Mechanistic insights into taste modulation of Xanthoceras sorbifolium leaf tea fermented by Aspergillus cristatus and Aspergillus neoniger: an integrated component–taste relationship analysis

Abstract

Xanthoceras sorbifolium leaf tea has notable pharmacological potential, but its taste limits broader acceptance. In this study, the tea was fermented with Aspergillus cristatus (AC) or Aspergillus neoniger (AN), and taste modulation was investigated by electronic tongue analysis and untargeted metabolomics. AC fermentation increased bitterness and sweetness but reduced umami and saltiness, whereas AN increased sourness and reduced saltiness and bitterness. Integrated metabolite–taste analysis identified 34 and 48 key differential metabolites in AC- and AN-fermented teas, respectively. Redundancy analysis highlighted flavonoids, amino acids, triterpenoid saponins, and glycosides as major contributors. Time-series clustering revealed their dynamic changes during fermentation. Molecular docking indicated favorable metabolite–receptor interactions, and microscale thermophoresis further validated the binding affinities of key metabolites to corresponding taste receptors. These findings explain the metabolic basis of fermentation-driven taste modulation and support targeted microbial processing to improve fermented herbal tea quality.

Highlights

• •Strain-specific fermentation drives distinct taste trajectories in XLT.
• •Sensory-metabolomics integration deciphers taste-metabolite relationships.
• •Flavonoids, amino acids and triterpenoid saponins drive taste variations.
• •Molecular docking links non-volatile metabolites to taste receptor interactions.

1. Introduction

Xanthoceras sorbifolium is a woody oil-bearing tree species endemic to northern China that has attracted increasing attention as an important economic crop due to its low investment requirements and considerable economic potential (Chen et al., 2024; Kuang et al., 2025; Zang et al., 2021). According to the National Forestry and Grassland Administration of China, Xanthoceras sorbifolium has become an important woody economic crop, with a cultivation area of approximately 267,000 ha and an annual industrial output value exceeding RMB 10 billion in 2025. Despite its broad distribution across 14 provinces, the economic value of X. sorbifolium is still largely dependent on seed oil production, leaving its leaves as an abundant but underutilized by-product resource (Chen et al., 2022; Han et al., 2025). As one of the most valuable by-products generated during the cultivation of Xanthoceras sorbifolium, the leaves have long been traditionally processed into herbal tea and consumed daily by people in northern China (Ren et al., 2025). Moreover, China has established corresponding food production standards, formally recognizing Xanthoceras sorbifolium leaves as a novel food ingredient for beverage production (Announcement No. 5, 2023), thereby underscoring their edible value.

Xanthoceras sorbifolium leaf tea (XLT) is rich in various bioactive components, mainly including saponins, flavonoids, polyphenols, tannins, and small peptides (Miao et al., 2023; Zhou et al., 2022). When consumed as a tea beverage, XLT can exert multiple pharmacological activities (Xiao et al., 2022). Studies have shown that XLT has preventive and therapeutic effects against diabetes (Jia et al., 2025). Long-term consumption can exert diuretic effects, alleviate rheumatism, and contribute to the regulation of blood glucose, blood lipid, and uric acid levels (Li, Shao, et al., 2025; Ta et al., 2023), and is therefore regarded as beneficial for the management of metabolic risk factors associated with hyperglycemia, hyperlipidemia, and hyperuricemia. XLT is widely recognized for its safety and considerable functional potential. However, compared with conventional teas, its market penetration remains limited by undesirable taste and low consumer acceptance, which may be largely attributed to the high abundance of saponins and other bitter compounds in the leaves (Cui et al., 2024).

Microbial fermentation is an important food-processing technology that has been extensively applied in tea production, and fermented teas currently account for a substantial proportion of global tea consumption (Ruan et al., 2025). Compared with traditional unfermented teas, microbially fermented teas exhibit distinctive sensory characteristics, including brighter infusion color, unique aroma profiles, enhanced sweetness, and a mellow, lingering aftertaste (Wang et al., 2023), which have contributed to their broad consumer acceptance. Aspergillus neoniger (AN) and Aspergillus cristatus (AC) are commonly isolated from dark tea and are recognized as dominant microorganisms involved in tea fermentation. Both species are generally regarded as safe for food applications and have been widely applied in food fermentation processes (Chen et al., 2023; Hu et al., 2025). In our previous studies, fermentation with these fungal strains was found to induce substantial alterations in the metabolic profiles of tea, accompanied by an enhancement of hypoglycemic activity (Xiang et al., 2021). In addition, fermentation of Kuding tea using AC and AN resulted in pronounced modifications of taste attributes, leading to an improvement in its originally unfavorable sensory characteristics (Zeng et al., 2024). Therefore, microbial fermentation mediated by AC and AN may represent a promising strategy for improving the sensory quality of XLT while simultaneously enhancing its functional properties.

Tea quality largely depends on its taste attributes, and the sensory characteristics of tea are primarily determined by its non-volatile components (Zhou et al., 2026). The generation of undesirable flavors in tea mainly stems from an imbalance in the composition of taste-active metabolites. Deviations from optimal concentrations (whether excessive or deficient) of compounds such as caffeine, catechins, flavonoids, phenolic acids, and organic acids significantly affect the sensory quality of tea infusions (Deng et al., 2022; Ye et al., 2022). Furthermore, there is evidence that amino acids, peptides, and soluble sugars contribute sweetness and umami to tea (Chang et al., 2026; Mo et al., 2026). These chemical components not only act directly on the sensory system but also modulate the overall flavor profile through complex synergistic or antagonistic interactions. Because fungal fermentation fundamentally alters this chemical matrix, it serves as a critical process for reshaping tea flavor. To systematically elucidate the underlying mechanisms, an integrated analytical strategy was established. An electronic tongue system was utilized to objectively quantify macroscopic taste variations (Liu, Qian, et al., 2024), while liquid chromatography-mass spectrometry (LC-MS)-based untargeted metabolomics was applied to profile the differential non-volatile metabolites at the microscopic level (Liang et al., 2026; Zhang et al., 2025). In addition, molecular docking was applied to simulate the binding modes between metabolites and taste receptors, while microscale thermophoresis (MST) was used to further validate their binding affinities and molecular interactions (Li, Xia, et al., 2025; Ma et al., 2025).

Therefore, this study first employed two representative microbial strains commonly used in traditional fermented teas to ferment XLT, and the fermentation-induced flavor changes were evaluated using electronic tongue analysis. Ultra-high-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry (UPLC-QTOF-MS/MS) was applied to characterize the key differential metabolites among different sample groups. Subsequently, untargeted metabolomics was integrated with electronic tongue analysis to perform component-taste correlation analysis, thereby linking flavor changes with their potential chemical basis and identifying the key compounds responsible for fermentation-induced flavor modulation. Time-series clustering analysis was subsequently performed to characterize the dynamic variation patterns of these compounds throughout fermentation. Finally, molecular docking analysis was conducted between the identified key flavor-related differential compounds and their corresponding taste receptors, and the predicted interactions were further validated using microscale thermophoresis (MST) to further explore the potential molecular mechanisms underlying flavor modulation at the receptor level.

2. Materials and methods

2.1. Materials and reagents

Acetonitrile (HPLC grade) was purchased from Merck Chemicals (Darmstadt, Germany). HPLC-grade deionized water was obtained using a Milli-Q water purification system (Bedford, MA, USA). Rutin and phenylethyl primeveroside standards (purity ≥98.0%) were purchased from Shanghai Yuanye Bio-Technology Co., Ltd. (Shanghai, China). The XLT samples were obtained from the Etuoke Qianqi Biyuan Xanthoceras Sorbifolium Planting Professional Cooperative in Ordos City, Inner Mongolia Autonomous Region, China, and were identified by Professor Wenjie Wang of Hubei University of Chinese Medicine. AC (GenBank ID: OP588748) and AN (GenBank ID: OP588747) were isolated and purified from dark tea samples (Zeng et al., 2023).

2.2. Fermentation of samples

Based on previously reported fermentation methods (Zeng et al., 2024), the procedure was slightly modified in this study. AC and AN, purified and cultivated to their optimal active growth phase, were prepared as spore suspensions, with the concentration adjusted to approximately 1 × 10⁷ CFU mL⁻¹. XLT was ground into a fine and homogeneous powder to ensure uniform contact with the spore suspension during fermentation. After sterilization at 121 °C for 40 min, the tea powder was thoroughly mixed with the corresponding spore suspensions and subsequently incubated in a constant temperature and humidity chamber for 10 days. According to the fungal species used, the samples were divided into the AC-fermented 10-day group (FAC) and the AN-fermented 10-day group (FAN). Both the unfermented day-0 group (UF0) and the unfermented day-10 group (UF10) received sterile water instead of the spore suspension. The UF10 group was incubated under the same constant-temperature and constant-humidity conditions as the fermentation groups. Each group was prepared in six independent replicates (n = 6) to ensure experimental robustness and reproducibility. Samples from both fermented and unfermented groups were stored in sterile polyethylene bags and kept at −80 °C prior to further analysis.

The samples used for the dynamic fermentation study were prepared as described above and divided into three groups: UF, FAC, and FAN. Samples were collected and immediately frozen on days 0, 5, 10, and 15 of incubation, yielding 12 groups: UF0, UF5, UF10, UF15, FAC0, FAC5, FAC10, FAC15, FAN0, FAN5, FAN10, and FAN15. Each group comprised six independent biological replicates (n = 6).

2.3. Sample extraction

For each group, 0.6 g of XLT sample was added to 120 mL of boiling water and immediately incubated in a 90 °C water bath for 20 min. A portion of the resulting extract was used for subsequent taste analysis, while the remaining extract was centrifuged at 12,000 rpm for 10 min at 4 °C for untargeted metabolomics analysis. Curcumin was used as the internal standard, with a final concentration of 200 ng mL⁻¹ in the samples. Aliquots of the supernatants from all samples were pooled to generate a quality control (QC) sample for LC-MS analysis.

2.4. E-tongue analysis

Electronic tongue analysis was performed using an α-ASTREE electronic tongue system (Alpha MOS, Toulouse, France), equipped with an Ag/AgCl reference electrode and seven taste sensors, including AHS (sourness), NMS (umami), CTS (saltiness), ANS (sweetness), SCS (bitterness), as well as two universal sensors, PKS and CPS. For each measurement, 100 mL of XLT extract was transferred into a 150 mL sample cup for detection. During the analysis, the sensors were rinsed with 100 mL of deionized water between each measurement. Each sample was measured over seven consecutive cycles, with each cycle lasting 2 min. The average response value of the three cycles with the highest stability was selected for subsequent statistical analysis.

2.5. Untargeted metabolomics through UPLC-QTOF-MS/MS analysis

Metabolite separation was performed using an ACQUITY UPLC H-Class system (Waters, Milford, MA, USA) equipped with a Waters ACQUITY UPLC BEH C18 column (100 mm × 2.1 mm, 1.7 μm). The mobile phases consisted of phase A, a mixture of water and formic acid (1000:1, v/v), and phase B, acetonitrile. The gradient elution program was set as follows: 0 min, 5% B; 20 min, 50% B; 27 min, 95% B; 30 min, 95% B; 31 min, 5% B; and 35 min, 5% B. The flow rate was maintained at 0.3 mL min⁻¹, the column temperature was set at 40 °C, and the injection volume was 2.0 μL.

Mass spectrometric analysis was carried out using a Waters Xevo G2-XS QTOF system equipped with an electrospray ionization (ESI) source (Waters, Milford, MA, USA). Data acquisition was performed in sensitivity mode, with both positive and negative ionization modes applied for qualitative metabolite analysis. In addition, untargeted metabolomics analysis was conducted in positive ion mode using centroid data. The optimized operating parameters were as follows: desolvation temperature, 500 °C; source temperature, 100 °C; capillary voltage, 3 kV; cone voltage, 20 V; collision energy, 30–40 eV; desolvation gas flow, 1000 L h⁻¹; and cone gas flow, 50 L h⁻¹. Full-scan mass spectra were acquired over an m/z range of 50–1500 Da with a scan time of 1.0 s, and data were collected in MS^E mode.

2.6. Taste receptor molecular docking

The three-dimensional structure of the sweet taste receptor T1R2/T1R3 (PDB ID: 9NOR) was obtained from the RCSB Protein Data Bank (https://www.rcsb.org/). As no experimentally resolved three-dimensional structures are currently available for the umami taste receptor T1R1 (UniProt ID: Q7RTX1)/T1R3 (UniProt ID: Q7RTX0), the bitter taste receptor TAS2R14 (UniProt ID: Q9NYV8), the sour taste receptor OTOP1 (UniProt ID: Q7RTM1), or the salty taste receptor TMC4 (UniProt ID: Q7Z404), their amino acid sequences were retrieved from the UniProt database (https://www.uniprot.org/) and subjected to structural prediction (Jiang et al., 2025) using AlphaFold 3.0 (https://alphafoldserver.com/). The predicted three-dimensional structures of the taste receptors are provided in Fig. S3. To evaluate the quality of the predicted three-dimensional protein structures, the taste receptor models were assessed using the SAVES v6.0 server (https://saves.mbi.ucla.edu/). Ramachandran plots and ERRAT scores were obtained to examine the stereochemical quality and overall reliability of each protein model.

Three-dimensional ligand structures were obtained from PubChem (https://pubchem.ncbi.nlm.nih.gov/) or constructed using ChemDraw 21 (PerkinElmer Informatics, Inc., Waltham, MA, USA). The protonation states of the ligands at the measured tea-infusion pH were predicted using the Tamarind Bio Protonation State tool (https://app.tamarind.bio/tools/protonation-state), and the structure with the highest Boltzmann population was retained for molecular docking. Molecular docking between the small-molecule ligands and taste receptor proteins was performed using CB-Dock2 (http://cao.labshare.cn/cb-dock/) (Liu et al., 2022; Zeng et al., 2026). CB-Dock2 was operated in blind-docking mode with its default cavity-detection algorithm and AutoDock Vina parameters. The binding pose within the highest-ranked cavity was selected according to the Vina score, which was used as an indicator of ligand–receptor binding affinity (Liu et al., 2022; Zeng et al., 2026). In addition, LigPlus was employed to obtain and visualize the 2D ligand-protein interaction profiles.

2.7. Microscale thermophoresis validation of metabolite–taste receptor interactions

The interaction between the protein and ligands, predicted by molecular docking, was experimentally evaluated using microscale thermophoresis (MST) (Huang & Zhang, 2021). Purified TMC4 protein was fluorescently labeled using the RED-NHS 2nd Generation Protein Labeling Kit (NanoTemper Technologies, Munich, Germany) according to the manufacturer's instructions. Binding measurements were performed in 1× PBST buffer (1× PBS containing 0.05% Tween-20). Phenylethyl primeveroside and rutin were separately subjected to 16-point twofold serial dilutions, with a nominal ligand concentration range from 1 mM to 3.05 × 10⁻⁵ mM, and subsequently mixed with equal volumes of labeled TMC4 protein. The samples were loaded into Monolith NT.115 Series Capillaries and analyzed using a Monolith NT.115 instrument (NanoTemper Technologies, Munich, Germany) at 25 °C. Measurements were performed using Nano-Red excitation at 100% excitation power and medium MST power. Three replicate measurements were conducted for each protein–ligand pair. Binding curves were fitted using MO.Affinity Analysis software (version 2.3, NanoTemper Technologies), and the apparent dissociation constant (Kd) was calculated to quantify the binding affinity.

2.8. Statistical analysis

Raw data from UPLC-QTOF-MS/MS were acquired using MassLynx v4.1 (Waters, Milford, MA, USA). Subsequently, the open-access software MS-DIAL (version 4.9) was employed for comprehensive data processing, including peak detection, peak alignment, spectral deconvolution, metabolite identification, and normalization (Tsugawa et al., 2015). The retention time for peak collection was set from 1 to 30 min. Based on the expected range of target components, peaks with a minimum peak height of 400 and an m/z range of 100–1500 Da were selectively retained, while the m/z range for fragment ions was set to 50–1500 Da. The mass tolerance was set to 0.012 Da for MS and 0.015 Da for MS/MS. Several common positive ion adducts, including [M + H]⁺, [M + Na]⁺, [M + K]⁺, and [M + H-H₂O]⁺, were used to determine the elemental composition of the peaks. In addition, under negative ion mode, [M-H]⁻, [M + HCOOH-H]⁻, and [M-H₂O-H]⁻ were also used in conjunction with positive ion adducts for auxiliary calibration. All sample features were aligned using a retention time tolerance of 0.13 min and an MS tolerance of 0.012 Da. For peak deconvolution, the σ window value and the MS/MS abundance cutoff were set to 0.6 and 100, respectively. Peak response values were normalized using an internal standard-based method combined with the LOWESS algorithm. Finally, to ensure that the screened metabolites accurately represented each group, a peak was required to be present in at least one group, and all samples within that group had to consistently display the corresponding peak.

Multivariate statistical analyses, including principal component analysis (PCA) and orthogonal partial least squares-discriminant analysis (OPLS-DA), were performed using SIMCA 14.1 software (Umetrics AB, Umeå, Sweden). Data visualization was carried out using Origin 2021 (OriginLab Corporation, Northampton, MA, USA) and R software (version 4.3.2; R Foundation for Statistical Computing, Vienna, Austria). Statistical analyses were conducted using SPSS version 26.0 (SPSS, Inc., Chicago, IL, USA). Differences were considered statistically significant at P < 0.05 based on the Mann-Whitney U test. Qualitative identification of these features was performed by comparing their fragment ion patterns with reported data from the literature (Kuang et al., 2025) and by querying public databases, including the Human Metabolome Database (https://www.hmdb.ca), FooDB (https://www.foodb.ca), and MassBank (https://massbank.eu/MassBank/). Quantitative determination of metabolites was performed using QuanLynx software v4.1 (Waters, Milford, MA, USA).

3. Results

3.1. Changes in taste of XLT after fermentation

As shown in Fig. 1A, XLT exhibited noticeable changes in infusion color after fungal fermentation. No obvious difference was observed between the two unfermented groups (UF0 and UF10). AC fermentation resulted in a brighter and redder XLT infusion, whereas the AN-fermented sample exhibited a distinctly darker, nearly black appearance.

Fig. 1.

Sensory profile analysis of Xanthoceras sorbifolium leaf tea (XLT) before and after fermentation. (A) Visual appearance of tea infusions from unfermented (UF0, UF10) and fermented (FAC, FAN) groups. (B) Radar plot illustrating the sensory characteristics of different groups utilizing the electronic tongue system. “UF0” and “UF10” represent the unfermented control groups at day 0 and day 10; “FAC” and “FAN” represent samples fermented by Aspergillus cristatus and Aspergillus neoniger for 10 days, respectively.

The radar plot derived from electronic tongue analysis is presented in Fig. 1B. Based on the responses of five taste sensors, XLT exhibited significant and strain-specific taste changes after fermentation with the two microbial strains compared with UF10. In the FAC group, XLT showed significant changes in four taste attributes-umami (NMS), sweetness (ANS), bitterness (SCS), and saltiness (CTS) (P < 0.01). In contrast, in the FAN group, XLT exhibited significant changes in sourness (AHS), bitterness (SCS), and saltiness (CTS) (P < 0.01). Specifically, in FAC, XLT showed decreased umami (NMS) and increased sweetness (ANS), whereas an increase in sourness (AHS) was observed only in FAN, indicating strain-dependent taste modulation induced by fermentation. Conversely, XLT exhibited a significant decrease in saltiness (CTS) in both fermentation groups, with a more pronounced reduction in FAN. Notably, XLT showed opposite trends in bitterness (SCS), increasing after AC fermentation but decreasing after AN fermentation.

3.2. Multivariate statistical analysis and qualitative characterization

The overall metabolic differences among the unfermented, fermented, and QC groups were visualized by PCA, as shown in Fig. 2A. Samples from the QC group clustered tightly, indicating good analytical stability and reproducibility throughout the detection process. No clear separation was observed between the two control groups (UF0 and UF10), suggesting that aseptic incubation for 10 days exerted minimal influence on the metabolite profiles of XLT. In contrast, samples from the fermented groups displayed obvious separation in the PCA score plot, demonstrating pronounced alterations in the metabolic profiles of XLT following fermentation and confirming the reliability of the fermentation procedure. Furthermore, as shown in Fig. S1, OPLS-DA score plots revealed clear discrimination between each fermented group and the UF10 group. In the result of 999 permutation tests, the results of scores and regression line indicating the absence of over-fitting in the original OPLS-DA model.

Fig. 2.

Overview of metabolomic profiling and differential metabolite classification. (A) Principal component analysis (PCA) score plot of the QC, unfermented (UF0, UF10), and fermented (FAC, FAN) groups. (B) Pie chart showing the chemical classification of all identified taste-related differential metabolites. (C) Number of upregulated and downregulated taste-related differential metabolites in the FAC group compared to UF10. (D) Number of upregulated and downregulated taste-related differential metabolites in the FAN group compared to UF10.

After processing the raw data using MS-DIAL, a total of 3675 detectable features were obtained in the normalized data matrix. These features were further screened based on variable importance in projection (VIP > 1.1) and fold change between fermented and unfermented groups (FC > 1.5 or FC < 0.667). As a result, 1311 and 1905 candidate features were retained in the UF10-FAC and UF10-FAN comparisons, respectively. Following metabolite identification, 39 and 60 metabolites were successfully identified in the UF10-FAC and UF10-FAN groups, respectively. After quantitative analysis, statistical significance testing, and correlation analysis with taste attributes, a total of 59 key taste-related differential metabolites were ultimately identified.

The classification and number of the 59 identified compounds are shown in Fig. 2B. Flavonoids constituted the largest proportion, with 20 compounds accounting for 33.9% of the total. The remaining compound classes, in descending order of proportion, were triterpenoid saponins (13.58%), others (11.86%), amino acids (8.47%), glycosides (8.47%), alkaloids (6.78%), nucleosides (5.08%), and terpenoids (5.08%). When analyzed separately, 34 and 48 compounds were identified in the UF10-FAC and UF10-FAN comparisons, respectively. Supplementary Table S1 provides the qualitative information of these metabolites, including molecular formula, accurate mass data, mass error, adduct type, MS/MS fragments, retention time, and identification number. Supplementary Table S2 summarizes the fold change (FC) values of differential metabolites in UF10-FAC and UF10-FAN, as well as the normalized relative quantitative results expressed as mean ± SD for each group. Overall, these results indicate that the two different fermentation strategies exerted distinct effects on the metabolic profiles of XLT.

3.3. Fermentation-induced changes in differential metabolites

To further investigate the variation of these differential metabolites among different fermentation groups, the relative contents of the metabolites were analyzed across the fermentation treatments. As shown in Fig. 2C, among the differential metabolites in FAC, 16 were upregulated and 18 were downregulated. As illustrated in Fig. 2D, 15 metabolites were upregulated and 33 were downregulated in FAN. Overall, the majority of differential metabolites in both groups exhibited a decreasing trend after fermentation; however, the proportion of upregulated and downregulated metabolites in the FAC group was relatively more balanced compared with that in the FAN group.

Fig. 3 were more clearly illustrated the changes in metabolite abundance shown in the classified heatmaps between the two fermentation groups. In Fig. 3A, following AC fermentation, most compounds showed a decreasing trend, except for amino acids, glycosides, and nucleosides, whereas a higher number of compounds from other categories exhibited increased levels. In contrast, as shown in Fig. 3B, apart from nucleosides and alkaloids, which contained more upregulated compounds, most metabolite classes in the AN-fermented group were dominated by downregulated compounds.

Fig. 3.

Hierarchical clustering heatmaps of taste-related differential metabolites. (A) Heatmap comparison between UF10 and FAC groups. (B) Heatmap comparison between UF10 and FAN groups. The color scale from blue to red represents the normalized relative abundance of metabolites from low to high. Rows represent individual metabolites, and columns represent sample replicates. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Amino acids in the AC group (M1, M4, M11, M14) and those in the AN group (M1, M4, M8, M14) both exhibited an overall decreasing trend. For flavonoids, 62.5% were upregulated in the AC fermentation group, whereas a large proportion of flavonoids were downregulated in the AN fermentation group, with 12 compounds accounting for 70.6% of the total flavonoids identified in FAN. A similar pattern was observed for triterpenoid saponins: 80.0% were upregulated in FAC, while 85.7% were downregulated in FAN. Interestingly, all three nucleosides (M2, M3, M6) identified in FAN were upregulated, whereas one of these nucleosides (M2), which was also identified in FAC, exhibited a downregulated trend in the FAC group.

3.4. Integrated correlation analysis between chemical components and taste changes of XLT

To link taste attributes with metabolomic variations, Spearman correlation analysis was performed between the relative quantitative data of differential metabolites and the response values of the five electronic tongue taste sensors to explore the associations between taste changes and metabolite up- or downregulation (Fig. 4). For each fermentation group, only taste attributes that exhibited significant changes were included in the correlation analysis.

Fig. 4.

Integrated correlation network analysis connecting differential metabolites with taste attributes. (A) Correlation network for the FAC group. (B) Correlation network for the FAN group. The connecting lines represent significant correlations based on Spearman's rank correlation coefficient. The line color indicates positive (brown) or negative (green) correlations, and the line width corresponds to the strength of the correlation coefficient. The squares on the heatmap represent pairwise correlations between metabolites. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

As shown in Fig. 4A, the correlations between 34 differential metabolites in FAC and the four taste attributes that exhibited significant changes, including sweetness (ANS), saltiness (CTS), umami (NMS), and bitterness (SCS), are visualized. Following AC fermentation, changes in sweetness and umami were observed that were not present in the AN fermentation. Most metabolites showed negative correlations with sweetness (17/26) and positive correlations with umami (16/28). Notably, all amino acids (M1, M4, M11, and M14) and all glycosides (M17, M20, M28, and M45) were negatively correlated with sweetness while exhibiting positive correlations with umami. As illustrated in Fig. 4B, the correlations between 48 differential metabolites in FAN and three significantly altered taste attributes, including sourness (AHS), saltiness (CTS), and bitterness (SCS), are presented. The increase in sourness was a characteristic change observed only after AN fermentation. Most metabolites were negatively correlated with sourness (17/32), among which 71% of triterpenoid saponins showed negative correlations with sourness (M47, M41, M58, M57, M53, and M49). In contrast, all nucleosides (M2, M3, and M6) exhibited positive correlations with sourness.

To further focus on the key metabolites or metabolite classes responsible for different taste attributes, redundancy analysis (RDA) was performed to interpret the relationships between metabolites and taste variations (Haq et al., 2023). In the RDA models (Fig. 5), the first two axes explained 98.85% of the total variance in FAC (RDA1: 98.09%, RDA2: 0.76%) and 99.77% in FAN (RDA1: 99.61%, RDA2: 0.16%), indicating that the metabolite–taste relationships in both fermentation groups were mainly driven by RDA1. RDA analyses of FAC and FAN both demonstrated that taste variation was predominantly explained by RDA1; however, the contribution of specific taste attributes and their associated metabolites differed between the two fermentation systems. As shown in Fig. 5A, sweetness (ANS), umami (NMS), saltiness (CTS), and bitterness (SCS) exhibited longer vectors in FAC, indicating that these taste attributes contributed more strongly to the overall sensory variation. In contrast, Fig. 5B shows that sourness (AHS), saltiness (CTS), and bitterness (SCS) displayed longer vectors in FAN, suggesting a greater influence of these taste attributes on the overall sensory profile. These observations are consistent with the previously identified significant taste changes.

Fig. 5.

Redundancy analysis (RDA) biplots illustrating the relationships between key metabolites and taste attributes. (A) RDA plot for the FAC group. (B) RDA plot for the FAN group. In an RDA plot, red arrows represent taste attributes, triangles indicate samples from different groups, and blue dots represent key differential metabolites. The angle between the arrows and the dots reflects the correlation (acute angle: positive correlation; obtuse angle: negative correlation). The projection distance of each dot onto the arrow vector represents the strength of the correlation, with a greater distance indicating a stronger correlation. The structures of the top five metabolites most strongly associated with each taste attribute are presented. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

For each taste attribute, key taste-related differential metabolites were required to rank among the top five positive contributors in the RDA, based on their projection lengths onto the corresponding taste vector, and to exhibit a Spearman correlation coefficient greater than 0.6. In FAC, the compounds associated with umami and saltiness were M4, M20, M17, M15, and M11, which were mainly composed of amino acids and glycosides. The compounds influencing sweetness and bitterness in FAC included M58, M13, M55, and M57, among which triterpenoid saponins accounted for 75% (M58, M55, and M57). In FAN, the compounds associated with sourness included M18, M31, M26, M2, and M56; most belonged to flavonoids and phenolic acids, whereas M56 was a cyclic pentapeptide isolated from AN. The compounds influencing saltiness and bitterness in FAN included M29, M22, M52, and M15. Among them, rutin (M29) and myricetin-3-rutinoside (M22), both belonging to flavonoids, contributed most strongly to the decreases in saltiness and bitterness.

3.5. Temporal changes in differential metabolites during fermentation

To evaluate the temporal stability and reproducibility of the differential metabolites during fermentation, samples from the UF, FAC, and FAN groups were collected at 0, 5, 10, and 15 days. The tea infusion colors at the different fermentation stages are shown in Supplementary Fig. S2, and the observed color changes were consistent with those described above. As shown in Fig. 6A, the PCA score plot revealed substantial overlap among the four UF groups and the FAC0 and FAN0 groups, indicating similar metabolite compositions and abundance profiles before fermentation. In the FAC group, FAC5 was clearly separated from FAC10 and FAC15, whereas the latter two groups were more similar. In the FAN group, the metabolic profiles of FAN5, FAN10, and FAN15 were relatively similar, with the confidence regions of FAN10 and FAN15 showing substantial overlap. Fig. 6B and C further illustrate the overall distributions and probability densities of the differential metabolites identified in the two fermentation groups. The median values of FAC10 and FAC15, as well as those of FAN10 and FAN15, were nearly identical, suggesting that the differential metabolite profiles of both fermentation systems had reached a relatively stable state by day 10.

Fig. 6.

(A) Principal component analysis (PCA) of the detected peak areas in quality-control samples (QC), unfermented samples (UF), and XLT samples fermented with AC or AN. The numbers 0, 5, 10, and 15 indicate fermentation times of 0, 5, 10, and 15 days, respectively. Raincloud plots of the differential metabolites in the AC (B) and AN (C) groups. For days 5, 10, and 15, the |log₁₀(FC)| values were calculated using UF0 as the reference group.

To characterize the temporal trajectories of metabolites during fermentation, fuzzy c-means clustering was performed. Previously identified differential metabolites showing poor reproducibility or unstable temporal patterns were excluded according to the following criteria: (1) no significant difference between the day-10 fermented samples and the corresponding unfermented controls; (2) a direction of change between the day-10 fermented samples and unfermented controls that was opposite to that observed in the preliminary analysis; or (3) a significant reversal in the direction of change across days 0, 5, and 10. As shown in Fig. 7A, the differential metabolites in FAC were grouped into four clusters according to similarities in their temporal patterns. Clusters 1 and 4 exhibited decreasing trends, whereas clusters 2 and 3 showed increasing trends. Notably, clusters 1, 2, and 4 underwent pronounced changes between days 5 and 10. M10 and M51 were excluded from subsequent analysis because of their irregular and poorly reproducible temporal patterns. As shown in Fig. 7B, the differential metabolites during FAN fermentation were classified into five clusters according to their temporal variation patterns, among which only Cluster 3 exhibited an overall increasing trend. In contrast to FAC fermentation, most changes in metabolite abundance during FAN fermentation occurred within the first 5 days, indicating that FAN-induced metabolic transformations occurred relatively earlier. Nevertheless, the changes in metabolite abundance tended to stabilize between days 10 and 15 in both fermentation systems. M7, M34, M42, and M59 were excluded from subsequent analyses because of their irregular temporal variation patterns during fermentation.

Fig. 7.

Temporal clustering analysis of metabolites during AC (A) and AN (B) fermentation. The line plots on the left show the metabolite clusters generated by fuzzy c-means clustering according to similarities in their temporal variation patterns. The heatmaps on the right visualize changes in metabolite abundance after Z-score normalization.

3.6. Molecular docking and MST validation of key metabolite–taste receptor interactions

All five protein models showed high ERRAT scores (Supplementary Table S4), indicating good consistency of non-bonded atomic interactions. Moreover, most residues were located in the most favored and additionally allowed regions of the Ramachandran plots (Supplementary Fig. S4), confirming satisfactory stereochemical quality and reasonable backbone conformations. Thus, these receptor models were considered reliable for subsequent molecular docking analyses. In the preceding analysis, 17 key taste-related differential metabolites were identified by combining Spearman correlation analysis (r > 0.6) with RDA-based screening of the top five metabolites projected onto each taste vector. To further validate their potential interactions with taste-related proteins, these 17 key differential metabolites were selected as ligands for molecular docking with their corresponding taste receptor proteins, including the umami receptor (T1R1/T1R3), sweet receptor (T1R2/T1R3), salty receptor (TMC4), sour receptor (OTOP1), and bitter receptor (TAS2R14). According to the docking results, all compounds exhibited high affinity toward their corresponding taste receptors, with docking scores (Fig. 8) lower than −5.0, suggesting that these metabolites may contribute to taste perception. Notably, most compounds showed binding energies lower than −7.0 (25/29), indicating strong binding affinities with the receptors, which may facilitate the generation of the corresponding taste sensations.

Fig. 8.

Molecular docking analysis of key taste-related metabolites with their corresponding receptors. (A) Docking results for key metabolites in the FAC group. (B) Docking results for key metabolites in the FAN group. Left: Heatmap of docking scores (kcal/mol); darker colors indicate lower binding energy and stronger affinity. Right: Representative 3D and 2D diagrams of ligand-receptor interactions for the compound exhibiting the highest binding affinity (marked with *) for each taste receptor.

For each taste attribute, the docking scores of the selected compounds were visualized graphically (Fig. 8), together with representative three-dimensional (3D) and two-dimensional (2D) ligand-receptor interaction conformations. The results indicated that the key metabolites were accommodated within the receptor binding pockets through multiple non-covalent interactions, including hydrogen bonding and hydrophobic interactions. Detailed information on the interacting amino acid residues and additional docking parameters is provided in Supplementary Table S6. In FAC (Fig. 8A), compounds associated with umami (NMS) and saltiness (CTS) exhibited strong binding affinities toward their corresponding receptors. Notably, the glycosides Icariside F2 (M17) and Phenylethyl primeveroside (M20) showed the strongest binding affinities among the NMS-related compounds, while Phenylethyl primeveroside (M20) also displayed high affinity in the CTS-related docking analysis, suggesting its potential involvement in the modulation of both umami and saltiness perception. For bitterness (SCS) and sweetness (ANS), triterpenoid saponins, including M58, M55, and M57, generally exhibited lower docking scores than other compound classes, indicating stronger interactions with the corresponding receptors. In FAN (Fig. 8B), compounds associated with sourness (AHS), saltiness (CTS), and bitterness (SCS) showed pronounced binding affinities toward their respective receptors. In the AHS-related docking analysis, the AN-specific secondary metabolite malformin (M56) exhibited the strongest binding affinity among the tested compounds. In the saltiness (CTS)- and bitterness (SCS)-related docking analyses, rutin (M29) showed the strongest binding affinity toward both receptors, while flavonoids such as myricetin-3-rutinoside (M22) and rutin (M29) displayed notably low docking scores, indicating strong ligand-receptor interactions. These observations were consistent with their strong correlations with the reductions in saltiness and bitterness observed in the sensory analysis.

To experimentally validate the binding affinities predicted by molecular docking, readily available reference standards that exhibited relatively high docking affinities were selected for MST analysis with their corresponding taste receptor. As shown in Fig. 9 and Table S7, both rutin and phenylethyl primeveroside were able to bind to TMC4. Rutin exhibited an apparent dissociation constant (Kd) of 20.85 ± 7.21 μM (signal-to-noise ratio, S/N = 15.612), whereas phenylethyl primeveroside showed a higher binding affinity, with an apparent Kd of 0.42 ± 0.11 μM (S/N = 12.887). Both compounds showed favorable binding to TMC4, with phenylethyl primeveroside exhibiting a markedly stronger binding affinity. These results demonstrate direct molecular interactions between key differential metabolites and the taste receptor protein and provide further experimental support for the molecular docking predictions.

Fig. 9.

Microscale thermophoresis (MST) analysis of the interactions between TMC4 and rutin (A) and phenylethyl primeveroside (B).

4. Discussion

In the present study, a systematic component-taste relationship-based approach was employed, revealing that two distinct fermentation strategies mediated by AC and AN resulted in significant differences in both flavor characteristics and metabolite composition of XLT. Comparisons between UF-AC and UF-AN identified 34 and 48 differential metabolites, respectively, which exhibited pronounced up- or downregulation during fermentation and showed significant correlations with the flavor attributes of XLT. Subsequently, integrated correlation and redundancy analyses identified 17 key differential metabolites associated with specific taste attributes. Molecular docking and MST were further conducted to validate and elucidate the potential mechanisms underlying the modulation of different taste characteristics in XLT.

During FAC fermentation, various extracellular enzymes secreted by AC may promote the formation and accumulation of theabrownins, thereby contributing to the characteristic orange-brown and reddish appearance of the fermented tea (Zhang & Zhang, 2024). In contrast, XLT in the FAN group showed a darker, nearly black infusion compared with unfermented XLT, which is consistent with the dark pigmentation commonly observed during the growth of AN (Mirhendi et al., 2016). XLT also exhibited distinct taste changes in response to fermentation by different fungal strains. AC fermentation primarily enhanced the sweetness and bitterness of the infusion while reducing umami and saltiness, whereas AN fermentation significantly increased sourness while decreasing saltiness and bitterness. Untargeted metabolomics analysis and dynamic fermentation experiments revealed extensive metabolic reprogramming in XLT after fermentation, mainly involving flavonoids, triterpenoid saponins, amino acids, glycosides, and nucleosides. Notably, no significant differences were observed between the unfermented groups (UF0, UF5, UF10, and UF15) and either FAC0 or FAN0 groups, confirming that the changes in taste, color, and metabolites in XLT were mainly driven by microbial activity rather than aseptic incubation conditions (Zeng et al., 2024). Although both fermentation systems showed an overall decreasing trend in metabolite abundance, FAC exhibited a more balanced pattern of upregulation and downregulation, whereas FAN was dominated by metabolite downregulation. Previous studies have shown that Aspergillus species secrete a complex array of hydrolytic and oxidative enzymes, including tannases, glycoside hydrolases, polyphenol oxidases, catechol oxidases, peroxidases, laccases, and other carbohydrate-active enzymes, which can cooperatively drive the hydrolysis, oxidation, polymerization, and subsequent degradation of tea-derived metabolites (Gamarra et al., 2010; Li, Song, et al., 2025). The substantial accumulation of gallic acid in FAC is highly consistent with tannase-mediated degalloylation. Tannase catalyzes the hydrolysis of galloyl ester bonds in galloylated polyphenols, releasing gallic acid together with corresponding non-galloylated derivatives (Fan et al., 2025; Lee et al., 2015). Notably, pronounced deglycosylation of glycosidic compounds was observed during AN fermentation, leading to significant decreases in flavonoid glycosides, triterpenoid saponins, and other glycosides. This phenomenon can be attributed to glycosidases secreted by AN, which cleave glycosidic bonds, thereby promoting the accumulation of soluble sugars and accelerating carbohydrate degradation (Kumar et al., 2020; Zeng et al., 2024). In contrast, triterpenoid saponins and flavonoids in AC generally exhibited relative accumulation after fermentation, a phenomenon also observed in the natural fermentation of Fu brick tea (Li et al., 2021; Zhang & Zhang, 2024). The markedly different variation patterns of these compounds between the two fermentation systems reflect fundamental differences between AC and AN in substrate utilization and secondary metabolic pathways (Liao et al., 2023). The dynamic fermentation results showed that the biotransformation processes mediated by both fungal strains had largely stabilized by day 10 of fermentation, after which the levels of most differential metabolites no longer changed significantly. Interestingly, under the experimental conditions used in this study, AN exhibited a faster growth, proliferation, and metabolic transformation rate than AC. Most AN-induced metabolite changes occurred during days 0–5 of fermentation, whereas AC-mediated metabolite transformations were mainly concentrated during days 5–10. Taken together, the metabolomic alterations induced by the two different fermentation strains ultimately led to the distinct sensory characteristics observed in fermented XLT.

Unlike industrial flavor modulation strategies that rely on masking or blocking undesirable tastes, flavor formation in natural fermented foods is predominantly driven by the accumulation and interaction of taste-active compounds. These compounds contribute to taste enhancement through additive or synergistic effects, which have been widely reported for amino acids, peptides, organic acids, and polyphenol-derived metabolites (Zhang et al., 2023; Zhao et al., 2016). Correlation and RDA analyses indicated that the decreases in umami and saltiness observed after AC fermentation were mainly mediated by amino acids, glycosides, and phenolic acids. The overall reduction of amino acids during fermentation can be attributed to deamination, decarboxylation, oxidation reactions, and their conversion into volatile organic compounds via Maillard reactions, as well as their utilization as carbon and nitrogen sources (Liao et al., 2025). The roles of amino acids in umami perception (Liu, Han, et al., 2024; Yu et al., 2026) and their synergistic modulation of saltiness in combination with peptides have been well documented (Li, Yin, et al., 2025; Liu et al., 2026). In contrast, triterpenoid saponins in AC were identified as major contributors to the increases in bitterness and sweetness. Their molecular structures, characterized by abundant glycosidic moieties, are associated with sweetness enhancement, and numerous studies have reported that triterpenoid saponins inherently possess sweet or bitter taste properties (Guo et al., 2024; Karolkowski et al., 2023; Yang et al., 2023; Zhang et al., 2017). The increase in sourness induced by AN fermentation was closely associated with flavonoids, phenolic acids, and the AN-specific secondary metabolite malformin (Gangaraj et al., 2024), highlighting the important role of fungal secondary metabolism in taste modulation. The pronounced decreases in saltiness and bitterness observed after AN fermentation were strongly linked to flavonoids. Enzymes secreted by AN participate in key metabolic pathways, including flavonoid biosynthesis and flavonoid and flavanol metabolism, thereby regulating flavonoid composition (Zeng et al., 2023). As major contributors to the yellow-green color of tea infusions, flavonoid glycosides exhibit extremely low taste thresholds, imparting mild astringency and enhancing caffeine-related bitterness (Ke et al., 2025). Sensory supplementation studies have demonstrated that even trace amounts of rutin can significantly increase bitterness (Li et al., 2019). Molecular docking analysis was subsequently employed to further elucidate the relationships between metabolites and taste modulation. All ligands exhibited docking scores lower than −5.0, with most showing values below −7.0, further confirming the strong binding affinities between key metabolites and their corresponding taste receptors and providing molecular-level support for their potential roles in taste regulation (Lao et al., 2024; Ma et al., 2025). Multiple non-covalent interactions, including hydrogen bonding and hydrophobic interactions, were involved in ligand-receptor binding. Interestingly, previously reported umami-related amino acids, including L-phenylalanine and L-tryptophan, exhibited relatively low docking scores, whereas the cyclic peptide malformin showed unexpectedly high binding affinity to the sour taste receptor. In the fields of molecular docking and molecular dynamics, peptides composed of multiple amino acid residues are generally considered to exhibit higher affinity and binding energy toward protein receptors (Wu et al., 2026; Zhao et al., 2026). OTOP1 functions as a proton-selective channel involved in sour taste perception; under acidic conditions, extracellular H⁺ permeates through OTOP1 into taste cells and initiates sour taste signaling (Teng et al., 2019). Previous studies have further suggested that some nonacidic organic molecules can interact with regulatory regions of OTOP1, thereby increasing its sensitivity to H⁺ and enhancing proton currents (Kong et al., 2025). Notably, the pH of the tea infusion did not decrease after AN fermentation (Supplementary Table S5), suggesting that the enhanced sourness was not simply attributable to an increase in bulk proton concentration. We therefore speculate that the stable sulfur-containing cyclic structure of malformin may facilitate its interaction with regulatory regions of OTOP1, potentially increasing channel responsiveness under the same acidic conditions and thereby promoting H⁺ influx and sour taste signaling. This may also help explain its favorable predicted binding affinity toward OTOP1. However, this proposed mechanism remains speculative and should be further validated by OTOP1-based electrophysiological or cell-functional assays.

The MST results provided further experimental evidence for the molecular interactions predicted by molecular docking. TMC4 has been reported as a transmembrane protein associated with salt taste transduction and is highly expressed in posterior taste buds (Kasahara et al., 2021). In the present study, phenylethyl primeveroside exhibited submicromolar binding affinity toward TMC4 (Kd = 0.42 ± 0.11 μM), which is comparable to the affinity range recently reported for several food-derived salty peptides interacting with TMC4 (approximately 0.18–3.43 μM) (Song et al., 2025). In comparison, rutin showed a relatively weaker affinity for TMC4 (Kd = 20.85 ± 7.21 μM), although previous studies have also reported favorable molecular docking interactions between rutin and TMC4 (Cai et al., 2026). It should be noted that MST primarily quantifies molecular binding affinity rather than ion-channel functional activity. Therefore, further electrophysiological or cell-based functional assays are still required to determine whether these ligand–TMC4 interactions can alter TMC4-mediated ion-channel responses and thereby influence salt taste perception. Taken together, the consistency among the taste changes observed by the electronic tongue, metabolite–taste correlations, molecular docking predictions, and MST validation provides a plausible molecular-level explanation for the taste modulation observed in this study.

In the past, Xanthoceras sorbifolium has mainly been utilized for landscaping and seed oil production, while its leaves have received limited attention and are generally regarded as low-value feed resources (Jia et al., 2025). Early statistics indicate that the cultivation area of Xanthoceras sorbifolium in several provinces of northern China exceeds 2 × 10⁴ acres, generating approximately 1 million tons of discarded leaves annually, resulting in substantial resource waste (Miao et al., 2023). Therefore, there remains considerable potential for the valorization and utilization of Xanthoceras sorbifolium leaves. As an emerging herbal tea, XLT possesses various health benefits; however, its consumption remains limited due to its less acceptable taste compared with commercially available teas. Therefore, improving its sensory quality is essential for enhancing consumer acceptance and reducing the waste of Xanthoceras sorbifolium leaves. However, this study still has several limitations. The present work mainly focused on changes in taste, specifically the relationships between electronic tongue responses and non-volatile metabolites, whereas overall flavor characteristics, particularly those related to volatile metabolites and electronic nose responses, were not extensively investigated. Moreover, the relative changes in taste before and after fermentation were evaluated using electronic tongue data without human sensory evaluation. Although the electronic tongue provides objective and reproducible measurements of taste-related changes, these instrumental responses cannot fully represent human sensory perception. Future studies should therefore incorporate trained sensory panels and consumer acceptance tests to validate the instrumental findings and provide a more comprehensive assessment of changes in the sensory quality of XLT during fermentation. In addition, although associations between differential metabolites and sensory attributes were identified, the causal contribution of individual metabolites to specific taste or flavor characteristics still requires further validation through targeted sensory and functional experiments. Overall, this study provides new insights into the mechanisms underlying flavor evolution during XLT fermentation and advances the understanding of flavor-related metabolic dynamics. More importantly, this work provides a scientific framework for improving the sensory quality of XLT and supports the sustainable valorization of Xanthoceras sorbifolium leaf resources.

5. Conclusion

In this study, we applied an integrated strategy combining electronic tongue analysis, untargeted metabolomics, time-series analysis, molecular docking, and MST to decode the flavor modulation of XLT during microbial fermentation. Unlike unfermented tea, the two fungal strains induced divergent sensory trajectories: AC fermentation significantly enhanced bitterness and sweetness while reducing umami, whereas AN fermentation was characterized by a distinctive increase in sourness. Metabolomic profiling revealed extensive metabolic reprogramming, identifying 34 and 48 key differential metabolites in the AC and AN groups, respectively. Correlation and redundancy analyses further pinpointed the chemical basis of these changes: the increased bitterness and sweetness in the AC group were primarily driven by the accumulation of triterpenoid saponins, while the reduction in umami was linked to the degradation of amino acids. In contrast, the sourness in the AN group was strongly associated with specific flavonoids and secondary metabolites like malformin. Moreover, molecular docking confirmed that these key metabolites exhibited high binding affinities (all scores < −5.0) with their corresponding taste receptors, providing a solid molecular explanation for the observed sensory shifts. MST analysis further provided experimental evidence for the docking-predicted interactions between selected taste-related metabolites and TMC4. Collectively, these findings not only clarify the metabolic mechanisms underlying flavor formation but also suggest that targeted fermentation can effectively modulate the unfavorable taste profile of XLT. Future work should focus on tracking the kinetic evolution of these taste-active compounds and exploring mixed-culture strategies to further optimize palatability. Additionally, consumer acceptance trials and process scale-up assessments will be crucial for the industrial application of this functional herbal tea.

CRediT authorship contribution statement

Yingxue Guo: Writing – review & editing, Writing – original draft, Software, Methodology, Formal analysis, Conceptualization. Zhaoxiang Zeng: Writing – review & editing, Writing – original draft, Visualization, Software, Conceptualization. Keying Zhao: Validation, Supervision, Data curation. Haihuan Wei: Methodology, Investigation. Xueyan Zhao: Supervision, Investigation. Yi-hung Chen: Validation, Data curation. Cong Wang: Software, Resources. Meiping Lv: Visualization, Investigation. Chengwu Song: Writing – original draft, Resources, Project administration, Conceptualization. Jinsong Shu: Writing – review & editing, Supervision, Methodology, Conceptualization. Shuna Jin: Writing – review & editing, Writing – original draft, Project administration, Investigation, Funding acquisition, Conceptualization.

Declaration of competing interest

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

Appendix A. Supplementary data

Supplementary material. Figure S1. The score plot of the OPLS-DA model and permutation test plot with 999 replications. Figure S2. Tea infusion color at different time points during dynamic fermentation. Figure S3. Three-dimensional structures of five taste receptors. Figure S4. Ramachandran diagram of predicted structures of taste protein receptors. Table S1. Identification or characterization results of taste-related differential metabolites in UPLC-QTOF-MS/MS analysis. Table S2. Relative quantitative analysis of taste-related differential metabolites in XLT during various processes. Table S3. Sensory analysis was performed to evaluate the taste differences of XLT fermented with different microbial strains. Table S4. ERRAT score results of taste protein receptors. Table S5. pH values of tea infusion. Table S6. Summary of docking scores and amino acid residues involved in hydrogen-bonding and hydrophobic interactions between taste receptors and ligands. Table S7. Binding affinities determined by microscale thermophoresis (MST).

Data availability

Data will be made available on request.