Resistance to acidic conditions was tested according to the method of Conway et al. (1987). Lactic acid bacteria were cultured in MRS medium overnight and harvested by centrifugation at 6,000 × g for 10 min. Cells were washed twice with phosphate-buffered saline (PBS; pH 7.2) and resuspended in an equal volume of PBS adjusted to pH 3.0 and 2.5 with HCl. Following incubation for 0, 90, and 180 min at 37°C, acid tolerance was evaluated by spreading cells on MRS agar and counting the number of viable cells (Log CFU/ml) after incubation at 37°C for 48 h (Maragkoudakis et al., 2006). Tolerance to bile salts was evaluated by suspending cells in PBS solution containing 0.3% (w/v) bile salt (Sigma, St. Louis, MO, United States) and incubating at 37°C (Gilliland et al., 1984). Bile tolerance was measured using the same method used for acid tolerance.

The adhesion assay was performed as described by Messaoudi et al. (2012). Caco-2 and HT-29 human colonic epithelial cell lines were obtained from the Korean Cell Line Bank (KCLB; Seoul, South Korea) and grown in Dulbecco’s modified Eagle’s medium (DMEM; Hyclone, Logan, UT, United States) supplemented with 10% fetal bovine serum (FBS; Hyclone), and 1% each of 10,000 U/mL penicillin and 10 mg/mL streptomycin (Hyclone) in 0.85% NaCl. The cells were cultured at 37°C in an atmosphere of 5% CO₂ and 95% air, and the medium was changed regularly at 2 days intervals. All cells used in this study were between passages 37 and 40. Caco-2 and HT-29 cells were seeded in 24-well tissue culture plates (2 cm² per well) at 4.7 × 10⁵ cells per well. Once the culture reached 80% confluency, the medium was changed to one without antibiotics. After cultivation of lactic acid bacteria in MRS medium, bacterial cells (10⁸ CFU/mL) were collected by centrifugation, washed twice with PBS (pH 7.4), and resuspended in DMEM without serum and antibiotics. The bacterial cells were then applied to a Caco-2 and HT-29 cell monolayer and incubated at 37°C in 5% CO₂ for 2 h. After incubation, non-adherent bacteria were removed by washing twice with PBS. Cells with adhered bacteria were treated with a detachment solution containing 0.1% Triton X-100 and 0.1% trypsin-EDTA (Sigma) for 15 min. To calculate the number of adherent bacteria, the suspensions of the detached cells were plated onto MRS agar and incubated at 37°C for 48 h. The adhesion ability was estimated using the formula (the adhered bacteria/100 cells), where the Caco-2 and HT-29 cells were counted with hemocytometer (Thoma, Hirschmann, Germany).

To verify the safety of the strain, hemolysis analysis was conducted as described by Ryu and Chang (2013). Bacterial cells were inoculated into a BHI agar plate supplemented with 7% horse blood (MB CELL, Seoul, South Korea) and incubated under anaerobic conditions at 37°C for 24 h. Hemolysis was observed on the medium using Listeria monocytogenes as a positive control. In addition, the presence of biogenic amine genes in the genomic DNA of the selected strains was analyzed by multiplex PCR. The hdc (histidine decarboxylase) and tyrdc (tyrosine decarboxylase) genes were amplified using the following primer pairs: hdc: HDC3 (5-GATGGTATTGTTTCKTATGA-3) and HDC4 (5-CAAACACCAGCATCTTC-3); tyrdc: TD2 (5-ACATAGTCAACCATRTTGAA-3) and TD5 (5-CAAATGGA AGAAGAAGTAGG-3); 16S rRNA gene: 27F (5′-AGAGTTTGA TCMTGGCTCAG-3′) and 1492R (5′-GGTTACCTTGTT ACGACTT-3′) (O’Sullivan et al., 2015). In the multiplex PCR, each biogenic amine gene was amplified simultaneously with the 16S rRNA gene as a positive control for the PCR reaction in the tube by adding each corresponding primer set. Genomic DNA from bacterial strains was used as a template in the PCR reaction. The amplification program was as follows: 95°C for 5 min, followed by 32 cycles of 95°C for 45 s, 58°C for 45 s, and 72°C for 75 s, with a final extension at 72°C for 5 min. The genomic DNA of L. reuteri ATCC 23272 and Enterococcus faecalis KCCM 11729 were used as positive controls for the hdc and tyrdc genes, respectively.

Antioxidative activity of the three fractions of bacterial cells was measured using the DPPH inhibition assay (Das and Goyal, 2015). For preparation of intact cells, cell-free extracts (CFE), and cell-free supernatant (CFS), bacteria were pre-cultured for 12 h, the main culture was performed for 12 h, and the optical density at 600 nm was adjusted to 1.0. The bacterial cells were washed twice and resuspended in 0.85% saline solution to obtain intact cells. For CFE, sonication was performed using a sonicator (VP-050N; Taitec Corp., Saitama, Japan) for 10 min (5 s on/5 s off pulse; at 35% amplitude), and the cell debris was removed by centrifugation (10,000 × g at 4°C for 5 min). CFS was prepared by centrifugation (10,000 × g at 4°C for 10 min) of the bacterial culture. The supernatant was neutralized (pH 7.0) with 1 M NaOH and passed through syringe filters to remove the remaining cells. The ethanolic DPPH solution (100 μL, 0.4 mM) was mixed with 100 μL of bacterial sample or water (control) and incubated at 37°C in the dark for 30 min. The absorbance of the mixture was measured at 517 nm using a microplate reader and compared with the reference compound, ascorbic acid (AA).

For the nitric oxide (NO) production assay, RAW 264.7 cells, a murine macrophage line, were obtained from the KCLB and maintained in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin at 37°C in 5% CO₂. To analyze the anti-inflammatory activity of bacterial strains, the effect of two bacterial fractions on the production of NO in LPS-induced RAW 264.7 cells was determined using Griess reagent as described by Yu et al. (2019). To prepare heat-killed bacteria, the absorbance of the strains at 600 nm was adjusted to 1.0, and the bacteria were heat-killed at 90°C for 30 min. After centrifugation at 10,000 × g for 5 min, the cell pellets were rinsed twice with PBS and suspended in DMEM. To prepare cell-free lysates, sonication was performed using a sonicator (VP-050N; Taitec Corp., Japan) for 10 min, and the cell debris was removed by centrifugation. The supernatant was filtered through a 0.22 μm microfilter membrane (polypropylene, Whatman, Kent, United Kingdom). RAW 264.7 cells (5 × 10⁵ cells per mL in 96-well plates) were pre-treated with Escherichia coli lipopolysaccharide (LPS; 1 μg/mL, Sigma) with or without heat-killed cells or cell-free lysate for 24 h. Then, the supernatant from each well was mixed with an equal volume of Griess reagent and placed in the dark for 10 min at room temperature. The absorbance of each well was measured at 540 nm using a microplate spectrophotometer (BioTek, Winooski, VT, United States). Nitrite levels in the growth medium were calculated using a standard curve constructed using NaNO₂ in DMEM. Methyl arginine was used as a positive control to inhibit NO production.

The epithelial barrier model and transepithelial electrical resistance (TEER) measurements were conducted using the method described by Zaylaa et al. (2018). Caco-2 cells were seeded on 12 well Transwell^® inserts (polyester membrane with 0.4 μm pore size, 12 mm diameter; Costar, Corning Life Science, Kennebunk, ME, United States) at a density of 5 × 10⁴ cells per cm². The medium was changed every 2 days until confluency, when the optimal TEER value (≥200 Ω⋅cm²) was reached. TEER values were measured every 2 days using Millicell-ERS (Millipore, Billerica, MA, United States). Cells were then incubated in medium without antibiotics before experiments and treated in the insert with bacteria (10⁸ CFU/mL) 30 min prior to hydrogen peroxide (H₂O₂) administration (100 μM, to both the upper and lower sides). The TEER was measured every 30 min after sensitization. Results were expressed as% TEER compared with the initial TEER value at T₀ (before the addition of H₂O₂) for each insert using the formula: TEER (Ω⋅cm²)/initial TEER (Ω⋅cm²) × 100 (%).

Intestinal permeability was tested using the fluorescein isothiocyanate (FITC)-dextran method described by Miyauchi et al. (2012) with slight modifications. The flux of FITC-dextran (molecular weight 4,000; FD4; Sigma, St. Louis, MO, United States) across the monolayer could be used as an indicator of cell permeability. After TEER measurement of H₂O₂-induced Caco-2 cells, FITC-dextran (100 μg/mL) was added to the upper chamber and placed in the dark for 4 h at room temperature. Aliquots (100 μL) of the sample were obtained from the lower chamber of each well and transferred to a black 96-well opaque plate. The fluorescence intensity was determined using a fluorescence spectrometer (LS55, Perkin Elmer Instruments, Waltham, MA, United States) at 485 nm and 535 nm, excitation and emission wavelengths, respectively. Cells not treated with bacteria were used as controls, and results were expressed as% of control.

Rice solutions (5% and 10% w/v) were mixed with distilled water and sterilized at 121°C for 15 min. All strains were inoculated (10⁷ CFU/mL) in rice solution and incubated at 30°C for 24 h under anaerobic condition. After fermentation, samples were collected and stored at −20°C. The growth rate was measured by viable cell counts on MRS agar plate. The plates were incubated at 37°C for 48 h, and the viability was expressed as Log CFU/mL. The pH of each sample was monitored using a pH meter (Orion Versa, Thermo, United States).

The viscosity was measured at 17°C using an RVDV-II+Pro viscometer (Brookfield Engineering, Middleboro, MA, United States). Approximately 250 mL of each sample was placed in a glass beaker, and spindle no. 2–4 was used. Viscosity data were recorded at 30 s intervals and expressed in centipoise (cP). For comparison, commercial products such as fruit juice (orange, tomato), drinking yogurt (product A, product B), and semi-solid yogurt (plain, Greek) were analyzed together. Additionally, the reported viscosity data of sauce (a ketchup and a French mustard) were added from the ProSys Filling System web¹.

To conduct high-performance liquid chromatography (HPLC) analysis, a 5 mL sample was centrifuged at 10,000 × g for 10 min and the supernatant was filtered using 0.2 μm filters (Whatman, United Kingdom). Organic acids (lactic acid and acetic acid) were measured using HPLC 1260 Infinity (Agilent Technologies, United States) with Aminex HPX-87H column (300 × 7.8 mm; BioRad, CA, United States) under the operation setting: 0.008N H₂SO₄ in water as mobile phase, 20 μL of sample injection volume and 0.6 mL/min of flow rate. Sugars (glucose, maltose, maltotriose, maltotetrose, maltopentaose, and maltohexaose) were measured using HPLC Acme 9000 (Younglin, South Korea) with VN-50 4D column (150 × 4.6 mm; Shodex, Japan) under the operation setting: 67% acetonitrile (v/v) as mobile phase, 10 μL of sample injection volume and 0.3 mL/min of flow rate. Organic acids and sugars were detected using UV (215 nm) and RI detectors, respectively. Lactic acid, acetic acid (Wako, Japan), glucose, maltose (Junsei, Japan), maltotriose, maltotetrose, maltopentaose, and maltohexaose (TCI, Japan) were used as standards.

To prepare crude enzymes, bacteria were cultured at 37°C in 5% rice solution for 3 days and in MRS for 18 h, and the supernatants of bacterial cultures were harvested. For phytase activity, a modified Chalmers broth containing 1% sodium phytate (MCP) was used because the MRS medium contained high phosphate content. The cell-free supernatant fraction was collected by centrifugation at 10,000 × g for 10 min, and the methods were slightly modified from Giri et al. (2018). For α-amylase activity, 0.1 mL of supernatant was incubated with 0.1 mL of 1% starch (w/v) at 37°C for 30 min. The reaction was terminated using 0.6 mL of 3,5-dinitrosalicylic acid (DNS) reagent at 10 min intervals from the start of incubation, followed by boiling for 5 min. Absorbance at 540 nm was measured using a spectrophotometer (BioTek, United States). A standard curve was constructed using 2–10 mM glucose. One unit of amylase activity was defined as the amount of enzyme releasing 1 μmol of reducing sugars per minute under the assay conditions. For glucoamylase activity, 0.1 mL supernatant was incubated with 0.1 mL of 1% starch (w/v) at 37°C for 30 min. Then, 1 mL of a mixture of glucose oxidase/peroxidase reagent and o-dianisidine reagent in a glucose assay kit (GAGO20; Sigma-Aldrich, United States) was added at 10 min intervals from the start of incubation, followed by incubation at 37°C for 10 min. The reaction was terminated by adding 12 N H₂SO₄, and the absorbance was measured at 540 nm. A standard curve was constructed using 0.5∼2.5 mM glucose. One unit of glucoamylase activity was defined as the amount of enzyme that released 1 μmol glucose per minute under the assay conditions. For phytase activity, 0.1 mL) was mixed with 0.4 mL of substrate (3 mM sodium phytate in 0.2 M sodium acetate buffer, pH 4.0) and incubated at 37°C for 30 min. The reaction was stopped using 5% trichloroacetic acid (0.5 mL) at 10 min intervals from the start of incubation. Then, 0.5 mL of color reagent (1.5% (w/v) ammonium molybdate in 5.5% (v/v) sulfuric acid solution: 2.7% (w/v) ferrous sulfate solution = 1:4) was added. The absorbance at 700 nm was measured, and a standard curve was constructed using 2–10 mM KH₂PO₄. One unit of phytase activity was defined as the amount of enzyme releasing 1 nmol phosphate per minute under the assay conditions.

Nuclear magnetic resonance (NMR) spectroscopy was used to analyze the metabolites produced in fermented rice solution. For the preparation of NMR samples, fermented rice was centrifuged, and the supernatant was collected. The pH of the supernatant was adjusted to 7.0 ± 0.2, by the addition of 5 N NaOH. The neutralized supernatant (350 μL) was mixed with 350 μL of 1 mM sodium trimethylsilyl propanesulfonate (DSS) as an internal standard in 10% D2O. The mixture (700 μL) was transferred to NMR tubes and subjected to ¹H-NMR analysis. ¹H-NMR spectra were recorded on an Avance 500-MHz spectrometer (Bruker BioSpin, Karlsruhe, Germany). The metabolite peaks in the NMR spectra were identified and their concentrations were calculated using the Chenomx NMR suite 8.4 library software (Chenomx Inc., Edmonton, AB, Canada).

The acid and bile tolerance levels of all strains were measured and compared with those of L. rhamnosus GG (LGG) as positive controls. L. fermentum DSM 20052 (LFT) and KCCM 35469 (LFC) were used as the type and commercial strains, respectively. As shown in Table 1, the viability of LGG decreased significantly after 90 min and 180 min of incubation at pH 2.5. In the case of LFT and LFC, their acid and bile tolerance levels also decreased. In contrast, most L. fermentum isolates showed higher viability than LGG, LFT, and LFC. Meanwhile, under bile salt conditions, only 10 strains (MG7011, MG4244, MG4261, MG5154, MG4500, MG4510, MG4531, MG4533, MG4535, and MG4536) showed higher levels of tolerance than the positive controls. Therefore, these 10 strains were selected for the next experiment.

The adhesion ability of the selected strains to intestinal epithelial cells was measured by incubating bacterial cells with epithelial cell lines, Caco-2 and HT-29 cells. As shown in Figure 1A, three strains (MG7011, MG5154, and MG4531) showed higher adhesion to Caco-2 cells than WCFS1 (4000 CFU/100 cells) and five strains (MG4244, MG4261, MG4500, MG4510, and MG4535) showed high adhesion ability comparable to LGG (2500 CFU/100 cells). As shown in Figure 1B, only four strains (MG7011, MG4244, MG4261, and MG4531) showed high adhesion abilities comparable to those of WCFS1 (2400 CFU/100 cells) and LGG (1100 CFU/100 cells). Consequently, the five strains (MG7011, MG4244, MG4261, MG4500, and MG4531) were selected for good adhesion to intestinal epithelial cells, Caco-2 and HT-29.

To evaluate the safety of L. fermentum strains, the hemolytic activity and biogenic amine genes were analyzed. As shown in Figure 2A, the five strains selected above, MG7011, MG4244, MG4261, MG4500, and MG4531, did not show any clear zones around the cell drop, whereas Listeria monocytogenes as positive control showed a clear zone that could be interpreted as hemolytic activity. In addition, as shown in Figure 2B, hdc and tydc genes were not detected in the five selected strains which produce histamine and tyramine, respectively, but clear bands appeared in positive controls. These results reveal that selected strains are considered as safe with non-hemolytic activity and no hdc and tyrdc genes.

To evaluate antioxidative activity, DPPH scavenging capacities of isolates were measured by using intact cells, cell-free extracts (CFE), and cell-free supernatants (CFS). In case of intact cells, all L. fermentum strains exhibited higher DPPH scavenging activity (>15%) than positive control, LGG (12.4%) as shown in Figure 3. Among them, MG4500 and MG4531 showed the highest DPPH scavenging activity (22.6% and 22.1%, respectively). In the case of CFE, all L. fermentum strains showed >28% activity, which was significantly higher than that of LGG (17.9%). In addition, all strains showed similar levels of DPPH scavenging activity as LGG when CFS were used. These results show that the five selected strains of L. fermentum have excellent antioxidative activity compared to LGG and WCFS1.

To evaluate the anti-inflammatory activities of the selected strains, the inhibitory activities of heat-killed and lysate cells were tested on NO production in LPS-induced RAW 264.7 cells. As shown in Figure 4, treatment with LPS markedly increased the NO production (6.47 ± 0.40 μM), compared with the control that was not treated with LPS (0.39 ± 0.19 μM). In addition, treatment with 5 μM and 20 μM methyl arginine, which is a nitric oxide synthase inhibitor, inhibited the production of NO in a dose-dependent manner (4.15 ± 0.28 μM and 2.82 ± 0.71 μM, respectively). In cases of heat-killed strains, all tested samples including LGG and WCFS1 exhibited significant inhibitory activities on NO production compared with the LPS-treated group (p < 0.05). Particularly, three heat-killed strains (MG7011, MG4261, MG4531; <2.4 μM of NO) showed significantly stronger inhibitory activity than LGG (3.0 μM). In lysates, LGG, MG4244, and MG4261 (3.8 μM, 2.8 μM, and 4.5 μM, respectively) showed significant inhibitory effects compared with the LPS-treated group (p < 0.01 for LGG and MG4261, p < 0.001 for MG4244).

To test the protective activities of selected strains for epithelial monolayers, bacterial cells and H₂O₂ were incubated with Caco-2 cells after confluency, and the TEER value and FITC-dextran flux were measured. As shown in Figure 5A, the addition of H₂O₂ to Caco-2 cell monolayers caused a decrease in TEER (42.34%) along with incubation time, but LGG and the three selected strains significantly protected against H₂O₂-induced epithelial damage. In Figure 5B, the protective effects of MG7011, MG4244, and MG4261 were compared, which resulted in an increase in TEER (74.06%, 81.15%, and 77.36%, respectively) compared to LGG (72.64%) after 120 min. Additionally, in Figure 5C, the paracellular permeability of Caco-2 cell monolayers was measured using FITC-dextran flux. Each well was incubated with FITC-dextran for 4 h, and the fluorescence transmitted through Caco-2 cells was measured as% based on the control (treated with only H₂O₂). All bacterial strains showed lower fluorescence (<79.3%) than the control group (p < 0.01), and MG4244 showed the lowest values (54.2%) compared to LGG (79.2%) (p < 0.05). The results show that MG7011, MG4244, and MG4261 strains have protective activities for epithelial cells against oxidative stress induced by H₂O₂ and lower the permeability of Caco-2 cell monolayers. In conclusion, the three strains, MG7011, MG4244, and MG4261, were selected as potential probiotic candidates.

To prepare different types of fermented rice, such as drink or semi-solid type, 5% and 10% of rice solutions were fermented by the selected strains at 30°C for 24 h, and fermentative properties (viable cell, pH, and viscosity) were measured. As shown in Figure 6, three strains grew well in both 5% and 10% rice solutions from the initial cell count of 6.7 Log CFU/mL to more than 7.2 Log CFU/mL after 24 h – especially, MG7011 showed superior cell growth. The scales of increase in viable cells and decrease in pH for each strain were consistent. Meanwhile, as shown in Figure 7, 5% rice solution fermented with MG7011, MG4244, or MG4261 showed lower viscosity levels respectively (269, 787, and 261 cP, respectively) than non-fermented rice (NF, 861 cP). In contrary, 10% rice solution fermented with three strain showed higher viscosity than that of 5% rice. In the case of 5% fermented rice, viscosities were observed between the two groups of commercial products as follows: higher than several drinking beverages (orange juice, tomato juice, drink A, and drink B), but lower than plain or Greek yogurts. In contrast, viscosities of 10% fermented rice were higher than plain or Greek yogurts and lower than commercial ketchup or mustard products. In summary, these results show that the viscosities of fermented rice can be controlled by the initial concentration of rice in between 5% and 10%, and MG7011 has the highest growth rate and pH change in both 5% and 10% rice solutions, revealing its good adaptability in rice medium.

Enzyme activities related to rice utilization were analyzed against MG7011, MG4244, and MG4261: α-amylase and glucoamylase to hydrolyzed starch, and phytase to hydrolyzed phytic acid (Table 2). In the case of α-amylase, MG7011 showed activity in 5% rice solution (0.11 U) and MRS (0.08 U), while MG4261 showed lower levels in MRS (0.02 U). In the case of glucoamylase, three strains showed 0.01∼0.02 U in rice solution. In the case of phytase, MG7011 showed activity in MCP (17.65 U). This result revealed that the MG7011 strain can produce α-amylase, glucoamylase, and phytase.

To analyze the changes in various metabolites during rice fermentation, NMR analysis was conducted. As shown in Figure 8, after fermentation of 5% and 10% rice solution by MG7011, MG4244 and MG4261 for 24 h, the concentrations of amino acids, alcohols, organic acids, phenolics, and myo-inositol increased. In the case of sugars, glucose, fructose, and maltose increased, and sucrose decreased in fermented rice by all strains. Sucrose in rice can be hydrolyzed into fructose and glucose, and they are converted into several products such as lactate, acetate, ethanol, and glucitol. Maltose is produced by the bacterial amylase from rice starch. All strains produced significant amounts (>2 folds) of γ-aminobutyric acid (GABA). As compared in Supplementary Figure 2, in amino acids, glycine and histidine decreased, and proline, threonine, and alanine increased due to bacterial metabolism. Most samples showed significant increases in certain phenolic compounds, including 4-hydroxybenzoate, chlorogenate, ferulate, and vanillate. In particularly, myo-inositol was produced by all strains, and MG7011 resulted in 4.4 folds higher synthesis in 5% rice. Indeed, MG7011 efficiently hydrolyzed phytic acid as shown in Table 2. Metabolite patterns in the 5% and 10% rice solutions were generally consistent. Based on the results obtained in rice fermentation as well as probiotic traits, we selected L. fermentum MG7011 as the best strain for probiotic starter for fermentation of rice beverage.