Cacao beans are the main raw material used for chocolate production (Schwan et al., 2014), and fermentation of the raw beans is a critical step in the formation of flavor compounds through a cascade of enzymatic and biochemical processes (Santander Muñoz et al., 2020).

Most of the world’s cacao production is linked to the Forastero variety, which is regarded as bulk cacao because of its strong basic cacao character (Rottiers et al., 2019a). Criollo and Trinitario are considered as fine flavor and are widely utilized for the elaboration of specialty chocolate; these varieties are mostly cultivated in South America and Asia (Laliberté et al., 2012; Kongor et al., 2016). Nacional cacao genotype is mostly produced in Ecuador and is characterized by its strong floral notes. Nonetheless, materials of pure Nacional cacao are scarce since the hybrids between Nacional and Trinitario have become prevalent in Ecuador (Samaniego et al., 2020). However, the Nacional x Trinitario complex maintains the unique traits of fine cacao (Rottiers et al., 2019b; Tigrero-Vaca et al., 2022). Beyond their sensory qualities, these fine flavor varieties also command higher economic value in the global cacao market (Parra, 2017).

In addition to genetic factors, post-harvest processing, particularly fermentation, plays a crucial role in determining cocoa bean quality by facilitating the development of flavor precursors (Streule et al., 2024). Years of research led to the conclusion that a proper cacao fermentative process requires a succession of specific yeasts, lactic acid bacteria (LAB) and acetic acid bacteria (AAB) (De Vuyst and Weckx, 2016). Within this frame, most of the studies have utilized amplicon analysis for profiling microbial communities present during Forastero, Criollo and Trinitario cacao fermentation (Camu et al., 2007; Garcia-Armisen et al., 2010; Lefeber et al., 2011; Díaz-Muñoz et al., 2021). In recent years, only a few studies have employed shotgun metagenomic sequencing for assessing the microbial communities during fermentation of Trinitario (Verce et al., 2021), Forastero and Criollo cultivars (Carolina et al., 2021; Mota-Gutierrez et al., 2021). To date, however, no such studies have been conducted on the fermentation of Nacional × Trinitario beans.

Microbial activity during fermentation has traditionally been studied through culture-dependent techniques (Camu et al., 2007; Lagunes Gálvez et al., 2007; Nielsen et al., 2007) and more recently, through next-generation sequencing (NGS) techniques (Mota-Gutierrez et al., 2018; 2021; Verce et al., 2021). Among these, amplicon sequencing, targeting the 16S rRNA gene for bacteria and the internal transcribed spacer (ITS) region for fungi, has been widely used to analyze microbial communities (Nilsson et al., 2019; Gao et al., 2021). However, this approach can introduce bias due to differential amplification efficiencies, which may distort relative abundance estimates (Bista et al., 2018).

Shotgun metagenomic sequencing offers a robust alternative by analyzing the total genomic content of all microorganisms present in a sample without relying on targeted genetic markers (Ranjan et al., 2016). This method provides a more comprehensive and accurate taxonomic resolution, capturing even rare and low-abundance species down to the species level (Brumfield et al., 2020).

Traditionally, shotgun metagenomics has been performed using Illumina platforms due to their high sequencing accuracy and reliability (Serra et al., 2019). However, these platforms are limited by short read lengths, which can introduce analytical bias in complex microbial communities (Kim et al., 2011), and they require extensive laboratory infrastructure, restricting their use in field-based studies.

Oxford Nanopore Technologies (ONT)’s portable MinION sequencer offers a promising alternative. It generates long reads that enable more detailed and contiguous microbial characterization (Shin et al., 2018) and has been applied to diverse studies ranging from bacterial genotyping to antibiotic resistance profiling (Brown et al., 2017; Quick et al., 2017; Kai et al., 2019). Nonetheless, nanopore sequencing has not yet been utilized to investigate microbial communities during the fermentation of Nacional × Trinitario cacao beans.

Understanding the microbial dynamics involved in cacao fermentation is vital for improving post-harvest practices and enhancing bean quality. In this context, the objective of this study was to characterize the microbial communities involved in the fermentation of Nacional × Trinitario cacao beans using both culture-dependent methods and NGS technologies, including Illumina and Oxford Nanopore sequencing platforms.

In the current study, we characterized the microbial communities present during Nacional x Trinitario cacao bean fermentation. For this purpose, three different approaches based on Illumina and Nanopore metagenomic sequencing, as well as Sanger sequencing of individual isolates, were used for characterizing the microbial composition of samples taken at 0 h, 24 h, 48 h, 72 h, and 96 h after the start of the spontaneous fermentation process. Additionally, the role of the microbiota in the fine flavor cacao fermentation process was investigated by performing a functional annotation of the hybrid assembly of Nanopore and Illumina sequencing data.

Culture-dependent assessment followed by Sanger sequencing revealed the presence of several LAB, AAB, yeasts, and spore-forming bacteria (e.g., Bacillus) at the start of fermentation. As fermentation progressed, microbial dynamics became dominated by LAB, followed by AAB. These findings are consistent with previous studies on cacao fermentation (Camu et al., 2007; Nielsen et al., 2007; Camu et al., 2008; Constante Catuto et al., 2024; Camu et al., 2007; 2008; Nielsen et al., 2007; Constante Catuto et al., 2024). Both NGS and Sanger sequencing were consistent in the detection of L. plantarum, L. pseudomesenteroides, L. brevis, L. hilgardii and L. mali. According to Papalexandratou et al. (2011b) and Vuyst and Weckx (2015), several LAB species can utilize citrate as an energy source, thereby shortening the fermentation period and producing flavor precursors via pyruvate metabolism.

Saccharomyces cerevisiae was detected at the onset (0 h–48 h) of fermentation by all three sequencing methods. This yeast’s metabolic activity is closely linked to carbohydrate metabolism, a major pathway contributing to flavor development in cocoa fermentation (Almeida et al., 2020; Santander Muñoz et al., 2020).

At the end of the fermentation process, A. pasteurianus was consistently identified as the dominant AAB species by all methods, in line with prior reports (Lefeber et al., 2011; Soumahoro et al., 2020). Conversely, viral taxa such as L. phage Lfelnf and L. virus Lfelnf were detected only by Illumina and Nanopore sequencing. These viruses are known to infect L. fermentum (Liu et al., 2015; Tochilina et al., 2019), and their presence in cacao fermentation has only been reported in a few studies (Illeghems et al., 2015; Agyirifo et al., 2019; Almeida et al., 2020). Agyirifo et al. (2019) proposed that bacteriophages may positively influence fermentation by lysing LAB cells and releasing enzymes that catalyze the formation of aroma compounds. The detection of these viral taxa adds to the growing but still limited evidence that phages may play an active and underexplored role in shaping microbial dynamics during cacao fermentation.

Additionally, NGS uniquely detected a high abundance of A. ascendens in the final fermentation stage (72 h–96 h). This species has rarely been associated with cacao fermentation, with only one report from Brazil involving Forastero and Trinitario beans (Carolina et al., 2021). Its consistent presence at the late stage of fermentation in our samples suggests it may have a more relevant role in the microbial succession of cacao fermentations than previously recognized. While further studies are needed to elucidate its functional contribution, this finding broadens the known diversity of AABs involved in the process. Together, these observations highlight several rarely described microorganisms in cacao fermentation such as specific bacteriophages and A. ascendens which enhance the novelty and significance of our findings.

It is also important to recognize that Sanger sequencing and NGS are fundamentally different technologies, producing distinct data types. As such, direct comparison is challenging and often requires generating consensus sequences from Nanopore and Illumina reads (Paul et al., 2018; Sahlin et al., 2021; Marshall et al., 2023).

Our findings indicate that throughput in Nanopore sequencing was limited compared to Illumina. Although Nanopore provided valuable insights into the microbial dynamics of cacao fermentation, the relatively low number of reads per sample (average of 1,136) represents a notable limitation. This likely stems from the challenging nature of fermented cacao matrices, which contain polysaccharides, polyphenols, and other inhibitory compounds that interfere with DNA extraction and library preparation quality (Ramos et al., 2014). Despite using the rapid sequencing kit (SQK-RBK004), its high DNA input requirement (400 ng) further constrained sequencing performance. Additionally, Nanopore reads exhibited a mean Phred-like quality score of ∼10.9, reflecting the known higher error rates compared to Illumina. This lower accuracy may reduce species-level assignment precision, particularly for low-abundance taxa (Delahaye and Nicolas, 2021).

While dominant groups such as S. cerevisiae, L. plantarum, and A. pasteurianus were generally detected across all methods, discrepancies were noted in early fermentation stages. Nanopore showed limited detection of LAB at the onset of fermentation, reflecting differences in sensitivity and taxonomic resolution. These inconsistencies highlight the importance of cautious cross-platform interpretation. In this regard, future studies should consider improved DNA extraction methods, newer ONT kits requiring less input, and enhanced bioinformatic strategies such as updated basecallers like ONT’s Dorado and optimized EPI2ME workflows with stronger error correction to improve sequencing quality and accuracy (Buddle et al., 2024; Hong et al., 2024).

Illumina and Nanopore sequencing, analyzed against NCBI and RefSeq databases, revealed similar structural changes in microbial communities over fermentation at order, family, genus and species level showing a prevalence of yeasts, LAB, and AAB, consistent with findings in Criollo, Forastero, and Trinitario fermentations (Camu et al., 2007; Agyirifo et al., 2019; Díaz-Muñoz et al., 2021).

Hierarchical clustering (Figure 1) revealed time-dependent grouping of samples across all taxonomic levels, indicating clear shifts in community structure during fermentation. Early stages (0 h–48 h) clustered separately from late stages (72 h–96 h), reflecting a temporal transition toward communities dominated by fewer, fermentation-adapted taxa. This pattern was consistent across both Illumina and Nanopore datasets.

Notably, Nanopore detected a higher abundance of Enterobacteriaceae throughout fermentation. This group may originate from soil or cacao plant tissues (Papalexandratou et al., 2011a; Almeida Câmara Leite et al., 2013), and similar patterns have been reported in fermented vegetables like Chinese spicy cabbage and Paocai (Cao et al., 2017; Liu et al., 2019). Both platforms showed strong agreement in taxonomic composition at the order and family levels across all time points, as confirmed by correlation and heatmap analysis, aligning with previous studies (Shin et al., 2016; Nygaard et al., 2020). Therefore, while either sequencing approach appears suitable for capturing broad microbial trends during fermentation, higher taxonomic resolution at the genus or species level may require complementary methods.

Differences in performance that resulted in dissimilar alpha diversity indicators between Nanopore and Illumina data were observed (Figure 2). These disparities in alpha diversity likely reflect the inherent sensitivity and limitations of each sequencing platform, Nanopore may miss or misclassify certain taxa due to relatively lower accuracy and throughput observed in this study, while Illumina may better resolve community richness but be constrained by read length (Van Uffelen et al., 2024). One limitation of this study is the limited number of biological replicates per time point, which may affect the resolution of microbial shifts.

Several factors are likely to contribute to discrepancies and weaker correlations (Figures 3 and 4) observed at finer taxonomic levels. First, DNA extraction from fermented cacao is notoriously difficult due to the presence of polyphenols and complex carbohydrates that inhibit enzymatic reactions (Schrader et al., 2012). Such matrix effects may differentially impact extraction efficiency for high-molecular-weight DNA, favoring one platform over another. Second, biases in sequencing chemistry, such as Nanopore’s relatively high error rates or Illumina’s shorter read lengths can influence taxonomic classification accuracy (Stevens et al., 2023).

Third, the use of different bioinformatics pipelines adds further variability. The ONT cloud-based workflow fastq What’s in My Pot (WIMP), as noted by Brown et al. (2017), although suitable for rapid taxonomic profiling, offers less flexibility for metagenomic analysis due to constraints such as limited database customization and fewer options for algorithmic fine-tuning, which can impact taxonomic resolution and quantification accuracy. In contrast, classifiers like Kraken2 allow more customizable and comprehensive reference databases, improving detection and relative quantification of microbial taxa in complex communities. These pipeline-specific differences can result under or overestimation of specific taxa within a sample (Acharya et al., 2019; Winand et al., 2020).

Additionally, only 1.77% of species-level taxa were shared between Illumina and Nanopore platforms (Figure 5), underscoring the limitations of using a single sequencing approach for fine-resolution microbial profiling. This low overlap likely reflects platform specific biases that affect the detection of low-abundance or poorly annotated taxa (Stevens et al., 2023), which in turn may lead to an underrepresentation of microbial diversity and ecological interactions. Therefore, species-level outcomes should be interpreted with caution, and integrating complementary sequencing strategies may be necessary to achieve a more complete representation of microbial diversity.

Llumina and Nanopore sequencing showed that the initiation of fermentation was characterized by diverse microbial genera, including LAB (Lactiplantibacillus, Levilactobacillus, Liquorilactobacillus), AAB (Gluconobacter), and environmental or enterobacterial taxa like Frateuria, Salmonella, and Escherichia. This microbial diversity stems from the cacao pulp’s exposure to external sources such as pod surfaces, soil, machetes, workers’ hands, and fermentation containers (Viesser et al., 2021). At this stage, the abundance of genes associated with pyruvate and pentose phosphate metabolism suggests a constant carbon supply (Agyirifo et al., 2019). LAB showed homolactic metabolism, producing ethanol, acetate, lactate, and acetoin (Gänzle, 2015), and were also involved in fatty acid biosynthesis via acetyl-CoA carboxylase (Gurav and Bokade, 2010). Additionally, Acetobacter, Komagataeibacter and Limosilactobacillus converted pyruvate to oxaloacetate, a precursor of aspartate, through pyruvate carboxylase (De Vries, 2006).

Results showed that, as fermentation progressed other microbial groups started to predominate, for instance, Illumina sequencing revealed that yeasts (Saccharomyces) populations reached a peak at 48 h into fermentation. The presence of this yeast has been widely reported in the fermentation of different cocoa varieties across the world (Jespersen et al., 2005; Moreira et al., 2013; Batista et al., 2015). In this context, previous research (Dzialo et al., 2017) suggest that yeast metabolic activities cause a portion of the carbon getting transported to the Krebs cycle, which in turn allows the formation of aroma precursors by means of a series of biochemical reactions related to the amino acid metabolism. Furthermore, it is worth noting that the abundance of the viral genus Hopescreekvirus increased at this fermentation time point. This genus has one known species and was only recently reported by (Liu et al., 2015; Greiner et al., 2018; González-Orozco et al., 2023).

On the other hand, Nanopore sequencing revealed that enterobacterial genera were predominant over the course of fermentation. Within this microbial group, only a subset of genera and species was identified, namely, E. coli, E. marmotae and S. enterica (Figures 2A,B). Other researchers have also reported the involvement of enterobacteria in the cacao fermentation (Garcia-Armisen et al., 2010; Papalexandratou et al., 2011a; 2013; Hamdouche et al., 2015). Enterobacterial taxa are thought to contribute to glucose conversion into lactic acid and citric acid (Grimont and Grimont, 2006). Nonetheless, their persistent detection at high levels via Nanopore warrants critical interpretation. Rather than solely reflecting a dominant, metabolically active population, this finding may highlight a platform-specific bias. Several technical factors could contribute to this observation.

First, a reference database bias likely influenced the taxonomic classification. Taxa like E. coli and Salmonella are among the most sequenced organisms, with thousands of complete, high-quality reference genomes available in public databases (Brown et al., 2021; Horesh et al., 2021). Bioinformatics classifiers, when faced with the longer but higher error rate reads from Nanopore, may preferentially assign them to these “best-match” genomes over the less-complete or more fragmented genomes of niche fermentative taxa (Winand et al., 2020; Wick et al., 2023). This can create an illusion of high abundance for well-characterized organisms. Second, the DNA from these Gram-negative bacteria may be more efficiently extracted and amplified compared to that from thick-walled Gram-positive bacteria (LAB) and yeasts, further skewing their representation in the sequencing library (De Bruin et al., 2019).

The final stage of fermentation (72 h–96 h) as revealed by Nanopore and Illumina sequencing was characterized by an increment in the relative abundance of AABs including Acetobacter and Komagataeibacter which was consistent with findings from other cacao studies (Serra et al., 2019; Pacheco-Montealegre et al., 2020). The enzyme pyruvate decarboxylase (EC: 4.1.1.1) was attributed to the metabolic activities of these genera and it has been implicated in the decarboxylation of pyruvate to acetaldehyde, a volatile compound known to contribute to fruity aroma notes in fermented products (Peters et al., 2013).

It is worth noting that Illumina sequencing also revealed that Lactobacillus was abundant at 72 h after the start of the fermentation process. LAB are important actors of the fermentative process since they produce lactic acid that diffuses into the seed, which subsequently allows the activation of endogenous enzymes that contribute to the generation of the distinctive chocolate flavor and aroma (De Vuyst and Weckx, 2016; Viesser et al., 2021). Overall, the main genera identified in the present study are in accordance with previous research (Meersman et al., 2013; Bortolini et al., 2016; Carolina et al., 2021).

Amino acid metabolism (Figure 6) emerged as a central functional pathway in our data, strongly linked to the microbial production of volatile flavor compounds (Lima et al., 2022). Genes coding for various enzymes (EC: 2.6.1.19, EC: 1.2.1.16) attributed to various LAB genera that form succinate which then enters the tricarboxylic acid cycle (Sahab et al., 2020; Xia et al., 2022). Additionally, genes related to valine, leucine, and isoleucine metabolism were detected, supporting the microbial synthesis of flavor-active compounds like benzaldehyde and 2-phenylethanol, known for imparting fruity, malty, and floral notes (Pires et al., 2014; Díaz-Muñoz et al., 2021; Quelal et al., 2023).

Moreover, genes attributed to Acetobacter and Limosilactobacillus were associated with the enzyme aspartate aminotransferase, which participates in the biosynthesis of diacetyl and acetoin, volatile compounds that contribute buttery and creamy aromas to fermented cacao (Ardö, 2006; Tian et al., 2020). The detection of glutamate-cysteine ligase in these genera also suggests microbial involvement in the production of γ-glutamyl peptides, which enhance umami and overall flavor complexity (Xie and Gänzle, 2021).

Beyond aroma development, our functional annotation also highlighted pathways relevant to the health-promoting potential of cacao. Genes encoding β-glucosidase enzymes, mainly from Paucilactobacillus and Limosilactobacillus were found to catalyze the release of polyphenol aglycones, reducing astringency and bitterness while enhancing antioxidant availability (Rodríguez et al., 2004; Llano et al., 2025). These enzymes also break down cellulose into glucose monomers, supporting applications like bioethanol production (Harun and Danquah, 2011; Tan and Lee, 2014).

Lastly, genes involved in the biosynthesis of theobromine, a methylxanthine alkaloid with antioxidant and cardiovascular protective effects, were detected, further linking microbial activity to both sensory and nutritional quality (Brunetto et al., 2007; Jean-Marie et al., 2021; Pagliari et al., 2022).

To optimize the application of MinION sequencing in future cacao metagenomic studies, several improvements should be considered. First, selecting a DNA extraction method that yields high molecular weight and inhibitor-free DNA is critical for capturing the full microbial diversity of cacao fermentation. In this sense, incorporating purification techniques, such as AMPure XP bead-based cleanup is recommended, as it enhances DNA quality and optimizes subsequent sequencing outcomes (Angthong et al., 2020).

Second, the use of specialized enzymatic lysis cocktails can improve DNA recovery from hard-to-lyse taxa such as fungi and Gram-positive bacteria, which are often underrepresented (Langsiri et al., 2025). Third, leveraging ONT’s latest flow cells and updated library preparation kits, which support ultralong reads (∼100 kb) with higher basecalling accuracy, may improve taxonomic resolution (Sharma et al., 2025).

Finally, recent bioinformatics tools such as Dorado’s dorado correct (v0.9.1), which integrates the HERRO deep learning algorithm, offer effective raw read error correction, enhancing reliability of species-level assignments in complex communities (Vereecke et al., 2025). Integrating these advances can significantly improve data quality, depth, and interpretability in long-read metagenomics of cacao fermentation.