Identification and characterization of umami peptides from fermented soybean meal hydrolysates: Combined machine learning, cellular functional characterization, and structural interpretation

Abstract

This study developed an efficient strategy for identifying umami and umami-enhancing peptides from fermented soybean meal hydrolysates (FSMH) to support the high-value utilization of soybean byproducts. An integrated approach combining virtual screening, molecular docking, sensory validation, intracellular calcium mobilization assays, and molecular dynamics simulation was employed. Sixteen peptides were synthesized for sensory validation, confirming 15 umami peptides. LE, AE, GEDLMVQ, and FEEINKV exhibited the strongest enhancement in both monosodium glutamate and inosinate/disodium guanylate systems, with GEDLMVQ showing the lowest enhancement threshold (0.0079 mM). Fermentation significantly increased most umami peptide concentrations, peaking at 24 h. Molecular docking implicated T1R1-VFTD as the main binding domain, mediated primarily by hydrogen bonding, hydrophobic contacts, and van der Waals forces. Representative peptides induced TAS1R1-associated Ca²⁺ responses with distinct kinetics. Molecular dynamics simulations provided supportive structural information for interpreting peptide–receptor interactions. These findings support a feasible strategy for obtaining sustainable flavor enhancers from soybean byproducts.

Highlights

• •Ten novel umami peptides and 14 umami-enhancing peptides were identified from FSMH.
• •GEDLMVQ showed the lowest umami and umami-enhancing thresholds in FSMH.
• •Co-fermentation increased most umami peptide levels, peaking at 24 h.
• •Three peptides induced distinct TAS1R1-associated Ca²⁺ responses.

1. Introduction

A sustainable global food system requires efficient utilization of agro-industrial byproducts to mitigate environmental impacts and address resource scarcity. Soybean meal, the primary byproduct of soybean oil extraction, is produced in large quantities worldwide and is predominantly used as low-value animal feed. Despite limitations such as protein denaturation, undesirable flavor, and antinutritional factors, its high protein content and balanced amino acid profile, particularly the abundance of acidic amino acids, soybean meal serves as an excellent precursor for umami peptides (Cao et al., 2024, Zhang et al., 2022). Valorizing soybean meal into high-value food ingredients represents a promising strategy to support circular bioeconomic objectives.

Umami, recognized alongside sweet, sour, salty, and bitter as the fifth basic taste, is described as a pleasant savory sensation or monosodium glutamate (MSG) like taste. A distinctive characteristic of umami, which plays an important role in flavor modulation, is its synergistic interaction with other taste modalities, including saltiness enhancement and bitterness suppression (Kim, Son, Kim, Misaka and Rhyu, 2015, Wang et al., 2023). Umami peptides, owing to their dual functions in directly imparting umami taste and enhancing flavor perception, have attracted growing research interest. These peptides have been identified from diverse plant and animal-derived foods, including meats, seafood, mushrooms, legumes, and fermented soybean products (Wang et al., 2023, Xu et al., 2024; Zhang et al., 2024), and hold considerable potential for improving sensory quality while reducing sodium intake (Gu et al., 2024). Although umami peptides have been isolated from soy sauce, fermented soybean paste, and other soybean-derived products (Zhang et al., 2024; Zhao et al., 2023), soybean meal has rarely been investigated as a direct source of umami peptides despite its favorable amino acid profile. Furthermore, the molecular mechanisms underlying their taste perception remain poorly characterized.

Enzymatic hydrolysis is widely employed for the industrial production of umami peptides due to its controllable reaction conditions, mild processing environment, and high safety. Commonly used proteases for legume processing include Alcalase, flavourzyme, and neutrase (Cao et al., 2024). However, exclusive reliance on enzymatic treatment frequently results in limited peptide diversity, low yields, and the generation of undesirable sensory attributes (Zhao et al., 2023). In contrast, microbial fermentation not only facilitates the release of proteases but also promotes the degradation of anti-nutritional factors and the formation of diverse flavor-active metabolites (An et al., 2023). Combining fermentation with enzymatic hydrolysis has been demonstrated to enhance proteolysis efficiency, increase peptide diversity, and enrich desirable flavor characteristics (Chen et al., 2024; Fan et al., 2025). Our previous work demonstrated that co-culture fermentation of soybean meal hydrolysates with Limosilactobacillus fermentum and Pichia fermentans significantly improved their flavor quality, as reflected by enhanced umami and saltiness, reduced bitterness and beany odor, and the most balanced overall flavor profile at 24–30 h. These flavor changes were associated with the accumulation of small peptides, free amino acids, nucleotides, organic acids, and volatile flavor compounds (Cao et al., 2025). Nevertheless, the specific umami peptides contributing to the improved taste characteristics and their molecular mechanisms remained unclear. Conventional identification of umami peptides, typically involving sensory-guided fractionation and chromatographic purification, is labor-intensive, time-consuming, and costly, thereby limiting the rapid discovery of novel candidates (Gu et al., 2024). In recent years, advanced computational strategies have emerged as efficient alternatives for umami peptide screening. Most previous studies have relied on single or at most two machine-learning models. The integrated use of multiple machine learning models, such as iUmami-SCM, UMPred-FRL, Umami-YYDS, and TastePeptides-DM, has been shown to improve prediction accuracy and discovery efficiency, although reports applying such combined approaches remain limited (Feng et al., 2024; Xiong et al., 2022; Xu et al., 2024). Moreover, molecular docking and molecular dynamics (MD) simulations provide structural information on peptide–receptor interactions, thereby improving the efficiency of umami peptide identification. However, these computational approaches mainly generate predictive or structural hypotheses and remain insufficient for establishing receptor-associated functional activity. This limitation is particularly relevant for taste receptors. Accordingly, integrating cellular functional assays combined with computational analyses can strengthen the mechanistic interpretation of umami peptide activity. This integrated strategy accelerates peptide discovery and provides molecular insights into peptide-receptor interactions. It also supports the sustainable valorization of soybean meal in alignment with circular bioeconomic principles.

Therefore, this study aimed to establish a rapid and efficient strategy for screening umami and umami-enhancing peptides from fermented soybean meal hydrolysates. This strategy integrated computational prediction with experimental validation to enhance the value of soybean meal. Potential umami peptides were screened using multiple machine-learning models and molecular docking. Selected peptides were subsequently subjected to chemical synthesis, quantitative analysis, sensory evaluation, and electronic tongue measurements to verify their taste properties and contributions to umami perception. To further assess receptor-associated functional responses at the cellular level, representative peptides were evaluated using intracellular Ca²⁺ mobilization assays in TAS1R1-transfected HEK293 cells. In addition, molecular dynamics simulations were performed to provide structural interpretations of the distinct functional behaviors of key peptides in the receptor model. This approach not only provides mechanistic insights into umami perception but also demonstrates the feasibility of converting soybean meal into high-value flavor-active ingredients for fermented food systems, thereby contributing to sustainable food waste valorization and the development of natural, health-oriented taste modulators.

2. Materials and methods

2.1. Materials and reagents

Soybean meal powder was provided by Guangdong Tianqi Biotechnology Co., Ltd. (Foshan, China). Commercial proteases (Flavourzyme 500 MG and Alcalase 2.4 L) were obtained from Novozymes Biotechnology Co., Ltd. (Tianjin, China). Synthetic peptides used as analytical standards were purchased from Nanjing Jietai Biological Technology Co., Ltd. (Nanjing, China). Food-grade monosodium glutamate (MSG), sodium chloride (NaCl), citric acid, sucrose, caffeine and tea polyphenols, were supplied by Guangdong Tianqi Biotechnology Co., Ltd. (Foshan, China). Chromatographic-grade methanol and acetonitrile were acquired from Fisher Scientific (Shanghai, China). The pcDNA3.1+/C-(K)DYK plasmid containing the TAS1R1 gene (clone ID# NM_138697.4) was purchased from GenScript (Piscataway, NJ, USA). Lipofectamine 3000 transfection reagent and P3000 enhancer reagent were obtained from ThermoFisher (Ottawa, ON, Canada). HEK293 (CRL-3216) cells were purchased from the American Type Culture Collection (Manassas, VA, USA). Dulbecco's modified Eagle's medium (DMEM) and Fluo-4 AM were obtained from Gibco Invitrogen (Burlington, ON, Canada). All other reagents used in this study were of analytical grade and purchased from Guangzhou Chemical Reagent Factory (Guangzhou, China).

2.2. Microorganisms and growth conditions

Limosilactobacillus fermentum SCUT221015 was provided by the Food Biotechnology Laboratory at South China University of Technology (Guangzhou, China) and deposited in the NCBI database under accession number PRJNA1138699. The strain was activated and cultivated in MRS broth (Hopebio, Qingdao, China) at 37 °C for 24 h under static conditions. Pichia fermentans (CICC 33372), obtained from the China Center of Industrial Culture Collection (Beijing, China), was grown in yeast extract peptone dextrose (YPD) medium (Huankai Microbial, Guangzhou, China) at 30 °C for 48 h with agitation at 180 rpm.

2.3. Preparation of fermented soybean meal hydrolysate samples

Fermented soybean meal hydrolysates (FSMH) were prepared through a two-step process involving enzymatic hydrolysis followed by fermentation, as described in previous study (Cao et al., 2023). Briefly, soybean meal powder was mixed with deionized water (1:7, w/v) and subjected to sterilization at 121 °C for 15 min. Upon reaching room temperature, the Flavourzyme (0.5%, w/v) and Alcalase (1.0%, w/v) were added, and the mixture was incubated at 55 °C for 18 h. Enzyme inactivation was conducted at 95 °C for 15 min. The resulting hydrolysates were centrifuged at 8000 ×g for 25 min at 4 °C, and the supernatant was collected for subsequent fermentation. The fermentation procedure was performed according to our previously reported method. The pH of the hydrolysates was adjusted to 6.5 prior to inoculation, and no external pH control was applied during fermentation. Then, a co-culture of L. fermentum SCUT221015 (1 × 10⁷ CFU/mL) and P. fermentans (5 × 10⁵ CFU/mL) was inoculated, and fermentation was carried out at 30 °C for 18 h (LP18), 24 h (LP24), 30 h (LP30), 36 h (LP36), and 48 h (LP48). A non-inoculated group (CK) was included as a control. After fermentation, all samples were centrifuged at 8000 ×g for 30 min at 4 °C, and the supernatants were filtered through a 0.22 μm membrane filter and stored at −20 °C until further analysis.

2.4. Identification of peptides by UPLC-QTOF-MS/MS

Peptide identification in FSMH samples was performed using ultrahigh-performance liquid chromatography coupled with quadrupole-time-of-flight tandem mass spectrometry (UPLC-QTOF-MS/MS) after ultrafiltration through the 3 kDa ultrafiltration membranes. Peptide analysis was performed on an Acquity UPLC system (Waters, Bray, Ireland) equipped with an HSS T3 column (2.1 × 100 mm, 1.8 μm, Waters). Mobile phase A was ultrapure water containing 0.1% (v/v) formic acid, and mobile phase B was acetonitrile. The elution program was set as previously described.1 Mass spectrometry was conducted using an electrospray ionization quadrupole-time-of-flight mass spectrometer (ESI-QTOF-MS/MS) (Bruker Impact II, Bruker Daltonics Inc., Billerica, MA, USA) operating in positive ionization mode. Briefly, the mass scanning range was m/z 50–2000 with auto MS/MS acquisition mode, in which precursor ions exceeding 1000 counts were automatically selected for fragmentation. Quadrupole ion energy was 3.0 eV and collision-induced dissociation energy was set to 10.0 eV. Electrospray conditions included a capillary voltage of 3.5 kV, drying gas temperature at 180 °C, drying gas flow of 6.0 L/min, and atomizer pressure of 1.5 bar. Hyphenation Star Application software (v3.2, Bruker Daltonics Inc., Billerica, MA, USA) was used to control the instrumentation. Mascot Distiller (v2.4.2.0, Matrix Science Inc.) was applied for automatic spectrum deconvolution, peak list extraction, and subsequent database searching. Mascot database searches were performed with no enzyme specified, carbamidomethylation of cysteine as a fixed modification, and oxidation of methionine as a variable modification. The peptide mass tolerance was set to ±10 ppm, and peptide identifications were accepted based on a Mascot score ≥ 20. Manual de novo peptide sequencing of low molecular peptides was performed using Data Analysis software (v3.0, Bruker Daltonics Inc., Billerica, MA, USA). Peptide identification was further verified against the UniProt protein database (https://www.uniprot.org/) using soybean-derived protein sequences. Only peptides corresponding to theoretical fragments of known soybean proteins were retained for further analysis.

2.5. Fourier transform infrared spectroscopy (FTIR) analysis

FTIR spectroscopy was performed using a Nicolet IS50 FTIR spectrometer (Thermo Fisher Scientific, Waltham, MA, USA) to analyze the structural characteristics of FSMH samples (Hu et al., 2025). Freeze-dried powders were mixed with potassium bromide (KBr) and pressed into thin sheets. Spectra were collected in the range of 4000–400 cm⁻¹ with a resolution of 4 cm⁻¹ and 32 scans per sample. Background spectra were recorded prior to each measurement. All spectra were baseline-corrected and deconvoluted using PeakFit software (v4.12, USA).

2.6. Screening of the potential umami peptides

The umami potential of the identified peptides was initially evaluated using four machine learning predictors, including iUmami-SCM (https://camt.pythonanywhere. com/iUmami-SCM), TastePeptides-DM (http://tastepeptides-meta.com/TPDM), UMPred-FRL (https://pmlabstack.pythonanywhere.com/ UMPred-FRL), and Umami-YYDS (http://tastepeptides-meta.com/Umami_YYDS). Peptides predicted as umami by all four models were retained as potential umami peptides. Subsequent screening of the selected peptides was conducted using multiple bioinformatics tools to assess their safety and physicochemical properties. ToxinPred (https://webs.iiitd.edu.in/raghava/toxinpred/design.php) was used to evaluate peptide toxicity, while solubility and stability were predicted using the Innovagen peptide tool (https://www.innovagen.com/proteomics-tools) and the ExPASy ProtParam tool (https://web.expasy.org/ protparam/), respectively.

2.7. Molecular docking

Homology modeling of the T1R1/T1R3 (T1R1, UniProtKB: Q7RTX1; T1R3, UniProtKB: Q7RTX0) heterodimer was performed according to previously established methods (Su et al., 2023), using the activated-state extracellular domain of the calcium-sensing receptor as the structural template (CaSR; 5K5S). The quality of the modeled structures was validated by Ramachandran plot and Verify 3D analysis, confirming that >99% of residues were in allowed regions and meeting stereochemical and 3D profile evaluation standards. The 2D structures of the selected peptides were drawn using Chem Draw 21.0, and using Chem 3D (PerkinElmer Informatics, Inc., MA, USA) to construct their three-dimensional structures, and the structure of each peptide was optimized by calculating their minimum energy. Molecular docking was conducted with AutoDock Vina (The Scripps Research Institute, CA, USA), focusing on the interactions between selected peptides and the T1R1/T1R3 receptor. Docking results were visualized in 3D using PyMOL (Schrödinger, LLC, NY, USA), while 2D interaction diagrams were shown using Discovery Studio 2021 (Accelrys, Inc., CA, USA).

2.8. Sensory analysis of synthetic peptides

2.8.1. Basic taste analysis

Sixteen potential umami peptides were synthesized by Nanjing Jietai Biotechnology Co., Ltd. (Nanjing, China) utilizing the Fmoc solid-phase process, with purity exceeding 95%, as determined by analytical HPLC (Table 1). The observed retention times of peptides in FSMH samples were consistent with those of the corresponding synthesized standards, providing support for the accuracy of sequence assignments. Sensory profiling was evaluated by twelve experienced panelists (six females and six males, aged 20–28 years) from South China University of Technology (Guangzhou, China). The assessments were conducted under standard laboratory conditions (23 ± 2 °C, normal lighting) following ISO 8589:2007 guidelines. All panelists were trained according to ISO 8586: 2023 and had at least one year of experience in food sensory analysis. Prior to the formal evaluation, a specific pre-training using the peptide solutions was conducted for the panelists. Triangle tests were performed in accordance with ISO 4120:2021. This study was conducted in accordance with the ethical principles of the Declaration of Helsinki. While formal ethics committee approval was not required, all participants received full information about the study and provided written informed consent before participation.

Table 1.

The basic information, taste descriptions and threshold of screened umami peptides.

Number	Sequence	Purity (%)	Mass/Da	Threshold value (mM)	Threshold value (mM)		Taste description
					In the presence of MSG	In the presence of I + G
1	LE	96.48	260.29	1.50*	0.10	0.04*	umami, salty, kokumi
2	GE	95.60	204.18	0.80*	1.47	0.73	umami, weak sour
3	VE	95.62	246.26	1.50*	0.81	0.09*	umami, weak sour, weak astringency
4	ER	96.99	303.31	1.50	0.66	0.49	umami, weak salty
5	AE	99.80	218.21	1.50*	0.29	0.15*	umami, weak sour, weak astringency
6	SE	95.75	234.21	1.49*	0.87*	0.05*	umami, weak salty, weak astringency
7	SQ	97.08	233.22	0.75	1.50	0.86	Sour, weak umami
8	VGE	97.35	303.31	2.77	1.32	0.66	umami, kokumi, weak salty, astringency
9	VEK	95.34	374.43	1.50	0.13	0.2	salty, weak umami, weak sour, weak astringency
10	SPT	95.80	303.31	–	0.99	1.15	weak umami, sour, astringency
11	AVE	95.82	317.34	2.71	1.26	0.47	weak sour, weak astringency, weak umami
12	VVVD	99.54	430.50	1.50	0.35	0.35	salty, kokumi, umami
13	TLTV	96.73	432.51	–	0.17	–	weak sour, astringency, weak umami
14	GEDLMVQ	97.57	790.88	0.75	0.0079	0.0079	umami, weak astringency
15	FEEINKV	99.43	877.98	1.50	0.23	0.05	umami, weak astringency, weak sweet

Note: Values marked with an asterisk indicate threshold values previously reported. ^14,21,33 The remaining peptides are newly identified umami or umami-enhancing peptides from FSMH in this study.

Quantitative descriptive analysis, slightly modified from the previous methods (Cao, Song, Mu and Su, 2025, Wang et al., 2023), was conducted using a 10-point scale, where 1 represented extremely weak and 10 indicated extremely strong. All panelists completed preliminary tests including taste discrimination, ranking, and triangle testing. Seven taste attributes (umami, saltiness, sweetness, sourness, bitterness, astringent and kokumi) were evaluated using reference standards: 0.05% MSG, 0.1% NaCl, 0.5% sucrose, 0.02% citric acid, 0.08% caffeine, 0.4% tea polyphenols and 0.15% GSH, each set as 5 points on the scale. Kokumi was defined as a sensation of mouth fullness, thickness, and continuity that enhances basic tastes (Wang, Han, et al., 2023). The synthetic peptides were dissolved in ultrapure water at a concentration of 10 mg/mL, and the pH of each solution was adjusted to 6.5. All assessments were conducted under double-blind conditions. Peptide samples were randomly coded with three-digit numbers and presented randomly to panelists, and the evaluations were conducted in triplicate. Assessors rinsed with room-temperature ultrapure water and rested for 60 s between different samples.

2.8.2. Umami and saltiness enhancing analysis

To assess the umami and saltiness enhancement properties of synthetic peptides, a series of taste reconstitution assays were performed following previous methods with modifications (Cao et al., 2024; Zhao et al., 2023). Peptide solutions (1 mg/mL) were individually prepared by dissolving synthetic peptides in three different taste bases: (1) 0.03% MSG, (2) 0.006% I + G (a 1:1 mixture of disodium 5′-inosinate and disodium 5′-guanylate) and (3) 0.3% NaCl solution. For reference calibration, the umami intensity of 0.05% MSG and 0.03% I + G solutions, and the saltiness of 0.5% NaCl solution, were defined as 5 points on the 10-point scale. All solutions were adjusted to pH 6.5.

2.8.3. Taste threshold determination

The taste thresholds of synthetic peptides were determined using taste dilution analysis (TDA), based on previous studies with minor modifications (Feng et al., 2024). Briefly, peptides were serially diluted with ultrapure water (1:1, v/v) to obtain a concentration gradient. Trained panelists evaluated each dilution series using a triangle test, in which one sample and two blanks (ultrapure water) were presented in random order, until the sample could no longer be distinguished from the blanks. Additionally, the comparative taste dilution analysis (cTDA) was used to evaluate the umami-enhancing thresholds of peptides (An et al., 2023). Peptides were separately dissolved in MSG (0.03%, w/v) and I + G solution (0.01%, w/v). The triangle test was repeated under identical conditions, with MSG or I + G solution replacing the blank groups.

2.8.4. Dose-response test

The umami-enhancing effects of the synthetic peptides were further assessed using a dose–response test (Cui, Li, et al., 2023). Two systems were established, in which each peptide was dissolved in 0.03% MSG or 0.006% I + G solutions to prepare a series of concentrations (0.25, 0.5, 1.0, 1.5, 2.0, and 2.5 mg/mL). Sensory evaluations were conducted using a 10-point intensity scale, with 0.05% MSG and 0.03% I + G solutions defined as 5 points in the MSG and I + G systems, respectively.

2.9. Electronic tongue analysis

The taste characteristics of the synthetic peptides were evaluated using an electronic tongue system (TS-5000Z, Intelligent Sensor Technology Inc., Kanagawa Pref., Japan), following the previous procedure (Cao et al., 2025). Sensors were conditioned, calibrated, and drift-corrected using the reference and cleaning solutions supplied by the manufacturer. Peptides were dissolved at the concentration of 2 mg/mL either in ultrapure water for basic taste assessment or in 0.1% (w/v) MSG solution for umami-enhancing evaluation. Each sample was analyzed using five taste sensors: AAE (umami), CT0 (salty), GL1 (sweet), CA0 (sour), and COO (bitter). In addition to the five primary sensor outputs, the instrument further provided supplementary taste parameters, including richness and astringency. All measurements were conducted in quadruplicate, with a detection at 23 ± 2 °C (120 s).

2.10. Quantitation of umami peptides in FSMH samples

The concentrations of selected umami peptides in FSMH samples were determined using an external standard method combined with UPLC-ESI-QTOF-MS/MS analysis. FSMH samples were diluted to appropriate concentrations and desalted by centrifugation through filtration tubes to remove salts and insoluble materials. Standard calibration curves were established using six concentration levels (0.10, 0.50, 1.00, 2.50, 5.00, and 10.00 μg/mL) of synthetic peptides, prepared in 0.1% (v/v) formic acid aqueous solution. Chromatographic separation and mass spectrometric detection were performed under the same conditions as described in section 2.4. Calibration curves for all 14 target peptides showed good linearity within this range, with coefficients of determination (R²) summarized in Table S8, and all concentrations measured in FSMH samples fell within the corresponding calibration ranges.

2.11. Calcium mobilization assay in TAS1R1-transfected HEK293 cells

HEK293 cells were cultured in high-glucose DMEM and passaged with 0.25% (w/v) trypsin-EDTA at 80–90% confluency. For transfection, cells were seeded in 48-well plates and transfected with the pcDNA3.1+/C-(K)DYK plasmid containing human TAS1R1 gene using Lipofectamine 3000 following the manufacturer's protocol. After confirming successful transfection, HEK293 cells were reseeded into black, clear-bottom 96-well plates for cellular calcium mobilization measurements. Non-transfected cells served as negative controls.

Intracellular calcium mobilization was measured using Fluo-4 AM as described previously. ¹⁹ Briefly, Fluo-4 AM was dissolved in DMSO and mixed 1:1 (v/v) with 20% Pluronic F-127, then diluted in HEPES-buffered synthetic physiological salt solution (HEPES–SPSS) to 200 μM (final concentration). After 20-fold dilution, the Fluo-4 AM loading mixture was applied to a 96-well plate at a final concentration of 10 μM, incubated at 37 °C for 20 min and subsequently washed with HEPES-SPSS. MSG (10 mM) was used as a positive control, and the selected umami peptides were applied individually at 5 mM to TAS1R1-transfected HEK293 cells to assess receptor mediated calcium responses, following the previous method (Wang et al., 2026). Fluorescence images were captured using an Olympus IX81 fluorescent microscope (Olympus, Tokyo, Japan) at 20 s intervals during the first minute and at 30 s intervals over the next four minutes.

2.12. Molecular dynamics simulation

Molecular dynamics (MD) simulations were conducted using GROMACS 2022 software. The peptide-receptor complexes obtained from molecular docking were parameterized using the AMBER14SB force field and solvated with the TIP3P water model in a periodic cubic box under constant temperature and pressure (NPT) conditions. The systems were solvated in a periodic cubic box with a minimum distance of 10 Å between the complex and the box edge. NaCl was introduced to achieve an ionic strength of 0.15 M. Energy minimization was performed using the steepest descent algorithm prior to equilibration. Hydrogen-involved bonds were constrained using the LINCS algorithm with a 2 fs time step. Temperature was maintained at 298 K using the V-rescale thermostat, and pressure was controlled at 1 bar. Prior to production simulations, the system was equilibrated using 100 ps of NVT followed by 100 ps of NPT simulations. For each peptide–receptor complex, three independent 100 ns production MD simulations were performed, and conformations were saved every 10 ps for subsequent analysis.

2.13. Statistical analysis

Statistical analysis was performed using SAS 9.3 (SAS Institute Inc., Cary, NC, USA). One-way analysis of variance followed by Duncan's multiple range test was used to determine significant differences among groups (p < 0.05). Hierarchical cluster and principal component analysis (PCA) were conducted using OriginPro 2021b (OriginLab, Northampton, USA). Wordcloud diagrams were generated using the online platform (https://www.bioinformatics. com.cn).

3. Results and discussion

3.1. Analysis of peptides identification

The diversity and abundance of peptides generated during fermentation significantly influence flavor characteristics, particularly the umami intensity, which depends on peptide chain length, amino acid composition, sequence arrangement, and spatial configuration (Zhao et al., 2023). To investigate the dynamic changes in peptide composition during the enzymatic hydrolysis followed by co-culture fermentation of soybean meal hydrolysates, peptides were identified and characterized using UPLC-QTOF-MS/MS. As illustrated in Fig. 1A, a total of 312 peptides were identified across all samples, including 115 peptides in CK (enzymatic hydrolysis only) and 197 peptides across the fermented groups. Dipeptides and tripeptides were predominant, along with a considerable number of heptapeptides, indicating extensive proteolysis induced by enzymatic hydrolysis. GE, LP, LE, and VE, demonstrated the highest peak areas, suggesting their relative dominance. Notably, from 18 h of fermentation (LP18), fermented soybean meal hydrolysates exhibited a higher peptide abundance and diversity compared to the CK, suggesting that co-culture with L. fermentum and P. fermentans promoted further hydrolysis of larger peptides into smaller peptides. These findings are consistent with previous reports indicating that microbial fermentation substantially increases the proportion of low-molecular-weight peptides in soybean protein hydrolysates (Cao et al., 2023; Cao et al., 2025). Moreover, low-molecular-weight peptides formed during enzymatic hydrolysis are readily utilized by microorganisms, thereby enhancing the proteolytic activity of microbial enzymes under suitable fermentation conditions (Chen et al., 2024).

Fig. 1.

Peptides profiling of different FSMH samples based on UPLC-QTOF-MS/MS data. (A) The peptide composition. (B) Frequency distribution of amino acids in the top 100 peptides. The single letter codes on the y-axis represent amino acid residues. (C) Peptide Venn diagrams and fingerprint comparison among CK, LP24, and LP48.

To further elucidate the effects of fermentation on amino acid composition within soybean meal hydrolysates, amino acid frequencies of the top 100 peptides ranked by peak area in each group were analyzed (Fig. 1B). Compared to CK, FSMH samples exhibited distinct amino acid profiles, with LP48 clearly distinguished from other fermentation treatments. Specifically, amino acids associated with umami (Glu, Asp, Gln) and sweet (Ala, Gly, Thr, Pro) increased notably in frequency across all fermented samples. This enrichment indicated the potential of controlled fermentation to enhance the umami characteristics, thereby providing feasibility for subsequent screening of umami peptides. Previous studies have highlighted peptides containing Glu, Asp and Gln as significant contributors to umami taste or its enhancement (Chen et al., 2024; J. Zhang, Sun-Waterhouse, et al., 2019). Furthermore, many umami peptides also contain sweet amino acids such as Gly, Ser, Thr, and Ala, which may synergistically enhance umami perception (Zhao et al., 2023). In contrast, certain hydrophobic amino acids (e.g., Arg, Val, His, Phe, Ile) showed reduced frequencies in fermented groups, suggesting microbial metabolic preferences that could mitigate bitterness and enhance overall taste acceptability (Su et al., 2023).

The overall amino acid composition of peptides largely determines the taste profile of the fermented products due to synergistic effects between molecules (Su et al., 2023). Fig. 1C showed a comparative peptide fingerprint analysis of the top 100 peptides among CK, LP24, and LP48 to identify specific differential peptides contributing to fermentation induced flavor alterations. Numerous short-chain peptides enriched in umami amino acids (highlighted in yellow) accumulated post-fermentation and may synergize with other flavor-active compounds to enhance overall flavor. Notably, peptides identified in LP24 and LP48 frequently contained umami and sweet amino acid residues in exposed orientations, which can more effectively stimulate corresponding umami taste receptors (Wang et al., 2020). Additionally, the peptide fingerprints indicated that CK contained more peptides with hydrophobic amino acids at terminal positions, whereas peptides identified uniquely in LP24 and LP48 exhibited a higher proportion of hydrophilic amino acids with hydrophobic residues primarily located centrally within the peptide chains, demonstrating greater potential for enhancing umami taste (Wang et al., 2024). Overall, these results indicated that fermentation significantly modulated the peptide profile of soybean meal hydrolysates, promoting the formation of potential umami peptides.

3.2. FTIR analysis

The flavor characteristics of umami peptides are closely associated with their spatial structure. FTIR was employed to investigate structural changes in functional groups and secondary structures of CK, LP24, and LP48 samples. As shown in Fig. S1, all samples displayed broad absorption bands at 3000–3500 cm⁻¹, which corresponds to O—H and N—H stretching vibrations (Ai et al., 2024). The increased intensity of these bands in LP24 and LP48 suggested changes in the hydrogen-bonding environment, consistent with strengthened intermolecular interactions. The amide I region (1700–1600 cm⁻¹), primarily arising from C O stretching and C—N stretching vibrations, is recognized as the important indicator of protein secondary structure (Yan et al., 2024). A red shift of the amide I peak from 1618 cm⁻¹ in CK to 1604 cm⁻¹ in LP48 indicated structural reorganization at the peptide backbone level, potentially associated with β-sheet formation. Additionally, the C—O stretching band shifted from 1095 cm⁻¹ (CK) to 1067 cm⁻¹ in fermented samples, likely due to microbial metabolic alterations.

Deconvolution of the amide I band was performed to quantitate secondary structure components (Table S1). The predominant β-type structures (β-sheets and β-turns) are consistent with previous findings in peanut protein hydrolysates (Zhang, Sun-Waterhouse, et al., 2019). Fermentation significantly increased β-sheet content (from 44.51% to 54.13%), while α-helix and β-turn contents decreased (from 11.93% to 9.67% and from 29.42% to 22.08%, respectively), suggesting that enzymatic hydrolysis and microbial fermentation actively promote peptide release and associated secondary structure transitions. Notably, previous studies have associated β-sheet formation with the taste characteristics of umami peptides (Ai et al., 2024). Additionally, LP24 exhibited an increase in random coil content, which indicated enhanced conformational flexibility, potentially facilitating structural rearrangement via hydrophobic, electrostatic, and hydrogen-bonding interactions. These structural modifications may contribute to the enhanced generation of umami peptides during fermentation.

3.3. Screening of potential umami peptides

To screen potential umami peptides from the 312 peptides identified, four well-established machine learning-based prediction models (iUmami-SCM, UMPred-FRL, Umami-YYDS, and Taste Peptides-DM) were employed (Charoenkwan et al., 2020; Charoenkwan et al., 2021; Cui, Zhang, et al., 2023). These predictors were selected as publicly available tools that have been specifically developed and experimentally validated for the recognition of umami or other taste-related peptides and that rely on distinct feature representations and learning algorithms. Consequently, their combined application provides complementary information and enhances the reliability of umami-propensity predictions. As illustrated in Fig. S2 and Table S2, the predicted scores varied among the different models, with the majority concentrated in favorable umami ranges. Preliminary screening based on each model's threshold yielded 128, 125, 153, and 162 potential umami peptides, respectively (Fig. 2A). In a second step, to enhance prediction accuracy, stricter screening criteria (iUmami-SCM > 600, UMPred-FRL > 0.6, Umami-YYDS > 0.9, Taste Peptides-DM > 0.8) were applied, refining the final set to 51 high-confidence potential umami peptides (Fig. S2E) (Xu et al., 2024). To minimize false-positive predictions, only peptides identified simultaneously by all four models were initially recognized as potential umami peptides (Gu et al., 2024). The statistical distribution of 51 potential umami peptides across different FSMH samples (Fig. 2B) indicated a significant increase in peptide abundance during fermentation, particularly after 24 h. Toxicity assessment helps determine whether peptides entail any potential harm to human health, thereby ensuring their safety and guiding the necessity for further investigations into their bioactivity or taste properties (Wang, Luan, et al., 2023). All 51 peptides were predicted as non-toxic by the ToxinPred database (Table S3). Additionally, since stability and solubility critically influence perceived umami intensity, predictions revealed high stability for 46 peptides and good water solubility for 38 peptides, highlighting their potential as umami-enhancing agents in food systems.

Fig. 2.

Machine learning-based screening and molecular docking analysis of potential umami peptides from FSMH samples. (A) Number of potential umami peptides identified by each model. (B) Distribution of potential umami peptides across different groups. (C—D) Key interacting residues of potential umami peptides with the T1R1 (C) and T1R3 (D) predicted by molecular docking.

3.4. Molecular docking of potential umami peptides and T1R1/T1R3

To better predict the umami potential of candidate peptides and elucidate their structure–activity relationships, molecular docking was performed between the umami receptor T1R1/T1R3 and 51 peptides pre-screened using four established machine-learning models. This aimed to screen for peptides with higher umami potential before synthesis and sensory validation, while also exploring the mechanism of peptide-receptor interactions. Binding energies ranging from −5.0 to −15.0 kcal/mol are generally considered indicative of stable ligand–receptor interactions, with lower values reflecting stronger binding affinities (Gu et al., 2024). In this study, all peptides exhibited binding energies ranging from −5.0 to −7.3 kcal/mol (Table S4), suggesting stable interactions with the T1R1/T1R3 receptor. The docking energies and interaction profiles were used to screen the peptides with greater umami potential, according to previous virtual-screening studies on umami peptides (Feng et al., 2024; Wang, Han, et al., 2023). T1R1, particularly its Venus flytrap domain (VFTD), is recognized as the primary binding site for umami ligands, whereas T1R3 plays a supportive role (Liang et al., 2022). In addition, subsequent studies have suggested that the VFTD of T1R3 can also contribute to umami peptide recognition (Dang et al., 2019). In the present study, a similar pattern was observed. 34 of the 51 docked peptides were predicted to bind within the T1R1-VFTD, 4 interacted with both T1R1 and T1R3, and 8 showed preferential binding to the T1R3-VFTD. The remaining 5 peptides were predicted to bind outside the VFTD regions of the receptor model. These results indicated that the T1R1-VFTD served as the predominant binding region for umami peptides, whereas the T1R3-VFTD can also participate in the recognition of a subset of ligands.

Hydrogen bonding, hydrophobic interactions, and van der Waals forces were the principal forces mediating peptide–receptor binding, while electrostatic interactions were also observed in certain peptide–receptor complexes. These results are consistent with previous reports on the binding mechanisms of umami peptides to the T1R1/T1R3 receptor complex (Chen et al., 2024; Wang et al., 2023). As shown in Fig. 2C, peptide binding to T1R1 primarily involved key residues such as Pro445, Ala438, Val437, Trp471, Leu446, Thr436, Arg391, Ser447, Gln327, Asp435, Lys434, and Arg329. On T1R3 (Fig. 2D), frequently interacting residues included Tyr431, Tyr394, Pro244, Val178, Pro246, His515, Leu223, Leu427, Leu161, His55, and Arg54. The numerical frequencies of these contacts were summarized in Table S5. Notably, residues such as Glu, Ser, Arg, Asp, and His have been reported to play essential roles in promoting the docking affinity between umami peptides and the T1R1/T1R3 receptor complex (Feng et al., 2024; Liang et al., 2022). These residue frequency plots provided an overview of the interactions of all 51 potential umami peptides at the receptor level, highlighting key amino acid residues on T1R1 and T1R3 that are repeatedly engaged by soybean meal-derived peptides. The resulting contact pattern is informative for further analyses of umami peptide recognition and for the rational selection of new potential umami peptides.

Based on the docking results, sixteen peptides were selected for detailed interaction analysis due to their high relative abundance in FSMH samples or structural novelty, as well as favorable solubility and stability (Fig. 3A–P). Among the selected peptides, ten peptides (LE, GE, ER, SQ, LTL, VGE, SPT, AVE, VVVD and FEEINKV), were docked within the VFTD of T1R1. AE and GEDLMVQ exhibited dual binding interactions with both T1R1 and T1R3. This variation in docking positions is consistent with the presence of several partially overlapping subsites within the T1R1/T1R3 VFTDs and with the marked differences in peptide length, charge distribution, and hydrophobicity. Notably, FEEINKV and GEDLMVQ demonstrated the strongest binding affinities, with binding energies of −7.2 and − 7.0 kcal/mol, respectively. Previous studies have suggested that residues such as Lys, Leu, and Arg within peptides may enhance binding affinity through interactions with taste receptors (Wang et al., 2023). Hydrogen bonding served as the predominant interaction between the 16 peptides and the T1R1/T1R3 receptor complex, and was consistently observed across all docking results. The formation of hydrogen bonds is crucial for stabilizing peptide–receptor binding conformations (Chen et al., 2024). Hydrophobic interactions, including alkyl, Pi–alkyl, and Pi–sigma contacts, were also frequently observed and have been recognized as critical contributors to peptide–receptor binding affinity (Chen et al., 2024). Furthermore, numerous van der Waals interactions were detected surrounding the peptide chains, further contributing to the overall stability of the ligand–receptor complexes (Dang et al., 2019). Residue frequency heatmaps (Fig. 3Q–R) identified frequently interacting residues. For T1R1, Asp435, Gln327, Ser448, Trp471, Arg391, Thr436, Ala438, and Arg329 exhibited high interaction frequencies, suggesting pivotal roles in binding. On T1R3, Pro244, Pro246, Tyr394, Arg247, and Arg252 were similarly involved in peptide recognition. These results are consistent with previous studies suggesting that hydrogen bonds formed by hydrophilic residues on the receptor may play a critical role in the umami perception of these peptides (Chen et al., 2024). Peptides that engage multiple key residues within the T1R1-VFTD (e.g., FEEINKV) or establish simultaneous contacts with both T1R1 and T1R3 (e.g., GEDLMVQ) are predicted to have a higher potential to elicit umami effects than peptides bound to more peripheral regions of the receptor (e.g., TLTV). These findings suggest that both the binding mode and the extent of binding-site engagement are likely to influence the final umami perception of these peptides.

Fig. 3.

Molecular docking analysis of representative umami peptides with the T1R1/T1R3. (A–P) Predicted binding conformations and key interaction forces of 16 peptides (LE, GE, VE, ER, AE, SE, SQ, LTL, VGE, VEK, SPT, AVE, VVVD, TLTV, GEDLMVQ, and FEEINKV) with the T1R1/T1R3. Heatmaps of key interacting residues in T1R1 (Q) and T1R3 (R). The interactions shown include hydrogen bonds, alkyl and π-alkyl interactions, as well as van der Waals contacts.

3.5. Sensory evaluation of synthetic peptides

3.5.1. Basic taste of peptides by sensory evaluation

To validate the sensory characteristics of the 16 selected synthetic peptides, quantitative descriptive analysis (QDA) was conducted. As shown in Fig. 4A, all peptides exhibited multiple taste perceptions rather than a singular attribute. This may be attributed to the structural and chemical complexity of peptides and their multifaceted interactions with taste receptors (Pan et al., 2025). Except for LTL, the remaining 15 peptides primarily exhibited umami and kokumi tastes, along with varying degrees of saltiness, sourness, and astringency, whereas bitterness and sweetness were largely absent. These findings highlight the effectiveness of integrating machine learning and molecular docking for more comprehensive and accurate umami peptide screening (Chen et al., 2024; Pan et al., 2025). Among these peptides, GEDLMVQ showed the highest umami intensity, followed by VVVD, LE, AE, SE, GE, and FEEINKV, possibly due to their enrichment in umami residues (Glu, Asp, Gln, and Asn) (Gu et al., 2024; Xu et al., 2024). The sequence composition of GEDLMVQ features an N-terminal cluster of acidic residues, a hydrophobic core (Leu, Met, Val) and a C-terminal amide-containing residue (Gln). Such an arrangement can facilitate complementary electrostatic and hydrogen-bonding interactions with positively charged and polar residues in T1R1/T1R3, while providing hydrophobic contacts that stabilize the peptide within the binding pocket. Notably, peptides with relatively high umami intensity largely coincided with those predicted by molecular docking to exhibit suitable binding energies and extensive interaction networks with key residues in T1R1/T1R3. This consistency supports the predictive reliability of the docking-based screening strategy. This consistency supports the predictive reliability of the docking-based screening strategy. Similar structural features have also been reported for umami peptides derived from other food sources, including meat, mushrooms, and fermented soybean products, which are typically enriched in acidic or amide residues (Feng et al., 2024, Wang et al., 2023).

Fig. 4.

Taste characteristics of selected umami peptides. (A) Bubble plots of basic taste attributes for 16 synthesized peptides. (B) Bubble plots showing umami and saltiness-enhancing effects in MSG, I + G, and NaCl systems. (C, D) Dose–response curves of umami intensity for peptides in MSG and I + G solutions.

(E) Principal component analysis (PCA) of basic taste profiles obtained using an electronic tongue. (F) PCA of umami-enhancing effects based on electronic tongue measurements.

In contrast, TLTV and SPT lack this combination of acidic and amide residues and displayed weak umami taste, consistent with the findings of Zhao et al. (2023) who reported that umami peptides identified from fermented broad bean paste lacking D, E, Q, or N can still exhibit umami but only to a limited extent. Moreover, a high proportion of sweet residues (Ala, Ser, Gly, and Thr) has been recognized as a key determinant of umami peptides (Xu et al., 2024), while basic residues at the C-terminus (e.g., Lys and Arg) may further enhance umami intensity by facilitating receptor interactions. Umami intensity was influenced by a combination of peptide length, charge distribution, and hydrophobicity. Longer peptides in this study tend to provide multiple interaction sites that enabled extensive hydrogen bonding and electrostatic interactions with the T1R1/T1R3 receptor, while hydrophobic residues further stabilized binding within the receptor pocket. Most peptides exhibited mild sourness and astringency, which may be attributed to acidic residues from peptide synthesis (Wang, Han, et al., 2023). Notably, no peptide displayed detectable bitterness, aligning with previous studies that umami peptides can suppress bitterness by interfering with bitter taste receptor binding, thereby enhancing flavor quality (Kim et al., 2015; Pan et al., 2025). In line with previous findings, our previous study systematically compared different fermentation processes and demonstrated that optimized fermentation resulted in FSMH with enhanced umami intensity and reduced bitterness (Cao et al., 2025). These findings suggest that the fermentation process, in combination with increased umami intensity, may contribute to bitterness reduction in fermented soybean meal hydrolysates. Overall, 15 peptides presented well-balanced taste profiles with desirable umami, moderate kokumi, and saltiness attributes, underscoring their potential as natural flavor modulators in food applications.

3.5.2. Umami and saltiness enhancing effect of peptides

Umami peptides not only contribute to umami taste but also enhance flavor, potentially reducing the need for MSG and sodium in food systems (Gu et al., 2024; Zhao et al., 2023). Exploring the potential synergistic effects of FSMH derived umami peptides with MSG, I + G, and NaCl is of practical importance for flavor enhancement and sodium reduction strategies, which may promote the industrial application of fermented soybean meal as a functional ingredient in food formulations. Umami and saltiness enhancement were therefore evaluated as secondary sensory modulations associated with umami perception. As shown in Fig. 4B, all peptides exhibited umami-enhancing effects to varying degrees, with more pronounced enhancement observed in the I + G system compared to MSG alone, which was consistent with previous findings (Cao et al., 2024). LE exhibited the strongest enhancement in MSG solution, while AE showed the highest intensity in I + G solution, suggesting strong synergistic interactions with MSG and I + G, respectively (Li et al., 2020). The umami-enhancing effects observed in two systems are consistent with the cooperative activation of the T1R1/T1R3 receptor by MSG and 5′-ribonucleotides, as previously reported, where co-binding of MSG and IMP/GMP stabilizes the active receptor conformation (Zhang, Cui, et al., 2022). In this context, the soybean meal derived peptides may act as positive modulators that further stabilize the activated T1R1/T1R3 state in the presence of MSG and I + G.

Dipeptides such as GE, VE, ER, and SE also showed notable synergistic effects in both systems. Notably, two heptapeptides (e.g., GEDLMVQ, FEEINKV) demonstrated strong umami enhancement. Similarly, Gu et al. (2024) reported that long-chain peptides (GESMTDGF) could contribute significantly to umami synergy. Although saltiness-enhancing effects were generally weaker than umami enhancement, peptides such as VEK, VVVD, GE, and LE significantly intensified the perceived saltiness of NaCl solutions. Particularly, tripeptide VEK markedly increased saltiness perception, similar to saltiness-enhancing umami peptides previously identified in fermented soybean paste (Zhao et al., 2023). The saltiness enhancement may be related to interactions between the umami peptides and NaCl in the mixture, in which the peptides improved overall sensory perception and attenuated undesirable taste (such as bitterness), thereby increasing the perceived saltiness without altering the sodium content (Wang et al., 2023). Overall, in model systems, the FSMH-derived peptides evaluated in this study enhanced umami intensity and moderately increased saltiness perception for a subset of sequences. Although confirmation in complex food matrices and formulation studies is still required, these results suggest that the screened peptides are promising flavor modifiers and may contribute to future sodium-reduction strategies in food applications.

3.5.3. Taste threshold analysis

Taste threshold analysis is essential for evaluating the sensory effectiveness and practical applicability of umami peptides in food systems. As summarized in Table 1, the umami thresholds and umami-enhancing thresholds of the selected peptides exhibited considerable variation. Thresholds previously reported are marked with asterisks in Table 1 for reference (Wang et al., 2024; Xiong et al., 2022; Zhang, Zhao, et al., 2019). Since LTL failed to elicit umami perception, and TLTV and SPT exhibited weak umami intensity, threshold measurements for these peptides were not determined. The umami thresholds of peptides ranged from 0.75 to 2.77 mM, which is consistent with the threshold range (0.45–1.73 mM) reported for other umami peptides screened from soybean-based food fermented with Tetragenococcus halophilus (An et al., 2023). Dipeptides such as ER and SQ exhibited umami thresholds that were comparable to or even lower than previously identified dipeptides (Xiong et al., 2022). The tripeptides displayed relatively higher thresholds (1.50–2.77 mM), slightly exceeding those of AEA (0.80 mM) and DQR (1.11 mM) isolated from peanut protein hydrolysates (Zhang, Zhao, et al., 2019). Conversely, peptides GEDLMVQ and SQ demonstrated the lowest thresholds (0.75 mM), indicating their strong umami potential.

Although many umami peptides exist in foods at concentrations below their individual taste thresholds, they can still play a pivotal role in modulating flavor perception by imparting, enhancing, or enriching subtle taste characteristics (Song et al., 2025). To assess their potential for flavor enhancement, the umami-enhancing thresholds of all selected peptides were further determined in the presence of MSG and I + G. Most peptides exhibited significantly lower enhancement thresholds compared to their umami thresholds, suggesting pronounced synergistic interactions with typical umami substances. This pronounced difference may be attributed to synergistic activation of the T1R1/T1R3 receptor. In the presence of MSG or I + G, umami peptides may act as co-ligands or allosteric modulators, stabilizing the receptor in an active conformation and thereby lowering the concentration required to elicit a perceptible taste response (Zhang, Sun-Waterhouse, et al., 2019). Notably, GEDLMVQ exhibited the lowest enhancement threshold (0.0079 mM) in both MSG and I + G systems, indicating its exceptional potential as a flavor enhancer. Similarly, LE, AE, VEK, and FEEINKV showed marked umami-enhancing effects in both systems, while VE and SE demonstrated stronger synergy with I + G. These results demonstrated the strong taste-modulating capabilities of the identified umami peptides and support their potential application in flavor enhancement strategies aimed at reducing reliance on conventional additives.

3.5.4. Dose-response analysis

Dose–response experiments were conducted to further elucidate the release and dissipation of peptides in umami enhancement models (Cui et al., 2023). Seven peptides (LE, AE, VE, VEK, VVVD, GEDLMVQ, and FEEINKV) that exhibited strong umami-enhancing effects in both MSG and I + G systems were selected for evaluation. Within the tested concentration range (0.25–2.5 mg/mL), most peptides displayed dose-dependent increases in umami intensity in both systems. In the MSG solution (Fig. 4C), VE and VEK reached their highest umami intensities at 2.0 mg/mL, after which the responses leveled off. In contrast, the other five peptides (LE, AE, VVVD, GEDLMVQ, and FEEINKV) peaked at 2.5 mg/mL. At the lowest concentration (0.25 mg/mL), all peptides exhibited relatively weak enhancement effects, likely due to insufficient interaction with umami taste receptors. Notably, GEDLMVQ showed the highest umami intensity at this concentration, in line with its lowest umami-enhancing threshold. Additionally, AE exhibited a sharp increase in umami intensity from 0.25 to 1.0 mg/mL, followed by a transient decline and then a slower upward trend at higher concentrations. Similarly, LE showed a slight decrease in intensity between 1.0 and 1.5 mg/mL before continuing to rise, ultimately approaching the intensity level of GEDLMVQ. These results suggested that the synergistic umami-enhancing effects of peptides do not strictly follow a linear pattern, corroborating previous reports (Cui, Li, Wu and Hu, 2023, Zhang, Sun-Waterhouse, Su and Zhao, 2019). Such non-linearity is commonly attributed to factors such as receptor saturation, mixture interactions and the contribution of other taste rather than to a simple proportional relationship between peptide concentration and perceived intensity.

In the I + G solution (Fig. 4D), all peptides demonstrated pronounced umami-enhancing capacities. AE and GEDLMVQ showed the most considerable effects, reaching maximum umami intensities at 2.0 mg/mL, beyond which the increases were not statistically significant. VE and VEK exhibited relatively lower intensities at the concentrations below 1.5 mg/mL. For VVVD and FEEINKV, umami intensity increased in a dose-dependent manner across the tested range, although the rate of increase appeared to slow at higher concentrations. These findings indicated that umami intensity was not enhanced consistently at increasing concentrations of some peptides, likely due to negative sensory attributes such as intrinsic astringency and sourness (An et al., 2023). Overall, these findings suggested that, in the present model systems, peptide concentrations of approximately 2.0–2.5 mg/mL are sufficient to elicit significant umami-enhancing effects.

3.6. Taste profile analysis of synthetic umami peptides by electronic tongue

Sensory evaluation can be influenced by environmental and individual factors, and instrumental methods are therefore useful to support the results of panel. The electronic tongue, which integrates taste sensors and analog signal processing, enables accurate detection of individual taste modalities and has been widely recognized for its effectiveness in evaluating taste characteristics (Pan et al., 2025). In this study, an electronic tongue system was used as a complementary tool to sensory evaluation and was employed to objectively assess the basic taste attributes and umami-enhancing potential of selected synthetic peptides. As shown in Table S6 and Fig. 4E, eight novel umami peptides were selected for analysis. Compared with MSG controls (0.1%, 0.2%), the tested peptides displayed distinct taste profiles. All eight peptides demonstrated significantly higher umami intensity than the 0.1% MSG solution (p < 0.05), and five peptides (ER, VGE, VVVD, GEDLMVQ, and FEEINKV) exhibited umami intensities surpassing the 0.2% MSG solution. Additionally, most peptides also presented notably higher saltiness and richness scores than the controls, indicating their potential as multifunctional taste enhancers. The astringency and sourness detected in some peptides may be attributed to residual acetate or the presence of acidic amino acids from peptides, which is consistent with previous sensory observations. Principal component analysis (PCA) further differentiated the peptides from the 0.2% MSG. The PCA model explained 90.9% of the total variance, with PC1 (69.1%) primarily associated with umami and saltiness, and PC2 (13.6%) related to sourness and richness. Peptides VVVD, GEDLMVQ, and FEEINKV were closely clustered with the 0.1% MSG, while the remaining peptides were distinctly separated from the 0.2% MSG. The ranking of umami intensities obtained from the electronic tongue was consistent with sensory evaluation, with peptides such as ER, VGE, VVVD, GEDLMVQ, and FEEINKV showing high umami scores in both evaluations. Accordingly, these peptides were located on the positive side of PC1 in the PCA score plot, close to the umami and saltiness vectors, indicating good agreement with sensory assessments. These results confirm the effectiveness of electronic tongue analysis in evaluating subtle variations in taste perception among synthetic peptides, thereby providing instrumental support for the sensory evaluation.

Furthermore, seven peptides with strong umami-enhancing performance assessed through sensory evaluation were subjected to electronic tongue analysis (Table S7 and Fig. 4F). VE, LE, and AE displayed the most pronounced umami-enhancing effects, with umami intensities significantly higher than those of the 0.1% MSG control, consistent with previous sensory results. Saltiness and richness were also significantly increased, suggesting synergistic modulation of multiple taste attributes by these peptides. The PCA result revealed clear differentiation among the peptides, with the model accounting for 98.3% of the total variance. VVVD, GEDLMVQ, and FEEINKV exhibited similar taste profile, while AE, LE, and VEK formed another distinct cluster. Notably, VE was clearly separated from other peptides due to its superior umami-enhancing performance. This may be attributed to its specific amino acid composition and molecular structure. Previous studies have suggested that small peptides can synergize with umami compounds, leading to an expansion of the binding cavity of the T1R1/T1R3 receptor and thereby enhancing ligand binding ability and umami perception (Dang et al., 2019). Overall, these findings demonstrate the applicability of electronic tongue analysis in characterizing the complex taste profiles of synthetic umami peptides, providing an effective complement to the development of taste-enhancing food ingredients.

3.7. Quantitative analysis of umami peptides in FSMH samples

To evaluate the contribution of the selected umami peptides to the umami perception of FSMH samples, 14 representative peptides, selected on the basis of the preceding sensory evaluation results and their relative abundance in FSMH, were quantitatively analyzed at three critical fermentation stages (Fig. 5). Overall, co-fermentation with L. fermentum and P. fermentans significantly increased the levels of most umami peptides in FSMH, with LE and VE being the most abundant across all samples. Notably, six peptides were detected only in the fermented samples. SE, SQ, and VGE showed a progressive increase in concentration with fermentation time, which may be attributed to enhanced microbial proteolysis and peptide release during active growth and metabolism. Similar trends were reported by Chen et al. (2024), who identified high umami peptides generated during Corynebacterium glutamicum fermentation of wheat gluten hydrolysates. Mu et al. (2025) observed the formation of umami peptides (EA and EV) as early as 12 h during Aspergillus oryzae fermentation of soybean meal. Notably, nine peptides reached their peak concentrations at 24 h, followed by varying degrees of decline at 48 h, suggesting that 24 h may represent the optimal peptide release point during fermentation. These results may be due to microbial utilization or enzymatic degradation of peptides during prolonged fermentation. These findings were in agreement with previous studies that peptide formation peaked at the mid-stage of fermentation, reflecting a dynamic equilibrium between peptide generation and breakdown (Cao et al., 2025). It should be noted that the quantitative analysis of umami peptides in FSMH was performed in a complex fermentation matrix, in which ion suppression effects may occur and potentially lead to an underestimation of absolute peptide concentrations. However, all samples were analyzed under identical matrix conditions, allowing reliable relative comparisons among peptides and fermentation stages.

Fig. 5.

Quantitative analysis of umami peptides in FSMH samples at different fermentation stages.

Although numerous studies have reported novel umami peptides from various food sources, their actual concentrations in the corresponding samples have rarely been quantitated, thereby limiting accurate evaluation of their contribution to umami perception. To quantitate the potential contribution of individual peptides, the measured concentrations of the 14 umami peptides in LP24 group were compared with their umami taste thresholds and umami-enhancing thresholds in MSG and I + G systems. Taste activity values (TAVs) were calculated as the ratios of peptide concentrations to the corresponding thresholds (Table S9). LE, GE, VE, and SQ were present at concentrations approximately 3–5 times higher than their umami thresholds, indicating the direct contribution to the overall umami taste of FSMH. Furthermore, AE, VEK, and GEDLMVQ were detected at levels exceeding their enhancement thresholds in MSG and I + G systems, highlighting their potential roles as key umami-enhancing substances. In addition, peptides such as ER, SE, VGE, AVE, and FEEINKV surpassed their enhancement thresholds in I + G solutions, indicating their potential to synergistically enhance umami perception in FSMH. Several peptides showed umami-enhancing TAVs >1. In particular, LE, GE, VE, AE, VEK, and GEDLMVQ exceeded their enhancement thresholds in the presence of MSG, while LE, GE, VE, AE, SE, SQ, VGE, VEK, AVE, GEDLMVQ, and FEEINKV exhibited umami-enhancing TAVs greater than 1 in the I + G system. These findings indicated that, in addition to a few peptides that can directly impart umami, a larger subset of peptides is likely to act as umami enhancers, thereby strengthening and modulating the overall umami perception of FSMH. In summary, fermentation significantly increased umami peptide levels in FSMH, with most peptides accumulating to sufficient concentrations for umami enhancement, thereby providing evidence at the peptide level for their contribution to the improved taste quality of fermented soybean products.

3.8. Umami peptides activation of TAS1R1 induced intracellular calcium mobilization

To further evaluate whether representative umami peptides could trigger receptor associated signaling at the cellular level, intracellular Ca²⁺ mobilization was evaluated in TAS1R1 transfected HEK293 cells using Fluo-4 AM staining. Based on the results of umami intensity, umami-enhancing ability, and quantitative abundance, three representative peptides (LE, GEDLMVQ, and FEEINKV) were selected for this assay. As shown in Fig. 6, treatment with MSG or the selected peptides induced detectable intracellular Ca²⁺ responses in TAS1R1 transfected cells, whereas no detectable changes in intracellular Ca²⁺ levels were observed in non-transfected HEK293 cells (Fig. S3). The analysis of relative fluorescence changes further revealed distinct activation profiles among the tested peptides (Fig. S4). MSG induced a rapid and transient Ca²⁺ response, reaching its maximum intensity at 40 s, followed by a gradual decline (Fig. 6A). LE also triggered a transient response, but with markedly lower amplitude. In contrast, GEDLMVQ and FEEINKV elicited substantially stronger and more sustained Ca²⁺ mobilization. GEDLMVQ induced rapid intracellular Ca²⁺ mobilization and exhibited the strongest response during the early phase, reaching a peak at 180 s before gradually declining over the remaining observation period (Fig. 6C). FEEINKV displayed a slower onset but a more sustained activation profile, reaching its maximum at 240 s and remaining at a high level throughout the later stage of the recording period (Fig. 6D). The absence of detectable Ca²⁺ responses in non-transfected cells further supports that these effects were dependent on TAS1R1 associated signaling (Fig. S3).

Fig. 6.

Fluo-4 AM staining depicting intracellular Ca²⁺ mobilization in HEK293 cells expressing.

TAS1R1 after treatment with MSG or umami peptides. Cells were treated with 10 mM MSG (A), or 5 mM of the peptides LE (B), GEDLMVQ (C), and FEEINKV (D) following Fluo-4 AM staining.

Umami perception is initiated by activation of the T1R1/T1R3 receptor, which triggers a G protein mediated signaling cascade involving phospholipase C activation and inositol-1,4,5- triophosphate (IP₃) generation. IP₃ then stimulates intracellular Ca²⁺ release, ultimately leading to membrane depolarization and transmission of taste signals (Wang et al., 2026). The results in this study indicated that the selected FSMH derived peptides were able to induce TAS1R1 associated intracellular Ca²⁺ mobilization. The rapid early response elicited by GEDLMVQ may reflect faster receptor engagement, whereas the slower but sustained activation induced by FEEINKV suggested a more persistent receptor associated interaction. By contrast, LE elicited a relatively weak Ca²⁺ response, suggesting that this peptide may function less effectively as an independent receptor activator under the present conditions. In combination with the sensory results, LE may contribute more prominently to umami enhancement in systems containing MSG or I + G than to direct receptor activation when tested alone. These findings indicated that representative FSMH-derived umami peptides can induce receptor associated signaling and support their ability to initiate umami taste transduction at the cellular level.

3.9. MD simulation analysis

Intracellular Ca²⁺ mobilization assays provided functional support for TAS1R1 associated responses induced by representative umami peptides. MD simulations were further performed to obtain a structural interpretation of the distinct response behaviors. While molecular docking reveals static receptor–ligand interactions, MD simulation provides dynamic insights into the binding process and structural evolution of complexes (Ai et al., 2024; Xiong et al., 2022). Three peptides (LE, GEDLMVQ, and FEEINKV) were selected for MD simulations with the T1R1/T1R3 receptor complex over 100 ns to evaluate the structural stability and compactness (Fig. 7). Overall, all three peptide–receptor complexes remained structurally stable during the 100 ns simulations. The root mean square deviation (RMSD) was used to evaluate the conformational stability of ligand–receptor complex, as lower RMSD values and smoother curves typically reflect higher structural stability (Feng et al., 2024). As shown in Fig. 7A, all complexes exhibited RMSD fluctuations within 0.6 nm throughout the simulation. An initial increase in RMSD was observed during the first 30 ns, after which the systems reached equilibrium and maintained structural stability. The radius of gyration (Rg) was further analyzed to assess the compactness and stability of each complex. A lower Rg value is indicative of a more compact and stable structure (Xu et al., 2024). As shown in Fig. 7B, all three complexes exhibited limited Rg fluctuations (<1.0 nm), corroborating the RMSD findings and indicating that peptide binding did not induce major global destabilization of the receptor complex during the simulation (Zhang, Cui, et al., 2022). The two heptapeptide complexes stabilized after approximately 20 ns, whereas the LE complex showed a marked decrease in Rg between 30 and 50 ns before reaching equilibrium. Notably, FEEINKV exhibited the lowest average Rg, indicating the most compact binding conformation.

Fig. 7.

The results of molecular dynamics. (A) Root mean square deviation (RMSD). (B) Radius of gyration (Rg). (C) Root mean square fluctuation (RMSF). (D) Buried solvent accessible surface area (Buried SASA). (E) The number of hydrogen bonds. (F) Binding energy.

To further assess the structural influence of peptide binding on the umami receptor, the flexibility and solvent accessibility of the complex were evaluated. Root mean square fluctuation (RMSF) reflects the mobility of amino acid residues, with lower RMSF values indicating smaller positional deviations of residues. As shown in Fig. 7C, all three peptide–receptor complexes curves exhibited RMSF values within 0.8 nm, indicating minimal perturbation to the overall receptor conformation (Feng et al., 2024). The fluctuation patterns were largely consistent across the three complexes, with notable differences primarily observed in the 350–380 region. This region corresponds to an inherently flexible solvent-exposed loop near the T1R1/T1R3 interface, and the RMSF peak thus reflects local loop mobility rather than global destabilization. Compared with LE, the two heptapeptides showed slightly higher RMSF values at specific residues, suggesting more extensive or more dynamic local interactions within the receptor model (Han et al., 2024). Buried solvent-accessible surface area (buried SASA) was further analyzed to estimate the extent of peptide–receptor interface burial. Higher buried SASA values suggest more extensive intermolecular interactions and deeper ligand embedding. Among the three complexes, FEEINKV exhibited the largest buried SASA, indicating a tighter and more extensively buried interaction mode within the binding site (Fig. 7D). Consistent with this observation, the relatively stable RMSD and compact Rg values of the FEEINKV–receptor complex further supported a comparatively tight and stable binding conformation, which may help explain its more sustained Ca²⁺ response observed experimentally. The buried SASA of the GEDLMVQ–receptor complex exhibited substantial fluctuations during the early stages of the simulation, which may be attributed to initial water–residue interactions surrounding the T1R1/T1R3 interface (Ai et al., 2024). In contrast, LE exhibited the lowest buried SASA, suggesting a more superficial interaction mode. This shallower interaction pattern was consistent with its relatively weak direct Ca²⁺ response when tested alone. Interestingly, LE produced the highest umami-scores in both MSG and I + G systems. This observation indicated that extensive burial at the receptor interface may not be a prerequisite for pronounced umami enhancement in mixed systems. Instead, a comparatively accessible interaction mode may still be compatible with cooperative enhancement in the presence of MSG or I + G. These results suggested that the functional role of LE under the present conditions may be more closely related to umami enhancement in combination systems than to strong independent receptor activation. In comparison, the intermediate burial and interfacial flexibility of GEDLMVQ were consistent with its stronger early phase Ca²⁺ response, whereas the tighter embedding of FEEINKV is consistent with its slower but more sustained activation kinetics.

Hydrogen bonding and binding energy were analyzed to further clarify the interaction strength and stability of the peptide–receptor complexes (Fig. 7E-F). Hydrogen bonds have been identified as a major force mediating the interactions between umami peptides and the T1R1/T1R3 receptor in this study. Among the three complexes, FEEINKV consistently exhibited the highest average number of hydrogen bonds throughout the simulation, with a noticeable increasing trend over time, significantly exceeding those formed by GEDLMVQ and LE. This result indicated that hydrogen bonding may contribute importantly to stabilization of the FEEINKV–receptor complex. To further investigate the binding affinity of the peptide–receptor complexes, molecular mechanics Poisson–Boltzmann surface area calculations were performed. This method provides valuable insights into the binding mechanisms between ligands and target proteins (Ai et al., 2024). FEEINKV showed the most significant binding energy profile, with values markedly larger than those of GEDLMVQ and LE. Notably, the binding energies of FEEINKV and GEDLMVQ began to increase after 30 ns, whereas LE showed relatively stable energy levels throughout the simulation. The stronger interaction pattern of FEEINKV, with its higher hydrogen bonding capacity and greater buried SASA, provides a structural basis for the slower but more sustained Ca²⁺ response observed experimentally. By contrast, GEDLMVQ showed a more dynamic interaction pattern, particularly at the interfacial region, which was associated with stronger early-phase Ca²⁺ response followed by gradual decline over time. LE formed a less buried and comparatively less interaction intensive complex, consistent with its weaker direct Ca²⁺ mobilization under the present conditions.

MD simulations indicated that all three peptides formed structurally stable complexes with the T1R1/T1R3 receptor model, with favorable binding free energies and well-defined interaction interfaces. These computational results provided structural interpretations that were broadly consistent with the experimentally observed Ca²⁺ responses and sensory results. FEEINKV displayed the most stable and extensively buried interaction mode, which was in line with its sustained Ca²⁺ activation profile. Such sustained receptor associated responses may be relevant to prolonged umami perception, although sensory persistence in the oral cavity is likely influenced by multiple additional factors. GEDLMVQ showed more dynamic conformational behavior, which was compatible with its stronger early phase response. Although LE adopted a more superficial binding mode and showed weaker direct Ca²⁺ mobilization, it exhibited a pronounced umami-enhancing effect in mixed systems. This discrepancy may be related to factors beyond receptor binding mode, including the influence of aqueous solubility, conformational behavior in solution, and potential synergistic interactions with other taste-active components. Sensory attributes such as umami intensity, umami enhancement and threshold are determined by multiple factors, including receptor affinity, activation efficacy, ligand–receptor association and dissociation kinetics, secondary taste notes and in mouth transport or degradation. The docking and simulation-based interpretations are model dependent and may be affected by uncertainties in receptor structure and scoring functions. Therefore, MD results and sensory data are expected to show some degree of correlation, but a strict rank order match is not necessarily observed. In this study, the MD analyses were used to provide structural hypotheses and interpretive support, whereas receptor associated functional activation was evaluated experimentally using Ca²⁺ mobilization assays. Overall, these MD findings provide supportive structural information for interpreting the experimentally observed umami and umami-enhancing effects of LE, GEDLMVQ, and FEEINKV and help rationalize the differences in their sensory profiles.

4. Conclusions

This study employed an integrated approach combining virtual screening, molecular docking, and synthesis-based validation to identify potent umami and umami-enhancing peptides from fermented soybean meal hydrolysate, providing a sustainable route for the valorization of soybean byproducts. Virtual screening combined with docking analysis yielded 51 potential umami peptides, of which 16 peptides were synthesized for further evaluation. Sensory evaluation demonstrated that 15 peptides exhibited umami characteristics and enhancing effects, leading to the identification of 10 novel umami peptides and 14 novel umami-enhancing peptides. LE, AE, GEDLMVQ, and FEEINKV showed superior enhancement in both MSG and I + G systems. Notably, GEDLMVQ displayed the lowest umami threshold (0.75 mM) and umami-enhancing threshold (0.0079 mM), with remarkable performance in both systems. Co-fermentation with L. fermentum and P. fermentans significantly increased the concentrations of most umami peptides in FSMH, with the highest accumulation observed at 24 h. At this stage, most peptides exceeded their respective umami or enhancing thresholds, supporting their potential contribution to flavor enhancement. Intracellular Ca²⁺ mobilization assays further showed that representative FSMH-derived peptides could induce TAS1R1-associated responses at the cellular level. GEDLMVQ exhibited a stronger early phase Ca²⁺ response, whereas FEEINKV displayed a slower but more sustained activation profile. By contrast, LE elicited a weaker direct Ca²⁺ response when tested alone. Molecular docking revealed that the T1R1-VFTD served as the principal binding site, with hydrogen bonding, hydrophobic interactions, and van der Waals forces as the dominant stabilizing forces. Molecular dynamics simulations further suggested that LE, GEDLMVQ, and FEEINKV could maintain structurally stable interaction modes in the T1R1/T1R3 receptor model. The simulation results were broadly consistent with the distinct Ca²⁺ response observed experimentally and provided supportive structural information for interpreting the different functional and sensory characteristics of these representative peptides. This work presents a rapid and efficient strategy for screening umami and umami-enhancing peptides from FSMH and provides structural insights into their interactions with umami receptors, thereby supporting the application of soybean meal as a high-value functional flavor enhancer in fermented food systems. Future studies should further evaluate the activities of these peptides in complex food matrices.

CRediT authorship contribution statement

Chenchen Cao: Writing – original draft, Visualization, Software, Methodology, Investigation, Formal analysis. Weizheng Sun: Validation, Supervision. Jianping Wu: Writing – review & editing, Methodology. Mouming Zhao: Supervision, Methodology. Guowan Su: Writing – review & editing, Funding acquisition.

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

Data availability

Data will be made available on request.