Near Infra-red Characterization of Changes in Flavan-3-ol Derivatives

Article pubs.acs.org/JAFC

Near Infra-red Characterization of Changes in Flavan-3-ol Derivatives in Cocoa (Theobroma cacao L.) as a Function of Fermentation Temperature Clotilde Hue,† Pierre Brat,§ Ziya Gunata,# Ivan Samaniego,⊥ Adrien Servent,§ Gilles Morel,§ André Kapitan,§ Renaud Boulanger,§ and Fabrice Davrieux*,§ †

Valrhona SA, 8 quai du Général de Gaulle, 26600 Tain, L’Hermitage, France CIRAD, UMR Qualisud, TA 80/16, 75 Avenue JF Breton, 34398 Montpellier Cedex 5, France # Université Montpellier II, UMR Qualisud, 2 place E. Bataillon, 34095 Montpellier Cedex 5, France ⊥ INIAP, Panamericana Sur Km. 1, Sector Cutuglagua, Cantón Mejía, Pichincha, Ecuador §

ABSTRACT: Flavan-3-ols were successfully extracted from cocoa by the Fast-Prep device and analyzed by HPLC-DAD, and their identifications were confirmed by injection of authentic standards. (−)-Epicatechin was the most abundant component with an average of 9.4 mg/g dried cocoa powder. More than 700 cocoa samples were used to calibrate the NIRS. An efficient calibration model was developed to accurately determine any flavan-3-ol compound of ground dried cocoa beans (SEP = 2.33 mg/g in the case of total flavan-3-ols). This performance enabled NIRS to be used as an efficient and easy-to-use tool for estimating the level of targeted compounds. The analysis of the PLS loadings of the model and pure epicatechin spectra gave proof that NIRS was calibrated on an indirect strong correlation resulting in the changes in flavan-3-ols during fermentation and their interaction with some major components, such as proteins. Total flavan-3-ol concentration fell from an average of 33.3 mg/ g for unfermented samples to an average of 6.2 mg/g at the end of fermentation. Changes in flavan-3-ol content were dependent upon the origin and highly correlated to the fermentation level expressed as the sum of temperatures (average R2 = 0.74), a good marker of the fermentation process and of the heterogeneity of the batch. KEYWORDS: cocoa, fermentation, flavan-3-ol content, NIRS, sum of temperatures

■

INTRODUCTION World cocoa (Theobroma cacao L.) production is shared between bulk cocoa (standard quality) and fine or flavor cocoa (highly aromatic cocoa), which accounts for about 5% of production (The International Cocoa Organization). The development of fine or flavor quality traits is linked to the cocoa variety, soil, climate, crop management, and mainly postharvest processing.1,2 Indeed, fermentation is considered as a key step in obtaining the characteristic flavor and taste of chocolate,3,4 but it remains empirical most of the time.5,6 In the ever more competitive market for fine or flavor cocoa, it is of great importance for chocolate manufacturers to be able to characterize and understand how fermentation occurs to improve their cocoa quality. After harvesting, cocoa seeds are extracted from the pods7,8 and placed in wooden boxes or bags or piled up to start fermentation. Cocoa seeds are surrounded by a mucilaginous pulp, which undergoes natural alcoholic, lactic, and then acetic fermentation processes, which are considered to initiate biochemical and enzymatic changes within the bean.8 The production of acetic acid, together with the increase in temperature, generates drastic changes within the cocoa beans, such as the death of the embryo and the breakdown of the cellular compartment.9 As a result, contact between biochemical constituents is facilitated. The bean color changes from purple to brown due to oxidation and polymerization of flavan-3-ols into high molecular weight structures.3,10−12 Cocoa © 2014 American Chemical Society

aroma precursors are formed by the degradation of soluble proteins.1,13,14 Flavan-3-ols are among the major secondary metabolites because they account for up to 18% of the fat-free dry weight in unprocessed seeds. 15 Kim and Keeney15 revealed that (−)-epicatechin is the predominant flavan-3-ol (∼35%) in cocoa. (+)-Catechin and proanthocyanidins have also been found in cocoa products.12,16−19 During fermentation, epicatechin and catechin are oxidized into quinones, while proteins and flavan-3-ols are condensed.20,21 The oxidation of flavan-3-ols (catechin, epicatechin, and their derivatives) and anthocyanidin pigments causes the color of fresh beans to turn from purple to brown.22,23 Nowadays, an international quality control, called a cut-test, is based on visual observation of the color of 300 beans to estimate the level of fermentation. The main objective of this paper was to determine the fermentation level of cocoa beans through an efficient estimation of flavan-3-ol levels using near-infrared spectroscopy (NIRS). NIRS has already been used to predict total proanthocyanidins in cocoa liquors from various geographical origins24 and the (−)-epicatechin content of various fermented and dried cocoa powders.25 Received: Revised: Accepted: Published: 10136

March 16, 2014 September 15, 2014 September 26, 2014 September 26, 2014 dx.doi.org/10.1021/jf501070d | J. Agric. Food Chem. 2014, 62, 10136−10142

Journal of Agricultural and Food Chemistry

Article

The geographical origin26,27 and the manufacturing process, including fermentation,28 need to be highlighted as key factors that might greatly influence polyphenolic distribution. To develop an efficient calibration model using NIRS, a solid database representative of the chemical variability of the product29,30 and a reliable reference analysis were prerequisites. To achieve this goal, we first built a large database of samples (with different geographical origins, phenotypes, and fermentation conditions). Second, different flavan-3-ol extraction procedures were studied to choose the one allowing a quick and reliable analysis to be applied on a large cocoa sample for NIRS calibration purposes. In addition to the already known extraction techniques involving sonication or Ultra-Turax homogenization,10,17,18,31,32 a Fast-Prep (FP) device was used for the first time for cocoa flavan-3-ol extraction. With the large and robust database we generated and with the optimized extraction protocols, we set out to build a quantitative and efficient model for predicting flavan-3-ol content. The changes occurring within the beans were correlated with the sum of temperatures (ST)33 during fermentation for different cocoa origins, and the changes in flavan-3-ols monitored by NIRS were studied.

■

matrix and 1/4 ceramic sphere as the lysing matrix. Cocoa powder (100.0 ± 0.1 mg) was defatted two times for 1 min with 1 mL of hexane. Following centrifugation (5 min, 21400g, 4 °C), the upper organic phase was discarded to remove lipids. The solid phase (defatted cocoa powder) was then extracted for 1 min with 1 mL of acetone/water/formic acid (70:30:0.1, v/v/v). After centrifugation (21400g, 5 min, 4 °C), the upper phase was collected and the solid phase was re-extracted using the same conditions. The two fractions were then gathered, and the solvent was eliminated by a rotavapor (50 °C). Flavan-3-ols were thereafter extracted with methanol/water (50:50 v/v) and injected into HPLC. The analysis was carried out in duplicate. Replicates were consolidated using the Student t test at 95%, based on the standard deviation of samples. Ultra-Turax Extraction. One gram of cocoa powder was added to 20 mL of hexane and finely ground for 2 min at 2000 rpm with an Ultra-Turax T25 digital (Staufen, Germany). After centrifugation (10 min, 21400g, 4 °C), the hexane phase was discarded. The defatted powder was then extracted with the blender for 2 min with 20 mL of acetone/water/formic acid (70:30:0.1, v/v/v). The organic phase was then recovered by evaporation (same conditions as above) and injected into HPLC. Pressurized Liquid Extraction (PLE). PLEs were carried out on an ASE350 automated sample extractor (Dionex, Idstein, Germany) using 10 mL stainless steel extraction cells. One gram of cocoa powder was mixed with 5 g of diatomaceous earth. The mixture was loaded into the cartridge. The remaining free space of the cell was filled with diatomaceous earth. Cocoa powder was first defatted with four cycles of hexane for 5 min each (60 °C, 1500 psi). The hexane phase was eliminated by a rotavapor (50 °C). Flavan-3-ols were thereafter extracted with one cycle (5 min) of an acetone/water/formic acid (70:30:0.1, v/v/v) mixture at 70 °C and 1451 psi. The organic phase was then recovered by evaporation (same conditions as above) and injected into HPLC. The volume flush was 20 mL of acetone/water/ formic acid (70:30:0.1 v/v/v). HPLC-DAD Analysis. Prior to injection, all of the samples were filtered through a 0.45 μm syringe filter. The HPLC analysis was performed on an HPLC system from the Agilent Technologies 1200 series (Agilent Technologies, Santa Clara, CA, USA). Separation was achieved at 30 °C using a 250 mm × 4.6 mm i.d., 5 μm, end-capped reversed-phase Lichrospher ODS-2 (Interchim, Montluçon, France). The mobile phase was water/acetonitrile/formic acid (99:0.8:0.2, v/v/ v) (solvent A) and acetonitrile (solvent B). Proanthocyanidins were eluted using the following gradient: 5% of B for 4 min, then from 5 to 35% B in 41 min and from 35 to 100% B in 15 min. At the end, the column was equilibrated to its initial state. The injection volume was 20 μL, and detection was carried out between 200 and 600 nm. Quantification. Quantification of flavan-3-ols was achieved through external calibration curves established at 280 nm with epicatechin, proanthocyanidin B1, and proanthocyanidin B2 standards (0−120 mg/ L), leading to R2 = 0.93, 0.92, and 0.99, respectively. HPLC-MS Analysis. After passing through the flow cell of the diode array detector, the column eluate was split and at 0.25 mL/min was directed to an LCQ ion trap mass spectrometer fitted with an electrospray interface (Thermo Finnigan, San Jose, CA, USA). Experiments were performed in both negative and positive ion modes. The scan range was 100−2000 and scan rate, 1 scan/s. The desolvation temperatures were 250 and 300 °C in the positive and negative ion modes, respectively. The high spray voltage was set at 5000 V. Nitrogen was used as the dry gas at a flow rate of 75 mL/min. Identification was achieved on the basis of the ion molecular mass, MS, and UV−visible spectra. The identifications of catechin, epicatechin, proanthocyanidin B1, and proanthocyanidin B2 were confirmed by direct injection of authentic standards. Total flavan-3-ols were quantified as the sum of individual compounds. NIRS Prediction. NIRS acquisitions were obtained on a Foss 6500 monochromator (Foss, Silver Spring, MD, USA) using a spin cell sample module. About 3 g of cocoa powder was analyzed by diffuse reflectance from 400 to 2500 nm in 2 nm steps. Data were saved as the average of 32 scans and stored as log(1/R), where R was the

MATERIALS AND METHODS

Chemicals. Hexane, acetone, methanol, and acetonitrile were of HPLC grade, purchased from Carlo Erba (Val de Reuil, France). Formic acid was also purchased from Carlo Erba. The reference compounds (−)-epicatechin and proanthocyanidin dimers B1 and B2 were obtained from Sigma Chemical Co. (St. Louis, MO, USA). Cocoa Bean Sampling. Thirty microfermentation trials were carried out in three different countries (Ecuador, Madagascar, and Dominican Republic), in line with producers’ usual practices. A microfermented sample was composed of about 700 g of fresh cocoa beans placed in nets and positioned in fermentation boxes at three different defined levels (close to the top, in the middle, close to the bottom).34 The temperature of each microfermented sample was recorded every 15 min by sensors (Thermo-boutons IP65 40/+85 +/−0.5, Laboratoires Humeau, France) placed in the nets. The nets were surrounded by the mass of fresh cocoa beans to ensure a sufficient volume of cocoa for fermentation to proceed. Fermentation was carried out for 6 days with mixing every 2 days (48 and 96 h). One sample per position and per fermentation was taken each day and then sun-dried to a final moisture content close to 7%. In addition, the cocoa mass was sampled every 2 days during mixing for each fermentation trial and then sun-dried like the microfermented samples. The experimental design led to 635 samples: 524 microfermented samples and 111 cocoa mass samples. This sampling was completed with 83 samples from Cameroon, Ghana, Indonesia, and Trinidad and Tobago. Those samples, used in other projects, were prepared and handled in the same way as the others, but no survey of the temperature was carried out. The total number of samples used for this study was 718. In addition to that, for the 524 microfermented samples, the ST was calculated by using the integral of the curve of temperatures from the beginning of fermentation to the time of sampling with a baseline fixed at 20 °C, as described previously.33 About 100 g of unshelled dried cocoa was ground in a “Valentin” blender (SEB, France) under liquid nitrogen, sifted to 0.5 mm, and stored at −20 °C prior to analysis. Flavan-3-ol Extraction and Quantification. To investigate the phenolic content, and for NIR calibration, wet chemistry was carried out on 193 samples selected from the set of 718 samples. The 193 samples were randomly selected under the constraint of having the same sample percentage per class (fermentation time, origin, position in the box). The 193 samples were analyzed in the laboratory and were used for NIR model development. Fast-Prep Extraction (FPE). A Fast-Prep-24 device (MP Biomedicals, Eschwege, Germany) at 6.5 m s−1 was used with a garnet 10137

dx.doi.org/10.1021/jf501070d | J. Agric. Food Chem. 2014, 62, 10136−10142


Article

Table 1. Statistical Parameters of the NIR Equation for Each Phenolic Constituenta Calibration R2

constituent

N

mean (mg/g)

SD (mg/g)

(−)-epicatechin proanthocyanidin B1 proanthocyanidin B2 total flavan-3-ols

158 157 156 155

9.23 2,18 1,82 18.28

6.30 1.45 1.03 11.59

1.04 0.97 0.31 0.96 0.24 0.95 1.61 0.98 Validation

SEC (mg/g)

SECV (mg/g)

terms (n)

outliersb (n)

RPD = SECV/SD

1.46 0.41 0.29 2.30

9 8 6 10

5 6 7 8

4.32 3.54 3.55 5.03

constituent

N

mean (mg/g)

SD (mg/g)

SEP (mg/g)

slope

bias

SEPc (mg/g)

R2 p

RPDp = SEP/SD

(−)-epicatechin proanthocyanidin B1 proanthocyanidin B2 total flavan-3-ols

30 30 30 30

10.54 2.38 1.96 20.82

6.19 1.43 0.97 11.48

1.81 0.29 0.24 2.33

0.97 1.02 0.98 0.98

−0.78 0.06 0.02 −0.62

1.67 0.29 0.24 2.28

0.93 0.96 0.94 0.96

3.41 4.85 4.12 4.93

a

N, number of samples; SD, standard deviation; SEC, standard error of calibration; R2, coefficient of multiple determination; SECV, standard error of cross-validation; RPD, ratio of performance to deviation; SEP, standard error of prediction; SEPc, standard error of prediction bias-corrected; R2p, coefficient of multiple determination for prediction; RPDp, ratio of performance to deviation for prediction. bStudent’s t test. 3-ol values from HPLC-DAD analysis and flavan-3-ol NIRS-predicted values was evaluated from the linear regression slope, the R2p, and the bias. On the basis of these results, the final NIRS calibration was developed using the 193 samples and applied to the 635 samples for which the ST was calculated to predict the polyphenolic content. The predicted values were studied as a function of fermentation time and the ST.

reflectance at each wavelength. Spectra were acquired in a random order. Statistical analyses were performed using Win-ISI II software (Infrasoft International, Port Matilda, PA, USA) and XLstat software (Addinsoft, Paris, France). Spectrum Pretreatment. Different pretreatments were tested. The one adopted was a mathematical correction for light scattering, using the standard normal variate and detrend correction.35 Then, the second derivative was calculated on five data points and smoothed using Savitzky and Golay polynomial smoothing on five data points on the whole data set including visible and near-infrared wavelengths. Principal Component Analysis (PCA). Prior to calibration development, a PCA was used to extract relevant information from the spectral matrix (n = 193). The generalized Mahalanobis distance (H) was calculated on the extracted PCs for each sample. This statistical distance is useful for defining boundaries of the population and a similarity index between spectra. No sample displayed a Mahalanobis distance greater than the distance limit (3), so all of the samples were kept in the database. NIR Calibration Development. To optimize the model parameters and assess the performance of the predictive model, the 193 samples analyzed in the laboratory were separated into two groups, one for calibration and one for validation. The validation subset comprised 30 samples: 14 samples corresponding to two independent geographical origins (Ghana and Indonesia) and 16 randomly selected samples. The calibration subset comprised the remaining 163 samples. The calibration equations for the phenolic content were constructed with the calibration subset using the modified partial least-squares regression (mPLS) algorithm of WinISI software.30 Calibration statistics included the following parameters: standard deviation (SD), coefficient of determination (R2), standard error of calibration (SEC), and standard error of cross-validation (SECV). Cross-validation was used during calibration development to select the optimum number of latent variables and to minimize overfitting of the equations. For SECV estimation, 25% of the samples were predicted using a calibration model developed with the other 75%. SECV estimation was repeated four times and the average calculated. In addition to R2, the ratio of performance to deviation (RPD = SD/ SECV) was used to evaluate the general quality of the fit obtained for each equation. Unlike SEC and SECV, RPD has no units and can therefore be compared between parameters.36 The Student t test was used to identify t-outlier samples during calibration development. Outlier detection was based on the standardized residuals (= error/ SECV) with a cutoff of 2.5. Two passes of outlier elimination were used.37 The standard error of prediction (SEP), corresponding to the standard deviation of residuals, was estimated by predicting the validation subset using a model developed on the calibration subset. The ratio of performance to deviation of prediction (RPDp) was also calculated as RPDp = SDval/SEP (where SDval was the standard deviation of validation samples). The quality of the fit between flavan-

■

RESULTS AND DISCUSSION Comparison of Flavan-3-ol Extraction Techniques. The three extraction methods were compared for their flavan-3ol extraction yield and their relative repeatability. It clearly appeared that epicatechin required two cycles to be extracted at >95% using Fast-Prep, whereas only one cycle was needed for PLE proanthocyanidin extraction (>95%) (results not shown). A three-way ANOVA was performed on the total flavan-3-ol content with fermentation time, repetition, and extraction type as factors. A strong effect of fermentation time (F = 1773.4; p < 0.0001) and a lesser effect of extraction type (F = 14.5; p = 0.001) appeared. The Newman−Keuls test (α = 5%) showed that Fast-Prep enabled the extraction of a significantly higher content (21.3 mg/g of dried cocoa) than PLE (20.5 mg/g of dried cocoa). The lowest level of flavan-3-ols was obtained with the Ultra-Turax technique (19.7 mg/g of dried cocoa). Seven samples covering the range of flavan-3-ol contents were extracted twice with the three different extraction methods to express the repeatability of (−)-epicatechin extraction yield. For total flavan-3-ol quantification, the smallest coefficient of variation was obtained with Fast-Prep extraction (0.75%); it was close to that achieved with PLE (1.27%), with the highest being for the Ultra-Turax method (2.67%), but remained very low whatever extraction device was used. The coefficients of variation were not significantly different for proanthocyanidin B1 and proanthocyanidin B2 (results not shown). The Fast-Prep extraction procedure was preferred considering that the repeatability of this device was greater than for the other devices, with the advantage of its rapidity and easyhandling. Determination of Polyphenolic Content. For the 193 samples selected, the flavan-3-ols were extracted by Fast-Prep and quantified by HPLC-DAD. The most abundant component was (−)-epicatechin, with an average of 9.4 mg/g of cocoa powder and a standard deviation of 6.3 mg/g. The values were regularly distributed between the 10138



Article

Figure 1. Scatter plot of NIRS-predicted values versus laboratory values for the validation set: (a) (−)-epicatechin; (b) total flavan-3-ols.

Figure 2. First PLS loadings as a function of the wavelengths observed for (−)-epicatechin calibration.

probably by surface contact and hydrogen-bonding,39 lead to regression models computed on a correlation between proteins and flavan-3-ols. Those kinds of correlations make it difficult to interpret the size and the sign of the regression coefficients.40 The contribution of flavan-3-ols in NIR spectra patterns is difficult to interpret, even on pure components. As an example, some of the absorption bands of pure (−)-epicatechin (data not shown) could also be assigned to N−H bond vibrations from proteins, polypeptides, amino acids, and amides: 1458, 1986, 2014, 2062, 2302, and 2474 nm.41 However, the interpretation of PLS loadings (generated from raw or derivative spectra) could efficiently help to bring out such correlations. The observation of the loadings associated with the first PLS term (Figure 2) for (−)-epicatechin calibration revealed high coefficients41 in the NIR region associated with (1) NH bonds at 2052 nm (proteins) and 2300 nm (amino acids); (2) CO bonds at 1900 nm (second overtone stretching) from −CO2H− (acids), at 2036 nm (CO stretching second overtone of CONH2), and at 1924 nm (stretching, second overtone of CONH); (3) CH bond of aromatics at 1444 nm (and phenolic group) and 1684 nm; and (4) O−H bonds at 1412 nm, 2084 nm (R−OH), and 2252 nm (starch). The remaining coefficients were associated with CH bonds from CH2 and CH3; a coefficient at 1668 nm could be assigned to CH from cis-RCHCHR1. These patterns in PLS loadings were similar for proanthocyanidins and total flavan-3-ols. The relationship between bands related to proteins and bands related to O−H and

minimum and maximum values. This distribution reflected the experimental design assuming the hypothesis that flavan-3-ol content is linked to the fermentation time. The distribution was then considered optimum for developing NIR calibration.30 NIRS Models and Validation. The models developed for each individual component were efficient; the number of latent variables (PLS terms) introduced in the models ranged from 6 to 10 (Table 1). The R2 values were between 0.95 and 0.98, and five to eight outliers (t test) were removed during calibration. These models were applied to estimate the SEP by predicting the validation set. The SEP values were between 0.24 mg/g (for proanthocyanidin B2) and 2.33 mg/g (for total flavan-3-ols). The regression between predicted values and laboratory values led to R2p values between 0.93 and 0.96 with nonsignificant bias and slope. The RPDp values were >3, reflecting the efficiency and the robustness of the models. The scatter plot of NIRS-predicted values versus laboratory (Figure 1) values for the validation set highlighted the quality of the calibration for the whole range of the component concentration. This accuracy was not expected due to NIR sensitivity, which is known to be around 0.1% for quantification. We put forward the hypothesis that NIRS was mainly calibrated on an indirect strong correlation due to the changes occurring during fermentation and the interactions of flavan-3-ols with some major components such as proteins. The amount of proteins in cocoa (10−15% of dry bean weight),38 their hydrolysis, and their complexation with flavan-3-ols during fermentation, 10139



Article

aromatic groups confirmed the close and strong relationship between proteins and flavan-3-ols. The evolution of these two compounds is dependent on fermentation conditions and follows similar kinetics. The accuracy of the model, despite low concentration levels, was directly linked to this strong relationship between proteins and flavan-3-ol compounds that occurs during fermentation and drastically changes spectrum profiles. Changes in Phenolic Compounds during Fermentation. The phenolic compounds of the 718 samples were predicted by NIRS using the model developed. The values ranged from 0 to 46.6 mg/g for total flavan-3-ols, 5.4 mg/g for proanthocyanidin B1, 4.3 mg/g for proanthocyanidin B2, and 25.5 mg/g for (−)-epicatechin, with standard deviations of 10.6, 1.3, 1.0, and 5.7 mg/g, respectively. As expected, and whatever the origin, the phenolic content decreased significantly with fermentation time: total flavan-3-ol concentration fell from an average of 33.3 mg/g for unfermented samples to an average of 6.2 mg/g at the end of fermentation (Figure 3). A similar decrease was observed in previous studies.12,15,42,43

Figure 4. Total flavan-3-ol content predicted values versus fermentation times per geographical origin.

behavior was similar for other constituents measured on those samples, such as ammonia nitrogen or free amino acids (data not shown). One hypothesis is that no fermentation occurred after 48 h, probably due to poor fermentation conditions. Fermentation Monitoring and Relationship with Phenolic Content. Temperatures were recorded at the exact place where samples in the net were located. The temperature profiles measured for the 30 microfermentations were similar to those in the literature with an average of 33 °C after 24 h, a rise to 40 °C at 96 h, and ending up close to 44 °C at the end of fermentation. To better understand the effect of temperature on cocoa fermentation, the ST was calculated for each microfermented sample. For the 524 microfermented samples, the calculated values of the ST ranged from 0 to 3662 °C. They were regularly distributed. At 24 h the ST was relatively uniform with an average value equal to 265 ± 51 °C. The dispersion increased after 72 h, whereas at 144 h, the average ST was close to 2811 ± 387 °C. Unsurprisingly, the dispersion of ST was greater with the progression of fermentation and reflected the high heterogeneity of the batches. On this basis, we investigated the usefulness of ST for keeping track of flavan-3-ol changes (Figure 5). There was a significant correlation (r = 0.80) between both parameters, which could be described as a logarithmic equation. Each fermentation batch analyzed separately led to a high correlation between total flavan-3-ol concentration and the ST (e.g., R2 = 0.87 for a particular batch randomly selected from the whole set). The minimum R2 observed per batch was 0.44 and the highest, 0.92, for an average value of 0.74. This result was expected and confirmed the correlation between ST and flavan-3-ol concentration. The ST factor explained changes occurring in the flavan-3-ol content more accurately than the fermentation time, whatever the geographical origin, phenotype, or position in the box studied. Flavan-3-ol derivatives and ST could be used by farmers and chocolate manufacturers to check fermentation levels. In conclusion, this study, based on a large collection of samples and a robust experimental design, confirmed the effect of fermentation on the polyphenolic content of cocoa beans and the feasibility of using it as a marker of fermentation time. The Fast-Prep technique, a repeatable, easy-to-use, and quick

Figure 3. Evolution of (−)-epicatechin, sum of proanthocyanidins B1 and B2, and total flavan-3-ol concentration during fermentation.

Each compound was highly correlated to the others, enabling us to focus only on the evolution of the total flavan-3-ol content during fermentation. A two-way ANOVA was performed to test the influence of cocoa origins and fermentation time on the total flavan-3-ol content. The ANOVA highlighted the strongest significant effect of the fermentation time (F = 96.3; p < 0.0001) compared to the origin (F = 52.5; p < 0.0001). According to the ANOVA-SNK (Newman−Keuls) test (α = 5%), for the fermentation time, the total flavan-3-ol concentration values were not significantly different for 120 and 144 h only. As for the origin, the Dominican Republic, Madagascar, and Indonesia samples had significantly lower flavan-3-ol content values than those from Trinidad and Tobago, Ecuador, and Ghana. Samples from Cameroon were not significantly different from Ghana, but different from the others. Those variations could be explained by differences in the cocoa varieties used, endogenous microflora, fermentation conditions (size and material of boxes, harvest and pod opening delays, etc.), environmental conditions, etc. The interaction between origin and fermentation time was also significant (Figure 4) and was due to differences in the kinetics of flavan-3-ol degradation depending on the origin. Samples from Indonesia displayed particular kinetics because no variation in flavan-3-ol content was observed after 48 h. This 10140



Article

Figure 5. Scatter plot of total flavan-3-ol concentration versus the sum of temperatures for the whole database (a) and for a particular batch (b). (5) Senanayake, M.; Jansz, E. R.; Buckle, K. A. Effect of different mixing intervals on the fermentation of cocoa beans. J. Sci. Food Agric. 1997, 74, 42−48. (6) Vincent, J. C. Influence de différents traitements technologiques sur la fermentation du cacao et le goût du chocolat. Cafe Cacao The 1970, XIV, 303−322. (7) Lima, L. J. R.; Almeida, M. H.; Nout, M. J. R.; Zwietering, M. H. Theobroma cacao L., “The Food of the Gods”: equality determinants of commercial cocoa beans, with particular reference to the impact of fermentation. Crit. Rev. Food Sci. 2011, 51, 731−761. (8) Lopez, A. S.; Dimick, P. S. Cocoa fermentation. In Biotechnology: Enzymes, Biomass, Food and Feed; Wiley-VCH Publisher: New York, 1995; pp 561−577. (9) Quesnel, V. C. Agents inducing the death of cacao seeds during fermentation. J. Sci. Food Agric. 1965, 16, 441−447. (10) Forsyth, W. G. C. Cacao polyphenolic substances. II. Changes during fermentation. Biochem. J. 1952, 51, 516−520. (11) Hansen, C. E.; del Olmo, M.; Burri, C. Enzyme activities in cocoa beans during fermentation. J. Sci. Food Agric. 1998, 77, 273−281. (12) Wollgast, J.; Anklam, E. Review on polyphenols in Theobroma cacao: changes in composition during the manufacture of chocolate and methodology for identification and quantification. Food Res. Int. 2000, 33, 423−447. (13) Buyukpamukcu, E.; Goodall, D. M.; Hansen, C. E.; Keely, B. J.; Kochhar, S.; Wille, H. Characterization of peptides formed during fermentation of cocoa bean. J. Agric. Food Chem. 2001, 49, 5822−5827. (14) Lerceteau, E.; Rogers, J.; Petiard, V.; Crouzillat, D. Evolution of cacao bean proteins during fermentation: a study by two-dimensional electrophoresis. J. Sci. Food Agric. 1999, 79, 619−625. (15) Kim, H.; Keeney, P. G. (−)-Epicatechin content in fermented and unfermented cocoa beans. J. Food Sci. 1984, 49, 1090−1092. (16) Hammerstone, J. F.; Lazarus, S. A.; Mitchell, A. E.; Rucker, R.; Schmitz, H. H. Identification of procyanidins in cocoa (Theobroma cacao) and chocolate using high-performance liquid chromatography/ mass spectrometry. J. Agric. Food Chem. 1999, 47, 490−496. (17) Adamson, G. E.; Lazarus, S. A.; Mitchell, A. E.; Prior, R. L.; Cao, G.; Jacobs, P. H.; Kremers, B. G.; Hammerstone, J. F.; Rucker, R. B.; Ritter, K. A.; Schmitz, H. H. HPLC method for the quantification of procyanidins in cocoa and chocolate samples and correlation to total antioxidant capacity. J. Agric. Food Chem. 1999, 47, 4184−4188. (18) Belšcǎ k, A.; Komes, D.; Horžić, D.; Ganić, K. K.; Karlović, D. Comparative study of commercially available cocoa products in terms of their bioactive composition. Food Res. Int. 2009, 42, 707−716. (19) Sánchez-Rabaneda, F.; Jáuregui, O.; Casals, I.; Andrés-Lacueva, C.; Izquierdo-Pulido, M.; Lamuela-Raventós, R. M. Liquid chromatographic/electrospray ionization tandem mass spectrometric study of the phenolic composition of cocoa (Theobroma cacao). J. Mass Spectrom. 2003, 38, 35−42. (20) Zak, D. L.; Keeney, P. G. Changes in cocoa proteins during ripening of fruit, fermentation, and further processing of cocoa beans. J. Agric. Food Chem. 1976, 24, 483−486.

extraction method, was tested and used to calibrate the NIRS. This made it possible to effectively enlarge the database. Chocolate manufacturers could use the NIRS model on dry cocoa sample to predict fermentation levels by estimating the polyphenolic content. To complete this study, a specific study on tanning occurring with flavan-3-ols and proteins would give more information on reactions taking place during fermentation and on the specificity of the NIRS response. In addition, samples from other geographical origins (such as Ivory Coast or Brazil) need to be analyzed to enhance the robustness of the model.

■

AUTHOR INFORMATION

Corresponding Author

*(F.D.) Centre de Coopération Internationale en Recherche Agronomique pour le Développement (CIRAD) Département PERSYST, UMR Qualisud TA 80/16, 34398 Montpellier Cedex 5, France. Phone: +33 4 67 61 44 32. Fax: +33 4 67 61 44 33. E-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We gratefully acknowledge Valrhona and their partner, the Cocoa Research Unit of the University of the West Indies, for providing cocoa and Gerard Fourny for preparing samples in Trinidad and Tobago. We also thank Peter Biggins, from CIRAD, for revising the English.

■

REFERENCES

(1) de Brito, E. S.; García, N. H. P.; Gallão, M. I.; Cortelazzo, A. L.; Fevereiro, P. S.; Braga, M. R. Structural and chemical changes in cocoa (Theobroma cacao L) during fermentation, drying and roasting. J. Sci. Food Agric. 2001, 81, 281−288. (2) Sukha, D. The influence of processing location, growing environment and polen donor effects on the flavour and quality of selected cacao (Theobroma cacao L.) genotypes. University of the West Indies, St. Augustine, 2008. (3) Aculey, P. C.; Snitkjaer, P.; Owusu, M.; Bassompiere, M.; Takrama, J.; Norgaard, L.; Petersen, M. A.; Nielsen, D. S. Ghanaian cocoa bean fermentation characterized by spectroscopic and chromatographic methods and chemometrics. J. Food Sci. 2010, 75, 300−307. (4) Voigt, J.; Heinrichs, H.; Voigt, G.; Biehl, B. Cocoa-specific aroma precursors are generated by proteolytic digestion of the vicilin-like globulin of cocoa seeds. Food Chem. 1994, 50, 177−184. 10141



Article

(21) Jumnongpon, R.; Chaiseri, S.; Hongsprabhas, P.; Healy, J. P.; Meade, S. J.; Gerrard, J. A. Cocoa protein crosslinking using Maillard chemistry. Food Chem. 2012, 134, 375−380. (22) Niemenak, N.; Rohsius, C.; Elwers, S.; Omokolo Ndoumou, D.; Lieberei, R. Comparative study of different cocoa (Theobroma cacao L.) clones in terms of their phenolics and anthocyanins contents. J. Food Compos. Anal. 2006, 19, 612−619. (23) Mayer, A. M. Polyphenol oxidases in plants and fungi: going places? A review. Phytochemistry 2006, 67, 2318−2330. (24) Whitacre, E.; Oliver, J.; van den Broek, R.; van Engelen, P.; Kremers, B.; van der Horst, B.; Stewart, M.; Jansen-Beuvink, A. Predictive analysis of cocoa procyanidins using near-infrared spectroscopy techniques. J. Food Sci. 2003, 68, 2618−2622. (25) Alvarez, C.; Perez, E.; Cros, E.; Lares, M.; Assemat, S.; Boulanger, R.; Davrieux, F. The use of near infrared spectroscopy to determine the fat, caffeine, theobromine and (−)-epicatechin contents in unfermented and sun-dried beans of Criollo cocoa. J. Near Infrared Spectrosc. 2012, 20, 307−315. (26) Caligiani, A.; Acquotti, D.; Cirlini, M.; Palla, G. 1H NMR study of fermented cocoa (Theobroma cacao L.) beans. J. Agric. Food Chem. 2010, 58, 12105−12111. (27) Counet, C.; Ouwerx, C.; Rosoux, D.; Collin, S. Relationship between procyanidin and flavor contents of cocoa liquors from different origins. J. Agric. Food Chem. 2004, 52, 6243−6249. (28) Cooper, K. A.; Campos-Giménez, E.; Jiménez Alvarez, D.; Nagy, K.; Donovan, J. L.; Williamson, G. Rapid reversed phase ultraperformance liquid chromatography analysis of the major cocoa polyphenols and inter-relationships of their concentrations in chocolate. J. Agric. Food Chem. 2007, 55, 2841−2847. (29) Davrieux, F.; Allal, F.; Piombo, G.; Kelly, B.; Okulo, J. B.; Thiam, M.; Diallo, O. B.; Bouvet, J. M. Near infrared spectroscopy for high-throughput characterization of shea tree (Vitellaria paradoxa) nut fat profiles. J. Agric. Food Chem. 2010, 58, 7811−7819. (30) Shenk, J.; Westerhaus, M. Population definition, sample selection, and calibration procedures for near infrared reflectance spectroscopy. Crop Sci. 1991, 31. (31) Kelm, M. A.; Johnson, J. C.; Robbins, R. J.; Hammerstone, J. F.; Schmitz, H. H. High-performance liquid chromatography separation and purification of cacao (Theobroma cacao L.) procyanidins according to degree of polymerization using a diol stationary phase. J. Agric. Food Chem. 2006, 54, 1571−1576. (32) Camu, N.; González, Á .; De Winter, T.; Van Schoor, A.; De Bruyne, K.; Vandamme, P.; Takrama, J. S.; Addo, S. K.; De Vuyst, L. Influence of turning and environemental contamination on the dynamics of populations of lactic acid and acetic acid bacteria involved in spontaneous cocoa bean heap fermentation in Ghana. Appl. Environ. Microbiol. 2008, 74, 86−98. (33) Hue, C.; Gunata, Z.; Bergounhou, A.; Assemat, S.; Boulanger, R.; Sauvage, F. X.; Davrieux, F. Near infrared spectroscopy as a new tool to determine cocoa fermentation levels through ammonia nitrogen quantification. Food Chem. 2014, 148, 240−245. (34) Sukha, D. A.; Butler, D. R.; Umaharan, P.; Boult, E. The use of an optimized organoleptic assessment protocol to describe and quantify different flavour attributes of cocoa liquors made from Ghana and Trinitario beans. Eur. Food Res. Technol. 2008, 226, 405− 413. (35) Barnes, R. J.; Dhanoa, M. S.; Lister, S. J. Standard normal variate transformation and de-trending of near-infrared diffuse reflectance spectra. Appl. Spectrosc. 1989, 43, 772−777. (36) Williams, P. What does the raw material have to say? NIR News 1993, 4, 13. (37) Burns, D. A.; Ciurczak, E. W. Handbook of near-Infrared Analysis; Dekker: New York, 1992. (38) Bertazzo, A.; Comai, S.; Brunato, I.; Zancato, M.; Costa, C. V. L The content of protein and non-protein (free and protein-bound) tryptophan in Theobroma cacao beans. Food Chem. 2011, 124, 93−96. (39) Baxter, N. J.; Lilley, T. H.; Haslam, E.; Williamson, M. P. Multiple interactions between polyphenols and a salivary proline-rich protein. Biochem. J. 1997, 36, 5566−5577.

(40) Westad, F.; Schmidt, A.; Kermit, M. Incorporating chemical band-assignment in near infrared spectroscopy regression models. J. Near Infrared Spectrosc. 2008, 16, 265−273. (41) Osborne, B. G.; Fearn, T.; Hindle, P. H. Practical NIR Spectroscopy with Applications in Food and Beverage Analysis; Longman Group: Harlow, Essex, UK, 1993. (42) Davrieux, F.; Boulanger, R.; Assemat, S.; Portillo, E.; Cros, E. Détermination du niveau de fermentation et des teneurs en flavan-3ols du cacao marchand par spectrométrie proche infrarouge. In 15th International Cocoa Research Conference, San Jose, Costa Rica, 2006. (43) Payne, M. J.; Hurst, W. J.; Miller, K. B.; Rank, C.; Stuart, D. A. Impact of fermentation, drying, roasting, and Dutch processing on epicatechin and catechin content of cacao beans and cocoa ingredients. J. Agric. Food Chem. 2010, 58, 10518−10527.

10142


Near Infra-red Characterization of Changes in Flavan-3-ol Derivatives

Recommend Documents