Chapter 30
Differentiating the Flavor Potential of Cocoa Beans by Geographic Origin 1
B. D. Glazier and P. S. Dimick
Downloaded by CORNELL UNIV on October 12, 2016 | http://pubs.acs.org Publication Date: January 15, 2000 | doi: 10.1021/bk-2000-0754.ch030
Department of Food Science, The Pennsylvania State University, University Park, PA 16802
Unique flavors are derived from cocoa beans that originate from different geographic regions. Continually monitoring these flavor differences using a trained sensory panel can be effective; however, this process is time consuming, costly, and subjective in nature. To simplify the screening process and, more objectively, gauge the flavor potential of cocoa beans, this study sought to develop a multivariate statistical model to enable predictions of cocoa bean flavor from analytical measurements of the volatiles emitted during nib roasting. Twelve cocoa bean samples were examined using traditional bean quality tests to assess fermentation, headspace/gas chromatography (GC) to provide analytical characterization, and subsequently, made into chocolate and evaluated by a trained sensory panel. Both the analytical and sensory data matrices allowed the cocoa beans and their corresponding chocolates to be distinguished and similarly grouped using principal component analysis (PCA). When the data matrices were subjected to partial least squares regression analysis (PLS) to test for correlation, a limited predictive model was generated.
Despite the global appeal of chocolate and its long history of consumption, scientists have not fully explained the mechanism by which chocolate flavor arises. Many of the known reactions, which contribute to the chemistry of chocolate flavor, have been described in review articles (1-3). Hoskin and Dimick (3) described the formation of chocolate flavor as dependent upon genotype, geographic origin, proper fermentation, drying, roasting, and processing of the cocoa beans. Chocolate flavor consists of both volatile and non-volatile chemical compounds that contribute to its complexity. Much of what scientists know about chocolate flavor 'Current affiliation: M & M / M a r s , 295 Brown Street, Elizabethtown, P A 17022
© 2000 American Chemical Society
Parliment et al.; Caffeinated Beverages ACS Symposium Series; American Chemical Society: Washington, DC, 2000.
293
294
Downloaded by CORNELL UNIV on October 12, 2016 | http://pubs.acs.org Publication Date: January 15, 2000 | doi: 10.1021/bk-2000-0754.ch030
is the result of research on the volatile aroma fraction. Maarse and Visscher (4) reported that as many as 503 volatile compounds occur in cocoa. Flament (5) provided identification for 462 volatile compounds present during chocolate production and new compounds are continually being discovered and identified (6). Differences between the volatiles from one sample to another have been successfully used to identify cocoa beans by origin, understand flavor development, and interpret process changes (7-10). Among the hundreds of volatile compounds present in foods, often a small number, in specific ratios, determine the characteristic odor. In a complex flavor like chocolate, finding this cause-and-effect relationship has been elusive. Even today, the chemist is unable to accurately duplicate this desirable flavor. With numerous countries supplying cocoa beans, combined with the variation that can occur during fermentation, the task of consistently formulating chocolate according to flavor specifications can be difficult. Besides compositional and physical differences, cocoa beans originating from different supply countries possess their own unique flavor. Traditionally, chocolate manufacturers have considered West African cocoa beans to be the industry flavor standard. However, other supply countries have undergone considerable efforts to improve the flavor quality of their cocoa beans. Today, a common practice among chocolate manufacturers is to select cocoa beans from several countries and blend them. Economic considerations and changing supply patterns have complicated the selection of cocoa beans and reinforced the importance of performing quality tests on incoming raw materials. Many of the quality tests for cocoa beans assess the degree of fermentation and by themselves provide limited insight into the flavor potential of the beans. Researchers can increase the value of these tests by correlating their results with sensory scores. Sensory tests, for raw materials and finished products, are necessary and effective for assessing their flavor quality. However, cocoa beans are difficult to assess for flavor until they have been roasted or semi-processed making them impractical, costly, and time consuming to evaluate. Because the flavor potential of cocoa beans is difficult to assess and often subjective in nature, the present study served to expand on existing bean quality tests and identify meaningful statistical relationships between analytical and sensory data that could be used for predictive modeling.
Materials and Methods Cocoa Beans A total of 12 cocoa bean samples were obtained from a commercial supplier. The samples and their descriptive names, when available, were from Brazil (Bahia), Ecuador (Arriba), Ghana, Ivory Coast, Indonesia (Sulawesi), Malaysia, and two from Dominican Republic (Sanchez and Hispaniola). Three lots from Ivory Coast (890P, 864, 889) and Indonesia (89IP, 892, 909) were carried through the study as separate samples to assess the variation between cocoa beans of the same origin. A l l of the samples were, to some extent, fermented and dried (cured) except the Dominican Republic (Sanchez) sample, which was unfermented.
Parliment et al.; Caffeinated Beverages ACS Symposium Series; American Chemical Society: Washington, DC, 2000.
295 Cocoa Bean Quality Tests
Downloaded by CORNELL UNIV on October 12, 2016 | http://pubs.acs.org Publication Date: January 15, 2000 | doi: 10.1021/bk-2000-0754.ch030
Cut-test Score The cut-test (11) is a visual assessment of 100 halved cocoa beans. Results are qualitative and based on criteria such as cotyledon color and defects. A modified version of the cut-test (12) was used to generate more quantitative results in the form of cut-test scores. One hundred halved cocoa beans were assessed as either slaty (unfermented), fully purple, three-quarter purple, half brown, three-quarter brown, or fully brown (fermented) and scored 1 through 6, respectively. Fermentation Index Fermentation indices (FI) were determined from extracts prepared from ground (40 mesh) raw cocoa. Extracts were prepared with 0.5g cocoa in 50 mL of methanol and hydrochloric acid (97:3). Extracts were cooled to 8°C and held for 16-18 hrs. before the cocoa solids were removed by vacuum filtration. Absorbance values were determined for the extracts from 200 nm to 900 nm using a scanning spectrophotometer. The fermentation index was obtained by calculating the ratio of absorbance at 460 nm to 530 nm (13). Triplicate determinations were conducted and reported as mean ± standard deviation. In addition to FI determination, complete spectrophotometric scans were included to provide a more revealing look at the differences between samples. pH and Titratable Acidity A 10 g sample of raw cocoa (40 mesh) was added to 90 mL of boiling water. The mixture was stirred vigorously and cooled to 20-25°C in an ice bath. The p H was determined using a p H meter standardized with p H 4.0 and p H 7.0 buffers. Following pH measurement, each sample was titrated with 0.1 Ν sodium hydroxide to an endpoint of p H 8.1. Triplicate determinations were performed for both p H and titratable acidity and reported as mean ± standard deviation.
Sensory Analysis Dark chocolates were prepared from the 12 cocoa bean samples using minimal processing and a simplified recipe to allow the cocoa derived flavors to dominate (14). Finished chocolates were 57% cocoa mass, 40% 10X confectionery sugar, and 3% deodorized cocoa butter. A trained sensory panel evaluated the chocolates for 8 attributes. The attributes evaluated were sour, bitter, chocolate, fruity, nutty, smoky, burnt, and astringent mouthfeel. Computer generated 15 point intensity scales (Compusense, Guelph, Ontario), anchored with reference standards, were used to collect the data (14). The 15 point scale went from 0=none to 15=very strong.
Parliment et al.; Caffeinated Beverages ACS Symposium Series; American Chemical Society: Washington, DC, 2000.
296 Analytical Characterization
Downloaded by CORNELL UNIV on October 12, 2016 | http://pubs.acs.org Publication Date: January 15, 2000 | doi: 10.1021/bk-2000-0754.ch030
Gas chromatograms of the volatiles emitted during nib roasting were used for analytical characterization. Nibs were roasted in sealed vials under static conditions and the headspace was auto-sampled and injected into a G C with a flame ionization detector. The capillary column used was a 30 M X 0.25 mm id Supelcowax™10 with a film thickness of 0.25 μπι (Supelco, Inc., Bellefonte, PA). Column temperature was held at 50°C for 5 min, then increased at 5°C/min to 220°C and held isothermally for 10 min. Chromatograms were integrated and divided into 8 retention time segments (A-H) that were used to quantify the volatiles and differentiate the samples (14). The time segments used to divide the chromatograms were chosen based on inspection. Retention time segments were selected at points where peaks could be clearly grouped and where sufficient peaks fell within each segment. The same 8 retention time segments were maintained throughout all of the samples.
Statistical Analyses Sensory data were analyzed by a two-way A N O V A with interaction, for each of the 8 defined flavor attributes. Cocoa bean origin, panelist and their interaction were the factors used in the general linear model ( G L M ) procedure within the S A S statistical program (SAS Institute, Inc. Cary, NC). Analytical data were analyzed by a one-way A N O V A for each of the 8 G C retention time segments. Cocoa bean origin was the factor used in the G L M procedure. Means for both sensory and analytical data were separated using Duncan's Multiple Range procedure at a=0.05. Multivariate Statistics Multivariate statistics examine many variables simultaneously, while reducing them to a few factors capable of providing the most information. Multivariate statistical techniques used for this study were Principal Component Analysis (PCA) and Partial Least Squares regression (PLS) (15-16). P C A was the initial step used to plot or map the cocoa bean samples by origin. Plots were generated for both the sensory and analytical data to determine whether the two maps would group the cocoa beans similarly. Secondly, P L S regression was used to identify correlation between the sensory and analytical data that could be used for predictive modeling. Other researchers (17-18) have successfully used P C A and P L S to develop robust statistical models capable of predicting sensory scores from analytical data. However, Chien and Peppard (79) emphasized that seemingly reliable correlations between sensory attributes and analytical data do not necessarily indicate a cause-and-effect relationship. Researchers can only validate these cause-and-effect relationships through further experiments specifically designed to test them.
Parliment et al.; Caffeinated Beverages ACS Symposium Series; American Chemical Society: Washington, DC, 2000.
297
Results and Discussion
Downloaded by CORNELL UNIV on October 12, 2016 | http://pubs.acs.org Publication Date: January 15, 2000 | doi: 10.1021/bk-2000-0754.ch030
Cocoa Bean Quality Tests Results from the cocoa bean quality tests were used to assess the degree of fermentation and acidity level within the cocoa bean samples. Individually, each of the cocoa bean quality tests provides directional information, but together they provide a more complete picture. The samples have been listed in Table I by cut-test score. Although previous research (72) showed a strong correlation between the cuttest score and fermentation index, the results from this study did not correlate well. In agreement with earlier work (20-27), linear relationships were observed between pH and titratable acidity with a correlation coefficient of -0.93. Aside from being correlated to one another, p H and titratable acidity were highly correlated with FI. The correlation coefficients which related FI to p H and titratable acidity were -0.89 and 0.95, respectively. In addition to the FI values derived from absorbance at 460 and 530 nm, complete spectrophotometric scans were included to more visually reveal differences between the cocoa bean extracts (Figures 1 and 2). The Ecuadaor (Arriba), Indonesian, and D.R.(Sanchez) samples, with their higher absorbance values at 530 nm, were determined to be less fermented. Greater variation in fermentation was observed within the 3 lots from Indonesia as compared to the more consistent fermentation of the 3 lots from Ivory Coast (Figure 1 ).
Table I. Summarized Results from the Cocoa Bean Quality Tests Sample L C . 889 Braz. Bahia L C . 809P Ecu. Arriba Ghana Malaysia L C . 864 Ind. 909 Ind. 892 D.R. Hisp. Ind. 89IP D.R. San.
Cut-test Score 594 593 589 578 566 483 441 438 429 352 337 231
Fermentation Index 1.1281 ± 0 . 0 1 0 9 1.4597 ± 0 . 0 0 1 0 1.1595 ± 0 . 0 1 6 1 0.7647 ± 0 . 0 1 9 4 1.2530 ± 0 . 0 2 5 0 1.9080 ± 0 . 0 1 2 7 1.0522 ± 0 . 0 0 3 0 0.7827 ± 0 . 0 2 3 9 0.6077 ± 0.0052 1.4870 ± 0.0056 0.7424 ± 0.0087 0.5929 ± 0 . 0 2 8 0
pH 5.78 5.31 5.76 6.37 5.39 5.21 5.78 6.10 6.09 5.24 6.09 5.91
±0.08 ±0.04 ±0.02 ±0.11 ±0.05 ±0.05 ±0.05 ±0.05 ±0.03 ±0.01 ±0.05 ±0.03
Titratable Acidity 0.141 ± 0 . 0 0 4 0.193 ± 0 . 0 0 8 0.126 ± 0 . 0 0 4 0.101 ± 0 . 0 0 4 0.149 ± 0 . 0 0 5 0.200 ± 0 . 0 0 9 0.129 ± 0 . 0 0 6 0.103 ± 0 . 0 0 4 0.097 ± 0 . 0 0 4 0.176 ± 0 . 0 0 2 0.086 ± 0 . 0 0 1 0.110 ± 0 . 0 0 1
N O T E : Fermentation Index equals the ratio of absorbance at 460 nm to 530 nm. Titratable Acidity expressed as meq NaOH/g sample.
Parliment et al.; Caffeinated Beverages ACS Symposium Series; American Chemical Society: Washington, DC, 2000.
Downloaded by CORNELL UNIV on October 12, 2016 | http://pubs.acs.org Publication Date: January 15, 2000 | doi: 10.1021/bk-2000-0754.ch030
298
¥ Sr !» Wavelength (nm)
Figure 1. Spectrophotometric scans ofFI extracts from Indonesia and Ivory Coast.
Wavelength (nm)
Figure 2. Spectrophotometric scans ofFI extracts from Malaysia, Dominican Republic (Hispaniola), Dominican Republic (Sanchez), Ghana, Ecuador (Arriba), and Brazil (Bahia).
Parliment et al.; Caffeinated Beverages ACS Symposium Series; American Chemical Society: Washington, DC, 2000.
299 Overall, the results from the cocoa bean quality tests served as excellent indicators of cocoa bean fermentation. The well fermented samples included the 3 lots from Ivory Coast along with samples from Brazil (Bahia) and Ghana. The moderately fermented samples were from Malaysia and Dominican Republic (Hispaniola). The poorly fermented or unfermented samples originated from Ecuador (Arriba), Indonesia (Sulawesi), and Dominican Republic (Sanchez). Although still underfermented, the Indonesia 909 sample was better fermented than the other Indonesian samples.
Downloaded by CORNELL UNIV on October 12, 2016 | http://pubs.acs.org Publication Date: January 15, 2000 | doi: 10.1021/bk-2000-0754.ch030
Sensory and Analytical Results Sensory results have been tabulated in Table II. Significant differences were observed at P