Next Article in Journal
Research Progress on Cannabinoids in Cannabis (Cannabis sativa L.) in China
Previous Article in Journal
Synthesis of New Liquid-Crystalline Compounds Based on Terminal Benzyloxy Group: Characterization, DFT and Mesomorphic Properties
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Link between Flavor Perception and Volatile Compound Composition of Dark Chocolates Derived from Trinitario Cocoa Beans from Dominican Republic

by
Santiago Guzmán Penella
1,2,3,*,
Renaud Boulanger
1,2,*,
Isabelle Maraval
1,2,
Gabi Kopp
3,
Marcello Corno
3,
Bénédicte Fontez
4 and
Angélique Fontana
2
1
CIRAD, UMR Qualisud, F-34398 Montpellier, France
2
Qualisud, Univ Montpellier, Avignon Université, CIRAD, Institut Agro, IRD, Université de La Réunion, 97490 Montpellier, France
3
Barry Callebaut AG, Hardturmstrasse 181, 8005 Zurich, Switzerland
4
MISTEA, Université Montpellier, INRAE, Institut Agro, 34060 Montpellier, France
*
Authors to whom correspondence should be addressed.
Molecules 2023, 28(9), 3805; https://doi.org/10.3390/molecules28093805
Submission received: 31 March 2023 / Revised: 25 April 2023 / Accepted: 26 April 2023 / Published: 28 April 2023
(This article belongs to the Section Food Chemistry)

Abstract

:
The chemical composition of dark chocolate has a significant impact on its complex flavor profile. This study aims to investigate the relationship between the volatile chemical composition and perceived flavor of 54 dark chocolate samples made from Trinitario cocoa beans from the Dominican Republic. The samples were evaluated by a trained panel and analyzed using gas chromatography-mass spectrometry (GC-MS) to identify and quantify the volatile compounds. Predictive models based on a partial least squares regression (PLS) allowed the identification of key compounds for predicting individual sensory attributes. The models were most successful in classifying samples based on the intensity of bitterness and astringency, even though these attributes are mostly linked to non-volatile compounds. Acetaldehyde, dimethyl sulfide, and 2,3-butanediol were found to be key predictors for various sensory attributes, while propylene glycol diacetate was identified as a possible marker for red fruit aroma. The study highlights the potential of using volatile compounds to accurately predict chocolate flavor potential.

1. Introduction

Chocolate is highly consumed in developed markets and is increasingly penetrating new markets, mainly in Asia and Latin America. In 2021, the global chocolate market was worth USD 46.6 billion and is expected to grow at a compound annual growth rate of 4.98% from 2022 to 2028, expecting to be valued at USD 65.49 billion by the end of this forecast period [1].
While 90% of the world’s total “fine” cocoa exports come from Latin America, the Dominican Republic is one of the three biggest exporting countries, along with Ecuador and Peru. In broad terms, fine “cocoa” is mostly produced from the Criollo or Trinitario varieties and is distinguished from “bulk” cocoa because of flavor alone. Fine cocoa often exhibits fruity, floral, herbal, woody nutty, and caramel-like notes [2].
Cocoa products from Dominican Republic beans are known to display multiple dominant flavor profiles, ranging from low cocoa and bitter, to winey, earthy, and spicy notes [3]. Furthermore, Trinitario cocoa has been traditionally known as fine-grade cocoa [4], and has been associated to varied sensory attributes, including a marked chocolate flavor and distinctive wine-like notes [5].
Flavor is one of the most significant consumer acceptance parameters. It is hence one of the main quality indicators for cocoa products. Cocoa flavor is highly complex as many diverse components may be linked to its resulting sensory perception. Both non-volatile and volatile compounds contribute to the overall flavor profile of cocoa and cocoa-derived products. Over 600 odor compounds have been reported to be found in cocoa and chocolate [6].
The aroma composition of cocoa products is tightly related to the unique postharvest processing conditions, as well as to the variety and the origin of the cocoa itself [7].
Among the non-volatile compounds in cocoa, alkaloids and polyphenols have arguably the highest impact on flavor perception, as they are both linked to bitterness. Additionally, polyphenols are associated to astringent sensations and contribute to green and fruity flavors [8]. Moreover, proteins and carbohydrates are non-volatile compounds that play an important role in the formation of volatile aroma compounds during the drying, roasting and conching processes by means of Strecker Degradation and a Maillard reaction [6].
On the other hand, volatile aroma compounds found in cocoa products include esters, alcohols, acids, and phenols, which are mostly derived from the fermentation and drying processes. These compounds tend to be linked to sweet, sour, fruity, and floral notes, with the exception of phenols, which may convey smoky and other generally undesirable hints [9]. Pyrazines, aldehydes, and ketones products resulting from Maillard reactions are other volatile compounds of interest. Some ketones and aldehydes are known to bestow floral notes to cocoa, but pyrazines are usually associated to the expression of nutty, earthy, roasted, and green notes [10].
The use of gas chromatography-mass spectrometry (GC-MS) for the characterization of the volatile composition of chocolate and other cocoa products is a well-established technique [1,5].
In recent years, multiple studies have been conducted in an attempt to better understand the impact of different variables, such as the plant genetic origin or processing conditions, on the volatile composition and on the flavor perception of cocoa and chocolate [11,12,13,14,15,16,17,18,19,20]. However, these studies did not always look for the relationship existing between the volatile composition and sensory perception in chocolate. Many of them have tried to identify compounds that could be used as markers for certain flavor attributes [21,22,23,24,25]. While some have looked into the single compound’s presumed impact on the actual perceived taste of chocolate itself [26,27,28,29], most have focused solely on the individual compound’s reported sensory descriptor, independent to the food matrix itself. Additionally, some studies have either contradicted or have not been able to corroborate previously reported results when trying to establish the individual contribution of single compounds to the overall perceived flavor profile in final products, as has for instance been the case of linalool’s impact on floral notes [30]. A rather limited number of studies have indeed attempted to describe and predict the perceived intensity of sensory attributes based solely on the volatile composition of the cocoa products [31,32,33]. Some others have classified samples into differentiated sensory groups based on their overall volatile composition and sensory profile, but without distinguishing between unique sensory attributes [34]. A 2022 study claimed to be the first to show how distinct differences in the flavor profiles of dark chocolates are reflected in their molecular compositions, while also factoring in the compounds’ odor activity values (OAC) and dose over threshold factors (DoT) [30].
In spite of the growing interest in researching both volatile composition and flavor in cocoa and chocolate, identifying the linkage between the two remains particularly challenging because of the established high chemical complexity of the product. Moreover, the differences among chocolates produced from beans of the same variety and origin are bound to be more nuanced compared to those found in chocolates derived from cocoa that have more obviously differentiated backgrounds. Furthermore, the ability to differentiate chocolates in terms of their flavor profile in a fast, repeatable, and reliable way is of major interest for the industry, notably in the case of highly-sought-after attributes, such as fruity or floral.
The focus of this study is to identify key volatile aroma compounds that may enable differentiation within a selection of dark chocolate samples derived from Trinitario beans grown in the Dominican Republic, as well as to ultimately develop sufficiently accurate predictive models that would allow the discrimination of samples with regard to their flavor potential. This characterization will be based on the perceived sensory profile of each of the samples, blind to their processing conditions, with the aim of understanding to what extent subtle differences in taste could be described by differences in volatile composition.

2. Results

2.1. Sensory Characterization of Dark Chocolate Samples

A clustered heatmap (Figure 1) allows for a summarized visualization of the raw (not pretreated data, as in not centered nor scaled) mean sensory scores of all 54 chocolate samples evaluated in this study, along with insights into the correlations that exist among the different attributes that were evaluated (sweetness, bitterness, acidity, astringency, cocoa, yellow fruits, red fruits, citrus, dried fruits, nutty, winey, black olives, green, earthy, floral, woody, spicy, and roasted). In Appendix A, Table A1, additional information on the overall sensory results obtained from the chocolate dataset is presented.
It can be seen that cocoa, acidity, bitterness, and astringency are attributes highly correlated with each other. Moreover, cocoa is the attribute that displays the highest overall mean sensory scores, which would indicate that the majority of the chocolate samples display a flavor profile that has strong cocoa notes.
All of the fruity sub-attributes (red fruits, yellow fruits, citrus, and dried fruits), together with the floral attribute, also present a high degree of correlation based on their mean sensory scores. These fruity mean sensory scores show their highest values mostly whenever the values for nutty, roasted, and spicy attributes are the lowest.
The earthy, woody, winey, and black olives attributes present the lowest mean sensory scores among all attributes. The flavor profiles of the set of chocolate samples evaluated are thus characterized for weakly expressing these attributes. Furthermore, earthy, woody, winey, and black olives seem to show their highest mean sensory score values whenever the values for bitterness and astringent are also high.
A principal component analysis (PCA) was performed on the mean sensory scores of all sensory attributes obtained for all samples. This was performed to characterize how the differences in the reported mean sensory scores were driving sample differentiation with regards to their sensory profiles. The two main components, F1 and F2, were able to explain 51.86% of the variance among samples.
In Figure 2, the PCA biplot showing the distribution of all chocolate samples in the space of F1 and F2 is presented, together with the loadings, corresponding to all sensory attributes (variables). The scree plot pertaining to this analysis is shown in Appendix A, Figure A1. It indicates that the first two components are responsible for a substantial amount of the variability in the data, while the third component explains a relatively small amount of variance.
The samples seem to be well scattered along both axes. The distribution along the F2 axis is presumably defined by the direction of cocoa and winey attributes. Furthermore, groups of samples can be observed in the whole F1-F2 plane, which could be potentially described in terms of the membership to each of the different quarters.
The samples exhibiting high roasted and nutty mean sensory scores are opposed to samples with high mean sensory scores for acid, floral, and citrus attributes. Similarly, samples whose mean sensory scores are high for sweetness, yellow fruits, and dried fruits, are opposed to samples high in woody, spicy, earthy, black olives, green, bitter, and astringent mean sensory scores.

2.2. Identification and Quantification of Volatile Aroma Compounds Present in the Dark Chocolate Samples

Volatile compounds present in the 54 dark chocolate samples were determined by GC-MS; 34 known compounds were identified to be present from well-defined peaks shown in the retrieved spectra. An unidentified 35th compound was found in several of the samples, but it could not be associated to any of the spectra already in the NIST Mass Spectral Library. All identified volatiles, together with the odor descriptors commonly associated individually to each, are listed in Table 1.
In Appendix A, Table A2, additional information is presented on the relative concentrations obtained for each of the identified volatile compounds among the chocolate sample dataset.
A PCA was performed on the relative concentrations of the identified volatile compounds. While the two main components (F1 and F2) were able to explain 54.74% of the variance among the samples, the contribution of the third component to the explanation of the variance of the dataset is also worth discussing. This is because the third component appears to explain a sufficiently large amount of variance (as seen in the scree plot pertaining to this PCA, shown in Appendix A, Figure A2). Together, F1 and F3 are capable of explaining up to 53.18% of the dataset variance. In Figure 3, the PCA biplot showing the distribution of the chocolate samples in the space of F1 and F2, as well as in the space of F1 and F3, is presented, together with the variable loadings, corresponding to all volatile compounds.
Within the space of the first two components, the samples are well scattered. A segregation of the samples is predominantly observed in a gradient along the F1 axis, pointing towards a somewhat well-defined formation of two distinct groups. F1 sample differentiation appears to be mainly driven by the relative concentrations of pentan-2-ol (whose contribution to the building of the component is of 3.46%), ethanol (3.36%), ethyl acetate (3.32%), 3-methylbutan-1-ol (3.09%), and dimethyl sulfide (2.83%) on one side. While on the opposite extreme of the F1 axis, the most contributing volatile compounds leading the differentiation of the samples are acetic acid (contributing to the component at 5.79%), 2,3,5,6-tetramethylpyrazine (5.70%), 2,3,5-trimethylpyrazine (5.60%), benzaldehyde (5.45%), 2,3-dimethylpyrazine (5.15%), and 2,3-butanediol (isomere A; 4.93%).
Differentiation of the samples along the F2 axis is less pronounced, compared to what was seen along the F1 axis. No evident grouping of samples is seen here and the gradient segregation is not as widely dispersed. Nonetheless, the volatile compounds contributing the most to this second component may be clearly identified as diethyl butanedioate (12.54%), 1,3-diacetoxypropane (9.10%), and the “unidentified compound” (7.81%) on one side. The following compounds may be found towards the opposite direction, along the components: axis butane-2,3-dione (11.13%), 2-methylbutanal (7.10%), 3-methylbutanal (5.63%), acetoin acetate (5.50%), and 2-methylpropanal (5.31%).
It may then be assumed that differences in the relative concentrations of the above-listed compounds for both F1 and F2 are contributing the most to the differentiation of samples, in relation to their overall volatile composition.
Furthermore, it has been observed that sample segregation is not necessarily associated with differences in the relative concentrations of entire chemical families, but rather appears to be linked to specific, unique compounds that are independent of their chemical family affiliation. This observation suggests that the sensory attributes of samples may be more closely associated with individual compounds, rather than the chemical families to which they belong.
In relation to the biplot on the F1 and F3 components, it must be noted that the dimension of the third component appears to be mostly represented by one single chocolate sample (contributing to 25.13% of the building of the component), which would indicate this sample is an atypical individual. This atypical sample seems to be characterized by compounds such as 3-methylbutan-1-ol (which is contributing to 10.55% of the third component), 2-methylpropan-1-ol (10.18%), isoamyl acetate (9.73%), ethyl acetate (8.88%), pentan-2-ol (8.75%), and ethanol (6.27%). It is only when looking at the third dimension that these compounds are visually separated along its axis from dimethyl sulfide. Other compounds whose distribution in relation to the other compounds changes markedly along the axis of the third component are 1,3-diacetoxypropane and the unidentified compound, both of which are also important contributors to the building of these components, and which now seem to be clearly distanced from diethyl butanedioate, for instance.

2.3. Identification of Key Aroma Compounds Based on Their Impact on the Sensory Perception of Dark Chocolate Samples

A partial least squares (PLS) regression was performed in an attempt to obtain a global view of the whole dataset. This allows the principal relationships existing between groups of volatile compounds (explanatory variables) whose concentrations may be able to predict mean sensory scores (response variables) to be summarized. The variables were centered and reduced. The quality of the PLS model obtained is explained using a bar plot presented in Appendix A, Figure A3; it shows that while the first two components summarize the correlations between explanatory and dependent variables well, the third component may still provide additional information. The obtained bi-dimensional correlation plot on axes t1 and t2 is presented below, in Figure 4.
This overview of the PLS regression showed a number of similarities with the two PCAs that had been already performed on both mean sensory scores and on the relative concentration of the volatile compounds. Taken as a whole, this PLS regression confirmed that the samples are broadly differentiated in two groups. One of these groups could be described as highly aromatic for being linked to large concentrations of plenty of different volatile compounds, which at the same time seem to describe higher fruity, floral, and sweet mean sensory scores. On the other hand, a group of samples seems to be characterized by lower concentrations of most of the identified volatiles and by exhibiting higher mean sensory scores for astringent and bitter attributes, among others.
The PLS regression demonstrated that all fruity attributes remain seemingly very highly linked, just as it was seen with the PCA. Not only is the link between the different fruity attributes maintained, but the relationship existing between them and the acid, sweet, and floral attributes appears to be further tightened. Moreover, these attributes seem to be very tightly correlated to the differences in the relative concentration of numerous volatile compounds, including but not limited to acetic acid, 2,3-butanediol, propylene glycol diacetate, and 2,3-dimethylpyrazine.
Being located closer to the center of the plot, attributes such as spicy, winey, and nutty appear to be amongst the most difficult to predict. This could indicate that differences in their mean sensory scores are not well described by differences in the relative concentrations of the volatile compounds. Noticeably for spicy, nonetheless, it was already seen in the PCA that its contribution to the first two principal components was moderate and was thus not driving sample segregation as much as the other attributes were. This might suggest that the mean sensory scores for spicy, winey, and nutty are overall too low and similar for them to drive sample differentiation altogether (or that they could be better represented in the other PCA components).
The attributes of cocoa and roasted remain closely linked, as seen in the PCA, and also appear to be largely described by lower relative concentrations of most of the volatile compounds.
Similarly, the PLS regression suggests that the attributes green, earthy, black olives, woody, bitterness, and astringency are as related as initially observed in the PCA. It would also seem that they may all be well characterized by higher relative concentrations of ethanol, ethyl acetate, pental-2-ol, dimethyl sulfide, and 3-methylbutan-1-ol.
Lastly, there is a subgroup of samples that exhibit high relative concentrations of compounds, including but not restricted to 2,3-dimethylpyrazine, 3-methylbutanal, 2-methylpropanal, and 2-methylbutanal, but which are seemingly not well characterized by any single sensory attribute.
The bidimensional correlation plot on axes t1 and t3 is also presented as complementary information (Appendix A, Figure A4). It shows how compounds such as acetaldehyde and butane-2,3-dione are closely grouped together and relatively separated from the rest along the t3 axis, being now more closely related to sensory attributes such as acidity and sweetness. This might be giving hints at these compounds’ potential ambivalent character, being linked to multiple sensory attributes at once.

2.4. PLS Predictive Models for Individual Sensory Attributes

It would be of interest to gain an in-depth understanding of the specific compounds contributing to predicting each attribute. For this purpose, PLS predictive models were built. Given the results of the prior global PLS regression analysis, which already established some key relationships between the explanatory and response variables, it is not expected that a substantial difference will be observed.
For each attribute, the samples were classified in two groups relative to their mean sensory scores: samples above the median and samples below the median. This classification would enable samples to be broadly differentiated in terms of their flavor potential per attribute, delimited by the current sample set.
As described in Section 4. Materials and Methods, training models were built and optimized by cross-validation. Predictive models were then built for the best fitted training models. The model performance metrics for each of the obtained models are presented in Table 2.
The five volatile compounds that had the largest impact in the building of each model (most important variables) were retrieved for reporting. A Pearson correlation analysis was performed on the raw data in order to understand the nature of the correlations existing between the explanatory (volatile compounds) and the response variables (sensory attributes). Both the most important compounds per predictive model and their corresponding Pearson correlation coefficients (obtained from the raw data) are presented in Table 3.
Acetaldehyde seems to be a key compound in predicting multiple distinct models, and it is either positively or negatively related to an increase in the attribute’s intensity. Other compounds whose importance is prevalent in the building of several models are the two isomeres of 2,3-butanediol, as well as dimethyl sulfide, followed by 3-methylbutan-1-ol, acetic acid, 3-methylbutanal, pentan-2-ol, ethyl acetate, and butyrolactone. The rest of the listed compounds seem to be more attribute-dependent, as they present the highest importance in describing single unique attributes.
A different and more restrictive classification of the samples was then performed. The samples were classified as “high” based on their mean sensory scores, with the highest quartile of the scores being considered for the classification. Subsequently, the models were developed as previously described based on this newly established classification of the samples. ROC was again used as the metric. If was found that the accuracy, specificity, and sensibility of most models decreased considerably, except for those of bitterness and astringency. The best models obtained for both bitterness and astringency are presented in Table 4.
There is only one compound whose importance markedly differs when attempting to predict whether a sample is highly bitter as opposed to merely above the median, and this is 2,3-butanediol (isomere B), which is now listed among the top five most important variables. This would suggest that these compounds have a stronger differentiating power than the rest when attempting to discriminate the samples that express a higher intensity of bitterness.
While for astringency, on the other hand, there is a more pronounced rearrangement of the importance of the variables responsible for predicting if a sample will be classified as highly astringent, as acetic acid, 2,3-butanediol (isomere B) and 3-methylbutan-1-ol now acquire much more importance.

3. Discussion

Given that the set of chocolate samples evaluated in the context of this study were produced from cocoa of the same variety and region, their flavor profiles were not expected to be as different as if chocolates produced from different regions and genotypes had been included. Nonetheless, while most samples displayed a dominant cocoa profile, important differences were found in the expression of most other attributes. These differences translated into certain samples displaying markedly differentiated fruity notes as opposed to others exhibiting more of a bitter/astringent/spicy/winey profile, covering a wide flavor range within the varietal limitations of the cocoa in this region.
The grouping of the samples based on their sensory profiles (Figure 2) showed certain parallels to what has been reported in similar studies conducted on the organoleptic properties of cocoa products, where PCA plots have also grouped together attributes such as astringent, bitter, and green, opposed to fruity, floral acid, or cocoa notes, for instance [39].
All of the identified compounds had been previously reported in the literature as present either in dry fermented cocoa beans, roasted beans, liquor, or dark chocolate from different varieties and regions, processed under unique conditions [14,17,18,20,32,38,39,40,41,42,43,44,45,46]. Hence, none of the 35 identified compounds point to being unique markers that could be potentially used for differentiating the Dominican Republic chocolates that make up the sample dataset of this study in relation to other cocoa products.
The predictive models obtained from the PLS regression are limited in their predictive capacity due to the reduced number of samples used, resulting in a data-dependent model. Despite this limitation, the models still display a relatively good level of predictive accuracy. It is important to be cautious when interpreting the results of the analyses aimed at identifying the key compounds that could predict the perceived intensity of the evaluated sensory attributes. There are several confounding factors that must be taken into consideration, such as variable exchangeability and causality [47].
Simply identifying a compound as having an important relation with a particular attribute does not necessarily imply causality. To establish causality, it would be necessary to consider other factors that may be contributing to the expression of the sensory attribute and to compare with previously reported findings. Variable exchangeability refers to the presence of highly correlated variables, both explanatory and response variables, which can lead to misleading conclusions about the importance of a particular compound in predicting sensory attributes.
Both variable exchangeability and causality will now be further discussed in relation to the results obtained in an attempt to determine whether causality may be assumed. By carefully considering both of these factors, a deeper understanding of the mechanisms of sensory perception may be gained and more accurate predictions about the impact of individual compounds on the perception of sensory attributes may then be conducted.
When attempting to identify the most important compounds responsible for contributing to the perception of each individual sensory attribute, it was seen that the existing underlying correlation was often negative. This could indicate that the presence of certain volatiles was detrimental to the perception of the given attribute, or in other words, that low concentrations of said compounds were needed to maximize the intensity of the described attributes.
The above statement seems particularly true for bitterness and astringency. It is worth noting that most of the compounds identified as important for predicting both bitterness and astringency hold a negative correlation with the intensity of both attributes, which would mean that the presence of these compounds in low concentrations would be needed for a sample to be highly bitter and astringent.
Bitterness (classified as one of the four primary tastes, along with sweet, sour, and salty) and astringency (a trigeminal sensation) have both been strongly linked in cocoa products to mostly non-volatile compounds [48]. Theobromine and caffeine, for instance, are alkaloids that contribute to the typical bitter taste of cocoa [6]. Polyphenols and flavonoids such as tannins, flavan-3-ols [(+)-catechin, (-)-epicatechin and (-)-epigallocatechin], and anthocyanins have also been associated with an astringent and bitter taste [49].
Nonetheless, it would appear that the perception of both bitterness and astringency increases whenever the concentration of compounds known to be linked to unpleasant notes is relatively high, as it happens with dimethyl sulfide and 3-methylbutan-1-ol, while the perceived intensity seems to decrease whenever the sample contains higher concentrations of compounds linked to pleasant notes, such as the sweetness associated with 2,3-butanediol, the chocolate taste linked to 3-methylbutanal and 2-methylpropanal, the acidity of the acetic acid, and the fruity-like odor of the acetaldehyde. This could suggest that these desirable compounds may have a particularly important masking effect over the non-volatile-associated bitterness and astringency, as well as over the high concentrations of volatile compounds responsible for the unpleasant notes that may be reminiscent to the bitter taste and the astringent sensation.
In order to better understand the real impact of these compounds’ potential “masking effect” on the reduced perception of bitterness and astringency in chocolate, further analysis would be needed to also take into account the concentration of non-volatile bitter and astringent compounds and to look deeper into the existing correlations.
It is noteworthy that bitterness and astringency exhibit a relatively high degree of correlation, with a Pearson correlation coefficient of 0.732. This correlation is reflected in the shared compounds that were selected as important for predicting both sensory attributes (acetaldehyde, dimethyl sulfide, and 2,3-butanediol (isomere A)). This could also potentially mean that the volatile compounds identified as important in describing bitterness are also likely to be important in describing astringency. This suggests that compounds such as 2-methylpropanal and 3-methylbutanal, which are related to a chocolate-like flavor, may be important when describing bitterness, as well as compounds such as 3-methylbutan-1-ol, which has a pungent taste, in describing astringency. This highlights the potential for overlapping contributions of certain volatile compounds in both attributes, further emphasizing the interconnection of bitterness and astringency in the flavor profile of dark chocolate.
The acidity, which is another taste that was accurately described and predicted by the set of identified and quantified volatiles in this study’s samples, was unsurprisingly largely influenced by high concentrations of acetaldehyde, which has an acidic taste. Surprisingly, on the other hand, the differences in the concentration of acetic acid appeared not to have an important effect on the increased perception of acidity, which would have been expected.
Other volatiles found to be seemingly important in describing the acid taste were not necessarily acidic in nature (butyrolactone, isoamyl acetate, and 2,3-butanediol), leaning more towards sweet and fruity notes. This could suggest that such compounds may have a potential enhancing effect on the perception of the volatile acidity brought in by acidic volatile compounds such as acetaldehyde and acetic acid, as well as by the non-volatile acidity, for which phosphoric and lactic acids are mostly responsible, together with oxalic, malic, succinic, and citric acids [50].
Nonetheless, another possible explanation would be that some of these non-acidic compounds are actually statistically exchangeable with the acetic acid or other volatile compounds with an acid-like taste. When looking into the existing correlations between the relative concentrations of volatile compounds, it may be seen that the acetic acid is very highly correlated with most of the compounds here identified as important in describing acidity: 2,3-butanediol (isomere A; Pearson correlation coefficient of 0.673), 2,3-butanediol (isomere B; 0.850), acetaldehyde (0.673), and butyrolactone (0.595). This could then imply that some of these compounds were selected as important variables in the PLS regression because they were providing similar information compared to that of the acetic acid. This redundancy would make it difficult to determine which variable is the most important predictor and could not allow us to discard acetic acid as a likely important contributor in the expression of acidity.
Derived from alanine, acetaldehyde is one of the most abundant Strecker aldehydes linked to chocolate notes [51], and it is one of the most abundant carbonyl compounds found in many fermented foods [52]. Acetaldehyde was found to be of high importance in describing and predicting most of the sensory attributes discussed in this study. This compound is mostly formed during alcoholic fermentation, by the decarboxylation of pyruvate, after which it may subsequently be transformed into ethanol by alcohol dehydrogenase enzymes [53]. Acetaldehyde is also an intermediate in the synthesis of acetic acid and acetoin, which may later be reduced to 2,3-butanediol [51]. While acetaldehyde’s associated flavor is commonly described as “oxidized”, studies conducted into its impact on the flavor perception of wines have shown that, at different concentrations, it may be linked to markedly differentiated sensory notes, ranging from fresh fruit aromas at low concentrations, to nutty, cocoa, ripe fruit, and even rotten-like off-flavors at higher concentrations [53].
The numerous aroma descriptors that have been associated to different concentrations of acetaldehyde could help to explain why it has also been identified as a key compound when attempting to explain different chocolate descriptors in this present study. Additionally, it must also be considered that interactions of acetaldehyde with other molecules may not only affect its flavor perception in chocolate, but may also alter the perception threshold of free acetaldehyde [53].
In addition, it has been noted that during cocoa fermentation, acetaldehyde reacts with epicatechin and procyanidin B2 to form ethyl-linked flavan-3-ol trimers [54]. This could be the reason behind the observed negative correlation between the relative concentration of acetaldehyde and the bitterness mean sensory scores, as it would suggest that a greater amount of the available acetaldehyde was involved in the formation of non-volatile bitter compounds through condensation reactions.
The statements above provide a reasonable explanation for the causal relationship between the increased acetaldehyde content and the heightened perception of certain attributes such as fruitiness, as well as the reduced perception of bitterness and astringency. Furthermore, these deduced causal relationships are reinforced by the lack of strong correlation between the relative concentration of acetaldehyde and that of any other compound, which reduces the likelihood of mistakenly interpreting the impact of this compound as it would not be easily exchangeable.
Similarly, while dimethyl sulfide (a sulfur compound widely present in food products) on its own has been described as exhibiting green and unpleasant cabbage-like aromas [55], studies about its impact on other aroma descriptors in wine as a matrix have been published. It has been suggested that dimethyl sulfide may play a role as a fruity flavor enhancer, especially in the case of blackberry and blackcurrant aromas [55]. Furthermore, the presence of dimethyl sulfide has also been linked to a decrease of the olfactory threshold of fruity notes and to an increase in overall flavor intensity. At higher concentrations, on the other hand, it has been reported as displaying notes more resembling black olives and truffles [55]. The results obtained seem to point to a similar direction in chocolate. Based on the retrieved Pearson correlation coefficients, it would seem that higher concentrations of this compound are presumably responsible for enhancing the expression of bitterness and astringency, as well as the perceived green and black olive aromas. It would also seem that its presence in low concentrations could indeed play an important role in enhancing a red fruit aroma and sweet taste in the dark chocolate samples evaluated.
Dimethyl sulfide seems to exhibit no strong correlation with any other compound, indicating that no other volatile compound is providing similar information as dimethyl sulfide. This suggests that the conclusions drawn about it are likely valid.
Another compound whose importance seems to be prevalent in describing and predicting several of the studied sensory attributes is 2,3-butanediol. The present study’s results show that higher concentrations of this compound appear to be related to a higher overall flavor intensity, particularly in the case of desirable notes such as citrus, red fruits, and sweetness; whereas lower concentrations seem to be linked to more intense bitterness and astringency, which are attributes that tend to be associated with lower quality chocolate. The obtained results would then be in agreement with previously reported propositions that suggest that the presence of this compound is desirable for high quality cocoa products [56].
As expected, the two identified isomers of 2,3-butanediol show a strong correlation with each other, suggesting that the information they are each providing is redundant when predicting any of the given attributes. Due to this exchangeability, the correlation they have with other compounds is also quite similar, as is the case with their correlation with acetic acid.
The identification of propylene glycol diacetate as a key predictor is noteworthy. This volatile compound has been described as having a fruity aroma [43] and it has now been found to be an important variable when determining the potential intensity of a red fruit aroma in chocolate samples. Its high correlation with the red fruit sensory attribute and its unique importance in predicting the expression of red fruits make it a valuable marker in this regard. It is the second most important predictor of red fruit expression after acetaldehyde, suggesting that there may be a combined effect on this particular fruity attribute. Additionally, propylene glycol diacetate does not appear to be strongly correlated to any compound other than 2,3,5,6-tetramethylpyrazine and acetophenone, having Pearson correlation coefficients of 0.739 and 0.733, respectively. Since neither of these two compounds are linked to expressing red fruit-related aromas, it may then be assumed that the information provided by them is not similar and that propylene glycol diacetate might indeed be responsible for the expression and perception of a red fruit aroma in the chocolate samples evaluated.

4. Materials and Methods

Dark Chocolate Samples—Barry Callebaut AG provided 54 different dark chocolate samples, which were produced following internal and non-disclosed processing parameters. All chocolates were produced from cocoa beans issued from a fermentation campaign of Trinitario beans carried out by Barry Callebaut AG in the Dominican Republic, in April 2020. The recipe of the chocolates was described as containing: 60% cocoa liquor, 30% sugar, and 10% deodorized cocoa butter; no lecithin was added. The tempered chocolates were molded in plastic molds into individual chocolate square-shaped pieces (3.5 cm × 3.5 cm), each weighing approximately 5 g. The molded chocolates were vacuum sealed and allowed to stabilize for a month at room temperature. The vacuum-sealed chocolates were then stored inside a freezing chamber at −20 °C. Prior to tasting, the chocolates were allowed to defreeze at room temperature for two days.
Sensory Analysis—All tasting sessions were carried out at CIRAD’s (Montpellier, France) sensory analysis laboratory, in individual boxes and under a red light. Blind sensory analysis was performed on 54 dark chocolate samples by a panel composed of 13 trained tasters, 6 of which were female and 7 were male, aged between 21 and 60 years (all members of CIRAD’s internal sensory panel).
The 13 panelists, who were already familiar with dark chocolate sensory evaluation, were trained throughout six training sessions for the purpose of this analysis. Six different dark chocolates were used for this training, each exhibiting distinctive and intense attributes in particular (e.g., cocoa, bitter, yellow fruits, etc.). The chocolates used for the training were either provided by Barry Callebaut AG and CIRAD, or they were bought from local chocolatiers in France. During the first two training sessions, open discussions were held on the perceived attributes of each of the chocolates, In the course of the next four sessions, the chocolates were then blindly tasted, repeatedly. The performance of the panelists was validated based on their repeatability and agreement with the rest of the panel.
Eight sessions distributed over six weeks were needed for the tasting and evaluation of the 54 chocolates. In each session, six chocolates were evaluated, one of which was a replicate sample taken randomly within that session’s set of samples. The sensory attributes (sweetness, bitterness, acidity, astringency, cocoa, fruity-yellow fruits, fruity-red fruits, fruity-citrus, fruity-dried fruits, nutty, winey, black olives, green, earthy, floral, woody, spicy, and roasted) were evaluated using a score that ranged from 0 to 10. For each attribute, the mean sensory score was calculated from the scores given by the eight most discriminant panelists. The most discriminant panelists were those who gave a wider range of scores to each attribute among different samples, effectively demonstrating their ability to best differentiate the subtleties between chocolate samples.
Volatile Analysis—Dark chocolate samples were frozen in liquid nitrogen before being milled using a conventional coffee mill. The retrieved powder was then sieved and stored in a freezer at −20 °C. Volatile compounds were extracted from 2 g of sieved sample powder by means of headspace solid-phase micro extraction (SPME-HS), using a 50/30-μm divinylbenzene/carboxene/polydimethylsiloxane (DVB/CAR/PDMS) fiber (Supelco Analytical Products—Sigma Aldrich, Merck, Darmstadt, Germany). The extracted volatile compounds were analyzed using an Agilent 6890 N gas chromatography–mass spectrometer (GC–MS) equipped with a capillary column DBWAX, 60 m length × 0.25 mm internal diameter × 0.25 μm film thickness (Agilent, Santa Clara, California, USA). The full procedure for volatile compound extraction and identification was previously described by Assi-Clair et. al. [57]. The relative concentration of each compound was calculated based on the exact weight of the sample and on the known concentration of the compound that was added as an internal standard (butan-1-ol). Each sample was analyzed in triplicate and the mean of the three obtained concentrations per sample was recovered for reporting (in μg/g of fresh matter) and for the subsequent statistical analysis.
Statistical Data Analysis—Statistical analysis, including chemometric analysis based on the principal component analysis (PCA) and on partial least squares (PLS) regression, was performed with XLSTAT STUDENT (Addinsoft, Paris, France).
PLS descriptive and predictive models were built in R, using the ‘caret’ package. All variables were centered and scaled. The dataset was split into training and testing sub-datasets. PLS training models were built, using the ‘trainControl()’ function. The training models were optimized by cross-validation following the “repeatedcv” method, by repeatedly partitioning the data into a fixed number of equally sized groups (20 folds were used) and then training and testing the model using each group as the validation set. This process was then repeated 30 times. The metric used in the cross-validation with the training data subset was the receiver operating characteristic (ROC).
The best models obtained from the training data subset were then validated using the testing data subsets. This operation was repeated in a loop in order to cover the entirety of the dataset, iteratively training and testing the PLS models on different subsets of the data. The ‘predict()’ function was used to obtain predictions for the test set based on the trained models for each attribute. In order to determine the best fitted predictive model per attribute, the following performance indicators were taken into consideration: accuracy, confidence interval, sensitivity, and specificity.

5. Conclusions

The results of this study provide new insights into the link between the volatile composition and the perceived flavor of dark chocolates processed from cocoa beans sourced from the Dominican Republic. This study helped identify certain volatile compounds that are important in predicting the intensity of the sensory attributes of interest in dark chocolate samples. Acetaldehyde, dimethyl sulfide, and 2,3-butanediol were found to be key predictors in identifying the intensity at which multiple sensory attributes may be perceived, including bitterness, astringency, citrus, acidity, red fruits, dried fruits, green, black olives, woody, and sweetness. While the relative concentrations of these compounds seem to be key in describing and predicting the flavor intensity of multiple attributes, propylene glycol diacetate was identified as a unique key compound in describing and predicting a single attribute (red fruit aroma).
This study has also demonstrated the potential to differentiate chocolates with relative accuracy based on their flavor profiles, using predictive models based solely on the volatile composition of the analyzed chocolates. The efficacy of these models could be improved by incorporating more samples, leading to more accurate flavor profile predictions. Furthermore, these models could potentially be used by industry players as a reliable, repeatable, and inexpensive tool, which would spare them the costs of using human sensory panels for the classification and evaluation of chocolate and other cocoa products. As such, the results of this study open up new avenues for the cocoa and chocolate industries to evaluate and optimize their flavor quality.

Author Contributions

Conceptualization, S.G.P., R.B., I.M., G.K., M.C., B.F. and A.F.; methodology, S.G.P., R.B., I.M. and A.F.; software, S.G.P. and B.F.; validation, S.G.P., R.B., I.M. and A.F.; formal analysis, S.G.P. and B.F.; investigation, S.G.P.; data curation, S.G.P.; writing—original draft preparation, S.G.P.; writing—review and editing, S.G.P., R.B., I.M., G.K., M.C., B.F. and A.F.; visualization, S.G.P.; supervision, S.G.P., R.B., I.M. and A.F.; project administration, R.B.; funding acquisition, R.B. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by a doctoral fellowship (Convention CIFRE N° 2019/1911), jointly financed by Barry Callebaut France and the ANRT (Association Nationale de la Recherche et de la Technologie).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data presented in this study are not available.

Acknowledgments

The authors thank the CIRAD panelists that contributed to the sensory evaluation of the chocolate samples, as well as the scientific staff at UMR Qualisud.

Conflicts of Interest

The study was funded by Barry Callebaut AG under the participation of CIRAD, UMR Qualisud, MISTEA, and Barry Callebaut AG.

Sample Availability

No sample is available from the authors.

Appendix A

Table A1. Average, standard deviation, minimum, and maximum values of the mean sensory scores of all samples per sensory attribute evaluated. The scores ranged from 0 to 10.
Table A1. Average, standard deviation, minimum, and maximum values of the mean sensory scores of all samples per sensory attribute evaluated. The scores ranged from 0 to 10.
Sensory AttributeMEANSTDEVMAXMIN
Sweetness4.2450.5355.1253.125
Bitterness4.1080.9916.7502.750
Acidity3.9970.6866.0002.375
Astringency4.0380.7625.3752.250
Cocoa5.4470.5296.5634.000
Yellow fruits2.5980.9834.7500.750
Red fruits2.7031.0335.1250.500
Citrus2.2880.8664.2500.375
Dried fruits2.0890.7103.6250.500
Nutty2.0820.5853.6251.000
Winey1.2580.4922.3750.250
Black olives1.0500.5062.5000.125
Green1.2740.6292.8750.375
Earthy1.6270.6253.3750.625
Floral2.2320.7394.0000.625
Woody1.5250.5252.7500.625
Spicy2.7290.6354.3751.500
Roasted2.5330.5193.7501.625
Table A2. Average, standard deviation, minimum, and maximum values of the relative concentrations of all identified volatile compounds in the entire sample dataset. The relative concentration was reported in μg/g of fresh matter.
Table A2. Average, standard deviation, minimum, and maximum values of the relative concentrations of all identified volatile compounds in the entire sample dataset. The relative concentration was reported in μg/g of fresh matter.
Volatile CompoundMEANSTDEVMAXMIN
Acetaldehyde0.3410.0990.5500.146
Dimethyl sulfide0.0920.0240.1450.048
2-methylpropanal0.3190.0740.4890.166
Ethyl acetate29.28819.12882.0806.820
2-methylbutanal 0.4670.1360.8380.255
3-methylbutanal 1.4540.3692.4370.927
Ethanol3.0321.9409.7980.499
2,2,4,6,6-pentamethylheptane 20.8817.42141.7527.318
Butane-2,3-dione 5.0611.0387.7463.044
Decane0.2290.1400.6020.017
Toluene0.7940.7632.9000.030
Isoamyl acetate1.3250.8256.2240.558
2-methylpropan-1-ol1.3040.8246.3000.605
Pentan-2-ol2.0811.0985.1840.805
3-methylbutan-1-ol 0.7450.3941.8720.213
Acetoin15.7473.98425.3358.412
Acetol7.9882.16013.7074.305
2,3-dimethylpyrazine 0.6870.3561.7970.178
Acetoin acetate2.8301.0635.5501.183
2,3,5-trimethylpyrazine 2.6881.2295.7240.857
Acetic acid297.58467.181416.201177.745
2,3,5,6-tetramethylpyrazine 27.51717.19492.4715.735
Benzaldehyde1.7670.7973.9130.595
Propylene glycol diacetate1.3090.9504.2430.192
2,3-butanediol (isomere A)37.81410.78455.34617.717
2,3-butanediol (isomere B)12.6794.40221.4775.786
Butyrolactone1.4560.2142.1121.115
2-phenylacetaldehyde0.5680.3571.9090.145
Acetophenone0.5000.2421.2920.155
1,3-Diacetoxypropane4.8287.71834.4680.098
Furfuryl alcohol0.1660.0380.2480.108
Diethyl butanedioate 0.1560.1530.7140.004
Unidentified Compound10.47015.05164.0910.331
2-phenylethanol 4.9582.09710.2071.677
2-Acetylpyrrole0.4510.2091.0330.165
Figure A1. Scree plot of the PCA performed on the mean sensory scores obtained for all sensory attributes; linked to Figure 2.
Figure A1. Scree plot of the PCA performed on the mean sensory scores obtained for all sensory attributes; linked to Figure 2.
Molecules 28 03805 g0a1
Figure A2. Scree plot of the PCA performed on the relative concentrations of the identified volatile compounds; linked to Figure 3.
Figure A2. Scree plot of the PCA performed on the relative concentrations of the identified volatile compounds; linked to Figure 3.
Molecules 28 03805 g0a2
Figure A3. Bar chart of the quality of the PLS regression as a function of the number of components. Q2 cumulated index stands for the global goodness of fit; R2Y and R2X cum stand for the correlations existing between the components and the explanatory (X, relative concentrations of volatile compounds) and dependent (Y, mean sensory scores) variables.
Figure A3. Bar chart of the quality of the PLS regression as a function of the number of components. Q2 cumulated index stands for the global goodness of fit; R2Y and R2X cum stand for the correlations existing between the components and the explanatory (X, relative concentrations of volatile compounds) and dependent (Y, mean sensory scores) variables.
Molecules 28 03805 g0a3
Figure A4. Correlation map generated by the PLS regression between the components, explanatory variables, and response variables, along the t1 and t3 axes, with the chocolate samples superimposed.
Figure A4. Correlation map generated by the PLS regression between the components, explanatory variables, and response variables, along the t1 and t3 axes, with the chocolate samples superimposed.
Molecules 28 03805 g0a4

References

  1. SQ. Global Chocolate Market to Worth 65.49 Billion. SkyQuest Technology Consulting Pvt. Ltd. 2022. Available online: https://www.globenewswire.com/en/news-release/2022/11/17/2558106/0/en/Global-Chocolate-Market-to-Worth-65-49-Billion-Global-Cocoa-Production-is-Pegged-at-4-9-Million-Tons-Europe-Consumes-Over-628-000-Tons-of-Cocoa.html (accessed on 2 March 2023).
  2. ICCO. Fine or Flavour Cocoa. 2023. Available online: https://www.icco.org/fine-or-flavor-cocoa/ (accessed on 27 February 2023).
  3. Reed, S. Sensory Analysis of Chocolate Liquor. Manuf. Confect. 2010, 90, 43–52. [Google Scholar]
  4. Afoakwa, E.O.; Paterson, A.; Fowler, M.; Ryan, A. Flavor Formation and Character in Cocoa and Chocolate: A Critical Review. Crit. Rev. Food Sci. Nutr. 2008, 48, 840–857. [Google Scholar] [CrossRef] [PubMed]
  5. Giacometti, J.; Jolić, S.M.; Josić, D. Cocoa Processing and Impact on Composition. In Processing and Impact on Active Components in Food; Elsevier: London, UK; Waltham, MA, USA; San Diego, CA, USA, 2015; pp. 605–612. [Google Scholar]
  6. Aprotosoaie, A.C.; Luca, S.V.; Miron, A. Flavor Chemistry of Cocoa and Cocoa Products-An Overview. Compr. Rev. Food Sci. Food Saf. 2015, 15, 73–91. [Google Scholar] [CrossRef] [PubMed]
  7. Kongor, J.E.; Hinneh, M.; Van de Walle, D.; Afoakwa, E.O.; Boeckx, P.; Dewettinck, K. Factors influencing quality variation in cocoa (Theobroma cacao) bean flavour profile—A review. Food Res. Int. 2016, 82, 44–52. [Google Scholar] [CrossRef]
  8. Noor-Soffalina, S.S.; Jinap, S.; Nazamid, S.; Nazimah, S.A.H. Effect of polyphenol and pH on cocoa Maillard-related flavour precursors in a lipidic model system. Int. J. Food Sci. Technol. 2009, 44, 168–180. [Google Scholar] [CrossRef]
  9. Jinap, S.; Rosli, W.I.W.; Russly, A.R.; Nordin, L.M. Effect of roasting time and temperature on volatile component profiles during nib roasting of cocoa beans (Theobroma cacao). J. Sci. Food Agric. 1998, 77, 441–448. [Google Scholar]
  10. Rodriguez-Campos, J.; Escalona-Buendía, H.; Contreras-Ramos, S.; Orozco-Avila, I.; Jaramillo-Flores, E.; Lugo-Cervantes, E. Effect of fermentation time and drying temperature on volatile compounds in cocoa. Food Chem. 2012, 132, 277–288. [Google Scholar] [CrossRef]
  11. Ooi, T.S.; Ting, A.S.Y.; Siow, L.F. Volatile organic compounds and sensory profile of dark chocolates made with cocoa beans fermented with Pichia kudriavzevii and Hanseniaspora thailandica. J. Food Sci. Technol. 2021, 59, 2714–2723. [Google Scholar] [CrossRef]
  12. Frauendorfer, F.; Schieberle, P. Key aroma compounds in fermented Forastero cocoa beans and changes induced by roasting. Eur. Food Res. Technol. 2019, 245, 1907–1915. [Google Scholar] [CrossRef]
  13. Chetschik, I.; Pedan, V.; Chatelain, K.; Kneubühl, M.; Hühn, T. Characterization of the Flavor Properties of Dark Chocolates Produced by a Novel Technological Approach and Comparison with Traditionally Produced Dark Chocolates. J. Agric. Food Chem. 2019, 67, 3991–4001. [Google Scholar] [CrossRef]
  14. Rodriguez-Campos, J.; Escalona-Buendía, H.; Orozco-Avila, I.; Lugo-Cervantes, E.; Jaramillo-Flores, M. Dynamics of volatile and non-volatile compounds in cocoa (Theobroma cacao L.) during fermentation and drying processes using principal components analysis. Food Res. Int. 2011, 44, 250–258. [Google Scholar] [CrossRef]
  15. Cevallos-Cevallos, J.M.; Gysel, L.; Maridueña-Zavala, M.G.; Molina-Miranda, M.J. Time-Related Changes in Volatile Compounds during Fermentation of Bulk and Fine-Flavor Cocoa (Theobroma cacao) Beans. J. Food Qual. 2018, 2018, 1758381. [Google Scholar] [CrossRef]
  16. Rottiers, H.; Sosa, D.A.T.; De Winne, A.; Ruales, J.; De Clippeleer, J.; De Leersnyder, I.; De Wever, J.; Everaert, H.; Messens, K.; Dewettinck, K. Dynamics of volatile compounds and flavor precursors during spontaneous fermentation of fine flavor Trinitario cocoa beans. Eur. Food Res. Technol. 2019, 245, 1917–1937. [Google Scholar] [CrossRef]
  17. Bastos, V.S.; Uekane, T.M.; Bello, N.A.; de Rezende, C.M.; Paschoalin, V.M.F.; Del Aguila, E.M. Dynamics of volatile compounds in TSH 565 cocoa clone fermentation and their role on chocolate flavor in Southeast Brazil. J. Food Sci. Technol. 2019, 56, 2874–2887. [Google Scholar] [CrossRef]
  18. Qin, X.W.; Lai, J.X.; Tan, L.H.; Hao, C.Y.; Li, F.P.; He, S.Z.; Song, Y.H. Characterization of volatile compounds in Criollo, Forastero, and Trinitario cocoa seeds (Theobroma cacao L.) in China. Int. J. Food Prop. 2017, 20, 2261–2275. [Google Scholar] [CrossRef]
  19. Tuenter, E.; Delbaere, C.; De Winne, A.; Bijttebier, S.; Custers, D.; Foubert, K.; Van Durme, J.; Messens, K.; Dewettinck, K.; Pieters, L. Non-volatile and volatile composition of West African bulk and Ecuadorian fine-flavor cocoa liquor and chocolate. Food Res. Int. 2019, 130, 108943. [Google Scholar] [CrossRef]
  20. Moreira, I.M.D.V.; Vilela, L.D.F.; Santos, C.; Lima, N.; Schwan, R.F. Volatile compounds and protein profiles analyses of fermented cocoa beans and chocolates from different hybrids cultivated in Brazil. Food Res. Int. 2018, 109, 196–203. [Google Scholar] [CrossRef]
  21. Utrilla-Vázquez, M.; Rodríguez-Campos, J.; Avendaño-Arazate, C.H.; Gschaedler, A.; Lugo-Cervantes, E. Analysis of volatile compounds of five varieties of Maya cocoa during fermentation and drying processes by Venn diagram and PCA. Food Res. Int. 2020, 129, 108834. [Google Scholar] [CrossRef]
  22. Tran, P.D.; Van de Walle, D.; De Clercq, N.; De Winne, A.; Kadow, D.; Lieberei, R.; Messens, K.; Tran, D.N.; Dewettinck, K.; Van Durme, J. Assessing cocoa aroma quality by multiple analytical approaches. Food Res. Int. 2015, 77, 657–669. [Google Scholar] [CrossRef]
  23. Escobar, S.; Santander, M.; Zuluaga, M.; Chacón, I.; Rodríguez, J.; Vaillant, F. Fine cocoa beans production: Tracking aroma precursors through a comprehensive analysis of flavor attributes formation. Food Chem. 2021, 365, 130627. [Google Scholar] [CrossRef]
  24. Magagna, F.; Guglielmetti, A.; Liberto, E.; Reichenbach, S.E.; Allegrucci, E.; Gobino, G.; Bicchi, C.; Cordero, C. Comprehensive Chemical Fingerprinting of High-Quality Cocoa at Early Stages of Processing: Effectiveness of Combined Untargeted and Targeted Approaches for Classification and Discrimination. J. Agric. Food Chem. 2017, 65, 6329–6341. [Google Scholar] [CrossRef] [PubMed]
  25. Barbosa-Pereira, L.; Rojo-Poveda, O.; Ferrocino, I.; Giordano, M.; Zeppa, G. Assessment of volatile fingerprint by HS-SPME/GC-qMS and E-nose for the classification of cocoa bean shells using chemometrics. Food Res. Int. 2019, 123, 684–696. [Google Scholar] [CrossRef] [PubMed]
  26. Schlüter, A.; Hühn, T.; Kneubühl, M.; Chatelain, K.; Rohn, S.; Chetschik, I. Comparison of the Aroma Composition and Sensory Properties of Dark Chocolates Made with Moist Incubated and Fermented Cocoa Beans. J. Agric. Food Chem. 2022, 70, 4057–4065. [Google Scholar] [CrossRef] [PubMed]
  27. Kadow, D.; Bohlmann, J.; Phillips, W.; Lieberei, R. Identification of main fine or flavour components in two genotypes of the cocoa tree (Theobroma cacao L.). J. Appl. Bot. Food Qual. 2013, 86, 90–98. [Google Scholar] [CrossRef]
  28. Frauendorfer, F.; Schieberle, P. Identification of the Key Aroma Compounds in Cocoa Powder Based on Molecular Sensory Correlations. J. Agric. Food Chem. 2006, 54, 5521–5529. [Google Scholar] [CrossRef]
  29. Liu, J.; Liu, M.; He, C.; Song, H.; Guo, J.; Wang, Y.; Yang, H.; Su, X. A comparative study of aroma-active compounds between dark and milk chocolate: Relationship to sensory perception. J. Sci. Food Agric. 2014, 95, 1362–1372. [Google Scholar] [CrossRef]
  30. Ullrich, L.; Casty, B.; André, A.; Hühn, T.; Steinhaus, M.; Chetschik, I. Decoding the Fine Flavor Properties of Dark Chocolates. J. Agric. Food Chem. 2022, 70, 13730–13740. [Google Scholar] [CrossRef]
  31. Owusu, M.; Petersen, M.A.; Heimdal, H. Relationship of sensory and instrumental aroma measurements of dark chocolate as influenced by fermentation method, roasting and conching conditions. J. Food Sci. Technol. 2013, 50, 909–917. [Google Scholar] [CrossRef]
  32. Waehrens, S.S.; Zhang, S.; Hedelund, P.I.; Petersen, M.A.; Byrne, D.V. Application of the fast sensory method ‘Rate-All-That-Apply’ in chocolate Quality Control compared with DHS-GC-MS. Int. J. Food Sci. Technol. 2016, 51, 1877–1887. [Google Scholar] [CrossRef]
  33. Rottiers, H.; Sosa, D.A.T.; Lemarcq, V.; De Winne, A.; De Wever, J.; Everaert, H.; Jaime, J.A.B.; Dewettinck, K.; Messens, K. A multipronged flavor comparison of Ecuadorian CCN51 and Nacional cocoa cultivars. Eur. Food Res. Technol. 2019, 245, 2459–2478. [Google Scholar] [CrossRef]
  34. Deuscher, Z.; Andriot, I.; Sémon, E.; Repoux, M.; Preys, S.; Roger, J.M.; Boulanger, R.; Labouré, H.; Le Quéré, J.L. Volatile compounds profiling by using proton transfer reaction-time of flight-mass spectrometry (PTR-ToF-MS). The case study of dark chocolates organoleptic differences. J. Mass Spectrom. 2019, 54, 92–119. [Google Scholar] [CrossRef]
  35. Afoakwa, E.O.; Paterson, A.; Fowler, M.; Ryan, A. Matrix effects on flavour volatiles release in dark chocolates varying in particle size distribution and fat content using GC–mass spectrometry and GC–olfactometry. Food Chem. 2009, 113, 208–215. [Google Scholar] [CrossRef]
  36. National Center for Biotechnology Information (NCBI). PubChem. Compound Summary. 2022. Available online: https://pubchem.ncbi.nlm.nih.gov/ (accessed on 21 March 2023).
  37. Indraprastha Institute of Information Technology Delhi (IIIT-Delhi). FlavorDB. FlavorDB Search—A Resource for Exploring Flavor Molecules. 2017. Available online: https://cosylab.iiitd.edu.in/flavordb/search (accessed on 20 March 2023).
  38. Bonvehí, J.S. Investigation of aromatic compounds in roasted cocoa powder. Eur. Food Res. Technol. 2005, 221, 19–29. [Google Scholar] [CrossRef]
  39. Machado Cuellar, L.; Ordoñez Espinosa, C.M.; Angel Sánchez, Y.K.; Guaca Cruz, L.; Suárez Salazar, J.C. Organoleptic quality assessment of Theobroma cacao L. in cocoa farms in northern Huila, Colombia. Acta Agron. 2018, 67, 46–52. [Google Scholar] [CrossRef]
  40. Grillo, G.; Boffa, L.; Binello, A.; Mantegna, S.; Cravotto, G.; Chemat, F.; Dizhbite, T.; Lauberte, L.; Telysheva, G. Analytical dataset of Ecuadorian cocoa shells and beans. Data Brief 2018, 22, 56–64. [Google Scholar] [CrossRef]
  41. Pieracci, Y.; Ascrizzi, R.; Pistelli, L.; Flamini, G. Comparison of the Chemical and Sensorial Evaluation of Dark Chocolate Bars. Appl. Sci. 2021, 11, 9964. [Google Scholar] [CrossRef]
  42. Kouassi, A.D.D.; Koné, K.M.; Assi-Clair, B.J.; Lebrun, M.; Maraval, I.; Boulanger, R.; Fontana, A.; Guehi, T.S. Effect of spontaneous fermentation location on the fingerprint of volatile compound precursors of cocoa and the sensory perceptions of the end-chocolate. J. Food Sci. Technol. 2022, 59, 4466–4478. [Google Scholar] [CrossRef]
  43. Colonges, K.; Solorzano, R.G.L.; Jimenez, J.; Lahon, M.; Seguine, E.; Calderon, D.; Subia, C.; Sotomayor, I.; Fernández, F.; Lebrun, M.; et al. Variability and genetic determinants of cocoa aromas in trees native to South Ecuadorian Amazonia. Plants People Planet 2022, 4, 618–637. [Google Scholar] [CrossRef]
  44. Calva-Estrada, S.; Utrilla-Vázquez, M.; Vallejo-Cardona, A.; Roblero-Pérez, D.; Lugo-Cervantes, E. Thermal properties and volatile compounds profile of commercial dark-chocolates from different genotypes of cocoa beans (Theobroma cacao L.) from Latin America. Food Res. Int. 2020, 136, 109594. [Google Scholar] [CrossRef]
  45. Michel, S.; Baraka, L.F.; Ibañez, A.J.; Mansurova, M. Mass Spectrometry-Based Flavor Monitoring of Peruvian Chocolate Fabrication Process. Metabolites 2021, 11, 71. [Google Scholar] [CrossRef]
  46. Deuscher, Z.; Gourrat, K.; Repoux, M.; Boulanger, R.; Labouré, H.; Le Quéré, J.-L. Key Aroma Compounds of Dark Chocolates Differing in Organoleptic Properties: A GC-O Comparative Study. Molecules 2020, 25, 1809. [Google Scholar] [CrossRef] [PubMed]
  47. Greenland, S.; Pearl, J.; Robins, J.M. Confounding and Collapsibility in Causal Inference. Stat. Sci. 1999, 14, 29–46. [Google Scholar] [CrossRef]
  48. Januszewska, R. Hidden Persuaders in Cocoa and Chocolate: A Flavour Lexicon for Cocoa and Chocolate Sensory Professionals; Elsevier Science: Amsterdam, The Netherlands, 2018. [Google Scholar]
  49. Bonvehi, J.S.; Coll, F.V. Evaluation of bitterness and astringency of polyphenolic compounds in cocoa powder. Food Chem. 1997, 60, 365–370. [Google Scholar] [CrossRef]
  50. Bonvehí, J.S.; Coll, F.V. Parameters affecting the quality of processed cocoa powder: Acidity fraction. Z. Lebensm.-Forsch. A 1997, 204, 287–292. [Google Scholar] [CrossRef]
  51. De Vuyst, L.; Leroy, F. Functional role of yeasts, lactic acid bacteria and acetic acid bacteria in cocoa fermentation processes. FEMS Microbiol. Rev. 2020, 44, 432–453. [Google Scholar] [CrossRef]
  52. Osborne, J.; de Orduña, R.M.; Pilone, G.; Liu, S.-Q. Acetaldehyde metabolism by wine lactic acid bacteria. FEMS Microbiol. Lett. 2000, 191, 51–55. [Google Scholar] [CrossRef]
  53. Garcia, L.; Perrin, C.; Nolleau, V.; Godet, T.; Farines, V.; Garcia, F.; Caillé, S.; Saucier, C. Impact of Acetaldehyde Addition on the Sensory Perception of Syrah Red Wines. Foods 2022, 11, 1693. [Google Scholar] [CrossRef]
  54. Fayeulle, N.; Vallverdu-Queralt, A.; Meudec, E.; Hue, C.; Boulanger, R.; Cheynier, V.; Sommerer, N. Characterization of new flavan-3-ol derivatives in fermented cocoa beans. Food Chem. 2018, 259, 207–212. [Google Scholar] [CrossRef]
  55. Lytra, G.; Tempere, S.; Zhang, S.; Marchand, S.; De Revel, G.; Barbe, J.-C. Olfactory impact of dimethyl sulfide on red wine fruity esters aroma expression in model solution. OENO One 2014, 48, 75. [Google Scholar] [CrossRef]
  56. Ascrizzi, R.; Flamini, G.; Tessieri, C.; Pistelli, L. From the raw seed to chocolate: Volatile profile of Blanco de Criollo in different phases of the processing chain. Microchem. J. 2017, 133, 474–479. [Google Scholar] [CrossRef]
  57. Assi-Clair, B.J.; Koné, M.K.; Kouamé, K.; Lahon, M.C.; Berthiot, L.; Durand, N.; Lebrun, M.; Julien-Ortiz, A.; Maraval, I.; Boulanger, R.; et al. Effect of aroma potential of Saccharomyces cerevisiae fermentation on the volatile profile of raw cocoa and sensory attributes of chocolate produced thereof. Eur. Food Res. Technol. 2019, 245, 1459–1471. [Google Scholar] [CrossRef]
Figure 1. Clustered heatmap of all samples (rows) in relation to their mean sensory scores per attribute (columns). The score scale ranges from 1 to 10; the highest granted mean sensory score was 6.
Figure 1. Clustered heatmap of all samples (rows) in relation to their mean sensory scores per attribute (columns). The score scale ranges from 1 to 10; the highest granted mean sensory score was 6.
Molecules 28 03805 g001
Figure 2. F1−F2 biplot of PCA on mean sensory scores showing distribution of samples (observations, in blue), and of sensory attributes (variables, in red).
Figure 2. F1−F2 biplot of PCA on mean sensory scores showing distribution of samples (observations, in blue), and of sensory attributes (variables, in red).
Molecules 28 03805 g002
Figure 3. Biplots of the PCA on the relative concentration of the identified volatile compounds, showing distribution of samples (observations), and of volatile compounds (variables): (a) F1−F2 biplot; (b) F1−F3 biplot.
Figure 3. Biplots of the PCA on the relative concentration of the identified volatile compounds, showing distribution of samples (observations), and of volatile compounds (variables): (a) F1−F2 biplot; (b) F1−F3 biplot.
Molecules 28 03805 g003
Figure 4. Correlation map generated by PLS regression between the components, explanatory variables, and response variables, along the t1 and t2 axes, with the chocolate samples superimposed.
Figure 4. Correlation map generated by PLS regression between the components, explanatory variables, and response variables, along the t1 and t2 axes, with the chocolate samples superimposed.
Molecules 28 03805 g004
Table 1. Volatile compounds identified in the 54 dark chocolate samples, along with their associated flavor descriptors.
Table 1. Volatile compounds identified in the 54 dark chocolate samples, along with their associated flavor descriptors.
Compound GroupCompound NameOdor/Flavor Attributes
Aldehydes2-methylbutanal Chocolate [35]
2-methylpropanalChocolate [35]
2-phenylacetaldehydeBerry, geranium, honey [36]
3-methylbutanal Chocolate [35]
AcetaldehydeTart (acidic), pungent fruity [36]
BenzaldehydeNutty, almond [35]
Esters1,3-diacetoxypropaneAcetic, fruit [37]
Diethyl butanedioate Cotton, fabric, floral, fruit [36]
Ethyl acetatePineapple [14]
Isoamyl acetateFruity, banana [36]
Propylene glycol diacetateFruit [36]
Alcohols and Phenols3-methylbutan-1-ol Pungent, repulsive taste [36]
2,3-butanediol (isomere A)Sweet [36]
2,3-butanediol (isomere B)Sweet [36]
2-methylpropan-1-olSweet, whiskey [36]
2-phenylethanol Rose, lilac, flowery, caramel [28]
AcetoinButtery [36]
Ethanol-
Furfuryl alcoholBitter [36]
Pentan-2-olFuel oil, green [36]
KetonesAcetoin acetateFruit [36]
AcetolPungent, sweet, caramel-like [36]
AcetophenoneMust, flower, almond, sweet [38]
Butane-2,3-dione Buttery [36]
Pyrazines2,3,5,6-tetramethylpyrazine Milk-coffee, roasted, chocolate [35]
2,3,5-trimethylpyrazine Cocoa, roasted, cooked [35]
2,3-dimethylpyrazine Cooked, nutty [35]
Other2,2,4,6,6-pentamethylheptane Unspecified
2-acetylpyrroleBread, cocoa, hazelnut, licorice, walnut [36]
Acetic acidSour, astringent, vinegar [35]
ButyrolactoneSweet, caramel-like [36]
DecaneGasoline-like [36]
Dimethyl sulfideUnpleasant, cabbage-like [36]
TolueneFuel-like [36]
Unidentified compoundUnspecified
Table 2. Training (estimated by cross-validation) and validation (tested in-loop, with the whole data set) test results of the best fitted models obtained for predicting samples classified as “above median”.
Table 2. Training (estimated by cross-validation) and validation (tested in-loop, with the whole data set) test results of the best fitted models obtained for predicting samples classified as “above median”.
Training Dataset Metrics Validated Model Performance for Prediction
Attribute ModelNumber of ComponentsROCSensitivitySpecificityAccuracy95% CISensitivitySpecificity
Bitterness40.85080.81170.73000.7963(0.6647, 0.8937)0.77140.8421
Astringency30.92630.80830.86080.7963(0.6647, 0.8937)0.82140.7692
Citrus10.81460.66000.72830.7222(0.5836, 0.8354)0.76920.6786
Acidity30.76210.68170.70080.7037(0.5639, 0.8202)0.75000.6667
Red fruits30.85790.70580.78670.7037(0.5639, 0.8202)0.74070.6667
Dried fruits20.78460.62830.82500.6852(0.5445, 0.8048)0.72730.6562
Green 20.81920.82250.67920.6852(0.5445, 0.8048)0.71430.6316
Black olives 10.65380.73420.59330.6667(0.5253, 0.7891)0.65620.6818
Woody 40.80380.75500.75000.6667(0.5253, 0.7891)0.60710.7308
Sweetness30.77670.70420.75500.6667(0.5253, 0.7891)0.66670.6667
Table 3. Compilation of the five variables with the highest relative importance in building the above-median-models, together with their variable importance in projection (VIP) and Pearson correlation coefficient.
Table 3. Compilation of the five variables with the highest relative importance in building the above-median-models, together with their variable importance in projection (VIP) and Pearson correlation coefficient.
Attribute ModelVariableVIPCorr.
Coeff.
Attribute ModelVariableVIPCorr.
Coeff.
BitternessAcetaldehyde100.00−0.782Dried fruitsEthyl Acetate100.00−0.554
Dimethyl sulfide64.980.5943-methylbutan-1-ol 96.36−0.566
2,3-butanediol (isomere A)49.53−0.694Ethanol91.13−0.439
3-methylbutan-1-ol 41.430.598Acetaldehyde89.820.505
Acetic acid37.21−0.636Pentan-2-ol87.1−0.553
AstringencyAcetaldehyde100.00−0.749Green Acetaldehyde100.00−0.571
3-methylbutanal 79.16−0.632Diethyl butanedioate 86.360.344
2,3-butanediol (isomere A)75.10−0.622Dimethyl sulfide82.360.529
2-methylpropanal72.22−0.4873-methylbutanal 77.51−0.589
Dimethyl sulfide59.420.531Butane-2,3-dione 60.84−0.411
Citrus2,3-butanediol (isomere B)100.000.596Black olives Dimethyl sulfide100.000.530
Acetic acid96.230.5962,3-butanediol (isomere A)92.87−0.591
3-methylbutan-1-ol 91.39−0.564Acetic acid88.45−0.486
Ethyl Acetate88.51−0.503Pentan-2-ol75.660.310
Pentan-2-ol87.34−0.5322,3-butanediol (isomere B)71.44−0.340
AcidityAcetaldehyde100.000.603Woody 3-acetyloxypropyl acetate 100.000.151
Butyrolactone73.300.451Unknown compound98.140.248
Isoamyl acetate65.340.179Acetaldehyde91.19−0.561
2,3-butanediol (isomere A)64.380.5572,3-butanediol (isomere A)84.28−0.626
2,3-butanediol (isomere B)64.240.567Butyrolactone78.3−0.257
Red fruitsAcetaldehyde100.000.628SweetnessAcetaldehyde100.000.605
Propylene glycol diacetate83.080.665Acetol89.51−0.042
2,3-butanediol (isomere A)57.040.607Dimethyl sulfide73.49−0.557
Dimethyl sulfide51.26−0.5062,3-butanediol (isomere A)67.460.582
2,3-butanediol (isomere B)47.270.5343-methylbutanal 59.840.512
Table 4. Validation test results of the models obtained for predicting “high” sensory scores for bitterness and astringency, along with the five most important variables, their VIP, and Pearson correlation coefficients.
Table 4. Validation test results of the models obtained for predicting “high” sensory scores for bitterness and astringency, along with the five most important variables, their VIP, and Pearson correlation coefficients.
Looped Tested Model Performance
Attribute ModelAccuracy95% CISensitivitySpecificityVariableVIPCorr. Coeff.
Bitterness0.7963(0.6647, 0.8937)0.89190.5882Acetaldehyde100.00−0.782
2,3-butanediol (isomere A)67.29−0.694
3-methylbutan-1-ol 65.250.598
2,3-butanediol (isomere B)62.41−0.548
Dimethyl sulfide57.280.594
Astringency0.8148(0.6857, 0.9075)0.89740.60002,3-butanediol (isomere A)100.00−0.622
Acetaldehyde98.52−0.749
Acetic acid97.44−0.519
2,3-butanediol (isomere B)88.46−0.378
3-methylbutan-1-ol 86.760.397
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Guzmán Penella, S.; Boulanger, R.; Maraval, I.; Kopp, G.; Corno, M.; Fontez, B.; Fontana, A. Link between Flavor Perception and Volatile Compound Composition of Dark Chocolates Derived from Trinitario Cocoa Beans from Dominican Republic. Molecules 2023, 28, 3805. https://doi.org/10.3390/molecules28093805

AMA Style

Guzmán Penella S, Boulanger R, Maraval I, Kopp G, Corno M, Fontez B, Fontana A. Link between Flavor Perception and Volatile Compound Composition of Dark Chocolates Derived from Trinitario Cocoa Beans from Dominican Republic. Molecules. 2023; 28(9):3805. https://doi.org/10.3390/molecules28093805

Chicago/Turabian Style

Guzmán Penella, Santiago, Renaud Boulanger, Isabelle Maraval, Gabi Kopp, Marcello Corno, Bénédicte Fontez, and Angélique Fontana. 2023. "Link between Flavor Perception and Volatile Compound Composition of Dark Chocolates Derived from Trinitario Cocoa Beans from Dominican Republic" Molecules 28, no. 9: 3805. https://doi.org/10.3390/molecules28093805

APA Style

Guzmán Penella, S., Boulanger, R., Maraval, I., Kopp, G., Corno, M., Fontez, B., & Fontana, A. (2023). Link between Flavor Perception and Volatile Compound Composition of Dark Chocolates Derived from Trinitario Cocoa Beans from Dominican Republic. Molecules, 28(9), 3805. https://doi.org/10.3390/molecules28093805

Article Metrics

Back to TopTop