Divergent metabolic shifts of lactic acid bacteria strains shape the quality and predicted antioxidant mechanisms of a fermented fish–rice matrix (Yucha) Authors: Jiaqi Wang, Xiaojing Li, Jiaxin Shi, Binbin Wang, Qiwei Liu, Junli Feng, Changlong Liu, Jing Xue, Guofang Shen, Miaoyu Li, Qing Shen Published: 2026 ABSTRACT AbstractThe functional potential of fish–rice co-fermented high-protein matrices remains difficult to harness because of the unpredictability of spontaneous fermentation. This study employed metabolomics and network pharmacology to decode the metabolic reprogramming orchestrated by three lactic acid bacteria strains. Inoculation rapidly stabilized physicochemical properties and mitigated lipid oxidation. Metabolomic profiling revealed distinct strain-specific metabolite profiles: YC202 was associated with differences in carbohydrate-related metabolites and the antioxidant dipeptide carnosine; YDN29 showed higher abundances of umami-associated amino acids; and FY27 showed higher abundances of selected polyunsaturated fatty acids. In vitro antioxidant assays further showed that LAB inoculation increased chemical antioxidant capacity. Bridging these chemical phenotypes to biological functions, network pharmacology integrated 129 bioactive metabolites with 421 oxidative damage-associated targets. The in silico analysis suggested that the identified metabolites may be associated with antioxidant-related PI3K-Akt and MAPK signaling pathways. These findings may support rational LAB selection for standardized fermented aquatic foods.Keywords: Yucha, Lactic acid bacteria, Untargeted metabolomics, Antioxidant activity, Network pharmacology, Food safety Introduction Yucha, a traditional fermented aquatic food prevalent in southern China and Southeast Asia, is typically prepared by co-fermenting freshwater fish with cooked rice (Z. Luo et al., 2026; Rodzi & Lee, 2021). Unlike single-substrate fermentations, yucha represents a highly complex dual-substrate (protein-carbohydrate) matrix. During the prolonged fermentation process, the dynamic succession of endogenous microorganisms and endogenous enzymatic activities synergistically degrade complex proteins and lipids into low-molecular-weight flavor compounds, conferring yucha with its characteristic umami taste, appealing texture, and extended shelf life (Hu et al., 2020; Hu et al., 2022). Beyond its culinary value, yucha is rich in essential amino acids, active peptides, and beneficial fatty acids, making it a promising dietary source of natural antioxidants and functional nutrients (Han et al., 2020). Despite its popularity, the traditional production of yucha relies predominantly on spontaneous fermentation driven by naturally occurring microbiota from the raw materials and environment (Zhang et al., 2016). This uncontrolled process has several drawbacks, including prolonged fermentation cycles, inconsistent sensory quality, and critical safety risks. Most notably, the proliferation of undesirable spoilage microorganisms during natural fermentation can trigger severe lipid oxidation, textural deterioration, and the excessive accumulation of total volatile basic nitrogen (TVB-N), hazardous biogenic amines, and nitrites (Zhao et al., 2021). To overcome these industrial bottlenecks, the application of selected starter cultures has emerged as a highly effective strategy (Zheng et al., 2024). Lactic acid bacteria (LAB), widely classified as generally recognized as safe (GRAS) microorganisms, are the ideal candidates. LAB can rapidly lower the environmental pH through prolific lactic acid production, thereby suppressing pathogenic competitors, ensuring product safety, and actively participating in the biotransformation of flavor and functional precursors (W. Luo et al., 2023). However, the exact metabolic pathways and the specific chemical footprints orchestrated by different LAB strains within the complex fish-rice matrix remain largely underexplored (Zhou et al., 2026). Furthermore, while recent studies have documented the potential antioxidant properties of fermented aquatic products, the underlying molecular mechanisms remain poorly understood (Lv et al., 2025; Sun et al., 2025). Traditional evaluation methods often focus on macroscopic free-radical scavenging assays in vitro, which are fundamentally inadequate for capturing the complex in vivo interactions of a multi-component food matrix (Lang et al., 2024). Network pharmacology, an emerging bioinformatics paradigm, bridges this gap by systematically mapping the “multi-component, multi-target, multi-pathway” interaction networks. It has proven to be a powerful tool in deciphering the holistic pharmacological mechanisms of complex natural products and functional foods (Shen et al., 2024). Therefore, this study aimed to evaluate the targeted quality improvement and to decode the antioxidant mechanisms mediated by three high-performing LAB strains (Lacticaseibacillus paracasei YC202, Latilactobacillus sakei YDN29, and Lactiplantibacillus plantarum FY27) isolated from commercial yucha. We systematically monitored the physicochemical parameters, textural properties, and safety indicators during fermentation. Crucially, to comprehensively map the intricate metabolic flux, ultra-performance liquid chromatography coupled with Q-Exactive Orbitrap high-resolution mass spectrometry (UPLC-Q-Exactive Orbitrap-MS/MS) was employed. Its exceptional mass accuracy and resolving power are vital for precisely annotating complex metabolites in high-protein matrices, thereby providing highly reliable chemical foundations for subsequent predictions (Huang et al., 2021). Ultimately, a network pharmacology approach was pioneered to decipher the core therapeutic targets and signaling pathways responsible for the antioxidant efficacy of the identified bioactive metabolites. This multifaceted study is anticipated to provide robust theoretical guidance for the targeted screening of starter cultures, serving as a novel paradigm for the industrial modernization of highly functional and standardized fermented high-protein foods. Materials and methods Culture propagation For strain isolation, yucha was homogenized in PBS, serially diluted, and spread on MRS agar. After incubation at 37 °C for 48 h, colonies with distinct morphologies were purified by repeated streaking and stored at −80 °C. PCR amplification of the groEL gene was carried out on genomic DNA extracted from every isolate, employing the specific primers groEL F (3’-GCYGGTGCWAACCCNGTTGG-5′) and groEL R (3’-AANGTNCCVCGVATCTTGTT-5′). The amplicons were then sequenced, and the obtained sequences were compared against the NCBI GenBank reference database using BLAST to determine the species affiliation. These strains were selected as representative candidate starters based on their acid and bile-salt tolerance (Table. S1), as well as their growth characteristics (Fig. S1), which indicated their potential adaptability and functional applicability in fermented-food systems. Strains stored at −80 °C were activated twice in MRS broth at 37 °C for 24 h. The cultures were centrifuged at 8000 ×g for 15 min, washed twice. Before inoculation, viable cell concentrations were determined by serial dilution and plate counting on MRS agar. Based on the viable counts of the bacterial suspensions, the inoculation volume for each strain was calculated according to the total mass of the fish-rice mixture to obtain a target initial level of 107 CFU/g. Fermentation process Fresh grass carp (Ctenopharyngodon idella) were eviscerated, cleaned, and cut into uniform pieces. Food-grade mined salt (rock salt) was used to prepare the 7% (w/w) brine. The fish pieces were marinated in 7% (w/w) brine at 20 °C for 3 h. Lowland rice was cooked at a rice-to-water ratio of 1:2 (w/v), after which the cooked rice and marinated fish were combined at a ratio of 140:80 (w/w), corresponding to a cooked-rice-to-marinated-fish ratio of 1.75:1. This ratio was maintained consistently for every fermentation jar across all treatments and biological replicates. The materials were then alternately layered in jars. Four fermentation treatments were prepared: a spontaneous-fermentation control without starter inoculation (Con), and three starter-inoculated groups containing Lacticaseibacillus paracasei YC202, Latilactobacillus sakei YDN29, or Lactiplantibacillus plantarum FY27. Each starter culture was inoculated at 107 CFU/g and distributed evenly across the fish-rice layers. All jars were incubated at 20 °C for 28 days. For each treatment and sampling time point, three independent fermentation jars were prepared, with each jar serving as one biological replicate (n = 3). Samples were collected on days 0, 7, 14, 21, and 28, sealed in sterile bags, and stored for analysis. Determination of pH value The pH of the yucha samples was measured using a pH meter (Leici, China). A 2.0 g fish sample was homogenized with 18 mL of ultrapure water for 2 min and left to stand for 10 min. The pH of the supernatant was then measured. Determination of texture Fish samples (1.5 cm × 1 cm × 1 cm, connective tissues removed) were analyzed using a TMS-Pro texture analyzer (FTC, USA). Hardness, cohesiveness, springiness, and chewiness were determined via texture profile analysis (TPA) using a TA25 probe. Parameters were: test/return speed, 1.0 mm/s; compression deformation, 50%; and compression distance, 8 mm. Samples were measured in triplicate. Safety and freshness index testing The level of thiobarbituric acid reactive substances (TBARS) was quantified following the method described by (Ahmed et al., 2025), with slight modifications. To evaluate lipid oxidation, 2.0 g of sample was homogenized with 20 mL of 7.5% trichloroacetic acid for 5 min and centrifuged (8000 ×g, 10 min). The supernatant was mixed with 0.02 mol/L thiobarbituric acid (TBA) solution, incubated at 90 °C for 30 min, and cooled. The absorbance was measured at 532 nm using a UV–Visible spectrophotometer (HACH, China). Results were expressed as mg malondialdehyde (MDA)/kg. Total volatile basic nitrogen (TVB-N) was determined via the semi-micro Kjeldahl method (GB 5009.228–2016). Samples underwent alkaline distillation; released volatile bases were absorbed in boric acid and titrated with standard hydrochloric acid. Results were expressed as mg N/100 g. Nitrite content was determined using the spectrophotometric method (GB 5009.33–2016, Method 1). Sample nitrite reacted with sulfanilamide under acidic conditions to form a diazonium compound, which coupled with N-(1-naphthyl) ethylenediamine dihydrochloride to produce a pink azo dye. Absorbance was measured, and results were expressed as mg/kg. Sensory evaluation Sensory analysis was performed by ten trained panelists (5 males and 5 females, aged 20–30 years) following a slightly modified version of the method (Yu et al., 2025). Before formal evaluation, the panelists were familiarized with the sensory attributes and scoring criteria used in this study. Each panelist independently evaluated the samples for colour, odour, taste, and texture. The maximum scores for colour, odour, taste, and texture were 20, 35, 35, and 10 points, respectively, giving a total possible sensory score of 100 points. The overall sensory score for each sample was calculated as the sum of the four attribute scores and is reported as the mean score of the ten panelists. Formal ethical approval was not required per Zhejiang Gongshang University regulations. Measurement of in vitro antioxidant activity DPPH radical scavenging capacity The DPPH radical-scavenging activity of the yucha extracts was determined with reference to (Hao et al., 2026), with minor modifications. Briefly, 1 mL of the centrifuged yucha extract (100 mg·mL−1) was diluted to 50 mL with absolute ethanol. Subsequently, 100 μL of the diluted extract was mixed with 100 μL of DPPH solution (0.02 mmol·L−1) in a 96-well microplate. The plate was incubated at room temperature for 30 min in the dark, and the absorbance was then measured at 517 nm. The DPPH control was prepared by mixing 100 μL of absolute ethanol with 100 μL of DPPH solution. To correct for the intrinsic colour of the extract, a sample blank containing 100 μL of diluted extract and 100 μL of absolute ethanol was included. Trolox standard solutions were reacted with DPPH under the same conditions to generate a calibration curve. The DPPH radical-scavenging capacity of each sample was calculated from its Trolox-equivalent value after blank correction and adjustment for the dilution factor and dry weight of the original yucha sample. Results were expressed as mg Trolox equivalents per 100 g dry weight (mg Trolox/100 g DW). ABTS radical scavenging capacity The ABTS scavenging activity was assessed through a modified decolorization assay (Song et al., 2025). The antioxidant potential of the extracts was assessed by their ability to reduce the ABTS radical cation generated from a reaction between ABTS (7 mM) and potassium persulfate (2.45 mM). The mixture was kept in the dark for 30 min to allow the formation of ABTS. The ABTS solution was diluted with PBS buffer (pH 7.04) to achieve an absorbance of 0.7 ± 0.02 nm at 734 nm. A negative control containing 50 μL of PBS buffer and 950 μL of ABTS solution, as well as a sample blank containing 50 μL of sample and 950 μL of PBS buffer, were included to correct for the intrinsic absorbance of the extracts. A standard curve was generated using varying concentrations of Trolox. Results were expressed as mg Trolox equivalents per 100 g dry weight (mg Trolox/100 g DW). Ferric-reducing antioxidant power (FRAP) assay The ferric-reducing capacity of the extracts was evaluated according to previously reported procedures (Guo et al., 2025). Briefly, 1 mL of sample extract was combined with 2.5 mL of 0.2 M phosphate buffer (pH 6.6) and 2.5 mL of 1% potassium ferricyanide. After incubation at 50 °C for 30 min, 2.5 mL of 10% trichloroacetic acid was added, and the mixture was centrifuged at 8000 ×g for 10 min. Subsequently, 2.5 mL of the upper phase was mixed with 2.5 mL of deionized water and 0.5 mL of 0.1% ferric chloride solution. The absorbance was measured at 700 nm. A Trolox calibration curve was generated using serially diluted Trolox solutions, and the results were expressed as mg Trolox equivalents per 100 g dry weight (mg Trolox/100 g DW). UPLC-Q-exactive orbitrap-MS/MS analysis Sample preparation Untargeted metabolomics analysis was performed using yucha samples collected on day 21 of fermentation. Yucha samples (∼25 mg) were mixed with extraction solvent (MeOH:ACN:H₂O = 2:2:1, v/v/v) and homogenization beads containing isotopically labeled internal standards. The homogenized composite, rather than fish tissue or rice analyzed separately, was used for metabolite extraction and subsequent UHPLC-MS/MS analysis. The mixture was homogenized (35 Hz, 4 min) and ultrasonicated in an ice-water bath (5 min); this process was repeated thrice. After incubation at −40 °C for 1 h, samples were centrifuged. The supernatant (300 μL) was filtered through a 96-well plate under positive pressure (6 psi, 3 min). Quality control (QC) samples were prepared by pooling equal aliquots of all sample supernatants. In the metabolomics analysis, “Blank” refers to the unfermented fish-rice mixture collected on day 0 prior to fermentation. “Con” refers to the spontaneous-fermentation control group, which was prepared without starter-culture inoculation. The three inoculated groups were designated YC202, YDN29, and FY27 according to the respective starter strains. UPLC-Q-exactive orbitrap-MS/MS conditions for metabolomics Untargeted metabolomic analysis of yucha samples was performed using a Vanquish ultra-high-performance liquid chromatography (UHPLC) system (Thermo Fisher Scientific, USA). Chromatographic separation was achieved on a Waters ACQUITY UPLC BEH Amide column (2.1 mm × 50 mm, 1.7 μm). The mobile phase consisted of phase A, water containing 25 mmol/L ammonium acetate and 25 mmol/L ammonia, and phase B, acetonitrile. The injection volume was 2 μL, and the autosampler temperature was maintained at 4 °C. The gradient elution conditions were as follows: 0–0.25 min, 95% B; 0.25–3.5 min, 95–65% B; 3.5–4.0 min, 65–40% B; 4.0–4.5 min, 40% B; 4.5–4.55 min, 40–95% B; 4.55–6.0 min, 95% B. Mass spectrometric detection was carried out using an Orbitrap Exploris 120 high-resolution mass spectrometer (Thermo Fisher Scientific, USA) controlled by Xcalibur software (version 4.4) for both MS and MS/MS data acquisition. The ion source parameters were set as follows: sheath gas flow rate, 50 Arb; auxiliary gas flow rate, 15 Arb; and capillary temperature, 320 °C. The full-scan resolution was 60,000, while the MS/MS resolution was 15,000. Fragmentation was performed using stepped collision energies of 20, 30, and 40 eV. The spray voltage was set to 3.8 kV in positive ion mode (ESI+) and − 3.4 kV in negative ion mode (ESI-). Statistical data analysis Raw data were converted to mzXML format using ProteoWizard (version 3.0.24054). Metabolite annotation was performed using a R package, with BiotreeDB (version 3.0; Biotree, China) as the reference database. The detected features were compared with the authentic-standard library in BiotreeDB based on accurate mass, retention time, adduct information, and MS/MS fragmentation spectra, according to the annotation criteria provided by Biotree. Only features assigned as Level 1 by the BiotreeDB annotation system were retained for subsequent analyses. Accordingly, the metabolites included in the differential-metabolite analysis, KEGG enrichment analysis, and network pharmacology analysis were referred to as high-confidence, Level 1-annotated metabolites matched against the BiotreeDB standard library. Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were performed using MetaboAnalyst 6.0. Model performance was evaluated using five-fold cross-validation, with 1000 permutation test. Differential metabolites were screened based on VIP > 1, p < 0.05, and |log₂FC| ≥ 1. KEGG enrichment analysis was conducted in R (v4.3.1) using the clusterProfiler package. Network pharmacology Prediction of potential active compounds To accurately identify the core bioactive constituents from the untargeted metabolomics dataset, a rigorous three-step screening pipeline was established. First, the differentially accumulated metabolites (DAMs) were prioritized and filtered based on their positive annotations in the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, ensuring the selected components possessed established biological relevance. Second, a matrix-specific manual curation was meticulously conducted. This critical step served to eliminate instrumental false positives and retain only those compounds that are chemically and biologically plausible within the complex yucha matrix (a fermented freshwater fish and rice system). Finally, these highly confident candidate metabolites were subjected to in silico pharmacokinetic validation via Absorption, Distribution, Metabolism, and Excretion (ADME) predictions using the SwissADME web server (http://www.swissadme.ch/). Only compounds demonstrating favorable profiles—specifically those exhibiting ‘High’ gastrointestinal (GI) absorption and satisfying at least two standard drug-likeness criteria (Lipinski, Ghose, Veber, Egan, and Muegge)—were definitively selected as the core bioactive compounds for subsequent target prediction and network construction. Candidate compounds were further subjected to target prediction analysis using SwissTargetPrediction (http://swisstargetprediction.ch/), with target proteins exhibiting a predicted probability >0.15 selected as targets for subsequent studies. Prediction of disease targets Genes related to “oxidative damage” were retrieved from the OMIM (https://www.omim.org/) and GeneCards databases (https://www.genecards.org/, relevance score ≥ median value). Overlapping targets between active compounds and disease genes were identified via Venn analysis. PPI network construction and the “bioactives–targets” network analysis Shared targets were identified using Venny 2.1.0. Protein-protein interactions (PPI) of these targets were analyzed using STRING (human species, confidence score > 0.9, isolated nodes hidden). Cytoscape 3.8.0 was used for topological analysis based on median values of six parameters (degree, betweenness, closeness, LAC, network centrality, and eigenvector centrality). Targets remaining after three screening rounds were designated as core regulatory nodes. The “Strain–metabolite–target” network was also visualized in Cytoscape 3.8.0. GO and KEGG enrichment analysis GO and KEGG enrichment analyses were conducted using the R package clusterProfiler (org.Hs.eg.db annotation). Significance was set at q < 0.05 following Benjamini-Hochberg correction. Data processing and analysis All experiments were performed in triplicate and the results were expressed as mean ± standard deviation (SD). One-way ANOVA with Tukey's multiple comparison test was applied to assess statistical significance (p < 0.05) for cross-sectional comparisons (among groups at identical times) and longitudinal variations (within the same group over time). Results Physicochemical analysis of fermented yucha pH and textural properties of yucha While the pH of the naturally fermented control (Con group) fluctuated and remained relatively high over the first 21 days, the LAB-inoculated groups exhibited a distinct and rapid decreasing trend during fermentation (Fig. 1A, Table S2). Overall, the LAB treatments maintained significantly lower pH values than the control from day 21 onwards. Specifically, the L. paracasei (YC202) and L. sakei (YDN29) groups displayed similar acidification patterns, dropping to around 3.8 on day 21 (control: 4.46 ± 0.08) and approximately 3.5 on day 28 (control: 3.97 ± 0.03). The L. plantarum group (FY27) showed slightly higher pH values of 4.09 ± 0.08 and 3.69 ± 0.03 on days 21 and 28, respectively. Physicochemical properties of yucha. (A) pH; (B) Springiness; (C) Chewiness; (D) Hardness; (E) TBARS; (F) TVB-N; (G) Nitrite content; (H) Sensory evaluation. Data are presented as mean ± SD (n = 3). Different uppercase letters denote significant differences across fermentation times, and different lowercase letters denote significant differences within the same fermentation time (p < 0.05). Textural properties, including springiness (Fig. 1B, Table S5), chewiness (Fig. 1C, Table S3), and hardness (Fig. 1D, Table S4), were also significantly influenced by LAB inoculation. Although the springiness of the Con group drastically deteriorated after 14 days, the LAB-inoculated groups maintained significantly superior springiness, chewiness, and hardness throughout the fermentation (p < 0.05). This suggests that specific LAB inoculation effectively prevents textural softening and preserves the structural integrity of yucha. Determination of freshness indicators TBARS and TVB-N are crucial indicators of lipid oxidation and protein decomposition, respectively. The TBARS value in the Con group showed a sharp peak at day 14 (Fig. 1E, Table S6), indicating severe early-stage lipid oxidation, and ultimately exceeded the Chinese national safety standard (≤ 2.5 mg/kg, GB 10136–2015) by day 14. In contrast, the LAB-inoculated groups effectively suppressed TBARS accumulation, maintaining significantly lower levels that remained strictly within the safety limit throughout the entire fermentation period (p < 0.05). Furthermore, the TVB-N value (Fig. 1F, Table S7) increased continuously in the Con group, reaching its maximum at day 28. Its TVB-N contents exceeded the national safety standard (≤ 30 mg/100 g, GB 10136–2015) on both day 21 (30.03 ± 0.38 mg/100 g) and day 28 (30.95 ± 0.91 mg/100 g). Conversely, all LAB-fermented groups exhibited significantly lower TVB-N contents (p < 0.05) and remained well below the national limit. Notably, the YC202 and YDN29 groups showed a transient decrease at day 21 followed by a slight increase at day 28, with the YC202 group demonstrating the strongest inhibitory effect on TVB-N accumulation after 14 days. This trend aligns seamlessly with their rapid pH reduction (Fig. 1A), which may reflect an altered fermentation environment that is less favorable for volatile‑nitrogen accumulation. Assessment of safety and sensory properties Nitrite content, a critical safety parameter, exhibited a continuous degradation dynamic during fermentation (Fig. 1G, Table S8). While nitrite levels in the control group decreased gradually, LAB inoculation significantly accelerated this degradation process (p < 0.05). Notably, nitrite in the inoculated samples was almost completely depleted by day 21. This reduction coincided with acidification during fermentation. However, because nitrite-reductase activity and the contribution of acid-mediated nitrite degradation were not directly determined, the mechanism underlying the accelerated nitrite decrease requires further investigation. Throughout the entire process, nitrite content remained well below the national limit of 20 mg/kg. Ultimately, these physicochemical and safety improvements translated into superior sensory qualities. Sensory evaluation (Fig. 1H) demonstrated that LAB-inoculated yucha consistently achieved higher overall acceptability scores compared to the natural control, achieving optimal maturation and palatability between days 21 and 28. Among the evaluated strains, the L. paracasei group (YC202) exhibited the most outstanding sensory quality, outperforming both the L. sakei (YDN29) and L. plantarum (FY27) groups. Antioxidant capacity of yucha The antioxidant activities of the fermented yucha samples are shown in Fig. 2 and Table S9. The DPPH radical-scavenging activity increased from 256.83 ± 0.69 mg Trolox/100 g DW in the control group to 280.88 ± 0.52, 283.34 ± 0.20, and 287.31 ± 0.22 mg Trolox/100 g DW in the YC202, YDN29, and FY27 groups, respectively (Fig. 2A). Each LAB-inoculated group showed a significantly higher DPPH value than the control group, with the FY27 group exhibiting the highest activity. In vitro antioxidant activities of yucha samples fermented with different LAB starters at day 21. (A) DPPH radical-scavenging activity; (B) FRAP capacity; (C) ABTS radical-scavenging activity. The bars represent the mean ± SD of three independent fermentation jars per treatment (n = 3). Different lowercase letters indicate significant differences among treatments (p < 0.05). A similar pattern was observed for FRAP (Fig. 2B). The reducing capacity increased from 205.92 ± 5.88 mg Trolox/100 g DW in the control group to 255.32 ± 2.82, 277.23 ± 0.96, and 281.96 ± 1.69 mg Trolox/100 g DW in the YC202, YDN29, and FY27 groups, respectively. The YDN29 and FY27 groups showed the highest FRAP values, and their values were not significantly different from each other, as indicated by the same lowercase letter. The ABTS radical-scavenging activity also increased significantly following LAB inoculation (Fig. 2C). The values were 120.60 ± 0.80, 149.74 ± 0.64, 156.65 ± 0.30, and 160.48 ± 0.37 mg Trolox/100 g DW for the Con, YC202, YDN29, and FY27 groups, respectively. Thus, FY27 showed the strongest ABTS scavenging activity, followed by YDN29 and YC202. Overall, LAB inoculation was associated with enhanced radical-scavenging and ferric-reducing capacities in fermented yucha. The relatively larger increases in FRAP and ABTS activity compared with DPPH activity suggest that the starter strains may have altered the abundance or composition of electron-donating and radical-scavenging compounds. Non-targeted metabolomic profiling of fermented yucha Identification and classification of metabolites To comprehensively investigate the chemical composition changes in yucha driven by different fermentation strategies, untargeted metabolomics via UPLC-Q-Exactive Orbitrap-MS/MS was performed in both positive (POS) and negative (NEG) ion modes. As depicted in the total ion chromatograms (Fig. 3A), observable variations in peak intensities and retention times were present among the groups, suggesting significant metabolic shifts induced by LAB inoculation. The identified metabolites were further categorized, and their class distribution is summarized in Fig. 3B. Across all samples (Blank, Control, YC202, YDN29, and FY27), amino acids and their derivatives, alongside lipids, constituted the most dominant chemical classes in terms of metabolite counts, followed by phenylpropanoids, carbohydrates, steroids and steroid derivatives, organic acids and derivatives. Although the absolute counts of identified metabolites within these classes varied slightly among the treatments, the overall structural composition of the metabolome remained generally consistent across the groups. Statistical analysis of yucha metabolomics. (A) Chromatograms in positive and negative ion modes; (B) Bar chart showing the distribution of metabolite class counts. A, Blank; B, Control; C, YC202; D, YDN29; E, FY27; (C) Heatmap of metabolite classes based on normalized peak areas; (D) Pairwise PCA score plot matrix of metabolites; (E) PLS-DA scores plot of metabolites. Metabolic shifts and multivariate statistical analysis While the categorical distribution of metabolites was similar, their relative abundances underwent profound alterations during the fermentation process, as visualized in the heatmap clustering based on normalized peak areas (Fig. 3C). The unfermented raw material (Blank) exhibited a higher relative abundance of amino acids and derivatives, as well as carbohydrates. After spontaneous fermentation, the metabolic profile shifted, characterized primarily by an accumulation of lipids and phenylpropanoids. In contrast, inoculation with specific LAB strains (L. paracasei YC202, L. sakei YDN29, and L. plantarum FY27) markedly promoted the accumulation of organic acids and derivatives, alongside steroids and steroid derivatives. This distinct transition reflects intense microbial metabolic activities, indicating that the inoculated LAB strains rapidly utilize raw material components and convert them into specific flavor-active and functional metabolites. To further decipher the global metabolic disparities and natural clustering behaviors among the different groups, multivariate statistical analyses were employed. The pairwise principal component analysis (PCA) score plot matrix (Fig. 3D) demonstrated significant spatial separation among the groups across multiple principal components, confirming that the metabolic profiles were fundamentally remodeled by the different fermentation treatments. This robust separation is further supported by the supplementary PCA scores plot (Fig. S2A), with the corresponding scree plot indicating that the first two principal components capture 51.5% and 21.3% of the total variance, respectively (Fig. S2B). This divergence was further validated by the partial least squares discriminant analysis (PLS-DA) model (Fig. 3E). In the PLS-DA scores plot, Component 1 (accounting for 67.1% of the variance) effectively separated the Blank and Control groups from the LAB-inoculated groups, indicating that LAB inoculation is the primary driver of metabolic variation. Furthermore, the three LAB-inoculated groups (FY27, YC202, and YDN29) clustered tightly within their respective groups. The FY27 group was clearly separated from the other two LAB-fermented groups, whereas YC202 and YDN29 showed a high degree of overlap, suggesting that their fermentation profiles may be more similar in terms of compound composition. To ensure the reliability and predictive capability of the classification, the PLS-DA model was rigorously validated. The model containing three components showed the highest predictive performance, with an R2 value of 0.99849 and a Q2 value of 0.98479 (Fig. S2C). Furthermore, a permutation test (n = 1000) yielded an observed statistic p = 0.001 (Fig. S2D), confirming that the model is highly significant and not overfitted. This clear segregation powerfully demonstrates that different LAB strains possess distinct metabolic capacities, ultimately driving unique metabolic trajectories and shaping the specific chemical fingerprints of the fermented yucha. Differential metabolite analysis of variously fermented yucha To further elucidate the specific metabolic alterations driven by different LAB strains, differentially accumulated metabolites (DAMs) were comprehensively screened based on the stringent criteria of VIP > 1, p < 0.05, and |log₂FC| ≥ 1. Global heatmap analysis A hierarchical clustering heatmap of the top 50 DAMs across all groups (Fig. 4A) provides a global perspective on these metabolic shifts. It is visually evident that the overall metabolic profiles of the LAB-inoculated groups underwent profound reprogramming compared to the unfermented Blank and spontaneously fermented Control. Furthermore, the heatmap reveals distinct, strain-specific expression patterns among the three LAB treatments, with unique clusters of metabolites being highly accumulated in each specific inoculated group. Differential metabolite profiles. (A) Heatmap of the top 50 differential metabolites. (B) Volcano plot and heatmap of differential metabolites (top 15) between control and YC202-fermented yucha. (C) Control vs. YDN29-fermented yucha. (D) Control vs. FY27-fermented yucha. Differential metabolites were screened by VIP > 1, p < 0.05, and |log₂FC| ≥ 1. Pairwise comparison and strain-specific metabolic signatures To pinpoint the precise metabolic contributions of each strain, pairwise comparisons between the natural Control and each LAB-inoculated group were visualized using volcano plots and heatmaps of the top 15 signature metabolites (Fig. 4B–D). A consistent trend across all inoculated groups was a substantially higher number of upregulated metabolites than downregulated ones, highlighting the robust metabolic productivity of the introduced LAB strains. Specifically, in the L. paracasei YC202 group (Fig. 4B), out of 1046 differential metabolites, 754 were significantly upregulated. The top signature metabolites included critical intermediates such as α-ketoglutaric acid (alpha-KG) and galactose, suggesting enhanced carbohydrate utilization and tricarboxylic acid (TCA) cycle activity during YC202 fermentation. In contrast, the L. sakei YDN29 group (Fig. 4C; 696 upregulated vs. 308 downregulated) was uniquely characterized by a pronounced accumulation of free amino acids and their derivatives, such as sarcosine, glycine, asparagine, and serine. This distinct signature strongly indicates that the YDN29 strain possesses exceptional proteolytic capabilities, accelerating protein degradation into small-molecule amino acids, which critically contribute to the umami and sweet flavor profiles of yucha. Notably, the L. plantarum FY27 group (Fig. 3D; 595 upregulated vs. 282 downregulated) exhibited a highly functional metabolic footprint. Key bioactive and functional metabolites were prominently upregulated in this group, most notably pyridoxal (vitamin B6) and specific flavor precursors like aspartate and butanoic acid. Collectively, these differential metabolite profiles powerfully demonstrate that while all LAB strains accelerate the fermentation process, they orchestrate the metabolic flux towards completely divergent pathways: YC202 towards active carbohydrate metabolism, YDN29 towards profound amino acid accumulation for flavor enhancement, and FY27 towards the biosynthesis of high-value functional compounds like α-ketoglutaric acid. KEGG pathway enrichment analysis of differential metabolites To deeply understand the biological functions and metabolic mechanisms underlying the metabolomic alterations induced by different LAB strains, KEGG pathway enrichment analysis was performed on the identified DAMs, accompanied by the relative quantification of core metabolites within key pathways (Fig. 5). KEGG enrichment analysis of differential metabolites and peak area of metabolites in key pathways. (A) YC202. (B) YDN29. (C) FY27. Data are presented as mean ± SD (n = 3). As shown in the KEGG enrichment bubble chart (Fig. 5A), the DAMs in the YC202-inoculated group were prominently enriched in amino acid and carbohydrate metabolism. Relative quantification of selected metabolites further characterized these pathway-associated differences. In arachidonic acid metabolism (Fig. 5A-a), YC202 fermentation was associated with significantly higher relative abundances of prostaglandin J2 and 15-deoxy-PGJ2, whereas prostaglandin A2 showed a markedly lower abundance compared to the control. In histidine metabolism (Fig. 5A-b), the relative abundances of carnosine and imidazole-4-acetaldehyde were remarkably increased, whereas 1-methylhistamine decreased. Furthermore, YC202 was associated with differences in metabolites related to glycerophospholipid metabolism (Fig. 5A-c) by significantly accumulating glycerophosphoethanolamine and glycerophosphocholine, accompanied by a reduction in acetylcholine. Regarding carbohydrate utilization (Fig. 5A-d), the levels of cellobiose and trehalose were significantly lower than the natural control, while maltose was substantially more abundant, indicating strain-associated differences in carbohydrate-related metabolite profiles. For the YDN29-inoculated yucha (Fig. 5B), pathways related to the biosynthesis of unsaturated fatty acids and amino acid metabolism were highly responsive. In the biosynthesis of unsaturated fatty acids pathway (Fig. 5B-a), YDN29 was associated with higher relative abundances of linoleic acid and cis-11,14-eicosadienoic acid, though it exhibited a lower level of eicosapentaenoic acid (EPA) compared to the spontaneous fermentation. In pantothenate and CoA biosynthesis (Fig. 5A-b), pantoic acid was more abundant in the YDN29 group, while pantothenoylcysteine was significantly suppressed. Moreover, in arachidonic acid metabolism (Fig. 5B-c), prostaglandin F2alpha and 20-hydroxyeicosatetraenoic acid were more abundant, contrasting with a notable decrease in 12-docosahexaenoyl-sn-glycero-3-phosphocholine. In alanine, aspartate, and glutamate metabolism (Fig. 5B-d), YDN29 significantly increased the content of aspartate and 2-oxosuccinamate, while the level of 4-aminobutyric acid (GABA) was observed to be substantially lower than that in the control group. The FY27-inoculated group (Fig. 5C) exhibited a distinct pattern of metabolite differences associated with amino acid and fatty acid metabolisms. Regarding the biosynthesis of unsaturated fatty acids (Fig. 5C-a), FY27 fermentation resulted in a markedly higher relative abundance of linoleic acid and docosapentaenoic acid (DPA), alongside a concurrent decrease in alpha-linolenic acid and arachidonic acid. In arginine and proline metabolism (Fig. 5C-b), specific metabolites such as 4-guanidinobutyric acid and N-carbamoylputrescine were markedly. Similar to YDN29, the FY27 strain was also associated with a higher abundance of pantoic acid and a lower abundance of pantothenoylcysteine in the pantothenate and CoA biosynthesis pathway (Fig. 5C-c). Lastly, in alanine, aspartate, and glutamate metabolism (Fig. 5C-d), FY27 was associated with higher abundances of 2-oxosuccinamate and aspartate; however, the abundances of GABA and N-acetylaspartic acid were notably reduced compared to the natural fermentation control. Taken together, the KEGG pathway analyses indicated strain-associated differences in lipid and amino acid metabolisms. These strain-specific metabolite profiles were associated with differences in the final nutritional and functional profiles of fermented yucha. Network pharmacology predicts antioxidant potential of yucha To systematically elucidate the underlying pharmacological mechanisms responsible for the antioxidant properties exhibited by different LAB-fermented yucha, a network pharmacology approach was employed. Screening of bioactive compounds and target intersections Initially, the potential bioactive compounds were systematically screened from the untargeted metabolomics dataset using a robust multi-step pipeline. First, the differentially accumulated metabolites (DAMs) were strictly filtered based on positive KEGG annotations to ensure their biological relevance. Subsequently, a rigorous matrix-specific manual curation was performed to eliminate instrumental false positives and retain only those compounds chemically plausible within the complex yucha (fermented fish-rice) system. Finally, these candidate metabolites were subjected to ADME evaluation to confirm their pharmacokinetic feasibility. Only compounds meeting specific criteria (high gastrointestinal absorption and favorable drug-likeness) were definitively identified as core bioactive compounds for downstream network analysis (Table S10). As shown in the Venn diagram (Fig. 6A), a core set of 26 bioactive compounds was universally shared among the three inoculated groups (YC202, YDN29, and FY27), indicating a conserved metabolic foundation for their functional properties. Concurrently, distinct strain-specific compounds were also identified (22 for YC202, 11 for YDN29, and 28 for FY27), reflecting the unique metabolic and functional capacities inherent to each strain. Analysis of antioxidant activity differences in yucha fermented by different strains. (A) Venn diagram of bioactive compounds produced by different strains; (B) Venn diagram showing the intersection between targets of shared bioactive compounds (from all strains) and disease targets of oxidative damage; (C) Upset plot showing the intersection between strain-specific bioactive compound targets and disease targets; (D) “Strain–metabolite–target” interaction network. Larger and darker circles represent higher degree values and stronger significance. Subsequently, the pharmacological targets corresponding to these bioactive compounds (designated as “Drug” targets) were intersected with “Disease” targets specifically related to oxidative damage (Table S11). As depicted in Fig. 6B, this critical intersection yielded 421 shared therapeutic targets. This substantial overlap robustly demonstrates that the bioactive metabolites generated during LAB fermentation possess significant potential to alleviate oxidative stress and modulate related genetic networks. Strain-specific target profiling and network construction To quantitatively dissect the specific contributions of individual strains to these antioxidant therapeutic targets, an UpSet plot was generated (Fig. 6C). The analysis detailed the complex overlapping relationships between strain-specific bioactive compound targets and the oxidative damage disease targets. It revealed that while the strains share a large subset of common targets, they also engage unique target clusters (with YDN29, FY27, and YC202 associated with 710, 691, and 634 total targets, respectively). This demonstrates that the protective effects of the different LAB strains against oxidative damage are mediated through both synergistic (shared) and highly strain-specific genetic pathways. Finally, a comprehensive “Strain–metabolite–target” interaction network was constructed and visualized using Cytoscape 3.8.0 (Fig. 6D) to elucidate this holistic pharmacological mechanism. This complex network, integrating 129 bioactive components and their 421 corresponding therapeutic targets, effectively maps the relationships from the three LAB strains (left) to their specific active metabolites (hexagons, middle), and subsequently to the downstream effector genes (rectangles, right). The dense topological connectivity powerfully illustrates a “multi-component, multi-target” synergistic paradigm. Protein-protein interaction (PPI) network and functional enrichment analysis The 421 shared targets (identified in Fig. 6B) were submitted to the STRING database to construct a PPI network (Fig. 7A). The median values of six parameters—degree, betweenness centrality, closeness centrality, LAC, network centrality, and eigenvector centrality—were used as screening thresholds. The initial network was progressively refined through three rounds of topological screening, yielding sequential sub-networks of 71 core targets (393 edges), 28 core targets (150 edges), and ultimately a highly interconnected core network of 9 essential targets (24 edges). As depicted in the progressive sub-networks (Fig. 7A, from left to right), larger node sizes and darker colors denote higher connectivity and biological importance. Ultimately, a densely connected core network of hub genes was successfully extracted. The most prominent hub targets included TP53, SRC, AKT1, MAPK3, CTNNB1, JUN, EGFR, ERBB2 and MAPK1. Functional analysis of shared targets between bioactive compounds and disease. (A) Protein–protein interaction (PPI) network. Larger and darker circles represent higher degree values and stronger significance; (B) GO enrichment analysis; (C) KEGG pathway enrichment analysis. To further explore the potential functional association of these targets, Gene Ontology (GO) and KEGG pathway enrichment analyses were performed. The GO enrichment analysis (Fig. 7B) categorized the targets into biological processes (BP), cellular components (CC), and molecular functions (MF). Strikingly, in the BP category, the targets were significantly enriched in stress-related functions such as “response to oxygen levels,” “response to decreased oxygen levels,” and “response to xenobiotic stimulus,” these terms represent general cellular stress-response annotations and may be relevant to oxidative-stress-associated processes. In terms of MF, “protein serine/threonine kinase activity” and “protein tyrosine kinase activity” were highly enriched, suggesting that kinase-mediated signal transduction is a primary functional route. The KEGG pathway enrichment analysis (Fig. 7C) provided an overview of the major signaling pathways annotated to the predicted target set. The bar chart showed that the predicted targets were enriched in several signaling pathways. In particular, the PI3K-Akt signaling pathway and the MAPK signaling pathway exhibited relatively high target counts and significant enrichment. In addition, pathways associated with metabolic and oxidative-stress-related conditions, including “Lipid and atherosclerosis”, “AGE-RAGE signaling pathway in diabetic complications”, and “Apoptosis”, were also among the highly ranked terms. Because these pathways were inferred from computationally predicted targets, the enrichment results indicate potential functional associations rather than confirmed antioxidant, cell-survival, or disease-related effects of fermented yucha. The biological relevance of these pathway associations requires further biochemical, cellular, or in vivo validation. Discussion The spontaneous fermentation of traditional aquatic products is often plagued by uncontrolled microbial successions, leading to texture deterioration, lipid oxidation, and the accumulation of biogenic amines and nitrites (Gao et al., 2023). In this study, targeted inoculation with specific LAB strains (L. paracasei YC202, L. sakei YDN29, and L. plantarum FY27) was associated with improvements in these quality attributes. The rapid pH decline induced by LAB was likely contributed to the physicochemical stabilization of yucha. A highly acidic environment not only suppresses the proliferation of spoilage microorganisms, but also promotes the denaturation and cross-linking of myofibrillar proteins (Chen, Tang, et al., 2022). This acid-induced gelation mechanism plausibly explains why the LAB-inoculated groups maintained superior springiness, chewiness, and hardness compared to the naturally fermented control, which suffered from severe proteolytic softening. Furthermore, the inoculated strains demonstrated profound inhibitory effects on lipid oxidation (TBARS) and significantly accelerated the degradation of nitrites. This dual protective effect is likely attributable to the robust antioxidant systems (reactive oxygen species scavenging capacity) and the specific nitrite reductase activities inherent to these selected LAB strains (Feng & Wang, 2020). By mitigating these spoilage and oxidative factors, LAB inoculation establishes a critical foundation for the safety and potentially improves storage stability of yucha. Consistent with these quality improvements, the LAB-inoculated groups also showed higher DPPH and ABTS scavenging activities and greater ferric-reducing capacity than the spontaneous-fermentation control. These assays provide complementary evidence for enhanced in vitro chemical antioxidant capacity, but do not by themselves demonstrate cellular protection or pathway activation. While all tested LAB strains improved the basic quality of yucha, untargeted metabolomics revealed that distinct strain-specific metabolite profiles at the sampled fermentation endpoint. Spontaneous fermentation primarily resulted in the unregulated accumulation of lipids and phenylpropanoids, indicative of early-stage spoilage and oxidative degradation. Conversely, LAB inoculation was associated with differences in the relative abundances of organic acids, specific amino acids, and their derivatives. Interestingly, each strain exhibited a highly specialized metabolic fingerprint. The L. paracasei YC202 strain was characterized by vigorous carbohydrate utilization, accumulating critical TCA cycle intermediates (including α-ketoglutaric acid) and antioxidant dipeptides like carnosine. This suggests a metabolism optimized for rapid energy production and redox balance (Komae et al., 2026). In contrast, L. sakei YDN29 showed higher relative abundances of the accumulation of free amino acids (such as aspartate, glycine, and serine), which are essential precursors for the characteristic umami and sweet tastes of fermented fish (Zang et al., 2020). This proteolytic activity likely contributes to the high sensory scores of the YDN29 group; however, the top-rated YC202 group suggests that balanced acidification and carnosine accumulation, rather than proteolysis alone, govern overall palatability. Meanwhile, L. plantarum FY27 exhibited a functional metabolic footprint, significantly enriching beneficial polyunsaturated fatty acids (PUFAs) like linoleic acid and docosapentaenoic acid (DPA), as well as specific amino acid derivatives. These strain-specific metabolic signatures underscore the necessity of rational starter culture selection, proving that specific LAB strains can be strategically harnessed to customize the flavor, nutritional, and functional attributes of fermented aquatic foods (Grujović et al., 2022). A defining innovation of this study is the integration of network pharmacology to decipher the functional properties of yucha. While traditional studies merely observe the end-point antioxidant capacity, our network analysis mechanistically bridged the 129 specific bioactive metabolites generated during LAB fermentation to 421 targets associated with oxidative damage. The topological analysis of the PPI network identified TP53, SRC, AKT1, and MAPK3 as the most critical core regulatory hubs. These genes are master regulators of cellular redox homeostasis, stress-induced apoptosis, and inflammatory responses (Chen, Choi, et al., 2022; Yang et al., 2023). For instance, TP53 and AKT1 are pivotal in determining cell survival under oxidative stress, orchestrating the balance between pro-oxidant damage and antioxidant defense mechanisms (Li et al., 2024). Further functional enrichment analysis suggested potential associations with the PI3K-Akt and MAPK signaling pathways. The PI3K-Akt pathway is universally recognized for its role in promoting cell survival and upregulating endogenous antioxidant enzymes (such as SOD and CAT) in response to oxidative stress (Wen et al., 2024). Simultaneously, the MAPK cascade is a primary responsive element to reactive oxygen species (ROS), modulating downstream transcriptional activity to mitigate cellular damage (Franci et al., 2024). The bioactive metabolites (such as PUFAs, carnosine, and specific organic acids) produced by strains like YC202, YDN29, and FY27 collaboratively engage these central kinase-driven pathways, thereby theoretically conferring robust protective effects against oxidative damage (Sebghatollahi et al., 2025). The increased in vitro antioxidant capacity observed after LAB inoculation is compatible with this predicted antioxidant potential. However, the network analysis does not establish direct metabolite-target binding, pathway activation, or cellular protection. Conclusion This study deciphered the complex metabolic biotransformations within the fish-rice co-fermented high-protein matrix. Untargeted metabolomics showed that different lactic acid bacterial (LAB) starter cultures were associated with distinct metabolite profiles in the dual-substrate system, including higher relative abundances of carnosine in the YC202 group, umami-associated amino acids in the YDN29 group, and selected polyunsaturated fatty acids in the FY27 group. Moreover, all LAB-inoculated groups exhibited higher DPPH and ABTS radical-scavenging activities and ferric-reducing capacity than the spontaneous-fermentation control, supporting an enhancement of in vitro chemical antioxidant capacity. These findings directly demonstrate strain-associated differences in the measured quality attributes and metabolite composition, whereas the predicted antioxidant mechanisms remain to be experimentally validated. Network pharmacology further predicted potential associations between these distinct metabolite profiles and biological networks, with TP53, SRC, and AKT1 identified as core network hubs within the PI3K–Akt and MAPK signaling pathways. Although the in vitro assays support antioxidant potential at the chemical level, they do not establish cellular protection or confirm the involvement of these predicted signaling pathways. While this predictive framework requires future in vivo experimental validation, it may provide a framework for rational LAB starter selection and the future development of standardized fermented aquatic foods. Its practical and functional significance should be further assessed through targeted biochemical assays, mechanistic validation, and shelf-life studies. CRediT authorship contribution statement Jiaqi Wang: Writing – original draft, Investigation, Data curation, Conceptualization. Xiaojing Li: Writing – review & editing, Visualization, Conceptualization. Jiaxin Shi: Writing – review & editing, Methodology, Investigation. Binbin Wang: Writing – review & editing, Data curation. Qiwei Liu: Writing – review & editing, Investigation. Junli Feng: Writing – review & editing, Data curation. Changlong Liu: Writing – original draft, Funding acquisition. Jing Xue: Writing – review & editing, Conceptualization. Guofang Shen: Writing – review & editing, Supervision, Project administration, Methodology. Miaoyu Li: Writing – review & editing, Supervision, Investigation. Qing Shen: Writing – review & editing, Supervision, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgement The authors gratefully acknowledge the financial support for this study from the Natural Science Foundation of Zhejiang Province (LLSQN25H020003), the China Postdoctoral Science Foundation (2025M781420), and Zhejiang Medical Association Clinical Research Fund Project (2024ZYC-A297). Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.fochx.2026.104448. Contributor Information Jing Xue, Email: xuejing@zjgsu.edu.cn. Guofang Shen, Email: shengf@126.com. Miaoyu Li, Email: limiaoyu66@163.com. Qing Shen, Email: sq@zjgsu.edu.cn, leonqshen@163.com. Supplementary data Supplementary material: Supplementary tables (Tables S1–S11) and figures (Figs. S1–S2) supporting the findings of this study. Data availability Data will be made available on request. References Associated Data Supplementary Materials Supplementary material: Supplementary tables (Tables S1–S11) and figures (Figs. S1–S2) supporting the findings of this study. Data Availability Statement Data will be made available on request.