Exploring microbial dynamics and metabolomic profiling of isoflavone transformation in black and yellow soybean tempe for sustainable functional foods

Abstract

Tempe, a traditional Indonesian fermented food, is rich in bioactive isoflavones and peptides, offering significant health benefits. This study explores how fermentation methods and soybean varieties shape isoflavone profiles and microbial communities. Two fermentation approaches were compared: Raprima™ starter culture and a co-culture of Rhizopus oligosporus and R. stolonifer. Metabolomic analysis showed that co-culture fermentation significantly increased genistein levels and enhanced isoflavone bioavailability. Proteobacteria (78 %) and Firmicutes (18 %) dominated bacterial communities, with yellow soybeans containing more Enterobacteriaceae. Co-culture fermentation enriched Stenotrophomonas, while Raprima™ favored Acinetobacter. The fungal community, primarily Mucoromycota (92 %), exhibited significant correlations with isoflavone transformation. Co-culture fermentation improved microbial synergy and metabolic efficiency, boosting isoflavone aglycone production. While yellow soybeans had higher isoflavone content, black soybeans, with elevated genistein, present a promising alternative. These findings emphasize fermentation's role in enhancing tempe's functionality for sustainable, nutritionally rich food development.

Highlights

• •Klebsiella, Lactococcus, and Staphylococcus correlate with specific isoflavones.
• •Rhizopus species play a crucial role in isoflavone transformation during fermentation.
• •Metagenomic analysis reveals key pathways for isoflavone glycoside conversion.
• •Bacterial-fungal synergy boosts isoflavone aglycone production in fermentation.
• •Enzymes like beta-glucosidase drive isoflavone glycoside hydrolysis efficiently.

1. Introduction

Tempe, a globally recognized traditional Indonesian fermented food, offers a unique combination of high protein content and bioactive compounds (Ahnan-Winarno et al., 2021), positioning it as a promising functional food for addressing modern health challenges (Teoh et al., 2024). Among these bioactive constituents, isoflavones are recognized as key contributors to tempe's functional properties (Frediansyah, 2024; Nurmilah et al., 2024). Isoflavones, secondary metabolites derived from the phenylpropanoid pathway (Shrode et al., 2022), are predominantly found in legumes like soybeans. They exist in two main forms: glycosides, which are bound to sugar molecules, and aglycones, which are the free, unbound state (Khosravi & Razavi, 2021). The aglycone form of isoflavone exhibits higher bioavailability and biological activity, making it the focus of research for its potential health benefits (Sohn et al., 2021).

In their natural state, soybeans and their derivatives primarily contain isoflavones as glycosides (>80 %), such as daidzin, genistin, and glycitin, which have low bioavailability (Azam et al., 2020; You et al., 2015). These glycosides must undergo hydrolysis, catalyzed by specific digestive microorganisms, to transform into bioavailable aglycones like daidzein, genistein, and glycitein (Braune & Blaut, 2016; Kumar, Sasi, et al., 2023). The enzymatic hydrolysis facilitated by microorganisms enhances the conversion of glycosides into aglycones, which are more readily absorbed and exhibit stronger biological activity. Isoflavones in tempe fermentation, primarily contributed by Rhizopus species (Tamang et al., 2025), which convert isoflavone glycosides into bioactive aglycones with health benefits, including antioxidant effects (da Silva et al., 2011; Teoh et al., 2024). Aditionally, equol, a nonsteroidal estrogen derived from daidzein, is produced through a microbial fermentation pathway. Equol demonstrates superior antioxidant and estrogenic properties compared to its glycoside precursor, effectively mitigating oxidative stress and promoting cellular longevity (Legette et al., 2014; X. Zhang et al., 2021). Soybeans enriched with daidzein have also been shown to reduce inflammation and oxidative damage, highlighting the functional potential of isoflavone aglycones for human health (Kojima et al., 2019).

Despite the evident health benefits, only 30–60 % of individuals possess the necessary gut microorganisms to hydrolyze isoflavone glycosides into bioavailable aglycones (Nakatsu et al., 2014). However, glycosides are not entirely non-bioavailable; they can be hydrolyzed by intestinal β-glucosidase enzymes or during fermentation processes, leading to the release of aglycones (Fujita et al., 2015; Qu et al., 2022). Once converted, aglycones are more readily absorbed and exhibit stronger biological effects, such as antioxidant, anti-inflammatory, and estrogenic activity (Nagino et al., 2016; H. Zhang et al., 2020). To overcome this limitation, food engineering innovations are essential to produce isoflavone-rich functional foods. Fermentation technology offers a transformative solution by enhancing the bioavailability and biological activity of isoflavones (Hwang et al., 2018). Fermentation has been particularly effective in increasing the proportion of bioavailable aglycones, with studies demonstrating a significant rise in daidzein and genistein levels following microbial processing (H. Zhang & Yu, 2019). Studies have demonstrated that fermentation with lactic acid bacteria strains, such as Lactobacillus acidophilus, L. casei, and Bifidobacterium longum, increases isoflavone aglycone content via β-glucosidase activity (De Queirós et al., 2020; Leksono et al., 2022; Qu et al., 2022). Similarly, solid-state fermentation using strains like Eurotium cristatum has been shown to effectively convert glycosides into aglycones (Chen et al., 2020a, Chen et al., 2020b). These microbial processes offer promising pathways to overcome the limitations of human microbiota in processing isoflavones, unlocking the full health potential of soy-based foods.

Yellow soybeans are widely used as the primary raw material for tempe production and are a major source of isoflavones. While global production meets much of the demand, many tempe-producing countries, including Indonesia, rely heavily on imported yellow soybeans. This highlights the importance of exploring alternative, locally adapted soybean varieties to enhance sustainability and reduce import dependency. This necessitates exploring alternative legumes. Black soybeans (Glycine max (L.) Merr) are a promising candidate, containing high levels of genistein and daidzein (Juliana et al., 2020; Shabbir et al., 2022). Furthermore, black soybeans are better suited to Indonesia's tropical climate, making them a viable and sustainable option for tempe production (Andajani & Sidhi, 2019; Hizbi & Ghulamahdi, 2019; Setyawan & Huda, 2022; Wijaya et al., 2022). Research findings reveal that black bean tempeh preserves the bioaccessibility of proteins and phenolic compounds, retains antioxidant activity, and exhibits increased ACE-inhibitory activity post-consumption (Wang et al., 2022).

While previous studies have demonstrated the role of fermentation in enhancing isoflavone bioavailability, detailed insights into how diverse microbial systems influence the metabolomic profiles of different soybean varieties during fermentation remain scarce. Our study focuses on Indonesian Mallika black soybeans due to their cultural significance and high isoflavone content. These soybeans generally exhibit higher total phenolic content and antioxidant capacity than yellow soybeans, making them particularly valuable for functional food applications (Mira et al., 2017). Yellow soybeans are included in this study as a commercial benchmark, given their widespread market dominance. Rather than examining the full genetic diversity of soybeans, this study compares these two representative varieties based on their isoflavone composition, which can vary even among soybeans of the same color.

A systematic investigation integrating metabolomic and metagenomic approaches was conducted to elucidate the dynamic interactions between metabolites and microorganisms during fermentation. This research analyzed changes in isoflavone compounds using Raprima™ starter culture and coculture techniques, aiming to optimize the fermentation process. By bridging knowledge gaps, the study contributed to the development of sustainable functional foods that address global challenges in nutrition and food security, while promoting the use of local resources in a climate-resilient manner.

2. Materials and methods

2.1. Sample preparation

The materials used for tempe production were non-genetically modified organisms (Non-GMO) yellow soybean (Rumah Tempe Indonesia), Mallika black soybean (Organics Land, Indonesia), and commercial tempe mold Raprima™ (dried Rhizopus oligosporus NRRL 2771), provided by the Indonesian Institute of Sciences (LIPI – Lembaga Ilmu Pengetahuan Indonesia), which merged with several entities to form the National Research and Innovation Agency (BRIN) in 2021.

Six treatments of soybeans were used in this study, each fermented under different conditions to assess their characteristics. For each treatment, the experiment was conducted with three biological replicates, and these replicates were combined to obtain an average for each replication. Black soybeans, with high total phenolic content, were divided into three groups: BSRP (fermented with Raprima™, a commercial tempeh starter), BSCC (fermented with a spore mixture of R. oligosporus and R. stolonifer), and BS (unfermented, used as a control). Yellow soybeans were also split into three groups: YSRP (fermented with Raprima™), YSCC (fermented with a spore mixture of R. oligosporus and R. stolonifer), and YS (unfermented, control). R. oligosporus and R. stolonifer culture was obtained from Institut Teknologi Bandung culture collection (ITB CC).

Fermentation was performed using black and yellow soybeans, following a similar protocol as described by Yarlina et al. (2024). The soybeans were first selected, ensuring they were clean, without defects, and of uniform color. A total of 150 g of each soybean type was soaked in 1500 mL of tap water at a 1:10 ratio for 36 h. After soaking, the soybeans were washed and boiled in water, then steamed for 45 min. They were drained and cooled before undergoing fermentation. Inoculation was performed by adding a 1.5 % (w/w) dose of Raprima™ tempe starter and 1.5 mL of spore suspension with 10⁸ concentration (R. oligosporus and R. stolonifer), and the soybeans were incubated at ∼30 °C for 36 h. The inoculated soybeans were then placed in plastic bags for incubation. The process flow diagram for tempe production can be seen in Fig. S1.

2.2. Sample extraction for Metabolomic analysis

The extraction process was performed on BSRP, BSCC, BS, YSRP, YSCC, and YS flours, following a modified method adapted from Wang et al. (2021). Samples were weighed and then transferred into microcentrifuge tubes and then transferred into microcentrifuge tubes. To extract metabolites, 1 mL of MS-grade methanol was added to each sample, which was vortexed for 30 s and sonicated at room temperature for 30 min. The mixture was then centrifuged at 1400 ×g for 5 min to separate the supernatant from the pellet. The supernatant was carefully collected and filtered through a 0.22 μm PTFE membrane. The prepared supernatant was then used for LC-HRMS analysis. MS-grade methanol served as the blank control in the metabolomics analysis.

2.3. Metabolomic analysis using LC-HRMS

The analysis utilized LC-HRMS (Thermo Scientific™ Vanquish™ UHPLC Binary Pump) coupled with Orbitrap high-resolution mass spectrometry (Thermo Scientific™ Q Exactive™ Hybrid Quadrupole-Orbitrap™), using a modified approach based on the work of (Windarsih et al., 2022). The chromatographic separation was achieved using a Thermo Scientific™ Accucore™ Phenyl-Hexyl analytical column (100 mm × 2.1 mm ID × 2.6 μm). The mobile solvents consisted of MS-grade water with 0.1 % formic acid (Solvent A) and MS-grade methanol with 0.1 % formic acid (Solvent B). The LC-HRMS method was performed at a flow rate of 0.3 mL/min with a 5 μL injection volume. The gradient program started with 5 % solvent B, which gradually increased to 90 % over 16 min. An isocratic hold at 90 % Solvent B was maintained for 4 min, followed by a re-equilibration from 90 % to 5 % solvent B over an additional 5 min, resulting in a total run time of 25 min. The column temperature was maintained at 40 °C, and a 3 μL injection volume was used. For untargeted screening, full MS/dd-MS2 mode was employed under both positive and negative ionization conditions. Nitrogen served as the sheath, auxiliary, and sweep gases, with settings of 32, 8, and 4 arbitrary units, respectively. The spray voltage was set to 3.30 kV, while the capillary temperature was held at 320 °C, and the auxiliary gas heater temperature was adjusted to 30 °C. The scan range spanned 66.7–1000 m/z, with resolutions of 70,000 for full MS and 17,500 for dd-MS2 in both ionization modes. The overall analytical error is <0.5 %. Additionally, we performed spectral matching and fragmentation analysis using ChemSpider, mzCloud, and GNPS databases to enhance compound identification and intensity measurement accuracy.

2.4. Molecular networking and metabolites annotation

The feature lists derived from all samples were exported from ProteoWizard as separate mzML files. These mzML files were then uploaded to the Global Natural Product Social Molecular Networking (GNPS) online platform, where molecular networks were constructed using the METABOLOMICS-SNETS-V2 workflow. Networks were generated based on a cosine similarity score exceeding 0.65 in both positive and negative ion modes, with a minimum of five matched fragment ions. Edges between nodes were retained only if each node was among the top 10 most similar nodes to the other. The resulting network spectra were matched against GNPS spectral libraries, with identification requiring at least six fragment matches. Cytoscape 3.10.3 was used to visualize the molecular networks.

2.5. Amplicon metagenomic and predictive functional analysis

Amplicon Metagenomic Analysis was employed to investigate the bacterial community composition, following the methodology outlined by Nurmilah et al. (2022) with modifications. DNA was extracted from the samples using the ZymoBIOMICS™ DNA Mini kit (Zymo Research, USA), adhering to standard genomic DNA (gDNA) extraction protocols. To amplify bacterial 16S rRNA and ITS genes, PCR was performed using the MyTaq™ HS Red Mix (Bioline, UK) on a Thermal Cycler (Thermo Fisher Scientific, USA). The V3–V4 region of bacterial 16S rRNA genes was amplified using primers with adapter overhangs and MyTaq™ HS Red Mix (Bioline, UK) on a Thermal Cycler (Thermo Fisher Scientific, USA). Similarly, fungal ITS1–ITS2 regions were amplified using primers with adapter overhangs. The PCR reaction conditions consisted of an initial denaturation at 90 °C, followed by denaturation at 95 °C, annealing at 50 °C, and a final extension at 72 °C. Following amplification, amplicon libraries were prepared using Illumina's two-step PCR protocol, which included the incorporation of Nextera XT tags and indices. The quality and quantity of the libraries were evaluated using multiple methods: the TapeStation 4200 (Agilent Technologies, USA), ddsDNA green Helixyte™, and qPCR with JetSeq Lo-Rox (Bioline, UK). Finally, sequencing was conducted on the Illumina MiSeq platform using a 2 × 300-bp paired-end format.

Illumina sequencing generated raw data through system control and real-time base calling using the Real-Time Analysis (RTA) software. The binary Base Call (BCL) files were converted to FASTQ format with the bcl2fastq tool. Paired-end reads were merged using VSEARCH, filtered for low-quality sequences, and trimmed to remove primer regions, yielding high-quality ITS and 16S rRNA reads. Taxonomic profiling was performed via the QIIME2 pipeline, where non-redundant reads were assigned taxonomy using SILVA for bacteria and UNITE for fungi. DADA2 detected and removed chimeras and denoised the data to generate Amplicon Sequence Variants (ASVs), which were further refined. Predictive functional analysis of bacterial and fungal communities was conducted with PICRUSt2 (Douglas et al., 2020) and FunFun (Krivonos et al., 2023). Functional pathways and genes were annotated using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database to identify potential metabolic pathways and functions.

2.6. Statistical analysis

All data were collected and analyzed using principal component analysis (PCA) and hierarchical cluster analysis (HCA) to explore patterns and relationships within the dataset. Chemometric analysis, including PCA and HCA, as well as statistical evaluations and graphical representations, were performed using OriginPro 2024 software (OriginLab Corporation, USA). PCA visualizations, such as scores and loadings plots, were generated to depict the contributions of individual principal components to the overall model. The isoflavone intensity from each sample was further analyzed using Tukey's test with a significance level of 0.05.

Clustering of both samples and features was performed using the group average method, with Euclidean distance as the similarity metric. The resulting heatmap and dendrograms provided a detailed visualization of the hierarchical relationships within the data. Additionally, the relationship between microorganism abundances and isoflavone intensities was analyzed using Spearman correlation to identify potential associations between microorganism dynamics and isoflavone compound levels.

3. Results and discussion

3.1. Discrimination of samples by Metabolomic profiling

The PCA plot reveals the clustering and separation of metabolite profiles in soybean extracts analyzed using LC-HRMS. The first and second components (component-1 and component-2), representing the primary axes of variance, explain 85,6 % of the total variability in the data (Fig. 1A). PCA was used as an unsupervised pattern recognition technique to identify sample grouping patterns, as well as similarities and differences between samples (Mashiane et al., 2021). The samples were grouped and labeled according to their treatment conditions, including soybean type (yellow or black) and fermentation method (Raprima™ or coculture). This study specifically focuses on Indonesian Mallika black soybeans due to their cultural significance and their typically high isoflavone content. The inclusion of yellow soybeans in our study serves as a commercially relevant benchmark, given their dominance in the soybean market. Importantly, our study aims to assess the functional differences between these two representative types of soybean.

Fig. 1.

(A) PCA plots for the main components and on all extract samples using LC-HRMS data. (B) HCA Heatmap of identified isoflavones in positive and negative modes of group samples.

The PCA of yellow (YS) and black soybean (BS) samples highlights intrinsic biochemical differences, primarily driven by variations in secondary metabolites, particularly isoflavones. Fermentation-induced transformations, including glycoside hydrolysis to aglycones (e.g., daidzein, genistein, glycitein) and microbial synthesis of novel metabolites, result in distinct clustering of fermented (YSRP, YSCC, BSCC) and unfermented (YS, BS) samples. Previous study show that PCA has been consistently used to differentiate chemical profiles, aroma compounds, and metabolites across various fermented soybean products, serving purposes such as quality control, identification of key compounds, and monitoring of fermentation progress (He et al., 2024; Q. Tan et al., 2024; Yang, Wang, et al., 2020). Microbial specificity plays a crucial role in shaping metabolite profiles (Park & Kim, 2020; G. Tan et al., 2022). Raprima™-fermented samples (YSRP, BSRP) exhibit consistent metabolic signatures, whereas co-culture fermentation (YSCC, BSCC) produces more diverse profiles due to enhanced enzymatic interactions. PCA further reveals that component-1 differentiates soybean types, while component-2 captures fermentation effects. Metabolite correlation vectors indicate that those associated with fermented clusters reflect biochemical modifications, whereas unfermented clusters retain compounds inherent to raw soybeans.

LC-HRMS analysis visualized via HCA heatmaps confirms that fermentation significantly alters metabolite distribution, leading to increased bioactive compound concentrations (Fig. 1B). Unfermented samples (YS, BS) display a homogeneous profile with minimal enzymatic activity, while most fermented samples (except BSRP, which aligns closely with raw samples) show a heterogeneous isoflavone distribution. Fermentation enhances bioactive metabolite production, particularly antioxidant-rich compounds in black soybeans and elevated isoflavones aglycones (e.g., daidzein, genistein) in yellow soybeans. Previous studies suggest that enzymatic hydrolysis and fermentation enhance metabolites with new functional properties, while untreated soybeans retain a homogeneous profile with minimal enzymatic activity (Z. Liu et al., 2024; Toy et al., 2020; Yang, Qu, et al., 2020). Co-culture fermentation amplifies these effects through microbial synergy, yielding higher concentrations of functional metabolites, as indicated by the heatmap (purple denotations). Coculture fermentation produces higher concentrations of bioactive compounds compared to fermentation with a single microbial strain (Hu et al., 2024; W. Zhang et al., 2023). These findings underscore the metabolic complexity introduced by fermentation, with co-culture strategies demonstrating superior bioactive compound enrichment.

3.2. Isoflavone dynamics

The intensity data from LC-HRMS reveals the distribution of eleven major isoflavones—daidzin, genistin, and glycitin—in both their aglycone and glucoside forms across black and yellow soybean samples subjected to different fermentation methods, as well as in unfermented controls (Fig. 2). Notably, acetyl glycitin was not detected in this study. While acetyl glycitin may be present at trace levels below the LC-HRMS detection limit, its absence could also stem from database limitations, ion suppression, or data processing thresholds. Manual inspection of raw m/z data confirmed no detectable peak matching its expected mass and retention time, suggesting it was either absent or below the method's sensitivity. In the unfermented samples, glucosides predominate, underscoring their role as the primary storage forms of isoflavones, with yellow soybeans exhibiting slightly higher intensities than black soybeans. This disparity suggests intrinsic compositional differences between the two soybean varieties. Fermentation markedly alters the isoflavone profiles, particularly by enhancing the intensities of aglycones (Hwang et al., 2018).

Fig. 2.

Comparison of Isoflavone Aglycone and Glucoside Peak Area in Black and Yellow Soybeans: Effects of Raprima™ and Coculture Fermentation. Data are presented as mean ± S.D. Different letters indicate significant differences between groups according to Tukey's test at p < 0.05. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Co-culture fermentation systems (BSCC and YSCC) result in the highest concentrations of daidzein and genistein aglycones, indicative of superior enzymatic hydrolysis of glucosides and potential de novo synthesis of bioactive metabolites. Microbial interactions in co-culture can activate silent gene clusters, enhancing metabolite synthesis (Li et al., 2025). Moreover, division of biosynthetic pathways between strains improves efficiency (Thuan et al., 2023). Interestingly, malonyl glycitin was not detected in any of the co-culture samples. The use of a mixed spore inoculum (R. oligosporus and R. stolonifer) maybe resulted in a distinct enzymatic profile compared to the commercial starter. This suggests that enzymes from R. stolonifer likely catalyzed the breakdown of the malonyl group in malonyl glycitin, converting it into other isoflavonoids such as glycitin or genistin. Previous studies revealed that fermentation with lactic acid bacteria such as Lactiplantibacillus plantarum and Streptococcus thermophilus increased daidzein levels by up to 107 % and genistein by up to 132 % in black soybean milk (Leksono et al., 2022), while fungal fermentation with Eurotium cristatum boosted daidzein by up to 10.4 times and genistein by up to 8.4 times compared to non-fermented soybeans (Chen et al., 2020).

Additionally, fermentation conditions—including pH, temperature, and duration—may have influenced the stability of the compound, accelerating hydrolysis or decarboxylation processes that led to the absence of detectable malonyl glycitin. Similar to previous studies, black soybeans fermented at 50 °C showed the most significant increase in daidzein and genistein aglycone levels (Wu & Chou, 2009). While black soybeans display lower overall aglycone levels, they still benefit from co-culture fermentation, demonstrating improved production of bioactive compounds compared to single-strain fermentation with Raprima™. These findings highlight the microbial synergy within co-culture systems, which appear more efficient at breaking down glucosides into aglycones. Malonyl glycitin and other malonyl isoflavones are highly heat-sensitive, rapidly degrading through decarboxylation and deesterification into acetyl-glucosides, regular glucosides, and eventually aglycones (Lee et al., 2022). Additionally, pH changes during fermentation can influence enzyme activity and the stability of malonyl glycitin (Lim, Lim, Kim, Kwon and Eom, 2020a, Lim, Lim, Kim, Kwon and Eom, 2020b).

Moreover, yellow soybeans consistently exhibit higher glucoside and aglycone intensities than black soybeans across all experimental conditions, likely attributable to inherent differences in the chemical matrix and substrate availability of the raw materials. The increased daidzein intensity in the YSRP sample suggests that fermentation facilitated the hydrolysis of isoflavone glycosides, including malonyl and acetyl derivatives, contributing to its higher levels. The presence of the Raprima™ microbial starter likely enhanced β-glucosidase activity, accelerating the conversion of daidzin to daidzein and resulting in a more pronounced peak area. Notably, genistein intensities are elevated in yellow soybeans fermented with co-culture, indicating selective microbial activity that favors the production of specific isoflavones. These results highlight the potential of co-culture fermentation to optimize isoflavone profiles and highlight its capacity to enhance the bioactivity of black soybeans, thereby reducing the compositional disparity between the two soybean varieties.

3.3. Molecular network insights

The GNPS molecular network analysis emphasize the significant impact of fermentation on the diversity and complexity of metabolites in soybean samples (Fig. 3A and B). Clusters observed in both positive and negative ionization modes distinctly separate unfermented and fermented samples. Prominent isoflavone derivatives, including daidzein, genistein, glycitein, as well as their malonylated and acetylated forms, were identified within the molecular networks (Fig. 3B). We identified eleven isoflavonoids, while acetyl glycitin was likely undetected due to its extremely low intensity in our samples. Notably, metabolites such as daidzein (C₁₅H₁₀O₄, m/z: 254.241) and genistein (C₁₅H₁₀O₅, m/z: 270.240) were markedly enriched in all fermented samples, with co-culture fermentations exhibiting the highest concentrations. This increase is likely attributed to the hydrolysis of isoflavone glucosides during fermentation, mediated by microbial enzymatic activity. Additionally, metabolites such as 6“-O-malonylglycitin and 6”-O-acetyldaidzin were uniquely associated with fermentation conditions, indicating enhanced transformations driven by microbial interactions. Additionally, metabolites such as 6”-O-malonylglycitin and 6”-O-acetylglycitin were uniquely associated with fermentation conditions, indicating enhanced transformations driven by microbial interactions.

Fig. 3.

(A) GNPS molecular network analysis shows the diversity and complexity of metabolites in soybean samples; (B) GNPS molecular network analysis shows the impact of fermentation on the diversity and complexity of isoflavones cluster in soybean samples.

During the early stages of fermentation, microbial enzymes such as malonyltransferase and acetyltransferase facilitate the addition isoflavone 7-O-glucosides (Yu et al., 2008). Isoflavone 7-O-glucoside-6“-O-malonyltransferase is crucial for the production of malonylated isoflavone. Malonyltransferase derived from Glycine max has been employed to introduce malonyl groups at the 6” position of the glucose moiety (Koirala et al., 2014). This enzymatic transformation enriches the chemical diversity of isoflavones without cleaving the sugar bond attached to the isoflavone core structure.

During the malonylation process, the enzyme malonyltransferase catalyzes the addition of a malonyl group (-CO-CH₂-COOH) to the hydroxyl group (-OH) on the sugar moiety of glycitin (Ahmad et al., 2017; Y. Liu et al., 2017). This reaction, which is controlled by microorganisms during fermentation, requires malonyl-CoA (malonyl coenzyme A) as the donor of the malonyl group. The resulting product, 6“-O-malonylglycitin, is a derivative with enhanced stability, improved solubility, and modified biological properties compared to the unmodified glycitin. Similarly, during acetylation, acetyltransferase enzymes facilitate the transfer of an acetyl group (-COCH₃) to the sugar component of glycitin, utilizing acetyl-CoA (acetyl coenzyme A) as the acetyl donor. This modification results in the formation of 6”-O-acetylglycitin, which alters the compound's polarity and bioavailability, potentially enhancing its functional activity. Both malonylation and acetylation are critical biochemical transformations that occur during fermentation, leading to the production of bioactive derivatives with distinct properties and health-promoting potential.

The co-culture fermentation process demonstrated a broader enzymatic activity spectrum, resulting in more diverse and complex metabolite profiles compared to single-culture fermentation. This emphasizes the transformative effect of fermentation in enhancing the production of bioactive metabolites, particularly under co-culture conditions that foster synergistic microbial interactions. The distinct metabolite profiles observed in black and yellow soybeans further highlight the importance of selecting specific soybean varieties for targeted functional food applications.

In the molecular network analysis, the isoflavone cluster revealed several unidentified metabolites potentially related to isoflavones. These signals may result not only from the presence of intermediate isoflavone derivatives that have not yet been cataloged in the GNPS database but also from analytical limitations such as in-source fragmentation, adduct formation, or ion suppression. This highlights the need for further investigation and validation. Additionally, we identified compounds within the isoflavone cluster, such as 5′-deoxy-5′-methylthioadenosine and luteolin-7-O-glucoside. These metabolites are likely secondary compounds produced through the same or related metabolic pathways, which may account for their co-occurrence in the analyzed samples. Moreover, these compounds could play roles in interconnected or mutually influential biological pathways, further supporting the complexity of the metabolic processes occurring during fermentation.

3.4. Microorganisms profile

The bacterial abundance at the phylum level is shown in Fig. 4. Proteobacteria and Firmicutes were the dominant bacteria identified in each sample, while Cyanobacteria and Bacteroidota were present in minor quantities. These findings are consistent with previous studies, which report that the dominant bacteria in tempe are Lactobacillus from Firmicutes, with the bacterial community being primarily influenced by the soybean soaking process rather than the starter culture used (Penido et al., 2013). Additionally, the bacterial abundance in tempe made from yellow soybeans was more dominant compared to that made from black soybeans. This suggests that the raw material plays a significant role in determining the diversity and abundance of bacteria identified in tempe samples.

Fig. 4.

Abundance profiles of pyhlum level; (A) bacteria (B) fungi in samples fermented using Raprima™ and Coculture.

Among the fungal communities, all samples were predominantly composed of the phylum Mucoromycota (98 %), with minor contributions from other phyla, including Ascomycota, Basidiomycota, and unidentified fungi. This observation is consistent with prior studies, which have shown that the fermentation process tends to enhance the abundance of Mucoromycota while diminishing the presence of Ascomycota and Basidiomycota (Yarlina et al., 2024). Interestingly, the phylum Ascomycota was more frequently detected in the BSRP samples, suggesting a distinctive microbial composition associated with this fermentation method.

Although this study employed the commercially available starter Raprima™, whose fungal components have been previously characterized, metagenomic analyses revealed notable differences in fungal composition post-fermentation, which were dependent on the raw materials used. These findings suggest that variations in both the raw materials and the starter culture play a significant role in shaping the fungal community involved in tempe fermentation. The fungal composition of tempe is thus influenced not only by the starter culture but also by the substrate serving as the raw material, as highlighted by previous research (Damayanti et al., 2021; Feng et al., 2007).

Fig. 5A and B illustrate the variations in microbial composition across the fermented samples in genus and species level. Overall, the bacterial composition in all samples was predominantly dominated by the Enterobacteriaceae family, although the specific genus could not be identified (Fig. 5A). Enterobacteriaceae levels were higher in yellow soybean samples (45 % and 38 %) compared to black soybean samples (32 % and 27 %). Tempe fermented using the co-culture system showed a notable presence of bacteria from the genus Stenotrophomonas (28 % and 21 %), second only to the Enterobacteriaceae family. In contrast, tempe fermented with Raprima™ exhibited a higher abundance of bacteria from the genus Acinetobacter (23 % and 18 %). This pattern was consistent across both yellow and black soybean samples, highlighting the significant influence of the fermentation starter culture on the bacterial abundance profile.

Fig. 5.

Abundance profiles of genera level microorganisms; (A) bacteria (B) fungi in samples fermented using Raprima™ and Coculture.

Commercial tempe samples predominantly exhibited high levels of Enterobacteriaceae, with 67 % of the samples surpassing 10⁵ CFU/g (Samson et al., 1987), suggesting that these bacteria are a ubiquitous component of the tempe microbiota. Ilham et al. (2021) further elucidated that Enterobacteriaceae, including Klebsiella, are primarily introduced from the raw soybean substrate. Although the soybean boiling process prior to fungal inoculation can reduce bacterial populations, these bacteria rapidly proliferate during the fermentation phase. Notably, some Klebsiella isolates in tempe were genetically identical to those derived from soybeans, reinforcing the notion that soybeans are the primary reservoir for this bacterium. Concurrently, Stenotrophomonas, a bacterium known for producing fibrinolytic enzymes (Putra, 2018), has been identified in tempe. Previous studies have highlighted its potential applications in plant growth promotion and biocontrol (Deng et al., 2022; Horch et al., 2023; Kumar, Rithesh, et al., 2023; Maraolo et al., 2023; Zhang et al., 2022; Zhao et al., 2024), suggesting that Stenotrophomonas may originate from the raw materials and establish a mutualistic relationship with the fermentation starter culture during the fermentation process.

Fig. 5B presents a comprehensive analysis of the metagenomic ITS amplicon data from tempe samples derived from Raprima™ and co-cultures, revealing the presence of three distinct species within the genus Rhizopus: R. microsporus, R. arrhizus, and an unidentified Rhizopus species. These findings emphasize the microbial diversity of Rhizopus species involved in the tempe fermentation process. Notably, R. arrhizus is the currently accepted name for R. oryzae, as established by taxonomic revisions favoring the original designation by Fischer . Additionally, R. microsporus var. oligosporus, commonly known as R. oligosporus, is recognized as a domesticated variant of R. microsporus, widely utilized in tempe fermentation.

However, the close genetic similarity between R. stolonifer and R. arrhizus within the same clade complicated their differentiation, highlighting a challenge in species-level resolution. The results suggest that tempe fermentation hosts a more intricate microbial community than previously recognized, with contributions from species beyond the target strains. The misidentification of R. stolonifer as R. arrhizus using ITS markers illustrates the limitations of this approach in distinguishing closely related species. Previous studies have noted the high genetic similarity among R. stolonifer, R. arrhizus, and R. delemar, further complicating species identification at the molecular level (Ahmad et al., 2017; Y. Liu et al., 2017). Despite these challenges, our findings emphasize the essential role of Rhizopus as a primary microbial player in tempe fermentation. To overcome these limitations, we advocate for the use of more advanced techniques, such as whole-genome sequencing, to identify additional genetic markers that can improve species-level resolution.

Most samples were dominated by the R. microsporus group, particularly in the YSRP sample, where its relative abundance reached 99.82 %. In contrast, black soybean samples fermented with the same starter exhibited a notable presence of Fusarium equiseti at 8.26 %. Previous studies have established R. microsporus (or R. oligosporus) as the predominant mold species in tempe fermentation using Raprima™, a dried R. oligosporus starter combined with ingredients such as Allium cepa and Oryza sativa (Yarlina et al., 2024). On the other hand, Fusarium equiseti is a well-known plant pathogen responsible for causing wilting in crops (Bibi et al., 2024; Tziros et al., 2022), suggesting that its presence in black soybean samples may be attributed to contamination from the raw material. Notably, the introduction of coculture significantly reshaped the fungal community composition, particularly by reducing the prevalence of Fusarium equiseti.

The use of coculture significantly suppressed Fusarium equiseti populations in black soybean samples, concomitant with an increase in the dominance of R. arrhizus. In both YSCC and BSCC samples, R. arrhizus exhibited considerable dominance. Specifically, in YSCC, R. arrhizus accounted for 93.72 % of the fungal community, with R. microsporus representing 5.65 %. In contrast, R. microsporus remained the predominant species in BSCC, comprising 76.90 % of the community, while R. arrhizus constituted 22.62 %. These findings underscore the role of coculture in not only suppressing pathogen proliferation but also promoting fermentation by enhancing enzymatic activity, which parallels the function of R. microsporus. This suppression is consistent with findings that coculturing systems can reshape microbial communities and enhance the abundance of beneficial fungi while reducing pathogenic Fusarium species (Chang et al., 2022).

The contrasting dominance of R. arrhizus in yellow soybean (YSCC, 93.72 %) versus black soybean (BSCC, 22.62 %) samples accentuate the significant influence of substrate composition on fungal community dynamics. Yellow soybeans appear to provide a more favorable environment for R. arrhizus growth, while black soybeans better support R. microsporus (76.90 % in BSCC compared to 5.65 % in YSCC). Variations in substrate properties, including protein content, fiber composition, and bioactive compounds, likely play important roles in shaping these microbial communities. This insight can guide the selection of appropriate substrate and starter combinations, optimizing fermentation processes and enhancing the microbiological quality and safety of the final product.

3.5. Metabolites and microorganisms correlation

Fig. 6A shows that several bacteria exhibit a significant positive correlation with various isoflavone components. For example, Klebsiella shows a positive correlation with glycitein, Lactococcus is positively correlated with glycitin, while Staphylococcus exhibits a significant positive correlation with genistin and daidzin. These significant positive correlations suggest that these bacteria may contribute to the formation of the metabolites produced. Lactococcus sp. and Klebsiella sp., identified as the most abundant species in over-fermented tempe, had relative abundances of 26.3 % and 13 %, respectively (Pangastuti et al., 2019). This finding is further supported by a previous study revealing that Klebsiella sp. contributes to the production of daidzein and genistein during soybean fermentation for tempe production (Kustyawati et al., 2020).

Fig. 6.

Spearman Correlation Analysis between the Abundance of Top Ten Metabolites, Isoflavones, and Dominant (A) Bacteria (B) Fungi Identified in Samples Fermented with Raprima™ and Coculture.

On the other hand, several bacteria exhibit a significant negative correlation with certain isoflavone components. For instance, Sphingobium and Stenotrophomonas show a significant negative correlation with genistin and daidzin. This negative correlation suggests that the abundance of these bacteria is associated with a decrease in the metabolites produced. Stenotrophomonas maltophilia exhibits proteolytic activity, meaning it can produce protease enzymes to break peptide bonds in protein molecules (Yarlina et al., 2023). This activity is important for enhancing the protein and amino acid content in tempe, although its relationship with isoflavone biotransformation remains unclear.

In Fig. 6B, Most Rhizopus groups exhibited a positive correlation with isoflavone aglycones, including daidzein, genistein, and glycitein. Notably, R. arrhizus showed a stronger positive correlation with genistein compared to other isoflavone aglycones. R. arrhizus also exhibited a significant negative correlation with several types of isoflavone glycosides, including daidzin, genistin, malonyl daidzin, and acetyl daidzin. This suggests a potential role of R. arrhizus in the conversion of isoflavone glycosides into aglycones. These results indicate that, in general, the Rhizopus group plays a role in isoflavone formation, as supported by transcriptomic results showing that Rhizopus is associated with GABA and isoflavones (Tamang et al., 2025). Meanwhile, yeast groups such as Trichosporon asahii, Candida parapsilosis, Candida spp., Purpureocillium spp., and Meyerozyma guilliermondii tended to exhibit a negative correlation with isoflavone aglycones. This aligns with the significant correlation observed between Candida parapsilosis and nearly all types of isoflavone glycosides. However, Purpureocillium exhibited a relatively strong positive correlation with genistein. These correlations suggest an association between fungal and yeast species and isoflavone biosynthesis, although the precise mechanisms underlying these interactions remain unclear.

3.6. Predictive functional pathways and genes analysis related to Isoflavone transformation

Further investigation was conducted to explore the potential roles of various pathways and enzymes in the bioconversion of isoflavones, despite the limited direct evidence linking the isoflavone pathway to microbial dynamics. This analysis was performed using the metagenomic prediction software PICRUSt2 and FunFun in conjunction with the KEGG database. The KEGG-based PICRUSt2 and FunFun analysis provided comprehensive insights into microorganism pathway and gene function expression. Fig. 7 highlights several pathways (level 3) associated with beta-glucosidase production, providing insights into the potential role of microorganisms in the transformation of isoflavone aglycones during fermentation. All identified pathways related to beta-glucosidase production generally belong to the carbohydrate metabolism group. Notable pathways include starch and sucrose metabolism (ko00500), amino sugar and nucleotide sugar metabolism (ko00520), glycolysis/gluconeogenesis (ko00010), and cyanoamino acid metabolism (ko00460) (S4 and S5).

Fig. 7.

Classification of predicted pathway of bacterial (B) and fungal (F) related to beta-glucosidase production identified in samples fermented with Raprima™ and coculture. (A) Starch and sucrose metabolism, (B) Amino sugar and nucleotide metabolism, (C) Glycolysis/Gluconeogenesis, and (D) Cyanoamino acid metabolism.

The starch and sucrose metabolism pathway generates oligosaccharides, which act as optimal substrates for beta-glucosidase induction, with enzymatic regulation varying across microorganisms (Deflandre & Rigali, 2022). The type of substrate plays a important role in enzyme production. For instance, substrates like specific oligosaccharides induce higher beta-glucosidase production compared to others (Liao et al., 2014). The glycolysis/gluconeogenesis pathway supplies essential energy and precursors for enzyme biosynthesis while also influencing carbon source availability through glucose hydrolysis (Xiong et al., 2011).

The amino sugar and nucleotide sugar metabolism pathway are crucial for generating precursors required for cell wall synthesis and nucleotides, supporting both the biosynthesis and activity of beta-glucosidase. Studies using substrates such as para-nitrophenyl-beta-D-glucopyranoside (pNPG) have demonstrated that enzyme activity, as indicated by initial velocity, depends on substrate concentration. Increased substrate availability enhances enzyme activity up to a saturation point (Klimeš et al., 2016; Kong et al., 2019). Moreover, the metabolism of cyanoamino acids and the synthesis of the enzyme beta-glucosidase are interrelated within multiple biological processes. In cyanoamino acid metabolism, beta-glucosidase participates in the degradation of molecules originating from cyanoamino acids, enabling the liberation of glucose and other sugars. For instance, BglC was identified via the KEGG pathway as a potential beta-glucosidase implicated in cyanoamino acid metabolism. BglC exhibits behavior that cannot be elucidated just by its action on its native substrates, cellobiose and cellotriose. A prior study examined if this beta-glucosidase would act on additional carbohydrates including a terminal glucose linked by a β-1,4 bond (Deflandre & Rigali, 2022). The interaction among these pathways establishes a strong metabolic basis for microorganisms to enhance the conversion of isoflavone compounds during fermentation. Further research is required to pinpoint specific genes within these pathways to clarify their exact functions.

Fig. 7 illustrates the relative abundance differences in significantly identified pathways between bacterial and fungal groups. Across all treatments, the fungal group appears to dominate significantly over bacteria, indicating that the transformation of isoflavone aglycones during tempe fermentation was predominantly driven by fungal activity. Nevertheless, the interaction between bacterial and fungal groups plays a crucial role in the transformation process. This is evident from the variations in abundance observed in the combinations of bacteria and fungi across the different fermentation treatments.

Bacterial groups in fermented samples using Raprima™ exhibit similar abundance compared to co-culture samples. Conversely, in the fungal group, the yellow soybean YSCC sample shows greater abundance than the YSRP sample, whereas the black soybean BSCC sample has lower abundance compared to YSRP. This variation highlights the influence of both raw materials and the type of starter culture used. The co-culture starter demonstrates a synergistic interaction between bacteria and fungi in yellow soybean, leading to a higher intensity of isoflavone aglycones. This synergy is primarily due to the complementary metabolic activities of the microorganisms involved. Increases in isoflavone aglycones are achieved through the deglycosylation of glycoside precursors and the upregulation of biosynthetic pathways induced by fungal elicitation (Jiao et al., 2017). This is evidenced by the consistent superior isoflavone profiles observed in samples fermented with co-culture starters, further supporting the existence of microbial synergy.

The interaction between fungi and bacteria in co-cultures can lead to enhanced microbial dynamics. These interactions can result in more efficient systems compared to monocultures (Espinosa-Ortiz et al., 2021; Pozdnyakova et al., 2023). In some co-cultures, there is an increase in enzyme activities and electron transfer processes, which can enhance the overall metabolic output and efficiency of the system (Chen et al., 2020a). However, differences in raw materials as microbial substrates are linked to specific nutritional components, such as certain polysaccharides, which can influence microbial activity (Hernández & Hobbie, 2010; Huang et al., 2023; Solvang et al., 2023; Song et al., 2020).

The transformation pathway of isoflavone glycosides into aglycones involves a series of hydrolysis steps mediated by various enzymes, which are predicted to be produced by different microorganisms during the fermentation process (Fig. 8). The initial step entails the cleavage of glycosidic bonds in branched isoflavone glycosides by oligo-1,6-glucosidase (EC:3.2.1.10), which specifically acts on α-1,6 linkages. Subsequently, alpha-glucosidase (EC:3.2.1.20) hydrolyzes α-1,4 linkages in linear isoflavone glycosides, converting them into aglycones. For isoflavones modified by phosphate addition during fermentation, 6-phospho-beta-glucosidase (EC:3.2.1.86) cleaves phosphoglucan bonds, releasing the aglycone. Maltose-6′-phosphate glucosidase (EC:3.2.1.122) can further break down more complex phosphate derivatives, facilitating additional conversion into aglycones.

Fig. 8.

Predicted metabolic pathway of isoflavone glycoside to aglycone transformation by microorganism enzymes identified in samples fermented with Raprima™ and coculture.

Beta-glucosidase (EC:3.2.1.21), encoded by bglX and bglB, plays a crucial role in hydrolyzing β-glycosidic bonds in various isoflavone glycosides, expediting the release of sugar residues. These genes were identified in several bacteria, fungi, and yeast strains (Magwaza et al., 2024), including genera Gilbertella, Mucor, Rhizomucor, and Rhizopus that could produce β-glucosidase (Lim et al., 2020b; Mei et al., 2019; Takó et al., 2010). Bacillus velezensis S141 and Bifidobacterium pseudocatenulatum IPLA 36007 also could produce β-glucosidase. In B. velezensis S141, bglA encodes β-glucosidase, an enzyme crucial for hydrolyzing isoflavone glycosides into aglycones (Kondo et al., 2023). Similarly, in B. pseudocatenulatum, this gene potentially supports β-glucosidase activity, facilitating the breakdown of isoflavone glycosides (Alegría et al., 2014). The bglX gene encodes a glycoside hydrolase in Pseudomonas aeruginosa, whose inactivation has been shown to reduce biofilm formation by targeting periplasmic glucans (Mahasenan et al., 2019). In the case of isoflavones bound to complex polysaccharides, glucan 1,6-alpha-glucosidase (EC:3.2.1.70) and glucan 1,3-beta-glucosidase (EC:3.2.1.58) break down glucan linkages that obstruct access to the aglycone core. Once all glycosidic bonds are cleaved by these enzymes, isoflavone aglycones such as daidzein, genistein, and glycitein are released.

Meanwhile, the contribution of malZ to isoflavone transformation during fermentation is likely indirect, stemming from its role in encoding maltodextrin glucosidase (Kim et al., 2022). This enzyme cleaves maltodextrin, ranging from maltotriose to maltoheptaose, into shorter oligosaccharides, with maltose and glucose as final products (Tapio et al., 1991; TapioSt et al., 2001). Although malZ does not directly participate in breaking down isoflavone glycosides, the glucose produced by its enzymatic activity can enhance fermentation efficiency and support the production of isoflavone aglycones. While PICRUSt2 and FunFun provide valuable insights into microbial functional potential, they rely on reference genome databases that may not capture the full diversity of microbial communities, especially underrepresented or novel taxa. Moreover, they predict gene functions based solely on genomic data, without accounting for post-transcriptional regulation or environmental factors. Therefore, experimental validation is strongly recommended.

4. Conclusion

This study highlights the influence of fermentation methods and soybean varieties on microbial communities and isoflavone profiles in tempe. Coculture fermentation with R. oligosporus and R. stolonifer significantly enhanced the bioavailability of aglycones, particularly genistein, when compared to the Raprima™ starter culture. Microbial analysis revealed a predominance of Proteobacteria (78 %) and Firmicutes (18 %), with yellow soybeans exhibiting higher levels of Enterobacteriaceae. Coculture fermentation also enriched Stenotrophomonas, whereas Acinetobacter dominated in Raprima™ fermentation. Spearman correlation analysis identified positive associations between Klebsiella and glycitein, Lactococcus and glycitin, and Staphylococcus with genistin and daidzin. Among fungi, Mucoromycota comprised 92 %, with R. arrhizus predominating in coculture samples. The latter showed significant negative correlations with near all isoflavone glycoside, likely due to its capacity to convert these compounds into aglycones. Co-cultures fermentation demonstrating enhanced microbial synergy and metabolic efficiency, leading to superior isoflavone aglycone production through complementary interactions and substrate-specific influences. While fermented yellow soybeans contained higher isoflavone levels, black soybeans, with their elevated genistein content and better adaptability to tropical climates, present a promising alternative to yellow soybeans.

Supplementary Materials

• •S1 - Metabolomics
• •S2 - Bacterial Abundance
• •S3 - Fungal Abundance
• •S4 – PICRUSt2 Prediction
• •S5 - FunFun Prediction
• •Supplementary Figure

Declaration of AI-assisted technologies in the writing process

During the preparation of this work, the authors used ChatGPT to improve English language and readability. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

CRediT authorship contribution statement

Siti Nurmilah: Writing – review & editing, Writing – original draft, Visualization, Investigation, Data curation. Andri Frediansyah: Writing – review & editing, Validation, Supervision, Conceptualization. Yana Cahyana: Writing – review & editing, Supervision. Roostita L. Balia: Writing – review & editing, Supervision. Bibin Bintang Andriana: Writing – review & editing, Validation. Gemilang Lara Utama: Writing – review & editing, Validation, Supervision, Software, Resources, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material 1: Metabolomics

Supplementary material 2: Bacterial Abundance

Supplementary material 3: Fungal Abundance

Supplementary material 4: PICRUSt2 Prediction

Supplementary material 5: FunFun Prediction

Supplementary material 6: Supplementary Figure

Data availability

Data will be made available on request.