Introduction

Cocoa fermentation is an essential step in chocolate production1,2. This process is critical for the biosynthesis of relevant flavors and bioactive compounds that determine the chocolate quality2,3,4. Incomplete/shorter fermentations result in cocoa with undesirable astringent flavors3, while overfermentations result in undesirable hammy and putrid flavors5,6. Despite their high relevance, very little information is available regarding the biochemistry and microbiology of cocoa fermentation. Additional research is still required to elucidate the synthesis of valuable compounds during this critical step in chocolate production.

Cocoa fermentation is a highly complex spontaneous process led by the interaction of heterogeneous microbial populations. Even though it is generally accepted that the origin of this microbial community is the environment(e.g., insects, worker hands, tools, and air) the exact origin of every microorganism group present in the coco fermentation has not been proven so far3,7. For this reason, the origin of the microbial community accountable for cocoa fermentation is still unclear3,8,9. Three significant phases during spontaneous cocoa fermentations have been previously proposed10,11,12,13 due to the interactions between three main groups of microorganisms (yeast, lactic acid bacteria, and acetic acid bacteria) and their metabolic products9,14. An initial anaerobic phase occurs at the beginning of the fermentation, in which yeasts produce ethanol using the fermentable sugars from the cocoa pulp9. Then, when the pulp begins to disappear and oxygen penetrates at a higher rate the fermentation mass, a second (aerobic) phase occurs. In this phase, Lactic Acid Bacteria (LAB) utilizes the remaining sugars (mainly fructose) to produce lactic acid that accumulates9,15. Finally, the last phase occurs when acetic acid bacteria (AAB) convert ethanol into acetic acid that strongly affects the microbial diversity of fermentation9,15,16. Based on this paradigm, different mathematical models of cocoa fermentation have been previously proposed to further study the dynamics among these phases10,11,12,13. Apart from the general metabolic activity, other specific biochemical processes in cocoa fermentation have been modeled. For instance, genome-scale reconstructions for selected strains of acetic acid and lactic acid bacteria isolated from cocoa fermentations have been proposed in various studies, even using advanced techniques as fluxomic to improve the models17,18,19.

However, recent metagenomics studies highlighted the relevance of other dominant groups of microorganisms during cocoa fermentations8,13,14,20,21,22,23. The results of these studies question the traditional view of cocoa fermentation. For instance, the dominance of other microorganisms such as Bacillus, Pseudomonas, Aspergillus, Malassezia, and Pestalotiopsis at several phases during cocoa fermentations have been recently demostrated8,13,14,20,21,22,23. The role of these microorganisms and their potential to produce valuable flavor and bioactive compounds during spontaneous cocoa fermentation still have to be elucidated. Therefore, a current essential challenge in cocoa research is identifying metabolites associated with these microorganisms to better understand their relevance during cocoa fermentation.However , only a few studies have dissected the variation in the metabolic profile during spontaneous cocoa fermentations24,25,26,27,28. Most of these studies focus on identifying metabolites during the fermentation of bulk cocoa (i.e., cocoa varieties with an ordinary flavor profile). In contrast, these kinds of analyses are rare and still in their infancy for fine-flavor cocoa. In this regard, to understand the metabolic dynamic of fine-flavor cocoa fermentation is a necessary previous step to connect the metabolites and microorganisms involved in this process.

Compared to bulk cocoa, fine-flavor offers a higher diversity of flavor attributes of high demand by the elite chocolatiers. Therefore, identifying changes in the metabolic fingerprint during spontaneous fermentation of fine-flavor cocoa is highly relevant to standardize this process and produce a high-quality bar of chocolate. Consequently, the goal of this study was to analyze the changes in the metabolic fingerprint during spontaneous fermentation of fine-flavor cocoa. Our data revealed two main phases of differential metabolic activity during spontaneous fermentation of fine-flavor cocoa that correlated with the observed variations on temperature and highlighted the relevance of comprehensive metabolomics studies to break down the current cocoa fermentation paradigm.

Experimental section

Cocoa beans fermentation

Cocoa fermentations were performed at the Luker Farm (Caldas, Colombia (5°4' N 75°41' W)) owned by CasaLuker S.A. Cocoa beans, from a standard, pre-designed, and a frequently-used mixture of Theobroma cacao clones (i.e., LUKER40, FSV41, FSA13, and TSH565), were selected for wooden box fermentations. Additional information regarding the used clones is available at the International Cocoa Germplasm Database (ICGD) (http://www.icgd.reading.ac.uk/).

Two independent fermentation boxes containing 400 kg of cocoa pulp-bean mass each was arranged using a ladder system in a pre-designed fermentation room as previously described13. Briefly, this fermentation room has an area of 84 m2 with a metallic ceiling and acrylic walls to prevent air current entering the fermentation zone. The average temperature and during day time are 33 °C while at night time is 25 °C. Humidity is around 60% with a maximum level of 94%. The fermentation mass was mixed from the fourth day on, every 48 h,to allow aeration . Standard cutting tests were used to evaluate the fermentation quality over time and determine its final point following the standard protocols13.

Fermentation mass sampling

Three biological replicates, each consisting of 10 seeds, were collected from different fermentation mass locations—two of them from one fermentation box and the third one from the other fermentation box, as previously described13. Sampling was made at the beginning and every 24 h until the end of fermentation. To avoid chemical degradation, samples were frozen immediately at − 80 °C after collection. Furthermore, to connect the medium conditions and metabolic activity, we recorded the fermentation cocoa mass temperature variation every 4 min for each fermentation box using precision sensors placed in the center of the fermentation boxes.

This sampling process followed the guidelines and legislation settled by the ministry of environment and sustainable development of Colombia. We obtained a permission to access genetic and derivate resources according to resolution No 284 of 2020.

Sample preparation

Fermentation samples were initially milled using a clean and precooled coffee grinder. Aliquots of 100 mg were further macerated in the presence of liquid nitrogen and subsequently defatted by the addition of 500 µL of n-hexane and vigorous vortex as previously reported29. Defatting was repeated three times for each sample to maximize the lipid removal. Metabolite extraction was performed by adding 1 mL of methanol-ultrapure water (70:30) to each sample. Then these extracts were placed into an ultrasonic bath at room temperature for 10 min following by vortex-mixed for 10 min. After that, the samples were centrifuged at 6000 g, 25 °C for 10 min, and the supernatant was collected for LC-QTOF-MS and GC-QTOF-MS analysis. For LC-QTOF-MS analysis, 30 µL of supernatant was mixed with 70 µL of Milli Q water and transferred to LC–MS vials. For GC-QTOF-MS, 30 µL of supernatant was evaporated to dryness using a SpeedVac concentrator system. Methoxymation was performed by adding 20 µL of O-methoxyamine hydrochloride (30 mg/mL in pyridine) to each sample and vigorously vortex-mixed 5 min. Then, samples were incubated at 70 °C for 1 h. Then, 20 µL of BSTFA with 1% of TMCS were added, vortex-mixed for 5 min, and placed in the oven at 70 °C for 1 h. Finally, 100 µL of heptane containing C18:0 methyl ester (5 µg/mL) as internal standard.

Metabolic fingerprinting by LC- QTOF-MS analysis

The metabolic analysis was performed using an Agilent 1260 Infinity LC System coupled with Q-TOF 6545 MS system (Agilent Technologies, Palo Alto, CA, USA). 5 µL of the extracted sample was injected onto InfinityLab Poroshell 120 EC-C18 column (2.1 × 150 mm 2.7 µm, Agilent) thermostated at 30 °C. The flow rate of mobile phase (A: Milli-Q water with 0.1% formic acid (v/v), B: acetonitrile with 0.1% formic acid (v/v)) was 0.4 mL/min. The gradient elution program started running at 5% B, increasing to 10% B in 7 min, then increasing to 95% B in 15 min and hold at 95% B for 2 min, ending going back to initials conditions in 1 min and held there for 8 min to allowed column re-equilibrium. The analysis was performed with positive and negative ionization mode using two reference masses in each polarity: m/z 121.0509 (C5H4N4) and m/z 922.0098 (C18H18O6N3P3F24) in positive mode an m/z 112.9856 (C2O2F3(NH4)) and m/z 1033.9881 (C18H18O6N3P3F24) for negative ionization mode. The system was operated in full scan mode from 100 to 1100 m/z. The capillary voltage was set to 3500; the drying gas flow rate was 8 L/min at 325 °C, gas nebulizer 50 psi, fragmentation voltage 175 V and skimmer 65 V and octopole radio frequency voltage (OCT RF Vpp) 750 V. Data were collected in the centroid mode at a scan rate of 1.00 spectrum per second.

Metabolic fingerprinting by GC-QTOF-MS analysis

Metabolic fingerprinting was performed using an Agilent Technologies 7250 GC/QTOF system (Agilent Technologies, Palo Alto, CA, USA). 2 µL of the derivatized sample was injected onto an HP-5MS UI column (30 m × 0.25 mm × 0.25 µm), using helium as a carrier gas at a constant gas flow of 0,7 mL/min. Injector temperature was set at 280 °C and split ratio to 30:1. The gradient temperature program started at 60 °C, held there for 1 min, and then started to increase to 325 °C with a rate of 10 °C/min. The GC–MS transfer line was set at 280, filament source at 250, and quadrupole temperature at 150. The electron ionization (EI) source was set at 70 eV, and the mass spectrometer was operated in full scan mode from 50 to 600 m/z at a scan rate of 5.00 scans/s.

Quality control samples

Quality control (QC) samples were prepared by mixing equal volumes of extracted samples. To determine the reproducibility and stability of the analytical platforms used, several QC runs were performed before analyzing all cocoa samples until system equilibration was achieved and every five randomized samples.

Data treatment

All raw data were processed as previously reported by Cala et al30. In summary, for LC-QTOF-MS data processing was performed withAgilent MassHunter Profinder B.10.0 Software for deconvolution, alignment, and integration.. GC–MS data processing consisted of a deconvolution step with Agilent MassHunter Unknowns Analysis B.10.00 and metabolite identification using the libraries Fiehn version 2015 and NIST 17. Agilent Mass Profiler Professional B.12.1 software was used for retention time alignment, and then withAgilent MassHunter Quantitative B.10.00 was performed the integration of each metabolite following Agilent guidelines. Both LC–MS and GC–MS data were inspected manually to clean thenoise. Finally, data were filter by presence and reproducibility, maintaining the metabolites present in 100% of the biological replicates in each group and a coefficient of variation in the QC lower than 20%.

Statistical analysis

To determine statistically significant differences between metabolomics profiles, multivariate (MVA) statistical analyses were performed using SIMCA 16.0 (Umetrics, Umea, Sweden). Principal component analysis (PCA) was applied to evaluate the acquired data quality, verifying that the QC samples were correctly clustered in these models to guarantee the stability of the analytical system. After that, PLS-DA and OPLS-DA models were built to maximize and inspect the differences between study groups and select responsible metabolites for separating the groups. Pareto scaling were used before the statistical analysis. For all platform data, the significant variables were selected by keeping only the variables that fulfilled:1) MVA criteria (variance significant in projection (VIP) > 1.5 with Jack-knife confident interval (JK) not including the zero value from orthogonal partial least-squares discriminant analysis (OPLS-DA) with CV-ANOVA < 0.05) and 2) Change percent > 30%.

Metabolites identification

Accurate masses of features representing significant differences in class separation identified by all platforms were searched in the following databases: KEGG (http:// genome.jp/keg) and Lipid MAPS (http://lipidmaps.org), METLIN (http://metlin.scripps.edu) using the CEU Mass Mediator tool31. Then LC − MS/MS analyses were performed to confirm the metabolite's identity.

Results and discussion

A Multiplatform metabolic fingerprinting of fine-flavor cocoa fermentation was obtained through LC-QTOF-MS and GC-QTOF-MS

We used LC-QTOF-MS and GC-QTOF-MS to elucidate the metabolic activity through the cocoa fermentation process. The total coverage of molecular features from the metabolic fingerprinting (MF) after data processing and filtering that we obtained consisted of 1497, 551, and 71 features by LC-QTOF-MS( +), LC-QTOF-MS(-), GM-and GC-QTOF-MS( +), respectively. Using quality control (QC) samples clustering, we evaluated the performance of the different analytical platforms in unsupervised PCA models (Supplementary Fig. S1). A clear QC grouping was observed in PCA analysis for all analytical platforms, assuring the quality of acquired data and supporting that separating groups is related to biological and not analytical variations. Overall, we observed a significant difference between the metabolic fingerprint of the beginning (day 0) and the end (day 8) of the fermentation (Fig. 1), generally characterized by an increase in the signals of fermentation day 8. For all platforms, the PCA analysis revealed a clear separation between the metabolic fingerprint of day 0 and day 8 of the fermentation and a change in metabolite throughout all fermentation days (Fig. 2). However, the analysis of the metabolic fingerprint for each day revealed only a slight separation between specific days such as day 0 and day 1, day 2 and day 3, and the last three days of fermentation (i.e., day 6, day 7, and day 8), suggesting that these days exhibit a similar metabolic profile. This same behavior has been previously reported in several studies24,25,28.

Figure 1
figure 1

Comparison of total ion chromatogram of metabolic profiles for day 0 (D0) and day 8 (D8) during fermentation of fine-flavor cocoa. (A) Metabolic fingerprinting (MF) by LC–MS( +); (B) MF by LC–MS(-); (C) MF by GC–MS.

Figure 2
figure 2

PCA score plots for all fermentation samples. (A) MF by LC–MS( +): R2(cum): 0.875, Q2 (cum): 0.826; (B) MF by LC–MS(-): R2(cum): 0.876, Q2 (cum): 0.77; (C) MF by GC–MS: R2(cum): 0.724, Q2 (cum): 0.229.

Two different phases of metabolic activity were detected during spontaneous fermentation of fine-flavor cocoa beans

PLS-DA models were built to explore and maximize the differences in metabolic fingerprint throughout the fermentation process (Fig. 3A–C). The PLS-DA models showed little discrimination in the metabolic fingerprint of the fermentation process on each day in all analytical platforms; however, in the PLS-DA score plots, it is possible to observe three clusters of fermentation days corresponding to D0-D3, D4, and D5-D8. Using this approach, additional PLS-DA models were built to explore the differences between these three groups (Fig. 3D–F). A clear separation was observed in the score plots for all PLS-DA models between these groups with high quality, proven by significant variance values explained (R2), variance predicted (Q2), and CV-ANOVA. These results suggest that the most prominent metabolites changes cluster in two phases (D0 to D3 and D5 to D8) during the fermentation process of cocoa beans. To further explore the trends in metabolites modifications along fermentation days, a heatmap was built for all metabolite features detected in all analytical platforms using MetaboAnalyst 5.0 (Fig. 4). The groups (days) in the diagram were allocated using a hierarchical clustering algorithm, joining them by similarity, as indicates the dendrogram on the top of the figure. This heatmap shows two similar metabolic fingerprint with differential alteration of a significant number of metabolites between them. The first metabolic fingerprint goes from day 0 to day 3 and the second one from day fifth on, supporting the presence of two major metabolic phases during the fermentation of fine-flavor cocoa. Interestingly, these two phases are concomitant with the temperature profile in the cocoa mass through the entire process (Fig. 5).

Figure 3
figure 3

PLS-DA models for all cocoa fermentation samples and the proposed metabolic phases. ALL DAYS: (A) MF by LC–MS( +): R2(cum): 0.838, Q2 (cum): 0.137, cv-anova: 0.995; (B) MF by LC–MS(-): R2(cum): 0.696, Q2 (cum): 0.083, cv-anova: 0.993; (C) MF by GC–MS: R2(cum): 0.442, Q2 (cum): 0.084, cv-anova: 0.951. PHASES: (D) MF by LC–MS( +): R2(cum): 0.907, Q2 (cum): 0.899, cv-anova: 6.296e−07; (E) MF by LC–MS(-): R2(cum): 0.903, Q2 (cum): 0.804, cv-anova: 0.00053; (F) MF by GC–MS: R2(cum): 0.415, Q2 (cum): 0.45, cv-anova: 0.0188.

Figure 4
figure 4

Heat map analysis of metabolite features detected in all analytical platforms for all fermentation days (MetaboAnalyst 5.068). The color spectrum ranging from red to green indicates the range of high to low signal intensities for each metabolite.

Figure 5
figure 5

Temperature dynamic during the fermentation of fine-flavor cocoa. The dotted line divides cocoa fermentation temperature dynamics into two phases.

The temperature of the fermenting mass is a crucial parameter during spontaneous fermentation of fine-flavor cocoa

Our data revealed two different temperature phases during spontaneous cocoa fermentation (Fig. 5) concomitant to the two phases of differential metabolic activity previously described (Figs. 3 and 4). The first phase is essentially exothermic. During this phase, the temperature of the fermenting mass significantly increased from 26.8 ± 0.3 °C to 49.4 ± 2.5 °C. Contrary, the second phase (from the fifth day on) is primarily isothermic. The temperature during second phase remained between 45 and 50 °C until the fermentation ended, a variation 81% lower compared with the exothermic phase. This kind of temperature profile was reported in different cocoa fermentations worldwide, with different cocoa varieties, fermentation methods, and weather conditions20,21,32,33.

Temperature rise, usually more than 25 °C over the initial temperature of cocoa mass, could be associated with exothermic reactions (e.g., ethanol and weak acids production) from the first days of fermentation10,34,35. A maximum temperature increase rate was observed on fermentation day 4, which corresponds with the first turning of the fermenting mass to allow oxygenation. Thus, oxygen availably could partially explain the increase in temperature as a substantial increase in oxygen concentration may lead to a dramatic rise in exothermic oxidation reactions36,37. The temperature over 45 °C is a critical control point to determine the quality of the process, independently of the fermentation method, the region, the cocoa type, or the required fermentation time20,33,37,38,39.

As mentioned before, from the fifth day on, the temperature experienced minor changes and remained between 45–50 °C until fermentation ends. During this final phase, most microbial populations, and eventually its associated biochemical activity, decline as previously observed13,14,40. The relative abundance of microbial groups as yeast and LAB significantly decreases the last days of cocoa fermentation13,15,40. This drop is caused by a considerable decrease of fermentable sugars in the medium (e.g., glucose, fructose)10,11. Also, temperature plays a relevant role in controlling microbial populations considering that LAB and yeast growth decreases dramatically over 40 °C41,42. However, the elevated presence of lactic acid and oxygen prompts the growth of AAB that can resist higher temperature, producing acetic acid (an exothermic process) and releasing heat to the medium at a considerably lower rate but sufficient to maintain cocoa mass temperature relatively high, with a maximum variation of 5 °C35,43.

The temperature rise has a critical impact on flavor development during cocoa fermentation, as it guarantees the formation of molecules responsible for a high-quality sensorial profile44. The production of these compounds evidence the influence of temperature in metabolic activity across cocoa fermentation and the quality of the cocoa10,44. A high temperature causes a significant diffusion of acetic acid, lactic acid, and ethanol into the beans promoting the degradation of flavonoids, reducing the bitterness and astringency of the cocoa1,39. Also, it changes more than 25 °C during the fermentation10,37,44, stimulating a wide range of biochemical reactions and the growth of specific microorganisms in each increasing stage. These microorganisms carried out most of the transformations occurring in cocoa fermentation37,45. To achieve a high-quality sensorial profile, a proper succession of some specific microorganism genera is crucial. The development of this microbial progression depends on substrates present in the media and temperature39.

Additionally, temperature plays a crucial role in proteolysis. The temperature reaches more than 45 °C during cocoa fermentation. This high temperature causes the activation of some native enzymes, and the protein heat denaturation produces a significant amount of peptides and amino acids25,46. These molecules are flavor precursors associated with the development of fruits and nutty notes in cocoa, qualities quite appreciated in the international markets10,47,48.

The variation in dominant groups of metabolites diverge between the two phases of differential metabolic activity

We analyzed the effect of the temperature dynamic in the metabolic profile of cocoa fermentation using a multiplatform metabolomics analysis. This approach allowed us to propose a new paradigm of the phases of cocoa fermentation based on the global metabolic activity instead of the variation in only a few metabolites. We propose two phases of cocoa fermentation based on the temperature dynamic: an exothermic phase from day 0 to day 3 and an isothermic phase from day 5 to day 8. Once these phases were established, univariate (p-value < 0.05 from hypothesis testing) and multivariate (OPLS-DA models) analyses were performed to select the differential metabolites between each phase. For all analytical platforms, the OPLS-DA analysis allowed modeling the differences between the two phases with statistically significant values for R2, Q2, CV-ANOVA (Fig. 6). As a result, 44, 65, 20 differential metabolites were identified in LC-QTOF-MS( +), LC-QTOF-MS(-), and GC-QTOF-MS( +), respectively.

Figure 6
figure 6

OPLS-DA models for two phases. (A) MF by LC–MS( +): R2(cum): 0.804, Q2 (cum): 0.962, cv-anova: 3.58e−13; (B) MF by LC–MS(-): R2(cum): 0.739, Q2 (cum): 0.96, cv-anova: 5.67e−07; (C) MF by GC–MS: R2(cum): 0.702, Q2 (cum): 0.864, cv-anova: 5.86e−12.

Table 1 shows metabolites altered between the exothermic and isothermic phases of cocoa fermentation. The biggest group of metabolites differentially expressed correspond to amino acids, dipeptides, and tripeptides. The concentration of these 23 molecules increased in the isothermic phase. The formation of amino acids and peptides during cocoa fermentation is due to protein hydrolysis and denaturation processes. Protein content (e.g., albumin, prolamin, globulin, and glutein) represents 10 –15% of dry unfermented cocoa beans46,47 and is significantly lower for fermented cocoa beans38,47,49. The drop in protein content, usually above 60%, is caused by proteolysis during fermentation that involves two groups of native cocoa proteases: endoproteases and carboxypeptidases38,47,49. The optimal temperature for these enzymes is 45–50°C50,51, precisely the same temperature range of the fermentation the isothermic phase. Also, this temperature itself is accountable for protein denaturation, breaking them down into peptides or even amino acids25,49. This protein degradation linked to enzymatic hydrolysis and denaturation by heat in cocoa fermentation has been widely documented in several studies25,46,47,48,49.

Table 1 Metabolites with statistical significance in the two phases of the fermentation process identified by metabolic fingerprinting using LC-QTOF-MS( ±) and GC-QTOF-MS( +).

Peptides and amino acids are highly relevant for the nutraceutical and sensorial properties of fine-flavor cocoa. Several cocoa peptides have been associated with bioactive properties such as antioxidant, antihypertensive, and antimicrobial52, although further research is still required. On the other hand, peptides and amino acids are precursors of pyrazines produced during drying and roasting through Maillard reactions5,49. These pyrazines are responsible for fine-flavor cocoa sensorial notes as fruity, floral, and cocoa5,53. In this regard, proteolysis could explain why cocoa fermentation in which temperature is consistently below 45 °C, the obtained chocolate usually has a poor sensorial profile.

Flavonoids are the second largest group of molecules differentially expressed with 18 compounds. In the isothermic phase, we observed a rise of different flavonoids such as dimers and trimers of procyanidins, polyphenol glycosides, and rhamnose-containing polyphenols. Also, our data show a decrease in some arecatannin types and butein and pelargonidin derivates. Although the overall content of flavonoids is expected to decrease throughout fermentation, these degradation processes can cause the emergence of polyphenolic dimers and trimers as well as polyphenolic acids as caffeic acid, benzoic acid, and coumaric acid54,55. For instance, oxidation and polyphenol oxidase hydrolysis of complex flavonoids such as anthocyanins, procyanidins, epigallocatechin, and kaempferol results in the formation of dimers and trimers of procyanidins and rhamnose-containing polyphenols54,55. However, these derivates tend to have lower bioactivity, leading to a significant drop in the bioactive properties of cocoa throughout fermentation, as previously reported8,38,56. Polyphenol degradation and eventual derivates formation can be more pronounced at high temperatures56,57.

Contrasting that, carbohydrates are the compound group with a higher number of decreasing concentration metabolites from the exothermic to isothermic phase. These metabolites are a primary substrate for many yeasts and bacteria, eventually generating a rise in different organic acids beyond traditional ones (e.g., lactic acid, acetic acid) like benzoic acid, caffeic acid, coumaric acid, oxoglutaric acid, and salicylic acid, as our data revealed. The carbohydrates transformation into organic acids is more intense during the exothermic than isothermic phase due to temperatures under 40 °C and high content of pulp rich in carbohydrates11,21,39. These conditions facilitate yeast and bacteria growth as Saccharomyces, Candida, Mallasezia, Hanseniospora, Lactobacillus, and Bacillus produce organic acids as part of their central and secondary metabolism13,35,36,40,58.

Other small metabolite groups as phenols, coumarins, terpenoids, fatty acyls, and lipids were also found to varybetween the two phases. These compounds are likely a byproduct of the degradation of complex molecules as polyphenols using the central and secondary metabolism of many microorganisms associated with cocoa fermentations13,40,54. For instance, the yeast of genus such as Mallasezia, previously reported in cocoa fermentation, is highly active in the metabolism of lipids and fatty acyls13. However, as the link between microorganisms-metabolites is still not elucidated, further research is required to dissect the connection between microbial populations and altered metabolites during cocoa fermentations.

Changes in the concentration of relevant metabolites (i.e., flavor precursors and bioactive compounds) were observed during spontaneous fermentation of fine-flavor cocoa

To better understand the development of some sensorial notes associated with fine-flavor cocoa (i.e., fruity, floral, and chocolate notes), we performed a search in the database FlavorDB53 using the list of altered and annotated compounds. We found a rise between the exothermic and isothermic phases of cedrol, irone, acetophenone, coumaric acid, maltol, vanillin, vanillin isobutyrate, and methylcoumarin (Fig. 7) associated with fine-flavor cocoa sensorial notes5,53. Nevertheless, we also observed a significant increase in phenol, coumarin, benzoic acid, and salicylic acid. These molecules produce undesirable flavors previously associated with green, astringent, and bitter attributes (Fig. 8)5,53.

Figure 7
figure 7

Box plots of metabolites linked with fine-flavor cocoa sensorial notes significantly differ throughout fermentation days (one-way ANOVA correcting for false discovery rate, FDR). Linalool xylosyl-glucoside (fruity, floral), irone (floral), coumaric acid (floral), cedrol (floral, woody,sweet), acetophenone (floral), maltol (caramel, fruity), methylcoumarin (fruity,floral), vanillin (vanilla, chocolate), and vanillin isobutyrate (caramel, chocolate, fruity).

Figure 8
figure 8

Box plots of metabolites linked with undesirable cocoa sensorial notes with significant differences throughout fermentation days (one-way ANOVA correcting for false discovery rate, FDR). Salicylic acid (phenolic, faint), phenol (phenolic), benzoic acid (urine, faint), coumarin (green, bitter).

Interestingly, we also observed an increase in some specific flavonoids with documented bioactive properties from the exothermic to isothermic phase. For example, catechin, epicatechin, Resveratrol 3-sulfate, Isoflavonoid O-glycoside, and a group of procyanidins dimers and trimers experienced a considerable increase over fermentation (Fig. 9). These compounds are widely known due to their antioxidant properties8,57. Similarly, other metabolites as dopamine, nicotinamide riboside, and aspartic acid were observed to increase throughout cocoa fermentation. An attractive potential emerges from these molecules because they can act as neurotransmitters59, cardiovascular regulator60, and hormone regulator61, respectively. These bioactive molecules could be produced from complex polyphenol degradation processes that can involve the action of weak acids and the temperature changes into the beans, oxidation reactions, and hydrolysis by polyphenol oxidases62,63 or in the secondary metabolism of some microorganisms64, but the biochemical mechanisms behind their production remain unclear.

Figure 9
figure 9

Box plots of bioactive metabolites with significant differences throughout fermentation days (one-way ANOVA correcting for false discovery rate, FDR).

Several studies associate non-conventional microorganisms’ groups with the production of flavor molecules64,65,66,67. For instance, unconventional strains of Saccharomyces, Candida, Pseudomonas, and Bacillus species, widely reported in cocoa fermentations worldwide, can also produce vanillin and derivates from feluric acid65,66. Different fungi species produce coumarin and derivates naturally64,67. However, to fully connect the metabolite production with the microbiome of cocoa fermentation, an integration of metabolomics and metagenomic data is required, and it will be the focus of future research.

Conclusions

Our work reveals a clear connection between cocoa mass temperature and the metabolic activity during the fermentation of fine-flavor cocoa. Using temperature dynamic as a relevant parameter during cocoa fermentation, we proposed a new cocoa fermentation metabolic paradigm that offers a complete insight into how temperature regulates biochemical reactions during cocoa fermentation, considering global metabolic activity. This shift is a crucial step to develop strategies based on temperature control to drive cocoa fermentations toward better quality chocolate. Nevertheless, further research is required to further dissect the link between microorganisms and metabolites. This is a crucial step in order to understand the impact in the sensorial profile of a significant number of cocoa compounds.

We also elucidated metabolic modifications throughout fermentation associated with proteolysis and secondary metabolism. Our results reveal a potential for bioprospection beyond chocolate production of some peptides and polyphenols with attractive bioactive properties that arise during cocoa fermentation—considering that a significant proportion of these molecules are lost in post-fermentation processes. However, the bioactive properties of most of these identified metabolites also still need to be entirely dissected.