Multi-omics profiling and correlation analysis of microbial succession and dynamic flavor formation in Zhejiang rosy vinegar

Abstract

The quality of Zhejiang rosy vinegar (ZRV) is defined by its complex chemical profile, yet the biochemical links between microbial metabolism and flavor evolution remain unclear. We employed an integrated multi-omics approach (amplicon sequencing, GC–MS, GC-IMS, and UHPLC-MS/MS) to elucidate the molecular underpinnings of ZRV flavor formation. Results revealed a microbial succession dominated by Acetobacter, Lactobacillus, Saccharomyces, and Monascus. Metabolomics demonstrated a dynamic shift in volatile profiles and the accumulation of non-volatile organic acids, driven largely by branched-chain amino acid metabolism. Integration identified critical biochemical linkages: Saccharomyces and Acetobacter activity shaped the early ester/acid framework, while Monascus and Pseudomonas drove acetoin accumulation. Furthermore, Lactobacillus and Monascus abundances were strongly associated with flavonoid enrichment, underpinning ZRV's antioxidant capacity. These findings provide mechanistic insights into microbial-chemical interactions governing flavor and functional compound synthesis, offering a robust statistical and chemical foundation for the industrial quality control of traditional fermented foods.

Graphical abstract

Highlights

• •Integrated multi-omics decodes the flavor-forming chemistry of ZRV.
• •Microbial succession was mapped to dynamic volatile and non-volatile profiles.
• •Metabolic activity of Saccharomyces and Acetobacter drives early ester synthesis.
• •Amino acid degradation and organic acids determine the core non-volatile taste.
• •Monascus and Lactobacillus are closely linked to robust antioxidant capacity.

1. Introduction

Traditional grain vinegars are complex fermented ecosystems, prized not only for their distinctive flavors but also for a rich profile of bioactive compounds, including organic acids, polyphenols, and free amino acids, which are believed to confer various health benefits (Cianciosi et al., 2022; Li et al., 2024; Xia et al., 2020). Among the celebrated varieties of Chinese traditional vinegars, Zhejiang rosy vinegar (ZRV) is particularly renowned for its unique rosy hue, profound umami taste, and mellow aroma. The quality of ZRV is intrinsically linked to the metabolic activities of a dynamic microbial consortium that transforms raw grain materials through successive stages of fermentation (Bai et al., 2024; Niu et al., 2023; Peng et al., 2025). However, traditional production relies on empirical methods, which often leads to significant batch-to-batch variability. This highlights an urgent need for a deeper scientific understanding of the underlying processes to ensure product consistency and quality (Shao et al., 2025).

Scientific interest in ZRV has grown, with numerous studies providing foundational insights into its complex ecosystem. Foundational work from our group, for instance, established that environmental microorganisms act as heterogeneous sources for the fermentation, creating a distinct ‘microbial terroir’ that shapes the community structure in different factories (Chang, Zhang, Fang, et al., 2025). Other studies have separately tracked microbial succession and its link to flavor (Fang et al., 2021; Fang et al., 2023) or characterized the non-volatile and volatile chemical profiles (Chang, Zhang, Qin, et al., 2025; Zhang et al., 2023). However, while these studies identified the key microbial players and the resulting chemical landscapes, a holistic, mechanistic framework connecting them is still missing. The critical knowledge gap lies in understanding precisely how the microbial succession contributes to the dynamic formation of ZRV's final sensory profile and bioactive potential.

Therefore, this study employed an integrated multi-omics strategy to systematically investigate the microbial contributions to quality formation during ZRV fermentation. Our approach integrated high-throughput sequencing to track microbial community dynamics with comprehensive profiling of volatile aroma (leveraging the complementary advantages of GC–MS for accurate identification and GC-IMS for trace-level high-sensitivity detection) and non-volatile taste/bioactive metabolites (UHPLC-MS/MS). By constructing a multi-level correlation network, we aimed to achieve several key objectives: (1) to characterize the successional patterns of key microorganisms; (2) to identify key volatile aroma-active compounds and track their evolution; (3) to profile the dynamic changes of non-volatile flavor determinants and bioactive compounds; and (4) to explore the statistical linkages between key microbial taxa and the specific accumulation of flavor and functional compounds. Ultimately, this integrated approach aims to explore the statistical relationships by which the microbiome and the chemical environment coincide with the volatile and non-volatile quality profiles of ZRV, providing a scientific foundation for process optimization and industrial modernization.

2. Materials and methods

2.1. Sample collection and reagents

2.1.1. Vinegar samples collection

ZRV samples were collected between May and November 2024 from a commercial producer located in Shaoxing, Zhejiang Province, China, representing a single, complete, and continuous traditional brewing cycle. Specifically, the fermentation was initiated through spontaneous (natural) fermentation, relying entirely on the indigenous microbiota associated with the raw materials and the traditional workshop environment, without the addition of any commercial starter cultures, pure-culture strains (such as yeast or acetic acid bacteria), or backsloshing materials (such as aged vinegar mash or mother vinegar). To capture the key transformations during production, sampling was conducted at five distinct fermentation stages defined by specific time points according to the traditional brewing schedule of ZRV (which typically consists of a 20-day solid-state fermentation phase and a subsequent 150-day liquid-state fermentation phase). Specifically, samples were collected at the following time points: (1) preliminary solid-state fermentation (Pre-SSF, day 7 of the SSF phase); (2) late solid-state fermentation (Late-SSF, day 20 of the SSF phase, marking the completion of SSF); (3) early liquid-state fermentation (Early-LSF, day 30 of the LSF phase); (4) mid-liquid-state fermentation (Mid-LSF, day 90 of the LSF phase); and (5) late liquid-state fermentation (Late-LSF, day 150 of the LSF phase, marking the completion of LSF). At each stage, three biological replicates were collected, with each replicate (approx. 150 g) originating from a separate fermentation workshop to account for spatial and microclimatic variations within the facility. Upon collection, all samples were immediately stored at −80 °C until further analysis. It should be noted that while microbial dynamics and core physicochemical indicators (reducing sugar, alcohol content, pH, and acidity) were monitored throughout the entire five-stage fermentation cycle, other quality-related attributes (amino acid nitrogen, color profiles, and electronic nose responses) along with high-resolution chemical analyses (volatile and non-volatile metabolites) were specifically focused on the liquid-state fermentation (LSF) stages (Early-LSF, Mid-LSF, and Late-LSF). This strategic sampling was adopted because the LSF phase represents the critical period for the accumulation and transformation of key flavor-active and bioactive compounds. Furthermore, the homogeneous liquid matrix during LSF allows for more robust, accurate, and reproducible instrumental measurements compared to the highly heterogeneous solid-state mash.

2.1.2. Reagents

All chemical standards, unless otherwise specified, were purchased from Aladdin Chemical Co., Ltd. (Shanghai, China). Chromatography-grade methanol, 2-ethylbutyric acid, as well as standard kits for organic acids and free amino acids, were obtained from Sigma-Aldrich Co. Ltd. (Shanghai, China).

2.2. Analysis of physical and chemical indicators

Reducing sugar, pH, and acidity of the samples were measured using ISO 15154-2003, ISO 1842-1991, and ISO 750-1981 methods, respectively. The alcohol content was measured using a modified procedure with potassium dichromate oxidation (Cai et al., 2018). Amino acid nitrogen concentration was determined using titration methods (Liu et al., 2023). The color difference of the ZRV was determined using a colorimeter (CR-400, Konica Minolta, Tokyo, Japan) according to CIELAB parameters, specifically L*, a*, and b* (Petchpoung et al., 2020).

2.3. Electronic nose analysis

The volatile profile of each vinegar was analyzed on an in-house constructed electronic nose (e-nose) (Yang et al., 2020). The analytical procedure began with placing 10 mL of ZRV into a 250 mL conical flask, which was then sealed and incubated for 30 min at 25 °C for headspace equilibration. Subsequently, a 15 mL volume of the headspace was manually drawn and injected into the instrument. Data was acquired for 120 s by a 14-sensor array (specifications in Table S1) under a gas flow of 0.60 mL/min. A 180 s cleaning cycle was run after each measurement. Each sample was analyzed in triplicate to ensure reproducibility.

2.4. Microbiological analysis

2.4.1. Determination of culturable microbial counts

Culturable populations of total bacteria, total fungi, yeast, lactic acid bacteria (LAB), and acetic acid bacteria (AAB) were quantified using the plate count method. Briefly, 25 g of solid sample or 25 mL of liquid sample was homogenized in 225 mL of sterile saline (0.85% NaCl), followed by 10-fold serial dilutions. Inoculations (1 mL) of appropriate dilutions were performed in triplicate using the pour-plate technique. The specific media and incubation conditions were: total bacteria on Plate Count Agar (PCA) at 36 °C for 48 h; total fungi on Potato Dextrose Agar (PDA) at 30 °C for 72 h; LAB anaerobically on MRS agar at 37 °C for 72 h; AAB on selective AAB medium at 30 °C for 72 h; and yeast on YEPD acidic agar at 28 °C for 5 days. Plates containing 30–300 colonies (10–150 for yeast) were counted. The results were expressed as lg CFU/g or lg CFU/mL (mean ± standard deviation).

2.4.2. Microbiome analysis

The microbial community dynamics across the five fermentation stages were investigated via high-throughput amplicon sequencing. Total genomic DNA was extracted from ZRV samples using the E.Z.N.A.® Soil DNA Kit (Omega Bio-tek, USA) according to the manufacturer's protocols. The bacterial 16S rRNA gene (V3–V4 region) was amplified using the 338F (5′-ACTCCTACGGGAGGGAGGA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) primer set, while the fungal ITS region was amplified using the ITS1F (5′-CTTGGTCATTTAGAGGAAGTAA-3′) and ITS2R (5′-GCTGCGTTCTTCATCGATGC-3′) primer set. The PCR products were sequenced on an Illumina MiSeq platform (2 × 300 bp).

Raw sequencing reads were processed using the QIIME 2 pipeline (v.1.9.1). Briefly, raw reads were de-multiplexed and trimmed to remove adapter sequences and primers using Cutadapt. High-quality sequences were subjected to the DADA2 plugin for denoising, chimera removal, and the generation of Amplicon Sequence Variants (ASVs). To normalize the sequencing effort and minimize the impact of uneven sequencing depth across samples, all libraries were rarefied to 20,000 reads per sample for downstream diversity analysis. Potential reagent or environmental contamination was monitored by including negative control (blank) samples during both DNA extraction and PCR amplification processes. Taxonomic assignment of the representative ASVs was performed using a pre-trained Naive Bayes classifier against the SILVA (v138) database for bacteria and the UNITE (v8.x) database for fungi, with a confidence threshold of 0.7.

2.5. Determination of bioactive compounds and antioxidant activity

2.5.1. Total polyphenol and total flavonoid contents

The total polyphenol content was determined via the Folin-Ciocalteu colorimetric assay, with absorbance measured at 765 nm. Similarly, total flavonoid content was quantified using the aluminum chloride method, and the absorbance was recorded at 510 nm. Results were calculated based on standard curves and expressed as mg of gallic acid equivalents (GAE)/mL for the total polyphenol and mg of rutin equivalents (RE)/mL for total flavonoid.

2.5.2. Organic acids

Organic acids were quantified via HPLC (Waters 2695) after filtering the samples through a 0.22 μm membrane. Chromatographic separation was achieved on a Waters Symmetry C18 column (4.6 × 250 mm, 5 μm) maintained at 30 °C. The analysis used an isocratic mobile phase of 5:95 (v/v) methanol and 6 g/L sodium dihydrogen phosphate (pH 2.6) at a flow rate of 0.8 mL/min, with detection at 210 nm.

2.5.3. Free amino acids

Free amino acids were analyzed on an A300 amino acid analyzer (membra Pure GmbH) following the protocol described by Hu et al. (2024). Briefly, after protein precipitation with sulfosalicylic acid, the supernatant was analyzed. Amino acids were detected via post-column ninhydrin derivatization at 570 nm (440 nm for proline).

2.5.4. Antioxidant activity

The antioxidant capacity was comprehensively evaluated through four assays: DPPH radical scavenging, ABTS radical scavenging, hydroxyl radical (·OH) scavenging, and reducing power (RP). The specific methodologies for each assay were followed as described in the cited literature: Hu et al. (2024) for DPPH, Zhao et al. (2018) for ABTS, Chang et al. (2022) for hydroxyl radical scavenging, and Yin et al. (2024) for RP.

2.5.4.1. DPPH radical scavenging activity

Briefly, 0.3 mL of diluted ZRV sample was mixed with 3.0 mL of distilled water and 6.0 mL of 0.1 mmol/L DPPH-methanol solution. After reacting in the dark for 30 min, the absorbance was measured at 517 nm (A₁). The sample background control (A₂) was prepared by replacing the DPPH-methanol with pure methanol. The blank (A₀) was prepared by replacing the sample with distilled water. The scavenging rate was calculated as follows:

$$\text{Inhibition}\left(\%\right)=\frac{\left[A_0-\left(A_1-A_2\right)\right]}{A_0}\times 100$$ (1)

2.5.4.2. ABTS radical scavenging activity

The ABTS^+· working solution was prepared and adjusted to an absorbance of 0.70 ± 0.02 at 734 nm. Diluted sample (1.0 mL) was mixed with 5.0 mL of ABTS^+· solution. After 10 min, the absorbance was read at 734 nm (A₁). The sample background control (A₂, sample + water) and blank control (A₀, water + ABTS) were measured. The scavenging rate was calculated using the same subtraction formula:

$$\left[A_0-\left(A_1-A_2\right)\right]/A_0\times 100$$

2.5.4.3. Hydroxyl radical (·OH) scavenging activity

Briefly, 1.0 mL of sample was mixed with 1.4 mL of 6 mmol/L H₂O₂, 0.6 mL of 20 mmol/L sodium salicylate, and 2.0 mL of 1.5 mmol/L FeSO₄. After incubation at 37 °C for 1 h, the absorbance was measured at 562 nm (A₁). The sample background control (A₂) was prepared without the addition of H₂O₂. The scavenging rate was calculated as:

$$\left[A_0-\left(A_1-A_2\right)\right]/A_0\times 100$$

2.5.4.4. Reducing power

1.0 mL of sample was incubated with 2.5 mL of phosphate buffer (0.2 mol/L, pH 6.6) and 2.5 mL of 1% potassium ferricyanide at 50 °C for 20 min, followed by addition of 1.0 mL of 10% trichloroacetic acid. After centrifugation, 2.5 mL of the supernatant was mixed with 2.5 mL of water and 0.5 mL of 0.1% FeCl₃. The absorbance was measured at 700 nm (A₁). A sample blank (A₂) without FeCl₃ was used to correct for the sample background.

2.6. Determination of volatile compounds

2.6.1. HS-GC–MS analysis

Volatile compounds were analyzed using an established HS-SPME-GC–MS method from our laboratory, with procedures based on Zhang et al. (2023). In a 20 mL headspace vial, 10 mL of vinegar was mixed with 10 μL of 2-ethylbutyric acid (18.4 mg/L) as an internal standard. After a 20-min equilibration at 50 °C, a DVB/PDMS-coated fiber was exposed to the headspace for 40 min at the same temperature. The fiber was then desorbed at 240 °C for 15 min in the GC inlet.

Analysis was conducted on an Agilent system (7890 A GC, 5975C MS). Separation was performed on a DB-WAX column (60 m × 0.25 mm, 0.5 μm) with a temperature program starting at 40 °C (20 min hold), ramping to 100 °C at 5 °C/min, and then to 210 °C at 3 °C/min. The MS operated in EI mode (70 eV) scanning from m/z 35–350. Compound identification relied on matching mass spectra with the NIST 14.0 database and comparing calculated retention indices (RIs) with literature values.

2.6.2. HS-GC-IMS analysis

A complementary volatile profile was obtained using a FlavourSpec® GC-IMS system (G.A.S., Germany). For this analysis, 1 mL of sample was incubated at 60 °C for 15 min. Volatiles were separated on an MXT-5 column at a constant temperature of 85 °C. The instrument utilized high-purity nitrogen for both carrier gas (programmed flow gradient) and drift gas (150 mL/min). Compound identification was confirmed by cross-referencing both the GC retention index (NIST 2020 database) and the IMS drift time using the VOCal software.

2.6.3. Odor activity values (OAVs) analysis

The sensory relevance of individual volatile compounds was assessed by calculating their OAVs, a standard methodology in flavor research (Grosch, 2001). The OAV was derived by dividing a compound's concentration by its published odor threshold in water. Compounds with an OAV ≥ 1 were considered key aroma-active contributors.

2.7. Untargeted metabolomic analysis by UHPLC-MS/MS

The untargeted metabolomic profiling of vinegar samples was conducted according to our previously published method (Zhang et al., 2023). In brief, metabolites were extracted with 80% methanol and analyzed on a Thermo Vanquish UHPLC-Q Exactive HFX Orbitrap MS system equipped with a BEH Amide column. Data acquisition, raw data processing, and metabolite annotation against the BiotreeDB database were carried out as previously described. Prior to multivariate analysis, the resulting data matrix was first normalized to the total peak area of each sample to correct for variations in sample loading. Subsequently, the normalized data was imported into SIMCA (v. 16.0.2) and subjected to Pareto scaling. The pre-processed data was then used for Principal Component Analysis (PCA) and other multivariate analyses.

2.8. Statistical analysis

All measurements were performed in triplicate (n = 3). For correlation analyses involving LSF-specific quality attributes, a total of 9 samples (3 biological replicates × 3 LSF stages) were used to calculate Pearson correlation coefficients. To account for multiple comparisons and control false positives in high-dimensional correlation matrices, raw p-values were adjusted using the Benjamini-Hochberg false discovery rate (BH-FDR) procedure. Statistical significance for all correlations was strictly defined as an absolute correlation coefficient (|r|) ≥ 0.70 and an adjusted p-value (FDR) < 0.05. All other statistical analyses were performed using SPSS (v. 25.0), SIMCA (v. 14.1), and various online tools. One-way ANOVA followed by Duncan's post-hoc test was used to determine significant differences (p < 0.05) among sample groups. For multivariate analysis of ‘omics data, principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were conducted. Key features were selected based on their variable importance in projection (VIP) scores. The expression patterns of these features were visualized as heatmaps using OmicStudio tools. To elucidate the underlying biological functions, metabolic pathway enrichment was performed via MetaboAnalyst 4.0, with pathways mapped to the KEGG database. Finally, correlation networks were constructed in Cytoscape (v. 3.10.0) to explore inter-variable relationships.

3. Results

3.1. Changes in physicochemical indicators

Fig. 1 provides a comprehensive illustration of the dynamic changes in the physicochemical properties, color characteristics, and electronic sensory attributes of ZRV during the fermentation process.

Fig. 1.

Changes in physicochemical factors, color characteristics, and electronic sensory properties during ZRV fermentation. (A) Reducing sugar, alcohol content, pH, and acidity, (B) Amino acid nitrogen, (C) Color difference, (D) Electronic nose. Data are presented as the mean ± standard deviation of three independent biological replicates (n = 3) for each fermentation stage.

Throughout fermentation, key physicochemical parameters exhibited significant fluctuations (Fig. 1A). The concentration of reducing sugars peaked during the Late-SSF stage (approx. 7.01 g/100 mL) and then sharply decreased. The ethanol content steadily increased during the solid-state fermentation, reaching its highest value (3.2% v/v) in the Early-LSF stage, after which it progressively declined to negligible levels. The pH value continuously decreased throughout the entire process, dropping from 4.74 in the Pre-SSF stage to 3.09 in the Late-LSF stage. Concurrently, total acidity increased from 0.11 g/100 mL to 6.30 g/100 mL, reflecting the acidification process driven by acetic acid production. These changes collectively indicate a clear transition from alcoholic fermentation to acetic acid fermentation, with the latter becoming predominant in the later stages.

As shown in Fig. 1B, the amino acid nitrogen content remained relatively stable (0.50–0.59 g/100 mL) throughout the liquid-state fermentation stage, with no significant differences observed (P > 0.05).

The color parameters showed marked changes during the liquid-state fermentation (Fig. 1C). Lightness (L*) decreased sharply from 29.89 to 1.73, indicating a significant darkening of the ZRV. The a* value shifted from negative (greenish tones) to positive (reddish tones), peaking in the Mid-LSF stage. In contrast, the b* value remained relatively stable, suggesting minor changes in the yellow or blue hue components. The C* value reached its maximum (1.19) in the Early-LSF stage, indicating the highest color saturation at this point. The hue angle (H°) decreased from approximately 120 to 70, reflecting a color shift from green to red.

E-nose technology was employed to detect and discriminate the volatile compounds in ZRV samples from different fermentation stages. As depicted in Fig. 1D, the ZRV samples elicited high response intensities on sensors S1 (sensitive to ammonia and amines) and S8 (sensitive to a broad range including cigarette smoke, cooking odors, volatile compounds, ammonia, hydrogen sulfide, and alcohol). Although the response intensities were relatively lower on sensors S4 (alcohols, organic solvents), S10 (LPG, combustible gases), S12 (CO, ethanol, organic solvents), S13 (smoke, cooking odors, H₂), and S14 (methane, natural gas), differences among the samples were still discernible. In the early stages of liquid-state fermentation, the response values to the sensors were higher, subsequently diminishing as the fermentation advanced.

3.2. Microbial succession analysis

To provide a comprehensive understanding of the microbial load, culturable microbial counts were monitored across the five fermentation stages (Fig. 2A). Initially, in the Pre-SSF stage, total fungi and yeast dominated the fermentation system with high counts of 9.48 ± 0.38 lg CFU/g and 8.40 ± 0.38 lg CFU/g, respectively, whereas LAB (3.85 ± 0.13 lg CFU/g) and AAB (3.78 ± 0.27 lg CFU/g) remained at lower levels. As fermentation progressed to Late-SSF, total bacteria and LAB counts increased dramatically, with LAB peaking at 9.28 ± 0.28 lg CFU/g, indicating a robust lactic acid fermentation during the solid phase. Upon transitioning to the liquid-state fermentation (LSF) phase, yeast and LAB counts progressively declined due to substrate depletion and high acidity. Conversely, AAB counts increased significantly, reaching a peak of 7.72 ± 0.33 lg CFU/mL during the Mid-LSF stage, driving the oxidation of ethanol into acetic acid. By the Late-LSF stage (day 150), all microbial groups exhibited a significant decline, with total bacteria dropping to 2.68 ± 0.73 lg CFU/mL, reflecting the exhausted and highly acidic environment at the end of fermentation.

Fig. 2.

Analysis of microbial counts microbial community composition and dynamics during ZRV fermentation. (A) Culturable microbial counts (lg CFU/mL) of total bacteria, total fungi, yeast, lactic acid bacteria, and acetic acid bacteria determined by plate count method. (B) Relative abundance of bacterial communities at the genus level. (C) Relative abundance of fungal communities at the genus level. (D) Microbial association network showing microorganisms with a relative abundance >1%. Nodes represent microbial genera, with size proportional to degree (number of connections) and color indicating the taxonomic group (bacteria or fungi). Edges represent correlations: solid lines indicate positive correlations and dashed lines indicate negative correlations. The thickness of the edge corresponds to the strength of the correlation. The correlation network in (D) was calculated based on all 15 samples across the five fermentation stages (5 stages × 3 biological replicates).

For the high-throughput sequencing of the bacterial community (Fig. 2B), the Pre-SSF stage was characterized by a high relative abundance of Arthrobacter (87.32%). As fermentation progressed, this initial community was quickly succeeded by higher relative abundances of Limosilactobacillus and Lactiplantibacillus in the Late-SSF and Early-LSF stages. Notably, Acetobacter emerged as a key genus during liquid fermentation, with its relative abundance peaking at 30.61% in the Mid-LSF stage. In the Late-LSF stage, the relative proportion of Acetobacter decreased substantially, while Acinetobacter increased to 36.80%.

In the fungal community (Fig. 2C), the solid-state fermentation was co-dominated by Saccharomyces (37.40%), Petromyces (28.94%), and Rhizopus (24.62%). The subsequent stages saw a rise of Saccharomyces (66.91% in Late-SSF), followed by a dramatic increase in Monascus to 88.05% in the Mid-LSF stage, establishing it as the dominant fungus.

Building on this observed community structure, a microbial co-occurrence network was constructed to illustrate the potential relationships among taxa with a relative abundance greater than 1% (Fig. 2D). The network revealed strong positive correlations among genera such as Acinetobacter, Pseudomonas, and Sphingomonas, as well as between them and the fungus Monascus. In addition, associations were observed among Lactobacillus, Lactiplantibacillus, and Saccharomyces. Conversely, negative correlations were found between Monascus and several other bacterial and fungal genera, including Lactobacillus, Lactiplantibacillus, Limosilactobacillus, and Aspergillus.

3.3. Changes in bioactive compounds and antioxidant activity

3.3.1. Total flavonoids and total polyphenols

As shown in Fig. 3A, during the liquid-state fermentation of ZRV, the total polyphenol content progressively decreased from 4.58 mg/mL in the Early-LSF stage to 2.08 mg/mL in the Late-LSF stage. In contrast, the total flavonoid content exhibited a different trend, first increasing and then decreasing, reaching its peak (0.47 mg/mL) in the Mid-LSF stage. The total flavonoid content nearly tripled from the Early-LSF to the Mid-LSF stage.

Fig. 3.

Analysis of the dynamic changes in functional components, organic acids, amino acids, and antioxidant activity during ZRV fermentation. (A) Total polyphenols and total flavonoids, (B) Organic acids, (C) Amino acids, (D) DPPH radical scavenging and ABTS radical scavenging capacity, (E) OH radical scavenging and reducing power, (F) Correlation analysis among all the above indicators. For (A)–(E), data are expressed as the mean ± standard deviation of three independent biological replicates (n = 3) for each stage. For (F), the correlation analysis was performed based on a total of 9 samples (3 liquid-state fermentation stages × 3 biological replicates).

3.3.2. Organic acids and free amino acids

Eight organic acids—acetic acid, succinic acid, pyruvic acid, malic acid, pyroglutamic acid, lactic acid, oxalic acid, and tartaric acid—were quantified in ZRV samples at different fermentation stages using HPLC (Fig. 3B). The proportions of these eight acids varied with fermentation time. The overall content of organic acids continuously increased throughout the process, reaching a peak concentration of 8400.14 mg/100 g in the Late-LSF stage. Acetic acid was the predominant organic acid, increasing rapidly from 193.31 mg/100 g in the Early-LSF stage to 4488.86 mg/100 g in the Late-LSF stage. Succinic acid was the second most abundant organic acid, increasing progressively with fermentation time to reach 3591.14 mg/100 g in the Late-LSF stage. Pyruvic acid content showed little variation during the liquid-state fermentation (from 178.46 mg/100 g to 174.40 mg/100 g), whereas malic acid content progressively decreased (from 81.99 mg/100 g to 36.98 mg/100 g).

The results for free amino acids (FAAs) during the liquid-state fermentation of ZRV are presented in Fig. 3C. The 17 FAAs were categorized based on their taste profiles as bitter (Arg, His, Ile, Leu, Met, Phe, Tyr, Val), tasteless (Cys, Lys, Pro), umami (Glu, Asp), or sweet (Ala, Thr, Gly, Ser). As fermentation progressed, the umami amino acids gradually decreased, while the total amount of the other three categories increased rapidly. Bitter amino acids were the most abundant in ZRV, accounting for 47.63% of the 17 amino acids by the Late-LSF stage. Compared to traditional grain vinegars, the contents of Glu, Asp, and Met did not change significantly during fermentation.

3.3.3. Antioxidant activity

The antioxidant performance of ZRV at different fermentation stages was evaluated by assessing its scavenging capacities against DPPH, ABTS, and hydroxyl radicals, as well as its reducing power. As shown in Fig. 3D, all ZRV samples exhibited high scavenging activities against both DPPH (approx. 70%) and ABTS (approx. 60%) radicals, but no significant differences were observed among the stages (P > 0.05). As shown in Fig. 3E, the hydroxyl radical scavenging capacity increased significantly in the Mid-LSF stage (P < 0.05) before declining slightly. In contrast, the reducing power progressively decreased from the Early-LSF stage (1.22) to the Late-LSF stage (0.64). These results indicate that ZRV possesses strong antioxidant capacity that remains relatively stable as fermentation time extends.

3.3.4. Correlations between bioactive compounds and antioxidant activity

The correlation matrix (Fig. 3F, Table S2), constructed based on the LSF stages (n = 9), revealed the complex statistical interrelationships between antioxidant capacities and various bioactive components. Reducing power showed strong positive correlations with malic acid (r = 0.943, FDR-adjusted P < 0.05), oxalic acid (r = 0.869, FDR-adjusted P < 0.05), and aspartic acid (Asp) (r = 0.924, FDR-adjusted P < 0.05), while exhibiting strong negative correlations with compounds such as tyrosine (Tyr) (r = −0.928, FDR-adjusted P < 0.05) and succinic acid (r = −0.894, FDR-adjusted P < 0.05). The OH· radical scavenging capacity was significantly and positively correlated with total flavonoids (r = 0.7742, FDR-adjusted P < 0.05), lactic acid (r = 0.729, FDR-adjusted P < 0.05), and alanine (Ala) (r = 0.729, FDR-adjusted P < 0.05), and strongly negatively correlated with pyruvic acid (r = −0.948, FDR-adjusted P < 0.05) and glutamic acid (Glu) (r = −0.836, FDR-adjusted P < 0.05). The ABTS radical scavenging capacity showed significant positive correlations with total polyphenols (r = 0.723, FDR-adjusted P < 0.05) and malic acid (r = 0.723, FDR-adjusted P < 0.05), and significant negative correlations with succinic acid (r = −0.725, FDR-adjusted P < 0.05) and threonine (Thr) (r = −0.721, FDR-adjusted P < 0.05).

3.4. Changes in volatile compounds

3.4.1. Identification of volatile compounds by GC–MS

Through GC–MS analysis, a total of 68 volatile compounds were identified in ZRV samples across the different liquid-state fermentation stages, as detailed in Table S3. These compounds included 14 acids, 6 alcohols, 2 aldehydes, 26 esters, 4 heterocyclic compounds, 5 ketones, 3 phenols, 2 pyrazines, and 6 other compounds. Fig. 4 illustrates the evolution of these volatile compounds during the liquid-state fermentation of ZRV. As shown in Fig. 4A, the number of volatile compounds increased as fermentation progressed, rising from 41 in the Early-LSF stage to 52 in the Late-LSF stage. The most significant changes were observed in the number of acids (which gradually increased) and esters (which first increased then decreased). The number of heterocyclic compounds gradually increased, while that of alcohols, aldehydes, phenols, and ketones remained relatively stable.

Fig. 4.

Analysis of volatile compounds in ZRV at different fermentation stages by GC–MS. (A) Number and types of volatile compounds. (B) The contents of each type of volatile compounds. (C) Heatmap analysis of volatile compounds. (D) OPLS-DA score plot of volatile compounds. (E) Cross-validation plot by 200 permutation tests (R² = 0.18, Q² = − 0.73). (F) Key differential volatile compounds identified by VIP score > 1. All analyses were performed using three independent biological replicates (n = 3) for each liquid-state fermentation stage (total n = 9 samples for multivariate statistics).

Meanwhile, as depicted in Fig. 4B, the total concentration of volatile compounds first increased and then decreased. The concentrations of acids (5853.78–25,952.29 μg/100 mL), ketones (275.90–7410.24 μg/100 mL), heterocycles (0–581.86 μg/100 mL), and pyrazines (0–76.94 μg/100 mL) increased with fermentation time. In contrast, the concentrations of esters (23,071.08–8494.05 μg/100 mL), alcohols (16,870.98–4157.85 μg/100 mL), and phenols (865.30–115.80 μg/100 mL) decreased. Based on these findings, a hierarchical clustering heatmap was generated to visualize the dynamic changes in the ZRV volatile fingerprint (Fig. 4C). The results show that 11, 15, and 20 volatile compounds reached their highest concentrations or were unique to the Early-LSF, Mid-LSF, and Late-LSF stages, respectively.

Using the GC–MS dataset, an OPLS-DA model was employed to identify inter-group differences in the flavor profiles of ZRV samples from different stages. As illustrated in Fig. 4D, samples from different groups showed clear separation along PC1 and PC2, indicating distinct differences in the volatile composition between the early, middle, and late stages of liquid-state fermentation. The tight clustering of replicate samples within the 95% Hotelling's T² confidence interval indicated excellent reproducibility. The high cumulative variance explained by the first two principal components demonstrated that the OPLS-DA model effectively characterized the differences among ZRV samples. To validate the goodness-of-fit and ensure the reliability of the model, a 200-permutation test was performed (Fig. 4E). The original model (represented by the rightmost points at x = 1.0) exhibited high R² and Q² values close to 1.0, indicating excellent predictive ability. Furthermore, the regression lines of the permuted models yielded y-axis intercepts of R² = 0.18 and Q² = −0.73. The significantly negative Q² intercept and the fact that all permuted Q² values were lower than the original one confirmed that the model was not overfitted. Based on this established model, VIP scores were calculated to identify key differential volatile compounds. Using a threshold of VIP > 1, a total of 9 volatile compounds were identified, including 5 esters, 1 alcohol, 1 acid, 1 ketone, and 1 phenol (Fig. 4F).

3.4.2. Identification of volatile compounds by GC-IMS

A complementary analysis of volatile compounds was performed using GC-IMS. Fig. S1 depicts the dynamic changes in volatile compounds detected by GC-IMS during the liquid-state fermentation of ZRV. First, a three-dimensional topographic plot was constructed using relative drift time (x-axis), retention time (y-axis), and signal peak intensity (z-axis) to visualize the evolution pattern of volatile compounds (Fig. S1A). The results showed that the number and intensity of ion signal peaks first decreased and then increased as fermentation progressed. Based on this, a two-dimensional top-view plot (Fig. S1B) was generated for precise observation. The red vertical line at 1.0 on the horizontal coordinate represents the reaction ion peak (RIP), and each dot on either side of the RIP represents a volatile compound. Deeper red colors indicate higher peak intensities (i.e., higher concentrations), while deeper blue colors indicate lower intensities. It was concluded that most volatile compound ion signals were detected in the region around 1.5 a.u. (x-axis) and between 100 and 900 s (y-axis). The distribution of volatile compounds (dots) changed continuously across the different fermentation stages.

Based on the volatile fingerprint plot (Fig. S1C) and Table S4, the characteristic volatile compositions of ZRV samples from different stages were detailed (p < 0.05). Among them, 3-methylbutyl butanoate, (E)-2-octenal-D, 1-octen-3-one-D, 2-decanone, and 2-heptylfuran were unique volatile compounds of the Mid-LSF stage. Acetic acid propyl ester, 3-methyl-1-pentanol-D, and 2-nonanone-D were unique to the Early-LSF stage. 1-Octen-3-one-M, ethyl hex-3-enoate, undec-2-enal, and 1-hexyl acetate were unique to the Late-LSF stage.

Importantly, Table S4 and Fig. S1C show that GC-IMS identified a total of 31 volatile compounds, 28 of which were detected exclusively by GC-IMS, thereby complementing the GC–MS data. As shown in Fig. 5A, the 31 volatile compounds included 8 esters, 8 ketones, 5 aldehydes, 5 others, 4 alcohols, and 1 acid. Esters and ketones were the most abundant classes. The heatmap in Fig. 5B displays the content changes of volatile compounds across the fermentation stages. Different clustering patterns were observed: Early-LSF samples showed higher contents of certain volatile compounds (e.g., 2-hexenal, 2-methylbutanal, butanoic acid ethyl ester), while Late-LSF samples showed elevated levels of others (e.g., 1-octen-3-one, ethyl hex-3-enoate, undec-2-enal). This suggests that the volatile composition changed dynamically. The OPLS-DA score plot (Fig. 5C) showed a clear separation of samples from different stages, demonstrating significant differences in their volatile profiles. The Hotelling's T² ellipse (95% confidence interval) further validated this clear clustering. A 200-permutation test was conducted to verify the robustness of the model (Fig. 5D). Similar to the GC–MS results, the original OPLS-DA model exhibited outstanding fitness with its R² and Q² values approaching 1.0 (rightmost points). The permutation test yielded a negative y-axis intercept for Q² (−0.878), confirming that the model was robust, statistically significant, and completely free from overfitting. VIP score analysis (Fig. 5E) was then employed to identify key differential volatile compounds (VIP > 1), such as butyl sulfide, ethyl hex-3-enoate, and undec-2-enoic acid, as the primary contributors to the inter-stage differences.

Fig. 5.

Analysis of volatile compounds in ZRV during fermentation by GC-IMS. (A) Number and classification of volatile compounds. (B) Heatmap analysis of volatile compounds. (C) OPLS-DA score plot. (D) Validation of the OPLS-DA model by 200 permutation tests (R² = 0.42, Q² = −0.88). (E) Key differential volatile compounds identified by VIP score > 1. All analyses were performed using three independent biological replicates (n = 3) for each liquid-state fermentation stage (total n = 9 samples for multivariate statistics).

3.4.3. OAVs analysis of GC–MS data

As shown in Table 1, a total of 26 key aroma compounds (OAV ≥ 1) were identified across the three fermentation stages, covering various chemical classes including esters, alcohols, acids, aldehydes, ketones, phenols, and pyrazines.

Table 1.

Key odor activity values (OAVs ≥1) for volatile compounds in ZRV detected by GC–MS.

Compounds	CAS	Odor description	Threshold (mg/L)	OAV
				Early-LSF	Mid-LSF	Late-LSF
Acetic acid	64–19-7	Sour	26	2.02	5.72	9.44
3-Methylbutanoic acid	503–74-2	Sour, sweat	0.49	< 1	1.92	5.80
Octanoic acid	124–07-2	Sweat	0.91	< 1	1.09	2.02
Decanoic acid	334–48-5	Oil	0.13	1.00	–	3.62
Palmitic acid	57–10-3	Fatty, waxy dairy type flavors	0.04	–	–	116.00
Oleic acid	112–80-1	Fatty	0.01	–	–	30.41
2-Methyl-1-propanol	78–83-1	Fruity	0.06	16.11	22.12	22.83
3-Methyl-1-butanol	123–51-3	Fermented, oily, fruity	0.004	2142.50	1502.50	1155.00
Phenethyl alcohol	60–12-8	Flowery, rose	1.4	10.41	11.54	10.69
Furfural	98–01-1	Almond, caramel, honey, roasted, fatty	0.77	–	< 1	3.29
Benzaldehyde	100–52-7	Caramel, fruity, bitter almond, burnt sugar	0.35	1.34	9.57	15.40
Ethyl acetate	141–78-6	Fruity, sweet	0.005	36,332.00	27,778.00	9370.00
Ethyl isobutyrate	97–62-1	Sweet	0.06	–	3.48	3.13
n-Propyl acetate	109–60-4	Fruity	0.24	2.58	–	–
Isobutyl acetate	110–19-0	Sweet, fruity, ethereal, apple, banana	0.3	5.87	6.73	4.93
n-Butyl acetate	123–86-4	Green, fruity	0.06	53.97	3.10	–
3-Methylbutyl acetate	123–92-2	Fruity	0.09	136.95	93.40	61.55
Ethyl caproate	123–66-0	Fruity, apple	0.002	222.73	268.18	90.91
Ethyl caprate	110–38-3	Fruity	0.005	144.00	320.00	–
Ethyl phenylacetate	101–97-3	Sweet, rose, honey, cocoa	0.65	1.28	5.89	3.85
Phenethyl acetate	103–45-7	Honey-like, floral, sweet, fruity	0.02	426.50	615.50	648.00
Ethyl palmitate	628–97-7	Fatty, rancid, fruity, waxy	2	2.11	9.87	0.58
Acetoin	513–86-0	Buttery, green	0.01	197.14	3096.43	4797.14
2-Nonanone	821–55-6	Fruity, sweet, earthy, herbal	0.08	–	1.50	–
p-Cresol	106–44-5	Fecal	0.004	–	–	28.21
2,3,5-Trimethylpyrazine	14,667–55-1	Nutty, chocolate	0.35	–	–	1.11

Note: In the table, “-” means not detected. “< 1” means that the OAV of the compound is <1, and the specific value is not shown. Threshold values and odor description are referenced from relevant literature (Chen et al., 2024; Chang, Zhang, Qin, et al., 2025; Wang et al., 2025) and websites (https://mffi.sjtu.edu.cn/database, https://www.chemicalbook.com).

Among all key aroma compounds, esters were the most numerous and significant contributors. Notably, ethyl acetate exhibited the highest OAV in all samples (9370.00–36,332.00). Its OAV decreased as fermentation progressed, but it remained the core substance providing a strong fruity-sweet aroma. Similarly, 3-methyl-1-butanol (OAVs 1155.00–2142.50) and phenethyl acetate (OAVs 426.50–648.00) were major flavor contributors. The OAV of phenethyl acetate steadily increased with fermentation time. Additionally, 3-methylbutyl acetate and ethyl hexanoate also played important roles in shaping the characteristic fruity aroma of ZRV.

The contributions of alcohols, ketones, and aldehydes to the flavor also changed significantly as fermentation progressed. Phenylethyl alcohol, with its pleasant floral and rose aroma, was an important flavor compound in all three stages, maintaining a high OAV. The contribution of acetoin (creamy, buttery aroma) was particularly prominent, with its OAV soaring from 197.14 in the early stage to 4797.14 in the late stage. Similarly, the flavor contribution of benzaldehyde (caramel, bitter almond notes) also increased significantly over time, with its OAV rising from 1.34 to 15.40.

The flavor profile of the late fermentation stage (Late-LSF) was the most complex, characterized by the emergence of several potent aroma compounds. Volatile fatty acids such as palmitic acid (OAV = 116.00) and oleic acid (OAV = 30.41) made a substantial contribution to the overall flavor only in the late stage. Furthermore, p-cresol (fecal odor), 2,3,5-trimethylpyrazine (nutty, chocolate aroma), and furfural (caramel, baked aroma) were also identified as key flavor compounds only in the late stage. The emergence of p-cresol introduced complex off-notes, while pyrazines and furfural added pleasant baked and nutty aromas.

3.5. Untargeted metabolomics analysis

To elucidate the dynamic biochemical changes during the liquid-state fermentation (LSF) of ZRV, an untargeted metabolomics analysis was performed using UHPLC-MS/MS. The results identified a diverse range of metabolites, which were subsequently classified (Fig. S2). Among them, organoheterocyclic compounds (23.33%), organic acids and their derivatives (19.39%), lipids and lipid-like molecules (12.55%), and benzenoids (12.52%) were the most predominant classes, collectively accounting for over 67% of the total identified metabolites. To visualize the overall differences in metabolic profiles across the fermentation stages, Principal Component Analysis (PCA) was performed on the pre-processed data (see Methods for normalization and scaling details). In both positive and negative ion modes, the PCA score plots revealed a clear separation among the sample groups (Fig. S3), with tight clustering of replicates within each stage. The groups were distinctly separated along the first principal component (PC1), which alone explained an exceptionally high percentage of the total variance (91.89% and 93.38%, respectively). This dominance of PC1 is characteristic of time-series studies like fermentation, where the progression of time itself is the single largest source of variation. Therefore, PC1 can be interpreted as representing the temporal trajectory of the fermentation process, capturing the large-scale, systematic changes such as the consumption of primary substrates and the accumulation of metabolic end-products. This strong, directional clustering along a single dominant axis confirms that the fermentation process induced profound and highly consistent time-dependent alterations in the ZRV metabolome, shaping a unique metabolic fingerprint for each stage. A comparison between the Early-LSF and Mid-LSF stages identified a total of 2159 differentially abundant metabolites (DAMs), with 1185 being significantly upregulated and 974 being significantly downregulated in the Mid-LSF stage (Fig. S4A). During the transition from Mid-LSF to Late-LSF, a total of 2262 DAMs were identified, with 633 being upregulated and a larger number (1629) being downregulated (Fig. S4B). Metabolic activity was particularly dynamic in the early phase of fermentation (from Early to Mid), characterized by a net accumulation of metabolites. In contrast, the later phase (from Mid to Late) was characterized by a net consumption or transformation of metabolites.

KEGG pathway enrichment analysis was performed (Fig. 6A). The results showed that pathways related to amino acid metabolism (e.g., valine, leucine, and isoleucine degradation; histidine metabolism) and organic acid metabolism (e.g., butanoate metabolism; TCA cycle) were significantly enriched. First, amino acid metabolism underwent drastic remodeling. The KEGG analysis highlighted the significant enrichment of pathways such as the degradation of valine, leucine, and isoleucine, as well as histidine metabolism. Specifically (Table S5), tryptophan and histidine, which were abundant in the early stage, decreased significantly. Concurrently, the levels of tyrosine and arginine continuously increased. Of particular interest were the degradation products of amino acids. For instance, α-ketoisovaleric acid, a typical α-keto acid derived from valine, surged in the Mid-LSF stage along with isovalerylglutamic acid (Fig. 6B, Table S5). Second, closely coupled with amino acid metabolism was the dynamic network of organic acids, which is central to both energy metabolism and flavor formation. The enrichment of pathways like the TCA cycle and butanoate metabolism, along with the aforementioned keto acids derived from amino acids, constituted this network. In terms of energy metabolism, the hub molecule pyruvate and the TCA cycle intermediate succinic acid both peaked in the Mid-LSF stage (Table S5). For succinic acid (a key TCA cycle intermediate), while its relative MS intensity in the untargeted metabolomics profile showed a transient peak in the Mid-LSF stage (Table S5), absolute HPLC quantification (Fig. 3B) confirmed its progressive accumulation throughout the fermentation, reaching its highest level in the Late-LSF stage. This variation highlights the necessity of absolute quantification, as relative ion intensities in non-targeted MS profiles can be affected by matrix effects during late-stage fermentation. Regarding flavor formation, the contents of butanoic and pentanoic acid continuously rose with fermentation. The most striking change was observed for another key α-keto acid, 2-ketobutanoic acid, whose content increased explosively in the Late-LSF stage to over seven times its initial level. Furthermore, pyrimidine metabolism was identified as the most significantly enriched pathway. From Early-LSF to Late-LSF, a decrease in uracil was accompanied by an increase in its downstream metabolites, cytidine and thymidine (Table S5).

Fig. 6.

Metabolic pathway analysis of differential metabolites during fermentation of ZRV. (A) Enrichment analysis plot, (B) integrated analysis of the 9 major metabolic pathways. Pathway analyses were based on differential metabolites identified from untargeted metabolomics (n = 3 independent biological replicates per liquid-state fermentation stage).

Beyond primary metabolism, the fermentation process also significantly impacted the profile of secondary metabolites, particularly flavonoids and alkaloids (Table S5). The relative abundances of some flavonoids, such as isovanillin and the bitter-tasting lactupicrin, increased significantly, especially in the Late-LSF stage. Conversely, other flavonoids like jaceoside and kuwanon E were more abundant in the Early-LSF stage and subsequently decreased. Similarly, the alkaloid profile was highly dynamic. Compounds such as trigonelline and 10-hydroxymorphine accumulated throughout the fermentation, while others like betonicine and methylecgonine peaked in the Mid-LSF stage before declining.

3.6. Correlation network highlights associations between microorganisms and the volatile/non-volatile flavor profiles

To elucidate the interactions among the microbial community, non-volatile metabolites, and volatile flavor compounds within the ZRV fermentation system, a multi-omics correlation network was constructed based on the LSF stages (n = 9) (Fig. 7). After strict BH-FDR correction, this network visually illustrates the strong associations of specific microorganisms with key precursors and characteristic flavors, suggesting their putative roles in flavor formation, revealing complex synergistic and competitive relationships. As shown in Fig. 7, both bacteria and fungi in the network established strong links with a variety of metabolites. Interestingly, Lactobacillus and its close relatives (Lactiplantibacillus, Limosilactobacillus) exhibited significant negative correlations with specific volatile esters (e.g., acetic acid propyl ester) and acetoin (r = −0.914 to −0.922, FDR-adjusted P < 0.05), but showed strong positive correlations with non-volatile markers like royal jelly acid. The role of Saccharomyces as a primary flavor contributor was corroborated, showing a strong positive correlation with ethanol (r = 0.820) and volatile alcohols like 3-methyl-1-pentanol-D. Acetic acid bacteria such as Acetobacter and Komagataeibacter were strongly associated with acidic metabolism, showing positive correlations with pyruvate (r = 0.774) and characteristic volatiles like diethyl succinate and 2,3-pentanedione. Furthermore, the network highlighted functional roles for other microorganisms: for instance, Monascus and Pseudomonas exhibited strong positive correlations with the key creamy-note compound acetoin (r = 0.776 and 0.835, respectively, FDR-adjusted P < 0.05), while Monascus was also strongly linked to the non-volatile metabolite tyrosine.

Fig. 7.

Multi-level bipartite correlation network mapping key microorganisms to volatile compounds and non-volatile metabolites in ZRV. The network was constructed based on Pearson correlation coefficients from 9 samples (3 LSF stages × 3 biological replicates). Edges represent significant cross-layer correlations (Microorganism–Volatile compounds and Microorganism–Non-volatile metabolites) after strict Benjamini-Hochberg false discovery rate (BH-FDR) correction (|r| ≥ 0.70, FDR-adjusted P < 0.05). Note: intra-layer correlations (e.g., volatile vs. non-volatile) are not depicted to specifically highlight microbial associations.

4. Discussion

The microbial succession observed in this study provides a robust framework for understanding quality formation in ZRV. The initial dominance of Arthrobacter (87.32% in Pre-SSF) likely contributes to the preliminary degradation of grain substrates (Yang et al., 2023), creating favorable conditions for subsequent microbes. The liquid fermentation phase is characterized by a clear functional handover: the rise of Saccharomyces (peaking at 66.91% in Late-SSF) is associated with the initial alcoholic fermentation, followed by the surge of Acetobacter (peaking at 30.61% in Mid-LSF), which likely contributed to the sharp increase in total acidity from 0.11 to 6.30 g/100 mL (Zhang et al., 2023). Notably, the dominance of Monascus throughout is critical. Given its known capacity to produce bioactive secondary metabolites like monacolin K and γ-aminobutyric acid (Gong et al., 2024; Husakova et al., 2025), its dramatic expansion to 88.05% in the Mid-LSF stage strongly correlates with the formation of ZRV's characteristic rosy color. This color development, evidenced by the decrease in lightness (L*) and the shift of the a* value to positive (Fig. 1C), is likely a combined result of Monascus pigments and the Maillard reaction between microbially-derived aldehydes and amino acids (Niu et al., 2023; Zhang et al., 2025).

During the terminal phase of liquid fermentation (Late-LSF), the bacterial community exhibited a unique shift, characterized by a substantial decrease in Acetobacter and an increase in Acinetobacter (Fig. 2B). This succession is driven by environmental stress and the mathematical nature of relative abundance. By day 150 (Late-LSF), the depletion of ethanol (the primary carbon source for Acetobacter) combined with extremely high acidity triggered the death or entry into a viable but non-culturable (VBNC) state of Acetobacter (Bartowsky & Henschke, 2008). Conversely, Acinetobacter species possess high resilience to nutrient starvation and organic acid toxicity, allowing their cells or DNA to persist longer in the late-stage mash (Li et al., 2025). Since high-throughput sequencing generates relative abundance data, the decline and lysis of the dominant Acetobacter mathematically inflated the relative proportion of Acinetobacter. In absolute terms, the total bacterial count at Late-LSF was extremely low (2.68 ± 0.73 lg CFU/mL, Fig. 2A), indicating that both genera existed in very low absolute quantities at the end of brewing.

Importantly, this dynamic microbial succession, including the early-stage self-purification and the late-stage bacterial shifts, is intimately coupled with the seasonal environmental shifts inherent to the traditional six-month brewing cycle of ZRV (spanning May to November) (Liu et al., 2022). The elevated ambient temperatures during late spring and summer (May to August) act as a thermal catalyst, accelerating starch saccharification by Monascus and alcohol fermentation by yeast during the SSF and early LSF stages (Huang et al., 2023). As the fermentation progresses into the cooler autumn months (September to November), the naturally declining ambient temperatures slow down the metabolic activity of Acetobacter, thereby preventing over-acidification and reducing the loss of delicate volatile esters (Jiang et al., 2026). This cooling phase also favors the growth of acid-tolerant Lactobacillus and Monascus, facilitating the accumulation of functional non-volatile metabolites. Thus, the long production period and seasonal fluctuations serve as macro-environmental drivers that direct the orderly succession of key functional microbes.

Building upon this ecological context, the formation of ZRV's complex flavor profile, where key aroma compounds (OAV ≥ 1) shape the sensory experience (Chen et al., 2024), is a dynamic process shaped by a complex interplay of microbial activities. Our correlation network (Fig. 7) provides evidence for these interactions by highlighting the significant statistical associations among key microbes and flavor compounds. Saccharomyces establishes the primary flavor framework by producing ethanol and a rich profile of fruity esters like ethyl acetate, which, with its massive OAV (up to 36,332), dominated the early fermentation stages (Sun et al., 2025). Furthermore, its correlation with higher alcohols is consistent with yeast's production of these important flavor compounds via amino acid metabolism, namely the Ehrlich pathway (Dickinson et al., 1998). As fermentation progresses, Acetobacter assumes its classic role, oxidizing ethanol to acetic acid. Most critically, other microbes add layers of complexity. For instance, the soaring concentration of acetoin (creamy, buttery notes), a key aroma compound whose OAV skyrocketed from 197 to over 4700, showed a strong positive correlation with the abundances of Monascus and Pseudomonas (Fig. 7, Table S6). This robust FDR-corrected association suggests that in the specific microbial terroir of ZRV, these genera, rather than lactic acid bacteria, may be the primary potential contributors to the development of these creamy, buttery notes through specific keto-acid pathways (Lee et al., 2016; Tian et al., 2020). The flavor evolution is dynamic, with the late-stage emergence of benzaldehyde (almond notes) from amino acid degradation (Russo et al., 2021) and pleasant baked/nutty aromas from pyrazines and furfural, which are typical products of the Maillard reaction or microbial metabolism (Wang et al., 2025).

The comprehensive resolution of this dynamic flavor framework was made possible by the synergistic application of GC–MS and GC-IMS. In flavor analysis, GC–MS is the gold standard for robust quantification (enabling OAV calculations) and identifying higher-boiling-point compounds. However, its limitation lies in the potential loss of highly volatile or trace compounds during sample pretreatment or due to lower sensitivity. Here, GC-IMS perfectly complemented GC–MS by operating under ambient pressure without complex pre-concentration, providing ultra-high sensitivity for trace volatiles (Wang et al., 2019). As demonstrated in our results, GC-IMS exclusively detected 28 unique trace compounds missed by GC–MS, particularly highly volatile aldehydes (e.g., 2-hexenal, undec-2-enal) and ketones (e.g., 1-octen-3-one). While their absolute concentrations were too low for standard GC–MS detection, these compounds typically possess extremely low odor thresholds. For example, 1-octen-3-one and specific aldehydes are known to impart delicate mushroom, green, and fatty nuances (Maseko et al., 2026). Therefore, although GC–MS outlined the primary aromatic macro-framework (e.g., the dominance of ethyl acetate and acetoin), the unique trace compounds detected by GC-IMS act as crucial “top-notes,” filling in the micro-sensory complexity of ZRV. Furthermore, regarding the common overlapping chemical classes, both methodologies exhibited highly congruent dynamic trends. Both the absolute quantification by GC–MS (Fig. 4A) and the signal intensity clustering by GC-IMS (Fig. 5B) confirmed the same successional pattern: a robust presence of esters in the early stages, shifting toward a complex matrix of specific ketones and aldehydes in the late stages. This cross-validation confirms that the combined GC–MS and GC-IMS approach provides a more holistic and multidimensional perspective on the flavor evolution of ZRV.

Beyond volatile aroma, the microbial community acts as a dynamic factory shaping the non-volatile profile, which dictates both the taste foundation and functional quality of ZRV. The peak in total flavonoid content in the Mid-LSF stage (Fig. 3A) is highly likely driven by microbial biotransformation rather than simple extraction from raw materials. Microbial metabolism can promote the release of bound polyphenols from grain cell walls and actively biotransform them. For instance, microbes like Lactobacillus and Rhizopus have been reported to catalyze the hydrolysis of flavonoid glycosides into free, more bioavailable aglycones via glycoside hydrolases (Jin et al., 2025). This is powerfully supported by our metabolomics data, where the consistent decrease of glycosides like jaceoside and lonicerin occurred as fermentation progressed. Crucially, our multi-omics correlation network (Fig. 7) revealed exceptionally strong associations between Lactobacillus and Monascus abundances and the accumulation of non-volatile organic acids and specific flavonoids. These dynamically changing bioactive components directly shaped the antioxidant profile of ZRV during fermentation. The distinct trends observed across antioxidant assays (Fig. 3D, E) reflect the specific chemical mechanisms involved in each measurement. Reducing power was significantly correlated with total polyphenols (r = 0.825, P < 0.05), explaining its parallel decline with polyphenol content. In contrast, the mid-stage peak in hydroxyl radical scavenging mirrored the peak in total flavonoids (r = 0.774, P < 0.05), highlighting the high efficacy of flavonoids in neutralizing reactive oxygen species. Notably, the stable DPPH and ABTS scavenging rates suggest a ‘compensation effect,’ where the accumulation of other antioxidant metabolites—such as specific amino acids (e.g., Tyr, Arg) and organic acids, as well as the formation of Maillard reaction products—effectively offsets the decrease in polyphenols, thereby sustaining the robust functional quality of ZRV throughout the fermentation process(Xu et al., 2007). The antioxidant contributions of these metabolites, particularly MRPs generated via microbial fermentation, have been widely documented in various food matrices(Zhao et al., 2021), providing a solid basis for our observation.

The metabolic processes potentially underlying the formation of both volatile and non-volatile flavor qualities were explored via untargeted metabolomics analysis. Organic acids (providing sourness and body) and free amino acids (providing umami and bitterness), sourced from raw materials and microbial proteolysis, form the core non-volatile taste profile (Shi et al., 2022). The significant enrichment of pathways related to branched-chain amino acid (BCAA) degradation (Fig. 6A) is consistent with a biochemical linkage between non-volatile precursors and volatile flavors. The consumption of amino acids like tryptophan and histidine provides essential nutrients, while the degradation of BCAAs likely generates a rich pool of intermediate molecules, such as α-keto acids (e.g., α-ketoisovaleric acid). While we lack isolated strain verification, our KEGG pathway evidence suggests that these non-volatile intermediates may be channeled into the synthesis of complex volatile compounds, including branched-chain alcohols and esters (Wang et al., 2024). Furthermore, the sharp increase in succinic acid and the seven-fold increase of 2-ketobutanoic acid in the late stage strongly suggest major metabolic shifts that coincide with the evolution of the final complex taste. The identification of pyrimidine metabolism as the most significantly enriched pathway indicates intense microbial cellular regulation characterizing the dynamic succession within the ecosystem (Xiong et al., 2022). Ultimately, it is this microbially-associated, interconnected metabolic network that appears to shape the volatile aroma and non-volatile taste profiles, defining the unique quality of ZRV.

While our multi-omics correlation networks construct a robust statistical framework linking microbial succession to flavor formation, transitioning from statistical association to exact metabolic causation remains a complex challenge. It is important to note that the correlation networks and matrices presented herein (including Fig. 2D, Fig. 3F, and Fig. 7) were constructed based on a limited number of samples (specifically, n = 9 across three LSF stages). Consequently, the observed relationships represent exploratory findings rather than definitive causal links. While we applied strict significance thresholds (P < 0.05 and ∣r∣ ≥ 0.7) to mitigate the risk of false discovery, these results should be interpreted as potential associations. Additionally, a limitation of this study is its reliance on a single representative commercial producer and a single production year. Given the previously established “microbial terroir” effect, where different ZRV factories exhibit distinct workshop environmental microbiomes and divergent fungal successional trajectories (Chang, Zhang, Fang, et al., 2025), the exact flavor profiles and metabolic rates may vary across production sites. Nevertheless, because the core bacterial and fungal taxa (such as Saccharomyces, Acetobacter, Lactobacillus, and Monascus) and the fundamental multi-stage biochemical handovers remain highly conserved across ZRV fermentations (Chang et al., 2025, Fang, Chai, Zhong and Jiang, 2021), the metabolic framework resolved here serves as a valuable baseline model. Therefore, future studies employing multi-factory and multi-year comparative metabolomics, targeted metatranscriptomics, isotope tracing, and the in vitro functional validation of isolated synthetic microbial communities (SynComs) are warranted to assess the generalizability of these metabolic flows across different terroirs and confirm the exact causal enzymatic contributions of these key taxa.

5. Conclusion

In conclusion, this study characterizes the statistical associations between microbial succession and the dynamic formation of volatile and non-volatile flavor profiles in ZRV. We identified distinct successional patterns that correlate with flavor development. For volatile quality, the presence of Saccharomyces and Acetobacter was associated with the primary ester and acid aroma framework, while the specific abundances of Monascus and Pseudomonas were statistically linked to the surge in the creamy-note compound acetoin. Regarding non-volatile quality, the dynamic shifts in primary metabolism—particularly the degradation of amino acids and the accumulation of organic acids (e.g., succinic and 2-ketobutanoic acids)—coincided with the final taste profile and likely contributed to the pool of precursors for volatile synthesis. Furthermore, multi-omics correlation networks suggest that the specific proliferation of Lactobacillus and Monascus is closely associated with the accumulation of non-volatile bioactive compounds (such as specific flavonoids) and the observed antioxidant capacity. By integrating microbiomics and comprehensive metabolomics, this study shifts the understanding of ZRV from empirical observation to a holistic statistical level. These findings provide a robust scientific foundation for future targeted process optimization and the biological control of volatile aromas and non-volatile taste attributes in traditional fermented foods.

CRediT authorship contribution statement

Guoli Chang: Writing – original draft, Visualization, Software, Methodology, Investigation, Formal analysis, Data curation. Shenchenyu Zhang: Software, Investigation. Xinpei Wang: Investigation, Data curation. Guanyu Fang: Resources, Investigation. Junli Zhu: Writing – review & editing, Supervision, Resources, Funding acquisition, Conceptualization. Zihan Qin: Writing – review & editing, Supervision, Funding acquisition.

Declaration of competing interest

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

Appendix A. Supplementary data

Supplementary material

Data availability

Data will be made available on request.