Professional Documents
Culture Documents
Aroma Coffee Varieties
Aroma Coffee Varieties
www.elsevier.com/locate/foodres
Abstract
Principal component analysis and hierarchical clustering methods were used to dierentiate coees from 8 dierent origins pro-
duced by the same manufacturer. The analysis was applied to 23 peaks of the aroma fraction. The peaks were tentatively identi®ed
by GC±MS. The aroma fraction can successfully be employed to characterise roasted coees of dierent origins. The eect of
roasting is emphasised when the same method is applied to green coees of the same origins and the same manufacturer. This paper
reports the possibility of characterising roasted coee by PCA of some of the aroma compounds through the use of a purge and
trap sampling device and GC±FID analysis. The aroma fraction was obtained after milling the beans over controlled conditions
and using the purge and trap sampling technique as sample introducing system for GC analysis. Thermal desorption took place
during 5 min and the vapours were cryofocused at the head of the capillary column for 5 min. # 2000 Canadian Institute of Food
Science and Technology. Published by Elsevier Science Ltd. All rights reserved.
Keywords: Green and roasted coees; Aroma fraction; Purge and trap sampling; Principal component analysis (PCA)
and may have dierent chemical and organolpetic quality, within the variety, (as determined by a cup test) is
properties. The producers must select among the dier- not aected by the dierences of the roasting procedures.
ent raw materials to produce quality and homogeneous Samples were delivered on hermetically sealed poly-
blends. There is a certain interest in the commercial propylene/aluminium/polypropylene packages hermeti-
quali®cation of coee. The % of Arabica and Robusta has cally sealed under vacuum. Immediately after sampling
been considered as one way to produce blends of dierent the bags were again sealed under the same conditions.
qualities (Frega, Bocci & Lercker, 1994). Aroma being
one of the most important criteria for the evaluation of 2.2. Sample preparation
quality, the aroma fraction, though produced mainly
during the roasting process (Baltes & Bochman, 1987; Nine grams of each roasted coee was graded in an
Baltes, Kunnert-Kirchno & Reese, 1989; Parliament, electrical ``Moulinex'' coee miller for 30 s immediately
1989; Tressl, 1989), might be used to dierentiate the cof- before the analysis. Two sieves of 500 and 250 mm then
fees from dierent geographic origins (Bicchi, Binello, sifted the powder. Samples of 1.0 g were taken from the
Pellegrine & Vanni, 1993; Hernandez &Rutledge, 1994; thinner fraction and use for the analysis.
Mosca, 1994; Tipler, Spragg, Aries & Lidiard, 1991). The The samples were placed in a sampling glass tube,
comparison made with the green coees from the same between two pieces of silanized glass wool and the tube
origins has allowed the con®rmation of the roasting used in the P&T device.
process in¯uence for the aroma development. This paper
describes the use of P&T/GC method for determination of 2.3. Gas chromatography
volatile pattern of green and roasted coees from Arabic
and Robusta varieties. This method gives good accuracy The volatile components from the coee samples were
for the aroma fraction determination, however, if a large separated in a Carlo Erba 6000 Vega gas chromato-
desorption time is used bandbroadening of the peaks graph equipped with a ¯ame ionisation detector and a
occurs (Shimoda & Shibamoto, 1990). Cryofocusing of SuOx-0.1 capillary column (34 m0.35 mm). A Carlo
the sample, in liquid nitrogen at the head of the column, Erba TDAS 5000 purge and trap injector was used with
during injection can greatly improve the chromatography direct connection to the gas chromatograph. All sam-
(Frank & Frank, 1990; Jurski, Stansky & Ubik, 1991; ples were pre-heated at 200 C for 150 s, followed by 600
Mosca, 1994). Tentative identi®cation of the peaks chosen s of desorption (at the same temperature). The interface
as variables for PC (principal component analysis) was between the injector and the chromatograph was heated
attempted by GCMS. at 210 C. Helium was used as carrier gas at 200 KPa.
200 C of heating for 150 s for headspace development
did not show any changes in colour either for the roasted
2. Experimental or the green coees. During desorption (600 s) the vola-
tiles were cryofocused in a 1 m deactivated pre-column
2.1. Materials 10 cm of which emerged in liquid nitrogen. The chroma-
tograph oven was programmed as follows: 50 C for 5
Samples were a gift from Nestle Portugal SA. The min, then 50 to 200 C at a linear rate of 5 C/min rate and
varieties and suppliers countries of the raw beans are 20 min at 200 C. The detector temperature was 250 C.
the usually used by Nestle on the Portuguese market.
Green and roasted Arabica and Robusta coee beans 2.4. Gas chromatography±mass spectrometry
were used in our analyses. Arabica coees were produced
in Brazil (B) (Santos 2/3FFRF, S. Paulo and Parana), Tentative identi®cation was attempted by GC±MS
Costa Rica (CR, Costa Rica HB, Paci®c Plain and Central using an ITD Finnigan Matt associated with a personal
Plateau) and Honduras (H, Honduras HG, St. Barbara, El computer. Ion energy of 70 eV and scan time of 1 s was
Paraiso and Comayagua). Robusta coees came from used for the electron ionisation. GC and sampling con-
Ivory Coast (IC, Costa do Mar®m, South from 3/3 of all ditions were the same as used for GC alone. The open
the 49 city halls), Angola (A, Angola 2BB, Cabinda and split interface was kept at 200 C.
Ambriz), Uganda (U, Uganda Std Grau 2, a 300 km
radius from Lake Victoria), India (I, IndiaCherry AB, 2.5. Data processing
Karnataka and kevala States) and Cameron (C,
CamaroÄes Grau 1, all provinces except the North Statistical analysis and group separation were accom-
region, specially the Mungo region). All coee samples plished with the SYSTAT for WindowsTM version 5.0
were from the same manufacturer. Roasting conditions from SYSTAT Inc. (Ill. USA) using an IBM 486 personal
of Arabica and Robusta were dierent. Arabica were computer (PC).
submitted to a light roast and Robusta medium to high The peaks in the GC pattern to be submitted to PCA
roast. According to the manufacturer organoleptic were the 23 major ones and covered the whole aroma
A.M. Costa Freitas, A.I. Mosca / Food Research International 32 (1999) 565±573 567
chromatogram. The peak areas were processed on a (Bicchi et al 1995; Bicchi, Binello, Legovich, Pellegrio &
Shimadzu CR3A computing integrator and manually Vanni, 1993; Gomes da Silva & Neves, 1997; Hernandez &
introduced on the PC for statistical analysis. Principal Rutledge, 1994; Martin, Pablos & Gonzalez, 1996; Vas-
component analysis (PCA) was applied to the mean concelos & Neves, 1989). The ®rst step in PCA is the choice
areas of three samples three replicates of the similarity measure. The similarity measure chosed
was the covariance matrix. The values obtained are a
measure of the association degree between variables. Each
3. Results and discussion matrix was prepared with the mean values (three repli-
cates) of the areas of each 23 peaks used as variables. The
From the chromatograms obtained with the 16 coee values were normalised using the z-transform.
varieties (5 Robusta and 3 Arabica,) the 23 major peaks The next step on PCA is the eigenvectors extraction
were chosen to use as variables for PCA analysis. The from the matrix to obtain the so-called principal com-
total area of these 23 peaks was correspondent to ponents (uncorrelated variables). Principal components
almost 80% of the total chromatographic areas when (PCs) are linear combinations of all the original vari-
roasted coees chromatograms were considered (Fig. 1). ables and expand the maximum variance in the vari-
Peaks were tentatively identi®ed after GC±MS analysis. ables. The PC loadings are coecients of the correlation
Tentative identi®cations are shown in Table 1. PCA is between the variables vectors and the PCs. The PC
now a well established tool for interpretation of chemical scores are the coordinates of the sample points on PCs.
data and several examples of applications and of the basic When these scores are plotted against the PCs, similar
features of the method are now available on the literature samples tend to group together.
Table 1
Tentative identi®cation by spectral library (Wiley, 1989) of the peaks responsible for dierentiation obtained when PCA analysis is applied
Fig. 2 and 3 show the plot of the scores obtained for body (Correia, 1990). This statement is in good agree-
the two ®rst PCs of the roasted Arabica and Robusta ment to the fact that Arabica coees are better grouped
coees used. The matrix used was 15 cases (each case when, as in this study, only the aroma results is used to
being the mean value of three replicates) 23 variables achieve group separation. When PCA is applied as a
for the Robusta variety and 9 (each case being the mean reduction feature to the results of the Arabica and
value of three replicates) cases 23 variables for the Robusta coees together, eight variables were found
Arabica varieties. that are highly uncorrelated and account for 53.4% of the
The aroma fraction of the Arabica varieties seem to total variance. The matrix used had 24 cases (each case
be giving better results in terms on grouping according being the mean value of three replicates)23 variables.
to geographical origin. The two principal components in The plot of the two principal components scores, (Fig. 6)
this analysis account for the explanation of 71.7% of shows a clear cut between Arabica and Robusta varieties.
the total variance. Brazilian coees are deeply aected Peaks nos. 11 and 16 clearly aect the Robusta varieties.
by peaks nos. 11 and 16, whilst peaks nos. 15, 13 and 9 Ivory Coast coees are similar to the Robusta ones
seem to allow the dierentiation of the Costa Rica according to this results whilst the Costa Rica varieties
varieties. Honduras varieties do not seem to be deeply are far apart from the others and deeply characterised
aected by this set of peaks. All other variables do not by peaks nos. 8, 19, 12 and 2. Brazilian coees are
have in¯uence on this dierentiation (Fig. 4). completely at the opposite side and we can consider
Grouping in Robusta variety is also in accordance them to be aected by peak nos. 20, 21 and 18 (Fig. 7).
with geographical origin although not so clearly. The It can also be seen in the ®gure that the distribution of
two principal components used for the representation the samples follows a pattern. The separation between
account for 49.8% of the total variance. Peaks nos. 2, the classes is related to the geographical origin, and the
17 and 7 are clearly associated with the coees from Arabica varieties are better classi®ed than the Robusta
Ivory Coast whilst peaks nos. 13 and 11 are the major ones. This is, again, consistent with the fact than aroma,
responsible for the group separation of the Indian vari- in a coee blend, is mainly due to the % of Arabica
eties. Uganda Angola and Cameron are more associated which is re¯ected by the smaller in¯uence volatiles have
to each other and are mainly dependent on peaks nos. 6 on identifying Robusta, varieties.
and 18 (Fig. 5). The in¯uence of roasting is clearly demonstrated when
Arabica is usually the coee used in blends for the green coees, of the same origins, are analysed together
aroma eect and Robusta the one used for taste and with the roasted ones. Green coees are completely
Fig. 2. Plot of the PCA scores obtained for the two ®rst principal components that characterize the Arabica varieties. B-Brasil; H-Honduras; Cr-
Costa Rica.
A.M. Costa Freitas, A.I. Mosca / Food Research International 32 (1999) 565±573 569
Fig. 3. Plot of the PCA scores obtained for the two ®rst principal components the characterise the Robusta varieties. I-India; A-Angola; Ic-Ivory
Coast; Cm-Camaron; U-Uganda.
Fig. 4. Plots of factor loadings for the ®rst and second PCA of a test of Arabica coees. For numbers identi®cations see Table 1.
570 A.M. Costa Freitas, A.I. Mosca / Food Research International 32 (1999) 565±573
Fig. 5. Plots of factor loadings for the ®rst and second PCA of a test of Robusta coees. For numbers identi®cations see Table 1.
Fig. 6. Plot of the PCA scores obtained for the two ®rst PCs that characterize roasted Arabica and Robusta varieties. Letters as in Figs 1 and 2.
A.M. Costa Freitas, A.I. Mosca / Food Research International 32 (1999) 565±573 571
Fig. 7. Plots of factor loadings for the ®rst and second PCA of a test set of roasted Arabica and Robusta coees. For numbers identi®cations see
Table 1.
Fig. 8. Plot of the PCA scores obtained for the two ®rst principal components that characterise a set of green and roasted Arabica and Robusta
varieties. Letters as in Figs 1 and 2 except for green coees that are not dierentiated.
572 A.M. Costa Freitas, A.I. Mosca / Food Research International 32 (1999) 565±573
separated from the roasted ones and grouping wasn't and roasted coees were very well separated, as expected.
found according to geographic or variety origin. Green Clustering of roasted coees showed a grouping by variety
coees are all concentrated in the same point of the plot (Arabica and Robusta) and within the variety by geo-
(Fig. 8). By the analysis of Fig. 9 green coees are graphical origin (Fig. 10). Brazilian coee is an exception
characterized by peaks 2, 3, 5, 6, 7 and 17 mostly whilst (Fig. 10), which is in good agreement to the known fact
roasted coees are mainly characterized by the other that Brazilian coee has an o-¯avour very unique and
peaks (Fig. 9). This is obvious from the GC pattern since characteristic (Spadone et al., 1990). Varieties from clo-
green coees have some of the 23 peaks used for PCA ser countries were also grouped together. Table 1 has
missing or in amounts not detectable (green coees have the tentative identi®cations of the peaks that account
only peaks through 7 and peak 17) due to the fact that
aroma is mostly developed during roasting and is deeply
in¯uenced by the technology and temperature used in
processing (Takeoka et al., 1988). In this case roasting has
not been standardised and Arabica and Robusta were
roasted with two dierent roasting processes according to
the manufacturer and roast was controlled by a cup test.
When the same roasting process is used a clear cut
between Arabica and Robusta varieties is also seen but
the separation among Arabica varieties is not clear
(Wada, Seiichi, Hitoshi, Mitsuya & Yutaka, 1987).
Roasting in¯uence is, however, well demonstrated since
neither geographic origin nor variety have any in¯uence
on green coees dierentiation. The matrix used had 32
(243 for the roasted coees cases and 83 for the
green ones)23 variables and the two ®rst PCs accoun-
ted for 57.4% of the total variance. To verify similarity
between the samples and to single out some classes
Hierarchical Cluster Analysis (HCA) was applied to the
reduced set of variables obtained with PCA analysis.
The samples were clustered using the single linkage
method and Euclidean distance was taken as a measure
of the proximity between samples. The similarity values Fig. 10. Cluster analysis of roasted and green coees. Letters as in
for the linkage were represented by a dendogram. Green Figs. 1 and 2.
Fig. 9. Plots of factor loadings for the ®rst and second PCA of a test set of green and roasted Arabica and Robusta varieties. For numbers identi-
®cations see Table 1.
A.M. Costa Freitas, A.I. Mosca / Food Research International 32 (1999) 565±573 573
for dierentiation of the samples studied and analysed. analysis of gas chromatograms to dierentiate cocoa masses by geo-
Tentative identi®cation was attempted by GC±MS. graphical origin and roasting conditions. Analyst, 119, 1171±1176.
Jurski, T., Stansky, K., & Ubik, K. (1991). Trapping system for trace
organic volatiles. J. Chromatogr., 586, 315±322.
Kallio, H., Leino, M., & Solarinne, L. (1988). Analysis of the headspace
Acknowledgements of foodstus near room temperature. In Sandra Pat, Proceedings of the
9 International Symposium on Capillary Chromatography (pp. 191±
We thank Nestle Portugal SA for the gift of the samples 200). Heidelberg: Huethig Verlag Publishers.
and Junta Nacional de InvestigacËaÄo Cientõ®ca e TecnoloÂ- Martin, M. J., Pablos, F., & Gonzalez, A. G. (1996). Application of
pattern recognition to the discrimination of roasted coees. Analy-
gica (J.N.I.C.T.) for ®nancial support, project PBIC/C/ tica Chimica acta, 320, 191±197.
CEN/1041/92 A. I. Mosca thanks J.N.I.C.T. for a grant. Miller, K. G., Poole, C. F., & Pawlowski, T. M. P. (1996). Classi®ca-
tion of the botanical origin of cinnamon by solid-phase micro-
extraction and gas chromatography. Chromatographia, 42, 639±646.
References Mosca, A. I. (1994). ComposicËaÄo quõÂmica do aroma do cafe e sua
correlaccËaÄo com origem e controlo de qualidade MsC. Thesis, Uni-
Baltes, W., & Bochman, G. (1987). Model reactions on roast aroma versidade de EÂvora- P.
formation. 1. Reaction of serine and threonine under the conditions Nishimura, O., & Mihara, S. (1990). Investigation of 2-hydroxy-2-
of coee roasting and identi®cation of new coee aroma com- cyclopenten-1-ones in roasted coees. Journal of Agricultural and
pounds. Journal of Agricultural and Food Chemistry, 35, 340±346. Food Chemistry, 38, 1038±1041.
Baltes, W., Kunnert-Kirchho, J., & Reese, G. (1989). Model reac- Parliament, T. H. (1989). Thermal generation of aromas, an overview.
tions on generation of thermal aroma compounds. In T. In T. H. Parliment, R. J. McGorrin, & Ho Chi-Tang, Thermal
H. McGorrin, R. J. McGorrin, & Ho Chi-Tang, Thermal generation Generation of aromas ACS Symposium series. (pp. 2±11). Washing-
of aromas (pp. 143±155). Washington DC: ACS Symposium series. ton DC: American Chemical Society.
American Chemical Society. Pawliszyn, J., & Arthur, C. L. (1990). Solid phase micro extraction
Bicchi, C. P., Binello, A. E., Pellegrino, G. M., & Vanni, A. C. (1995). with thermal desorption using fused silica optical ®bers. Analytical
Characterisation of green and roasted coees through the chloro- Chemistry, 62, 2145±2148.
genic acid fraction by HPLC-UV and principal component. J. Shimoda, M., & Shibamoto, T. (1990). Isolation and identi®cation of
Agric. Food Chem., 43, 1549±1555. headspace volatiles from brewed coee with on-column GCMS
Bicchi, C. P., Binello, A. E., Legovich, M. M., Pellegrino, G. M., & method. Journal of Agricultural and Food Chemistry, 38, 802±804.
Vanni, A. C. (1993). Characterisation of roasted coee by S-HSGC Spadone, J. P., Takeoka, G., & Liardon, R. (1990). Analytical Inves-
and HPLC-UV and principal component analysis. Journal of Agri- tigations of rio o-¯avor in green coee. Journal of Agricultural and
cultural and Food Chemistry, 41, 2324±2328. Food Chemistry, 38, 226±233.
Bondorovich, H. A., Friedel, P., Krampl, J. A., Renner, J. A., She- Stoll, M., Winter, M., Gautschi, F., Flament, I., & Willhalm, B.
pard, F. W., & Gianturco, M. A. (1967). Volatile constituents of (1967). Sur l'arome de cafeÂ. Helvetica Chimica Acta., 50, 628±692.
coee. Pyrazines and other compounds. Journal of Agricultural and Takeoka, G., Flath, R., GuÈntert, M., & Jennings, W. (1988). Nectar-
Food Chemistry, 15, 1093±1098. ine volatiles: vacuum steam distillation versus headspace sampling.
Cadwallader, K. R., & Xu, Y. (1994). Analysis of volatile components Journal of Agricultural and Food Chemistry, 36, 553±560.
in fresh grapefruit juice by purge and trap/gas chromatography. Tipler, A., Spragg, R. A., Aries, R. E., & Lidiard, D. P. (1991). The
Journal of Agricultural and Food Chemistry, 42, 782±784. Characterization of Coee using headspace extraction, capillary gas
Coleman, W. M. (1992). Automated purge-and-trap-gas chromato- chromatography, selective detection systems, pattern recognition
graphy analysis of headspace volatiles from natural products. J. software and principal component analysis. In Sandra Pat, Pro-
Chromatograhic Sci., 30, 159±163. ceedings of the 13th International Symposium on Capillary Chroma-
Correia, A. M. N. G. (1990) A in¯ueÃncia da torra na evolucËaÄo dos tography (pp. 337±347). Heidelberg: Huethig Verlag Publishers.
aÂcidos clorogeÂnicos do cafeÂ. Ph. D. Thesis, Universidade TeÂcnica de Tressl, R. (1989). Formation of ¯avor components in roasted coee.
Lisboa, Instituto Superior de Agronomia.P. In T. H. Parliment, R. J. McGorrin, & Ho Chi-Tang, Thermal gen-
DeMaria, C. A. B., Tru, L. C., Moreira, R. F. A., & Werneck, C. C. eration of aromas, ACS Symposium series. (pp. 285±301). Washing-
(1994). Composition of green coee fractions and their contribution ton DC: American Chemical Society.
to the volatile pro®le formed during roasting. Food Chemistry, 50, Vasconcelos, A. H. P., & Neves, H. J. (1989). Characterization of ele-
141±145. mentary wines of vitis linifera varieties by pattern recognition of free
Frank, W., & Frank, H. (1990). The micro-trap: an alternative to amino acid pro®les. J. Agric. Food Chem., 37, 931±937.
cryofocusing in capillary gas chromatography. Chromatographia, Vitzthum, O. G., & Werkho, P. (1975). Cycloalkapyrazines in Coee
29, 571±574. aroma. J. Agric. Food Chem., 23, 510±516.
Frega, N., Bocci, F., & Lercker, G. (1994). High resolution gas chro- Wada, K., Seiichi, O., Hitoshi, S., Mitsuya, S., & Yutaka, O. (1987).
matographic method for determination of robusta coee in com- Classi®cation of Various Trade Varieties of Coee by Coupling of
mercial blends. J. High Res. Chromatogr., 17, 303±307. Sensory Data and Multivariate Analyses. Agric. Biol. Chem., 51,
Goldman, I. M., Seibl, J., Flament, I., Gautschi, F., Winter, M., 1745±1752.
Willhalm, B., & Stoll, M. (1967). Sur l'aroÃme de cafeÂ, II Pyrazines et Wiley/NBS Library, registry of mass spectral data (1989).
pyridines, Hel. Chemica Acta, 50, 694±705. Yang, X., & Peppard, T. (1994). Solid-phase microextraction for ¯a-
Gomes da Silva, M. D. R., & Neves, H. J. C. (1997). Dierentiation of vour analysis. J. Agric. Food Chem., 42, 1925±1930.
strawberry varieties through purge and trap HRGC-MS, HRGC- Yu, X., Chi-Tang, Ho, & Rosen, J. D. (1993). Determination of coee
FTIR and principal component analysis. J. High Resol. Chroma- volatiles with ITS 40 mass spectrometer. In G. Charalambous, Food
togr., 20, 275±283. and ¯avors, ingredients and composition (pp. 23±30). Amsterdam:
Hernandez, C. V., & Rutledge, D. N. (1994). Multivariate statistical Elsevier Science Publishers BV.