Linking antioxidant activity to strain-specific exometabolomic profiles in probiotic Lactobacillus paracasei and Lactobacillus plantarum fermentation supernatants Authors: Marianna Ciaccia, Matteo Radice, Kristina Haslinger, Rosangela Limongelli, Gennaro Agrimi, Isabella Pisano Published: 2026 ABSTRACT Probiotic microorganisms are increasingly exploited for their capacity to generate bioactive metabolites with potential health-promoting properties. In this study, we investigated the in vitro antioxidant activity of sixteen probiotic bacteria and characterized the extracellular metabolomic profiles of the spent fermentation broths (cell-free supernatants) derived from two strains showing markedly different antioxidant behaviors. Antioxidant potential was quantified using the DPPH radical scavenging assay, and total phenolic content was assessed by the Folin–Ciocalteu method. Among the tested strains, Lactiplantibacillus plantarum 8VEG3C displayed the most favorable antioxidant profile within the tested panel, with the lowest IC50 value (2.47% v/v) and the highest phenolic content, whereas Lacticaseibacillus paracasei LC01 showed a substantially weaker antioxidant activity. To explore the molecular basis of these functional differences, untargeted extracellular metabolomic analysis was performed by liquid chromatography–tandem mass spectrometry (LC–MS/MS), coupled with multivariate statistical analysis and putative, database-supported annotation. The two strains generated clearly distinguishable extracellular signatures. The L. plantarum 8VEG3C supernatant was enriched in fatty acyl-related metabolites, including oleic acid, 7-oxooctadecanoic acid and γ-dodecalactone, together with aromatic amino acid-derived compounds such as indole lactate. In contrast, L. paracasei LC01 was enriched in amino acid- and small peptide-related metabolites including the putatively annotated 2-(6-hydroxyhexanoylamino)-3-phenylpropanoic acid, compatible with a more prominent involvement of the proteolytic system of lactic acid bacteria. These findings support the use of untargeted metabolomics as a functional screening platform for the rational selection of probiotic strains with postbiotic potential. Introduction Microbial strains labelled as “probiotic” belong to species such as Bacillus clausii, Bifidobacterium animalis, Bifidobacterium breve, Lactobacillus acidophilus, Lactobacillus delbrueckii, Lacticaseibacillus casei group, Lactiplantibacillus plantarum group, Limosilactobacillus reuteri, and Saccharomyces boulardii. They have been extensively studied for their ability to modulate host-microbiota interactions, enhance mucosal immunity, and contribute to metabolic regulation (W. J. Choi et al., 2020). In addition, the antioxidant activity of some probiotics has been proposed as one of the mechanisms underlying prevention or treatment of cardiometabolic disorders (Cavalcanti Neto et al., 2018), gastrointestinal cancers (Thoda et al., 2025) and neurogenerative diseases (Taghizadeh Ghassab et al., 2024). Indeed, such activity would limit oxidative stress that could favour inflammation and cause cell damage (Chen et al., 2019). Some beneficial traits of probiotics are not alternative dependent on the viability of the microorganisms themselves, leading to increasing attention in the field of postbiotics (Salminen et al., 2021a). Although the International Scientific Association of Probiotics and Prebiotics (ISAPP) issued a consensual definition of postbiotics (“preparation of inanimate microorganisms and/or their components that confers a health benefit on the host”), neither the Food and Drug Administration nor the European Food Safety Authority have so far developed a regulatory framework dedicated to postbiotics. Therefore, for instance, researchers may refer to “postbiotics” as compounds (more commonly in mixture) secreted by microorganisms during growth and that possess activities, such as antimicrobial and antioxidant, that prolong shelf-life of food while ensuring salubrity (Rahman et al., 2025). These bioactive molecules may include short-chain fatty acids (SCFAs), exopolysaccharides, peptides, bacteriocins, organic acids, and vitamin derivatives (Scott et al., 2022). SCFAs play a pivotal role in epithelial barrier maintenance, energy balance, and anti-inflammatory signalling through host receptors such as GPR41, GPR43, and GPR109A (Liu et al., 2024). Among these bioactive molecules, peptides deserve particular attention, as their release during fermentation is not spontaneous but relies on a specialized proteolytic apparatus expressed by lactic acid bacteria. In these microorganisms, protein hydrolysis is mediated by a highly organized proteolytic system that includes a cell-envelope proteinase (PrtP), which initiates the degradation of extracellular proteins into oligopeptides, dedicated oligopeptide and di-/tripeptide transport systems, which import these fragments into the cytoplasm, and a set of intracellular peptidases, including PepN, PepX and PepO, which further hydrolyse the internalized peptides into smaller peptides and free amino acids (Kieliszek et al., 2021). The activity of this system has been directly demonstrated in probiotic lactobacilli: for example, a proteinase from Lactobacillus rhamnosus OXY was purified and biochemically characterized, showing the capacity to hydrolyse food proteins into peptide fragments compatible with the release of bioactive sequences (Waśko et al., 2012). Through this proteolytic machinery, lactic acid bacteria can generate short peptides with a wide range of biological activities, including antimicrobial, immunomodulatory and antioxidant effects, thereby contributing substantially to the functional properties of the corresponding fermented preparations and their postbiotic derivatives. Given the chemical and functional diversity of these bioactive molecules, there is a growing need for advanced tools capable of systematically characterizing the metabolic outputs of microbial strains under defined environmental and nutritional conditions (Kumar et al., 2024). Most of the studies about postbiotics either did not profile their chemical composition or reported a partial characterization of microbial metabolites. For instance, phenyl lactic acid, known for its potential antioxidant activity, was detected as a predominant component in the culture broth of L. reuteri DS0333, which showed anti-photoaging activity (Song et al., 2025). Yet, the ISAPP indicates, among the criteria for a postbiotic preparation, the need for detailed description of the composition of the preparation (Salminen et al., 2021b). In this context, metabolomics has emerged as a powerful systems-level platform for profiling the small-molecule composition of microbial cultures and exploring their functional interactions with the host. Unlike genomic or taxonomic approaches, which provide information on potential capabilities, metabolomics captures the actual biochemical activity of microbes, thereby offering direct insight into their phenotypic expression and bioactive potential (Go et al., 2024). Two alternative metabolomic strategies can be used: targeted and untargeted. Targeted metabolomics focuses on the quantification of predefined metabolites within known biochemical pathways, allowing detailed examination of specific functional traits, such as antioxidant capacity, anti-inflammatory effects, or production of essential micronutrients. In contrast, untargeted metabolomics enables broad-spectrum screening of the microbial metabolome without prior assumptions, facilitating the discovery of novel or unexpected compounds that may underlie functional traits (Alzahrani et al., 2021). This is typically performed using high-throughput analytical techniques such as nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS), including liquid chromatography–coupled mass spectrometry (LC–MS) and gas chromatography–coupled mass spectrometry (GC–MS). These platforms provide high sensitivity, structural information, and the ability to detect a wide range of metabolites including SCFAs, organic acids, amino acid derivatives, volatile compounds, and signalling molecules (Pinu et al., 2017). In both strategies, raw spectral data are processed through normalization, scaling, and noise reduction techniques, followed by statistical analyses including principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA), and clustering algorithms to identify discriminative metabolite signatures. These signatures can then be annotated and interpreted using pathway enrichment tools such as MZmine, MS-DIAL, XCMS and Compound Discoverer (Compound Discoverer 3.2 – The Ultimate Toolbox to Analyze Untargeted Metabolomics Data | My Compound Discoverer, n. d.; Dührkop et al., 2019; Tsugawa et al., 2015) which provide functional context and highlight relevant biosynthetic and regulatory networks. It was shown that two strains of L. plantarum converted catechin galloyl esters into gallic acid and pyrogallol derivatives and produced flavonoid aglycones, enhancing the antioxidant activity (Choi et al., 2023). However, despite the growing interest in postbiotics, most studies still lack a comprehensive characterization of the extracellular small-molecule profiles responsible for functional activities such as antioxidant capacity. It remains unclear which specific metabolites differentiate high-performing strains from those exhibiting limited antioxidant potential, and whether these differences can be systematically captured through untargeted metabolomics. To address this knowledge gap, the present study applied an untargeted LC–MS/MS metabolomics workflow to the spent fermentation broths of two probiotic lactobacilli that showed markedly different in vitro antioxidant activities. Materials and methods Microorganisms and culture conditions The following probiotic bacteria were used: Bifidobacterium animalis subsp. lactis 13 A, Bifidobacterium breve 15 A, Lactobacillus acidophilus LA3, Lacticaseibacillus casei LC4P1, Lactobacillus delbrueckii subsp. bulgaricus SP5, Lacticaseibacillus paracasei 14 A, L. paracasei LC01, Lactiplantibacillus plantarum 8VEG3C, L. plantarum ONI17, L. plantarum VEGI1, L. plantarum LPAL, L. plantarum 3ON, L. plantarum ONI3, L. plantarum VEGII1, Limosilactobacillus reuteri ATCC23272, and Lacticaseibacillus rhamnosus LRB. All the bacteria belonged to the Culture Collection of Department of Soil, Plant and Food Sciences (University of Bari), except for LC4P1, SP5, LPAL, and LRB, which were provided by Sacco srl (Cadorago, CO, Italy) and ATCC23272, which was purchased from the American Type Culture Collection. Strains were routinely cultured in de Man, Rogosa, Sharpe (MRS) broth (Oxoid Ltd., Basingstoke, Hampshire, England, UK) pH 6.2 ± 0.2, at 37 °C, except for L. casei LC4P1, and L. paracasei LC01 and 14 A, which were cultured at 30 °C. Preparation of spent fermentation broth Probiotic bacteria were cultured in MRS broth, under static, microaerophilic conditions, and uncontrolled pH, at 30 or 37 °C depending on the strain, for 24 h (stationary phase of growth, when all had reached a cell density of about 10^9 cfu/ml). Liquid cultures were then centrifuged (9000 × g for 10 min at 4 °C) and the resulting supernatant was collected and stored at −20 °C until analysis. DPPH∙Assay The antioxidant activity of the spent fermentation broths, as well as of non-inoculated MRS broth, was assessed in vitro, using the 2,2-diphenyl-1-picrylhydrazyl (DPPH∙) radical scavenging assay, based on the method described by Xiao et al. (2020), with minor modifications. Briefly, 100 μL of various concentrations of the sample including the unfermented broth control (20 –10 – 5 – 2.5 – 1.125 – 0.6 – 0.3 – 0.15%v/v in methanol) were mixed with 900 μL of freshly prepared DPPH∙ solution (0.1 mM in methanol) in a cuvette. As a negative control, 100 μL methanol, were mixed with the DPPH solution. The reaction mixture was incubated in the dark at room temperature for 30 min, after which the absorbance was recorded at 517 nm (Asample), reading against a negative control (Acontrol). For each experiment, a positive control with Trolox (≥97%, Sigma-Aldrich, Merck, Germany) at a final concentration of 100 μM as antioxidant in DPPH was prepared. The percentage of radical scavenging activity was calculated using the following formula:%Inhibition=(Acontrol−AsampleAcontrol)x100 The IC50 value, defined as the concentration of sample required to inhibit 50% of the DPPH radicals, was determined by plotting the percentage inhibition against the sample concentrations and calculating the value via nonlinear regression analysis. All measurements were performed in duplicate. Folin–Ciocalteu assay The total phenolic content (TPC) of the samples was determined using the Folin–Ciocalteu colorimetric method. Briefly, 100 μL for each sample of spent fermentation broth or unfermented broth was mixed with 1.0 mL of 10% (v/v) Folin–Ciocalteu reagent. After 5 min of incubation at room temperature, 800 μL of 7.5% (w/v) sodium carbonate solution was added. The mixture was vortexed and incubated in the dark at room temperature for 30 min. Absorbance was measured at 765 nm using a UV-Vis spectrophotometer. A calibration curve was generated under the same experimental conditions with gallic acid as a standard, so that the results were expressed as milligrams of gallic acid equivalents (mg GAE) per gram of weight of the sample. All measurements were performed in triplicate. Total antioxidant capacity calculation For each sample, DPPH radical scavenging activity and total polyphenol content (Folin–Ciocalteu assay) were first averaged across technical replicates. The mean values obtained for each assay were then independently normalized to a 0–1 scale using min–max normalization. Normalized DPPH values were inverted so that higher values corresponded to higher antioxidant activity. Subsequently, the normalized total antioxidant capacity was calculated for each sample as the arithmetic mean of the normalized DPPH and Folin–Ciocalteu values. This composite index was used for comparative purposes. The data were normalized using the following formulas applied in Microsoft Excel: For the Folin test values:NormalzedValue(Folin)=X−MIN(XFolinrange)MAX(XFolinrange)−MIN(XFolinrange)·(1−0) For the DPPH test values:NormalzedValue(DPPH)=(X−MIN(XDPPHrange)MAX(XDPPHrange)−MIN(XDPPHrange))·(0−1)+1 Extraction of extracellular metabolites For extraction of extracellular metabolites, spent fermentation broths and unfermented broths were transferred to 25 mL glass bottles, then extracted with 4 mL solvent (9:1 ethyl acetate-methanol (v/v) and 0.1% formic acid), spiked with 5 μL caffeine standard solution with a concentration of 10 mg mL−1, and mixed at 4 °C for 1 h. The organic phase was subsequently collected and dried under a gentle stream of N2. The dried extracts were resuspended in 500 μL of 1:1 MeOH-MilliQ water (v/v). High-resolution liquid chromatography-coupled mass spectrometry (HR-LC-MS/MS) analysis HR-LC-MS/MS analysis was performed with a Shimadzu Nexera X2 high performance liquid chromatography (HPLC) system with binary LC20ADXR coupled to a Q Exactive Plus hybrid quadrupole-orbitrap mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA). Chromatographic separation was performed using an XBridge™ C18 column (3.5 μm particle size, 2.1 × 50 mm, Waters, Milford, MA, USA). The mobile phase consisted of a gradient of solution A (0.1% formic acid in MilliQ water) and solution B (0.1% formic acid in Acetonitrile). A linear gradient was used: 0–2 min 5% B, 2–10 min linear increase to 50% B, 10–17 min linear increase to 90% B, 17 - 20 min decreases to 5% B. The injection volume was 10 μL, and the flow was set to 0.5 mL min−1. MS and MS/MS analyses were performed with electrospray ionization (ESI) in positive mode at a spray voltage of 3.5 kV and sheath and auxiliary gas flow set at 60 and 11, respectively. The ion transfer tube temperature was 300 °C. Spectra were acquired in data-dependent mode with a survey scan at m/z 100–1250 at a resolution of 70,000, followed by MS/MS fragmentation of the top 5 precursor ions at a resolution of 17,500. Data processing and annotation Raw data obtained from the LC-MS/MS analysis were processed using Compound Discoverer™ software (version 3.2, Thermo Fisher Scientific). The data were analysed following the predefined workflow template: Data Workflow Templates\LC\Metabolomics\Untargeted Metabolomics with Statistics - Detect Unknowns with ID using Online Databases and mzLogic. This workflow included retention time alignment, feature detection, compound grouping, background subtraction, and putative identification through spectral matching against multiple online databases such as ChemSpider and mzCloud. Annotation confidence was further supported using mzLogic, which integrates MS/MS fragmentation patterns and in silico prediction tools to improve structural elucidation. The compound annotation was performed using SIRIUS (version 6.1.0), which allows formula prediction and compound classification based on MS/MS fragmentation trees. The analysis was carried out with the following parameters: instrument type set to Orbitrap, mass accuracy set to 5 ppm for MS2, and a minimum of 3 fragments required for tree construction. The CSI: FingerID module was enabled for compound structure prediction, with searching performed against non in silico database. Only identifications with a score ≥80% were considered for annotation. Statistical analysis Antioxidant activity data are expressed as mean ± standard deviation (SD). DPPH radical scavenging activity was measured in duplicate (n = 2) and total phenolic content by the Folin–Ciocalteu assay was measured in triplicate (n = 3). IC50 values for DPPH radical scavenging activity were calculated by non-linear regression analysis of the concentration–response curves. DPPH and Folin–Ciocalteu mean values were subsequently normalized to a 0–1 scale using min–max normalization in Microsoft Excel to enable direct comparison across strains on a uniform reference frame. For the untargeted LC–MS/MS dataset, differential abundance analysis was performed within Compound Discoverer™ (version 3.2, Thermo Fisher Scientific). Metabolites were retained for downstream discussion only if they passed pre-defined exploratory selection thresholds a fold-change (log2FC) ≥ 3 and a p-value <0.001 for the fermented-versus-MRS comparisons, and a log2FC ≥ 2 and a p-value <0.05 for the pairwise comparison between L. plantarum 8VEG3C and L. paracasei LC01. Features attributable to the unfermented MRS medium were removed by background subtraction within the Compound Discoverer workflow. Multivariate structure of the metabolomic dataset was evaluated by Principal Component Analysis (PCA) on scaled peak-area matrices, and the fraction of variance explained by the first two principal components is reported for each dataset. Metabolite annotation was based on spectral matching against the ChemSpider and mzCloud databases, supported by mzLogic-based structural elucidation and by the SIRIUS platform (version 6.1.0) with the CSI:FingerID module enabled. Only annotations with a SIRIUS/CSI:FingerID confidence score ≥80% and an MS/MS mass accuracy within 5 ppm were retained. All annotations should therefore be considered putative, database-supported annotations rather than definitive identifications, in the absence of confirmation with authentic reference standards. Given that the DPPH screening was performed in duplicate to enable the parallel evaluation of sixteen strains, the ranking of intermediate strains is considered preliminary and hypothesis-generating, and it was cross-checked against the more robust Folin–Ciocalteu dataset (n = 3) and against the exometabolomic differentiation observed for the two selected strains. Results Strain-dependent antioxidant activity of probiotic supernatants The antioxidant activity of cell-free probiotic supernatants was evaluated by the DPPH radical scavenging assay and expressed as IC50 values. Marked variability was observed among the sixteen tested strains, indicating a pronounced strain-dependent antioxidant behaviour under the fermentation conditions adopted in this study. IC50 values ranged from 2.47% v/v to 9.44% v/v. L. plantarum 8VEG3C displayed the strongest antioxidant activity, with the lowest IC50 value (2.47% v/v), followed by other L. plantarum strains, including ONI17 and LPAL, which also showed comparatively strong radical scavenging activity. L. acidophilus LA3 showed the weakest performance (IC50 9.44% v/v), and L. plantarum 3ON and L. paracasei LC01 also exhibited relatively weak antioxidant behaviour, with IC50 values around 8% v/v. The non-inoculated MRS broth showed an intermediate IC50 value of 5.75% v/v. These data are shown in Fig. 1 (mean ± SD, n = 2).Fig. 1Antioxidant activity. DPPH assay results for the analysed supernatant samples. Data are presented as mean ± SD. Antioxidant activity. DPPH assay results for the analysed supernatant samples. Data are presented as mean ± SD. Total phenolic content of fermented supernatants Total phenolic content (TPC) was determined by the Folin–Ciocalteu assay and varied substantially among the analysed supernatants (Fig. 2). Measured concentrations ranged from approximately 0.5 mg/mL GAE to 1.4 mg/mL GAE. The lowest value was recorded for L. reuteri ATCC23272, whereas the highest phenolic content was detected in L. plantarum 8VEG3C. Additional L. plantarum strains, including 3ON, ONI17, and ONI3, also displayed relatively elevated TPC values. Conversely, lower phenolic levels were observed for strains such as L. paracasei LC01. The unfermented MRS broth showed an intermediate baseline value of approximately 0.7 mg/mL GAE. Fig. 2 reports the data as mean ± SD, with n = 3 per strain.Fig. 2Total phenolic content - Results of the Folin-Ciocalteu assay for the analysed supernatant samples. Data are presented as mean ± SD. Total phenolic content - Results of the Folin-Ciocalteu assay for the analysed supernatant samples. Data are presented as mean ± SD. Integrated comparison of antioxidant activity and phenolic content The joint evaluation of DPPH scavenging activity and total phenolic content revealed that these two readouts did not follow a fully overlapping ranking across the strain panel. Some probiotics displayed concordant profiles, combining strong radical scavenging activity with elevated TPC values, most notably L. plantarum 8VEG3C and L. plantarum VEG1. However, several strains showed discordant behaviour: L. plantarum 3ON displayed relatively high polyphenol levels but only moderate radical scavenging activity, whereas L. plantarum VEG11 combined comparatively stronger antioxidant activity with a less prominent phenolic profile. Because no formal correlation analysis was performed, these differences are described as a partial discordance in strain ranking rather than as evidence of the presence or absence of a statistical correlation between the two assays. The integrated comparison based on normalized DPPH and Folin–Ciocalteu values is summarized in Fig. 3. On this basis, two strains representing opposite ends of the observed antioxidant spectrum, L. plantarum 8VEG3C and L. paracasei LC01, were selected for subsequent untargeted exometabolomic profiling.Fig. 3Total Antioxidant Activity. Integrated comparison of normalized DPPH radical scavenging activity and normalized total phenolic content. Each parameter was independently normalized to a 0–1 scale using the mean of n = 2 for the DPPH assay and n = 3 technical replicates for the Folin–Ciocalteu assay. The composite antioxidant index was calculated as the arithmetic mean of the two normalized values, as described in Section 2.5. . Total Antioxidant Activity. Integrated comparison of normalized DPPH radical scavenging activity and normalized total phenolic content. Each parameter was independently normalized to a 0–1 scale using the mean of n = 2 for the DPPH assay and n = 3 technical replicates for the Folin–Ciocalteu assay. The composite antioxidant index was calculated as the arithmetic mean of the two normalized values, as described in Section 2.5. . Comparative extracellular metabolomic profiles of L. plantarum 8VEG3C and L. paracasei LC01 Untargeted LC–MS/MS analysis of L. plantarum 8VEG3C and L. paracasei LC01 fermented supernatants, together with unfermented MRS controls, generated 3392 features processed with Compound Discoverer. Of these, approximately 1500 features received a putative, database-supported annotation through the integration of ChemSpider, mzCloud and the SIRIUS/CSI:FingerID pipeline, with a confidence score ≥80%. To restrict the discussion to a robust and interpretable subset, metabolites were selected on the basis of statistics-driven criteria. In the fermented-versus-MRS comparisons (Fig. 4a and b) and in the pairwise comparison between the two strains (Fig. 6), only features passing a fold change threshold (log2FC ≥ 3 or ≥ 2, respectively) and a p-value threshold (p < 0.001 for the initial screen and p < 0.05 for the pairwise comparison) were retained. Using these criteria, 38 compounds were significantly enriched in the L. plantarum-fermented medium and 39 in the L. paracasei-fermented medium relative to the MRS control. In parallel, 129 features showing decreased abundance according to the predefined exploratory criteria by L. plantarum and 98 by L. paracasei, indicating a pronounced modification of the extracellular chemical environment by both strains.Fig. 4Exometabolomics data. Metabolomic comparisons were performed between MRS and (a) L. plantarum–fermented medium and (b) L. paracasei–fermented medium. Each dot represents a metabolite, plotted by log2 fold change (fermented vs. MRS) and −log10 p-value. Grey dots indicate non-significant changes. Light red and light green dots indicate non-significant metabolites above or below the fold change thresholds, respectively. Dark red and dark green dots indicate significantly upregulated or downregulated metabolites in fermented media compared to MRS. . Exometabolomics data. Metabolomic comparisons were performed between MRS and (a) L. plantarum–fermented medium and (b) L. paracasei–fermented medium. Each dot represents a metabolite, plotted by log2 fold change (fermented vs. MRS) and −log10 p-value. Grey dots indicate non-significant changes. Light red and light green dots indicate non-significant metabolites above or below the fold change thresholds, respectively. Dark red and dark green dots indicate significantly upregulated or downregulated metabolites in fermented media compared to MRS. . Multivariate analysis further supported these findings. Principal component analysis distinguished the two fermented supernatants from the corresponding MRS controls (Fig. 5). In the L. paracasei dataset (Fig. 5a), PC1 and PC2 explained 46.8% and 28.8% of the variance, respectively, corresponding to a cumulative explained variance of 75.6%. In the L. plantarum dataset (Fig. 5b), PC1 explained 51.7% and PC2 23.5% of the variance, with a total explained variance of 75.2%.Fig. 5PCA. Principal Component Analysis (PCA) of the untargeted LC–MS/MS metabolomic dataset. (a) PCA score plot comparing L. paracasei LC01–fermented medium and the unfermented MRS control. (b) PCA score plot comparing L. plantarum 8VEG3C–fermented medium and the unfermented MRS control. The percentage of variance explained by the first two principal components (PC1 and PC2) is indicated on the corresponding axes. PCA. Principal Component Analysis (PCA) of the untargeted LC–MS/MS metabolomic dataset. (a) PCA score plot comparing L. paracasei LC01–fermented medium and the unfermented MRS control. (b) PCA score plot comparing L. plantarum 8VEG3C–fermented medium and the unfermented MRS control. The percentage of variance explained by the first two principal components (PC1 and PC2) is indicated on the corresponding axes. Compound-class assignment of the annotated features identified two clearly distinct extracellular signatures (Table 1). L. plantarum 8VEG3C was enriched in fatty acyls, including unsaturated medium-chain and long-chain fatty acids, whereas L. paracasei LC01 metabolites were predominantly represented within the class of carboxylic acids and derivatives, mostly falling under the amino acids and small peptides subcategories.Table 1Compound-class assignment of the discriminant features associated with L. plantarum 8VEG3C and L. paracasei LC01 based on SIRIUS classification.Table 1StrainFormulaClass (final)Annotation hierarchy (Pathway > Superclass > Class > Most specific class)L. plantarumC12H22O2Fatty AcylsFatty acids > Fatty Acids and Conjugates > Unsaturated fatty acids > Medium-chain fatty acidsL. plantarumC18H34O2Fatty AcylsFatty acids > Fatty Acids and Conjugates > Unsaturated fatty acids > Long-chain fatty acidsL. plantarumC18H34O3Fatty AcylsFatty acids > Octadecanoids > Other Octadecanoids > Long-chain fatty acidsL. plantarumC12H19N2O2Carboxylic acids and derivativesAlkaloids > Nicotinic acid alkaloids > Pyridine alkaloids > Amino acidsL. paracaseiC12H19N2O2Carboxylic acids and derivativesAlkaloids > Nicotinic acid alkaloids > Pyridine alkaloids > Amino acidsL. plantarumC14H16N2O3Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Dipeptides > Alpha amino acids and derivativesL. paracaseiC14H16N2O3Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Dipeptides > Alpha amino acids and derivativesL. plantarumC15H21NO4Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Dipeptides > Phenylalanine and derivativesL. paracaseiC15H21NO4Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Dipeptides > Phenylalanine and derivativesL. plantarumC16H33N3O5Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Tripeptides > PeptidesL. paracaseiC16H33N3O5Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Tripeptides > PeptidesL. plantarumC7H11NO3Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Aminoacids > Alpha amino acid estersL. plantarumC15H24N2O4Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Pyridine alkaloids > N-acyl-alpha amino acids and derivativesL. plantarumC17H17NO7Carboxylic acids and derivativesPolyketides > Flavonoids > Depsides > N-acyl-alpha amino acids and derivatives Table 1Compound-class assignment of the discriminant features associated with L. plantarum 8VEG3C and L. paracasei LC01 based on SIRIUS classification.Table 1StrainFormulaClass (final)Annotation hierarchy (Pathway > Superclass > Class > Most specific class)L. plantarumC12H22O2Fatty AcylsFatty acids > Fatty Acids and Conjugates > Unsaturated fatty acids > Medium-chain fatty acidsL. plantarumC18H34O2Fatty AcylsFatty acids > Fatty Acids and Conjugates > Unsaturated fatty acids > Long-chain fatty acidsL. plantarumC18H34O3Fatty AcylsFatty acids > Octadecanoids > Other Octadecanoids > Long-chain fatty acidsL. plantarumC12H19N2O2Carboxylic acids and derivativesAlkaloids > Nicotinic acid alkaloids > Pyridine alkaloids > Amino acidsL. paracaseiC12H19N2O2Carboxylic acids and derivativesAlkaloids > Nicotinic acid alkaloids > Pyridine alkaloids > Amino acidsL. plantarumC14H16N2O3Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Dipeptides > Alpha amino acids and derivativesL. paracaseiC14H16N2O3Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Dipeptides > Alpha amino acids and derivativesL. plantarumC15H21NO4Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Dipeptides > Phenylalanine and derivativesL. paracaseiC15H21NO4Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Dipeptides > Phenylalanine and derivativesL. plantarumC16H33N3O5Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Tripeptides > PeptidesL. paracaseiC16H33N3O5Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Tripeptides > PeptidesL. plantarumC7H11NO3Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Aminoacids > Alpha amino acid estersL. plantarumC15H24N2O4Carboxylic acids and derivativesAmino acids and Peptides > Small peptides > Pyridine alkaloids > N-acyl-alpha amino acids and derivativesL. plantarumC17H17NO7Carboxylic acids and derivativesPolyketides > Flavonoids > Depsides > N-acyl-alpha amino acids and derivatives Compound-class assignment of the discriminant features associated with L. plantarum 8VEG3C and L. paracasei LC01 based on SIRIUS classification. The list of significantly upregulated metabolites in the two strains is reported in Table 2. For L. plantarum 8VEG3C, the annotated set included lipid-derived compounds such as oleic acid (C18H34O2), 7-oxooctadecanoate (C18H34O3), 12,13-dihydroxy stearic acid, and γ-dodecalactone (C12H22O2), together with additional annotated metabolites, including indole lactate, indoleacetaldehyde, ethyl pyroglutamate, nicotinic acid, and (S)-S-acetyl-pantetheine. Among the features detected in the L. paracasei LC01 supernatant, the annotated set included amino acid-related metabolites and small peptide derivatives, such as 2-(6-hydroxyhexanoylamino)-3-phenylpropanoic acid (C15H21NO4), among other putatively annotated compounds.Table 2Putatively annotated metabolites significantly enriched in L. plantarum 8VEG3C- and L. paracasei LC01-fermented media relative to the unfermented MRS control.Table 2FormulaNamem/z measuredm/z calculatedppm errorrtL. plantarum 8VEG3CC13H24N2O5S(s)-S-acetyl-pantetheine343.1297343.1303−1.673.83C7H14N2O2SS-(2-acetamidoethyl)N-ethylcarbamothioate191.0850191.0854−1.853.83C6H5NO2Nicotinic acid124.0398124.03980.130.32C8H15NO4Sn-(1-carboxyethyl)-d-methionine222.0795222.0800−2.141.84C15H31N3O5S(2 S)-4-methyl-2-[[(2 S)-1-[[(2 S)-1-oxaldehydoylpyrrolidifine-2-carbonyl]amino]-3-phenylpropanoyi]amino]pentanoic acid366.2058366.2062−1.113.83C11H11NO3Indole lactate206.0813206.0817−1.790.79C18H34O37-oxooctadecanoate299.2581299.2586−1.744.32C11H10O63-Acetyl-2-acetyloxy-4-hydroxybenzoic acid239.0549239.0555−2.4810.89C18H36O412,13-dihydroxy stearic acid317.2686317.2691−1.6511.84C18H19NO42-[[2-(4-methylphenoxy)acetyl]amino]-3-phenylpropanoate314.1385314.1392−2.1112.88C7H11NO3Ethyl pyroglutamate158.0814158.0817−1.440.90C25H30O8(3S,4aR,4bS,6S,8R,8aR,10aR)-6-Acetoxy-3-furan-3-yi-4a,8a-dimethyl-1,5-dioxo-dodecahydro-2-oxa-phenanthrene-8-carboxylic acid allyl ester459.2014459.2018−0.949.63C10H9NOIndoleacetaldehyde160.0758160.0762−2.499.76C18H34O2Oleic acid283.2631283.2637−1.8510.64FormulaNamem/z measuredm/z calculatedppm errorrtL. paracasei LC01C13H24N2O5S(+)-S-acetyl-pantetheine321.1481321.1484−0.933.83C7H14N2O2SS-(2-acetamidoethyl) N-ethylcarbamothioate191.0851191.0854−1.73.83C19H33N3O6S(2 S)-4-methyl-2-[[(2 S)-2-[[(2 S)-1-oxaldehydoylpyrrolidine-2-carbonyl]amino]-3-phenylpropanoyl]amino]pentanoic acid432.2164432.2168−0.915.72C15H31N3O5Stert-butyl N-[[[(2 S)-1-(2-methylpropylamino)-4-methylsulfonyl-1-oxobutan-2-yl]amino]methyl]carbamate366.2058366.2062−1.193.83C8H15NO4Sn-(1-carboxyethyl)-d-methionine222.0796222.0800−1.461.84C12H22N2O4SN-Formyl-L-methionyl-L-leucine291.1371291.1378−2.524.95C11H11NO3Indolelactate206.0812206.0817−2.084.32C18H19NO42-[[2-(4-methylphenoxy)acetyl]amino]-3-phenylpropanoate314.1385314.1392−2.247.57C16H33N3O5methyl 4-[4-hydroxybutyl-[4-methoxy-3-(methylamino)-4-oxobutyl]amino]-2-(methylamino)butanoate348.2493348.2498−1.574.91C25H30O8(3S,4aR,4bS,6S,8R,8aR,10aR)-6-Acetoxy-3-furan-3-yl-4a,8a-dimethyl-1,5-dioxo-dodecahydro-2-oxa-phenanthrene-8-carboxylic acid allyl ester481.1837481.1838−0.249.61C13H24N2O5L-Valine, N-[(1,1-dimethylethoxy)carbonyl]-L-alanyl-289.1760289.1763−0.964.0C14H26N2O5(2 S)-3-methyl-2-[[(2 S)-3-methyl-2-(propan-2-yloxycarbonylamino)butanoyl]amino]butanoic acid303.1914303.1920−1.874.91C10H9NOIndoleacetaldehyde160.0757160.0762−2.934.33C12H15NO41-Carboxyethylphenylalanine238.1077238.1079−0.984.62C16H23N3O3L-prolyl-L-tyrosine ethylamide306.1813306.1817−1.521.84C11H10O63,5-Bis(carboxymethyl)benzoic acid239.0547239.0555−3.480.91 Table 2Putatively annotated metabolites significantly enriched in L. plantarum 8VEG3C- and L. paracasei LC01-fermented media relative to the unfermented MRS control.Table 2FormulaNamem/z measuredm/z calculatedppm errorrtL. plantarum 8VEG3CC13H24N2O5S(s)-S-acetyl-pantetheine343.1297343.1303−1.673.83C7H14N2O2SS-(2-acetamidoethyl)N-ethylcarbamothioate191.0850191.0854−1.853.83C6H5NO2Nicotinic acid124.0398124.03980.130.32C8H15NO4Sn-(1-carboxyethyl)-d-methionine222.0795222.0800−2.141.84C15H31N3O5S(2 S)-4-methyl-2-[[(2 S)-1-[[(2 S)-1-oxaldehydoylpyrrolidifine-2-carbonyl]amino]-3-phenylpropanoyi]amino]pentanoic acid366.2058366.2062−1.113.83C11H11NO3Indole lactate206.0813206.0817−1.790.79C18H34O37-oxooctadecanoate299.2581299.2586−1.744.32C11H10O63-Acetyl-2-acetyloxy-4-hydroxybenzoic acid239.0549239.0555−2.4810.89C18H36O412,13-dihydroxy stearic acid317.2686317.2691−1.6511.84C18H19NO42-[[2-(4-methylphenoxy)acetyl]amino]-3-phenylpropanoate314.1385314.1392−2.1112.88C7H11NO3Ethyl pyroglutamate158.0814158.0817−1.440.90C25H30O8(3S,4aR,4bS,6S,8R,8aR,10aR)-6-Acetoxy-3-furan-3-yi-4a,8a-dimethyl-1,5-dioxo-dodecahydro-2-oxa-phenanthrene-8-carboxylic acid allyl ester459.2014459.2018−0.949.63C10H9NOIndoleacetaldehyde160.0758160.0762−2.499.76C18H34O2Oleic acid283.2631283.2637−1.8510.64FormulaNamem/z measuredm/z calculatedppm errorrtL. paracasei LC01C13H24N2O5S(+)-S-acetyl-pantetheine321.1481321.1484−0.933.83C7H14N2O2SS-(2-acetamidoethyl) N-ethylcarbamothioate191.0851191.0854−1.73.83C19H33N3O6S(2 S)-4-methyl-2-[[(2 S)-2-[[(2 S)-1-oxaldehydoylpyrrolidine-2-carbonyl]amino]-3-phenylpropanoyl]amino]pentanoic acid432.2164432.2168−0.915.72C15H31N3O5Stert-butyl N-[[[(2 S)-1-(2-methylpropylamino)-4-methylsulfonyl-1-oxobutan-2-yl]amino]methyl]carbamate366.2058366.2062−1.193.83C8H15NO4Sn-(1-carboxyethyl)-d-methionine222.0796222.0800−1.461.84C12H22N2O4SN-Formyl-L-methionyl-L-leucine291.1371291.1378−2.524.95C11H11NO3Indolelactate206.0812206.0817−2.084.32C18H19NO42-[[2-(4-methylphenoxy)acetyl]amino]-3-phenylpropanoate314.1385314.1392−2.247.57C16H33N3O5methyl 4-[4-hydroxybutyl-[4-methoxy-3-(methylamino)-4-oxobutyl]amino]-2-(methylamino)butanoate348.2493348.2498−1.574.91C25H30O8(3S,4aR,4bS,6S,8R,8aR,10aR)-6-Acetoxy-3-furan-3-yl-4a,8a-dimethyl-1,5-dioxo-dodecahydro-2-oxa-phenanthrene-8-carboxylic acid allyl ester481.1837481.1838−0.249.61C13H24N2O5L-Valine, N-[(1,1-dimethylethoxy)carbonyl]-L-alanyl-289.1760289.1763−0.964.0C14H26N2O5(2 S)-3-methyl-2-[[(2 S)-3-methyl-2-(propan-2-yloxycarbonylamino)butanoyl]amino]butanoic acid303.1914303.1920−1.874.91C10H9NOIndoleacetaldehyde160.0757160.0762−2.934.33C12H15NO41-Carboxyethylphenylalanine238.1077238.1079−0.984.62C16H23N3O3L-prolyl-L-tyrosine ethylamide306.1813306.1817−1.521.84C11H10O63,5-Bis(carboxymethyl)benzoic acid239.0547239.0555−3.480.91 Putatively annotated metabolites significantly enriched in L. plantarum 8VEG3C- and L. paracasei LC01-fermented media relative to the unfermented MRS control. The pairwise volcano plot analysis (Fig. 6, Table 3) yielded eleven significantly differential metabolites between the two strains. Seven were more abundant in L. plantarum 8VEG3C, including ethyl pyroglutamate (log2 FC 5.33; p = 0.0006), 4-hydroxy-N-(4-hydroxy-3-methoxybenzoyl)-3,5-dimethoxybenzamide (log2 FC 3.92; p = 0.019), 7-oxooctadecanoic acid (log2 FC 3.29; p = 0.044), oleic acid (log2 FC 2.84; p = 0.03), and γ-dodecalactone (log2 FC 2.19; p = 0.038). Four metabolites were more abundant in L. paracasei LC01, including 2-(6-hydroxyhexanoylamino)-3-phenylpropanoic acid (log2 FC −2.76; p = 0.036) and two unknown features (C14H16N2O3 and C12H19N2O2).Fig. 6Volcano plot analysis. Differential metabolite abundance between L. paracasei LC01– and L. plantarum 8VEG3C–fermented media. Each dot represents a metabolite, plotted according to log2 fold change (L. paracasei LC01 vs. L. plantarum 8VEG3C) and −log10 p-value. Grey dots indicate metabolites not meeting the statistical significance and fold change thresholds. Coloured dots indicate significantly different metabolites, with green representing metabolites at higher abundance in the L. paracasei LC01 and red representing metabolites at higher abundance in the L. plantarum 8VEG3C fermented medium compared each other.Table 3Putatively annotated metabolites showing significantly different abundance between L. plantarum 8VEG3C- and L. paracasei LC01-fermented media.Table 3StrainFormulaNamem/z measuredm/z calculatedppm errorrtLog2 fold changeP-valueL. paracasei LC01C14H16N2O3unknown261.1235261.1239−1.580.09−2.790.050L. paracasei LC01C15H21NO42-(6-hydroxyhexanoylamino)-3-phenylpropanoic acid280.1543280.1549−1.987.3−2.760.036L. paracasei LC01C16H33N3O5N-[2-[2-[2-(1-hydroxypropan-2-ylamino)-2-oxoethoxy]ethoxy] ethyl]-3-methyl-2-(methylamino) pentalamide348.2493348.2498−1.574.91−2.170.009L. paracasei LC01C12H19N2O2unknown246.1336246.1344−3.462.03−2.480.044L. plantarum 8VEG3CC12H22O2dodecalactone199.1694199.1698−2.2110.722.190.038L. plantarum 8VEG3CC17H17NO74-hydroxy-N-(4-hydroxy-3-methoxybenzoyl)-3,5-dimethoxybenzamide348.1076348.1083−2.168.123.920.019L. plantarum 8VEG3CC6H12N4O9Sunknown354.9965354.99620.891.042.970.031L. plantarum 8VEG3CC7H11NO3Ethyl pyroglutamate158.0815158.0817−1.450.75.330.0006L. plantarum 8VEG3CC18H34O37-oxooctadecanoic acid299.2581299.2586−1.7310.893.290.044L. plantarum 8VEG3CC18H34O2Oleic acid283.2632283.2637−1.8412.592.840.03L. plantarum 8VEG3CC15H24N2O4Ethyl 4-[[1-(cyclopropanecarbonyl) pyrrolidine-2-carbonyl]amino] butanoate297.1808297.1814−2.125.022.710.041 Volcano plot analysis. Differential metabolite abundance between L. paracasei LC01– and L. plantarum 8VEG3C–fermented media. Each dot represents a metabolite, plotted according to log2 fold change (L. paracasei LC01 vs. L. plantarum 8VEG3C) and −log10 p-value. Grey dots indicate metabolites not meeting the statistical significance and fold change thresholds. Coloured dots indicate significantly different metabolites, with green representing metabolites at higher abundance in the L. paracasei LC01 and red representing metabolites at higher abundance in the L. plantarum 8VEG3C fermented medium compared each other. Table 3Putatively annotated metabolites showing significantly different abundance between L. plantarum 8VEG3C- and L. paracasei LC01-fermented media.Table 3StrainFormulaNamem/z measuredm/z calculatedppm errorrtLog2 fold changeP-valueL. paracasei LC01C14H16N2O3unknown261.1235261.1239−1.580.09−2.790.050L. paracasei LC01C15H21NO42-(6-hydroxyhexanoylamino)-3-phenylpropanoic acid280.1543280.1549−1.987.3−2.760.036L. paracasei LC01C16H33N3O5N-[2-[2-[2-(1-hydroxypropan-2-ylamino)-2-oxoethoxy]ethoxy] ethyl]-3-methyl-2-(methylamino) pentalamide348.2493348.2498−1.574.91−2.170.009L. paracasei LC01C12H19N2O2unknown246.1336246.1344−3.462.03−2.480.044L. plantarum 8VEG3CC12H22O2dodecalactone199.1694199.1698−2.2110.722.190.038L. plantarum 8VEG3CC17H17NO74-hydroxy-N-(4-hydroxy-3-methoxybenzoyl)-3,5-dimethoxybenzamide348.1076348.1083−2.168.123.920.019L. plantarum 8VEG3CC6H12N4O9Sunknown354.9965354.99620.891.042.970.031L. plantarum 8VEG3CC7H11NO3Ethyl pyroglutamate158.0815158.0817−1.450.75.330.0006L. plantarum 8VEG3CC18H34O37-oxooctadecanoic acid299.2581299.2586−1.7310.893.290.044L. plantarum 8VEG3CC18H34O2Oleic acid283.2632283.2637−1.8412.592.840.03L. plantarum 8VEG3CC15H24N2O4Ethyl 4-[[1-(cyclopropanecarbonyl) pyrrolidine-2-carbonyl]amino] butanoate297.1808297.1814−2.125.022.710.041 Putatively annotated metabolites showing significantly different abundance between L. plantarum 8VEG3C- and L. paracasei LC01-fermented media. Overall, the results indicate that L. plantarum 8VEG3C and L. paracasei LC01 generate clearly distinguishable extracellular metabolomic signatures, dominated by fatty acyl-related compounds in the former and by amino acid- and peptide-related metabolites in the latter. Discussion Antioxidant potential is strain-dependent and only partially explained by total phenolics The DPPH and Folin–Ciocalteu screening confirmed that antioxidant potential is strongly strain-dependent, even within the same species, in agreement with previous reviews on the antioxidant properties of probiotic lactic acid bacteria (Faraki et al., 2020; Feng et al., 2020). L. plantarum 8VEG3C consistently emerged as the best-performing strain in both assays, while other members of the same panel showed discordant profiles between radical scavenging activity and total phenolics. The partial discordance between the DPPH and Folin–Ciocalteu rankings suggest, but does not demonstrate, that phenolic content alone may not fully account for the antioxidant-associated profiles observed under the tested conditions. Other extracellular constituents, including redox-active metabolites, peptides and lipid derivatives, may contribute to the observed activity; however, their individual or combined effects were not directly evaluated in the present study (Amaretti et al., 2013; Faraki et al., 2020). The finding that L. plantarum VEG11 combined stronger DPPH activity with a less prominent phenolic content, and that L. plantarum 3ON showed the opposite pattern, is fully consistent with this multifactorial interpretation. Because the DPPH screening (Fig. 1) was performed with n = 2, the ranking of intermediate strains should be considered preliminary and hypothesis-generating, whereas the more robust n = 3 Folin–Ciocalteu data (Fig. 2) provide a more stable comparative reference. Distinct extracellular metabolic signatures in L. plantarum 8VEG3C and L. paracasei LC01 The untargeted exometabolomic analysis revealed clearly distinguishable chemical signatures for the two selected strains (Fig. 4, Fig. 5, Fig. 6, Table 1, Table 2, Table 3). L. plantarum 8VEG3C was characterized by a broader set of strain-associated metabolites and by a marked enrichment in fatty acyl-related compounds, whereas L. paracasei LC01 displayed a profile dominated by amino acid- and small peptide-related metabolites. This divergence supports the interpretation that the two strains implement distinct extracellular metabolic strategies during fermentation of the same medium and that their functional differences are reflected in the composition of their secreted metabolomes. The higher metabolic versatility of L. plantarum 8VEG3C is in line with the well-documented capacity of this species to biotransform aromatic and lipidic substrates through a broad repertoire of esterases, decarboxylases and reductases (). Conversely, the peptide-oriented profile of L. paracasei LC01 is compatible with a stronger reliance on proteolytic processing during fermentation. This is discussed in more detail in Section 4.3 in relation to the proteolytic apparatus of lactic acid bacteria and its role in the release of bioactive peptides (Chourasia et al., 2023; Waśko et al., 2012). Putative biochemical mechanisms and pathway-oriented interpretation The compounds enriched in L. plantarum 8VEG3C provide a coherent, pathway-level interpretation of its superior antioxidant-associated phenotype. Oleic acid, 7-oxooctadecanoate and 12,13-dihydroxy stearic acid are consistent with an active bacterial lipid remodeling, and γ-dodecalactone is a known product of the microbial biotransformation of unsaturated fatty acid substrates such as a product of bacterial oleic-acid biotransformation through sequential hydration, oxidation, β-oxidation, reduction and cyclization steps. (Boratynski et al., 2020). Together, these annotations suggest that fatty acid metabolism represents one of the dominant extracellular metabolic axes of this strain and a plausible contributor to its antioxidant-associated profile through the release of redox-relevant lipid derivatives (Boratynski et al., 2020; Kim et al., 2021). The annotation of indole lactate and indoleacetaldehyde, both derived from tryptophan catabolism, further indicates that aromatic amino acid metabolism may also contribute to the extracellular chemical landscape of L. plantarum 8VEG3C; both compounds have been previously associated with anti-inflammatory and antioxidant-related activities and with the modulation of host redox and receptor signalling (Tang et al., 2016). The presence of (S)-S-acetyl-pantetheine is consistent with the role of pantetheine derivatives in coenzyme A metabolism and thiol-based redox balance (Wu et al., 2023), while nicotinic acid is compatible with the involvement of pyridine cofactor biosynthesis in NAD(P)+/NAD(P)H homeostasis. A different biochemical scenario emerges from the L. paracasei LC01 profile. The predominance of amino acid- and small peptide-related metabolites (Table 1, Table 2, Table 3) is compatible with an active proteolytic processing of proteinaceous substrates by this strain. In lactic acid bacteria, this activity is supported by a well-characterized cell-envelope proteinase system, together with intracellular peptidases and dedicated peptide transporters, which cooperatively hydrolyse proteins into oligopeptides and free amino acids (Kieliszek et al., 2021). Cell-envelope proteinases of probiotic lactobacilli, including strains of L. rhamnosus, have been shown to release peptide fragments compatible with the generation of bioactive sequences, some of which display antioxidant activity in vitro (Chourasia et al., 2023; Waśko et al., 2012). Consistently, the extracellular profile of L. paracasei LC01 included peptide-related annotations such as 2-(6-hydroxyhexanoylamino)-3-phenylpropanoic acid and additional amino acid-derived compounds (Table 1, Table 2), while The fermented medium also showed consumption of several amino acid-related features relative to the MRS control. This pattern is compatible with a proteolysis-driven metabolic program, in which extracellular nitrogenous substrates are actively hydrolyzed and rerouted (Chourasia et al., 2023; Waśko et al., 2012). It should be emphasized, however, that no direct enzymatic assay, peptide sequencing or functional validation of proteolytic activity was performed in the present study; the data therefore support an indirect and hypothesis-generating association between the peptide-enriched extracellular signature and a possible strain-dependent proteolytic contribution, rather than a demonstrated mechanism in the investigated strains. Overall, these pathway-oriented interpretations, based on SIRIUS/CSI:FingerID compound-class assignments and on the statistics-driven selection of discriminant features, are consistent with a model in which the antioxidant-associated phenotype of L. plantarum 8VEG3C is supported by lipid- and aromatic-related metabolism, whereas the extracellular profile of L. paracasei LC01 is more strongly shaped by nitrogen and peptide metabolism. Postbiotic relevance and connection with host-related pathways The bioactive metabolites highlighted in this study are of potential postbiotic interest. Lipid-derived compounds such as oleic acid and its microbial derivatives have been described as modulators of host inflammation and redox balance (Boratynski et al., 2020), while indole-related metabolites derived from tryptophan catabolism are known to interact with host receptors and to affect the Nrf2–Keap1 antioxidant response pathway (Tang et al., 2016; Tian et al., 2022). Short peptides and amino acid derivatives generated by lactic acid bacteria can also contribute to antioxidant and immunomodulatory effects and are among the most extensively studied classes of postbiotic molecules (S. R. Choi et al., 2023). Within this framework, the extracellular metabolome of L. plantarum 8VEG3C and L. paracasei LC01 provides two chemically complementary profiles that may be relevant for future postbiotic development, one dominated by lipid- and aromatic-related metabolites and the other by peptide- and amino acid-related metabolites. These interpretations remain associations supported by the annotated dataset and by the cited literature, and they should be validated in dedicated targeted and functional studies. Study limitations and future perspectives Several limitations must be explicitly acknowledged. First, metabolite annotation is based on an untargeted LC–MS/MS workflow and, in the absence of authentic standards analysed under identical experimental conditions, all reported identifications correspond to putative, database-supported annotations rather than to definitive identifications, in line with accepted metabolomics reporting standards (Salek et al., 2013; Sumner et al., 2007). As recently emphasized in the metabolomics literature, overinterpretation of untargeted annotations can lead to mechanistic claims not fully supported by the analytical evidence (Theodoridis et al., 2026). For this reason, the present study is best interpreted as a comparative, hypothesis-generating investigation of extracellular metabolomic patterns associated with different antioxidant-associated phenotypes, rather than a definitive mechanistic dissection. Second, the DPPH screening in Fig. 1 was performed with n = 2 replicates, which is underpowered for robust regression or full statistical modeling and, consequently, allows only descriptive strain-level comparisons. The relative ranking of the best- and worst-performing strains is nevertheless supported by the wide separation of IC50 values (2.47% vs 9.44% v/v) and by the concordance with the more robust Folin–Ciocalteu dataset (Fig. 2, n = 3) and with the exometabolomic differentiation observed for the two selected strains (Fig. 4, Fig. 5, Fig. 6, Table 1, Table 2, Table 3). Third, some biologically relevant processes were not directly investigated in this work. Proteolytic activity was inferred indirectly from the extracellular metabolomic signature and from the reported consumption of amino acids, but no proteolytic assay or peptide sequencing was performed. Future studies should therefore focus on (i) targeted LC–MS/MS validation of the most discriminant metabolites, in particular γ-dodecalactone, oleic acid, 7-oxooctadecanoate, indole lactate, indoleacetaldehyde, (S)-S-acetyl-pantetheine, and the peptide-related features of L. paracasei LC01, using authentic standards; (ii) direct characterization of the proteolytic system of the tested strains and biochemical validation of candidate antioxidant peptides; (iii) verification of the biological relevance of these strain-specific extracellular metabolomes in more physiologically informative in vitro and in vivo models and (v) extension of metabolomic profiling to a larger panel of independently fermented strains, enabling formal correlation and multivariate modeling between DPPH activity, Folin–Ciocalteu values and metabolite abundance. Overall, the present work provides an integrated framework combining functional screening and untargeted exometabolomics for the selection of candidate postbiotic-producing probiotic strains, and it identifies L. plantarum 8VEG3C as a priority for follow-up investigation. Conclusion Untargeted metabolomic profiling proved effective in revealing strain-specific extracellular metabolites produced during fermentation, offering insights into the metabolic diversity of probiotic strains and their ability to generate antioxidant compounds. L. plantarum 8VEG3C showed the strongest and most concordant antioxidant-associated profile, while ONI17 and LPAL displayed comparatively strong DPPH scavenging activity. This aligns with the known metabolic versatility of L. plantarum, including its ability to convert phenolic substrates into bioactive derivatives. In contrast, strains like L. acidophilus LA3 and L. paracasei LC01 showed lower antioxidant capacities, underscoring the variability in functional properties among probiotics. Further LC-MS/MS-based metabolomic analysis revealed distinct extracellular metabolic profiles between L. plantarum 8VEG3C and L. paracasei LC01, with notable differences in pathways related to lipid and amino acid metabolism, respectively. Overall, these findings highlight the importance of strain-level characterization in evaluating probiotic functionality and suggest that targeted selection based on metabolic outputs could enhance the development of probiotic-based antioxidant interventions. Further in vivo validation is warranted to confirm the health-promoting potential of the identified metabolites. CRediT authorship contribution statement Marianna Ciaccia: Conceptualization; Data curation; Methodology; Validation; Visualization; Writing – original draft; Writing – review and editing. Matteo Radice: Methodology; Validation; Visualization; Writing – review and editing. Kristina Haslinger: Data curation; Visualization; Writing – original draft. Rosangela Limongelli and Gennaro Agrimi: Validation; Visualization; Writing – review and editing; Isabella Pisano: Writing – review and editing; Supervision; Validation; Visualization; Conceptualization; Data curation; Methodology. 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.