MiCoP: Enhancing the sensory and nutritional properties of faba bean protein through fermentation with lactic acid bacteria and Bacillus spp.

Enhancing the sensory and nutritional properties of faba bean protein through fermentation with lactic acid bacteria and Bacillus spp N. Larsen a,* , M.G. Henriksen a , C. Crocoll b , Q. Li a , R. Lametsch a , M.M. Poojary a , L. Krych a , M.A. Petersen a , L. Jespersen a a Department of Food Science, University of Copenhagen, Rolighedsvej 26, 1958 Frederiksberg C, Denmark b Department of Plant and Environmental Sciences, University of Copenhagen, Thorvaldsensvej 40, 1871 Frederiksberg C, Denmark ARTICLE INFO Keywords: Faba protein Fermentation Aroma compounds Amino acids Kokumi Vicine-convicine ABSTRACT Faba beans are rich in protein and fiber, but their consumption is limited by off-flavors, and antinutritional compounds. This study explored fermentations with lactic acid bacteria (LAB) and Bacillus spp. as a strategy to improve the sensory and nutritional qualities of faba bean protein concentrate (FPC). Fermentations were performed with the strains isolated from African fermented foods, including Limosilactobacillus fermentum NSB2, Lactiplantibacillus argentoratensis 12-27B and Bacillus velezensis G17, and with Bacillus subtilis Natto. Strain identification was performed using Oxford Nanopore whole-genome sequencing. Samples of fermented FPC (8 % w/v) were analyzed after 8 h, 24 h, and 48 h fermentation for volatile aroma compounds (VOCs), organic acids, free amino acids, γ-glutamyl peptides, and vicine-convicine. All strains effectively reduced aldehydes associated with beany off-flavors, such as hexanal. LAB, particularly L. fermentum NSB2, promoted alcohol formation, B. velezensis G17 produced the greatest amounts of diacetyl and acetoin, while B. subtilis Natto generated the largest amounts of esters. L. argentoratensis 12-27B was the most efficient producer of lactic acid, whereas B. velezensis G17 uniquely generated high levels of propionic acid. Bacillus spp. exhibited strong proteolytic activity during FPC fermentation, producing free amino acids, including essential ones. Several γ-glutamyl dipeptides were increased in fermentations, highest with Bacillus spp., suggesting potential kokumi enhancement. Notably, L. argentoratensis 12-27B and B. velezensis G17 significantly reduced vicine and convicine, the primary antinutritional compounds in faba beans. These findings highlight the potential of targeted fermentation to enhance the sensory quality and nutritional value of faba bean protein for plant-based food applications. 1. Introduction Faba bean (Vicia faba) is a highly nutritious legume, rich in proteins, dietary fibers, and bioactive compounds, such as minerals and vitamins (Sharan et al., 2021). However, broader utilization of faba bean (FB) remains limited due to undesirable sensory attributes, particularly beany and rancid off-flavors and the presence of antinutritional factors. The beany aroma is mainly attributed to the volatile organic compounds (VOCs), including aldehydes, such as hexanal, as well as specific alcohols and methoxypyrazines (Fischer et al., 2022; Ritter et al., 2022). Among them, aldehydes are considered key contributors to the aroma profile in FB due to their high concentrations and low odor thresholds (Karolkowski et al., 2023; Oomah et al., 2014; Ritter et al., 2024). Pyrimidine glycosides, vicine and convicine, are of particular concern as antinutrients due to their potential to trigger hemolytic anemia in individuals with glucose-6-phosphate dehydrogenase (G6PD) deficiency (Khazaei et al., 2019; Rizzello et al., 2016). FB seeds are commercially processed into various ingredients, including flour, protein concentrate, or protein isolate (Sharan et al., 2021). Among these, FPC, produced via dry fractionation, is especially relevant for food applications due to its high protein content. Yet, FPC has been reported to exhibit the most intense off-flavors and bitterness among faba-derived ingredients (Schutyser et al., 2015; Tuccillo et al., 2022b). Microbial fermentation has emerged as a promising strategy to improve nutritional and sensory qualities of FB ingredients. The metabolic activity of starter cultures during fermentation lead to the production of aroma compounds that mask off-flavors, increase in protein digestibility, release of amino acids, and degradation of antinutritional factors (Fischer et al., 2022; Ritter et al., 2022; Sed´ o Molina et al., 2024). Among these, proteolytic activity is particularly important in developing * Corresponding author. E-mail address: [email protected] (N. Larsen). Contents lists available at ScienceDirect International Journal of Food Microbiology journal homepage: www.elsevier.com/locate/ijfoodmicro https://doi.org/10.1016/j.ijfoodmicro.2025.111549 Received 16 July 2025; Received in revised form 4 November 2025; Accepted 24 November 2025 International Journal of Food Microbiology 446 (2026) 111549 Available online 25 November 2025 0168-1605/© 2025 Published by Elsevier B.V. flavor-enhancing molecules, such as amino acids contributing to umami taste (e.g., glutamic and aspartic acid) and γ-glutamyl peptides that evoke kokumi sensations like mouthfulness and complexity (Hillmann et al., 2016; Li et al., 2022; Wang et al., 2022; Zhang et al., 2019). Lactic acid bacteria (LAB) and Bacillus spp. are commonly used starter cultures in legume fermentations. LAB, such as Lactiplantibacillus plantarum and Lactobacillus acidophilus have been shown to reduce beany and “haylike” aromas and bitterness in pea protein isolates and soy-based cakes (Razavizadeh et al., 2022; Shi et al., 2021). Bacillus spp. (B. subtilis, B. amyloliquefaciens, B. pumilus, B. licheniformis, etc.) known for their strong proteolytic capabilities, contributed to aroma development, umami perception and texture in traditional West African fermented products like soumbala, and soybeans (Ouoba et al., 2004; OwusuKwarteng et al., 2020; Zhang et al., 2021). Recent studies focusing on FB fermentation have demonstrated that LAB can substantially improve nutritional properties of faba-derived ingredients. L. plantarum, for instance, reduced vicine content by 85 %, enhanced protein digestibility, and increased free amino acids in faba protein (Coda et al., 2015). Similar benefits have been reported for Pediococcus pentosaceus, Leuconostoc kimchi and Weissella spp. in FB flour fermentations (Rizzello et al., 2019; Verni et al., 2019). Despite these promising findings, the impact of fermentation on FB protein remains insufficiently characterized. The outcomes are highly strain-specific and influenced by microbial metabolism, substrate characteristics, and process parameters. In this study, we hypothesized that the sensory and nutritional properties of FB protein could be significantly enhanced through microbial fermentation by modulating of VOCs, organic acids, free amino acids, kokumi peptides, and by reducing vicine and convicine. Fermentations were carried out using strains from two LAB species, Limosilactobacillus fermentum NSB2 and Lactiplantibacillus argentoratensis 1227B, and two Bacillus spp., B. velezensis G17 and B. subtilis Natto. The aim was to identify microbial candidates capable of improving the sensory and nutritional value of FB protein for use in plant-based food applications. 2. Materials and methods 2.1. Fermentation medium Faba protein concentrate was provided by Vestkorn Milling A/S (Tau, Norway). According to the manufacturer's specification, the composition per 100 g of FPC was 65 g protein, 10 g carbohydrates (including 1.9 g sugars), 4.5 g fat and 14 g dietary fiber. For fermentation experiments, 8 % (w/v) suspension of FPC was prepared in distilled water and sterilized by autoclaving at 121 ◦C for 20 min immediately prior to inoculation. 2.2. Bacterial strains and chemicals Three bacterial strains in this study were previously isolated from African fermented foods as follows: L. fermentum NSB2 from sorghum malt (Sawadogo-Lingani et al., 2010), L. argentoratensis 12-27B from Fura (fermented millet dough) (Owusu-Kwarteng et al., 2012) and B. velezensis G17 isolated from Kantong, a fermented Kapok tree seed product (Kpikpi et al., 2014). Additionally, B. subtilis Natto was obtained from commercial Natto spores (LUVI Fermente KG, Lenzing, Austria). Before the experiments, LAB strains were grown on Man, Rogosa Sharpe agar (MRS, Becton Dickinson A/S, Denmark) and incubated at 30 ◦C for 48 h. Strains of Bacillus spp. were grown on TSA (Tryptone Soya agar) at 37 ◦C for 24 h. All reagents for the experiments were purchased from Sigma-Aldrich (Denmark) or ThermoFisher Scientific (Denmark) unless otherwise stated. 2.3. Strain identification and genome annotation Taxonomic identification of bacterial strains was conducted using whole-genome sequencing (WGS) of native DNA with the Oxford Nanopore Technology (ONT) (Soto-Serrano et al., 2024; Wiedenbein et al., 2023). Shortly, the genomic DNA was extracted from bacterial cultures using the Bead-Beat Micro AX Gravity kit (cat. 106–100 M1; A&A Biotechnology, Gdynia, Poland). The DNA library was prepared according to the Native Barcoding Sequencing protocol (SQKNBD114.24, version: NBE_9169_V144_revN_ 15Sep2022). Sequencing was performed on ONT GridION and PromethION 2 Solo platforms using R10.4.1 Flow cells in combination with MagAttract HMW DNA kit (Qiagen, Germantown, MD, USA). Quality scores of the ONT reads were assessed using a custom function (https://github.com/ farhadm1990/ bactflow). Strains were identified using GTDB-TK v2.4.0 (https://github.com/ Ecogenomics/GTDBTk) with Genome Taxonomy Database (GTDB) release r220 (Chaumeil et al., 2022). Genome annotation for the tested strains and reference strains were performed using BAKTA v1.9.4 (Schwengers et al., 2021). Strain L. argentoratensis 12-27B was previously identified as L. paraplantarum based on the 16S rRNA gene sequencing and multiplex PCR (Owusu-Kwarteng et al., 2012). In this study, L. argentoratensis 12-27B was differentiated from closely related species (Table S1, Supplementary data) based on sequence analysis of the pheA, rpoA, and recA genes, using type strains for comparison, as well as average nucleotide identity (ANI) analysis conducted with the MUMmer-based ANIm method implemented in PYANI v0.2.12 (Li et al., 2020b; Pritchard et al., 2016). Genome sequences were deposited into the Genebank database under accession numbers CP193773-CP193774 for L. fermentum NSB2, CP193775-CP193776 for L. argentoratensis 12-27B, and CP193772 for B. subtilis Natto. Strain B. velezensis G17 (Syn. Pro81) has been identified previously and deposited in GenBank under accession no. CP121712 (Wiedenbein et al., 2023). 2.4. Fermentation conditions and sample collection Before fermentation, the LAB strains were propagated in MRS broth at 37 ◦C, while Bacillus spp. were grown in TSB at 37 ◦C with shaking (220 rpm). Cells were harvested by centrifugation, washed, and resuspended in a 0.9 % (w/v) NaCl solution. The FPC was inoculated with cell suspensions at a final concentration of 10 5 CFU/mL and incubated at 30 ◦C with shaking (200 rpm). Control fermentations, without inoculation, were conducted in parallel. Samples were collected at 0 h (control) and after 8, 24, 48, and 72 h for determination of viable counts and pH, and at 8, 24, and 72 h for biochemical analyses. Each time point was fermented as an independent batch. All fermentations were carried out as single-strain cultures in biological triplicates. Collected samples were either stored at −60 ◦C for the analyses of VOCs and γ-glutamyl peptides or freeze-dried for the analyses of organic acids, free amino acids (FAAs), and vicine-convicine (v-c) content. 2.5. Measurements of viable counts and pH Bacterial viable counts were determined by plating serial dilutions of cell suspensions onto MRS agar for LAB or TSA for Bacillus spp. Agar plates were incubated at conditions described above. The pH in fermented samples was recorded at time-points of sample collection using FiveEasy Plus FP20 pH meter (Mettler Toledo, Glostrup, Denmark). 2.6. Analysis of volatile compounds by GC–MS VOCs were analyzed using gas chromatography–mass spectrometry (GC–MS) as previously described (Srimahaeak et al., 2022). Briefly, for headspace sampling, 35 g of fermented FPC and unfermented controls were combined with 1 mL of an internal standard solution in water (4N. Larsen et al. International Journal of Food Microbiology 446 (2026) 111549 2 methyl-1-pentanol, 5 mg/L) and homogenized. VOCs were captured on Tenax-TA traps (Markes International, Llantrisant, UK) by purging with nitrogen (100 mL/min) for 45 min at 40 ◦C. Separation of volatiles was performed on a DB-Wax capillary column (30 m ×0.25 mm ×0.5 μ m) using an Agilent 7890 A GC system interfaced with a 5975C VL MSD with a Triple-Axis detector (Agilent Technologies, Palo Alto, CA, USA). Data extraction was carried out using the PARADISe software (Quintanilla-Casas et al., 2023). Volatile compounds were identified based on retention indices (RI) of n-alkanes (C6–C22) and the NIST05 database. Identifications were further confirmed by comparing the RI of authentic reference compounds or published RI values. VOC concentrations ( μ g/kg) were determined semi-quantitatively by normalizing the peak area of each compound to that of the internal standard and multiplying by the internal standard concentration. 2.7. Determination of organic acids by HPLC Organic acids (acetic, lactic, citric, and propionic) were extracted using the method described by Vong et al. (2016). Briefly, 150 mg of freeze-dried sample was mixed with 1.5 mL of 0.01 M sulfuric acid and incubated at 30 ◦C for 1 h with shaking at 1400 rpm. The supernatant was collected by centrifugation and stored at 5 ◦C for 24 h to allow complete protein precipitation. Concentration of organic acids in the supernatants was determined by HPLC using an Aminex HPX-87H column (Bio-Rad Laboratories, Denmark) and a refractive index detector (RID; Agilent 1100 series, Waldbronn, Germany) as described by Srimahaeak et al. (2022). Organic acid concentration was expressed in mg per g of FPC. 2.8. Determination of free amino acids by UHPLC FAAs were extracted following a protocol by Hildebrand et al. (2020). Briefly, 100 mg of freeze-dried sample was mixed with 5 mL of 6.25 % w/v trichloroacetic acid and incubated at 4 ◦C for 3 h. The supernatant was collected by centrifugation (10,000 ×g for 10 min), diluted with sterile Milli-Q water (1:1, v/v), and neutralized with 1 M NaOH. To each diluted sample, 6-aminocaproic acid (50 μ M) was added as an internal standard (1:1, v/v), and the mixture was filtered through a 0.2 μ m cellulose syringe filter (Phenomenex, Denmark). The amino acid analysis was carried out using an ultra-high performance liquid chromatography (UHPLC) (UltiMate 3000 UHPLC, Thermo Fischer Scientific, CA, USA) equipped with a reversed-phase UHPLC column (Advanced Bio AAA column, Agilent, Glostrup, Denmark) and fluorescence detection (FLD) as described by Schmidt et al. (2020). Primary amino acids were derivatized with a solution of 7.5 mM o-phthaldialdehyde (OPA)/ 225 mM 3-mercapto-propionic acid (MPA) prepared in 0.1 M borate buffer. Secondary amino acids were derivatized with 9.7 mM 9-fluorenyl-methyl chloroformate solution prepared in acetonitrile. Standard curves were constructed using amino acid standards for protein hydrolysates (A9781, Merck Life Science A/S, Denmark) in 0.1 M HCl added 6aminocaproic acid (1:1 v/v) to achieve concentration range of 0–100 μ M. The amounts of amino acids were expressed as mg per 100 g of FPC. 2.9. Quantification of γ-glutamyl peptides by UHPLC-MS/MS UHPLC-MS/MS analysis of targeted γ-glutamyl peptides was carried out as described by Li et al. (2020a). Shortly, FPC samples were homogenized using Mixer Mill MM 400 (Retsch, GmbH, Germany). The supernatant was collected, mixed with 10 % (w/v) trichloroacetic acid (TCA), and incubated at 4 ◦C for 30 min. Identification of γ-glutamyl dipeptides in the supernatant was performed using a Dionex Ultimate 3000 UHPLC system (Thermo Fisher Scientific, Hvidovre, Denmark) coupled with a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific, Roskilde, Denmark). External calibration curves were constructed using reference γ-glutamyl dipeptides (γ-Glu-Glu, γ-Glu-Gln, γ-Glu-Gly, Glu-Ala, γ-Glu-Trp, γ-Glu-Val, γ-Glu-Leu, γ-Glu-Met, γ-GluPhe, and γ-Glu-Tyr), as well as tripeptides and a tetrapeptide (γ-Glu-ValGly, glutathione (GSH) and γ-Glu-GSH). Quantification of peptides was performed by comparing the peak areas of the extracted ions with the corresponding calibration curves using the Thermo Xcalibur Quan Browser 4.0 software (Thermo Fisher Scientific Inc., USA). Concentration of the γ-glutamyl peptides was expressed in μ mol per kg of FPC. 2.10. Determination of vicine and convicine by LC-MS Vicine and convicine were extracted following Purves et al. (2018) with modifications. Briefly, 50 mg of freeze-dried, ground sample was mixed with 500 μ L of 70:30 acetone:water containing 2000 μ M uridine (internal standard). After 30 min shaking and centrifugation, the supernatant was collected and diluted with extraction solvent (without uridine). Chromatographic separation was adapted from Bj¨ ornsdotter et al. (2021) and performed on an Agilent 1290 Infinity II UHPLC with an ACQUITY BEH Amide column (Waters). Detection was performed on an Ultivo Triple Quadrupole MS (Agilent Technologies) with Jetstream ESI in negative mode. Quantification was carried out using MassHunter QQQ software (v10.1), based on external calibration curves. Concentrations of vicine and convicine were expressed as mg per g of FPC. 2.11. Data analysis and statistics Differences between the experimental data were evaluated using an unpaired two-tailed Student's t-test at a 95 % confidence level based on three biological replicates. Additionally, significant differences in the HPLC data for organic acids, the UHPLC data for FAA and the UHPLCMS/MS data for γ-glutamyl peptides, were analyzed using one-way analysis of variance (ANOVA) with Least Squares Means differences (confidence level of 95 %), followed by Tukey's posthoc test (JMP software, SAS Institute Inc., Cary, USA). A heatmap of VOCs profiles detected by GC–MS was generated using the gplots package in R software (version 3.6.1; http://www.r-project.org/). Data were standardized by z-score transformation per compound, and hierarchical clustering (UPGMA, average linkage, correlation distance) to visualize strainand time-dependent differences. 3. Results 3.1. Growth of bacteria in FPC Bacterial strains were able to grow in FPC, reaching viable cell counts of 7.97–8.78 Log CFU/mL after 72 h of fermentation, with the highest counts observed for L. fermentum NSB2 and L. argentoratensis 12-27B (Fig. 1A). Additionally, L. argentoratensis 12-27B showed the strongest acidification capacity, lowering the pH of FPC from 6.33 to 5.34 within 24 h. However, the lack of measurements between 8 and 24 h limit detailed characterization of the acidification dynamics (Fig. 1B). Conversely, fermentation with B. velezensis G17 led to a slight increase in pH to 6.49 at the end of fermentation. 3.2. Volatile compounds Fig. 2 illustrates changes in VOC classes during FPC fermentation. Aldehydes were abundant in FPC, accounting for 61.9 % of the total VOCs, followed by alcohols (16.2 %), ketones (13.8 %), and the minor amounts of esters and acids (<0.4 %). Relative proportions of aldehyde consistently declined across all fermentations, with the most substantial reduction observed for Bacillus spp. (<1 % after 72 h). LAB fermentations, particularly L. fermentum NSB2, significantly increased alcohol content up to 80.5 %. In contrast, Bacillus spp. fermentations were characterized by rise in ketones, largest with B. velezensis G17 (89.7 % after 72 h). Other VOC classes, such as alkanes, alkenes, furans and heterocyclic compounds (in total 7.5 % in FPC), remained largely unchanged during fermentation or were not associated with aroma N. Larsen et al. International Journal of Food Microbiology 446 (2026) 111549 3 compounds. A total of 84 VOCs were identified in FPC fermentations, with 49 compounds showing significant changes by at least one of the strains. Changes in concentrations of individual VOCs before fermentation (0 h) and at the sampling time points (8, 24 and 72 h) are illustrated in the heatmap (Fig. 3), with the corresponding values shown in Table S2 (Supplementary data). The predominant aldehyde in FPC before fermentation was hexanal (>1000 μ g/kg). Other C3-C7 straight chain aldehydes, branched-chain aldehydes and aromatic benzaldehyde were found in lesser amounts (from 3 to 100 μ g/kg). Fermentation with LAB and Bacillus spp. resulted in a major reduction of all aldehydes already after 8 h fermentation. For example, hexanal was reduced by more than 100-fold compared to control incubations, reaching 5–10 μ g/kg after 72 h. Due to their volatile nature, aliphatic aldehydes also declined in non-inoculated controls although to a lesser extent (2–4 fold after 72 h). VOCs belonging to other chemical classes remained relatively stable in controls throughout incubation. Fermentation with LAB, particularly L. fermentum NSB2, led to an increase in several straight-chain C3-C8 alcohols within the first 24 h of fermentation. Abundant alcohols, such as 1-hexanol (208 μ g/kg), 3methyl-1-butanol (69 μ g/kg), and 1-pentanol (36 μ g/kg), were increased by 2–3 fold during incubation. Concurrently, less abundant alcohols (<10 μ g/kg), including 2-propanol, ethanol, butanol, heptanol, and benzyl alcohol, increased by 2–20 fold predominantly in fermentations with L. fermentum NSB2. Conversely, fermentations with Bacillus spp. resulted in only minor increases in 3-methyl-1-butanol, 1-heptanol, and benzyl alcohol. Fermentations with Bacillus spp. resulted in production of various ketones, particularly 3-hydroxy-2-butanone (acetoin), 2,3-butanedione Fig. 1. Changes in viable counts (Log CFU/mL) (A) and pH values (B) during fermentation of faba protein concentrate (FPC, 8 % w/v in water) with L. fermentum NSB2, L. argentoratensis 12-27B, B. subtilis Natto, and B. velezensis G17. Mean values and SD (bars) from triplicate fermentation experiments are shown. Fig. 2. Distribution of the major classes of volatile compounds (VOCs, %) in faba protein concentrate (FPC, 8 % w/v in water) before (0 h) and after 8 h, 24 h and 72 h fermentation with L. fermentum NSB2 (Lf NSB2), L. argentoratensis 12-27B (La 12-27B), B. subtilis Natto (Bs Natto), B. velezensis G17 (Bv G17) and in non-inoculated controls. Means obtained from triplicate experiments are shown. N. Larsen et al. International Journal of Food Microbiology 446 (2026) 111549 4 (diacetyl), and 2-butanone. Among these, acetoin, initially present at 25 μ g/kg in FPC, became the predominant ketone, reaching 1351 μ g/kg and 6146 μ g/kg after 72 h of fermentation with B. subtilis Natto and B. velezensis G17, respectively. Similar ketones increased by 5–10 fold in fermentations with L. argentoratensis 12-27B. Additionally, butanone derivatives, such as 2-acetoxy-3-butanone (acetoin acetate) and 3-hydroxy-3-methyl-2-butanone, which were initially present at trace levels (<1 μ g/kg), increased by B. velezensis G17 to 57 μ g/kg and 20 μ g/ kg, respectively. L. fermentum NSB2 was the only strain to reduce 2-propanone (acetone) from 118 μ g/kg to 3.9 μ g/kg by the end of fermentation. Production of esters was primarily observed for B. subtilis Natto with the most significant increases in methyl derivatives of acetic, butanoic, and propanoic acids (initially <5 μ g/kg and to 9–22 μ g/kg after 72 h). 3.3. Organic acids Fig. 4 shows changes in the concentration of acetic, lactic, citric and propionic acids during FPC fermentation. Acetic acid increased from 1.5 mg/g to 6.0–11.7 mg/g after 72 h in all fermentations. Similarly, lactic acid, initially present at 1.1 mg/g in FPC, was produced by all strains, with the highest concentration of 17 mg/g observed after 24 h fermentation with L. argentoratensis 12-27B. Citric acid, initially predominant at 10.8 mg/g, was reduced to trace levels in all fermentations, most rapidly by L. fermentum NSB2 (0.58 mg/g after 24 h). An exception was L. argentoratensis 12-27B, which retained higher residual level of citric acid (4.4 mg/g after 72 h). Propionic acid was primarily produced by B. velezensis G17 (3.5 mg/g after 24 h) and to a lesser extent by B. subtilis Natto (1.7 mg/g after 72 h), while it was not detected in LAB fermentations. 3.4. Free amino acids Table 1 presents the concentration of FAAs in FPC before (0 h) and after 24 and 72 h fermentation. Changes in FAAs after 8 h of fermentation and in control incubations were insignificant (data not shown). Arginine (Arg) was the predominant FAA, comprising 842 mg/100 g out of 1014 mg/100 g of total amino acids in unfermented FPC. Concentrations of aspartic acid (Asp), glutamic acid (Glu), glycine (Gly), and lysine (Lys), were moderate, ranging from 11 to 67 mg/100 g. Other FAAs, serine (Ser), histidine (His), alanine (Ala), tyrosine (Tyr), valine (Val), tryptophan (Trp), phenylalanine (Phe), and leucine (Leu) were either present in smaller quantities (<10 mg/100 g) or not detected (threonine (Thr), methionine (Met), and isoleucine (Ile)). Essential FAA accounted for a minor fraction (50.8 mg/100 g) of the total FAA. Total amino acids and most individual FAAs were significantly reduced by LAB. Decrease of total FAAs by L. fermentum NSB2, was linked to utilization of arginine (Arg), which decreased to 1.97 mg/100 g after 72 h. Arginine levels were also reduced by more than twofold in fermentations with Bacillus spp. The ability of L. fermentum NSB2 and Bacillus spp. to degrade arginine was consistent with the presence of the key genes involved in arginine catabolism (Table S3, Supplementary Fig. 3. Heat map and hierarchical cluster analysis of individual VOCs in faba protein concentrate (FPC, 8 % (w/v) in water) before (0 h) and after 8 h, 24 h and 72 h fermentation with L. fermentum NSB2 (Lf NSB2), L. argentoratensis 12-27B (La 12-27B), B. subtilis Natto (Bs Natto), B. velezensis G17 (Bv G17) and in non-inoculated controls as detected by GC–MS. Sample codes denote the strains and the incubation period. The heat map was generated using mean values from triplicate experiments. The color scale indicates values normalized to a range of −4 (blue, lowest concentration) to +4 (red, highest concentration) within a row. N. Larsen et al. International Journal of Food Microbiology 446 (2026) 111549 5 Data). Genes from the arc cluster (arcA and arcC) were identified in L. fermentum NSB2, while the roc cluster (rocF and rocD) was present in Bacillus spp. Similar genes were lacking in L. argentoratensis 12-27B genome, consistent with its inability to metabolize arginine during fermentation. Fermentation with Bacillus spp. led to production of ten FAAs, including essential ones (His, Ile, Leu, Lys, Phe, Thr, Trp, and Val), which resulted in a significant increase in essential and total FAA content. The most abundant FAAs were Ala, Tyr, Val, Trp, Phe, Ile, Leu, and Lys, particularly in samples fermented with B. subtilis Natto, where their concentrations ranged from 107 mg/100 g (Trp) to 502 mg/100 g (Leu) after 72 h. 3.5. Production of γ-glutamyl peptides In total eight γ-glutamyl dipeptides were detected in FPC samples (Table 2). Five dipeptides were identified in FPC before fermentation, among them, γ-Glu-Leu (2.4 μ mol/kg) and γ-Glu-Phe (2.1 μ mol/kg) were the most abundant, while γ-Glu-Val, γ-Glu-Tyr, and γ-Glu-Trp were found in lower amounts (<1 μ mol/kg). After 24 h of fermentation with L. fermentum NSB2, peptides γ-Glu-Phe, γ-Glu-Tyr, and γ-Glu-Trp increased by more than 2-fold, while γ-Glu-Gly increased from undetectable levels to 1.5 μ mol/kg. Fermentation with L. argentoratensis 1227B resulted in a modest increase in a single peptide γ-Glu-Ala (1.0 μ mol/kg). Peptide γ-Glu-Gln was produced by B. subtilis Natto (7.1 μ mol/kg) and B. velezensis G17 (41.5 μ mol/kg). Additionally, B. subtilis Natto generated high levels of γ-Glu-Ala (18 μ mol/kg) by the end of fermentation. Conversely, several other peptides, including γ-Glu-Val, γ-Glu-Tyr, γ-Glu-Leu, and γ-Glu-Trp, decreased in Bacillus spp. fermentations. 3.6. Vicine and convicine Before fermentation, FPC contained 11.3 mg/g of vicine and 5.7 mg/ g of convicine (Fig. 5). The levels of vicine were significantly reduced in fermentations with L. argentoratensis 12-27B and B. velezensis G17 to 4.7 mg/g and 2.8 mg/g after 72 h, respectively. Concurrently, convicine levels were decreased by these strains, largest by B. velezensis G17 (1.2 mg/g after 72 h). No significant changes in vicine or convicine were observed in fermentations with L. fermentum NSB2 and B. subtilis Natto. 4. Discussion In this study we investigated the impact of LAB and Bacillus spp. on VOCs, organic acids, FAAs, γ-glutamyl peptides and vicine-convicine during fermentation of FPC. Among the tested LAB, species L. argentoratensis was previously classified as a subspecies of L. plantarum, and only recently elevated to species level (Li et al., 2020b). Given its close phylogenetic relationship and phenotypic similarity to L. plantarum, we consider it appropriate to interpret our findings on L. argentoratensis in the context of existing data on L. plantarum. FPC is a nutrient-rich substrate with a high protein content, along with soluble carbohydrates, fibers, and fats, supporting the growth of all tested bacteria and production of diverse metabolites. Among the VOCs Fig. 4. Concentration of acetic acid (A), lactic acid (B), citric acid (C) and propionic acid (D) (mg/g dry matter) after 8 h, 24 h and 72 h fermentation of faba protein concentrate (FPC, 8 % w/v in water) with L. fermentum NSB2 (Lf NSB2), L. argentoratensis 12-27B (La 12-27B), B. subtilis Natto (Bs Natto), and B. velezensis G17 (Bv G17) and in non-inoculated controls. Dashed lines indicate mean concentration of organic acids in FPC before fermentation. Different lowercase superscript letters within the same row indicate statistically significant differences among samples, as determined by one-way ANOVA (standard least squares method) followed by LSMeans Tukey's HSD post hoc test (P ≤0.05). N. Larsen et al. International Journal of Food Microbiology 446 (2026) 111549 6 in unfermented FPC, aldehydes were predominant, particularly straightchain compounds like hexanal. The hexanal concentration (1.34 μ g/g) was consistent with previous reports on FB protein (Karolkowski et al., 2023) and declined markedly during fermentation. Fermentation induced species-specific shifts in VOC profiles, with the most pronounced changes observed at the final time point. Aldehydes can be enzymatically reduced to alcohols by microbial alcohol dehydrogenases (ADH) or oxidized to carboxylic acids and esters by aldehyde Table 1 Concentration of free amino acids (FAAs, mg/100 g) in faba bean protein concentrate (FPC) before fermentation (0 h), and after 24 and 72 h fermentation with L. fermentum NSB2, L. argentoratensis 12-27B, B. subtilis Natto and B. velezensis G17 1 . FAAs Concentration, mg/100 g Before fermentation, 0 h L. fermentum NSB2 L. argentoratensis 12-27B B. subtilis Natto B. velezensis G17 24 h 72 h 24 h 72 h 24 h 72 h 24 h 72 h Asp 21.2 ±2.1 a 14.2 ± 3.2* abc 11.1 ± 5.3* bc 10.6 ± 0.9** bc 18.0 ±3.7 ab nd nd 9.36 ± 2.07** c nd Glu 66.6 ±2.9 a 52.8 ±5.3* ab 47.7 ± 8.8* ab 35.2 ± 0.8*** bc 21.4 ± 2.2*** cd 22.2 ± 4.4*** cd 64.8 ±14.8 a 8.73 ± 3.42*** d 11.8 ± 2.6*** d Ser 9.34 ±2.95 ab 3.01 ± 0.36* bc 3.30 ± 1.52* bc 1.27 ± 0.90* c 1.62 ± 0.90* bc 6.30 ± 0.46 abc 4.79 ±2.33 bc 13.6 ±6.2 a 4.34 ±2.10 bc His nd nd nd nd nd 2.56 ±1.55 c 63.0 ±7.7 a 2.55 ±1.34 c 35.0 ±5.3 b Gly 10.9 ±1.8 bc 5.99 ± 0.59* bc 7.32 ± 0.61* bc 10.9 ±1.9 bc 24.1 ±7.8* a 3.14 ± 2.04* c 15.7 ±5.48 ab 10.6 ±1.4 bc 3.95 ± 0.53** c Thr nd 0.76 ±0.57 c 0.89 ±0.64 c nd nd nd 45.7 ±12.4 a nd 16.7 ±2.7 b Arg 842 ±65 a 7.01 ± 3.01*** d 1.97 ± 1.61 d 788 ±28 a 741 ±35 ab 571 ±48** b 363 ±85** c 699 ±103 ab 325 ±18*** c Ala 9.80 ±2.20 c 1.32 ± 0.19** c 6.38 ±2.07 c 4.80 ± 1.40* c 13.1 ±0.2 c 7.14 ±1.87 c 198 ±36*** a 37.9 ± 7.8** bc 59.3 ± 7.7*** b Tyr 3.45 ±1.38 b nd nd nd nd 7.79 ±2.13 b 140 ±12*** a 19.4 ±4.6** b 127 ±13*** a Val 3.11 ±1.28 d 16.4 ± 1.5*** cd 16.2 ± 5.1** cd 8.94 ± 1.99* d 4.89 ±2.34 d 4.83 ±3.03 d 183 ±22*** a 44.3 ± 3.5*** c 148 ±16*** b Met nd nd nd nd nd nd 24.5 ±3.5 nd nd Trp 3.10 ±1.38 c 2.82 ±1.38 c 5.52 ±1.33 c 0.93 ±1.06 c nd nd 107 ±36*** a 4.24 ±0.11 c 38.2 ±0.5** b Phe 9.26 ±1.57 c nd nd nd nd 33.9 ±19.6 c 298 ±34*** a 112 ±35** b 265 ±24*** a Ile nd nd nd nd nd nd 168 ±15 a nd 118 ±15 b Leu 9.24 ±1.59 c nd nd nd nd 8.10 ±2.24 c 502 ±70*** a 61.5 ± 10.5*** c 339 ±34*** b Lys 26.1 ±2.2 b 33.7 ±1.6* b 36.3 ±1.0* b 23.1 ±1.3 b 22.6 ±1.4 b 27.7 ±6.2 b 250 ±4 7*** a 43.3 ±15.9 b 63.2 ±7.8** b Essential FAA 2 50.8 ±5.0 d 53.7 ±5.3 d 58.9 ±6.4 d 33.0 ± 1.9** d 27.5 ± 1.1** d 77.1 ±21.1 d 1642 ± 90*** a 267 ±67** c 1024 ± 121*** b Total FAA 1014 ±80 c 138 ±7*** e 137 ± 13*** e 884 ±31 cd 846 ±38* cd 694 ±57** d 2430 ± 179*** a 1066 ±53 c 1554 ± 133** b 1 Mean values (±SD) from triplicate experiments are shown. nd =not detected. Asterisks denote significant differences after fermentation compared to before fermentation determined by the Student's t-test (*P ≤0.05; **P ≤0.01; ***P ≤0.001). Different lowercase superscript letters within the same row indicate statistically significant differences among samples, as determined by one-way ANOVA (standard least squares method) followed by LSMeans Tukey's HSD post hoc test (P ≤0.05). 2 Concentration of total essential FAA including His, Thr, Val, Met, Trp, Phe, Ile, Leu and Lys. Table 2 Concentration of γ-glutamyl peptides ( μ mol/kg) in faba bean protein concentrate (FPC) before fermentation (0 h), and after 24 and 72 h fermentation with L. fermentum NSB2, L. argentoratensis 12-27B, B. subtilis Natto and B. velezensis G17 1 . γ-Glutamyl peptides Concentration, μ mol/kg Before fermentation, 0 h L. fermentum NSB2 L. argentoratensis 12-27B B. subtilis Natto B. velezensis G17 24 h 72 h 24 h 72 h 24 h 72 h 24 h 72 h γ-Glu-Gly nd 1.50 ± 0.60* a 1.24 ± 0.31** a nd nd nd nd nd nd γ-Glu-Gln nd nd nd nd nd nd 7.14 ±3.97* b nd 41.47 ± 8.76*** a γ-Glu-Ala nd nd nd 0.84 ± 0.23** b 1.02 ± 0.28** b 1.02 ± 0.24** b 18.48 ± 3.37*** a 0.34 ± 0.07** b 0.34 ± 0.09** b γ-Glu-Val 0.42 ±0.09 bc 0.86 ± 0.18* a 0.70 ± 0.13* ab 0.24 ± 0.15 c 0.41 ± 0.14 bc 0.23 ± 0.13 c nd nd nd γ-Glu-Tyr 0.37 ±0.06 c 1.27 ± 0.40* a 1.06 ± 0.29* ab 0.33 ± 0.14 c 0.50 ± 0.18 bc 0.16 ± 0.11* c nd 0.036 ± 0.017** c 0.025 ± 0.01*** c γ-Glu-Leu 2.40 ±0.20 ab 3.16 ± 0.44 a 2.78 ± 0.77 ab 1.76 ± 0.44 abc 1.98 ± 0.83 abc 1.64 ± 0.35 bc 0.83 ± 0.13** c 0.69 ± 0.16** c 0.57 ± 0.18** c γ-Glu-Phe 2.12 ±0.34 b 5.63 ± 0.93** a 5.36 ± 0.99** a 1.33 ± 0.56 b 1.27 ± 0.56 b 1.27 ± 0.44 b 1.29 ±0.35 b 0.43 ± 0.04*** b 1.72 ±0.66 b γ-Glu-Trp 0.97 ±0.19 b 1.98 ± 0.56* a 1.90 ± 0.60* a 0.75 ± 0.35 b 0.50 ± 0.35 b nd nd nd nd 1 Mean values (±SD) from triplicate experiments. nd =not detected. Asterisks denote significant differences after fermentation compared to before fermentation determined by the Student's t-test (*P ≤0.05; **P ≤0.01; ***P ≤0.001). Different lowercase superscript letters within the same row indicate significant differences among samples, as determined by one-way ANOVA (standard least squares method) followed by LSMeans Tukey's HSD post hoc test (P ≤0.05). N. Larsen et al. International Journal of Food Microbiology 446 (2026) 111549 7 dehydrogenases (ALDH), depending on oxygen availability (Engels et al., 2022; Fischer et al., 2022; Nugroho et al., 2024). The distinct patterns of aldehyde conversion observed among the tested species suggest variations in ADH and ALDH activity. Reduction of aldehydes, particularly hexanal, is considered beneficial, as it transforms potent offflavor compounds into alcohols and acids with higher odor thresholds (Sed´ o Molina et al., 2024; Tuccillo et al., 2022a). The observed reduction of aldehydes concurrently with accumulation of straight-chain (C3–C8) alcohols and methyl-derivatives of butanol and propanol aligns with previous studies on LAB fermentation of pea and bean-based substrates (Engels et al., 2022; Sed´ o Molina et al., 2024; Tuccillo et al., 2022b; El Youssef et al., 2020). Most of the straight-chain alcohols were found in higher amounts in fermentations with L. fermentum NSB2, compared to L. argentoratensis 12-27B. Likewise, high production of C6 - C8 alcohols by L. fermentum has been reported in faba-bean based milk (Tangyu et al., 2023). Additionally, L. argentoratensis 12-27B appeared less effective in utilization of several aldehydes, particularly benzaldehyde, compared to L. fermentum NSB2. This observation aligns with previous studies reporting higher benzaldehyde concentrations in synthetic medium fermented with L. plantarum than with L. fermentum (Sugahara et al., 2022). In all FPC fermentations, branched chain aldehydes were converted to their corresponding alcohols, mainly, 2/3-methylbutanol and 2-methylpropanol and/or corresponding acids. Notably, 3-methyl-butanoic acid, a major off-flavor compound, giving cheesy sweety aroma in raw faba-beans (Ritter et al., 2024), was produced at particularly high levels in fermentation with LAB and B. subtilis Natto. These results are in accordance with Serrazanetti et al. (2011), who reported a shift towards production of 3-methyl-butanoic acid under acid stress conditions during LAB fermentation of sourdough. In contrast, fermentations with B. velezensis G17 predominantly resulted in the accumulation of branched chain alcohols rather than their related acids. The formation of esters, including methyl and hexyl esters of acetic acid, as well as methyl derivatives of butanoic and propanoic acids, was predominantly observed in FPC fermentations with B. subtilis Natto. These esters are synthesized by esterases and alcohol acyltransferases, indicating their high enzymatic activity in B. subtilis Natto. Esters contribute to the overall aroma complexity, with compounds such as methyl acetate reported to impart sweet and fruity notes during the fermentation of pea protein-based products (El Youssef et al., 2020). Fermentations with L. argentoratensis 12-27B and Bacillus spp. led to a significant increase in various ketones, which can arise from the oxidation of corresponding secondary alcohols or through pyruvate metabolic pathways (Jia et al., 2017; Okoye et al., 2022). In FPC fermented with Bacillus spp., acetoin and diacetyl were the predominant ketones, with B. velezensis G17 being the highest acetoin producer. Our findings are supported by previous reports of elevated acetoin and diacetyl levels in soy-based products fermented with Bacillus spp. (Chen et al., 2022; Keong et al., 2023). These compounds contribute to a creamy, fruity, and buttery aroma commonly associated with dairy products and certain fruits (Tangyu et al., 2023; Utz et al., 2022). Notably, 2-butanone, characterized by its sweet, pungent, and acetonelike odor (Khrisanapant et al., 2019), was particularly abundant in L. argentoratensis 12-27B fermentations. In contrast, fermentations with L. fermentum NSB2 generally resulted in a reduction of acetoin and diacetyl, suggesting a metabolic shift favoring the conversion of pyruvate into acetyl-CoA, with subsequent pathways prioritizing acetate or ethanol production rather than acetolactate (Okoye et al., 2022; Wang et al., 2021). The major ketones in unfermented FPC, acetone and 2butanone, were also significantly reduced by L. fermentum NSB2. Acetone can be metabolized through several pathways, including conversion into acetaldehyde and subsequently into acetic acid via monooxygenase activity, or reduction to isopropanol via secondary alcohol dehydrogenase (Hausinger, 2007). In this study, the latter pathway appears to be predominant, as indicated by the concurrent increase of 2propanol and 2-butanol in L. fermentum NSB2 fermentations. All LAB and Bacillus spp. were capable to metabolize citrate and produce acetic acid, most efficiently in case of L. fermentum NSB2. Fermentations with L. argentoratensis 12-27B resulted in comparatively greater lactic acid levels, reflecting its role as the most efficient acidifier of FPC. Differences between the LAB species in organic acid profiles and VOC production can be attributed to their sugar metabolism. L. fermentum is an obligately heterofermentative,e producing multiple end products from hexose fermentation, while L. argentoratensis is facultatively heterofermentative and can shift between homoand heterolactic pathways (Illeghems et al., 2015; Popova-Krumova et al., 2024; Wang et al., 2021). Unlike LAB, Bacillus spp., particularly B. velezensis G17 produced moderate amounts of propionic acid. The relatively low synthesis of acetic acid and lactate by B. velezensis G17 suggests a metabolic shift towards propionic acid, likely via pyruvate pathways (Gonzalez-Garcia et al., 2017). Although Bacillus spp. are not typical propionic acid producers, the ability of B. subtilis to generate propionic acid has been previously documented in FB and soybean fermentations (El-Moghazy et al., 2011; Keong et al., 2023). In this study, the presence of propionic acid was accompanied by elevated levels of its corresponding methyl esters, indicating that Bacillus-mediated propionic acid production may contribute to the formation of aroma-active esters, thereby influencing the volatile profile of FPC. The increase in FAAs during Bacillus fermentations is consistent with Fig. 5. Concentration of vicine (A) and convicine (B) (mg/g dry matter) after 24 h and 72 h fermentation of faba protein concentrate (FPC, 8 % w/v in water) with L. fermentum NSB2 (Lf NSB2), L. argentoratensis 12-27B (La 12-27B), B. subtilis Natto (Bs Natto), and B. velezensis G17 (Bv G17). Mean values and SD (bars) from triplicate experiments are shown. Asterisks indicate values significantly different from the values before fermentation (the Student's t-test, *P ≤0.05, **P ≤0.01, ***P ≤0.001). Dashed lines indicate mean concentration of vicine and convicine in FPC before fermentation. N. Larsen et al. International Journal of Food Microbiology 446 (2026) 111549 8 previous reports showing that Bacillus spp. produce extracellular proteinases capable of hydrolyzing plant proteins (Chen et al., 2022; Ciurko et al., 2021). The released FAAs included several essential amino acids and probably played a role in the formation of aroma-active VOCs (Ritter et al., 2022; Smit et al., 2009). However, from a nutritional perspective, the concentration of FAA in fermented products remains low compared to the total amino acids bound within faba bean proteins, which become available upon digestion. Their main relevance is therefore functional and sensory, rather than nutritional. Elevated levels of leucine, isoleucine, and valine, precursors of branched-chain alcohols, likely contributed to the increased levels of 2/3-methylbutanol, 2methylpropanol, and related compounds, particularly in B. subtilis Natto fermentations. In contrast, L. plantarum and related LAB primarily rely on membrane-bound proteases and are often auxotrophic for several amino acids (Engels et al., 2022; Harper et al., 2022; Illeghems et al., 2015), which might explain their limited FAAs release in this study. These findings contrast with earlier reports showing increased FAAs during FB fermentation with L. plantarum (Coda et al., 2015; Verni et al., 2019). However, other studies reported only modest increases in specific amino acids in white beans or bean-based beverages fermented with L. plantarum or L. fermentum (Meng et al., 2023; Thompson et al., 2020). Such discrepancies likely reflect strain-specific proteolytic activity and differences in fermentation conditions, including matrix composition and microbial background in non-sterile systems. Arginine was the most abundant FAA, consistent with previous studies on faba beans (Coda et al., 2015; Verni et al., 2019). As a source of carbon, nitrogen, and energy, arginine is metabolized by bacteria through multiple pathways, ultimately yielding L-glutamate (Hern´ andez et al., 2021; Lu, 2006). In this study, arginine was almost depleted by L. fermentum NSB2, partially utilized by Bacillus spp., and remained unaltered by L. argentoratensis 12-27B. Accordingly, genomic analysis revealed genes encoding arginase, a metallohydrolase that converts arginine to L-ornithine and urea, in all tested strains except L. argentoratensis 12-27B. L. fermentum NSB2 also harbored the arc operon, enabling the arginine deiminase (ADI) pathway, which supports nitrogen metabolism and ATP generation via ornithine generation. In Bacillus spp., ornithine was likely converted to glutamate via the ARG pathway, supported by the presence of the roc operon and the observed arginine reduction. Glutamate, can subsequently enter the tricarboxylic acid cycle (TCA) via 2-ketoglutarate, catalyzed by NADP-specific glutamate dehydrogenases (rocG in Bacillus spp. and gdhA in L. argentoratensis 12-27B). Among the γ-glutamyl peptides detected in unfermented FPC, γ-GluLeu, γ-Glu-Val, γ-Glu-Tyr, and γ-Glu-Phe have previously been identified as kokumi-active in white beans and soybeans (Dunkel et al., 2007; Shibata et al., 2017). To our knowledge, this is the first report of these peptides in FPC. Fermentation of FPC led to the production of several γ-glutamyl dipeptides, with species-specific variations, suggesting a potential impact on kokumi perception (Li et al., 2024). Notably, Bacillus spp. markedly increased γ-Glu-Gln, a contributor to the long-lasting and complex flavors in cheese, and γ-Glu-Ala associated with bitterness (Wang et al., 2022). A recent study found that γ-Glu-Gln, but not γ-GluAla, was produced by B. subtilis and B. velezensis in synthetic medium, emphasizing the importance of the substrate in peptide formation (Li et al., 2024). In this study, high production of γ-Glu-Ala by B. subtilis Natto correlated with elevated levels of alanine. Peptide γ-Glu-Phe, which was specifically increased in L. fermentum NSB2 fermentations, is known for kokumi-enhancing and debittering properties in miso and cheese (Hillmann et al., 2016; Valer´ on et al., 2023). Hydrolysis of the pyrimidine glycosides vicine and convicine (v-c) is essential to prevent adverse effects in individuals with G6PD deficiency. These compounds are known to be partially thermostable (Khazaei et al., 2019), and the intense heat applied in FPC production had little effect on v-c levels in this study. Vicine levels remained high (11 mg/g), comparable to those reported in unheated FB protein (16 mg/g; Coda et al., 2015). Fermentation with B. velezensis G17 led to a significant v-c reduction (up to 80 %), whereas L. argentoratensis 12-27B exhibited more modest effect. Accordingly, previous studies reported more than 90 % v-c reduction in FB doughs fermented with L. plantarum strains, linking it to β-glucosidase activity (Kahala et al., 2023; Rizzello et al., 2016; Verni et al., 2019). To our knowledge, this is the first study demonstrating v-c degradation by B. velezensis, suggesting that species beyond L. plantarum may contribute to detoxification. Further studies are needed to assess the β-glucosidase activity of B. velezensis and its role in v-c hydrolysis. 5. Conclusions This study demonstrated that L. fermentum NSB2, L. argentoratensis 12-27B, B. subtilis Natto, and B. velezensis G17 effectively fermented FPC, each exhibiting distinct metabolic effects. Strain-specific differences in VOCs profiles, organic acids, and free amino acids reflected diverse metabolic pathways. Bacillus spp. exhibited strong proteolytic activity, enhancing essential amino acids and γ-glutamyl peptides associated with kokumi flavor, while B. velezensis G17 and L. argentoratensis 12-27B reduced vicine and convicine, improving nutritional quality and safety of FPC. Limitations of the study include testing only one strain per species and the absence of sensory evaluation, leaving species-level effects and perceptual improvements unconfirmed. Nevertheless, the findings highlight the potential of targeted microbial fermentation to enhance the sensory and nutritional properties of FPC, providing a bases for optimized fermentation strategies. Future work should focus on optimizing fermentation conditions, investigating mixed-culture interactions, and validating sensory outcomes in final products. Supplementary data to this article can be found online at https://doi. org/10.1016/j.ijfoodmicro.2025.111549. CRediT authorship contribution statement N. Larsen: Writing – original draft, Visualization, Validation, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. M.G. Henriksen: Writing – review & editing, Software, Methodology, Formal analysis. C. Crocoll: Writing – review & editing, Validation, Methodology, Formal analysis, Data curation. Q. Li: Writing - review and editing, Validation, Methodology, Formal analysis, Data curation. R. Lametsch: Writing – review & editing, Validation, Methodology, Formal analysis, Data curation. M.M. Poojary: Writing – review & editing, Visualization, Validation, Methodology, Formal analysis, Data curation. L. Krych: Writing – review & editing, Validation, Software, Methodology, Formal analysis, Data curation, Conceptualization. M.A. Petersen: Writing – review & editing, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation. L. Jespersen: Writing – review & editing, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization. Declaration of Generative AI and AI-assisted technologies in the writing process During the preparation of this work the authors used ChartGPT to improve English grammar and readability. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication. Declaration of competing interest We declare that we do not have any commercial or associative interest that represents a conflict of interest in connection with the work submitted. N. Larsen et al. International Journal of Food Microbiology 446 (2026) 111549 9