Candidate compounds associated with sourness and astringency in Corni Fructus: an integrated analysis of chemical profiles, sensory evaluation, and molecular docking

Abstract

Corni Fructus (CF), a medicinal and edible herb, is characterized by sour and astringent properties in both taste and the Traditional Chinese Medicine Five-Flavor Theory. However, the chemical basis of these sensory attributes remains unclear. This study integrated UPLC–MS/MS profiling, electronic tongue analysis, human sensory evaluation, and molecular docking to identify sour/astringent components and explore their sensory relevance. A total of 82 features were characterized, including 52 organic acids, 28 tannin-related features, and 2 unclassified features. Sensory and E-tongue analyses revealed region-dependent differences in sourness, astringency, and bitterness. Correlation analysis and PLSR suggested malic acid and related derivatives may contribute to sourness, whereas galloylated tannins, including 3-O-galloyl-d-glucose, were associated with astringency- and bitterness-related responses. Molecular docking provided preliminary support for interactions between compounds and taste-related protein models. These findings provide candidate markers for flavor-oriented quality evaluation of CF and offer a foundation for sensory validation and medicinal-edible product development.

Graphical abstract

Highlights

• •Integrated profiling screened taste-associated markers in Corni Fructus.
• •A total of 82 features were identified, including 52 organic acids, 28 tannin-related features, and 2 unclassified features.
• •Malic acid derivatives were candidate markers associated with sourness.
• •Galloylated tannins were associated with astringency- and bitterness-related responses.
• •Molecular docking suggested potential compound–protein interactions.

1. Introduction

Corni Fructus (CF, Shanzhuyu in Chinese), the dried ripe pulp of Cornus officinalis Sieb. et Zucc. (Cornaceae), has long been valued as a medicinal and edible resource (Yang et al., 2023). In Traditional Chinese Medicine (TCM), it is used to tonify the liver and kidneys and to astringe essence, serving as a principal herb in numerous classic formulas such as Liuwei Dihuang Wan (Sun, Xue, et al., 2023). Beyond its medicinal use, CF has also been consumed in central China for centuries in soups, main dishes, fruit wine, juices, and preserved products (Parveez Zia & Alibas, 2021). With its official approval in 2023 as a dual-use medicinal and edible material, CF has attracted increasing attention for its nutritional and functional value as well as its potential in food development and quality evaluation (Lu et al., 2022). Therefore, clarifying the chemical basis of its characteristic taste is important for the standardized utilization of CF in medicinal-edible products.

Among its sensory characteristics, CF is distinguished by pronounced sourness and astringency. Compared with the taste attributes defined in modern sensory science, the five flavors described in the TCM Five-Flavor Theory (TCMFFT) constitute a broader concept linking taste perceptions to therapeutic functions. In the TCMFFT, sourness is associated with convergence and the securing of essence, whereas astringency is linked to consolidation and restraint of fluid loss (Chen et al., 2024; Qiao et al., 2022; Wang et al., 2023; Wang, Wei, et al., 2024). From a modern food chemistry perspective, however, these sensory properties should be understood in terms of measurable taste-active compounds and their contributions to flavor quality. Although CF is widely recognized as sour and astringent, the chemical basis underlying these oral sensations remains insufficiently clarified.

In foods, sourness primarily arises from organic acids such as malic, tartaric, and citric acids, which influence proton activity in the oral cavity and thereby contribute to perceived acidity and flavor brightness (Kang et al., 2023; Zhu et al., 2024). Astringency, in contrast, is mainly attributed to polyphenolic compounds, especially gallic acid, its derivatives, and hydrolyzable tannins, which interact with salivary and mucosal proteins to produce a drying and puckering sensation (Chen et al., 2025; Ye et al., 2022). However, it remains unclear which specific compounds are most closely associated with the sour and astringent characteristics of CF. Identifying such candidate compounds would be valuable for flavor-oriented quality assessment and future product development.

An integrated analytical strategy combining metabolite profiling, instrumental taste evaluation, and computational interaction analysis provides a useful framework for investigating flavor-related compounds in complex plant-derived matrices (Hou et al., 2024; Kang et al., 2025). UPLC–MS/MS facilitates rapid qualitative and quantitative characterization of major organic acids and tannin-related constituents (Li et al., 2025; Wu et al., 2025). Electronic tongue (E-tongue) technology offers an objective evaluation of sourness and astringency, supporting the development of standardized gustatory fingerprint profiles (Marx et al., 2017; Sun, Zhang, et al., 2023). Concurrently, molecular docking enables the simulation of interactions between representative compounds and taste-related targets, providing mechanistic insights into flavor perception (Xiao et al., 2024). Notably, OTOP1 and PKD2L1 are closely associated with sour taste transduction, whereas β-casein is widely used as a model protein for studying polyphenol–protein interactions relevant to astringency (Liman & Kinnamon, 2021; Zheng et al., 2025).

Previous studies on CF have mainly focused on its phytochemical composition, pharmacological activity, or general quality evaluation, whereas the integrated identification of compounds associated with sourness and astringency has remained limited. Therefore, the study applies a multidimensional approach integrating UPLC–MS/MS profiling, electronic tongue analysis, human sensory evaluation, and molecular docking to investigate compounds associated with the sour and astringent characteristics of CF from different producing regions. This work aims to provide a food-chemistry-oriented basis for understanding the characteristic flavor of CF and to offer candidate markers for future quality evaluation and mechanistic validation.

2. Materials and methods

2.1. Sources of Corni Fructus samples

A total of 27 batches of CF were collected from four major producing regions in China: Xixia (Henan), Foping (Shaanxi), Chun'an (Zhejiang), and Yangcheng (Shanxi). Samples were grouped as CFA (Henan, n = 10), CFB (Shaanxi, n = 7), CFC (Zhejiang, n = 6), and CFD (Shanxi, n = 4). All samples were collected during the harvest season from September to October 2025 and were qualified commercial materials consisting of fully mature dried fruit pulp. The samples were processed by blanching in boiling water, removing the seeds, and air-drying.

2.2. Chemicals and reagents

HPLC-grade methanol, acetonitrile, and formic acid were purchased from Thermo Fisher Scientific (Waltham, MA, USA). Ultrapure water was obtained commercially, and analytical-grade methanol (≥99.5%) was supplied from Tianjin Zhiyuan Chemical Reagent Co., Ltd. (Tianjin, China). Reference standards (≥98%) of malic acid, fumaric acid, quinic acid, tartaric acid, D-glucuronic acid, citric acid, gallic acid, caftaric acid, caffeic acid, shikimic acid, and p-coumaric acid were purchased from Chengdu Zhibiao Chemical & Pure Biotechnology Co., Ltd. (Chengdu, China). Lactic acid (≥98%) was purchased from Shanghai Yuanye Bio-Technology Co., Ltd. (Shanghai, China), and sodium hydroxide (≥96%) from Tianjin Yongda Chemical Reagent Co., Ltd. (Tianjin, China).

2.3. UHPLC–MS/MS conditions

Chromatographic separation was performed on a Thermo Scientific Vanquish™ UHPLC system equipped with an ACQUITY™ Premier HSS T3 column (2.1 × 100 mm, 1.8 μm; VanGuard™ FIT). The mobile phase consisted of acetonitrile (A) and 0.1% (v/v) formic acid in water (B) at a flow rate of 0.3 mL/min. The column temperature was set at 30 °C and the injection volume was 2 μL. The gradient program was as follows: 0–2 min, 100% B; 2–5 min, 100–95% B; 5–15 min, 95–87% B; 15–20 min, 87–70% B; 20–30 min, 70–40% B; 30–35 min, 40–20% B; 35–40 min, 20% B.

Mass spectrometric analysis was conducted using an Orbitrap Exploris™ 120 mass spectrometer equipped with heated electrospray ionization (H-ESI) in negative ion mode over an m/z range of 70–1000. Full-scan and data-dependent MS² resolutions were set at 120,000 and 30,000, respectively. Source parameters were as follows: spray voltage 2.8 kV, capillary temperature 320 °C, auxiliary gas temperature 350 °C, sheath gas 45 arb, auxiliary gas 15 arb, and normalized collision energies of 15, 30, and 45 eV. Compounds were identified by comparing retention times, accurate masses, and MS/MS spectra with reference standards. Related compounds were annotated based on characteristic fragmentation patterns, and unknowns were tentatively assigned using Compound Discoverer 3.2, online databases, and literature.

2.4. Quantitative analysis of 12 reference-confirmed compounds

CF powder (1.0 g) was extracted twice with 70% methanol (50 mL each) by ultrasonication (250 W, 30 min). The combined extracts were diluted, centrifuged 13,000 rpm (16,200×g) for 10 min, and the supernatant was analyzed by UHPLC–MS/MS. Reference standards of malic acid, fumaric acid, quinic acid, tartaric acid, lactic acid, D-glucuronic acid, citric acid, gallic acid, caftaric acid, caffeic acid, shikimic acid, and p-coumaric acid were prepared in methanol (1 mg/mL) and serially diluted to construct external calibration curves for quantification. Prior to selecting 70% methanol as the extraction solvent, we compared the overall total ion chromatogram (TIC) profiles obtained using water and 30%, 50%, 70%, and 100% methanol and evaluated the extracted-ion chromatographic peak areas and trends of eight representative analytes, including tartaric acid, quinic acid, malic acid, fumaric acid, gallic acid, caftaric acid, 2,3-di-O-galloyl-d-glucose, and erythrulose gallate. Considering chromatographic response, peak shape, and compound coverage, 70% methanol showed favorable overall analytical performance (Supplementary Table S13 and Figs. S4–S5).

2.5. Determination of total organic acid content

Total organic acid content was determined by potentiometric NaOH titration as previously reported (Chen et al., 2011; Wang, Zhou, et al., 2024). CF powder (5.0 g) was ultrasonically extracted with distilled water (50 mL, 30 min), filtered, and an aliquot of the extract was titrated with 0.1 mol/L NaOH at 1–2 mL/min under continuous pH monitoring. The titration endpoint was determined using the second-derivative method, and total organic acid content was calculated accordingly.

2.6. Sensory evaluation

The sensory protocol was approved by the Ethics Review Committee of Henan University of Chinese Medicine (No. HNSZYYky2025081001), and written informed consent was obtained from all participants. Twenty healthy panelists aged 21–35 years with normal taste function were recruited. After training with sourness, bitterness, and astringency reference solutions, panelists evaluated room-temperature CF infusions prepared as for E-tongue analysis. Samples were coded with random three-digit numbers and presented blindly in randomized order. Panelists rinsed with purified water and rested for 2–3 min between samples. Sourness, astringency, bitterness, and taste harmony were scored on a 10-point scale in triplicate. Data were analyzed by one-way ANOVA or mixed-effects models, with p < 0.05 considered significant.

2.7. Electronic tongue analysis

Taste profiles of CF were evaluated using an SA402B electronic tongue (Insent Co., Ltd., Japan) equipped with five sensors (COO, AE1, CAO, CTO, and AAE) corresponding to bitterness, astringency, sourness, saltiness, and umami. Measurements were based on changes in membrane potential induced by interactions between taste-active compounds and artificial lipid membranes. Sample preparation followed a reported method (Sun, Fan, et al., 2023). Briefly, CF powder (4.0 g) was extracted with purified water (100 mL) by soaking (20 min) and ultrasonication (50 °C, 30 min), cooled, filtered, and analyzed. Sensor conditioning and measurement procedures were conducted according to the manufacturer's protocol. Each sample was analyzed in quadruplicate; the first measurement was discarded, and the mean of the remaining three replicates was used for data analysis (Liu et al., 2025).

2.8. Molecular docking

Molecular docking was performed to evaluate interactions between taste-related receptors and CF constituents. Sourness-related protein models, including OTOP1 (UniProt ID: Q7RTM1), PKD1L2 (UniProt ID: Q7Z442), PKD1L3 (UniProt ID: Q7Z443), and PKD2L1 (UniProt ID: Q9P0L9), together with β-casein, used as an astringency-related protein model, were selected as targets based on previous studies (Kaplan et al., 2025; Turner & Liman, 2022; Wilson et al., 2019). The 3D structures of 82 compounds were generated using Chem3D 2020, saved in MOL2 format, and subsequently energy-minimized. Structural data of the four sour taste receptors were retrieved from the UniProt database (https://www.uniprot.org/). For β-casein, homology modeling was performed using the I-TASSER platform (https://zhanggroup.org/I-TASSER-MTD/), based on the threading method and targeting human saliva-derived β-casein (Zhou et al., 2019; Zhou et al., 2022). Ligands and receptors were prepared and optimized using Discovery Studio 2019. Docking was conducted at predefined binding sites, and docking poses were ranked by binding energy. Ligands with binding energies below −40 kJ/mol were considered potential binders and further analyzed to identify favorable receptor–ligand interactions.

2.9. Cross-matrix comparison of chemical profiles

To compare chemical profile consistency between the 70% methanol and analytical aqueous extraction systems, 70% methanol and aqueous test solutions were prepared in parallel from the same 27 batches and analyzed by UHPLC–MS/MS. Pearson correlation analysis was performed on the peak areas of 82 matched analytes (52 organic acids, 28 tannins, and 2 unclassified features), followed by false discovery rate (FDR) correction. The statistical significance of the overall correlation was further evaluated using a permutation test with 1000 iterations.

2.10. Statistical data analysis

For statistical and multivariate analyses of chemical components, Principal Component Analysis (PCA) and Partial Least Squares Discriminant Analysis (PLS-DA) were performed using MetaboAnalyst (https://www.metaboanalyst.ca/). One-way Analysis of Variance (ANOVA) and significance testing were conducted with SPSS Statistics 27.0 (IBM, Chicago, USA). Heatmaps and correlation matrices were generated via OmicStudio (https://www.omicstudio.cn/home) and Chiplot (https://www.chiplot.online/), respectively. Box plots and radar charts were created using Origin 2025 (OriginLab, MA, USA). Additionally, PLSR analysis was repeated with the SPSSAU online platform to assess multicollinearity between E-tongue data and chemical components. Molecular docking and visualization were carried out using Discovery Studio 2019 (Dassault Systèmes, Paris, France).

3. Results and discussion

3.1. Annotation and characterization of organic acid- and tannin-related constituents

Diagnostic ions were first identified by summarizing characteristic fragments of known organic acids and tannins. Using full-scan and MS/MS data from total ion chromatograms (TIC), together with literature information, 82 constituents were tentatively identified, including 52 organic acids, 28 tannins, and 2 unclassified features (Table 1). As shown in Fig. 1, the identified constituents exhibited distinct distribution patterns among the four producing regions. Hierarchical clustering analysis revealed two major metabolite clusters, including a tannin-enriched cluster with relatively higher levels in the CFC group and an organic acid-dominated cluster with higher relative abundance in the CFA and CFD groups. Representative qualitative identification results are summarized below.

Table 1.

Targeted Metabolomics Identification of Chemical Components in CF.

No.	Identification	RT (min)	Formula	Ionization mode	MS1	Error (ppm)	MS2
1⁎	Lactic acid	0.99	C3H6O3	[M-H]−	89.02473	9.66	71.01415, 59.01408, 43.01906
2⁎	D-Glucuronic acid	0.99	C6H10O7	[M-H]−	193.03555	3.73	175.02591, 157.01505, 131.03580, 113.02470, 103.00392, 89.02463, 85.02969, 72.99332, 71.01405, 59.01403
3⁎⁎	Glutaric acid	0.99	C5H8O4	[M-H]−	131.03542	7.52	113.02471, 85.02966
4⁎⁎	Gluconic acid	1.01	C6H12O7	[M-H]−	195.05167	6.10	177.04124, 159.03050, 129.01985, 99.00912, 75.00909, 59.01409
5⁎⁎⁎	Tartaric acid glucoside	1.08	C10H16O11	[M-H]−	311.06244	3.21	191.05687, 179.05681, 161.04602, 149.00987, 143.03543, 130.99922, 103.001414
6⁎⁎	Isomers of citric acid	1.08	C6H8O7	[M-H]−	191.02031	5.92	173.01009, 129.02022, 85.02983
7⁎	Tartaric acid	1.09	C4H6O6	[M-H]−	149.00978	7.82	130.99924, 103.00410, 87.00912, 72.99345
8⁎⁎⁎	Butanedioic acid, 2,3-bis[(2-carboxyacetyl)oxy]-, (2R, 3R)- (ACI)	1.10	C10H10O12	[M-H]−	321.00850	−2.82	191.05681, 149.00977, 130.99915, 103.00414, 87.00912
9⁎⁎⁎	Quinic acid glucoside	1.14	C13H24O12	[M-H]−	371.12030	3.62	191.05684, 179.05663, 173.04633, 143.03558, 127.04065, 113.02488, 101.02479
10⁎⁎⁎	Derivatives of tartaric acid	1.14	C13H16O12	[M-H]−	363.05399	−6.51	243.01585, 191.05650, 149.00946, 130.99918, 103.00368, 87.00886
11⁎	Quinic acid	1.16	C7H12O6	[M-H]−	191.05682	6.57	173.04608, 127.04075, 111.04556, 93.03496, 87.00912, 85.02982
12⁎⁎	Quinic acid dimer	1.16	C14H24O12	[M-H]−	383.11978	2.15	191.05687, 173.04626, 127.04065, 85.02989
13⁎⁎⁎	Derivatives of quinic acid	1.17	C16H22O12	[M-H]−	405.10327	−0.09	273.05994, 213.03889, 191.05675, 173.04630, 127.04063, 111.04582, 85.02989
14⁎⁎⁎	Derivatives of quinic acid	1.21	C10H12O8	[M-H]−	259.04422	−4.54	191.05679, 173.04605, 127.04077, 85.02978
15⁎⁎⁎	Derivatives of quinic acid	1.36	C19H34O17	[M-H]−	533.17371	3.62	191.05690, 173.04645, 127.04076, 85.02989
16⁎⁎	2-Methylcitric acid	1.38	C7H10O7	[M-H]−	205.03607	6.05	187.02562, 169.01477, 161.04617, 143.03560, 131.03549, 125.02493, 113.02487
17⁎⁎⁎	Derivatives of (S)-malic acid 1-O-D-glucopyranoside	1.41	C22H38O21	[M-H]−	637.18408	2.10	295.06754, 179.05626, 161.04585, 143.03514, 133.01456, 119.03520, 113.02464, 101.02492, 89.02453
18⁎⁎	Trans-aconitic acid malic acid glucoside	1.43	C16H20O15	[M-H]−	451.07110	−2.88	295.06848, 271.00797, 227.01826, 197.00739, 179.05667, 161.04626, 154.99677, 143.03548, 133.01477, 115.00414, 89.02471, 71.01412
19⁎⁎	Tartaric acid aconitic acid monoester	1.47	C10H10O11	[M-H]−	305.01303	−4.79	149.00978, 133.01483, 115.00418, 103.0041
20⁎⁎⁎	Malic acid aconitic acid monoester	1.47	C10H10O10	[M-H]−	289.01852	−3.65	154.99654, 133.01483, 115.00420
21⁎	Malic acid	1.48	C4H6O5	[M-H]−	133.01488	8.87	115.00420, 89.02479, 71.01416
22⁎	Fumaric acid	1.48	C4H4O4	[M-H]−	115.00432	10.30	71.01421
23⁎⁎⁎	Derivatives of malic acid	1.50	C18H18O15	[M-H]−	473.05310	−7.72	353.01187, 311.00104, 221.00508, 154.99673, 133.01482
24⁎⁎⁎	Derivatives of malic acid	1.51	C22H24O20	[M-H]−	607.07530	−4.90	383.02170, 353.01151, 311.00140, 221.00499, 154.99661, 133.01472, 115.00410
25⁎⁎⁎	Derivatives of malic acid	1.51	C12H8O10	[M-H]−	311.00027	−11.75	133.01489, 115.00424
26⁎⁎⁎	Malic acid aconitic acid diester	1.51	C16H14O15	[M-H]−	445.02180	−8.20	289.01868, 154.99672, 133.01474, 115.00410, 71.01408
27⁎⁎	Isomers of citric acid	1.54	C6H8O7	[M-H]−	191.02036	6.18	173.00940, 111.00896, 85.02962
28⁎	Shikimic acid	1.63	C7H10O5	[M-H]−	173.04610	6.36	155.03552, 137.02493, 111.04559, 93.03489, 83.05052, 73.02978
29⁎⁎⁎	Quinic acid glucoside	1.72	C13H22O11	[M-H]−	353.11014	4.96	295.06796, 257.02853, 227.01758, 197.00717, 191.05655, 179.05652, 173.00954, 101.02473
30⁎⁎⁎	Derivatives of (S)-malic acid 1-O-D-glucopyranoside	1.73	C20H32O20	[M-H]−	591.14252	2.78	295.06747, 235.04648, 205.03635, 191.05629, 179.05634, 161.04572, 143.03505, 133.01436, 119.03520, 101.02463, 89.02449
31⁎⁎⁎	(S)-malic acid 1-O-D-glucopyranoside	1.74	C10H16O10	[M-H]−	295.06784	4.46	179.05682, 161.04626, 133.01494, 119.03553, 115.00426, 113.02495, 101.02486, 89.02483, 71.01423, 59.01414
32⁎⁎⁎	Derivatives of malic acid	1.75	C13H16O12	[M-H]−	363.05585	−1.39	317.05060, 295.06857, 257.02887, 227.01830, 197.00757, 179.05696, 154.99690, 133.01497
33⁎⁎	2-furoic acid	2.79	C5H4O3	[M-H]−	111.00899	6.94	67.01912
34⁎	Citric acid	2.84	C6H8O7	[M-H]−	191.02042	6.49	173.00983, 129.01945, 111.00885, 87.00883, 85.02959
35⁎⁎	Isomers of 7-O-galloyl-D-sedoheptulose	3.00	C14H18O11	[M-H]−	361.07809	2.77	301.05536, 271.04663, 241.03577, 211.02501, 169.01453, 125.02477
36⁎⁎	Isomers of 7-O-galloyl-D-sedoheptulose	5.04	C14H18O11	[M-H]−	361.07809	2.77	301.05737, 271.04648, 241.03590, 211.02504, 169.01450, 125.02464
37⁎⁎	Isomers of 7-O-galloyl-D-sedoheptulose	5.81	C14H18O11	[M-H]−	361.07813	2.88	271.04648, 241.03558, 211.02544, 169.01447, 125.02475
38⁎	Gallic acid	6.20	C7H6O5	[M-H]−	169.01480	6.51	125.02498
39⁎⁎	Isomers of 4-O-galloylquinic acid	6.23	C14H16O10	[M-H]−	343.06778	3.66	271.04712, 211.02599, 191.05640, 169.01524
40⁎⁎	Isomers of 3-O-galloyl-d-glucose	6.43	C13H16O10	[M-H]−	331.06778	3.79	271.04660, 241.03607, 211.02533, 169.01454, 151.00404, 125.02474
41⁎⁎	Isomers of 7-O-galloyl-D-sedoheptulose	6.80	C14H18O11	[M-H]−	361.07809	2.77	271.04657, 211.02527, 169.01463
42⁎⁎	Isomers of 3-O-galloyl-d-glucose	6.95	C13H16O10	[M-H]−	331.06784	3.97	271.04651, 211.02536, 169.01448, 125.02472
43⁎⁎	Isomers of 4-O-galloylquinic acid	7.40	C14H16O10	[M-H]−	343.06778	3.66	271.04712, 211.02599, 191.05640, 169.01524
44⁎⁎	2,3-di-O-galloyl-d-glucose	7.46	C20H20O14	[M-H]−	483.07843	1.96	439.08786, 331.06772, 313.05728, 287.07800, 271.04547, 169.01463, 125.02480
45⁎⁎	Erythrulose gallate	7.51	C11H12O8	[M-H]−	271.04675	5.00	211.02547, 169.01488, 125.02498
46⁎⁎	Isomers of 7-O-galloyl-D-sedoheptulose	7.51	C14H18O11	[M-H]−	361.07791	2.27	301.05771, 271.04648, 241.03619, 211.02528, 169.01450, 125.02504
47⁎⁎	Isomers of 3-O-galloyl-d-glucose	7.58	C13H16O10	[M-H]−	331.06799	4.43	271.04626, 241.03516, 211.02518, 169.01447, 125.02472
48⁎⁎	Isomers of 4-O-galloylquinic acid	8.23	C14H16O10	[M-H]−	343.06784	3.83	271.04587, 211.02472, 191.05629, 173.04578, 169.01448, 125.02486
49⁎⁎	Isomers of 3-O-galloyl-d-glucose	8.93	C13H16O10	[M-H]−	331.06790	4.15	169.01456, 125.02489
50⁎⁎	2,3-di-O-galloyl-d-glucose	9.27	C20H20O14	[M-H]−	483.07822	1.52	331.06906, 287.07803, 169.01453, 125.02507
51⁎⁎	Isomers of caftaric acid	9.68	C13H12O9	[M-H]−	311.04205	5.59	179.03511, 149.00937, 135.04541, 103.00387, 87.00890
52⁎⁎	Derivatives of erythrulose gallate	10.14	C23H20O15	[M-H]−	535.07214	−0.49	271.04749, 241.03679, 211.02563, 169.01485
53⁎⁎	Isomers of 1-O,7-O-digalloyl-D- sedoheptulose	10.19	C21H22O15	[M-H]−	513.08960	3.02	361.07928, 343.06955, 271.04614, 241.03532, 211.02582, 169.01454, 125.02522
54⁎⁎	Isomers of caffeic acid	10.62	C9H8O4	[M-H]−	179.03522	4.38	135.04550
55⁎	Caftaric acid	10.63	C13H12O9	[M-H]−	311.04187	5.02	179.03525, 149.00937, 135.04539, 103.00401, 87.00891
56⁎⁎⁎	Unknown	10.65	C16H12O11	[M-H]−	379.02899	−3.03	201.01657, 179.03560, 170.99155, 161.02466, 149.00894, 135.04602, 87.00929
57⁎⁎	2,3-di-O-galloyl-d-glucose	11.01	C20H20O14	[M-H]−	483.07910	3.34	439.08951, 331.06839, 313.05759, 287.07806, 271.04681, 169.01466, 125.02482
58⁎⁎	Isomers of 1-O,7-O-digalloyl-D- sedoheptulose	11.11	C21H22O15	[M-H]−	513.08954	2.90	361.07855, 343.06689, 271.04663, 241.03612, 211.02493, 169.01459, 125.02542
59⁎⁎	Derivatives of erythrulose gallate	11.15	C23H20O15	[M-H]−	535.07196	−0.82	445.03925, 271.04626, 241.03700, 211.02519, 169.01567
60⁎⁎	Isomers of 1-O,7-O-digalloyl-D- sedoheptulose	12.15	C21H22O15	[M-H]−	513.08966	3.14	361.07895, 343.06842, 271.04639, 241.03600, 211.02515, 169.01447, 125.02451
61⁎⁎	Isomers of caftaric acid	12.92	C13H12O9	[M-H]−	311.04114	2.67	179.03566, 149.00983, 135.04581, 130.99930, 103.00414, 87.00912
62⁎⁎	Cis-p-coumaroyltartaric acid	13.19	C13H12O8	[M-H]−	295.04608	2.32	163.04027, 158.97873, 130.98419, 119.05045, 87.00890
63⁎⁎	Isomers of p-Coumaric acid	13.20	C9H8O3	[M-H]−	163.04016	3.93	119.05064
64⁎⁎	2,3-di-O-galloyl-d-glucose	13.89	C20H20O14	[M-H]−	483.07840	1.89	439.09003, 331.06812, 313.05710, 287.07785, 271.04626, 169.01466, 125.02474
65⁎	Caffeic acid	13.91	C9H8O4	[M-H]−	179.03563	6.67	135.04579
66⁎⁎	Isomers of caftaric acid	13.92	C13H12O9	[M-H]−	311.04193	5.21	179.03525, 149.00937, 135.04539, 103.00401, 87.00891
67⁎⁎⁎	Unknown	13.92	C16H12O11	[M-H]−	379.02884	−3.43	201.01717, 179.03516, 170.99118, 161.02481, 149.00926, 135.04536, 87.00903
68⁎⁎	Isomers of p-Coumaric acid	13.92	C9H8O3	[M-H]−	163.04051	6.07	119.05054
69⁎⁎	Derivatives of erythrulose gallate	14.50	C23H20O15	[M-H]−	535.07172	−1.27	271.04623, 241.03650, 211.02585, 169.01440
70⁎⁎	Isomers of 1-O,7-O-digalloyl-D- sedoheptulose	14.51	C21H22O15	[M-H]−	513.08948	2.79	361.07944, 343.06836, 271.04645, 241.03560, 211.02525, 169.01451, 125.02510
71⁎⁎	2,3-di-O-galloyl-d-glucose	14.62	C20H20O14	[M-H]−	483.07880	2.72	439.09067, 331.06857, 313.05554, 287.07724, 271.04684, 169.01445, 125.02510
72⁎⁎	3,4-dimethoxyphenol-1-β-D-(6′-O-galloyl)	16.18	C21H24O12	[M-H]−	467.12064	3.61	327.07449, 313.05722, 169.01451, 125.02488
73⁎⁎	Cis-p-coumaroyltartaric acid	16.41	C13H12O8	[M-H]−	295.04608	2.32	163.04027, 158.97873, 130.98419, 119.05045, 87.00890
74⁎⁎	Isomers of p-coumaric acid	16.43	C9H8O3	[M-H]−	163.04016	3.93	119.05064
75⁎⁎	Isomers of p-coumaric acid	16.84	C9H8O3	[M-H]−	163.04031	4.85	119.05051
76⁎⁎	Cis-p-coumaroyltartaric acid	16.86	C13H12O8	[M-H]−	295.04684	4.90	163.04071, 158.97920, 149.00981, 130.99884, 119.05080, 103.00433, 87.00912
77⁎⁎	Malyl syringate	18.31	C13H14O9	[M-H]−	313.05760	5.24	197.04584, 153.05591, 138.03088, 121.03047, 115.00416
78⁎⁎	Derivatives of 1,7-di-O-galloyl-D-sedoheptulose	18.40	C28H26O19	[M-H]−	665.10070	2.54	513.08862, 495.07739, 393.04642, 343.06931, 325.05771, 271.04712, 241.03476, 211.02538, 169.01434
79⁎	p-coumaric acid	18.76	C9H8O3	[M-H]−	163.04021	4.23	119.05045
80⁎⁎	Malyl syringate	19.26	C13H14O9	[M-H]−	313.05710	3.64	197.04581, 153.05600, 121.02980, 115.00410, 71.01408
81⁎⁎	1,2,3,6-tetra-O-galloyl-β-d-glucose	19.89	C34H28O22	[M-H]−	787.10200	3.30	465.06769, 313.05795, 295.04562, 169.01405
82⁎⁎	Azelaic acid	21.44	C9H16O4	[M-H]−	187.09807	5.53	169.08771, 125.09768

^⁎Confirmed by comparison with the reference standards.

^⁎⁎Previously reported in the literature.

^⁎⁎⁎Putatively annotated unknown derivatives.

Fig. 1.

Heatmap of 82 annotated constituents in CF. Each row was standardized by z-score scaling across the 27 samples, representing the relative variation of each compound among samples.

3.1.1. Lactic acid and fumaric acid

Lactic acid (1) was identified based on its deprotonated molecular ion at m/z 89.02473. Its MS/MS spectrum was dominated by a fragment at m/z 71.01415, corresponding to dehydration. Fumaric acid (22) was characterized by [M−H]⁻ at m/z 115.00432 and a characteristic fragment ion at m/z 71.01421, arising from the loss of CO₂.

3.1.2. Tartaric acid and derivatives

Tartaric acid (7) generated a [M−H]⁻ ion at m/z 149.00978, with characteristic fragment ions at m/z 130.99924 (−H₂O), 103.00410, and 87.00912. Based on the presence of tartaric acid-derived diagnostic ions, compounds 5, 8, 10, 19, 62, 73, and 76 were assigned as tartaric acid-related derivatives. In particular, compounds 62, 73, and 76 produced fragment ions characteristic of both tartaric acid and coumaric acid, supporting their tentative assignment as cis-p-coumaroyltartaric acid isomers (Fathallah et al., 2023). Compound 19 was tentatively assigned as tartaric acid aconitic acid monoester (Bhatkar et al., 2023), whereas compound 5 was annotated as a tartaric acid glucoside.

3.1.3. Quinic acid, shikimic acid, and derivatives

Quinic acid (11) was confirmed by reference comparison and exhibited [M−H]⁻ at m/z 191.05682 with sequential fragment ions at m/z 173.04608, 127.04075, and 111.04556. Compounds 9, 12, 13, 14, 15, and 29 showed similar fragmentation behaviors and were therefore assigned as quinic acid derivatives, including glycosides and higher-mass substituted forms. Compound 12 was tentatively identified as a quinic acid dimer (Trujillo-Mayol et al., 2020). In addition, compound 28 was confirmed as shikimic acid by reference comparison.

3.1.4. Malic acid and derivatives

Malic acid (21) showed a [M−H]⁻ ion at m/z 133.01488 (C₄H₅O₅) and fragmented through dehydration and decarboxylation to yield ions at m/z 115.00420, 89.02479, and 71.01416. Compounds 17, 18, 30, and 31 were assigned as malic acid glycosides or oligomeric derivatives, all displaying the characteristic malic acid fragments. Among them, compound 31 was assigned as a malic acid glucoside, and compound 18 was identified as a trans-aconitic acid malic acid glucoside (Rotzoll et al., 2005). These results indicate that malic acid-related structures constitute an important subgroup of the organic acids in CF.

3.1.5. Citric acid and derivatives

Citric acid (34) was detected as a [M−H]⁻ ion at m/z 191.02042 (C₆H₇O₇) and produced fragments resulting from sequential dehydration, decarboxylation, and CO loss (Hurtado-Fernández et al., 2011). Compounds 6 and 27 showed highly similar fragmentation patterns and were assigned as citric acid isomers. Compound 16, showing [M−H]⁻ at m/z 205.03607, was tentatively identified as 2-methylcitric acid or its isomer according to its characteristic neutral-loss pattern (Idir et al., 2025).

3.1.6. Caffeic acid and derivatives

Caffeic acid (65) generated a [M−H]⁻ ion at m/z 179.03563 (C₉H₇O₄), with a dominant decarboxylation fragment at m/z 135.04579. Compound 54 was identified as its isomer. Compounds 51, 55, 61 and 66 showed a common precursor ion at m/z 311.04129 and generated fragment ions corresponding to caffeic acid and tartaric acid-derived moieties, supporting their assignment as caffeic acid–tartaric acid conjugates (Martinović & Cavoski, 2020). Furthermore, compound 55 was confirmed as caftaric acid, while compounds 51, 61, and 66 were assigned as its isomers. Compounds 56 and 67 were also tentatively annotated as related caffeic acid–tartaric acid conjugates.

3.1.7. Gallic acid and tannin derivatives

Gallic acid (38) was identified by its [M−H]⁻ ion at m/z 169.01480 and a characteristic fragment at m/z 125.02498. Based on this fragmentation pattern, a series of galloylated derivatives and hydrolyzable tannins were further annotated. Compound 45 was assigned as erythrulose gallate, and compounds 40, 42, 47, and 49 were tentatively identified as 3-O-galloyl-d-glucose isomers (Hatano et al., 1989). Compounds 39, 43, and 48 were assigned as 4-O-galloylquinic acid isomers (Yeap Foo, 1995), while compounds 35, 36, 37, 41, and 46 were tentatively annotated as 7-O-galloyl-D-sedoheptulose isomers (Zhang et al., 1999).

In addition, compounds 44, 50, 57, 64, and 71 were tentatively assigned as 2,3-di-O-galloyl-d-glucose based on their common precursor ions and typical galloyl-loss fragmentation pathways (Hatano et al., 1989). Compounds 53, 58, 60, and 70 were tentatively annotated as digalloyl-sedoheptulose derivatives (Ho Lee et al., 1989), whereas compounds 52, 59, and 69 were proposed to be erythrulose gallate-related derivatives. Compound 72 was tentatively identified as 3,4-dimethoxyphenol-1-β-D-(6′-O-galloyl)-glycoside (Tian et al., 2015).

Compound 78 showed a [M−H]⁻ ion at m/z 665.10070 (C₂₈H₂₅O₁₉) and fragmented to yield ions corresponding to digalloyl-sedoheptulose, along with additional diagnostic fragments indicative of an extra terminal galloyl group. This compound was tentatively proposed to be a methylol tannin-type structure, namely 1,7-di-O-galloyl-D-sedoheptulose with an additional galloyl substituent. Finally, compound 81 was tentatively annotated as a tetra-O-galloyl-glucose derivative (Sun, Xue, et al., 2023). Overall, tannin-related constituents in CF were dominated by galloyl glucose, galloyl quinic acid, and galloyl sedoheptulose derivatives.

3.1.8. Other organic acids

Several additional organic acids were also detected. p-Coumaric acid (79) was confirmed by reference comparison, and compounds 63, 68, 74, and 75 were assigned as its isomers based on their shared precursor ion and characteristic CO₂ neutral loss (Serrano-García et al., 2025). Compounds 77 and 80, showing identical precursor ions and similar MS/MS spectra, were tentatively identified as malyl syringate isomers according to diagnostic neutral-loss patterns and literature comparison (Wang et al., 2025). Compound 4 was tentatively assigned as gluconic acid, compound 2 as D-glucuronic acid (Ancillotti et al., 2017), and compound 3 as glutaric acid by database matching. In addition, compound 33 was tentatively identified as 2-furoic acid (Cao et al., 2025), and compound 82 as azelaic acid, based on accurate mass data, characteristic fragment ions, and database/literature matching.

3.2. Multivariate statistical analysis of sourness and astringency in CF

To elucidate the chemical basis of sourness and astringency in CF from four geographical origins, targeted LC–MS metabolomics data were analyzed using univariate and multivariate statistical approaches. PCA revealed evident differences in the profiles of organic acids and tannins among samples (Fig. 2A). The first two principal components (PC1 and PC2) explained 31.3% and 23.4% of the total variance, respectively, together accounting for 54.7% of the total variation. The PCA score plot showed partial overlap between the CFA and CFB groups, whereas the CFC and CFD groups were more clearly separated, suggesting greater compositional similarity between CF samples from Henan and Shaanxi and more pronounced differences in samples from Zhejiang. These results indicate that regional variation in CF was associated with differences in the relative abundance of sourness- and astringency-related constituents.

Fig. 2.

Multivariate statistical analysis of targeted metabolomics for CF from four origins. (A) PCA score plot showing group clustering (CFA, CFB, CFC, CFD). (B) PLS-DA score plot distinguishing the four groups. (C) LOOCV cross-validation of the PLS-DA model. (D) Model validation based on 100 permutation tests.

The PLS-DA score plot further improved group separation (Fig. 2B). CFA and CFB exhibited similar metabolic profiles, while both were clearly distinguished from CFC. The clearer separation observed in the supervised model further supported region-dependent differences in targeted metabolite composition, especially for the CFC group. Model validation using leave-one-out cross-validation indicated that four latent variables provided the optimal model, yielding an R² value of 0.89 and a Q² value of 0.67, reflecting good model fit and predictive ability. These parameters indicate satisfactory explanatory and predictive performance of the model. In addition, 100 permutation tests confirmed that the PLS-DA model was not overfitted (Fig. 2C, D). Overall, the multivariate results demonstrate that the four producing regions differed in their chemical profiles, providing a basis for further interpretation of sourness and astringency in CF.

3.3. Determination of the contents of 12 recognized components

Quantification of 12 compounds using their respective reference standards revealed pronounced differences in component concentrations between the CFC group and the other three groups (Fig. 3A-L). The CFC samples exhibited substantially higher levels of tartaric acid, citric acid, and shikimic acid, while the contents of caffeic acid, caftaric acid, and p-coumaric acid were significantly lower. Consistent with these trends, total organic acid measurements (Fig. 5A) also demonstrated that the CFC group possessed markedly higher overall organic acid levels. These high-abundance acids may contribute to the acidic chemical background of CFC samples. The regression equations and detection limits are provided in Supplementary Table S1.

Fig. 3.

Contents of 12 reference-confirmed compounds in 70% aqueous methanol extracts of CF samples from four producing regions. Concentrations are expressed as mg/g dry weight. Differences were analyzed by one-way ANOVA with LSD post-hoc test (^⁎p < 0.05, ^⁎⁎p < 0.01).

Fig. 5.

Total organic acid content and E-tongue analysis of CF from four origins. (A) Total organic acid content, expressed as % (w/w) on a dry-weight basis. (B) Cluster analysis integrating total organic acid content and E-tongue data. (C) Radar chart of mean E-tongue scores. (D) Box plots of five basic taste intensities. (E–H) Radar charts of individual origins: (E) Henan, (F) Shaanxi, (G) Zhejiang, and (H) Shanxi.

3.4. Sensory quantitative descriptive analysis of flavor in CF

3.4.1. Human sensory evaluation

Human sensory evaluation assessed taste differences among CF samples from different regions, scoring sourness, astringency, bitterness, and taste harmony on a 10-point scale. Significant variations were observed (p < 0.05) (Fig. 4A): CFC samples had the highest sourness (8.02 ± 0.60), CFA and CFB samples showed stronger astringency (7.02 ± 0.57 and 7.01 ± 0.60), and bitterness was more pronounced in CFC. Regional differences in taste reflected underlying chemical composition. Specific data are provided in Supplementary Tables S2–4.

Fig. 4.

Human sensory evaluation and its correlation with E-tongue responses.

Sensory results largely aligned with E-tongue profiles, especially for sourness (p < 0.001). Correlations for astringency and bitterness were moderate (p < 0.05) (Fig. 4B), indicating E-tongue cannot fully capture human astringency and bitterness perception. These findings support E-tongue as a comparative taste fingerprinting tool, while highlighting the need to interpret astringency and bitterness alongside sensory data.

3.4.2. Instrumental taste profiling by E-tongue

E-tongue analysis provided an objective characterization of CF taste attributes (Fig. 5C). Across the 27 samples, sourness values ranged from 1.18 to 7.78 and bitterness from 0.06 to 4.75, while astringency (−2.95–0.19), aftertaste-B (−1.72–5.98), and aftertaste-A (−4.21–1.12) showed broader variability. Cluster analysis using the Ward.D2 method, incorporating E-tongue responses and total organic acid content, classified the samples into three distinct clusters, with the CFC group clearly separated from the others (Fig. 5B). To further clarify taste differences, radar charts were employed to visualize the mean taste profiles of each group (Fig. 5D). The CFC group exhibited significantly higher acidity (4.54–7.78), as well as elevated bitterness (0.16–4.75) and aftertaste-B (0.32–5.99) scores relative to the overall mean. Conversely, the CFC group showed relatively low astringency (−2.95 to ∼1.10) and aftertaste-A (−4.21 to ∼0.90) scores. In comparison, the CFA group had higher astringency (−2.09–0.19), and its aftertaste-A (−2.25–1.12) was slightly stronger than its astringency.

Notably, radar charts (Fig. 5E–H) illustrated distinct flavor disparities among the four CF groups: CFA and CFB exhibited similar profiles, characterized by pronounced astringency and lower acidity, whereas CFC and CFD showed markedly higher acidity. The CFC group was further distinguished by elevated bitterness values and comparatively lower astringency and aftertaste-A. Collectively, these findings reveal clear distinctions in sourness and astringency among CF samples of different geographic origins. Specifically, samples from Henan were characterized by pronounced astringency and aftertaste-A, whereas those from Zhejiang displayed dominant sourness and bitterness notes. Importantly, taste profiles alone were sufficient to clearly discriminate CF samples according to their geographic origin. Detailed E-tongue data for each group are provided in Supplementary Table S5.

3.5. Integration of chemical profiles with E-tongue responses

Therefore, the correlations and PLSR models based on 70% methanol extracts should be interpreted as exploratory, hypothesis-generating associations rather than direct evidence of sensory contribution. Pearson correlation analysis revealed clear associations between E-tongue responses and identified metabolites (Fig. 6). Sourness was positively correlated with malic acid and its derivatives, as well as lactic acid and quinic acid glucoside. Bitterness and aftertaste-B showed strong positive correlations with multiple galloylated compounds, including 1,7-di-O-galloyl-D-sedoheptulose, erythrulose gallate, and 4-O-galloylquinic acid. Astringency showed significant positive correlations with 3-O-galloyl-d-glucose and p-coumaric acid, but a significant negative correlation with 1-O,7-O-digalloyl-d-glucose, implying that the degree of astringency is largely associated with specific gallic acid polymers. Furthermore, aftertaste-A exhibited positive correlations with tartaric acid, aconitic acid, caftaric acid, p-coumaric acid, and p-coumaroyltartaric acid.

Fig. 6.

Correlation heatmap between E-tongue responses and chemical components. Colors indicate the strength and direction of correlations between sensory responses and component levels. Correlation analysis was performed by Pearson. ^⁎p < 0.05, ^⁎⁎p < 0.01, ^⁎⁎⁎p < 0.001, ^⁎⁎⁎⁎p < 0.0001.

To quantify the relationships between chemical components (X) and taste attributes (Y), PLSR models were established, and model performance was evaluated using the coefficient of determination (R²) and root mean square error (RMSE) (He et al., 2025). As shown in Fig. 7, the predictive performance varied among taste attributes. The sourness model showed strong predictive performance on the test set (R² = 0.945, RMSE = 0.293), suggesting that variation in sourness-related E-tongue responses was statistically associated with variation in the measured methanol-extractable organic acids. The bitterness and aftertaste-B models also showed acceptable predictive ability, with test-set R² values of 0.820 and 0.826, respectively, indicating that galloylated compounds may be associated with these attributes.

Fig. 7.

PLSR-based predictive model linking E-tongue responses to chemical components. Model performance was as follows: sourness, R² = 0.945 and RMSE = 0.293; bitterness, R² = 0.820 and RMSE = 0.503; astringency, R² = 0.597 and RMSE = 0.471; aftertaste-A, R² = 0.522 and RMSE = 0.786; and aftertaste-B, R² = 0.826 and RMSE = 0.680.

In contrast, the predictive performance of the astringency and aftertaste-A models was comparatively limited, with test-set R² values of 0.597 and 0.522 and RMSE values of 0.471 and 0.786, respectively. These results indicate that the current chemical variables only partially explained the variation in astringency-related responses. Therefore, compounds such as 3-O-galloyl-d-glucose, p-coumaric acid, galloylquinic acid derivatives, and related hydrolyzable tannins should be interpreted as candidate astringency-associated markers rather than confirmed causal contributors. The lower robustness of the astringency-related models may reflect the complex oral nature of astringency, which involves not only compound concentration but also tannin polymerization, salivary protein interactions, oral lubrication, aftertaste formation, and individual sensory variability. Detailed regression coefficients are provided in Supplementary Table S6.

$${\mathbf{R}}^{\mathbf{2}}=1-\frac{{\sum }_{i=1}^n{\left({\widehat{y}}_i-y_i\right)}^2}{{\sum }_{i=1}^n{\left(\overline{y}-y_i\right)}^2}$$ (1)

$$\mathbf{RMSE}=\sqrt{\frac{1}{n}{\sum }_{i=1}^n{\left(y_i-{\widehat{y}}_i\right)}^2}$$ (2)

where n is the total number of the test set, yᵢ and ŷᵢ are the true and predicted values of the test set, respectively, and ȳ denotes the average of the true values.

3.6. Molecular docking analysis of sourness- and astringent-related components

Compounds showing exploratory associations with selected E-tongue responses were subjected to hypothesis-generating molecular docking analysis (Supplementary Tables S7–S10). Sourness-related compounds, including trans-aconitic acid, malic acid glucoside, quinic acid glucoside, and malic acid diesters, showed favorable binding energies with OTOP1, PKD1L2, and PKD2L1, providing preliminary support for their possible involvement in sourness-related responses. Bitterness- and aftertaste-B-associated compounds, including 4-O-galloylquinic acid, 2,3-di-O-galloyl-d-glucose, 7-O-galloyl-D-sedoheptulose, erythrulose gallate, and digalloyl sedoheptulose derivatives, also showed favorable docking interactions, suggesting their potential contribution to bitterness-related attributes. For astringency- and aftertaste-A-related compounds, including 3-O-galloyl-d-glucose, 2,3-di-O-galloyl-d-glucose, and caftaric acid derivatives, docking with β-casein was used only as a protein-interaction model relevant to polyphenol–protein binding, rather than as direct evidence of sensory perception. Given the relatively limited predictive performance of the astringency and aftertaste-A PLSR models, these tannin-related compounds should be regarded as potential astringency-associated candidates requiring further validation by sensory recombination, omission, and salivary protein-binding experiments. Representative docking poses are shown in Fig. 8A–H.

Fig. 8.

Molecular docking of selected candidate taste-associated compounds with taste-related targets/protein models. (A–D) Docking of 7-O-galloyl-D-sedoheptulose, erythrulose gallate, 1-O,7-O-digalloyl-D-sedoheptulose, and 1,7-di-O-galloyl-D-sedoheptulose with the sour receptor OTOP1. (E) Docking of trans-aconitic acid and malic acid glucoside with the sour receptor PKD2L1. (F) Docking of malic acid aconitic acid diester with the sour receptor PKD1L2. (G, H) Docking of 3-O-galloyl-d-glucose and 2,3-di-O-galloyl-d-glucose with the astringency-related protein model β-casein.

It should be noted that the predictive robustness differed substantially among taste attributes. Sourness was predicted with high accuracy, suggesting that the measured organic acids captured the major chemical variation associated with sourness-related responses. By contrast, astringency and aftertaste-A showed weaker model performance, indicating that these attributes cannot be fully explained by the current set of chemical variables. Astringency is a complex oral sensation involving drying, puckering, and lingering aftertaste, which may be affected by tannin molecular size, polymerization degree, salivary protein precipitation, oral lubrication, and individual panelist variability. Therefore, although galloylated tannins and related phenolic acids were statistically associated with astringency-related responses, they should be considered candidate markers rather than definitive contributors. Further recombination, omission, threshold, and saliva-binding assays are needed to confirm their causal roles.

Several methodological limitations should also be considered when interpreting the proposed candidate markers. Annotation confidence varied across the 82 compounds. Compounds confirmed with authentic standards had the highest confidence, whereas those assigned from accurate mass, MS/MS fragmentation, database matching, or literature were considered tentative. Putative derivatives require further confirmation using authentic standards, purified isolates, or orthogonal structural methods. The absence of pooled-QC-based feature filtering is also a methodological limitation. Therefore, markers lacking authentic-standard confirmation should be regarded as provisional.

3.7. Cross-extraction concordance between 70% methanol and analytical aqueous extracts

For the analytical aqueous extraction, water was substituted for 70% methanol while the remaining extraction and LC–MS sample-preparation procedures were kept consistent with the analytical workflow. This extract was distinct from the sensory infusion prepared according to Section 2.7. Of the 82 matched analytes, 79 were suitable for correlation analysis. The two extraction systems showed a significant but modest overall correlation (r = 0.308, permutation P < 0.001; Figs. S1-S3), with class-level correlations of 0.343 for organic acids and 0.266 for tannins (Table S11). Among them, 61 (77.2%) showed positive correlations and 24 remained significant after FDR correction (q < 0.05). Representative compounds, including malic acid, fumaric acid, tartaric acid, caffeic acid, gallic acid, and erythrulose gallate showed stronger concordance (r = 0.683–0.929, q < 0.001; Table S12). These results indicate partial and analyte-dependent concordance between the two extraction systems and support the use of 70% methanol for exploratory chemical screening.

4. Conclusion

In conclusion, this study integrated human sensory evaluation, E-tongue fingerprinting, chemical profiling, chemometric analysis, and molecular docking to screen candidate compounds associated with the sour and astringent characteristics of CF. Malic acid-related constituents and galloylated tannins showed exploratory associations with sourness and astringency- or bitterness-related responses, respectively. Cross-extraction analysis showed modest, analyte-dependent concordance between 70% methanol and analytical aqueous extracts (r = 0.308), with stronger agreement for malic acid, tartaric acid, fumaric acid, gallic acid, caffeic acid, and erythrulose gallate (r = 0.683–0.929, q < 0.001). Because correlation and PLSR analyses were based on 70% methanol extracts, these associations remain matrix-dependent and hypothesis-generating and require further matrix-matched sensory validation.

CRediT authorship contribution statement

Yadi Hou: Writing – original draft, Formal analysis, Data curation, Conceptualization. Hui Guo: Writing – review & editing, Project administration, Investigation, Funding acquisition. Yalin Liu: Validation, Methodology, Funding acquisition. Yixiao Li: Validation, Investigation, Data curation. Manliang Chen: Investigation, Data curation. Haiyang Du: Resources, Investigation, Data curation. Jun Chi: Visualization, Validation, Methodology. Zhimin Wang: Supervision, Resources, Funding acquisition. Liping Dai: Supervision, Project administration, Methodology, Funding acquisition.

Ethics declaration

Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed.

This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place. This study was conducted in accordance with the Declaration of Helsinki (2024), the Measures for Ethical Review of Life Science and Medical Research Involving Humans (2023), China, and relevant institutional guidelines for research involving human participants. This study was approved by the Ethics Review Committee of Henan University of Chinese Medicine. (Approval No. HNSZYYky2025081001)

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: Supplementary tables and figures presenting chemical quantification, sensory evaluation, electronic tongue analysis, PLSR modeling, molecular docking, cross-extraction correlation analysis, and extraction solvent comparisons of Corni Fructus.

Data availability

Data will be made available on request.