3420 J. Agric. Food Chem.

2000, 48, 3420−3424

Discriminate Analysis of Roasted Coffee Varieties for Trigonelline,
Nicotinic Acid, and Caffeine Content

S. Casal,† M. B. P. P. Oliveira,*,† M. R. Alves,‡ and M. A. Ferreira†

CEQUP/Serviço de Bromatologia, Faculdade de Farmácia, Universidade do Porto. Rua Anı́bal Cunha 164,
4050-047 Porto, Portugal, and ESTG, Instituto Politécnico de Viana do Castelo, Av. do Atlântico,
Apartado 574, 4900 Viana do Castelo, Portugal

Arabica and robusta roasted coffees from several geographical origins, in a total of 29 samples,
were characterized for their contents in caffeine, trigonelline, and nicotinic acid by a recently
developed HPLC/diode-array detector method. All samples were subjected to the same roasting
procedure in order to eliminate the variations due to this process. Characterization was achieved
by applying multivariate and nonparametric analysis to the chromatographic results. The two coffee
varieties were clearly separated by their trigonelline and caffeine contents. Nicotinic acid could not
be used as a variety discriminate factor. There was no association with the geographical origin of
the samples.

Keywords: Roasted coffee; caffeine; trigonelline; nicotinic acid; multivariate and nonparametric
analysis

INTRODUCTION the presence of 16-O-methylcafestol in robusta coffee
(Speer et al., 1991). Chlorogenic acids can also be
More than 80 species of the genus Coffea L. (Rubi- successfully applied to this discrimination and, either
aceae) are known. The most important are Coffea alone or in conjunction with caffeine, have been used
arabica and Coffea canephora, which account, respec- to characterize commercial and noncommercial species
tively, for about 75% and 24% of the world production. of coffee (Clifford, 1989; Correia et al., 1995; Andrade
Coffee is an expensive raw material, especially arabica et al., 1997; Bicchi et al., 1995). Caffeine is probably the
coffee, and over the years many fraudsters have been single most analyzed chemical factor in coffee and has
tempted to falsify the product declaration due to the also been frequently used in the discrimination of green
increasing practice of selling coffees on the basis of their coffee varieties (Clifford, 1987; Macrae, 1989). Several
botanical and/or geographic origin (Prodolliet, 1996). approaches to coffee authenticity have also been re-
International coffee trade is conducted almost exclu- ported based on spectroscopic techniques (Kemsley et
sively with green coffee. In this state, arabica and al., 1995; Briandet et al., 1996; Downey et al., 1997),
robusta coffees are easily distinguished by their appear- aroma discrimination (Aishima, 1991), solid-phase mi-
ance (e.g., size, shape and color). Once roasted and/or croextraction-gas chromatography (Bicchi et al., 1997),
ground, the form in which the coffee is commercially and volatile profiles (Martı́n et al., 1996). Several of
available to the consumers, this visual criterion is these works associated chemical analysis with pattern
eliminated. Efficient methods are thus required for recognition techniques.
authentication of roasted coffee beans and for detecting The aim of this work represents a contribution to the
trading fraud. discrimination of the most representative coffee variet-
Efforts have been made to characterize the two coffee ies and eventually to access their geographical origins.
species using chemical data. Nevertheless, the chemical The compounds analyzed simultaneously were trigo-
composition depends not only on the species and variety nelline (N-methylnicotinic acid), nicotinic acid, and
in question, but also on the degree of roasting and, to a caffeine by a recently developed HPLC/diode-array
lesser extent, on other factors such as agricultural detector method (Casal et al., 1998). It was our objective
practices, degree of maturation, storage conditions, and to find out whether nicotinic acid could be used as a
geographical origin. These factors are not under the discriminate factor and to confirm the ability of caffeine
analyst’s control which makes it extremely difficult to and trigonelline contents to discriminate arabica and
quote average values for any type of coffee (Macrae, robusta roasted coffees.
1989). The main difference between the arabica and
robusta roasted coffee seems to be the composition of MATERIALS AND METHODS
the unsaponifiable matter (Folstar, 1989), specifically
Coffee Samples. Roasted beans samples from both Coffea
canephora var. robusta and Coffea arabica were studied. Coffea
* To whom correspondence should be addressed (e-mail robusta samples (n ) 20) were from several geographical
bromato@ff.up.pt; telephone 351-22-2078902; fax 351-22- origins: Ivory Coast (IC), India (IN), Honduras (H), Vietnam
2003977). (VN), Angola (A), Uganda (UG), and Cameroon (CM). Coffea
† CEQUP/Serviço de Bromatologia. arabica samples (n ) 9) were from Brazil (Br), Mexico (MX),
‡ ESTG, Instituto Politécnico de Viana do Castelo. Colombia (CO), Guatemala (GO), and Costa Rica (CR).

10.1021/jf990702b CCC: $19.00 © 2000 American Chemical Society
Published on Web 07/08/2000

Detection was accomplished with a diode-array detector. 1979) as implemented in the Statistica for Figure 1. and caffeine in roasted coffee samples.1M) and methanol performed at a constant 1980). displaying. 98 ( 1% for nicotinic acid. Principal Component Analysis was carried out after standardization of data to mean zero and unit variance and following standard procedures (Mardia et al. tions. A standard method was used in the roasting F ) SUM [R(xi) . 2-g portions of each powdered of the total possible pairs of observations and computing the coffee sample (25-mesh) were extracted with a total of 100 mL quantity of boiling water (Casal et al. No. a Summary results are presented in Figure 2. and an AS-950 autosampler. Kendall’s Tau was calculated by checking the number of Sample Preparation for Chemical Analysis. in the linear relationship between the concentration and the UV form of conventional box and whiskers plots. Univariate nonparametric analysis of trigonelline. A typical chromatogram obtained with a roasted chromatographic comparisons with authentic standards. 48. The solution was filtered into appropriate vials for use in the autosampler. nicotinic acid. 264 nm for nicotinic acid. . The compounds were identified by their retention times. from Brazil. isorb ODS2 (5 µm particle size. HPLC analysis was achieved as described T ) (Nc . Under the assay conditions described. (1998) with an analytical HPLC unit (Jasco) consisting of two PU-980 pumps. The solvent system used was a gradient of phosphate with appropriate tables to check for significance (Conover. Statistical Analysis.Discriminate Analysis of Roasted Coffee J. 25. and the percentage recovery was calculated after triplicate analyses. To study the recovery of the procedure..(n + 1)/2][R(yi) .(n + 1)/2]/n(n2 .1)/12 procedure in order to eliminate the variations due to this process. flow rate of 1. 1979).. Cluster analyses were performed after standardization of variables to unit variance and mean zero and using Euclidean distances as the distance measure and following the un- weighted pair group average method as the linking method (Mardia et al.1)/2 by Casal et al. These recovery values were 101 ( 1% for trigonelline. Agric. known quantities of the three standards were added to one sample of roasted coffee. standard method.. Spearman’s Rho was calculated by substitution of the actual values observed by the respective ranks. 8.0 (0. observations and n equals the total number of observations Sample moisture was determined in order to compare the (or pairs).0 mL min-1 at room temperature.0 × 0. Food Chem. Quantification was based on the external Figure 1. and 99 ( 1% for caffeine. buffer pH 4. For extrac- concordant (Nc) and discordant (Nd) pairs of observations out tion of the compounds in study.. Samples were hermetically sealed and stored at -20 where R(xi) and R(yi) represent the ranks for each pair of °C until used for chemical analysis. results in a dry weigh basis. and chromatograms RESULTS AND DISCUSSION were recorded at 265 nm. and sample from Brazil (Coffea arabica) is represented in their UV spectra. HPLC Analysis. a MD-910 diode-array where n refers to the total number of possible pair of observa- detector. Figure 2. 1998). A reversed-phase Spher.46 cm) column was Both measures of correlations (F and T) were then compared used.Nd)/n(n . and at 276 nm for caffeine. Vol. 2000 3421 absorvance was obtained at 268 nm for trigonelline. followed by computa- A local importer and roaster of coffee supplied all coffee tion of samples and was able to confirm their botanical and geo- graphical origin. Typical HPLC profile of a roasted arabica coffee Windows Statistical package.

3422 J. but levels in both varieties. 1989). Macrae.00 -0. Trigonelline and nicotinic acid . The levels found are in accordance with nic 0. first quartile. Figure 4. 1989). Principal component analysis of coffee samples on the three chemical parameters. the Table 1. All values are expressed Caffeine in the Roasted Coffee Samplesa in g‚kg-1 on a dry weight basis (dwb). caf ) caffeine.00 the literature (Clifford.00 tion between arabica and robusta coffees is mainly due to genetic differences found in the varieties of coffee atrig ) trigonelline. No. caf -0.00 caffeine levels. Agric. Nicotinic Acid.01 1.00 1. for each variable and for each coffee variety. third quartile. overall correlations pooled within groups correlations It can be seen that there is a clear distinction between trig nic caf trig nic caf arabica and robusta samples in what concerns the trig 1.29 -0.00 -0. Figure 3.42b 1. Cluster analysis of roasted coffee samples. Food Chem. Vol. 2000 Casal et al.. Overall and Pooled within Groups Simple minimum values. median. 48. 8.21 1. In the same figure we can observe that the differences the nicotinic acid levels vary widely and assume similar in trigonelline between varieties are also apparent. as it is stable on roasting (Macrae. 1987. Linear Correlations of Trigonelline. b Significant correlations at p ) 0. and and maximum values observed. nic ) nicotinic acid. The varia.05 beans.11 1.85b -0.

391752a . Analysis of Responses from Semiconductor Gas Sensor Array. Vanni. Vol.51137a . M. The results show that component. - nicotinic acid 0.. B. only negative correlation between As our objective was to find correlation with the botani. stances in arabica and robusta coffees. Aroma Discrimination by Pattern Recognition tween nicotinic acid and caffeine is observed. - nicotinic acid -0. i. -0. Spearman Rho and Kendall Tau Measure Correlations: (a) Overall Correlations.. M. overcome the high variations observed within each Ferreira. R. is not correlated with the first two param. varieties separately. Figure 4. nated this variation factor by subjecting all samples to Therefore it seems that the negative correlation the same exactly roasting procedure (14 min from 160 between trigonelline and caffeine is only expressing to 220 °C). must be seen with caution since all data is analyzed simultaneously. without taking notice of the existence of clusters. 1997. -0. It is important to note that. but also a negative correlation between caf. with the linear correlation results. can be observed within each coffee variety and also tion. Consequently.. in Figure 4. -0. The second principal tion of pure roasted coffees. J. Bicchi. by Solid-Phase Microextraction . which explains roughly one-third of the total nicotinic acid cannot be used in the discrimination of information. as displayed in analysis in the coffee industry. general differences between the levels of these sub- A cluster analysis was carried out. As the nature of the green already observed in Figure 2. as origin of the roasted coffee. C. formed at lower levels. which Andrade. recommended to confirm this possibility. M. significant nega. 8. Analyzing coffee partially converted into nicotinic acid (Macrae. caffeine can be used for identification of the geographical eters under study. or strong nega- tive correlation. being represented in the second tially thought (Macrae. Neither trigonelline nor component. M.192982 -0. groups. robusta and Tau. 3. 2000 3423 Table 2. with all coffee samples together and S. Leitão. 1989). 39. the former being lower and the latter higher trigonelline and caffeine can be used in the discrimina- in robusta than in arabica samples. P. A. is caused sponding dendogram show in Figure 3. G. and trigonelline and the absence of correlation between the methodology proposed is appropriate for routine these two variables and nicotinic acid.276739 . When coffee varieties are analyzed separately. the LITERATURE CITED correlation between trigonelline and caffeine is not significant.045752 -0. and (c) Correlations within robusta Variety Spearman Rho Kendall Tau trigonelline nicotinic acid caffeine trigonelline nicotinic acid caffeine all samples trigonelline .686275a . centration in the roasted product. 48. Casal is indebted to Subprograma Ciência e Tec- also analyzing the pooled within groups correlation. C. in accordance cal and eventually geographical origin we have elimi.. T. However.303441 -0.03625 . tive correlation between trigonelline and caffeine are Characterization of Roasted Coffee and Coffee Beverages observed. -0.. No. Agric. P.516291a -0. It becomes clear that the first principal component..05 levels are known to be related with the intensity of feine and nicotinic acid becomes apparent even when roasting being the former progressively degraded and all samples are considered together. which. nicotinic acid and caffeine are significant. while trigonelline is lower. Food Chem. Oliveira. could not be attributed to the geographical there is no correlation between these two components. 0. Using more powerful.158372 - caffeine -0. results are obtained as represented in Table 2 were Coffea arabica. 511-514. Seabra. found. B. M. Panero. 1989). although within groups correlation are analyzed. between the levels of trigonelline and From the results obtained we can concluded that caffeine. Agric.. An important feature of these results is the negative As cluster analysis shows the existence of clusters correlation between nicotinic acid and caffeine. a correlation analy. a principal component analysis was done. there is a great variability beans’ processing employed may affect the final con- within each variety cluster in what concerns the nico. 0.4-Dimethoxycinnamic acid levels as a tool coffee group. represents an opposition. origin or to any other specific reason. Food Chem. 1991. Food Chem. one can see that apparent. (b) Correlations within arabica Variety.018303 - caffeine -0. represents differences in composition of these two coffee varieties contradicting what was ini- nicotinic acid. further studies are tinic acid levels.. which explains two-thirds of the total CONCLUSIONS information. 61 (4). Pellegrino.. 752-756. as shown when all data is taken together. O. and a significant negative correlation be. A.Discriminate Analysis of Roasted Coffee J. nologia do 2° Quadro Comunitário de Apoio (BD/9580/ taking arabica and robusta samples as two distinct 96). when the pooled but other clusters.232098a - a Significant correlations at p ) 0. Two distinct by the fact that caffeine is higher in robusta samples clusters representing each coffee variety are obvious.071895 - caffeine 0. which without explaining the reasons assisting their forma.740277a -2.Gas Chromatography and .207642a - arabica trigonelline .e. R.108359 . - nicotinic acid -0. ACKNOWLEDGMENT sis was carried out.001032 -0. The conjunction of these chemical parameters may The strong negative correlation between caffeine and be helpful in the identification of coffee varieties.542484a - robusta trigonelline . Aishima. such as the Spearman’s Rho and Kendall’s for differentiation of Coffea canephora var. with the corre. The results are summarized in Table 1. nonparametric statistics. Considering overall correlation.

Elsevier Applied Science Publish... 45. S. Kemsley. 1996. 191-197. R. M. received May 11. J. Infrared Spectroscopy and Chemometrics.. Ashurst. N.. Clarke. Lipids. 1989. Food Macrae. W. Near- and Mid-Infrared Spectroscopies in Food Authentication: JF990702B . A. Montag. Oliveira. 1. R.. Wilson. Macrae. Caffeoyl- tyrosine and Angola II as characteristic markers for Angolan Received for review June 28. M. J. Agric.. biochemistry..3424 J. indicator for coffee. pp Conover.. E. K. 1. 53.. HPLC/Diode-array detector method for simultaneous de- termination of trigonelline. H. In Coffee: Chemistry. 1989. A. 16-O-methylcafestol. R. and coffee products.. Food Chem. C.. nicotinic acid and caffeine in Mardia. 2000. M. ASIC: Paris. Agric. Kent and Bibby.. P. A quality R. Accepted May 13. G. Prodolliet... Bicchi. R. Authenticity of coffee. Component Analysis. Dennis. J. pp 304-338. Clifford. Croom Helm: Beckenham. Coffee Varietal Identification. A. Eds.. E. Blackie Academic & 374. N. production of beans and beverage. 1987. Food Chem. Correia. 1980. A.. B. Tewis. Clifford. between Coffea arabica and Coffea canephora variant ro- Briandet. J.. A. Elsevier Applied Science Publishers: London. J.. 1995. Vol. Elsevier Applied Science Casal. tional sur le Café-San Francisco. Kemsley. R. P. Development of an Publishers: London. J. 1995. Pablos. Briandet. and Chim. G. G. Academic coffee. In 14th Colloque Scientifique Interna- ers: London. K.. 1989.. Speer. C.. pp 115-137. Wilson.. F. Discrimination 1555. Eds. Food Chem. 170-174. Macrae. M. K. G. A. pp 203-222. Vanni. M. 48. M. González. Professional: London. Clarke. 320. A. Principal Component Analysis. Chemical and Physical aspects of green coffee recognition to the discrimination of roasted coffees. R. M. Chlorogenic acids. R. H. Clifford. 4357-4361. Ruault. Multivariate Analyses. 1.. pp 305. 43. Agric. E.. 8. M. Clarke. Application of pattern Clifford. Eds. S. Martı́n. R. Pellegrino. Binello.. Wilson... Macrae. 4680-4686. In Coffee: Chemistry. R. & Sons: New York. J. Eds. N. Downey. 3187. John Wiley 237-244. J. Chrom. 1996.. Eds. P. 1998. 21 (20). Anal. Practical Nonparametric Statistics. Revised manuscript robusta coffees. Vol. 3195. Food Chem. R. Arabica and Robusta in Instant Coffee by Fourier Transform 54. In Coffee: botany. Acta 1996. R. H. N. Press: London. Chem. Wilson. Liq. N. Leitão. 1979.. 309-313. Folstar. C. K. Technol. J. Discrimination of busta beans using infrared spectroscopy. 1995. M. J.. 45 (12). pp 153-202.. K. Food Chem. Vol.. R. 44. M. Kemsley. Relat. M. 1999. No. Ferreira. Chlorogenic Acid Fraction by HPLC-UV and Principal Vol. 2000.. 321-326. Agric. 1549. 1997. U.. Food Chem. Agric. Nitrogenous components. 1991.. In Coffee: Chemistry. 1997.. C. In Food Authentication. J.K. J. Characterization of Green and Roasted Coffees through the R. 2000 Casal et al.