Introduction

Soybean-based fermented foods have gained global popularity not only for their distinctive flavors but also for their remarkable health-promoting properties (An et al., 2023). Diverse fermentation methodologies developed across different regions have produced a wide range of soybean-derived fermented products, each with unique aromatic profiles, textural characteristics, and functional properties. Examples include Chinese Douchi (Zhao et al., 2025), Sufu (Liu et al., 2025), and Korean Ganjang (Lim et al., 2025). Within China, Sojae Semen Praeparatum (SSP) represents a special type of Douchi. It is uniquely prepared through natural fermentation combined with herbal juices. This traditional Chinese medicinal product has dual functionality, serving both therapeutic purposes and culinary applications (Li et al., 2021). Previous studies have shown that SSP contains typical bioactive compounds, such as flavonoids, polyphenols, and organic acids, commonly found in other fermented soybean products (Lin et al., 2021). In addition, SSP possesses a complex profile of volatile components (Guo et al., 2024; Wang et al., 2024). These key chemical constituents (isoflavonoids, polyphenols, organic acids, and volatile compounds) and critical physicochemical properties (e.g., pH, total acids (TAs), ammonia nitrogen (AN), and moisture content)) are recognized as fundamental indicators reflecting the fermentation progress and product quality of Douchi (Sui et al., 2024; Liu et al., 2020). Collectively, these chemical constituents provide the essential material basis for SSP’s therapeutic efficacy in treating fever-related conditions. Therefore, monitoring the dynamic changes of these specific parameters during SSP fermentation is crucial to understand the underlying biochemical processes and establish quality control benchmarks.

Microorganisms play a significant role in shaping these compositional changes during SSP fermentation. For instance, Xiong et al. (2019) have identified nine bacterial strains capable of producing γ-aminobutyric acid in SSP, while Zhao et al. (2022) demonstrate that co-fermentation with Rhizopus chinensis 12 and Bacillus sp. DU-106 substantially increases isoflavone content. More recently, Guo et al. (2024) and Wang et al. (2024) link core bacterial genera to the formation of SSP flavor compounds. Conventional microbial sequencing, however, provides only relative quantification, which limits the analysis of microbial dynamics. This limitation is particularly critical because key processes, such as koji-making and washing, can drastically alter total microbial biomass. Relying solely on relative abundance may introduce systematic bias by ignoring absolute differences in microbial load across samples. To address this, we applied absolute quantitative sequencing (AQS) using synthetic spike-in sequences. These spike-ins were added to samples at known concentrations and processed together with sample DNA. Microbial absolute copy numbers were then calculated by calibrating target reads against spike-in reads, enabling direct measurement of microbial load and unbiased comparisons between samples in spite of variations in biomass (Yao et al., 2024).

In spite of the well-established fermentation process of SSP, significant regional variations persist in the microbial composition and abundance under natural fermentation conditions. For example, Zhu et al. (2015) have identified Bacillus subtilis and Aspergillus spp. as the dominant microorganisms during the koji-making stage of SSP in Jiangxi Province using Biolog technology. In contrast, Liu et al. (2021) report Neurospora crassa and Aspergillus spp. as the dominant fungal taxa at the same stage in Fujian Province based on ITS sequencing. Studies from Chengdu by Guo et al. (2024) and Wang et al. (2024) further highlight geographical divergence: Nostocaceae and Enterobacteriaceae dominate the koji-making stage, whereas Leuconostoc and Aspergillus prevail post-fermentation. These findings collectively underscore substantial regional heterogeneity in SSP fermentation microbiota, which can directly affect the stability and consistency of the final product. Identifying the core microbial community in SSP is therefore essential for developing relatively controlled inoculated fermentation processes. To address the current research gap, which has focused primarily on southern China, this study selected Beijing—a representative city in northern China—as the research site. This approach allows for comparison with previous studies while significantly expanding the documented diversity of SSP fermenting microbiota. The results not only enhance our understanding of potential core starter cultures but also provide more options for inoculated fermentation aimed at improving quality control.

In short, given the critical role of microbial succession in shaping the metabolic profile and final quality of fermented foods, this study aims to elucidate key microbial and functional changes and their potential interactions during SSP fermentation in Beijing, China. We employed AQS to characterize the temporal succession of bacterial and fungal communities and conducted comprehensive profiling of functional components, such as isoflavones, total polyphenols (TPPs), total saponins (TSs), and volatile compounds. Furthermore, this study seeks to reveal the potential associations between microbial community dynamics and the evolution of SSP functional profiles throughout the fermentation process.

Materials and Methods

Fermentation of SSP and sampling

SSP was fermented following the protocol described by Li et al. (2014) at the Beijing University of Chinese Medicine, Laboratory C101, Department of TCM Resources and Identification. The process (Figure 1) consisted of the following steps: A mixture of Mori Folium (90 g) and Artemisia Annuae (100 g) was decocted twice in 2 L of distilled water, each for 1 h. The combined extracts were concentrated to 1 L. Black beans (1 kg) were soaked in the herbal decoction for 6 h, steam-cooked for 1.5 h, and cooled to room temperature. The beans were then transferred to rattan trays, covered with three layers of sterile cotton gauze and residue herbs, and incubated during the koji-making stage at 30 ± 2°C and 75% relative humidity for 6 days, until a yellow mycelium and spore layer, commonly known as yellow cladding, was formed. Following winnowing and washing, the beans were placed in glass jars for post-fermentation at 30 ± 2°C for 18 days. Finally, SSP decoction pieces were obtained after steaming for 0.5 h and drying at 60°C.

Figure 1. Manufacture process of Sojae Semen Praeparatum (SSP).

Triplicate samples (~200 g each, n = 3) of fresh SSP were collected at predefined intervals from the central regions to avoid edge effects: every 2 days during the koji-making stage (days 0, 2, 4, 6; designated K0, K2, K4, K6) and every 3 days during the post-fermentation stage (days 0, 3, 6, 9, 12, 15, 18; designated P0, P3, P6, P9, P12, P15, P18). The final steamed product was labeled PS18. Half of the fresh SSP samples were flash-frozen in sterile polyethylene bags and stored at −80°C for DNA extraction and subsequent microbial community analysis, while the remaining samples were used for physicochemical characterization.

To ensure process consistency, five consecutive SSP batches were produced prior to sampling. Product quality was assessed according to the Chinese Pharmacopoeia (Ch. P.) 2020 (Commission, 2020) through evaluations of sensory characteristics, purity test, and quantification of bioactive compounds (genistein and daidzein).

Determination of physiochemical properties

The content of TAs was determined by acid-base titration using sodium hydroxide (Liang et al., 2019). AN content was measured using the formaldehyde titration method (Zhao et al., 2019). Water-soluble extracts (WSE), alcohol-soluble extracts (ASE), and moisture content were determined following the Chinese Pharmacopoeia (Ch. P.) 2020 (Commission, 2020). pH values were recorded using a pH meter (FiveEasy Plus, Mettler Toledo Instruments Co., Ltd., Shanghai, China) according to Kim et al. (2018). All assays were performed in triplicates (n = 3).

The content of TPPs in SSP was determined using a modified Folin–Ciocalteu method (Gao et al., 2000). Briefly, 1 g of SSP powder was ultrasonically extracted with 20 mL of 80% (v/v) methanol. Following centrifugation, the extraction was repeated, and the combined supernatants were then vacuum-dried at 40°C. The residue was reconstituted in 25 mL of 80% (v/v) methanol and centrifuged, and the resulting supernatant was designated as the test solution. For analysis, 0.1 mL of the test solution was mixed with 2.9 mL of water and 0.5 mL of 1 mol/L Folin–Ciocalteu reagent. After standing for 1 min, 0.5 mL of 15% (w/w) sodium carbonate solution was added, and the mixture was incubated in the dark at 25°C for 3 h. The mixture was subsequently diluted to 10 mL with water, and absorbance was measured at 760 nm. A standard curve was prepared using gallic acid solutions ranging from 10 to 50 mg/mL. Each SSP sample was analyzed in triplicate (n = 3).

Total polysaccharides (TPSs) were extracted and quantified using a modified phenol-sulfuric acid method (Liu et al., 2022). Briefly, 2 g of SSP powder was defatted by soaking in 10 mL of petroleum ether (b.p. 30 ~ 60°C) for 30 min, and the supernatant was discarded; this process was repeated thrice. After complete solvent evaporation, the powder was reflux-extracted twice in 100 mL of distilled water for 1 h each. The combined extracts were filtered, concentrated under vacuum at 60°C, and redissolved in distilled water to a final volume of 25 mL. A 5 mL aliquot was precipitated by adding four volumes of ethanol. After centrifugation, the precipitate was dissolved in 10 mL of distilled water. Deproteinization was performed by mixing the solution with 1/4 volume of Sevag reagent (chloroform: n-butanol = 4: 1, v/v), followed by centrifugation at 5000 × g for 5 min. The supernatant was collected, and the Sevag treatment was repeated thrice. The final polysaccharide extract was stored at 4°C prior to analysis. TPSs’ content was determined using the phenol-sulfuric acid method. Briefly, 0.1 mL of polysaccharide extract was diluted to 2 mL with distilled water, mixed with 1 mL of 5% (w/v) phenol, and then rapidly mixed with 5 mL of concentrated sulfuric acid. The mixture was incubated in a boiling water bath for 10 min, cooled immediately in ice water, and absorbance was measured at 485 nm. A standard curve was prepared using glucose solutions ranging from 10 to 50 mg/mL. Each SSP sample was analyzed in triplicate (n = 3).

TSs’ content was determined using a modified vanillin-phenol-perchloric acid method (Wang et al., 2014). Briefly, 1 g of SSP powder was defatted with petroleum ether as described for polysaccharide extraction. The defatted powder was ultrasonically extracted twice with 20 mL of ethanol, followed by centrifugation. The combined supernatants were concentrated under reduced pressure at 45°C. The concentrate was dissolved in 20 mL of distilled water and partitioned thrice with an equal volume of water-saturated n-butanol in a separatory funnel. The combined n-butanol phases were vacuum-dried at 60°C. The residue was dissolved in 25 mL 80% (v/v) ethanol and stored at 4°C prior to analysis. For quantification, 0.2 mL of the test solution was evaporated to dryness at 60°C. After cooling, the residue was dissolved in 0.2 mL of 5% (w/v) vanillin in glacial acetic acid, and 0.8 mL of perchloric acid was added. The reaction mixture was incubated in a 70°C water bath for 15 min, cooled in and ice bath, and then diluted to 10 mL with ethyl acetate. The solution was kept on ice, and absorbance was measured at 550 nm within 1 h. A standard curve was established using oleanolic acid solutions ranging from 10 to 50 mg/mL. Each SSP sample was analyzed in triplicate (n = 3).

UPLC analysis of isoflavones

The six isoflavones (daidzein, daidzin, genistein, genistin, glycitein, and glycitin) in SSP were quantified using ultra-performance liquid chromatography (UPLC). Briefly, 1 g of SSP powder was ultrasonically extracted with 25 mL of 75% (v/v) methanol at 40°C for 40 min and centrifuged at 6000 × g and 4°C for 15 min. The supernatant was filtered through a 0.22 μm nylon membrane before UPLC analysis. Accurately weighed reference standards (10 mg each) were dissolved in methanol to prepare individual stock solutions (250 μg/mL). Working calibration standards were prepared by combining appropriate aliquots of stock solutions and preforming serial dilutions with methanol. UPLC analysis was performed on an ACQUITY H-Class system (Waters Cor., Milford, MA, USA) equipped with an ACQUITY UPLC BEH C18 column (2.1 mm × 100 mm, 1.7 µm; Waters Corp., Dublin, Ireland). The mobile phase consisted of 0.1% formic acid in acetonitrile (solvent A) and 0.1% formic acid in water (solvent B). Gradient elution was carried out as follows: 0 ~ 5 min, 12 ~ 20% A; 5 ~ 8 min, 20% A; 8 ~ 13 min, 20 ~ 35% A; 13 ~ 15 min, 35% A; 15 ~ 16 min, 35 ~ 50% A; 16 ~ 17 min, 50 ~ 12% A; 17 ~ 18 min, 12% A. The flow rate was 0.35 mL/min, the detection wavelength was 260 nm, the column temperature was 30°C, and the sample temperature was 20°C. The injection volume was 1 μL. Each sample was analyzed in triplicate (n = 3).

GC-MS analysis of volatile compounds

Volatile metabolites were analyzed using gas chromatography-mass spectrometry (GC-MS). One gram of SSP powder was spiked with 348.8 μg of methyl undecanoate (internal standard) and extracted with 20 mL ethyl acetate under ultrasonication (250 W, 40 min). The extract was filtered through a 0.22 μm nylon membrane prior to GC-MS analysis. The analysis was performed on a GC-MS system (7890B GC coupled with 5977C MSD, Agilent Technologies Inc., Santa Clara, CA, USA). Separation was achieved on a BP-Inowax-MS capillary column (30 m × 0.25 mm, 0.25 μm; NanoChrom Technologies Co., Ltd., Suzhou, China) using helium as the carrier gas at a constant flow rate of 1.0 mL/min. The inlet temperature was maintained at 230°C, and the hydrogen, air, and tail blow gas flow rates were set to 30, 300, and 25 mL/min, respectively. Samples were injected at a volume of 1 μL with a split ratio of 50: 1. The oven temperature program followed the conditions listed in Table 1. The mass spectrometer operated with an ion source temperature of 230°C, a transfer line temperature of 240°C, and a scan range of 40 ~ 500 m/z. Each SSP sample was analyzed in triplicate (n = 3).

Extraction of microbial genome and sequencing

Microorganisms in the SSP samples were pre-enriched as described by Xu et al. (2020), followed by genomic DNA extraction using the Fast DNA^TM SPIN Kit for soil (MP Biomedicals, California, USA). The preparation procedure was as follows. Briefly, 2 g of fresh SSP sample was homogenized with 6 mL of sterile washing solution (0.1 mol/L Tris-HCl, pH 8.0, 0.25% Tween 80) by vortexing for 2 min. After centrifugation (200 × g, 4°C, 5 min), the pellets were resuspended and washed once under the same conditions. The combined supernatants were filtered through a single layer of sterile gauze and centrifuged again (10,000 × g, 4°C, 10 min). The resulting pellet was weighed and processed according to the manufacturer’s instruction for DNA extraction. All solutions were membrane-filtered (0.22 μm) and stored at 4°C. Biological replicates were processed in quintuplicate (n = 5).

DNA integrity was assessed by 1% agarose gel electrophoresis. DNA concentration and purity were measured using a fluorometer (Qubit 3.0, Thermo Fisher Scientific Inc., Waltham, MA, USA) and a spectrophotometer (NanoDrop 2000, Thermo Fisher Scientific Inc., Waltham, MA, USA). Multiple spike-ins were artificially synthesized, each containing the same conserved regions as natural 16S rRNA genes but with variable regions replaced by random sequences with approximately 40% GC content. A spike-in mixture with known gradient copy numbers was then added to the DNA samples at appropriate ratios. For bacterial sequencing, the V3-V4 hypervariable regions of the 16S rRNA gene and spike-ins were amplified using primers 341F (5’-CCTACGGGNGGCWGCAG-3’) and 805R (5’-GACTACHVGGGTATCTAATCC-3’). For fungal sequencing, the ITS1 region and spike-ins were amplified using primers (5’-CTTGGTCATTTAGAGGAAGTAA-3’) and (5’-GCTGCGTTCTTCATCGATGC-3’). The resulting PCR products were verified, purified, and quantified, after which sequencing was performed on a NovaSeq 6000 platform (Illumina Inc., San Diego, CA, USA) by Genesky Biotechnologies Inc., Shanghai, China (Jiang et al., 2019).

Statistical analysis

The content (mass fraction) of each physicochemical property was calculated based on the sample mass and analyzed using Origin 2021 software (OriginLab Corp., Northampton, MA, USA). Volatile compounds were identified by comparing mass spectra with those in the NIST 14.0 database. Quantification of volatile components was performed by normalizing chromatographic peak areas to the internal standard (methyl undecanoate) with a known concentration. Statistical analyses were conducted using SPSS 20.0 (IBM Corp., Armonk, NY, USA). Group differences were evaluated using one-way multivariate analysis of variance, with p-values indicating statistical significance. Mean values of the measured indicators were used for fold change (FC) calculations and correlation analyses.

For microbial analysis, QIIME 2 was used to process the raw sequencing data (Bolyen et al., 2019). After trimming adapter and primer sequences using the Cutadapt plugin, the DADA2 plugin was applied to denoise the sequences, identify amplicon sequence variants (ASVs), exclude low-quality reads and potential chimeras, and merge paired-end sequences (Callahan et al., 2016). Taxonomic annotation of the resulting ASVs was performed using the Naive Bayes classifier with reference to the SILVA v.138 database. After identifying the spike-in sequences and quantifying their read counts, a standard curve was generated for each sample by plotting the read counts against the known spike-in copy numbers. This standard curve was then used to calculate the absolute copy number of each ASV in the sample based on its corresponding reads. Because the spike-in sequences were exogenous and did not represent members of the native microbial community, they were removed from subsequent analyses. All ASVs for each species were systematically recorded using Microsoft Excel 2021. Diversity analyses were performed using R software (version 3.5.1), whereas microbial composition analysis, Spearman correlation analysis, and principal component analysis (PCA) were conducted using Origin 2021.

Results

Quality evaluation of five batches of SSP

SSP samples collected at different time points were shown in Figures S1 and S2. Characteristics of the five batches were presented in Figure S3. All samples exhibited similar shape and size, with brownish-black, rugose surfaces featuring a black elongated oval hilum. The texture was soft yet slightly brittle. The cross-section appeared reddish-brown to dark brown with an oily luster. Samples exhibited a strong sauce aroma and a slightly sour-sweet taste. The biuret color reaction for purity testing yielded negative results (Figure S4). Total daidzein and genistein contents across batches were 0.169%, 0.202%, 0.172%, 0.173%, and 0.152%, respectively.

Changes of physicochemical properties

UPLC chromatograms of the six isoflavones were presented in Figure 2. Validation routines of UPLC were provided in Text A1. During fermentation, the contents of isoflavone glycosides decreased (Figure 3A), while those of free isoflavones increased (Figure 3B). TPPs increased from 0.34% (K0) to 1.56% (P18), and then decreased slightly to 1.37% (PS18) after steaming. TSs increased during the koji-making stage, while TPSs remained relatively stable throughout the fermentation process (Figure 3C). TAs, AN, WSE, and ASE showed increasing trends over time (Figure 3D; Figure 3E). The pH values decreased during fermentation, showing an inverse pattern relative to TAs (Figure 3F). Moisture content decreased during the koji-making stage (Figure 3G) and remained at approximately 55% during post-fermentation. PCA showed that PC1 (14.5%) and PC2 (74.5%) together explained 89% of the total variance (Figure 4A). The loading plot (Figure 4B) presented the contributions of genistein, daidzein, WSE, ASE, TAs, AN, and TPPs to PC1.

Figure 2. UPLC chromatogram of six isoflavones in SSP sample P12 (post-fermentation day 12). Peaks: a, daidzin; b, glycitin; c, genistin; d, daidzein; e, glycitein; and f, genistein.

Figure 3. Physicochemical dynamics of SSP during fermentation. (A) Three isoflavone glycosides. (B) Three free isoflavones. (C) Total polyphenols (TPPs), total polysaccharides (TPSs), and total saponins (TSs). (D) Total acids (TAs) and ammonia nitrogen (AN). (E) Water-soluble extracts (WSE) and alcohol-soluble extracts (ASE). (F) pH. (G) Moisture content. K0, K2, K4, and K6: samples collected at 0, 2, 4, and 6 days during the koji-making stage; P0, P3, P6, P9, P12, P15, and P18: samples collected at 0, 3, 6, 9, 12, 15, and 18 days during the post-fermentation stage. PS18: steamed product after 18 days of post-fermentation. Data represent mean ± SD, n = 3.

Figure 4. Principal component analysis (PCA) of physicochemical profiles of SSP during fermentation. (A) PCA score plot with 95% confidence ellipses. (B) PCA loading plot. K0, K2, K4, and K6: samples collected at 0, 2, 4, and 6 days during the koji-making stage; P0, P3, P6, P9, P12, P15, and P18: samples collected at 0, 3, 6, 9, 12, 15, and 18 days during the post-fermentation stage. PS18: steamed product after 18 days of post-fermentation. n = 3 for each sample.

Changes of volatile components

The total ion chromatogram (TIC) presented in Figure 5 showed the identification and quantification of 54 volatile components, including 13 organic acids, 16 esters, 3 alcohols, 12 amines, 2 aldehydes, 3 ketones, 1 phenol, and 4 other components (Figure S5). The total volatile content increased during fermentation, reaching 17.36 mg/g (P9), followed by a decline during post-fermentation and a further decrease to 9.55 mg/g after steaming (PS18). Except for esters, all volatile categories showed reductions in total content following steaming. Organic acids decreased from 15.26 mg/g to 7.55 mg/g, whereas ester content increased from 0.47 mg/g to 1.05 mg/g (Table S1).

Figure 5. Total ion chromatogram (TIC) of SSP sample P12 obtained by GC-MS. IS denotes the inner standard (methyl undecanoate).

Variations across different stages were presented in the scatter plot (Figure 6). After the koji-making stage (Figure 6A), 37 components were significantly upregulated (FC > 2, p < 0.01). Following the post-fermentation stage (Figure 6B), 20 components were significantly upregulated (FC > 2, p < 0.01). After steaming (Figure 6C), 5 components increased, while 18 components, including amines and organic acids, significantly decreased (FC > 2, p < 0.01). An additional threshold, requiring component concentrations at each stage endpoint to exceed 10 μg/g, was applied to identify key metabolites, which were marked in the bubble plot (Figure 7). After the koji-making stage (Figure 7A), 20 key metabolites were identified, including 9 organic acids, 6 esters, 2 alcohols, 2 amines, and 1 phenazine. The post-fermentation stage (Figure 7B) yielded 11 key metabolites: 3 esters, 3 amines, 2 alcohols, 2 cyclic peptides, and 1 acid. After steaming (Figure 7C), 5 key metabolites were identified, all of which were esters.

Figure 6. Scatter plots comparing volatile metabolites across SSP preparation stages. (A) Koji-making stage. (B) Post-fermentation stage. (C) Steaming stage. Numerical identifiers correspond to peak assignments in Figure 5. K0 and K6: samples collected at 0 and 6 days during koji-making. P0 and P18: samples collected at 0 and 18 days during post-fermentation. PS18: steamed product after 18 days of post-fermentation. Data represent mean ± SD, n = 3.

Figure 7. Bubble plots showing changes in volatile metabolites during SSP preparation. (A) Koji-making stage. (B) Post-fermentation stage. (C) Steaming stage. Only compounds with content > 10 μg/g, FC > 2, and p < 0.01 were labeled. K0 and K6: samples collected at 0 and 6 days during koji-making. P0 and P18: samples collected at 0 and 18 days during post-fermentation. PS18: steamed product after 18 days of post-fermentation. Data represent mean ± SD, n = 3.

Quality control of sequencing data

Gel electrophoresis of microbial genomic DNA showed clear bands (Figure S6), and DNA purity and concentration were suitable for sequencing. Illumina sequencing generated 10,489,297 V3-V4 reads and 8,506,274 ITS reads. The raw data were deposited in the NCBI database (BioProject No. PRJNA1184323). After quality filtering, more than 80% of the initial sequences were retained, showing consistent length distributions and no miscellaneous fragments. The dilution curves (Figure S7) approached a plateau as sequence numbers increased, indicating that the sequencing depth was adequate for community assessment.

The succession of microbial community during SSP fermentation

A comparison of absolute and relative abundances was presented in Figure 8. A total of 86 bacterial genera from 61 families were identified (Table S2). At the family level, Lactobacillaceae, Enterococcaceae, Bacillaceae, and Enterobacteriaceae accounted for 45.3%, 20.7%, 20.1%, and 13.3%, respectively. During the koji-making stage, Bacillus showed a relative abundance of 96.7%, and its absolute abundance increased from 1,331,972 copies (K0) to 3,946,827 copies (K6). Minor populations of Franconibacter, Enterococcus, Acinetobacter, and Streptophyta were also detected, each remaining below 4% of Bacillus. After removal of the yellow cladding, the abundance of all taxa was reduced. At the onset of post-fermentation, Enterobacter (18.6%), Enterobacteriaceae Unassigned (23.2%), Enterococcus (24.2%), and Pediococcus (25.7%) had relative abundances in a similar range. From P12 onwards, Pediococcus and Enterococcus further increased. Their relative abundances reached 64.4% and 25.7%, respectively. During the same period, Bacillus and other genera decreased to relative abundances below 5%.

Figure 8. Bacterial community composition at the genus level during SSP fermentation. (A) Absolute abundance. (B) Relative abundance. K0, K2, K4, and K6: samples collected at 0, 2, 4, and 6 days during the koji-making stage; P0, P3, P6, P9, P12, P15, and P18: samples collected at 0, 3, 6, 9, 12, 15, and 18 days during the post-fermentation stage. PS18: steamed product after 18 days of post-fermentation. Data represent mean ± SD, n = 5.

Fungal succession at the species level was shown in Figure 9. All fungal sequences were annotated to 58 genera from 48 families (Table S2). At the family level, Aspergillaceae accounted for 98.3%, followed by Rhizopodaceae (0.7%), Chaetomiaceae (0.5%), Nectriaceae (0.2%), and Cladosporiaceae (0.1%). During the first 2 days of the koji-making stage, Aspergillus flavus reached a relative abundance of 94.9%. Other fungi, including Aspergillus Unassigned, Rhizopus arrhizus, and Chaetomium pachypodioides, had relative abundances below 5%. After the removal of the yellow cladding, the absolute abundance of A. flavus was reduced and remained low in subsequent stages.

Figure 9. Fungal community composition at the species level during SSP fermentation. (A) Absolute abundance. (B) Relative abundance. K0, K2, K4, and K6: samples collected at 0, 2, 4, and 6 days during the koji-making stage; P0, P3, P6, P9, P12, P15, and P18: samples collected at 0, 3, 6, 9, 12, 15, and 18 days during the post-fermentation stage. PS18: steamed product after 18 days of post-fermentation. Data represent mean ± SD, n = 5.

Correlation of microbial communities and components

Correlation at the koji-making stage

Spearman correlation analysis was performed between all microbial taxa and physicochemical components during the koji-making stage. Twelve bacterial genera were identified with significant associations (Figure 10). Bacillus, Terribacillus, and Paenibacillus showed strong positive correlations (|r| > 0.8, p < 0.05) with WSE, AN, TAs, free isoflavones, and TPPs. They were also significantly correlated with a wide range of volatile compounds, predominantly esters and acids. Pediococcus was positively correlated with TAs. An unclassified Enterobacteriaceae genus showed a unique positive correlation with several esters, including methyl palmitate, methyl elaidate, 2-hydroxyethyl oleate, and isopropyl linoleate. Among fungi, six species exhibited associations with the physicochemical properties measured in this study. These included four Aspergillus species (A. flavus, A. Unassigned, A. navahoensis, and A. pseudodeflectus). A. flavus and A. Unassigned displayed correlation profiles similar to Bacillus. In contrast, Chaetomium pachypodioides and Fusarium equiseti displayed different correlation patterns. C. pachypodioides was mainly correlated with ethyl esters.

Figure 10. Spearman correlation between SSP components and microorganisms during the koji-making stage. Vertical axis: numerical identifiers corresponding to peak assignments in Figure 5. Correlation analysis performed across time points using component data (n = 5) and microorganism data (n = 3). Significant correlations indicated by |r| > 0.8, p < 0.05.

Correlation at the post-fermentation stage

Spearman correlation results between microbes and components during the post-fermentation stage were shown in Figure 11. The dominant genera Pediococcus and Enterococcus exhibited significant positive correlations (|r| > 0.8, p < 0.05) with AN, TAs, TPPs, free isoflavones, and WSE. They were also significantly correlated with aldehydes and amines among the volatile components, a pattern not observed during the koji-making stage. Among the nondominant bacteria, Staphylococcus showed positive correlations with methyl linolenate, heptadecanoic acid, and ASE, while Nitrospira was correlated with ketones and certain esters. Fungal contributions were less extensive, with four Aspergillus species showing significant associations with glycerol content.

Figure 11. Spearman correlation between SSP components and microorganisms during the post-fermentation stage. Vertical axis: numerical identifiers corresponding to peak assignments in Figure 5. Correlation analysis performed across time points using component data (n = 5) and microorganism data (n = 3). Significant correlations indicated by |r| > 0.8, p < 0.05.

Discussions

Indicators reflecting biochemical transformation and product quality evaluationas of SSP

As a functional food, SSP requires objective and reliable quality control indicators to ensure product safety, consistency, and efficacy. The current Ch. P. 2020 specifies two main criteria: (1) a minimum total content of 0.04% for daidzein and genistein, and (2) the absence of detectable protein residues via the biuret reaction. However, our findings highlight several limitations in these criteria for assessing fermentation adequacy and product uniformity.

Notably, by the end of the koji-making stage, the total isoflavone content already reaches 0.049% (Figure 3B), satisfying the pharmacopoeial standard even before the onset of post-fermentation. This observation indicates that the current standard fails to adequately reflect the biochemical maturity of SSP, which is consistent with previous findings (Wang et al., 2024). In our data, complete hydrolysis of isoflavone glycosides requires at least 12 days of post-fermentation. Because free isoflavones possess higher biological activity than their glycoside counterparts, sufficient fermentation is necessary to ensure full conversion of glycosides into free forms, thereby optimizing the utilization of bioactive resources. Therefore, we propose including at least one isoflavone glycoside (e.g., daidzin or genistin) as a negative marker to better evaluate fermentation completeness and product quality.

In addition, the biuret reaction becomes unreliable after 6 days of post-fermentation, as dark-colored extracts interfere with the colorimetric response (Figure A4). To address this limitation, we identify AN as an alternative indicator that effectively differentiates fermentation stages by reflecting the degree of protein hydrolysis. This concept draws inspiration from quality monitoring systems applied in Douchi production, where AN serves as a key marker for proteolytic progress. Interestingly, AN tends to stabilize toward the end of both the koji-making and post-fermentation stages. This plateau likely indicates that microbial protein utilization reaches a steady state, suggesting that nitrogen metabolism slows as the microbial community approaches equilibrium. Similar trends of AN during fermentation have also been reported in other soybean-based fermented foods such as Douchi, where AN dynamic reflects the progression of protein hydrolysis (Kumari et al., 2023; Li et al., 2023). Therefore, incorporating AN as a quantitative parameter in industrial production helps determine the optimal fermentation endpoint, reduce batch-to-batch variation, and enhance process consistency. Compared with the color-based biuret assay—which is subjective and easily affected by dark extracts—AN measurement provides an objective, reliable, and quantifiable metric for monitoring protein degradation during fermentation.

Furthermore, we observe that TPPs, WSE, ASE, and TAs increase progressively during fermentation. Although often considered secondary indicators, these parameters effectively reflect the underlying biochemical transformations that determine SSP quality. Maintaining an appropriate TAs level helps suppress undesirable microbial growth, which in turn stabilizes the fermentation environment and enhances both safety and reproducibility. Meanwhile, TPP accumulation is generally associated with enhanced antioxidant capacity, as seen in other fermented soybean products where phenolic compounds and free radical–scavenging activities increase over time (He et al., 2024; Kim et al., 2018). WSE and ASE provide complementary insights into fermentation dynamics. WSE, enriched in soluble proteins, peptides, and free amino acids, accumulates with proteolysis and contributes to both flavor complexity and bioactivity. ASE primarily reflects the release and transformation of lipid-soluble and moderately polar compounds, including phenolics and aroma precursors (Ahnan-Winarno et al., 2021). Integrating these indicators into process monitoring allows producers to assess fermentation efficiency and product functionality more accurately, laying the foundation for a multiparameter quality evaluation framework in SSP production.

Process parameters guiding fermentation control and production optimization

The quality of SSP is closely linked to the control of critical fermentation parameters. In this study, moisture content and TSs emerge as particularly influential. Although TSs do not exhibit a clear overall trend during fermentation, practical observations have revealed their critical role during washing yellow cladding. After the initial fermentation, the soybean seed coat becomes thin and fragile, which makes it challenging to remove surface-adhered mycelia and hydrophobic spores. In our practice, we find that leveraging the foaming properties of TSs allows for effective detachment of these residues using minimal water and gentle stirring with smooth wooden tools. This approach not only markedly improves washing efficiency but also reduces water usage, offering an environmentally friendly solution.

Moisture content constitutes another essential parameter. Unlike fermented soybean products intended for pastes, SSP requires intact and discrete beans. In our study, during the post-fermentation stage, beans are taken out and spread every 3 days to maintain a water content of around 55% (Figure 3G). The resulting products demonstrate that, although the seed coat becomes thinner and wrinkled, the beans remain intact (Figure 1). This stability largely stems from the balanced moisture level. Maintaining this equilibrium prevents excessive softening or agglomeration, preserves the granular integrity of the final product, and contributes to improved yield. The observed 55% moisture level may serve as a practical reference for industrial SSP production.

Volatile metabolites: both beneficial and potentially risky components

Volatile compounds are key determinants of SSP’s characteristic flavor. Our analysis reveals a progressive accumulation of total volatile compounds during fermentation, predominantly esters and acids, consistent with previous reports (Li et al., 2021). By applying thresholds of content >10 μg/g and FC >2, we identify compounds with both beneficial and potentially risky implications. Beneficial compounds include lipids such as linoleic acid and oleic acid, which modulate lipid metabolism (Furse et al., 2022), and ethyl linoleate, which is known for cholesterol- and lipid-lowering effects (Iorizzo et al., 2024).

However, we also observe significant accumulation of amine derivatives, including N-(2-phenylethyl) formamide, acemide, oleamide, (R-(Z))-12-hydroxy-9-octadecenamide, and N-(2-phenylethyl) acetamide. Notably, the concentrations of the latter two compounds exceed 100 mg/kg during fermentation, which may pose health risks, such as skin or respiratory irritation and organ toxicity (PubChem, 2025). In addition, the biogenic amine phenylethylamine is detected. Biogenic amines are recognized biological hazards, as excessive intake can cause gastrointestinal discomfort, blood pressure fluctuations, and respiratory allergic reactions (Park et al., 2019). Although international regulations stipulate histamine limits for fish products (e.g., US: 50 mg/kg; EU: 100 mg/kg) (Özogul & Özogul, 2019), no analogous standards exist for fermented soybean products. In this study, phenylethylamine reaches a maximum concentration of 9.67 mg/kg. As this represents only one component of the biogenic amine profile, the cumulative risk from total amines is likely underestimated. This concern is further reinforced by reports showing total biogenic amine levels exceeding 100 mg/kg in commercial Douchi products (Toro-Funes et al., 2015), indicating that amine accumulation may pose a potential risk in spontaneously fermented SSP.

Nevertheless, the characterization of volatile compounds in this study remains incomplete because of limitations of our GC-MS methodology. In comparison, Guo et al. (2024) employ SPME-GC-MS with peak area normalization and detect a broader spectrum of alcohols, including 2,3-butanediol and 1-octen-3-ol. Notably, 2,3-butanediol is also detected in our study, confirming partial overlap between the two analyses. To achieve a more comprehensive understanding of SSP’s volatile profile, future studies should incorporate advanced complementary analytical techniques.

Dominant microbes in SSP fermentation

Microbial adaptability to fermentation conditions plays a crucial role in shaping microbial dominance. This study shows that Bacillus spp. and Aspergillus flavus dominate during the aerobic stage, whereas facultative anaerobes such as Enterococcus spp. and Pediococcus spp. prevail in the oxygen-limited post-fermentation stage. This pattern of microbial succession clearly reflects a close alignment between microbial physiological traits and the specific environmental conditions at each fermentation stage. Our findings challenge previous reports suggesting that Bacillus and A. flavus maintain dominance throughout the entire fermentation process. For example, Zhu et al. (2015) report Bacillus as the predominant genus across all stages. However, their culture conditions appear to favor Bacillus growth, potentially masking the actual dominance of anaerobic bacteria during the post-fermentation phase. Similarly, Wang et al. (2024) report that A. flavus and Rhizopus spp. are dominant during the post-fermentation stage. However, given that both fungi are obligate aerobes, it is unlikely that they truly predominate under oxygen-limited conditions. This discrepancy likely results from methodological limitations. When the absolute abundance of these fungi is low, the absence of other fungal taxa may still yield artificially high relative abundance values, leading to a misleading inference that they are dominant species. Therefore, relative quantification alone cannot accurately capture the true microbial abundance across different fermentation stages. This interpretation is further supported by our comparative data (Figure 9), and similar methodological artifacts have been documented in other studies (Bruijning et al., 2023; Tkacz et al., 2018).

During the post-fermentation stage of SSP, Enterococcus and Pediococcus play dominant and functionally significant roles, consistent with observations in the anaerobic fermentation of Korean doenjang (Kim et al., 2018). Both genera are well recognized for their ability to produce diverse bioactive metabolites. For instance, specific strains of Enterococcus (e.g., E. avium and E. faecium) are associated with the synthesis of γ-aminobutyric acid (GABA) (Xiong et al., 2019). Likewise, Pediococcus pentosaceus is reported to enhance the accumulation of polyphenols, alcohols, and flavonoids in mulberry leaf (Zhuo et al., 2023). In addition, certain Pediococcus strains produce pediocins that inhibit spoilage organisms and pathogens, including Listeria monocytogenes (Porto et al., 2017). Our correlational analyses have further demonstrated significant associations between these genera and beneficial metabolites, such as isoflavones and flavor-active acids and esters. The co-occurrence of Pediococcus and Enterococcus, together with their similar correlation patterns toward key metabolites, suggests a potential functional synergy in shaping the characteristics of fermentation products. Although the underlying mechanism of this cooperation has not yet been elucidated, it represents a promising direction for future research.

Dominant microbes promote both beneficial and potentially risky components

In naturally fermented ecosystems, dominant microbes play indispensable roles in shaping the biochemical profile of the product, yet their metabolic versatility also confers a degree of risk. Consistent with previous findings, Bacillus, Pediococcus, and Enterococcus facilitate the accumulation of bioactive components such as free isoflavones, TPPs, AN, and TAs; however, their proliferation is also associated with the formation of potentially harmful metabolites, including 5-isobutyl-2-isopropyl-3-methoxypyrazine, N-(2-phenylethyl) acetamide, N-(2-phenylethyl) formamide, and N-(2-hydroxyethyl) acetamide (Azizoglu et al., 2023; Lee et al., 2021). In particular, Pediococcus and Enterococcus, though widely recognized as beneficial fermentation starters, include strains with strong amino acid decarboxylase activity that can convert amino acids into biogenic amines—especially under prolonged or anaerobic fermentation conditions (Barbieri et al., 2019; Montanari et al., 2022). This explains the observed increase in phenylethylamine levels during post-fermentation, which aligns with previous evidence that extended fermentation favors amine accumulation (Jia & Yu, 2022). Therefore, precise control of fermentation duration is critical to prevent excessive amine formation while maintaining the desired product quality. Interestingly, our results reveal that subsequent steaming treatment markedly reduced amine levels while enhancing ester formation, indicating that thermal post-processing serves as an effective strategy to mitigate potential toxic risks while improving flavor characteristics.

Furthermore, several studies have reported that specific strains of Lactobacillus and Bifidobacterium possess amine-degrading capabilities (Shao et al., 2022), suggesting that microbial modulation offers another feasible approach to enhance product safety. Taken together, these findings underscore the dual attributes of dominant microbial communities in spontaneous fermentation. Future work should focus on the targeted screening and application of low amine-producing or amine-degrading strains to gradually transition traditional spontaneous fermentation toward controlled, starter-guided fermentation systems, thereby enabling more predictable and safer production of SSP with improved functional quality.

Moreover, the substantial presence of Aspergillus flavus during the koji-making stage warrants particular attention because of its potential to produce aflatoxins (Jallow et al., 2021). In a previous market survey, Qiu et al. (2019) detect total aflatoxin exceeding 10 ng/g in 2 out of 13 samples, surpassing both the Ch. P. 2020 limit (5 ng/g) and the EFSA threshold for cereal products (4 ng/g) (Garcia-Alvarez-Coque et al., 2020). However, Li et al. (2020) report aflatoxins levels below 1 ng/g in 22 commercial SSP products. In addition, aflatoxin levels accumulate in SSP up to P6 but become undetectable by P12 (Ma et al., 2022), suggesting that aflatoxin content is controllable through proper fermentation. Taken together, although A. flavus may pose a risk, aflatoxin levels in SSP are generally controllable. This control is likely achieved through appropriate fermentation practices, including the physical removal of the yellow cladding and the subsequent degradation by post-fermentation microbes such as Bacillus and Pediococcus (Ren et al., 2020). To ensure that aflatoxin levels remain at safe, low concentrations, it is essential to further investigate the effectiveness of the washing process in removing aflatoxins and the role of dominant microbes in post-fermentation in degrading these toxins. By rigorously controlling the washing process, selecting appropriate post-fermentation strains, and optimizing post-fermentation conditions, the risk of toxin accumulation can be minimized.

Conclusion

This study elucidates the dynamic changes of microbial communities during SSP fermentation through absolute quantification sequencing. It reveals stage-specific patterns of dominant microbial taxa and their functional associations with key quality components, while also highlighting potential safety risks posed by certain microbes, emphasizing the importance of integrating quality and safety management in SSP production. The findings provide a scientific basis for the quantitative development of standardized and safe starter cultures and for improving SSP quality. By accounting for fluctuations in total microbial load, absolute quantification sequencing offers a more accurate assessment of microbial community dynamics than conventional methods. This approach is particularly valuable in complex microbial ecosystems, where it can be used to monitor the accumulation of key risk genera, identify core functional microbiota, and optimize microbial consortia. Nevertheless, its taxonomic resolution is currently limited for precise species-level identification and quantification. To address this limitation, long-read genome sequencing can be integrated, or genus-specific identification and quantification methods can be developed for dominant taxa, thereby achieving higher taxonomic resolution, accurate species-level abundance determination, more precise risk assessment, improved process control, and enhanced product quality in fermentation systems.

Data Availability Statement

The sequencing datasets generated for this study have been deposited in NCBI database (BioProject No. PRJNA1184323). All other data supporting the findings of this study are included in the article and its Supplementary Materials. Further inquiries can be directed to the corresponding author.