Study area

The cocoa beans were selected from mature pods from 18 cocoa farms in three municipalities in the northern region of the department of Huila (Colombia): Algeciras, Campoalegre and Rivera. Each of these zones has different climatic characteristics as follows: Algeciras (2° 31> 22.58" N and 75° 18' 57.42" W), with an altitude between 800 to 3000 m.a.s.l., an average temperature of 22 °C, and an average annual rainfall of 1,500 mm; Campoalegre (2° 41' 7.58'' N and 75° 19' 35.91' W), has an altitude between 525 and 1400 m.a.s.l., an average temperature of 27 °C, and an average annual rainfall of900 mm; and the municipality of Rivera (2° 46' 42.15" N and 75° 15' 34.93" W), has an altitude ranged from 500 to 1000 m.a.s.l., an average temperature of 27 °C, and an average annual rainfall of 2500 mm.

Bromatological analysis

The following variables were established in triplicate for each cocoa sample: moisture (using the AOAC 931.04 18^th and gravimetric technique), fat (extracted by the Soxhlet method, AOAC 963.15), pH (using a potentiometer, AOAC 970.2118) and acidity (through titration).

Secondary metabolites and total polyphenols

For theobromine and caffeine quantification a high resolution liquid chromatograph (Thermo Fisher Scientific™ UltiMate™ 3000) equipped with a reverse phase column (Agilent ZORBAX Eclipse XDB 150 x 2.1 mm), and calibrated at 273 nm was used. As the mobile phase, methanol (A) with water (B) was used in a ratio of 40:60. The mobile phase flow was 0.7 mL. min^-1. Identification of peaks was made by comparing theobromine and caffeine with HPLC standards. Moreover, for samples under optimal conditions and to allow polyphenols extraction, these were degreased by the Soxhler method with n-Hexane (AOAC 7.060/84) and analyzed with the Folin-Ciocalteu spectrophotometric method using gallic acid as reference material.

Volatile fraction compounds

To determine volatile fraction compounds, a gas chromatography by solid phase microextraction in headspace mode (HS-GC-MS) analysis of the methylated fraction was performed through gas chromatograph (GC 7890 Agilent Technologies) with a flame ionization detector (GC-FID). A capillary column AgilentHP-5 (5 % polydiphenyl, 95 % dimethylsiloxane), 30 m long and 320 μπι internal diameter and 0.25 μπι film thickness in SpIit 3:1 mode was used. Moreover, linalool, 2,3-butanediol, 2,5-dimethylpyrazine, 2-phenethylacetate, acetophenone, 2,3,5-trimethylpyrazine, ethylbenzoate, phenethyl alcohol, acetylpyrrole, tetramethylpyrazine, benzaldehyde, phenylacetaldehyde (Sigma-aldrich), and acetic acid (Merck) were used as reference standards. Concentration results are presented as relative area percentage.

Sensory analysis

To establish sensory properties 3 kg of dry cocoa beans were used, processed according to postharvest conditions (fermentation and drying) established by each producer. Each sample was labeled and sent to the Food Laboratory of CICTA (NTC ISO / IEC 17025: 2005) at Universidad Industrial de Santander, to establish a description of flavors: basic (acidity, bitterness, astringency, salty), specific (cocoa, floral, fruity and walnut) and acquired (green).

Then, a tasting panel established a sensory profile with the average evaluation scores based on a quantitative assessment scale according to their rating (0 = absent, 1-2 = low, 3-5 = medium, 6-8 = high, 9-10 = very high, strong). Likewise, from the conclusions established by the panel, the results of the sample quality were labeled as acceptable, unpleasant and contaminated.

Statistic analysis

For each of the chemical compounds identified and sensory attributes found in cocoa liquor, descriptive statistics analysis (means and frequencies of variables) and an analysis of variance were performed using a Tukey test (P < .05). For each data set (i. bromatological analysis, secondary metabolites and total polyphenols, ii. volatile fraction compounds, and iii. sensory attributes) a principal components analysis (PCA) was performed to determine similarity between the classifications established by a sensory panel, and also explores the relationship among variables of the same data set. Moreover, to explore the relationships between data sets a PCA was used; then the matrices were connected through a co-inertia analysis, which is used as a sorting technique that correlates the first ordering axes of two matrices; in addition, to establish the significance of co-inertia values a Monte Carlo test was performed. The combinations generated between matrices were: i. Bromatological analysis, secondary metabolites and total polyphenols, with sensory attributes. ii. Volatile fraction compounds with sensory attributes, and iii. Bromatological analysis, secondary metabolites and total polyphenols, with volatile fraction compounds. Finally, multivariate analyzes (PCA and co-inercia) were generated using the Ade4 package (^{Dray, Dufour & Chessel, 2007}) and using the R software version 3.4.0 (^{R Development Core Team, 2017}) through the statistical software InfoStat interface (^{Di Rienzo, Casanoves, Balzarini, Gonzalez, Tablada & Robledo, 2017}).

Results

Bromatological analysis, secondary metabolites and total polyphenols

Statistical differences were found for pH and acidity (P < .05) among cocoa quality classifications, showing a negative correlation between these variables (r= -.95, P < .0001). Moreover, the moisture content showed a range of 5.77 ± 0.15 %, the fat content ranged between 52.9 ± 0.42 and theobromine values were close to 12 mg.g^-1 (Table 1).

Table 1 Bromatological analysis, secondary metabolites and total polyphenols found in cocoa samples 

E.E.: Experimental error

Volatile fraction compounds

Fourteen compounds were identified with their concentration as a percentage of the relative area (Table 2). The most representative volatile fraction compounds were acetic acid, tetramethylpyrazine and 2,3-butanediol.

Table 2 Volatile compounds present in the aromatic fraction of cocoa samples 

E.E.: Experimental error

Sensory attributes

Principal components analysis (PCA) for sensory attributes explains 85.4 % of the total data variability. Axis 1 expresses the relationship between sensory attributes with the factor of acceptance and rejection of the sample (Figure 1, 2 and 3). Axis 2 contrasts a sensory attribute related to unpleasant classification (salty). The results show a high effect of the sensory attributes on the classification of the sample (p-value: .001, Figure 1).

Figure 1 PCA projection in the factorial plane F1/F2 of sensory attribute variables and the classification carried out by a tasting panel, fa) Senso ry attributes; fb) Classification sorting included in the sample within the factorial plane of a PCA. 

Figure 2 PCA projection in the factorial plane F1/F2 of the bromatological analysis, secondary metabolites and total polyphenols variables and the clas sification carried out by a tasting panel, (a) Bromatological analysis, secondary metabolites and total polyphenols; (b) Classification sorting included in the sample within the factorial plane of a PCA. 

Figure 3 PCA projection in the factorial plane F1/F2 of volatile fraction compound variables resulting from the HS-SPME GC-MS analysis. (a) Vol atile fraction compounds, fb) Classification sorting included in the sample within the factorial plane of a PCA. 

Relationship among data sets

According to the co-inercia analysis only a significant covariance was found between volatile fraction compounds and variables of the bromatological analysis, secondary metabolites and total polyphenols (RV: 0.5, p-value: .001, Monte de Carlo test Figure 4 e and f).

Figure 4 Co-inercia analysis between i. bromatological analysis, secondary metabolites and total polyphenols, ii. volatile fraction compounds, and iii. sen sory attributes. a and b. Relationship between matrices of i and iii; c and d. Relationship between matrices ii and iii; e and f. Relationship between ma trices ii and i. 

Volatile fraction compounds

Fourteen compounds with olfactory activity that contributes to aroma in cocoa were identified (Table 2). The most representative compounds of these were acetic acid, tetramethylpyrazine and 2,3-butanediol, similar to those reported in cocoa liquors in different countries (^{Liu et al., 2017}).

Moreover, other compounds that were identi fied in samples analyzed were observed in lower amounts, but have been previously reported as characteristic for cocoa (^{Aprotosoaie, Luca & Miron, 2016}).

Furthermore, high percentage of relative area found for acetic acid that was generated by mi crobial activity during the fermentation process of the beans must be highlighted, since the su crose of the pulp was hydrolyzed to glucose and fructose by the action of invertases from yeasts; this in turn generates an increase in acetic acid concentrations (^{Chetschik, Kneubuhl, Chatelain, Schluter, Bernath & Huhn, 2017}).

Sensory attributes

Principal components analysis (PCA) results for sensory attributes explains more than 80 % of the total data variability. The first axis shows a relationship between sensory attributes with acceptance and rejection factors, and the second axis shows a relationship between sensory attributes and an unpleasant classification (Salty). By relating different sensory attributes with management practices, we found that pods collected in an immature state, generate a staining sensation in the mouth due to sugar content reduction. The green taste cataloged as a medium profile affects astringency level (^{Nazaruddin, Seng, Hassan & Said, 2006}). Likewise, unfavorable sensorial attributes such as astringency, acidity and green were caused by incomplete processing practices (^{Vera, Vallejo, Párraga, Morales, Macias & Ramos, 2014}) . Nonetheless, when drying process in beans is interrupted, chemical processes will be affected; this in turn influences the high values of acidity and bitterness found during the tasting process (^{Aprotosoaie, Luca & Miron, 2016}). However, specific flavors were also found in lower proportion (cocoa, fruit, floral) that altogether cataloged the product as medium level.

Relationship among data sets

According to the co-inercia analysis only a significant covariance was found between volatile fraction compounds and bromatological analysis, secondary metabolites and total polyphenols variables (RV: .5, p-value: .001, Monte Carlo test) (Figure 4). In this sense, a relationship was found between pH and acetic acid content as a result of the fermentation process (^{Do Carmo Brito, Chisté, da Silva Pena, Gloria & Lopes, 2017}). Likewise, a relationship between fat content and volatile compounds belonging to the functional group of terpenes (linalool and linalool oxide) were found, and were then characterized as floral (^{Afoakwa, Paterson, Fowler & Ryan, 2009}). Concentration of less volatile compounds increased, in particular pyrazines, tri- and tetramethylpyrazine as well as furans (linalool oxide) suggesting structural and rheological effects as main chocolate flavor determinants (^{Afoakwa, Paterson, Fowler & Ryan, 2008}). On the other hand, antioxidant activity found in cocoa samples was high, results that agree with other studies (^{Albertini, Schoubben, Guarnaccia, Pinelli, Della Vecchia, Ricci & Blasi, 2015}) , due to presence of phenolic compounds, especially catechins and epicatechins. Finally, special sensory attributes (fruity and floral) were found, possibly attributed to the presence of esters, which have notes of fresh-fruity odor, while the floral odor impression can be attributed to linalool and its oxides (^{Aprotosoaie, Luca & Miron, 2016}), giving importance to the cocoa samples analyzed.