Raw CPH

CPH (Theobroma cacao L.) samples were purchased in Ejido Hidalgo located in Tapachula Chiapas, Mexico (14°54’00” N92°16’00”O), in May 2014. Fresh CPH was triturated into a paste using a semi-industrial blender (Crypto Peerless K55, Birmingham, England), and kept at −20°C until use.

Sample preparation

CPH paste was taken out of the freezer and immediately dried by three different methods as was described in a previous work [18].

NMR sample preparation for ¹H NMR analyses

Five hundred mg of fresh or dry CPH (dry base) were extracted with 10 mL of sodium phosphate buffer (pH 6.5, 0.2 mol/L) in a centrifuge tube. All samples were stirred at 400 rpm for 60 min at room temperature. The extracts were centrifuged at 7,250 g for 30 min at 4°C (Hermle Z326K, Wehingen, Germany). Extracts were filtered and dissolved in 40 µL of phosphate buffer pH 6.5 ± 0.05 (0.2 mol/L) and 60 µL of D₂O (99.9% deuterated, Cambridge Isotope Laboratories), containing 10 mmol/L of 3-(trimethylsilyl) propionic-2,2,3,3-d₄ acid sodium salt (TSP), 98% atom deuterium and transferred in a 5 mm NMR sample tube. TSP was used as internal standard for chemical shift referencing (δ = 0 ppm). To avoid degradation of chemical compounds, the sample analyses were completed within 12 h after extraction.

¹H NMR acquisition

One-dimensional (1D) ¹H NMR spectra were acquired at 500 MHz on a Varian NMR System 500 spectrometer (Agilent Technologies, Inc., Santa Clara, CA) operating at 11.7 T and equipped with a 5 mm simple resonance inverse probe (ONENMR). The experiments were carried out without spinning and with water suppression by presaturation (PRESAT) pulse sequence. All the spectra were acquired at 32 K complex points (np), 128 transients with a spectral width of 6,510.4 Hz, acquisition time of 1.5 s, and a relaxation delay of 3 s. All data spectra were Fourier transformed (FT) with FT size of 64 K. To assure good stability and repeatability of the system three groups of samples and three replicates of each group were randomly measured to guarantee a standardized extract giving a total of 9 NMR spectra per sample.

The full NMR spectra were manually phased, baseline corrected, and referenced to TSP. The spectra were transferred to Chenomx NMR suite 8.0.1 software (Chenomx^® Inc., Canada), and reduced into spectral binning at 0.004 ppm in a chemical shift range from 0.66 ppm to 9.49 ppm after the residual water region was removed from the bucketing (4.75–4.95 ppm), all the spectra and the integrals were normalized to the total area.

For compounds assignments Chenomx NMR and MestreNova 10.0 software (Mestrelab Research, Santiago de Compostela, Spain) were used. The integration and identification of the metabolites were carried out based on the coupling constant and chemical shifting of each compound; as described by Villalón-López et al. [4] as well as available literature data [22–32].

¹H NMR data processing and multivariate statistical analysis

The complete data sets were imported into SIMCA-P13 software (v. 13.0; Umetrics, Umeå, Sweden) for statistical analysis. An unsupervised principal component analysis (PCA) was initially executed to explore the data, Hotelling’s T² and DModX tests for outlier detection at 95% confidence interval were used. A supervised pattern recognition using an orthogonal projection on latent structure-discriminant analysis (OPLS-DA) [30]; was used to assess the main metabolite changes in CPH samples.

To facilitate the OPLS-DA model interpretation, S-plots were obtained employing the color-coded value of the predictive component loading of the OPLS-DA with two classes. The statistically significant changes in the signals were identified based on the visual inspection of the pool ¹H NMR normalized spectra.

Identification of metabolites in dehydrated CPH extracts by ¹H NMR analysis

A ¹H NMR spectra of fresh and the three different dried CPH samples with 10x zoom are shown in Figure 1A, while HAD-CPH, MWD-CPH, and FD-CPH in Figure 1B, 1C, and 1D, respectively. The principal resonances of the spectrum were assigned according to FooDB database and literature [33], and database chemical shift data [22–31]. The results are summarized in Table 1. Even though the signals are broad and show a certain degree of overlap, the spectral contributions and the chemical shifts of some organic acids, carbohydrates, and methylxanthines could be clearly recognized. It is important to highlight that to obtain the fingerprint the amount and extraction process was standardized. It can be appreciated that the presence of many metabolites is consistent in the three drying methods with different relative intensities (Figure S1). According to the chemical shifts of their ¹H NMR spectra, 25 compounds were identified in fresh and dehydrated CPH (Figure 1). To confirm some signal assignments 2D NMR experiments were also performed (data not shown).

Display full size

Figure 1. 

¹H NMR spectra (500 MHz, 25°C) of fresh and dehydrated CPH extracts with the following signal assignments. (A) Fresh; (B) HAD; (C) MWD; (D) FD; 1: formic acid; 2: theophylline; 3: theobromine; 4: chlorogenic acid; 5: tyrosine; 6: fumaric acid; 7: maleic acid; 8: sucrose; 9: glucose; 10: gluconic acid; 11: fructose; 12: caffeine; 13: gallic acid; 14: tartaric acid; 15: acetoine; 16: myo-inositol; 17: choline; 18: γ-aminobutyric acid (GABA); 19: aspartic acid; 20: succinic acid; 21: levulinic acid; 22: acetic acid; 23: lactic acid; 24: theanine; 25: myristic acid

Regarding cacao by-products there is little information about their composition; in addition, the use of phosphate buffer instead of organic solvents for extraction is worth noticing since it is an environmentally friendly alternative. In the present study, some compounds such as gallic acid (polyphenolic acids), glucose, and some amino acids like tyrosine and aspartic acid were successfully identified (Figure 1). To the best of our knowledge, it is the first time that theanine is identified in cacao by-products. Therefore, it is necessary for further analysis to know more about this compound which possesses important health-related benefits promoting relaxation and improving concentration and learning ability [34]. Glucose and fructose have been found in cacao beans during fermentation and roasting [35], as well as during pectin extraction from CPH [36, 37].

In all spectra the chemical shifts of acetic acid and lactic acid were found, these two molecules are characteristic compounds formed during cacao beans fermentation [38]. The intensity of the signal decrease in different levels and this might be related to the conditions of each drying method. GABA signal was detected in both MWD and FD and did not appear in HAD.

Multivariate analysis in CPH components resulting from dehydration treatment

A principal component analysis (PCA) was carried out supported with multivariate analysis using an inconspicuous variation, showing that both the PC-1 and PC-2 explain 96.9% of the total variance with a Q² of 55.5% (Figure 2A), where is clearly revealed a difference in CPH composition between fresh and drying methods (Figure S2). On the other hand, Figure 2B shows a PCA that explains 96% of the total variance with a Q² of 50.3%, indicating similarities between FD and MWD.

Display full size

Figure 2. 

PCA model of the two principal components score plot of CPH. (A) PCA including fresh and dehydrated CPH; (B) PCA of dehydrated CPH (HAD, MWD, and FD)

A supervised model OPLS-DA was performed (Figure 3), as well as, an OPLS/O2PLS overview plot for describing the predictive variation between X and Y loadings and scores (Figure S3). The results show a well-separated cluster for the three samples, indicating that the metabolome profile of CPH is clearly different among the drying methods (HAD, MWD, and FD). Also, the scatter plot of the scores of PC1 versus PC2 as evidenced in Figure 3, imply the variations generated by storage conditions, extraction conditions, and dehydration method that together explain 90% of the total system variance; a permutation analysis plotting was carried out for R² and Q² from 100 permutation test to validate the OPLS-DA model (Figure S4) and shows the correlation coefficient on the permuted and observed data. The OPLS-DA model clearly classified the dehydration treatments into three different clusters explaining 98.23% of the system variance; and showing that all the NMR signals are highly imported for the fingerprint differentiation, also those that are unidentified.

Display full size

Figure 3. 

OPLS-DA score plot of CPH with different dehydration treatments. The OPLS-DA model with goodness of fit (R²X) was 0.984 and the goodness of predictability was indicated by a Q² value of 0.957

Variations in CPH components resulting from dehydration treatment

For multivariate analysis of CPH, the entire set of identified metabolites was considered. However, henceforth, only the metabolites that have high relative abundance and significant differences among the drying methods are relevant for further discussion.

Therefore, only the metabolites on which overall differences and similarities among the samples are shown in Figure 4. A clear distinction between HAD and MWD is observed (Figure 4A) indicating different ¹H NMR profiles (Figure 4B). The differences between HAD and MWD treatments were primarily observed in sugar and organic acids content. The analysis showed that fumaric acid, aspartic acid, succinic acid, acetic acid, and myristic acid content differ in HAD and MWD (Figure 4C). The organic acids could be present in dehydrated CPH due to an early stage of fermentation. In addition, the acetoin, lactic acid, and theanine (glutamic acid analogue) signals intensity disappeared in comparison to the signal intensity observed in the fresh sample (Figure 1A). While the signal of GABA spectra disappeared only in HAD treatment and is maintained in MWD treatment. These results cannot be explained exclusively by biochemical reactions that occurred during the drying process, nor by prolonged exposure to applied temperature in HAD treatment.

Display full size

Figure 4. 

A comparison between HAD and MWD treatments with OPLS-DA score plot. It is calculated from the binning of the NMR data showing (A) Inner relation plot of the score vectors from the same component, R²X(cum) = 0.471, R²Y(cum) = 0.994, Q²(cum) = 0.977, type scaling UV (unit variance); (B) the average of line spectra color colored according to the differences in the loading plots of OPLS-DA; (C) the overlay of S-line scores, color colored according the predictive component loading

Differences between HAD and MWD treatments were mainly observed in sugar, organic acids, theophylline, myo-inositol, and choline intensity signals. HAD loses acetoine, GABA, and theanine signals, while FD did not show choline, levulinic acid, or myristic acid (Figure 5).

Display full size

Figure 5. 

A comparison metabolites fingerprint signals between HAD and MWD treatments with OPLS-DA score plot calculated from the binning of the NMR data showing R²X(cum) = 0.471, R²Y(cum) = 0.994, Q²(cum) = 0.977, type scaling UV. (A) Inner relation plot of the score vectors from the same component; (B) line spectra color colored according to the differences in the loading plots of OPLS-DA; (C) S-line scores, color colored according to the predictive component loading

MWD and FD showed similar metabolite profiles mainly on sugars and some organic acids, indicating that both MWD and FD may be less destructive dehydration treatments for CPH, preventing the loss of functional metabolites. FD preserves gallic acid contrary to MWD that did not show this signal; these results are contradictory to those found in a previous work in which more polyphenolics compounds were found [18]; these discrepancies may be due to the selected solvent which has neutral pH, in addition since the extractant is an aqueous solution it may not allow extraction of other polar compounds. FD clearly shows a higher concentration of theanine which suggests that this molecule is negatively affected by temperatures up to 50°C and is best preserved at low temperatures, therefore it was not found in HAD (60°C) and MWD (~90°C, Figure 6).

Display full size

Figure 6. 

A comparison of metabolites fingerprint signals between MWD and FD treatments with OPLS-DA score plot calculated from the binning of the NMR data showing R²X(cum) = 0.381, R²Y(cum) = 0.996, Q²(cum) = 0.977, type scaling UV. (A) Inner relation plot of the score vectors from the same component; (B) line spectra color colored according to the differences in the loading plots of OPLS-DA; (C) S-line scores, color colored according to the predictive component loading

Some metabolites clearly observable and not overlapped with the S-line plot were selected to determine their relative abundance by comparison with internal standard (TSP; Figure 7). The identified metabolites that generated the main differences between fresh and dehydrated samples are displayed in Figure 7, as mentioned the relative abundance was referred to TSP standard.

Display full size

Figure 7. 

Relative abundance referred to TSP standard concentration, of some metabolites

Fresh CPH has clearly a higher relative abundance of acetic, succinic, and lactic acid, as well as levulinic acid; it may be due to the presence of some bacteria in an early stage of fermentation in fresh CPH. On the other hand, aspartic acid is less affected by temperature, even with an increase in its abundance after the dehydration process. While theanine is more affected by MWD than HAD, it is better preserved by FD method. Myristic acid content must be highlighted since there is little information about its presence in any part of cacao.

Conclusions

A ¹H NMR untargeted metabolomics to obtain the fingerprint of CPH fresh or dehydrated has been successfully established in this study. Both the extraction procedure and NMR parameters were optimized to obtain a fingerprint spectrum, despite the fresh sample having a slightly early fermentation stage (for future studies will be necessary a better control of temperature and storage conditions as far as possible). The use of chemometrics tools to analyze the obtained data allows us to identify 25 metabolites, most of these metabolites belong to methylxanthines, sugars, some amino acids, fatty acids, and organic acids. Multivariate analysis is a method that allowed us to differentiate the CPH composition of fresh samples compared to those obtained by different dehydration processes. Our results highlight that FD and MWD methods are similar in composition maintaining most of the compounds after drying.

Therefore, MWD may be a suitable commercial method to dehydrate CPH to preserve important health-related metabolites. Moreover, the results provide complementary information about CPH composition and how it is affected by the dehydration method selected and the temperature used during this process. Nevertheless, further studies are needed to evaluate i) different solvents in order to identify more possible metabolites present in CPH, and ii) the real content of the distinctive metabolites by their quantification.