The isolate Gt28L was obtained from star apple fruits (C. cainito) collected in July 2016 from the Amazon rainforest in the Sucumbíos Province, Ecuador, following the protocol described by Tenea and Ortega (2021). The strain was routinely cultured in de Man, Rogosa and Sharpe (MRS) broth (Difco, Detroit, MI, United States) and preserved at -80°C in MRS supplemented with 20% (v/v) glycerol for long-term storage. The microorganism has been deposited in the CCMBIOGEM Laboratory Culture Collection and is available upon reasonable request, in accordance with institutional policies. The schematic overview of the bacterial characterization is depicted in Figure 1.

Whole-genome de novo sequencing and assembly of the Gt28L strain were performed following the methodology previously described by Tenea and Ortega (2021), using sequencing services provided by Macrogen Inc. (Seoul, Republic of Korea). Genomic DNA libraries were prepared via random fragmentation followed by adapter ligation. Raw sequence quality was evaluated using FastQC v0.11.5, and adapter removal and quality trimming were carried out with Trimmomatic v0.36. High-quality reads were further assessed for base quality distribution, GC content, and read abundance prior to assembly. Genome assembly was conducted using SPAdes v3.15.5 with multiple k-mer sizes, supported by k-mer frequency profiling using Jellyfish v2.2.10. The de Bruijn graph-based assembly generated 14,106,374 reads, achieving an average sequencing coverage of 212 × , a GC content of 44.54%, and a Q30 score of 92.18%. Assembly validation through read mapping and BUSCO analysis resulted in 12 contigs totaling 3,231,781 bp, with an N50 of 662,366 bp. BUSCO assessment indicated a high level of genome completeness (99.19% complete, and 0.81% duplicated genes). Additional assembly statistics are provided in Supplementary Tables 1–3. Structural and functional annotation of the Gt28L genome was conducted using a previously established pipeline (Tenea and Ortega, 2021), with PROKKA (Seemann, 2014) for gene prediction and InterProScan and EggNOG (Huerta-Cepas et al., 2019) for functional categorization. Annotations were refined using the Global Catalog of Microorganisms pipeline¹ (Shi et al., 2021), referencing key databases including SwissProt, MetaCyc (Caspi et al., 2020), VFDB (virulence factors database), PHI (Pathogen-Host Interaction database), KEGG (Kyoto Encyclopedia of Genes and Genomes), and COG (Clusters of Orthologous Groups of proteins). Moreover, ARGs were identified using CARD and the RGI tool (Jia et al., 2017; Zankari et al., 2012), while ResFinder 4.1 (Bortolaia et al., 2020) was employed to detect acquired resistance and chromosomal mutations (≥ 90% identity, ≥ 60% length). The pathogenic potential was assessed using PathogenFinder (Cosentino et al., 2013).

Species identification was performed using BLAST against the NCBI NT database, with ANI (Average Nucleotide Identity) values ≥ 95–96% confirming species identity (Richter and Rosselló-Móra, 2009). A circular genome map was generated using Proksee server (Grant et al., 2023). For phylogenetic analysis, the genome FASTA file of strain Gt28L was submitted to the gcType platform, (see text footnote 1) which integrates the Type Strain Genome Database (Shi et al., 2021). The closest type strains were identified by comparing the Gt28L genome against all available type strain genomes in the TYGS database using the Mash algorithm (Meier-Kolthoff and Göker, 2019). Genomes with the smallest Mash distances were automatically selected for downstream analyses. Intergenomic distances were subsequently calculated using the Genome BLAST Distance Phylogeny (GBDP) method, applying the coverage algorithm and the recommended distance formula to ensure accurate phylogenetic resolution (Meier-Kolthoff et al., 2013). Species-level identification was further refined using multiple genome similarity metrics, including ANIb (Richter and Rosselló-Móra, 2009), FastANI, OrthoANIb, and OrthoANIu, to assess nucleotide identity between the query and reference genomes. For phylogenetic reconstruction, 16S rRNA gene sequences were aligned using MAFFT or MUSCLE, and phylogenetic trees were inferred using MEGA, FastTree, and RAxML. In addition, 56 conserved single-copy marker genes were extracted from the selected genomes to construct a robust phylogenomic tree.

FASTA-formatted contigs were subjected to analysis with BAGEL4 to identify BGCs involved in the production of antimicrobial peptides, including bacteriocins (de Jong et al., 2006). To complement this, antiSMASH v6.0.1 (Antibiotics and Secondary Metabolite Analysis Shell) was employed for the genome-wide prediction of BGCs for secondary metabolite, with a particular focus on clusters typical of anaerobic bacteria (Blin et al., 2021).

Pangenome analysis was performed using Roary (v1.007001)² (Page et al., 2015), with MAFFT v7.427 (Katoh and Standley, 2013) for sequence alignment. Protein-coding genes were clustered into core (hardcore and softcore) and accessory (shell and cloud) genomes based on their presence across the analyzed strains. Genomic data for Lactobacillus (n = 9) were retrieved from NCBI (Supplementary Table 4) selected based on their origin from plant- or fruit-derived (vegetal) substrates and the availability of high-quality assemblies with complete metadata, to ensure ecological relevance and robust comparative analyses. Gene clusters were classified as core (> 99%), soft core (95–99%), shell (15–95%), or cloud (< 15%), providing insights into conserved and variable genomic content and adaptive potential.

Adhesion of Gt28L to Caco-2 cells was evaluated following Wang et al. (2022) with minor modifications. Caco-2 cells (HTB-37, ATCC), an in vitro model for differentiated enterocytes, were cultured in RPMI 1640 medium (Thermo Fisher, #11875093) supplemented with 10% FBS and penicillin/streptomycin at 37°C, 5% CO2 for 14–21 days to allow full differentiation. Cells (2 × 105/mL) were seeded in 24-well plates (Fisherbrand, #FB012929) until confluent (∼80–100%). Gt28L cells were inoculated at MOI 10:1 or 100:1 and incubated for 2 h at 37°C. Non-adherent bacteria were removed by PBS washing, and adherent cells were lysed with 0.1% Triton X-100 (in PBS). Released bacteria were serially diluted and plated on MRS agar. Adhesion ability (%) was calculated as (At/Ax) × 100, where At = adherent CFU and Ax = initial CFU (Wang et al., 2022). E. coli ATCC11229 was used as a non-probiotic control; adherence > 20% was considered high. For microscopy, Caco-2 cells were cultured on coverslips in 24-well plates and co-incubated with bacteria as described. After 2 h, wells were washed with PBS, fixed with methanol (15 min), air-dried, and stained with Giemsa. Coverslips were dried overnight, and adhesion was visualized using a Zeiss PrimoStar 3 microscope at 100 × magnification.

All experiments were performed in triplicate, and data are expressed as mean ± standard deviation (SD). Statistical analyses were conducted using GraphPad Prism version 10 (GraphPad Software, San Diego, CA, United States). Antioxidant activity data were analyzed using an unpaired t-test. Comparisons among antioxidant assays (CUPRAC, FRAP, and TEAC) were performed using two-way analysis of variance (ANOVA), followed by Tukey’s post-hoc test. Data normality was assessed using the Shapiro–Wilk test and was confirmed for p > 0.05. For nitric oxide scavenging assays and hemolytic activity, two-way ANOVA with Geisser–Greenhouse correction was applied, followed by Tukey’s post-hoc test. Statistical significance was defined as p < 0.05.

The genome of Gt28L isolated from the Amazonian star apple fruit exhibits a compact and streamlined architecture suggestive of ecological adaptation to a fruit-associated niche. Bioinformatic analysis of the whole-genome sequencing data resulted in a complete genome assembly of 3,231,781 bp, with an average GC content of 44.54%, consistent with the species-specific genomic composition of L. plantarum and indicative of a stable genomic architecture (Table 1). The strain harbors 64 tRNA genes, indicating a potentially constrained codon usage adapted to the consistent nutrient landscape of its environment. Essential rRNA and tmRNA genes are conserved, while the absence of CRISPR-Cas arrays, frequently detected in other L. plantarum strains, may reflect a reduced phage burden in this niche. Despite lacking CRISPR-Cas, Gt28L retains three intact prophages (PHAGE_Lactob_Lj965, PHAGE_Lactob_phig1e, and PHAGE_Lactob_Sha1), which could enhance its ecological competitiveness, genomic plasticity, and stress resistance (Ventura et al., 2003; Luo et al., 2014; Martínez et al., 2022). Importantly, the genome lacks plasmids, transferable ARGs and known virulence factors, underscoring its genomic stability and biosafety. Taxonomic classification based on BLASTN analysis of the full contig sequence showed 100% identity at the genus level to Lactobacillus spp. Average Nucleotide Identity (ANI) analysis revealed 99.35% nucleotide identity and 89.60% alignment coverage with L. plantarum strain r-JN-1 (Supplementary Figure 1), supporting species-level assignment. A detailed genome map is shown in Figure 2, and functional annotation statistics across multiple databases are summarized in Table 2. The most abundant functional category is “General function prediction only” (R), comprising 747 genes, reflecting the presence of genes with predicted roles yet lacking experimental characterization, a typical feature of underexplored probiotic genomes (Figure 3). This is followed by “Function unknown” (S, 352 genes), highlighting a substantial reservoir of potentially novel or strain-specific genes warranting further functional validation (Douillard and de Vos, 2014). Among the well-defined categories, genes related to “Translation, ribosomal structure, and biogenesis” (J, 252 genes), “Amino acid transport and metabolism” (E, 200 genes), and “Carbohydrate transport and metabolism” (G, 172 genes) are prominently represented. These functional groups are critical for metabolic adaptability, efficient nutrient acquisition, and survival in competitive environments such as the gastrointestinal tract or fermented matrices. Additional enrichment in “Transcription” (K, 246 genes) and “Posttranslational modification, protein turnover, chaperones” (O, 145 genes) suggests effective regulatory and stress-response mechanisms. Furthermore, the presence of genes linked to “Defense mechanisms” (V, 107 genes) and “Cell wall/membrane/envelope biogenesis” (M, 130 genes) reinforces the strain’s potential for probiotic activity, including resilience to environmental stress and host interaction.

The distribution of antimicrobial resistance (AMR) features across three categories, ARG families, resistance mechanisms, and drug classes is illustrated in Supplementary Figure S2. The analysis indicates that the detected resistance traits are predominantly associated with intrinsic, chromosomally encoded mechanisms, which are commonly observed in lactic acid bacteria and are not linked to horizontal gene transfer. Specifically, antibiotic target alteration and antibiotic efflux represent the most frequently identified resistance mechanisms. At the ARG family level, glycopeptide resistance–associated gene clusters and ATP-binding cassette (ABC) efflux systems were the most abundant. These findings reflect the expected intrinsic resistance profile of the strain, rather than the presence of clinically relevant or transferable resistance determinants. The comparative resistome analysis of the Gt28L strain identified several ARGs, including those related to ribosomal protection (fusA, rpoB) and translation elongation (EFTu), with prevalence values reaching over 70% (Supplementary Table 5). The presence of vancomycin-resistance associated genes such as vanY and vanT, typically chromosomally encoded and intrinsic in L. plantarum, should not raise safety concerns, as these ARGs are non-transferable and part of the species’ natural resistome (Ammor et al., 2007). vanY was confirmed by RGI analysis with 95.90% sequence identity. Importantly, efflux-related genes such as qacJ, mdtG, and poxtA, which may play roles in resistance to biocides and antimicrobial agents, were also detected. These features suggest environmental adaptability and intrinsic resilience but do not imply pathogenic potential (Hu et al., 2013; Ojha et al., 2023). While L. plantarum is generally considered a safe microorganism with low AMR risk (Douillard and de Vos, 2014), the detection of multiple resistance markers at significant prevalence levels emphasizes the need for thorough genomic screening prior to industrial or clinical use. In addition, similar with other Lactiplantibacillus species (Li and Bi, 2025), tetO and tetA genes were annotated with EggNOG. Moreover, PathogenFinder analysis predicted Gt28L as a non-human pathogen with a 99.95% probability, indicating a highly confident classification.

Furthermore, to assess safety, the genome of Gt28L was screened against the VFDB (Supplementary Table 6). Although several sequences returned VFDB annotations, these hits corresponded only to low-identity matches (40–76.8%) involving housekeeping, metabolic, stress-response, or cell-surface biosynthesis genes commonly found in L. plantarum and other GRAS/QPS lactic acid bacteria (Colautti et al., 2022). None of the annotations represented bona fide virulence determinants such as toxins, hemolysins, cytolysins, invasins, or pathogenicity-associated secretion systems. The genes including Clp proteases, chaperones (DnaK/GroEL), glycolytic enzymes, ABC transporters, and exopolysaccharide biosynthesis genes are essential for bacterial physiology and are well-established features of safe, commensal LAB (Colautti et al., 2022). Likewise, domain-level similarity to pavA- or lap-like adhesion proteins reflects beneficial mucosal adherence functions, not virulence. Apparent matches to Dot/Icm-like secretion proteins do not imply pathogenic potential, as Gt28L lacks the structural modules required to assemble a functional virulence-associated T4SS.

Phenotypic assays further confirmed the safety of Gt28L the strain was non-hemolytic and non-cytotoxic to epithelial cells. The presence of such homologous genes does not demonstrate virulence, as sequence similarity alone cannot determine their functional role or pathogenic potential (Altavas et al., 2024). Together, these results indicate that the VFDB hits do not correspond to true virulence factors and affirm the overall safety of the Gt28L strain.

Comparative analysis with reference strains L. plantarum WCFS1 and L. plantarum ST-III revealed a shared core of such genes, suggesting these elements are part of the intrinsic genomic repertoire of the species (Wang et al., 2011). Notably, these genes are involved in key probiotic functions such as epithelial adhesion, stress resistance, and carbohydrate metabolism, supporting their functional conservation for host adaptation rather than pathogenicity. Similar findings have been reported in other Lactobacillus strains, reinforcing the concept that the presence of these genes alone does not imply virulence but reflects ecological fitness and commensal lifestyle (Petrova et al., 2016).

The phylogenetic analysis places Gt28L within a monophyletic clade comprising other well-characterized L. plantarum strains, including strains historically designated as L. arizonensis and L. argentoratensis. Recent taxonomic studies have reclassified these species as heterotypic synonyms of L. plantarum (Zheng et al., 2020), confirming that Gt28L is consistent with the L. plantarum species complex (Supplementary Figure 3). The tree topology, supported by high bootstrap values, indicates that Gt28L shares a recent common ancestor with strains such as GCM100194569 and GCM100103221, reflecting conserved genomic architecture within this group. Heatmap metadata further support its genomic stability, with Gt28L exhibiting a genome size (∼3.2 Mb) and GC content (∼44.5%) characteristic of L. plantarum, and low standard deviation values indicating minimal structural variability. These parameters reinforce the notion that Gt28L maintains evolutionary coherence with commensal L. plantarum species, while also exhibiting a streamlined genome potentially shaped by niche-specific pressures in the Amazonian fruit environment. Such genomic conservation, combined with its distinct ecological origin, positions Gt28L as a valuable reference for studying functional diversification within the L. plantarum lineage.

Genome mining revealed a well-defined BGC encompassing multiple genes involved in the synthesis, modification, transport, and immunity of class IIb bacteriocins (Figure 4). The cluster spans approximately 55 open reading frames (ORFs), including several putative bacteriocins, notably plantaricin_E, plantaricin_F, plantaricin_N, and plantaricin_J (orf00022, orf00031, orf00033, and orf00037, respectively). These core peptides are characteristic of the two-peptide bacteriocin system previously described in L. plantarum C11 and WCFS1, where synergistic action between peptides is required for antimicrobial activity (Anderssen et al., 1998; Heeney et al., 2019). Notably, the presence of multiple predicted core peptides within a single cluster indicates a potentially broad antimicrobial spectrum or multi-targeted mechanism. Associated with these structural genes are dedicated transport and immunity proteins, including LanT and HlyD, which are typically involved in the secretion and protection mechanisms of bacteriocin-producing strains (Nissen-Meyer et al., 2010). The presence of a glycosyltransferase gene (GlyS, orf00030) suggests potential post-translational modification, which may enhance bacteriocin stability or activity, as previously noted in glycosylated bacteriocins of other LAB strains (Todorov, 2009). Compared to other L. plantarum strains, the Gt28L cluster displays a higher degree of gene multiplicity and operon organization. For example, in L. plantarum WCFS1, the plantaricin locus contains fewer structural genes and is regulated by a well-characterized quorum-sensing system (Heeney et al., 2019), which appears absent or divergent in Gt28L. Moreover, the inclusion of two sets of plantaricin_E/F-like peptides in Gt28L suggests a possible gene duplication or horizontal acquisition event, offering a broader antimicrobial capacity. In terms of genomic arrangement, the Gt28L cluster shares synteny with the pln loci of several L. plantarum strains (Gizachew and Engidawork, 2024), but includes unique ORFs with no currently assigned function, indicating possible strain-specific adaptations or novel components. The abundance of predicted transcriptional terminators and promoters further supports the existence of multiple transcriptional units, suggesting tight regulation and potential responsiveness to environmental stimuli.

In addition, several a diverse secondary metabolite were predicted with antiSMASH (Table 3). A complete list of predicted compounds, category, location, and similarity scores are shown in Supplementary Table 7. Notably, contig 1.1 showed weak to moderate sequence similarity (21-32%) to known polyketide and non-ribosomal peptide clusters, including those associated with falcarindiol, chrysoxanthones, viguiepinol, and bacilysin, which are typically found in Streptomyces, Penicillium, and Bacillus species. Although these clusters displayed low similarity percentages, they may represent cryptic or evolutionarily conserved fragments, possibly involved in the biosynthesis of structurally similar metabolites with antimicrobial potential (Wang et al., 2011). In contrast, contigs 2.1 and 2.2 revealed multiple RiPP-like BGCs showing 16–22% similarity to characterized bacteriocins such as coagulin, microcin M and E492, ubericin K, and sublancin 168. Sublancin 168 was previously predicted in the genome of L. plantarum UTNG21A but was absent in the reference strain L. plantarum WCFS1, reinforcing the hypothesis that strain-specific biosynthetic capabilities, including the production of novel antimicrobial compounds, may be shaped by selective pressures within their native ecological niches (Tenea and Ascanta, 2022). Additionally, contig 7.1 contained a terpene biosynthetic cluster with high similarity (up to 45%) to carotenoid-producing pathways from Rhodobacter and Streptomyces, indicating potential antioxidant or stress adaptation roles. Taken together, these genomic features highlight the capacity of Gt28L to produce a diverse array of secondary metabolites, including multiple bacteriocin-like peptides and terpenes. Importantly, we recently demonstrated the antimicrobial activity of Gt28L in vitro against the multidrug-resistant E. coli L1PEag1 strain, which validates and supports the in silico predictions derived from BGC analysis (Tenea et al., 2025). These findings reinforce the strain potential as a source of novel antimicrobial compounds and its relevance for probiotic and biocontrol applications.

The pangenome analysis of L. plantarum strains revealed a highly diverse genomic structure, indicative of substantial adaptive potential. Among the total gene pool, only 743 genes (∼14%) were identified as core genes shared by 99–100% of strains (Figure 5A), reflecting the minimal set of conserved functions essential for survival. A larger fraction comprised shell genes (2,755 genes, ∼52%), present in 15–95% of strains, likely conferring niche-specific advantages such as carbohydrate metabolism, stress response, or bacteriocin synthesis. Cloud genes accounted for 3,658 genes (∼34%), representing highly variable, strain-specific elements often associated with horizontal gene transfer, prophages, or mobile genetic elements. The absence of soft-core genes underscores the pronounced dichotomy in gene conservation and the overall flexibility of the L. plantarum genome, consistent with its open pangenome and broad ecological range across fermented foods, plant surfaces, and animal gastrointestinal tracts (Choi et al., 2021; Carpi et al., 2022). The UpSet plot shows that most gene clusters (2,270) belong to the core genome, and Gt28L does not dominate any large unique cluster sets (Figure 5B). Roary matrix analysis and the phylogenetic tree further demonstrate Gt28L divergence, with a distinctive pattern of accessory gene presence and absence (Supplementary Figure 4), suggesting potential ecological or functional specialization. The Gt28L genomic architecture reflects adaptation to its fruit-associated niche and likely underpins its unique probiotic traits compared to other L. plantarum strains. Its compact, streamlined genome, combined with an abundant accessory gene pool, facilitates rapid adaptation to variable nutrient and stress conditions. Enrichment in genes related to amino acid transport and metabolism, carbohydrate utilization, and stress-response pathways supports environmental resilience and competitive colonization, while the absence of plasmids, transferable antibiotic resistance genes, and canonical virulence factors ensures biosafety. Extensive bacteriocin and RiPP-like clusters, alongside secondary metabolite pathways such as terpene and carotenoid biosynthesis, may provide broad antimicrobial and antioxidant capabilities, enhancing survival in microbially diverse niches. Compared to reference strains like WCFS1 or ST-III, often optimized for gut or dairy environments, Gt28L exhibits a strain-specific integration of accessory genes that support both ecological versatility and metabolically driven probiotic functions, including adhesion and stress tolerance. Thus, these genomic features highlight how Gt28L distinctiveness translates into functional traits enabling adaptation to its native niche while promoting health-supportive and antimicrobial properties.

The FTIR spectra of the extracellular metabolites produced by the Gt28L strain (Figure 7) were significantly different from the uninoculated and freeze-dried MRS medium in certain spectral regions, demonstrating biochemical changes caused by fermentation.

The stretching vibrations of carbon are very prominent in the vibrational spectrum of benzene and its derivatives; therefore, in the present study, the doubly degenerate e1u ring stretching mode (1,493 cm^–1) and the non-degenerate a1g mode of benzene (998 cm^–1) were attributed to C–C skeletal vibrations. The FTIR spectrum of the Gt28L sample highlights the presence of several functional group’s characteristic of extracellularly secreted secondary metabolites. The analysis was performed in the range of 600–4,000 cm^–1, and the main absorption bands identified as different from the uninoculated MRS medium are highlighted in the range of 698–851 cm^–1 and are specific to out-of-plane C–H deformations of the aromatic nucleus (Olsztyñska-Janus and Komorowska, 2012), indicating the presence of phenolic and aromatic compounds (e.g., phenolic acids, stilbenes, tannins). At the wavenumber 795 cm^–1, the Ring A torsion and C–H aromatic bend of pyrocatechol, flavone, flavanol and epicatechin could be attributed (Hossain et al., 2022), which confirms the presence of catechin and epicatechin. The stretching vibrations of carbon were very prominent in the vibrational spectrum of benzene and its derivatives, so in the present study, the e1u ring stretching mode (1,492 cm^–1) and the a1g mode of benzene (994 cm^–1) were attributed to C–C skeletal vibrations (Sajan et al., 2006). From Figure 7, a series of specific bands of Gt28L can be observed in the range of 1,112–1,359 cm^–1, bands associated with C–O–C, C–N bonds, and O–H vibrations, reflecting the possible presence of polysaccharide compounds, residual sugars, and/or secondary amines (Andreeva and Burkova, 2017). The range 1,462–1,577 cm^–1 was attributed to Amide II; specific bands were found in both non-inoculated MRS and Gt28L, but the bands at 1,493 cm^–1 and 1,516 cm^–1 are specific to the extracellular metabolites of LAB and are attributed to C = C (aromatic) vibrations and NH (amide II) deformations, associated with phenolic compounds and microbial peptides (Adt et al., 2006). The range 1,637–1,665 cm^–1 is specific to Amide I, the typical band for C = O stretching (amide I), highlighting protein content (possible extracellular peptides) and is found in both spectra (Ji et al., 2020). The peak identified only for Gt28L at 1,665 cm^–1 is attributed to C = O stretch vibrations specific to flavonoids (Rocha et al., 2021). The bands at 2,874 and 2,930 cm^–1 are due to the stretching of methylene groups (CH₂, CH₃), corresponding to aliphatic compounds, potentially fatty acids or biosurfactants secreted by bacteria (Dutta et al., 2013). The broad band between 3,140 and 3,263 cm^–1 was associated with the stretching of O–H and N–H, indicating a high content of bound water, hydroxyl groups, or proteins/polysaccharides.

The Gt28L sample showed new or amplified signals in the aromatic and carbonyl compounds typical areas. The existence of these bands in the fermented sample, as well as their absence in the medium, provides evidence that LAB secretes secondary metabolites extracellularly. FTIR spectroscopy indicates that the Gt28L sample contains a diverse spectrum of bioactive components, including phenolic compounds, proteins, polysaccharides, and organic acids.

The antioxidant activity of LAB-produced secondary metabolites opens fresh opportunities in biotechnology, nutrition, and medicine, providing novel sources of bioactive chemicals with significant implications for human health and industrial sustainability. At the same time, LAB can metabolize or inactivate the existing antioxidants during growth (Rusu et al., 2023; Chandra et al., 2020; Kumano, 2024), as also suggested by our results. Gt28L CFS had significantly higher IC50 values (Figure 8A) compared to MRS broth (p < 0.001), implying a weaker capacity to neutralize free radicals, correlated with a reduced amount of effective phenolic antioxidants. The Shapiro-Wilk test showed that the data had a normal distribution (W = 0.8181, p = 0.0850), meeting the normality criterion at a significance level of α = 0.05. Figure 8B presents the concentration-dependent antioxidant activity for both samples, with significantly higher values of DPPH radical scavenging capacity in the case of MRS compared to Gt28L, suggesting again a higher content of antioxidants or the presence of more efficient ones in the uninoculated culture medium.

Using the FRAP, CUPRAC, and ABTS methods (Figure 8C), the antioxidant activity expressed in μM Trolox equivalents per gram of freeze-dried sample was significantly lower than that of MRS (p < 0.0001), indicating a lower electron transport. This difference could be also the result of the consumption of antioxidant compounds initially present in MRS correlated with a limited production of new antioxidants, particularly flavonoids or compounds with reduced groups, in the post-fermentation supernatant. The CUPRAC method was the most efficient in quantifying antioxidants (p < 0.0001), suggesting a predominance of compounds capable of electron transfer, specific to polyphenols, as well as of some non-phenolic compounds such as antioxidant peptides or compounds with carbonyl/conjugated groups. The FRAP and TEAC assay results were marginally lower, indicating a reduced amount of chemicals that operate by electron donation with a lower redox potential. The ANOVA conditions were satisfied since the data was normally distributed (Shapiro-Wilk, p > 0.1) and the variances were homogenous (Brown-Forsythe test, p > 0.05) (fFigure 8C).

Nitric oxide (NO) scavengers help to maintain cardiovascular, neuronal, and cellular health by lowering oxidative stress and inflammation. In general, natural antioxidants decrease excess NO by limiting the formation of reactive nitrogen species (RNS), lowering oxidative stress, inflammation, and cellular damage. This impact is critical for preventing chronic illnesses and maintaining good health (Forman and Zhang, 2021; Mladenov et al., 2023). The NO radical scavenging activity of the CFS from Gt28L and of the MRS growth medium was significantly lower than that of gallic acid, which was used as a reference antioxidant (p < 0.01 for all tested dosages), as shown in Figure 8D. The NO scavenging activity of all three samples varied with concentration. Gallic acid had the strongest radical scavenging activity (85.49 ± 0.29% at 50 mg/mL), followed by MRS (58.62 ± 1.60%) and Gt28L (32.60 ± 1.90%). However, the differences between MRS and Gt28L remained significant (p < 0.05, p < 0.1). While both MRS and Gt28L have some antioxidant activity, MRS outperforms Gt28L at all tested doses. This suggests that LAB secondary metabolism during fermentation may diminish the total antioxidant capacity by degrading or modifying some of the phenolic or antioxidant compounds found in MRS. These results suggest moderate anti-inflammatory and antioxidant activities. However, the presence of other bacterial secondary metabolites may provide additional functional or probiotic effects beneficial in the development of supplements or functional food.

The hemocompatibility of Gt28L-derived metabolites was evaluated through hemolytic and anti-hemolytic assays using sheep erythrocytes. According to ISO 10993-4: ISO, 2017, a hemolysis index below 5% denotes acceptable blood compatibility. At the highest tested concentration (50 mg/mL), both the Gt28L and MRS samples exceeded this threshold; however, Gt28L displayed a significantly lower hemolysis rate (9.2 ± 0.9%) compared with the MRS control (28.2 ± 0.2%, p < 0.0001). At concentrations ≤ 20 mg/mL, the hemolysis level for Gt28L dropped below 5%, confirming its non-hemolytic and biocompatible profile (Figure 9A). The higher activity observed for MRS medium is consistent with the presence of Tween 80, a known membrane-disrupting surfactant (Muhialdin et al., 2021; Reffai and Fechtali, 2025). The anti-hemolytic potential was assessed against AAPH-induced oxidative damage. Both Gt28L and MRS significantly reduced erythrocyte lysis, achieving > 95% protection at 5–20 mg/mL, surpassing ascorbic acid (18.8 ± 1.1% hemolysis at 5 mg/mL, p < 0.01). The protective effect of Gt28L was concentration-dependent and slightly superior to MRS at all tested doses (p < 0.05). This activity likely reflects the presence of extracellular phenolic and peptide compounds with antioxidant and membrane-stabilizing functions.

The good hemocompatibility of the Gt28L sample, despite its moderate antioxidant activity, suggests that its fermented metabolites could be safely incorporated into systemic formulations, such as functional foods or dietary supplements, that require hematological safety. Even in trace amounts, flavonoids, phenolic acids, and stilbenes likely reflect cytoprotective properties (Kaźmierczak et al., 2025; Shevchenko, 2024). This effect is manifested by inhibiting membrane lipid oxidation, reducing oxidative stress, and stabilizing the structure and integrity of the cell membrane. The anti-hemolytic effect of the post-fermentation supernatant (Gt28L), MRS medium, and ascorbic acid was assessed by evaluating their capacity to protect erythrocytes against oxidative hemolysis produced by the peroxyl radical generator AAPH. Figure 9B indicates that, at all tested concentrations, AAPH generated considerable hemolysis (> 40%), demonstrating its oxidizing effect. In contrast, hemolysis was significantly decreased when erythrocytes were pre-treated with Gt28L, MRS, or ascorbic acid. The samples Gt28L and MRS significantly reduced hemolysis and offered better antioxidant protection (<3% hemolysis) than ascorbic acid at 5 mg/mL (18.80 ± 1.10% hemolysis) (p < 0.01). The differences between Gt28L and MRS were significant (p < 0.05) at all doses tested, with Gt28L offering better protection. Thus, by markedly reducing AAPH-induced oxidative stress, Gt28L exhibited notable membrane-protective properties. This protective effect is likely associated with the synthesis of extracellular phenolic metabolites possessing antioxidant activity (e.g., phenolic acids, stilbenes, flavonoids) (Rajashekar, 2023). The general trend was a decrease in hemolysis with the decrease of the sample concentration, suggesting that at higher doses, certain components may exhibit a pro-oxidative effect. Antioxidant compounds may exhibit concentration-dependent dual behavior, acting as pro-oxidants at elevated concentrations and thereby exacerbating oxidative stress rather than attenuating it (Manful et al., 2025). Consequently, the assessment of subhemolytic concentrations is critical to discriminate genuine cytoprotective effects from dose-dependent pro-oxidant activity. The CFS of Gt28L is intended for functional food and postbiotic applications, where its biological activity is expected to exert localized, transient effects within the gastrointestinal tract. Accordingly, the concentrations evaluated in this study are appropriate for assessing safety and functional efficacy under conditions that approximate physiological exposure during gastrointestinal transit, supporting their relevance for in vitro safety and bioactivity assessment.