Cured cocoa beans are the principal raw material for the production of cocoa-related products, such as chocolate (Afoakwa et al., 2008; De Vuyst and Leroy, 2020; Santander Muñoz et al., 2020). Curing involves fermentation and drying. These post-harvest processing steps of the cocoa pulp-bean mass, derived from the cocoa fruits of the cocoa tree (Theobroma cacao L.), are decisive for the flavour potential of the cured cocoa beans, as both flavour precursors that are important for further cocoa processing and flavour-active molecules are then formed (Kongor et al., 2016; De Vuyst and Leroy, 2020). The microbial succession during spontaneous cocoa fermentation processes (CFPs) is mainly determined by yeasts, lactic acid bacteria (LAB), and acetic acid bacteria (AAB); their different growth extent and metabolic activities result in complex flavour profiles of both cured cocoa beans and chocolates produced thereof (Schwan and Wheals, 2004; Lima et al., 2011; De Vuyst and Weckx, 2016; Figueroa-Hernández et al., 2019; Santander Muñoz et al., 2020; De Vuyst and Leroy, 2020). In the past, several starter culture-initiated (SCI) CFPs dealt with solely yeast species (Sanchez et al., 1985; Cempaka et al., 2014; Ramos et al., 2014; Meersman et al., 2015; Batista et al., 2016; Menezes et al., 2016), LAB species (Kresnowati et al., 2013), or AAB species (Samah et al., 1993). Other SCI CFPs dealt with starter culture mixtures that combined yeast, LAB, and/or AAB species (Schwan, 1998; Lefeber et al., 2012; Crafack et al., 2013, 2014; Sandhya et al., 2016; Miguel et al., 2017; Moreira et al., 2017; Díaz-Muñoz et al., 2021, 2023). Indeed, knowledge of the most prevalent yeast, LAB, and AAB species and their metabolism has seeded intensive research on candidate starter culture strains and the formulation of mixed-strain starter cultures to initiate CFPs (Lefeber et al., 2010, 2011a,b, 2012; Illeghems et al., 2012, 2013, 2015a,b; Papalexandratou et al., 2013; Batista et al., 2015; Apriyanto, 2017; Visintin et al., 2017; De Vuyst and Leroy, 2020). Consequently, the increasing economic importance of cocoa products and an improved fermentation outcome by applying starter cultures has increased the relevance of SCI cocoa fermentation trials (Saltini et al., 2013; De Vuyst and Weckx, 2016; Pereira et al., 2016; Ozturk and Young, 2017; Castro-Alayo et al., 2019; De Vuyst and Leroy, 2020; Díaz-Muñoz et al., 2021, 2023; Lima et al., 2021).

The yeast species diversity of the fermenting cocoa pulp-bean mass is rather broad compared to that of the LAB and AAB species (Jespersen et al., 2005; Daniel et al., 2009; Papalexandratou and De Vuyst, 2011; Maura et al., 2016; De Vuyst and Leroy, 2020; Díaz-Muñoz et al., 2021, 2023). Previous spontaneous and SCI CFPs regarding yeasts encompassed strains of species of Hanseniaspora, Pichia, and Saccharomyces and have shown that these species are key participants among the yeast communities (Daniel et al., 2009; Papalexandratou and De Vuyst, 2011; Meersman et al., 2013; Papalexandratou et al., 2013; De Vuyst and Weckx, 2016; Maura et al., 2016; De Vuyst and Leroy, 2020; Díaz-Muñoz et al., 2021, 2023). Moreover, the cruciality of strains of LAB and AAB species for successful CFPs has been a subject of debate, although their contributions are relevant (Ho et al., 2015, 2018; Meersman et al., 2015; Moreira et al., 2017; De Vuyst and Leroy, 2020; John et al., 2020). However, given the high inter-and even intraspecies diversity of the yeast communities, yeasts are supposed to have a greater impact on the fermentation extent of the cocoa pulp-bean mass and to influence the cured cocoa bean quality substantially. Indeed, many volatile organic compounds (VOCs) and non-VOCs of flavour impact are produced via the yeast metabolism during fermentation of the cocoa pulp-bean mass, such as higher aldehydes, higher alcohols, and esters (Schwan and Wheals, 2004; Aculey et al., 2010; Rodriguez-Campos et al., 2011, 2012; Ho et al., 2014; Ramos et al., 2014; Koné et al., 2016; Menezes et al., 2016; De Vuyst and Leroy, 2020; Díaz-Muñoz et al., 2021, 2023).

As Hanseniaspora species, such as Hanseniaspora opuntiae, possess a low ethanol and heat tolerance, have a good acid tolerance, and appear to be non-competitive with citrate-converting fructose-loving LAB species (Daniel et al., 2009; Papalexandratou and De Vuyst, 2011; Hamdouche et al., 2015; De Vuyst and Leroy, 2020; Díaz-Muñoz et al., 2023), they frequently occur during the initial phase of CFPs, under low-ethanol and low-pH conditions and in the presence of citrate. They seldomly prevail during the whole fermentation course (Daniel et al., 2009; Papalexandratou et al., 2013; Díaz-Muñoz et al., 2021), as they have to compete for substrates with more ethanol-tolerant yeast species, such as Saccharomyces cerevisiae and Pichia species. Among the latter, Pichia kudriavzevii often prevails toward the end of CFPs, either solely or together with S. cerevisiae (Papalexandratou et al., 2011b,c, 2013; Díaz-Muñoz et al., 2021). Both yeast species are the main contributors to flavour formation upon fermentation (Jespersen et al., 2005; Daniel et al., 2009; Meersman et al., 2013; Pereira et al., 2017; Díaz-Muñoz et al., 2021). For instance, the application of a mixed-strain starter culture composed of S. cerevisiae IMDO 050523 and P. kudriavzevii IMDO 020508 leads the VOC production toward isoamyl acetate or 3-methyl butanal, 2-phenyl ethanol and ethyl decanoate, respectively (Díaz-Muñoz et al., 2021). However, the production of flavour-active compounds by yeast species is often strain-dependent (Meersman et al., 2015, 2016; Papalexandratou and Nielsen, 2016; Pereira et al., 2017; Assi-Clair et al., 2019). Hence, it is not clear to what extent all these yeast species do contribute, either solely or in combination, to the flavour quality of cured cocoa beans and chocolates produced thereof. The investigation of several potential starter culture mixtures, including strains of Hanseniaspora or Pichia species, may help to elucidate this (Pereira et al., 2012, 2017; Ho et al., 2018; Ooi et al., 2020, 2021; Díaz-Muñoz et al., 2021, 2023). Therefore, different strains of both yeast species should be tested in a systematic way, applying both whole-community analysis and fine-scale monitoring of the strains inoculated in the cocoa pulp-bean mass as well as a temporal follow-up of the substrates degraded and metabolites produced, including VOCs, the latter not only in the cocoa pulp and beans but also during cocoa bean processing up to the final chocolates. Fine-scale monitoring has become possible by means of an amplicon sequence variant (ASV) approach (Callahan et al., 2019; Díaz-Muñoz et al., 2021, 2023). However, it is not always straightforward to assign the LAB and AAB ASVs to specific strains when only considering the V4 region of the 16S rRNA gene. Therefore, targeting the full-length 16S rRNA gene and the internal transcribed spacer (ITS1) region of the fungal rRNA transcribed unit will enable monitoring of the inoculated strains in a more accurate fashion (Callahan et al., 2019).

Finally, studies dealing with CFPs performed with the Trinitario cocoa variety are scarce (Cevallos-Cevallos et al., 2018; Ho et al., 2018; Castro-Alayo et al., 2019; Rottiers et al., 2019; Romanens et al., 2020; Verce et al., 2021; Alvarez-Villagomez et al., 2022), in particular when carried out in Costa Rica (Di Mattia et al., 2013; Mayorga-Gross et al., 2016; Díaz-Muñoz et al., 2021, 2023; Verce et al., 2021). Given the fine flavour quality of Trinitario cocoa, the impact of the use of starter cultures may have been underestimated compared to their use for the improvement of the fermentation of bulk cocoa varieties, such as Forastero (Rodriguez-Campos et al., 2011, 2012; Lefeber et al., 2012; Cevallos-Cevallos et al., 2018; Santos et al., 2020; Chagas-Junior et al., 2021). Yet, thanks to an increased knowledge of the functional roles of strains of individual core species for successful cocoa fermentation courses, a detailed investigation is possible now.

Apart from cocoa curing, subsequent processing of the cured cocoa beans, encompassing roasting of the latter and conching of the chocolate batter, co-defines the organoleptic properties and consumers’ acceptance of the final chocolates, as various physical and chemical transformations during chocolate manufacturing will determine their taste and aroma (Afoakwa et al., 2008; Caligiani et al., 2016; Santander Muñoz et al., 2020; Alvarez-Villagomez et al., 2022). For instance, high temperatures and the consequent dehydration of cocoa as a result of cured cocoa bean roasting reduces the concentrations of volatile organic acids, such as acetic acid, whereas non-volatile organic acids, such as citric acid, lactic acid, succinic acid, and oxalic acid, remain mostly untouched. Consequently, the strong acid taste encountered diminishes (Afoakwa et al., 2008). Furthermore, Maillard reactions are of pivotal importance, as they yield aldehydes (e.g., 2-methyl-1-propanal) and pyrazines (e.g., tetramethylpyrazine) that are associated with typical chocolate notes (Caligiani et al., 2007; Humston et al., 2010; Rodriguez-Campos et al., 2011; Farah et al., 2012; Magi et al., 2012; Kadow et al., 2013). Also, compounds associated with floral and roasted notes arise from Maillard reactions, such as phenethyl acetate and phenethyl alcohol from phenylalanine (Ascrizzi et al., 2017; Moreira et al., 2018; Barišić et al., 2019). During conching of the chocolate batter, for which temperature and time are crucial, remaining moisture and VOCs related to sweet, floral, and fruity notes are removed (Owusu et al., 2012, 2013; Rocha et al., 2017). This could be considered undesirable, but the addition of cocoa butter can compensate for the loss of some of these compounds (Souza and Block, 2018). A decrease of the tetramethylpyrazine and benzaldehyde concentrations and a general increase of the aldehyde concentrations characterizes the conching step (Ziegleder et al., 2005; Bastos et al., 2019). For instance, aldehydes usually increase in relative abundance because of alcohol-to-aldehyde oxidations that are promoted through increased oxygen ingress during conching. Aldehydes can be further oxidized to their corresponding organic acids. Finally, acetate esters decrease in relative abundance, while less volatile esters increase (Clark et al., 2020).

The present study aimed at the investigation of the impact of a strain of each of the yeast species H. opuntiae and P. kudriavzevii, different from the strains tested before (Díaz-Muñoz et al., 2021, 2023), on Costa Rican Trinitario CFPs and chocolates made from the concomitant cured cocoa beans. It concerned the microbial community and metabolite dynamics of the cocoa pulp-bean mass fermentation, the metabolite dynamics of the drying cocoa beans, and the VOC profiles of the chocolate production steps. To evaluate a possible strain dependency, the data obtained were compared to those from previous SCI CFPs carried out under the same conditions but in different years (Díaz-Muñoz et al., 2021, 2023).

The current study dealt with the microbial diversity (cocoa pulp-bean mass) and VOC and non-VOC dynamics (cocoa pulps, cocoa beans during fermentation, and drying cocoa beans) of a spontaneous (NC) and five SCI CFPs, differing in the participating yeasts, carried out in duplicate with Trinitario cocoa in plastic vessels in Costa Rica. Besides a well-characterized S. cerevisiae strain, IMDO 050523 (Lefeber et al., 2012; Moens et al., 2014; Díaz-Muñoz et al., 2022), strains of H. opuntiae (IMDO 020003) and P. kudriavzevii (IMDO 060005) were involved, which were different from those used previously, namely H. opuntiae IMDO 040108 (Díaz-Muñoz et al., 2023) and P. kudriavzevii IMDO 020508 (Díaz-Muñoz et al., 2021). Based on the total cocoa pulp-bean mass DNA, a suitable high-resolution ASV approach, taking into account the full-length 16S rRNA gene (bacteria) and the ITS1 region of the fungal rRNA transcribed unit (yeasts), allowed to monitor the starter culture strains added along the fermentation courses. Furthermore, the chocolate-making allowed to determine the impact of the starter cultures used on the non-VOC and VOC profiles along cured cocoa bean processing.

Although biological duplicates did not differ significantly for all CFPs performed, the NC (F1 and F2) fermentation processes displayed the greatest variability, likely due to their spontaneous nature (Camu et al., 2007, 2008). Although the temporal succession of yeasts, LAB, and AAB, which is key for CFPs, is exposed to numerous environmental factors affecting the microbial species diversity and activities, the cocoa variety and post-harvest practices may be of influence too (Schwan and Wheals, 2004; Rottiers et al., 2019; De Vuyst and Leroy, 2020; Díaz-Muñoz and De Vuyst, 2022). The microbial community dynamics influence the pH and temperature courses and vice versa, as exemplified in the NC fermentation processes of the present study. In particular, whereas the increase of the temperature is mainly determined by the extent of AAB growth and will determine the survival of yeasts, the pH course depends on the LAB species diversity and metabolism, as, for instance, the consumption of citric acid in the beginning of CFPs results into a slightly higher pH and the heterofermentation of glucose leads to lactic acid and acetic acid, in turn decreasing the pH. Moreover, the predominance of Liml. fermentum IMDO 0611222 in all SCI CFPs had a substantial impact on the pH and temperature courses.

The spontaneous CFPs were particularly characterized by the growth of mainly Weissella, LAB that are naturally present in cocoa pulp-bean mass but do not prevail in general (Camu et al., 2007; Papalexandratou et al., 2019). As Weissella is not capable to reduce fructose to mannitol, due to the absence of the enzyme mannitol dehydrogenase (Gänzle, 2015), in contrast to Liml. fermentum (Verce et al., 2020), mannitol concentrations are then low and, hence, also acetic acid concentrations. Although A. ghanensis is known to be a fast oxidizer of ethanol to acetic acid, which is however twice as slow as A. pasteurianus (Moens et al., 2014), the F2 fermentation process did not show low acetic acid concentrations. Compared to the SCI CFPs, additional LAB species appeared in the NC fermentation processes besides W. ghanensis, namely Leuc. pseudomesenteroides, Lc. lactis, and Lacp. plantarum. Except for Lc. lactis, these LAB species often belong to the background LAB of fermenting cocoa pulp-bean mass (Camu et al., 2007; Papalexandratou et al., 2011a,b,c, 2013; Moreira et al., 2013; Díaz-Muñoz et al., 2021, 2023). Their presence is the result of environmental contamination, as they can be associated with the cocoa pods, plantain leaves, and even the farmers’ hands and machetes used, all providing an inoculation source for the cocoa pulp-bean mass, whether or not further prevailing during spontaneous CFPs (Camu et al., 2007, 2008; Papalexandratou et al., 2011a,b,c, 2013; Maura et al., 2016; Visintin et al., 2016; Lee et al., 2019; Díaz-Muñoz et al., 2021, 2023). Although Leuconostoc spp. may decrease the proportion of lactobacilli at the beginning of a fermentation course (Corsetti and Settanni, 2007; Jung et al., 2012), they do not survive long-term acidification (Corsetti et al., 2007; Van der Meulen et al., 2007). Yet, their citrate consumption, which is also the case for Weissella (Lefeber et al., 2011b), may explain their (transient) occurrence during CFPs, in casu the NC and PC fermentation processes (Lefeber et al., 2011a; Ouattara et al., 2017; Díaz-Muñoz et al., 2023). Whereas Lacp. plantarum often occurs in spontaneous CFPs (Papalexandratou et al., 2011a,b,c, 2013; Díaz-Muñoz et al., 2023), it does not perform well as a starter culture strain (Lefeber et al., 2010; Moreira et al., 2017). Lactococcus lactis seldomly occurs in spontaneous CFPs (Ostovar and Keeney, 1973).

Keeping the heat inside the fermenting cocoa pulp-bean mass is necessary, as a high temperature is – together with the diffusion of ethanol and in particular acetic acid into the cocoa beans – needed to kill the seed embryo. This was made possible during the vessel CFPs by coverage of the fermenting cocoa pulp-bean mass with plantain leaves, which resulted in an average fermentation temperature well above 40°C (except for the AFSC V fermentation processes) compared to the average maximal fermenting cocoa pulp-bean mass temperature of 37°C for all vessel CFPs carried out without banana leaves cover before (Díaz-Muñoz et al., 2021). Also the flow of sweatings from the vessels, which was promoted in the current study by regular mixing of the cocoa pulp-bean mass and an increased number of openings in the bottom of the vessels, determines the growth and activities of the residing microorganisms, in particular the LAB and AAB. A continuous flow of sweatings makes the fermenting matrix more aerobic, which, in combination with increased lactic acid concentrations produced by LAB, improves the growth of AAB (Camu et al., 2007, 2008; Papalexandratou et al., 2011a,b,c; Díaz-Muñoz et al., 2021). Again, AFSC V with its lower cocoa pulp-bean mass temperature compared to the other CFPs was an exception, which favoured the growth of yeasts and LAB and delayed the growth of AAB, implying no further pH increase after 24 h of fermentation due to the accumulation of lactic acid that was not overoxidized.

Through starter culture addition, more uniform presumptive counts were obtained for the yeast, LAB, and AAB communities and the microbial communities of the SCI CFPs were restricted to a few microbial species. The most abundant yeast genera of the SCI CFPs were Saccharomyces, Hanseniaspora and Pichia, which were part of the starter culture mixtures applied. Concerning the bacterial communities, Tatumella and/or Pantoea species occurred during initial stages of the CFPs and Liml. fermentum and A. pasteurianus along most part of the fermentation courses. The early growth of enterobacteria may contribute to the pulp liquefaction through pectinase activity (Papalexandratou et al., 2011c, 2013, 2019; Illeghems et al., 2015b; De Vuyst and Leroy, 2020). The in-depth ASV analysis showed that Liml. fermentum and A. pasteurianus in the SCI CFPs were mainly represented by the strains IMDO 0611222 and IMDO 0506386, respectively, which proved the successful inoculation of these starter culture strains thanks to their fitness to overrule the background microbiota and their effectiveness to steer CFPs. This confirmed the success of previous on-farm CFPs carried out with strains of these species in Ghana, Ivory Coast, Malaysia, and Costa Rica (Lefeber et al., 2012; Díaz-Muñoz et al., 2021, 2023). Indeed, although the highly fragmented genome assembly of Liml. fermentum IMDO 0611222 did not allow the identification of all 16S rRNA copies present in its genome, five different 16S rRNA gene copies present in the genome of this strain could be recovered by aligning the ASVs corresponding with this species to the genome of the inoculated strain (Verce et al., 2020). Nevertheless, this ASV-based strain-level monitoring is not always achieved with the same resolution, as was the case for A. pasteurianus in the present CFPs. This could be ascribed to a lack of intraspecies resolution of its 16S rRNA gene. As already shown for microbial community profiling by denaturing gradient gel electrophoresis of PCR amplicons targeting the 16S rRNA gene (Díaz-Muñoz et al., 2021), the use of the full-length 16S rRNA genes to infer high-quality ASVs is suitable to monitor AAB at species level but too conserved to resolve intraspecies variability. Nevertheless, care needs to be taken at interpreting relative abundances based on amplicon-based sequences from rRNA genes, especially for eukaryotes. Indeed, the number of copies of rRNA genes (locating the ITS1 region) varies considerably between fungal genera and this can influence any downstream processing (Ganley and Kobayashi, 2007). Additionally, it is challenging to determine the real number of rRNA genes present in fungal genomes, as most of them are highly fragmented. Even with highly contiguous fungal genomes, assembled using third-generation sequencing technologies, it is possible that the number of rRNA genes is underestimated during the assembly, as is the case for the S. cerevisiae IMDO 050523 genome (Díaz-Muñoz et al., 2022). However, the approximate number can be estimated using the sequencing coverage information (Ganley and Kobayashi, 2007; C. Díaz-Muñoz, Q. Vanderauwera, L. De Vuyst, and S. Weckx, unpublished results). Finally, the polyploid, heterozygous nature of fungal genomes makes it difficult to completely haplotype-phase the regions surrounding rRNA genes, which would result in a mixed nucleotide sequence from different alleles. This could be a possible explanation for the fact that only one of the three P. kudriavzevii ASVs showed 100% identity to the inoculated strain. Similarly, the S. cerevisiae 14 ASV that was also present in the CFPs that were not inoculated with the S. cerevisiae IMDO 050523 starter culture strain indicated that this presence had to be taken with care due to the low numbers of S. cerevisiae ASVs retrieved from the samples under study.

The AFSC V fermentation processes were characterized by the prevalence of Limosilactobacillus until their end because of inferior pulp liquefaction, providing less aerobic conditions, due to the sole presence of H. opuntiae as yeast starter culture. The lack of pectinase activity by Hanseniaspora species, avoiding pectin degradation, has been reported before (Seixas et al., 2019; Steenwyck et al., 2019). Nonetheless, pulp degradation could still be mitigated through endogenous enzymes (Camu et al., 2007; De Vuyst and Leroy, 2020; Díaz-Muñoz et al., 2021) or enterobacterial species present at early stages of the fermentation process (Illeghems et al., 2015b). This probably resulted in both an accumulation of non-volatile lactic acid and longstanding anaerobic conditions, thereby preventing the growth of A. pasteurianus IMDO 0506386, as a well-balanced ratio of the microbial consortium members is necessary to allow good AAB growth (Adler et al., 2014). Further, the accumulation of lactic acid probably led to a lower internal pH of the AFSC V cocoa beans. Up to now, few studies dealt with CFPs restricted to growth of H. opuntiae solely (Arana-Sánchez et al., 2015; Visintin et al., 2016; Díaz-Muñoz et al., 2023). Moreover, H. opuntiae has been reported to correlate negatively with A. pasteurianus (Mota-Gutierrez et al., 2018). Furthermore, Hanseniaspora could easily be outperformed by Saccharomyces, as was the case in the AFSC VI fermentation processes, notwithstanding the similar initial cell densities. Indeed, it has been hypothesized that Hanseniaspora growth can be inhibited by Saccharomyces because of cell–cell contact (Pietrafesa et al., 2020). It is likely that Hanseniaspora species lack competitiveness toward S. cerevisiae, and perhaps P. kudriavzevii, in particular because of their low heat and ethanol tolerance and their slow nutrient uptake (Díaz-Muñoz et al., 2021) that explains their occurrence in the initial fermentation phases of CFPs (Daniel et al., 2009; Papalexandratou et al., 2013; Díaz-Muñoz et al., 2021). Also another strain of H. opuntiae, IMDO 040108, as the sole yeast member of a cocoa starter culture mixture, has led to underfermented Costa Rican cocoa beans (Díaz-Muñoz et al., 2023). Despite their low competitiveness, the lack of fructose 1,6-bisphosphatase, phosphoenolpyruvate carboxykinase, and isocitrate lyase could cause the inability of Hanseniaspora species to grow on alternative carbon sources, such as glycerol and ethanol (Casal et al., 2008; Seixas et al., 2019). Finally, Pichia seemed to outperform Saccharomyces in the AFSC VIII fermentation processes, perhaps because of the ability of Pichia strains to produce killer toxins (Branco et al., 2014). A prevalence of P. kudriavzevii instead of S. cerevisiae during fermentation of cocoa pulp-bean mass has been shown before (Papalexandratou et al., 2011c).

The fungal communities present in the fresh cocoa pulp-bean mass, on genus level encompassing – in decreasing order of relative abundance and based on ASVs – Pichia, Saccharomyces, Hanseniaspora, Wickerhamomyces, Acremonium, and Ambrosiozyma are expected to be inoculated from the close environment as their bacterial counterparts. For example, Acremonium, an endophytic fungus, has been isolated hitherto from healthy cocoa leaves, stems and pods, and often represents yet undescribed species because species of this genus represent many anamorphs (Evans et al., 2003; Rojas et al., 2017). Further, the yeast Ambrosiozyma can produce cocoa flavour compounds at a multifold level, among which fusel alcohols (e.g., 2-methyl-1-butanol), ethyl esters (e.g., ethyl isobutryate), and acetate esters (e.g., butyl acetate), in comparison to Saccharomyces yeasts (Gamero et al., 2016). Similarly, species of Wickerhamomyces can be interesting cocoa flavour producers (Koné et al., 2016).

Carbohydrate consumption happened within 48 h of fermentation in all SCI CFPs of the present study, in contrast to the 96 h necessary for the spontaneous CFPs. An accelerated glucose and fructose consumption particularly occurred in the presence of the S. cerevisiae starter culture strain. Indeed, the species S. cerevisiae is known to be a fast carbohydrate consumer and, consequently, a high ethanol producer (Schwan, 1998; Moreira et al., 2013; Batista et al., 2015; Maura et al., 2016). Ethanol is a key metabolite during cocoa pulp-bean mass fermentation, as it has to be cross-fed to the AAB to be oxidized to acetic acid, another key metabolite of CFPs. As H. opuntiae was not able to fulfil a complete alcoholic fermentation as the sole yeast starter culture, particularly due to a slow carbohydrate consumption (AFSC V fermentation processes), it leads to underfermentation, confirming a former study applying another strain of H. opuntiae as the sole yeast member of a starter culture mixture for a SCI CFP of Costa Rican cocoa (Díaz-Muñoz et al., 2023) and former studies on citrus wines (Hu et al., 2020). It probably further explains scarce attempts to use H. opuntiae in cocoa starter cultures (Batista et al., 2015; Ho et al., 2018). In contrast, P. kudriavzevii IMDO 060005 as the sole yeast starter culture as well as in combination with S. cerevisiae IMDO 050523 caused a fast carbohydrate consumption, in contrast with another strain of the same species, P. kudriavzevii IMDO 020508, applied as member of a cocoa starter culture mixture in previous Costa Rican CFPs (Díaz-Muñoz et al., 2021). Given the enhanced prevalence of P. kudriavzevii IMDO 060005, this strain may be a better candidate cocoa yeast starter culture than the other one. Finally, carbohydrate consumption could also be ascribed to the LAB starter culture strain, Liml. fermentum IMDO 0611222, given the increased lactic acid and mannitol concentrations in the SCI CFPs. Also, this strain was responsible for an efficient citric acid consumption thanks to its citrate lyase activity (Bekal et al., 1998; Camu et al., 2007; Lefeber et al., 2011a,b; De Vuyst and Weckx, 2016; Ouattara et al., 2017; Verce et al., 2021). Other organic acids, such as acetic acid and succinic acid, also originated from the citrate metabolism of LAB, although yeasts and Pantoea could be responsible for succinic acid production too (Illeghems et al., 2015b; Dzialo et al., 2017; Figueroa-Hernández et al., 2019; Verce et al., 2020). Whereas acetic acid diffused into the beans, thereby decreasing the internal bean pH, lactic acid was taken up to a lesser extent and citric acid barely or not at all. However, lactic acid favoured AAB growth, which implies an indirect effect on the internal bean pH (Camu et al., 2007; Papalexandratou et al., 2011c). Hence, the final pH increase of the fermenting cocoa pulp-bean mass was the result of diffusion of both acetic acid and lactic acid into the cocoa beans, volatilization because of the heat and mixing of the fermenting cocoa pulp-bean mass, and further oxidation through AAB growth (Schwan and Wheals, 2004; De Vuyst and Leroy, 2020). The consumption of malic acid could be assigned to malolactic fermentation (Díaz-Muñoz et al., 2021), which could be confirmed by the presence of the malic enzyme in Liml. fermentum (Verce et al., 2020). Alternatively, it could be converted via fumaric acid into succinic acid with fumarate hydratase and fumarate reductase (Verce et al., 2020). Indeed, the lack of fumaric acid and an increased concentration of succinic acid throughout the cocoa fermentation course suggested its complete conversion via this pathway in all CFPs. The increase in malic acid concentrations at the end of all CFPs, except for AFSC V, could be attributed to A. pasteurianus, which is capable to produce malic acid (Bartowsky and Henschke, 2008; Illeghems et al., 2013). Finally, the occurrence of gluconic acid in all CFPs may be ascribed to its production from glucose by Tatumella and Pantoea (Kageyama et al., 1992; Adachi et al., 2003), by Gluconobacter (Papalexandratou et al., 2011b), or by A. pasteurianus 0506386 (Illeghems et al., 2013), and was present at high relative abundances in all SCI CFPs, in particular at the end (Moens et al., 2014). As gluconic acid can be consumed by S. cerevisiae (Peinado et al., 2003), and given the absence of this yeast species in the starter culture mixture of the AFSC VII fermentation processes, it occurred in higher concentrations in those CFPs.

The amino acids detected and quantified in the cocoa beans of all CFPs were isoleucine, leucine, valine, alanine, phenylalanine, and tyrosine. However, no significant dynamics in the amino acid concentrations were revealed during fermentation, probably due to the suboptimal pH course taking place in the cocoa beans. These cocoa flavour precursors are the result of aspartic endoprotease activity, releasing hydrophobic oligopeptides and likely optimally active in the initial fermentation phase given its optimal pH of 3.5, followed by exocarboxypeptidase activity, liberating a multifold of hydrophilic peptides and hydrophobic amino acids, optimally taking place at a pH of 5.5–6.0, which was never reached in any of the CFPs (Biehl et al., 1993; Voigt et al., 1994, 2018; De Vuyst and Leroy, 2020).

Although enterobacteria, in casu Tatumella and Pantoea, are known to produce biogenic amines (Delgado-Ospina et al., 2020), the increase of the concentrations of the biogenic amines tryptamine and tyramine in the cocoa beans may be ascribed to the LAB metabolism in the cocoa pulp. This was reflected in their increasing trend with a decreasing cocoa pulp-bean mass pH. An increase of the concentrations of these biogenic amines in cocoa beans has been described before (Oracz and Nebesny, 2014; De Brito et al., 2017). The low tryptamine and tyramine concentrations in the cocoa beans of the AFSC VII and AFSC VIII fermentation processes may be associated with the ability of S. cerevisiae and P. kudriavzevii to degrade biogenic amines in the cocoa pulp (Delgado-Ospina et al., 2021).

Finally, the wide array of VOCs produced in the cocoa pulp-bean mass, either by the plant itself or through microbial biosynthesis (Serra-Bonvehí, 2005; Camu et al., 2007; Moreira et al., 2016; Qin et al., 2017; Utrilla-Vazquez et al., 2020), define the flavour of cured cocoa beans when their uptake through diffusion has taken place (Kadow et al., 2013; Chetschik et al., 2018; Castro-Alayo et al., 2019; De Vuyst and Leroy, 2020). The nature of these VOCs is particularly determined by the yeast activities that have taken place (Ho et al., 2014; Díaz-Muñoz et al., 2021, 2023). Indeed, less variation occurred when H. opuntiae IMDO 020003 was used as yeast starter culture, which confirmed previous findings with the H. opuntiae IMDO 040108 strain (Díaz-Muñoz et al., 2023). Yet, the latter strain has been associated with a dedicated production of the alcohols 2/3-methyl-1-butanol and 2,3-butanediol (Díaz-Muñoz et al., 2021). Whereas VOCs formed through microbial metabolism in the cocoa pulps (in particular higher alcohols and esters) can diffuse into the cocoa beans, migration of ethanol and acetic acid and increasing temperatures cause the liberation of endogenous enzymes that induce flavour precursor formation, which becomes more relevant in chocolate-making processes (Ho et al., 2014; De Vuyst and Leroy, 2020). However, when H. opuntiae IMDO 020003 was combined with S. cerevisiae IMDO 050523 (AFSC VI), more cocoa-related higher alcohols and esters (e.g., phenylethyl alcohol and isoamyl acetate, respectively) were produced. This could pinpoint the importance to always include a strain of S. cerevisiae in a starter culture mixture to ferment cocoa. Yet, studies on wine fermentation have reported an increased production of higher alcohols when strains of the latter yeast species were combined with other yeast species (Luan et al., 2018). Hence, regarding its effect on the flavour profile, the role of H. opuntiae should not be neglected, as the variability to produce VOCs among different H. opuntiae strains is high (Capece et al., 2005; Tristezza et al., 2016). Also, H. opuntiae has been reported to increase the aroma complexity in wine (Jayani et al., 2005; Whitener et al., 2017), as non-Saccharomyces yeasts often have a higher potential to produce enzymes (e.g., esterases) to impact flavour (da Silva et al., 2005; Maturano et al., 2012; Qin et al., 2017). Nonetheless, the genus Pichia is known to contribute significantly to the flavour richness of cocoa (Koné et al., 2016; De Vuyst and Leroy, 2020). The application of P. kudriavzevii IMDO 020508 has shown an increased contribution of higher aldehydes, higher alcohols, and esters in Costa Rican cocoa (Díaz-Muñoz et al., 2021). The less pronounced impact on beneficial cocoa-related VOCs by the P. kudriavzevii IMDO 060005 strain used in the present study could be explained by the strain-dependent flavour potential of this species (Pereira et al., 2017). Apart from the unquestionable importance of yeasts in producing VOCs, also LAB and AAB do contribute to the cocoa flavour (De Vuyst and Leroy, 2020; Díaz-Muñoz et al., 2021, 2023). For instance, benzaldehyde present in the cocoa pulp-bean mass of all CFPs of the present study was probably produced from the amino acid phenylalanine by LAB (Hamdouche et al., 2019; Verce et al., 2020). Further, higher alcohols are also products of the amino acid metabolism by LAB (Smid and Kleerebezem, 2017). Finally, AAB are correlated with acetic acid, acetoin, and even phenylethyl alcohol production (Illeghems et al., 2013; Moens et al., 2014; Mota-Gutierrez et al., 2018; De Vuyst and Leroy, 2020). Next, dedicated compounds associated to specific CFPs in the current study, as explained above for the cocoa pulps, were also revealed in the cocoa beans, which could be ascribed to migration processes, not only ethanol and acetic acid but also glycerol and mannitol as shown in the present study (De Vuyst and Leroy, 2020; Díaz-Muñoz et al., 2021). Moreover, the presence of certain VOCs in the cocoa shells indicates an on-going diffusion process. Finally, tetramethylpyrazine was formed during drying, likely from acetoin (Díaz-Muñoz et al., 2021).

During chocolate-making noteworthy compounds that occurred and/or increased in concentration were 2-phenyl ethanol and phenyl ethyl acetate, which are linked to Maillard reactions typically occurring during roasting (Ascrizzi et al., 2017). The concentration decrease (e.g., tetramethylpyrazine) or disappearance (e.g., benzaldehyde) of some compounds was an indication of conching (Bastos et al., 2019). Conching also resulted in the oxidation of alcohols to the concomitant aldehydes and consecutive organic acids (Clark et al., 2020). Further, volatile aldehydes (e.g., heptanal), furanones (e.g., 4-hydroxy-2,5-dimethyl-3-(2H)-furanone), pyrazines (e.g., trimethylpyrazine), pyrrols (e.g., 2-acetyl-1-pyrroline), and pyridines originated from Strecker degradations (Stadler et al., 2003; Frauendorfer and Schieberle, 2008; Aprotosoaie et al., 2016). Differences in VOC profiles as a result of the starter culture mixtures applied were most distinguishable among the cocoa liquors instead of the final chocolates, indicating a great effect of further cocoa liquor processing, among which the conching step and the addition of commercial cocoa butter.

An amplicon sequence variant (ASV) approach based on full-length 16S rRNA gene (bacteria) and ITS1 (yeasts) sequencing led to more accurate and reliable data concerning precise monitoring of the microbial strains used for starter culture inoculation of cocoa pulp-bean mass. However, starter culture addition led to an enhanced fermentation course if S. cerevisiae IMDO 050523 was present in the mixture inoculated, thereby producing desirable flavour compounds, such as higher alcohols and esters. Hanseniaspora opuntiae IMDO 020003 as the sole yeast strain in the starter culture mixture led to underfermentation and an inferior VOC profile, mainly due to its low competitiveness during fermentation of the cocoa pulp-bean mass. Whilst P. kudriavzevii may contribute to a richer VOC profile, the P. kudriavzevii IMDO 060005 strain tested in the present study did not, confirming strain dependency concerning VOC production of this yeast species. Even though differences in VOCs could be revealed in the cocoa liquors, no significant effect on the final chocolates could be obtained, mainly due to a great impact of cocoa liquor processing during chocolate-making. Hence, optimization of the starter culture mixture and cocoa liquor processing seem to be of pivotal importance.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://www.ebi.ac.uk/ena/browser/view/PRJEB57747.

DV and CD-M performed the field experiments, wet lab work, the statistical analyses, and drafted the manuscript. CD-M performed the culture-independent and bioinformatic analysis. DV performed the metabolite analyses. CH, SW, and LV contributed to the coordination of the field experiments. SW and LV designed and supervised the experimental set-up. DV, CD-M, SW, and LV revised the manuscript. DV and LV edited the manuscript. LV was responsible for funding. All authors read and approved the final version of the manuscript.

This work was supported by the Research Council of the Vrije Universiteit Brussel (SRP7 and IOF3017 projects) and the Research Foundation Flanders (SBO project REVICO, S004617N). DV is the recipient of a PhD fellowship from the VUB.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2023.1232323/full#supplementary-material