. (2013). Quantification of Coffea arabica and Coffea canephora var. robusta in roasted and ground coffee blends. Talanta, 106, 169–173. https://doi.org/10.1016/j.talanta.2012.12.003. Chong, I. G., & Jun, C. H. (2005). Performance of some variable selection methods when multicollinearity is present. Chemometrics and Intelligent Laboratory Systems, 78(1–2), 103–112 10.1016/j.chemolab.2004.12.011. Dias, R. C. E., & Yeretzian, C. (2016). Investigating Coffee Samples by Raman Spectroscopy for Quality Control – Preliminary Study. International Journal of Experimental Spectroscopic Techniques, 1(2), 1–5. El-Abassy, R. M., Donfack, P., & Materny, A. (2011). Discrimination between Arabica and Robusta green coffee using visible micro Raman spectroscopy and chemometric analysis. Food Chemistry, 126(3), 1443–1448. https://doi.org/10.1016/j.foodchem. 2010.11.132. Eravuchira, P. J., El-Abassy, R. M., Deshpande, S., Matei, M. F., Mishra, S., Tandon, P., ... Materny, A. (2012). Raman spectroscopic characterization of different regioisomers of monoacyl and diacyl chlorogenic acid. Vibrational Spectroscopy, 61, 10–16. https:// doi.org/10.1016/j.vibspec.2012.02.009. Ferreira, D. F. (2011). Sisvar: A computer statistical analysis system. Ciência e Agrotecnologia, 35(6), 1039–1042. https://doi.org/10.1590/S141370542011000600001. Figueiredo, L. P., Borém, F. M., Cirillo, M. A., Ribeiro, F. C., Giomo, G. S., & Salva, T. J. G. (2013). The potential for high quality bourbon coffees from different environments. Journal of Agricultural Science, 5(10), 87–98. https://doi.org/10.5539/jas.v5n10p87. Figueiredo, L. P., Borém, F. M., Ribeiro, F. C., Giomo, G. S., Malta, M. R., & Taveira, J. H. S. (2018). Sensory analysis and chemical composition of ‘bourbon’ coffees cultivated in different environments. Coffee Science, 13(1)https://doi.org/10.25186/cs.v13i1. 1400. Figueiredo, L. P., Borem, F. M., Ribeiro, F. C., Giomo, G. S., Taveira, J. H. S., & Malta, M. R. (2015). Fatty acid profiles and parameters of quality of specialty coffees produced in different Brazilian regions. African Journal of Agricultural Research, 10(35), 3484–3493. https://doi.org/10.5897/AJAR2015.9697. Keidel, A., Von Stetten, D., Rodrigues, C., Máguas, C., & Hildebrandt, P. (2010). Discrimination of green arabica and Robusta coffee beans by Raman spectroscopy. Journal of Agricultural and Food Chemistry, 58(21), 11187–11192. https://doi.org/10. 1021/jf101999c. Kennard, R. W., & Stone, L. A. (1969). Computer aided design of experiments. Technometrics, 11(1), 137–148. https://doi.org/10.1080/00401706.1969.10490666. Lingle, T. (2011). The coffee cupper’s handbook (4th ed.). Long Beach: Specialty Coffee Association of America. Marquetti, I., Link, J. V., Lemes, A. L. G., dos Scholz, M. B. S, Valderrama, P., & Bona, E. (2016). Partial least square with discriminant analysis and near infrared spectroscopy for evaluation of geographic and genotypic origin of arabica coffee. Computers and Electronics in Agriculture, 121, 313–319. https://doi.org/10.1016/j.compag.2015.12. 018. Mendes, T. O., Porto, B. L. S., Almeida, M. R., Fantini, C., & Sena, M. M. (2019). Discrimination between conventional and omega-3 fatty acids enriched eggs by FTRaman spectroscopy and chemometric tools. Food Chemistry, 273(1), 144–150. https://doi.org/10.1016/j.foodchem.2017.12.084. Monakhova, Y. B., Ruge, W., Kuballa, T., Ilse, M., Winkelmann, O., Diehl, B., ... Lachenmeier, D. W. (2015). Rapid approach to identify the presence of Arabica and Robusta species in coffee using1H NMR spectroscopy. Food Chemistry, 182, 178–184. https://doi.org/10.1016/j.foodchem.2015.02.132. Oliveira, K. S., De Souza Callegaro, L., Stephani, R., Almeida, M. R., & De Oliveira, L. F. C. (2016). Analysis of spreadable cheese by Raman spectroscopy and chemometric tools. Food Chemistry, 194, 441–446. https://doi.org/10.1016/j.foodchem.2015.08.039. Olsen, E. F., Rukke, E. O., Flåtten, A., & Isaksson, T. (2007). Quantitative determination of saturated-, monounsaturated- and polyunsaturated fatty acids in pork adipose tissue with non-destructive Raman spectroscopy. Meat Science, 76(4), 628–634. https://doi. org/10.1016/j.meatsci.2007.02.004. Organization, I. C. (2017). World coffee consumption. Ozaki, Y., Cho, R., Ikegaya, K., Muraishi, S., & Kawauchi, K. (1992). Potential of nearinfrared fourier transform raman spectroscopy in food analysis. Applied Spectroscopy, 46(10), 1503–1507. https://doi.org/10.1366/000370292789619368. Ribeiro, D. E., Borem, F. M., Cirillo, M. A., Prado, M. V. B., Ferraz, V. P., Helena, M. R. A., & Taveira, J. H. S. (2016). Interaction of genotype, environment and processing in the chemical composition expression and sensorial quality of Arabica coffee. African Journal of Agricultural Research, 11(27), 2412–2422. https://doi.org/10.5897/ AJAR2016.10832.
1.3a 0.6b 0.6b 0.1a
The data are the means of a triplicate ± standard deviation. G1 = Mundo Novo IAC 502/9, G2 = Yellow Bourbon IAC J9, G3 = Yellow Bourbon/Origin SSP, and G4 = Yellow Bourbon/ Origin CM.
3.2. Sensory analysis Table 1 presents the final scores of the sensory attributes of the four genotypes. The distinction between the coffees was shown from the classification model using the Raman spectra, and we found G1 (Novo Mundo) and G4 (Bourbon) to be statistically different from G2 and G3 (Bourbon). In this study, the sensory scores were very close, which may have resulted in difficulties with differentiating by sensory scores. Therefore, we believe that such differentiation can become even more evident when more distinct coffee profiles are compared, which justifies new studies comparing Raman and sensory analyses of coffee. 4. Conclusion With the aid of chemometry, Raman spectroscopy has enabled the distinction between coffee genotypes. The compounds that most contributed to this discrimination (PCA) were kahweol and fatty acids, especially when observing the bands at 1567 and 1479 cm−1, which were attributed to the kahweol, and 1442 and 1302 cm−1, which were attributed to fatty acids. Cafestol was also observed at 1502 cm−1 but with a lower contribution. The discriminant analysis by partial least squares (PLS-DA) performed for discriminating the samples (training and test set) of the Mundo Novo (G1) and Bourbon (G2, G3, and G4) genotypes was efficient, and it clearly separated both groups. Therefore, the model built with Raman spectroscopy data is suitable for discriminating Bourbon genotype samples. Sensorial analysis did not allow to discriminate coffee genotypes as well as discriminated by Raman. Conflict of interest The authors declare that they have no conflicts of interest. Acknowledgments The authors thank the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), A Fundação de Amparo à Pesquisa do Estado de Minas Gerais (Fapemig), a Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (Capes) e ao Instituto Nacional de Ciencia e Tecnologia do Café (INCTCafé). References Almeida, M. R., Oliveira, K. S., Stephani, R., & Oliveira, L. F. C. (2012). Application of FTRaman spectroscopy and chemometric analysis for determination of adulteration in milk powder. Analytical Letters, 45(17), 2589–2602. https://doi.org/10.1080/ 00032719.2012.698672. Alves, G. E., Borém, F. M., Isquierdo, E. P., Siqueira, V. C., Cirillo, M. A., & Pinto, A. C. P. (2017). Physiological and sensorial quality of Arabica coffee subjected to different temperatures and drying airflows. Acta Scientiarum Agronomy, 39(2), 225. https:// doi.org/10.4025/actasciagron.v39i2.31065. Arana, V. A., Medina, J., Alarcon, R., Moreno, E., Heintz, L., Schäfer, H., & Wist, J. (2015). Coffee’s country of origin determined by NMR: The Colombian case. Food Chemistry, 175, 500–506. https://doi.org/10.1016/j.foodchem.2014.11.160.
266
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al. Rubayiza, A. B., & Meurens, M. (2005). Chemical discrimination of arabica and robusta coffees by Fourier transform raman spectroscopy. Journal of Agricultural and Food Chemistry, 53(12), 4654–4659. https://doi.org/10.1021/jf0478657. Silveira, F. L., Silveira, L., Villaverde, A. B., Pacheco, M. T. T., & Pasqualucci, C. A. (2010). Use of dispersive raman spectroscopy in the determination of unsaturated fat in commercial edible oil- and fat-containing industrialized foods. Instrumentation Science and Technology, 38(1), 107–123. https://doi.org/10.1080/10739140903430339. Stephani, R., de Oliveira, K., de, S., Almeida, C. E. R., Perrone, Í. T., Carvalho, A. F., Oliveira, L. F. C., & Almeida, M. R. (2017). Raman spectroscopy as a tool to identify modification of whey protein concentrate (WPC) during shelf life. Food Packaging and
Shelf Life, 11, 1–9. https://doi.org/10.1016/j.fpsl.2016.10.001. Torres, J. K. F., Stephani, R., Tavares, G. M., Carvalho, A. F., Costa, R. G. B., Almeida, C. E. R., ... Perrone, Í. T. (2017). Technological aspects of lactose-hydrolyzed milk powder. Food Research International, 101, 45–53. https://doi.org/10.1016/j.foodres.2017.08. 043. Wermelinger, T., D’Ambrosio, L., Klopprogge, B., & Yeretzian, C. (2011). Quantification of the robusta fraction in a coffee blend via Raman spectroscopy: Proof of principle. Journal of Agricultural and Food Chemistry, 59(17), 9074–9079. https://doi.org/10. 1021/jf201918a.
267
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Raman spectroscopy for the differentiation of Arabic coffee genotypes a
b
c
Luisa Pereira Figueiredo , Flávio Meira Borém , Mariana Ramos Almeida , ⁎ Luiz Fernando Cappa de Oliveirad, Ana Paula de Carvalho Alvesb, Cláudia Mendes dos Santosb, , b Paula Almeida Rios
T
a
Departamento de Ciência dos Alimentos, Universidade Federal de Lavras, P.O. Box: 3037, Lavras, MG 37200-000, Brazil Departamento de Engenharia, Universidade Federal de Lavras, P.O. Box: 3037, Lavras, MG 37200-000, Brazil c Departamento de Química, Universidade Federal de Minas Gerais, Av. Antônio Carlos, n° 6627, Belo Horizonte, MG 31270-901, Brazil d Departamento de Química, Universidade Federal de Juiz de Fora, Rua José Lourenço Kelmer, s/n – Campus Universitário, Juiz de Fora, MG 36036-900, Brazil b
ARTICLE INFO
ABSTRACT
Keywords: Coffea Arabica L. Chemometric analyses Kahweol Fatty acids PCA PLS-DA
The objective of this study was to evaluate the ability of Raman spectroscopy to identify the genotype of green coffee beans. Four genotypes of Arabic coffee: one Mundo Novo line (G1) and three Bourbon lines (G2, G3, and G4). The harvest was selected using a wet processing method. Raman spectra of the samples were obtained using a FT-Raman RFS/100 spectrometer in the spectral range of 3500–400 cm−1. The data were treated using chemometric unsupervised classification tools and supervised analysis. Using the unsupervised analysis (PCA), the apparent tendency of agglomeration between samples G1 and G3 was verified. These differences were present in the spectral bands that are characteristic of fatty acids and kahweol. Based on this information, a classification model to discriminate (PLS-DA) the Mundo Novo and Bourbon samples was utilized. Raman spectroscopy allowed the building of an adequate model to differentiate between coffee genotypes.
1. Introduction Coffee has great worldwide economic importance. It is estimated that between 75 and 125 million people are connected to the coffee sector and that the global coffee consumption is approximately 155 million 60-kg bags per year (International Coffee Organization, 2017). Despite the market for commodity coffee representing the majority of the coffee transactions worldwide, the specialty coffees segment has been prominent in the international market. The coffee market grows approximately 2% each year, while the specialty coffee segment advances between 10% and 15% annually (Associação Brasileira De Cafés Especiais, 2017). In general, any cultivar of Coffea Arabica has the potential to produce high-quality coffees. However, the Bourbon cultivar is highlighted because of its intrinsic attributes and globally recognized sensorial characteristics, and it is used for the production of specialty coffees in many regions of the world. Coffee quality is expressed in many ways as a function of the crop location and is directly influenced by environmental and technological aspects (Alves et al., 2017). Through the interaction between the distinct Bourbon genotypes and different environments, as well as the phenotypic and agronomic variability, we verified that there are groups of Bourbon genotypes that are more
adapted for the production of quality coffees for each environment (Figueiredo et al., 2013). Based on the higher valuation of specialty coffees and the increasing demand for this product because of its genetic characteristics, we justify the development of techniques that aim at discriminating coffees as a function of these parameters to associate them to a sensorial profile and for detecting contaminations or frauds. Genetic factors also influence the chemical composition of coffee, which affects the quality of the final product. Thus, the quality and acceptability of coffee are directly related to the chemical composition of the beans (Figueiredo et al., 2018). Therefore, the intrinsic quality of the Bourbon cultivar is associated with the production of distinguished coffees that justifies studies aimed at understanding the relationship between these intrinsic chemical factors and the production of specialty coffees. Many studies have determined the chemical composition of coffee beans using chromatographic techniques, which have aimed at identifying their relation with the sensorial quality of coffee (Bertrand et al., 2012; Borém et al., 2016; Borém, Isquierdo, & Taveira, 2013; Figueiredo et al., 2015; Ribeiro et al., 2016). However, when considering particular techniques for determining specific groups of compounds, they are often inefficient for relating and discriminating coffees, environments, and types of processing.
Corresponding author. E-mail addresses: [email protected] (L.P. Figueiredo), [email protected] (L.F.C.d. Oliveira), [email protected] (C.M.d. Santos), ",0,0,2 >[email protected] (P.A. Rios).
⁎
https://doi.org/10.1016/j.foodchem.2019.02.093 Received 8 October 2018; Received in revised form 31 January 2019; Accepted 21 February 2019 Available online 28 February 2019 0308-8146/ © 2019 Elsevier Ltd. All rights reserved.
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al.
However, undirected methods and the use of spectroscopic techniques, such as near and mid-infrared spectroscopy (NIR and MIR), nuclear magnetic resonance (NMR), and Raman spectroscopy, have been employed to analyze and distinguish coffee genotypes and provide information on their sensorial quality (Arana et al., 2015; Cagliani, Pellegrino, Giugno, & Consonni, 2013; Marquetti, Link, Lemes, dos Scholz, & S., Valderrama, P., & Bona, E. , 2016; Monakhova et al., 2015; Rubayiza & Meurens, 2005; Wermelinger, D’Ambrosio, Klopprogge, & Yeretzian, 2011). The employment of Raman spectroscopy in the analysis of different foods has been recently demonstrated in the literature (Boyaci et al., 2015). Compared to traditional techniques, this method has the advantages of demanding minimum to no sample manipulation, the water is not an interfering substance, and it can provide information on the concentration, structure, and interaction of the molecules (Boyaci et al., 2015; El-Abassy, Donfack, & Materny, 2011; Wermelinger et al., 2011). Raman spectroscopy was employed on dairy products to identify and quantify adulterations in cream cheese (Oliveira, De Souza Callegaro, Stephani, Almeida, & De Oliveira, 2016) and powdered milk samples (Almeida, Oliveira, Stephani, & Oliveira, 2012). Technological aspects of the powdered milk were evaluated using Raman spectroscopy (Torres et al., 2017), which was also used to identify the chemical changes in whey protein concentrate (WPC) samples (Stephani et al., 2017). The use of Raman spectroscopy in coffee samples has been reported in the literature for the quick discrimination between Arabic and Robusta species (Dias & Yeretzian, 2016; El-Abassy et al., 2011; Rubayiza & Meurens, 2005; Wermelinger et al., 2011). We found no studies using this technique to discriminate between coffee genotypes. The objective of this study was to employ Raman spectroscopy to characterize green coffee beans of genotypes Bourbon and Novo Mundo. Chemometric tools were used to build a classification model. The differences in the chemical composition of green beans between the genotypes were evaluated and related to the sensorial quality of the roasted coffee for two consecutive harvests.
nitrogen was added to facilitate grinding and to avoid oxidation of the samples. The samples were then conditioned in falcon tubes and stored in a freezer at −80 °C until analysis. 2.4. Roasting and sensory evaluation All procedures were performed according to the protocol described by the Specialty Coffee Association (SCA) (Lingle, 2011). In total, 100 g of each coffee sample was roasted in a Probat TP2 (Curitiba, Brazil) for no longer than 24 h before tasting. The roasting was terminated when the coffee samples attained the desired roasting, which was visually determined using a system of color classification that employed standardized discs (SCA/Agtron Roast Color Classification System; reference color number 65 for milled beans and 55 for whole beans). The temperature and time of roasting were monitored by a thermometer and cronometer, respectively, with a time range of roasting between 8 and 12 min. Samples were weighed to obtain a predetermined ratio of 8.25 ± 0.25 g per 150 ml of water and then milled in a Mahlkönig Guatemala (Hamburg, Germany). Ten sensory attributes were evaluated by a panel of trained judges and scored on a scale of 10 points according to SCA (Lingle, 2011). The sensory attributes included aroma, uniformity, absence of injuries, sweetness, flavor, acidity, body, balance, completion and overall impression. The final sensory grade was generated from the sum of all of the evaluated attributes. For each evaluation, five cups of coffee representing each genotype were evaluated with one session of sensory analysis for each repetition and a total of three repetitions. Each environment was evaluated separately, and the results of the sensory analyses were scored on a scale representing the quality level in intervals of 0.25 points. 2.5. Raman measurements The Raman spectra of the coffee samples were obtained using a FTRaman RFS/100 spectrometer (Bruker) using an excitation Nd:YAG laser with λ = 1064 nm. The spectra were recorded in the spectral range of 3500 to 50 cm−1 at a spectral resolution of 4 cm−1. Optimized analyses were obtained with a laser power of 50 mW and 256 accumulations. The Raman spectra were obtained in duplicate to evaluate the intensity and position of the observed bands.
2. Material and methods 2.1. Samples We evaluated four genotypes of Arabic coffee (Coffea arabica L.) harvested during two agricultural years (2010 and 2011). The genotypes evaluated were Mundo Novo IAC 502/9 (G1), Yellow Bourbon IAC J9 (G2), Yellow Bourbon that originated from São Sebastião do Paraíso/SSP (G3), and Yellow Bourbon that originated from Carmo de Minas/CM (G4).
2.6. Sensory analysis The sensory attribute data were initially submitted to analysis of variance (ANOVA), and when significant differences were detected by the F test, the Scott-Knott test was applied at a significance level of 5% using the SISVAR software (Ferreira, 2011).
2.2. Coffee harvesting and processing The coffee fruits were selectively harvested and chosen if completely ripe to ensure their uniformity, integrity and high quality. Then, the coffee fruits were peeled to obtain the pulped coffee. Drying was conducted via patio drying immediately after processing according to the method of Borém, Reinato, and Isquierdo (2013) until coffee beans were at a moisture content of 11% (wet basis, w. b.).
2.7. Data treatment The baseline of the Raman spectra was corrected using the concave rubber band function of the OPUS 6 software from Bruker Optik GmbH (Ettlingen, Germany); this function removes thermal background. After correcting the baseline, the Raman spectra were organized in the form of a matrix in which the lines represented the samples and the columns represented the Raman intensities for each wavenumber. The spectral range employed was from 1800 to 650 cm−1. The data were imported into the MATLAB environment and chemometric tools, including principal component analysis (PCA) for exploratory analysis and partial least square discriminant analysis (PLS-DA) for classification, were used. Before the chemometric analysis, the Raman spectra were normalized to unit vector length to reduce systematic variations and mean centering. For the exploratory data analysis (PCA), the number of principal components was selected based on the percentage of variance explained for each component. Regarding the supervised analysis employing the PLS-DA, a vector y was built with numbers 1 and 0, where class 1 refers
2.3. Sample preparation After drying, the samples were packed in paper bags, covered with plastics bags, identified and stored in chambers at a controlled temperature of 18 °C for 60 days. Then, the samples were hulled, and the defects were removed to standardize the samples and minimize interferences that were unrelated to the genetic material. Chemical analysis and roasting were performed for beans that retained on screen 16 and higher (16, 17 and 18/64 in.). For the chemical analysis, the green coffee beans were ground for approximately 1 min in an 11A primary grinder (IKA, Brazil), and liquid 263
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al.
to the Mundo Novo coffees and class 0 refers to the Bourbon coffees. The set of data was divided into 70% for a training set (42 samples) and 30% for the test set (21 samples). The division of the samples to the training and test sets was performed using the Kennard-Stone algorithm (Kennard & Stone, 1969). In the training set, the model parameters were calculated, and we defined the threshold based on Bayesian statistics. The number of latent variables (LV) for the PLS-DA model was selected by cross validation using a random subset, and the choice was based on the smallest cross-validation classification error. The test set was used to validate the model. To evaluate the performance of the classification model and its prediction capacity, figures of merit, such as sensitivity, selectivity, classification efficiency, and Matthew coefficient of correlation, were calculated (Mendes, Porto, Almeida, Fantini, & Sena, 2019) To include the chemical characterization in the developed PLS-DA model, the most relevant variables for the model were calculated by VIP (variable importance in projection) scores and associated with spectral information. The VIP scores show the importance of the variables for the prediction ability of the model, and when the VIP value is greater than one, the variable is considered to be important for the model (Chong & Jun, 2005).
The mean Raman spectra of the four coffee genotypes (G1, G2, G3, and G4) are shown in Fig. 1. The region between 1800 and 800 cm−1 renders information on the composition of coffee beans. This spectral profile of the green coffee beans is dominated by the intense bands of chlorogenic acid observed at 1606, 1637, and 1680 cm−1. They are attributed to a coupled form of C]C and C]O bonds present in the structure of the chlorogenic acid (Eravuchira et al., 2012; Keidel, Von Stetten, Rodrigues, Máguas, & Hildebrandt, 2010). At 1657 cm−1, we verified a band corresponding to the double C bond (C]C), and three bands at 1440, 1302, and 1266 cm−1 corresponded to the deformation vibrations of the CH3, CH2, and CH groups, respectively (Silveira, Silveira, Villaverde, Pacheco, & Pasqualucci, 2010). Two weak Raman bands at 1567 and 1478 cm−1 were observed in the coffee bean spectra. These bands are characteristic of Arabica coffees and are due to kahweol content (Keidel et al., 2010). In the work of Rubayiza and Meurens (2005) Raman spectra of cafestol and
kahweol, which are diterpenes that are characteristic of lipids, were compared. The kahweol Raman spectrum has two bands at 1567 and 1478 cm−1, whereas the cafestol spectrum presents a dispersion band at 1500 cm−1. Arabica coffees show Raman bands at 1567 and 1478 cm−1, which are absent from Robusta coffees. These bands were employed to discriminate the Arabica from the Robusta coffees (Keidel et al., 2010). In this work, all the samples were of Arabica coffee, and kahweol was present in the compositions; however, the amount of this diterpene can vary between genotypes. We also observed bands at 2900 cm−1, which represent the symmetric and asymmetric stretching of the CeH bond and can be attributed to the presence of fatty acids in the beans. These bands can also be used as markers for the presence of these compounds in the coffee beans (El-Abassy et al., 2011) The Raman spectra presented very similar spectral profiles, making the discrimination between genotypes via impossible visual analysis. Thus, multivariate analysis was employed using the PCA as the first exploratory analysis stage. Principal component analysis was applied to evaluate the differences between the genotypes based on the Raman spectra. Fig. 2 shows the scores plot for the two first principal components, PC1 and PC2, which represent 59.78% of the total spectral variation in which PC1 explained 36.21% and PC2 explained 23.57% of the data variance. The low variance explained by PC1 and PC2 can be attributed to the considerable variability of the samples. The G1 coffee samples (Mundo Novo genotype) were concentrated on the positive side of the PC1, whereas the G3 coffee samples (Bourbon genotype) were agglomerated on the negative side of PC1. At the center of the scores plot, we observe a higher concentration of the G4 (Bourbon genotype) coffee samples distributed over all the space of the components. For the G2 (Bourbon genotype) coffees, we found no clear separation between samples. The tendency of a more precise separation occurred between the G1, which was the Mundo Novo genotype, and G3, which was the Bourbon/Origin SSP genotype. The loading of PC1 (Fig. 3) shows the relation between the principal component and the spectral regions, which revealed which variables are responsible for the differences observed between samples. A variable with a higher positive or negative weight has considerable importance for the component in question. The bands responsible for the separation that occurs in the PC1 are 1657, 1568, 1479, 1441, 1304, and 1265 cm−1 and refer to the samples on the positive side of the graphic, whereas the band at 1614 cm−1 refers to the samples on the negative side. Despite the Raman spectra spectral range being between
Fig. 1. Raman spectra of the Arabica coffee beans: G1 = Mundo Novo IAC502/ 9, G2 = Yellow Bourbon IAC J9, G3 = Yellow Bourbon/Origin SSP, G4 = Yellow Bourbon/Origin CM.
Fig. 2. Scores plot of PC1 × PC2 of the genotype samples: Mundo Novo IAC502/9 (G1) (▾); Yellow Bourbon IAC J9 (G2) (●); Yellow Bourbon/Origin SSP (G3) (■); and Yellow Bourbon/Origin CM (G4) (+).
3. Results and discussion 3.1. Raman spectroscopy
264
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al.
Fig. 3. Loadings of the PC1.
Fig. 4. Results of the PLS-DA model for the training and test samples.
1800 and 650 cm−1, the loadings of the PC1 (Fig. 3) revealed that the distinction between the coffees is restricted to the spectral region between 1200 and 1600 cm−1. Thus, the discrimination between the analyzed coffees, even upon consideration of the complete Raman spectrum, can be described by the general change in the bands characteristic of fatty acids, chlorogenic acids, and kahweol. As shown in Fig. 3, we observed the presence of a band at 1657 cm−1 attributed to the double carbon bond of cyclohexane (Rubayiza & Meurens, 2005) and to the presence of chlorogenic acids (Eravuchira et al., 2012), which were also observed in the region of 1614 cm−1. The bands at 1568 and 1479 cm−1 are related to the presence of kahweol and can be used as markers for kahweol in the coffee grain structure (Rubayiza & Meurens, 2005; Wermelinger et al., 2011). The bands at 1441 and 1304 cm−1 are attributed to the deformation vibrations of CH3, CH2, and CH bands (Rubayiza & Meurens, 2005) and the presence of fatty acids (Ozaki, Cho, Ikegaya, Muraishi, & Kawauchi, 1992; Silveira et al., 2010). At 1265 cm−1, we observed a band that was attributed to the deformation vibrations of CH3, CH2, and CH bonds (Rubayiza & Meurens, 2005), the presence of fatty acids (Ozaki et al., 1992; Silveira et al., 2010) and the double carbon bond (Olsen, Rukke, Flåtten, & Isaksson, 2007). In a study, Figueiredo et al. (2015) used chromatographic techniques to determine the content of fatty acids in the samples of green coffee of the four genotypes, and they observed that most of the fatty acids do not have differences between the studied genotypes, except of linoleic acid (C18: 2), which caused G1 and G4 to differ from G2 and G3 and arachidic acid (C20: 0); this also caused G4 to differ from the others. Our Raman data are consistent with the results obtained by Figueiredo et al. (2015) in the fatty acid analysis (Fig. 2). Based on this information, we built a classification model to discriminate (PLS-DA) the samples of the Mundo Novo genotype (G1) from the Bourbon genotype (G2, G3, and G4). The Bourbon coffees present higher quality and, generally, higher sensory analysis scores. The discrimination between the genotypes is essential given that the Bourbon coffees have a higher aggregated value because they service distinguished markets. Thus, a methodology to discriminate the genotypes is essential to avoid frauds. A partial least square discriminant analysis was performed to classify the Mundo Novo (G1) and Bourbon (G2, G3, and G4) samples. The result is presented in Fig. 4. The number of latent variables chosen for the model by crossvalidation was five, which represents 74.81% of the variance explained for matrix X and 95.28% in matrix Y. The threshold calculated for separating the classes was 0.6934. Below this value, the samples were classified as Bourbon coffees, and above, they were classified as Mundo Novo coffees. We observed in Fig. 4 that all samples (training and test
sets) were correctly classified, resulting in no false positive or false negative classifications. These results corroborate the good accuracy of the model with a sensitivity and selectivity of 100%, an efficiency rate of 100% and a Matthews correlation coefficient estimated at 1, which represents a perfect classification. Thus, the model built using the Raman spectroscopy data is suitable for classifying the genotype of new samples. To verify which regions of the spectra differentiate the samples that allow discrimination between genotypes, we calculated the VIP (variable importance in projection) score values (Fig. 5). The VIP scores show the importance of the variables for the prediction ability of the model. The VIP scores of the highest intensities are responsible for the separation. Analysis of the spectrum shows that the bands at 1567 and 1479 cm−1, which are related to kahweol, presented a higher contribution for differentiating between genotypes, followed by the bands at 1502 and 1442 cm−1, which are related to the cafestol compound and lipids, respectively (Rubayiza & Meurens, 2005; Wermelinger et al., 2011). Deformation forms of the torsion type of the CH2 group were at 1302 cm−1 (El-Abassy et al., 2011). The kahweol Raman bands are important for discrimination between Mundo Novo and Bourbon genotypes.
Fig. 5. VIP scores for the PLS-DA model showing the Raman bands responsible for the separation of the coffee genotypes. 265
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al.
Table 1 Final sensorial score of the four coffee genotypes. Coffee genotype
Final Sensorial Score
G1 G2 G3 G4
80.38 81.61 81.76 79.87
± ± ± ±
Associação Brasileira De Cafés Especiais (2017). Café especial: produção no Brasil deve ultrapassar Colômbia em menos de dois anos. Retrieved August 15, 2017, from http://siteantigo.bsca.com.br/noticia.php?id=453%3E. Bertrand, B., Boulanger, R., Dussert, S., Ribeyre, F., Berthiot, L., Descroix, F., & Joet, T. (2012). Climatic factors directly impact the volatile organic compound fingerprint in green Arabica coffee bean as well as coffee beverage quality. Food Chemistry, 135(4), 2575–2583. https://doi.org/10.1016/j.foodchem.2012.06.060. Borém, F. M., Figueiredo, L. P., Ribeiro, F. C., Taveira, J. H. S., Giomo, G. S., & Salva, T. J. G. (2016). The relationship between organic acids, sucrose and the quality of specialty coffees. African Journal of Agricultural Research, 11(8), 709–717. https://doi. org/10.5897/AJAR2015.10569. Borém, F. M., Isquierdo, E. P., & Taveira, J. H. S. (2013). Coffee processing. Borém, F. M Handbook of coffee post-harvest technology (pp. 49–68). . Borém, F. M., Reinato, C. H. R., & Isquierdo, E. P. (2013). Coffee drying. Borém, F. M Handbook of coffee Post-Harvest Technology (pp. 91–109). . Boyaci, I. H., Temiz, H. T., Geniş, H. E., Acar soykut, Yazgan, N. N., Bozkurt, B., ... Şeker, F. C. (2015). Dispersive and FT-Raman spectroscopic methods in food analysis. RSC Advances, 5(70), 56606–56624. https://doi.org/10.1039/C4RA12463D. Cagliani, L. R., Pellegrino, G., Giugno, G., & Consonni, R. (2013). Quantification of Coffea arabica and Coffea canephora var. robusta in roasted and ground coffee blends. Talanta, 106, 169–173. https://doi.org/10.1016/j.talanta.2012.12.003. Chong, I. G., & Jun, C. H. (2005). Performance of some variable selection methods when multicollinearity is present. Chemometrics and Intelligent Laboratory Systems, 78(1–2), 103–112 10.1016/j.chemolab.2004.12.011. Dias, R. C. E., & Yeretzian, C. (2016). Investigating Coffee Samples by Raman Spectroscopy for Quality Control – Preliminary Study. International Journal of Experimental Spectroscopic Techniques, 1(2), 1–5. El-Abassy, R. M., Donfack, P., & Materny, A. (2011). Discrimination between Arabica and Robusta green coffee using visible micro Raman spectroscopy and chemometric analysis. Food Chemistry, 126(3), 1443–1448. https://doi.org/10.1016/j.foodchem. 2010.11.132. Eravuchira, P. J., El-Abassy, R. M., Deshpande, S., Matei, M. F., Mishra, S., Tandon, P., ... Materny, A. (2012). Raman spectroscopic characterization of different regioisomers of monoacyl and diacyl chlorogenic acid. Vibrational Spectroscopy, 61, 10–16. https:// doi.org/10.1016/j.vibspec.2012.02.009. Ferreira, D. F. (2011). Sisvar: A computer statistical analysis system. Ciência e Agrotecnologia, 35(6), 1039–1042. https://doi.org/10.1590/S141370542011000600001. Figueiredo, L. P., Borém, F. M., Cirillo, M. A., Ribeiro, F. C., Giomo, G. S., & Salva, T. J. G. (2013). The potential for high quality bourbon coffees from different environments. Journal of Agricultural Science, 5(10), 87–98. https://doi.org/10.5539/jas.v5n10p87. Figueiredo, L. P., Borém, F. M., Ribeiro, F. C., Giomo, G. S., Malta, M. R., & Taveira, J. H. S. (2018). Sensory analysis and chemical composition of ‘bourbon’ coffees cultivated in different environments. Coffee Science, 13(1)https://doi.org/10.25186/cs.v13i1. 1400. Figueiredo, L. P., Borem, F. M., Ribeiro, F. C., Giomo, G. S., Taveira, J. H. S., & Malta, M. R. (2015). Fatty acid profiles and parameters of quality of specialty coffees produced in different Brazilian regions. African Journal of Agricultural Research, 10(35), 3484–3493. https://doi.org/10.5897/AJAR2015.9697. Keidel, A., Von Stetten, D., Rodrigues, C., Máguas, C., & Hildebrandt, P. (2010). Discrimination of green arabica and Robusta coffee beans by Raman spectroscopy. Journal of Agricultural and Food Chemistry, 58(21), 11187–11192. https://doi.org/10. 1021/jf101999c. Kennard, R. W., & Stone, L. A. (1969). Computer aided design of experiments. Technometrics, 11(1), 137–148. https://doi.org/10.1080/00401706.1969.10490666. Lingle, T. (2011). The coffee cupper’s handbook (4th ed.). Long Beach: Specialty Coffee Association of America. Marquetti, I., Link, J. V., Lemes, A. L. G., dos Scholz, M. B. S, Valderrama, P., & Bona, E. (2016). Partial least square with discriminant analysis and near infrared spectroscopy for evaluation of geographic and genotypic origin of arabica coffee. Computers and Electronics in Agriculture, 121, 313–319. https://doi.org/10.1016/j.compag.2015.12. 018. Mendes, T. O., Porto, B. L. S., Almeida, M. R., Fantini, C., & Sena, M. M. (2019). Discrimination between conventional and omega-3 fatty acids enriched eggs by FTRaman spectroscopy and chemometric tools. Food Chemistry, 273(1), 144–150. https://doi.org/10.1016/j.foodchem.2017.12.084. Monakhova, Y. B., Ruge, W., Kuballa, T., Ilse, M., Winkelmann, O., Diehl, B., ... Lachenmeier, D. W. (2015). Rapid approach to identify the presence of Arabica and Robusta species in coffee using1H NMR spectroscopy. Food Chemistry, 182, 178–184. https://doi.org/10.1016/j.foodchem.2015.02.132. Oliveira, K. S., De Souza Callegaro, L., Stephani, R., Almeida, M. R., & De Oliveira, L. F. C. (2016). Analysis of spreadable cheese by Raman spectroscopy and chemometric tools. Food Chemistry, 194, 441–446. https://doi.org/10.1016/j.foodchem.2015.08.039. Olsen, E. F., Rukke, E. O., Flåtten, A., & Isaksson, T. (2007). Quantitative determination of saturated-, monounsaturated- and polyunsaturated fatty acids in pork adipose tissue with non-destructive Raman spectroscopy. Meat Science, 76(4), 628–634. https://doi. org/10.1016/j.meatsci.2007.02.004. Organization, I. C. (2017). World coffee consumption. Ozaki, Y., Cho, R., Ikegaya, K., Muraishi, S., & Kawauchi, K. (1992). Potential of nearinfrared fourier transform raman spectroscopy in food analysis. Applied Spectroscopy, 46(10), 1503–1507. https://doi.org/10.1366/000370292789619368. Ribeiro, D. E., Borem, F. M., Cirillo, M. A., Prado, M. V. B., Ferraz, V. P., Helena, M. R. A., & Taveira, J. H. S. (2016). Interaction of genotype, environment and processing in the chemical composition expression and sensorial quality of Arabica coffee. African Journal of Agricultural Research, 11(27), 2412–2422. https://doi.org/10.5897/ AJAR2016.10832.
1.3a 0.6b 0.6b 0.1a
The data are the means of a triplicate ± standard deviation. G1 = Mundo Novo IAC 502/9, G2 = Yellow Bourbon IAC J9, G3 = Yellow Bourbon/Origin SSP, and G4 = Yellow Bourbon/ Origin CM.
3.2. Sensory analysis Table 1 presents the final scores of the sensory attributes of the four genotypes. The distinction between the coffees was shown from the classification model using the Raman spectra, and we found G1 (Novo Mundo) and G4 (Bourbon) to be statistically different from G2 and G3 (Bourbon). In this study, the sensory scores were very close, which may have resulted in difficulties with differentiating by sensory scores. Therefore, we believe that such differentiation can become even more evident when more distinct coffee profiles are compared, which justifies new studies comparing Raman and sensory analyses of coffee. 4. Conclusion With the aid of chemometry, Raman spectroscopy has enabled the distinction between coffee genotypes. The compounds that most contributed to this discrimination (PCA) were kahweol and fatty acids, especially when observing the bands at 1567 and 1479 cm−1, which were attributed to the kahweol, and 1442 and 1302 cm−1, which were attributed to fatty acids. Cafestol was also observed at 1502 cm−1 but with a lower contribution. The discriminant analysis by partial least squares (PLS-DA) performed for discriminating the samples (training and test set) of the Mundo Novo (G1) and Bourbon (G2, G3, and G4) genotypes was efficient, and it clearly separated both groups. Therefore, the model built with Raman spectroscopy data is suitable for discriminating Bourbon genotype samples. Sensorial analysis did not allow to discriminate coffee genotypes as well as discriminated by Raman. Conflict of interest The authors declare that they have no conflicts of interest. Acknowledgments The authors thank the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), A Fundação de Amparo à Pesquisa do Estado de Minas Gerais (Fapemig), a Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (Capes) e ao Instituto Nacional de Ciencia e Tecnologia do Café (INCTCafé). References Almeida, M. R., Oliveira, K. S., Stephani, R., & Oliveira, L. F. C. (2012). Application of FTRaman spectroscopy and chemometric analysis for determination of adulteration in milk powder. Analytical Letters, 45(17), 2589–2602. https://doi.org/10.1080/ 00032719.2012.698672. Alves, G. E., Borém, F. M., Isquierdo, E. P., Siqueira, V. C., Cirillo, M. A., & Pinto, A. C. P. (2017). Physiological and sensorial quality of Arabica coffee subjected to different temperatures and drying airflows. Acta Scientiarum Agronomy, 39(2), 225. https:// doi.org/10.4025/actasciagron.v39i2.31065. Arana, V. A., Medina, J., Alarcon, R., Moreno, E., Heintz, L., Schäfer, H., & Wist, J. (2015). Coffee’s country of origin determined by NMR: The Colombian case. Food Chemistry, 175, 500–506. https://doi.org/10.1016/j.foodchem.2014.11.160.
266
Food Chemistry 288 (2019) 262–267
L.P. Figueiredo, et al. Rubayiza, A. B., & Meurens, M. (2005). Chemical discrimination of arabica and robusta coffees by Fourier transform raman spectroscopy. Journal of Agricultural and Food Chemistry, 53(12), 4654–4659. https://doi.org/10.1021/jf0478657. Silveira, F. L., Silveira, L., Villaverde, A. B., Pacheco, M. T. T., & Pasqualucci, C. A. (2010). Use of dispersive raman spectroscopy in the determination of unsaturated fat in commercial edible oil- and fat-containing industrialized foods. Instrumentation Science and Technology, 38(1), 107–123. https://doi.org/10.1080/10739140903430339. Stephani, R., de Oliveira, K., de, S., Almeida, C. E. R., Perrone, Í. T., Carvalho, A. F., Oliveira, L. F. C., & Almeida, M. R. (2017). Raman spectroscopy as a tool to identify modification of whey protein concentrate (WPC) during shelf life. Food Packaging and
Shelf Life, 11, 1–9. https://doi.org/10.1016/j.fpsl.2016.10.001. Torres, J. K. F., Stephani, R., Tavares, G. M., Carvalho, A. F., Costa, R. G. B., Almeida, C. E. R., ... Perrone, Í. T. (2017). Technological aspects of lactose-hydrolyzed milk powder. Food Research International, 101, 45–53. https://doi.org/10.1016/j.foodres.2017.08. 043. Wermelinger, T., D’Ambrosio, L., Klopprogge, B., & Yeretzian, C. (2011). Quantification of the robusta fraction in a coffee blend via Raman spectroscopy: Proof of principle. Journal of Agricultural and Food Chemistry, 59(17), 9074–9079. https://doi.org/10. 1021/jf201918a.
267