Riboswitch-targeted improvement of lysine production in Bacillus and Priestia species Authors: Tessa S. Canoy, Pablo Atienza Lopez, Emma S. Wiedenbein, Helga Sato, Frej Larsen, Eoghan T. Reilly, Stefan E. Seemann, Jan Gorodkin, Lene Jespersen, Dennis S. Nielsen Published: 2026 ABSTRACT Lysine is an essential amino acid often limited in plant-based foods, making its enrichment in fermented foods a key nutritional goal. Bacillus and Priestia species are natural lysine producers, but their biosynthetic capacity is constrained by riboswitch regulation of the aspartokinase gene lysC. In this study, we employed mutagenesis using the lysine analogue aminoethylcysteine (AEC) to generate lysine-overproducing strains. AEC-resistant mutants carried distinct mutations in the lysC riboswitch and in lysine biosynthetic and transporter genes. Several mutants showed significantly increased lysine production in defined media. One riboswitch mutant produced over 150-fold more free lysine than its wild type in defined media and threefold during oat fermentation, establishing proof of concept for lysine enrichment in food matrices. Structural predictions revealed that lysC riboswitch mutations occurred near the lysine binding pocket, suggesting altered ligand-binding dynamics. These findings highlight riboswitch-targeted mutagenesis as a promising non-recombinant strategy to enhance lysine production in food fermentations. Introduction Plant-based diets are increasingly promoted for their health benefits and environmental sustainability, yet their protein quality is often limited by an imbalance of essential amino acids1. Among these, lysine is a critical amino acid for human health, supporting protein synthesis, calcium absorption, and immune function2. However, cereals such as wheat, rice, and oat are relatively low in lysine, making it a key nutritional bottleneck in plant-based foods3. Improving lysine content in plant-based matrices is therefore central to enhancing their nutritional quality and supporting sustainable food systems4. Fermentation provides a promising route to enhance lysine in plant-based substrates5. Species of Bacillus and Priestia contribute to nutritional improvement through de novo amino acid biosynthesis as well as extensive protease activity, which creates alkaline conditions and promotes release of free amino acids6. These genera are key players in traditional food fermentations, such as dawadawa in West Africa and natto in Asia, where they enhance flavour, texture, protein digestibility, and micronutrient bioavailability7. Importantly, several studies have reported increased lysine content following fermentation with Bacillus and Priestia species, including in soy milk8, chickpea milk9, and African locust bean10. In species of Bacillus and Priestia, lysine biosynthesis proceeds via the diaminopimelate (DAP) pathway, which converts aspartate into L-lysine through four known route variants: namely the acetylase, succinylase, dehydrogenase, and aminotransferase routes11. Most Bacillus and Priestia species utilize the acetylase variant, characterized by N-acetyl intermediates12. The pathway is tightly regulated at the level of lysC, encoding aspartokinase, the enzyme catalyzing the first committed step of lysine biosynthesis13. Expression of lysC is controlled by a lysine-responsive riboswitch in the 5′ untranslated region14, which represses transcription in response to lysine binding, thereby limiting production via negative feedback15,16. S-2-Aminoethylcysteine (AEC), also known as thialysine, is a synthetic analogue of lysine in which a sulphur atom replaces the second carbon in the side chain. AEC binds to the lysine riboswitch, repressing lysC transcription17. Some bacterial mutants resistant to AEC harbour mutations in the riboswitch region that reduce its responsiveness to lysine or AEC18, resulting in constitutive expression of lysC19, which may lead to increased lysine production. Improved lysine production of AEC-resistant mutants has been identified in Bacillus subtilis, Priestia megaterium (formerly classified under Bacillus), Bacillus licheniformis, and Bacillus pumilus20–23. While previous studies have primarily linked AEC resistance to mutations in the lysC riboswitch, we considered that other genetic mechanisms related to lysine metabolism might also contribute. We therefore aimed to characterize riboswitch sequences and structures to identify mutations potentially linked to regulatory malfunction, alongside broader genomic changes in AEC-resistant mutants of Bacillus and Priestia species. We hypothesized that AEC-resistant strains would accumulate higher lysine levels due to reduced regulatory repression or altered lysine metabolism. This approach allowed us to explore the genetic basis of lysine overproduction and identify promising candidates for nutritional enhancement of plant-based foods. Results Comparative lysine and lysine riboswitch genomics Initially, 78 Bacillus and Priestia strains24 were screened for presence of genes involved in lysine biosynthesis and transport. To facilitate this, all 78 strains were re-sequenced by long-read sequencing (Supplementary Data 1). All strains harboured genes for the acetylase route of the DAP pathway for lysine biosynthesis. Two P. megaterium strains (DSM319, DSM2894) and one B. licheniformis (DSM1969) also possessed the dapL gene for the aminotransferase route. Six strains possessed the dapE gene, the final gene in the succinylase pathway, whereas other succinylase genes dapC and dapD were not detected in any of the strains. Two lysine transporters were identified: lysine permease lysP (11 out of 12 Priestia strains) and L-lysine exporter family protein lysE/argO (64 strains). We identified eight genes under lysine riboswitch control in at least one Bacillus or Priesta strain with functions in lysine biosynthesis or in (putative) lysine transporters (Fig. 1). Many strains had multiple gene copies located in distinct operons, with only a subset of the copies regulated by lysine riboswitches (Supplementary Fig. 1). A common factor among the 78 strains was the riboswitch regulation of lysC. Most (n = 65) contained multiple lysC copies in two different operons, of which one operon was under riboswitch control, while thirteen strains had a single riboswitch-controlled lysC copy. Riboswitch-controlled lysC genes were markedly different from the lysC copies not controlled by riboswitches, falling into two distinct clades (Fig. 2).Fig. 1Genes for lysine biosynthesis and membrane transport proteins are under riboswitch control in a species-dependent manner.Maximum likelihood phylogenetic tree based on 92 Up-to-date Bacterial Core Gene (UBCG) method50 of 78 Bacillus and Priestia strains representing the best taxonomic matches to type strains (Supplementary Data 1). Columns show presence-absence of genes that are under lysine riboswitch regulation in ≥1 Bacillus or Priestia strains and the selection of wild-type strains for AEC mutagenesis and mutations resulting from AEC exposure. Genes involved in lysine biosynthesis and transport, including EC numbers, are shown in the right-bottom corner.Fig. 2Riboswitch-controlled (R-) lysC copies and non-riboswitch controlled lysC copies of 78 Bacillus and Priestia strains fall into separate phylogenetic clades.Maximum likelihood phylogenetic tree of lysC protein sequences, aligned using MUSCLE and constructed with IQ-TREE53 (1,000 bootstrap replicates, normalized to 100 for visualisation). The tree was visualized in iTOL55 and rooted at the midpoint. LysC sequences located adjacently within the same operon were concatenated prior to analysis. Genes for lysine biosynthesis and membrane transport proteins are under riboswitch control in a species-dependent manner. Maximum likelihood phylogenetic tree based on 92 Up-to-date Bacterial Core Gene (UBCG) method50 of 78 Bacillus and Priestia strains representing the best taxonomic matches to type strains (Supplementary Data 1). Columns show presence-absence of genes that are under lysine riboswitch regulation in ≥1 Bacillus or Priestia strains and the selection of wild-type strains for AEC mutagenesis and mutations resulting from AEC exposure. Genes involved in lysine biosynthesis and transport, including EC numbers, are shown in the right-bottom corner. Riboswitch-controlled (R-) lysC copies and non-riboswitch controlled lysC copies of 78 Bacillus and Priestia strains fall into separate phylogenetic clades. Maximum likelihood phylogenetic tree of lysC protein sequences, aligned using MUSCLE and constructed with IQ-TREE53 (1,000 bootstrap replicates, normalized to 100 for visualisation). The tree was visualized in iTOL55 and rooted at the midpoint. LysC sequences located adjacently within the same operon were concatenated prior to analysis. In the Priestia genus, other lysine biosynthetic genes were riboswitch regulated as well, including lysA and dapA in P. megaterium and Priestia aryabhattai. In addition to the riboswitch control of lysine biosynthesis genes, most strains (n = 74) had one lysine transporter gene under riboswitch control. For 11/12 Priestia strains, this was lysine permease lysP. In Bacillus, three transporter genes were regulated by lysine riboswitches, arginine/ornithine antiporter arcD (COG0531), putative arginine/ornithine antiporter ydgI (COG0531), and putrescine/ornithine transporter potE—with typically only one under riboswitch control per strain. AEC-induced mutagenesis Eleven wild-type strains were selected for AEC mutagenesis based on a diversity of genomic characteristics (Fig. 1), only including species with Qualified Presumption of Safety (QPS) status to support their use in food applications. To maximize mutant diversity, we employed two chemically defined media to generate mutants, one with and one without essential amino acids, based on prior evidence linking AEC resistance to auxotrophies for amino acids like leucine, threonine and methionine5. A total of 21 AEC-resistant mutants were isolated from AEC medium without essential amino acids (n = 4) and from AEC medium with essential amino acids (n = 17) (Table 1) and were re-sequenced together with wild-types (n = 14) derived from the original 11 strains, with some strains represented by two biological replicates (Supplementary Data 1). All mutants were able to grow on medium without essential amino acids, except mutants 109E11A and 109E11B and corresponding wild-type 109-WTA (Supplementary Table 1). Compared to the original wild-types, the 14 re-sequenced wild-type replicates that were re-cultivated for mutant generation acquired additional mutations, ranging from 8 to 1,793 (median: 97) (Supplementary Table 2).Table 1Strain information on AEC-induced mutants and their corresponding wildtypesWild-type strains (reference name)AEC-resistant mutantsIsolated mediumWild-type replicateP. megaterium 20-WT (DSM3228)20-F9AEC medium without EAA20-WTA20-G9AEC medium20-WTBP. megaterium 25-WT (DSM17641)25-F8AEC medium without EAA25-WTA25-H8AEC medium25-WTAB. pumilus 27-WT (DSM13835)27-H10AEC medium27-WTBB. velezensis 77-WT (PRO77)77-E7AEC medium77-WTAB. velezensis 96-WT (PRO96)96-E6AEC medium96-WTAB. paralicheniformis 109-WT (PRO109)109-E11AAEC medium109-WTA109-E11BAEC medium109-WTAB. licheniformis 15-WT (DSM13T)15-E12AAEC medium15-WTA15-E12BAEC medium15-WTAB. subtilis 19-WT (DSM21393)19-E3AEC medium19-WTA19-F3AEC medium without EAA19-WTAB. subtilis 2-WT (DSM10T)2-E1AEC medium2-WTA2-H1AEC medium without EAA2-WTB2-F2AEC medium2-WTAB. subtilis 122-WT (PRO122)122-E5AEC medium122-WTAB. subtilis 64-WT (PRO64)64-AM3AEC medium64-WTA64-E2AEC medium64-WTB64-H5AEC medium64-WTA64-H6AEC medium64-WTAAll wild-type strains were from the DSMZ collection (reference name starting with DSM) or from West-African origin (reference name starting with PRO), as described by Wiedenbein et al. (2023)24. EAA: essential amino acids, AEC: S-(2-aminoethyl)-L-cysteine. Wild-type and mutant strains selected for liquid fermentations in microtitre plate are indicated in bold. Table 1Strain information on AEC-induced mutants and their corresponding wildtypesWild-type strains (reference name)AEC-resistant mutantsIsolated mediumWild-type replicateP. megaterium 20-WT (DSM3228)20-F9AEC medium without EAA20-WTA20-G9AEC medium20-WTBP. megaterium 25-WT (DSM17641)25-F8AEC medium without EAA25-WTA25-H8AEC medium25-WTAB. pumilus 27-WT (DSM13835)27-H10AEC medium27-WTBB. velezensis 77-WT (PRO77)77-E7AEC medium77-WTAB. velezensis 96-WT (PRO96)96-E6AEC medium96-WTAB. paralicheniformis 109-WT (PRO109)109-E11AAEC medium109-WTA109-E11BAEC medium109-WTAB. licheniformis 15-WT (DSM13T)15-E12AAEC medium15-WTA15-E12BAEC medium15-WTAB. subtilis 19-WT (DSM21393)19-E3AEC medium19-WTA19-F3AEC medium without EAA19-WTAB. subtilis 2-WT (DSM10T)2-E1AEC medium2-WTA2-H1AEC medium without EAA2-WTB2-F2AEC medium2-WTAB. subtilis 122-WT (PRO122)122-E5AEC medium122-WTAB. subtilis 64-WT (PRO64)64-AM3AEC medium64-WTA64-E2AEC medium64-WTB64-H5AEC medium64-WTA64-H6AEC medium64-WTAAll wild-type strains were from the DSMZ collection (reference name starting with DSM) or from West-African origin (reference name starting with PRO), as described by Wiedenbein et al. (2023)24. EAA: essential amino acids, AEC: S-(2-aminoethyl)-L-cysteine. Wild-type and mutant strains selected for liquid fermentations in microtitre plate are indicated in bold. Strain information on AEC-induced mutants and their corresponding wildtypes All wild-type strains were from the DSMZ collection (reference name starting with DSM) or from West-African origin (reference name starting with PRO), as described by Wiedenbein et al. (2023)24. EAA: essential amino acids, AEC: S-(2-aminoethyl)-L-cysteine. Wild-type and mutant strains selected for liquid fermentations in microtitre plate are indicated in bold. Variant calling was performed by comparing mutants with the wildtype replicate they were isolated from (Table 1). All mutations were functionally annotated to COG categories based on protein sequences, and the non-coding RNA mutations were manually added as a custom category. Beyond mutations with unknown functions, genes involved in amino acid transport and metabolism (COG category E) had the highest mutation rate, spanning 138 mutations in 16 mutant strains (Fig. 3a).Fig. 3AEC-induced mutagenesis of Bacillus and Priestia strains leads to increased lysine biosynthesis.a Mutations identified in 21 AEC-resistant mutants, based on variant calling against the specific wildtype replicate from which each mutant was derived. For mutants 64-E2, 64-H5, 64-H6, and 64-AM3, comparisons were made against a joint assembly of biological replicates 64-WTA and 64-WTB (see Supplementary Data 1). Mutations were functionally annotated with EggNOG according to COG categories56, with non-coding RNA as manually added classification. Proteins not functionally annotated by EggNOG were classified together with COG category S (function unknown). b Mutations (SNPs, deletions, insertions) of seven selected AEC-resistant mutants, NA refers to hypothetical proteins. c Lysine production by seven selected AEC-resistant mutants and their wildtypes on three chemically defined media after 22 h of cultivation, error-bars represent the standard deviation of three biological replicates. Significant differences in lysine levels between mutants and wildtypes are displayed as * (p < 0.05), ** (p < 0.01) and ** (p < 0.001), based on Holm-adjusted P values from estimated marginal means comparisons. Non-significant comparisons are shown with exact adjusted p-values. AEC-induced mutagenesis of Bacillus and Priestia strains leads to increased lysine biosynthesis. a Mutations identified in 21 AEC-resistant mutants, based on variant calling against the specific wildtype replicate from which each mutant was derived. For mutants 64-E2, 64-H5, 64-H6, and 64-AM3, comparisons were made against a joint assembly of biological replicates 64-WTA and 64-WTB (see Supplementary Data 1). Mutations were functionally annotated with EggNOG according to COG categories56, with non-coding RNA as manually added classification. Proteins not functionally annotated by EggNOG were classified together with COG category S (function unknown). b Mutations (SNPs, deletions, insertions) of seven selected AEC-resistant mutants, NA refers to hypothetical proteins. c Lysine production by seven selected AEC-resistant mutants and their wildtypes on three chemically defined media after 22 h of cultivation, error-bars represent the standard deviation of three biological replicates. Significant differences in lysine levels between mutants and wildtypes are displayed as * (p < 0.05), ** (p < 0.01) and ** (p < 0.001), based on Holm-adjusted P values from estimated marginal means comparisons. Non-significant comparisons are shown with exact adjusted p-values. Many AEC-resistant strains had mutations in transporter genes, including arcD (n = 8) and ydgI (n = 4) which were exclusively observed in Bacillus strains, and lysP (n = 2), which was only mutated in P. megaterium (Supplementary Data 2). All these mutants carried a single mutation event in one of these genes, with no other detectable mutations in the genome for mutants 109-E11A (ydgI) and 122-E5 (arcD). Both arcD and ydgI were functionally annotated as putative lysine transporter yvsH. For two mutants (19-E3, 2-F2), mutations in the arcD locus led to the separation of the ancestral arcD gene into two distinct genes: arcD and a newly identified putrescine:ornithine antiporter potE gene (Supplementary Fig. 2). Among the four mutants isolated from AEC medium lacking essential amino acids, two carried mutations in transporter genes, lysP (25-F8) and arcD (19-F3), whereas the remaining two (2-H1 and 20-F9) contained unique mutations in hypothetical proteins, indicating distinct mutation types within this group. Two mutants contained mutations in genes involved in the DAP pathway, dapF (n = 2), and dapA (n = 1). Mutations in the lysC riboswitch element were identified for three strains (77-E7, 20-G9, 27-H10) (Fig. 3b). Other common mutations were in the aspartate-tRNA ligase aspS (n = 4), altronatedehydratase uxaA (n = 3), and HIP-CoA ligase fadD3 genes (n = 3). Seven AEC-resistant mutants and their wild-types were selected for fermentation on defined media in microtitre plate. These included two lysC riboswitch mutants, four transporter mutants, and one mutant carrying both a lysC riboswitch and a transporter gene mutation. All seven mutants significantly increased lysine levels compared to their wildtypes in lysine-supplemented defined media, while three mutants (27-H10, 20-G9 and 19-F3) also increased lysine levels in lysine-free media (Fig. 3c). As expected, all wildtypes produced less lysine than the mutants when grown on AEC medium, except for wild-type 25-WT, which uniquely acquired AEC resistance during cultivation on AEC medium (based on growth observations). Mutant 20-G9 showed the highest lysine production levels in microtitre plate (Fig. 3c), increasing lysine levels 156× (medium without lysine) and 4.6× (medium with lysine) compared to its wildtype. Pre-cultivation effects on growth, lysine production and genetic stability During growth on AEC medium, some mutants failed to grow despite previously acquired AEC resistance. We hypothesized that AEC resistance was lost during pre-cultivation on BHI medium. We therefore pre-cultivated three lysC riboswitch mutants (77-E7, 27-H10, 20-G9) on either BHI or AEC medium for 16 h, followed by a second pre-cultivation step on BHI (4 h) to ensure sufficient biomass, then transferred them to defined media for analysis of growth (Fig. 4a) and lysine production (Fig. 4b). Following pre-cultivation on BHI, none of the 27-H10 replicates (n = 3) grew on AEC medium, and only one of three replicates of 77-E7 showed growth (Fig. 4a). In contrast, all replicates of 27-H10 and 77-E7 grew on AEC medium after pre-cultivation on AEC medium. All 20-G9 replicates grew on AEC medium regardless of whether they were pre-cultivated on BHI or AEC medium.Fig. 4Pre-cultivation medium (BHI or AEC-containing medium) affects the ability of some Bacillus and Priestia mutants to subsequently grow on AEC-containing medium and to over-produce lysine.a Growth of mutants on chemically defined medium with or without AEC, following pre-cultivation in either BHI or defined AEC medium, error-bars represent the standard deviation of three biological replicates. b Lysine production by mutants after 22 h of cultivation on different chemically defined media, following pre-cultivation in either BHI or defined AEC medium. Horizontal lines (red) indicate average lysine levels of the corresponding wildtypes recultivated on BHI. Strain 27-10 shows medium-dependent lysine differences between pre-cultivation conditions (p-values from estimated marginal means). Strain 20-G9 shows no effect of pre-cultivation (p = 1), while strain 77-E7 shows a significant main effect of pre-cultivation regardless of medium (p = 0.022; no interaction; two-way ANOVA). Pre-cultivation medium (BHI or AEC-containing medium) affects the ability of some Bacillus and Priestia mutants to subsequently grow on AEC-containing medium and to over-produce lysine. a Growth of mutants on chemically defined medium with or without AEC, following pre-cultivation in either BHI or defined AEC medium, error-bars represent the standard deviation of three biological replicates. b Lysine production by mutants after 22 h of cultivation on different chemically defined media, following pre-cultivation in either BHI or defined AEC medium. Horizontal lines (red) indicate average lysine levels of the corresponding wildtypes recultivated on BHI. Strain 27-10 shows medium-dependent lysine differences between pre-cultivation conditions (p-values from estimated marginal means). Strain 20-G9 shows no effect of pre-cultivation (p = 1), while strain 77-E7 shows a significant main effect of pre-cultivation regardless of medium (p = 0.022; no interaction; two-way ANOVA). In mutant 27-H10, the loss of AEC resistance had no effect on lysine production when grown in medium containing lysine (p = 1, estimated marginal means), but it led to reduced lysine production in medium lacking lysine (p < 1 × 10-5) and AEC medium (p < 1 × 10-5). Mutant 77-E7 pre-cultivated on AEC medium generally produced more lysine than when pre-cultivated on BHI medium (p = 0.022). For mutant 20-G9, pre-cultivation did not affect lysine levels (p = 1, two-way ANOVA). Growth (Fig. 4a) and lysine production (Fig. 4b) of 20-G9 after pre-cultivation on AEC medium showed high variability, as indicated by large standard deviations. Interestingly, mutants 77-E7 and 27-H10 which showed replicate-dependent loss of AEC resistance during pre-cultivation on BHI (Fig. 4a) still produced more lysine than the wild type strains in lysine-supplemented medium, and in the case of 27-H10 also in medium without lysine (Fig. 3c). Whole-genome sequencing was performed on the 77-E7, 27-H10, and 20-G9 mutants after 16 h cultivation in either BHI or AEC medium. The resulting genomes were compared to the previously obtained genomes of these mutants to assess the impact of additional cultivation on genetic stability. Mutations in the lysC riboswitch were preserved in mutants 27-H10 and 20-G9 regardless of pre-cultivation medium. In contrast, the lysC riboswitch element of mutant 77-E7 resembled that of the wild-type (77-WT) after both BHI and AEC pre-cultivation. The ydgI gene mutation in 77-E7 remained stable across all conditions. Across all mutants, pre-cultivation led to a few additional mutations, mostly in hypothetical proteins (Supplementary Data 3). Lysine production by mutant 20-G9 Given the strong lysine overproduction by mutant 20-G9 in defined media, its performance was further assessed in liquid shaking culture fermentation and solid-state fermentation of oat. In liquid fermentation, both 20-WT and 20-G9 reached higher cell counts on medium without lysine than medium with lysine after 22 hours (20-WT: 7.2 ± 0.0 vs. 6.3 ± 0.1 log CFU/mL, p = 1 × 10⁻⁵, estimated marginal means; 20-G9: 6.7 ± 0.2 vs. 5.4 ± 0.1 log CFU/mL, p = 0.0023; Fig. 5a). On AEC-medium, 20-WT showed poor growth at 8 h, with cell counts of 3.5 ± 0.4 log CFU/mL. At 22 h, in an independent sample, 20-WT reached a similar cell count to 20-G9 (20-WT: 8.8 ± 0.2, 20-G9: 8.7 ± 0.6 log CFU/mL; p = 1), suggesting that 20-WT acquired AEC resistance.Fig. 5Riboswitch-mutant 20-G9 shows significantly higher free lysine production than its wild-type 20-WT after fermentation on chemically defined media (n = 3) and solid-state fermentation of oat (n = 3).a Growth during liquid fermentation on chemically defined media, b Intra- and extracellular free lysine production after 8 h and 22 h cultivation in shaking culture. Error bars represent the standard deviation of three biological replicates, except for one analysis marked with a diamond symbol where one biological replicate was excluded (see Methods); data shown for the remaining two replicates. c Free lysine production during solid-state fermentation of oat, d Total lysine production (free and protein-bound) during solid-state fermentation of oat. Significant comparisons are based on linear mixed models and estimated marginal means, exact P values are reported in the main text. Riboswitch-mutant 20-G9 shows significantly higher free lysine production than its wild-type 20-WT after fermentation on chemically defined media (n = 3) and solid-state fermentation of oat (n = 3). a Growth during liquid fermentation on chemically defined media, b Intra- and extracellular free lysine production after 8 h and 22 h cultivation in shaking culture. Error bars represent the standard deviation of three biological replicates, except for one analysis marked with a diamond symbol where one biological replicate was excluded (see Methods); data shown for the remaining two replicates. c Free lysine production during solid-state fermentation of oat, d Total lysine production (free and protein-bound) during solid-state fermentation of oat. Significant comparisons are based on linear mixed models and estimated marginal means, exact P values are reported in the main text. Extra- and intra-cellular lysine levels differed significantly between 20-G9 and 20-WT across media and time points. Mutant 20-G9 consistently yielded higher extracellular lysine levels than 20-WT, with the strongest difference observed in the medium without lysine at 22 hours (p < 1×10⁻⁵; Fig. 5b). At 8 h, mutant 20-G9 showed elevated intracellular lysine levels in AEC and lysine-supplemented media but lower intracellular levels in the medium without lysine. At 22 h, intracellular lysine levels were higher in 20-G9 than 20-WT across all media (AEC: 1.04, p = 9 × 10⁻⁵; lysine: 0.93, p = 3 × 10-4; no lysine: 0.64, p = 0.015). For both 20-WT and 20-G9, intracellular lysine production increased significantly from 8 h to 22 h in all media. Similar trends were noticed when considering cell dry weight, with 20-G9 producing 24-fold higher yields than 20-WT (p = 0.012, Supplementary Fig. 3). Lysine yields at 22 h were comparable between the no-lysine medium (30.5 ± 2.5 μM/g dry cell weight) and AEC-medium (31.5 ± 1.4 μM/g dry cell weight; p = 0.90). Whole oats were selected as a candidate substrate for in situ lysine fortification via fermentation. Steaming at 100 °C for 5 min prior to fermentation reduced viable counts below the detection threshold (2.3 log CFU/g). During fermentation, pH dropped significantly from 6.37 ± 0.02 to 5.10 ± 0.04 within 46 h (p < 1 × 10-5), with no strain-dependent differences (p = 0.95, linear mixed model) (Supplementary Table 3). Both strains, 20-WT and 20-G9, showed similar growth (p = 0.67, linear mixed model), increasing from 4.8 ± 0.1 to 8.7 ± 0.1 log CFU/g (20-G9) and 8.7 ± 0.6 log CFU/g (20-WT) by 46 h (Supplementary Table 3). Fermentation with 20-G9 significantly increased free lysine levels at 8 h (0.66 ± 0.08 µmol/g, p < 1 × 10-5) and 22 h (0.63 ± 0.02 µmol/g, p < 1 × 10-5, estimated marginal means) compared to initial lysine levels in oat (0.24 ± 0.01 µmol/g) (Fig. 5c). Although levels declined by 46 h (0.31 ± 0.01 µmol/g), they remained significantly higher than those after 46 h of 20-WT fermentation (p = 3 × 10⁻⁵). In contrast to 20-G9, 20-WT reduced free lysine at 8 h (0.14 ± 0.02 µmol/g, p = 1×10⁻4), with levels at 22 h (0.22 ± 0.01 µmol/g, p = 0.92) and 46 h (0.19 ± 0.02 µmol/g, p = 0.094) comparable to non-fermented oat. Total lysine, representing the sum of free and protein-bound lysine, was significantly higher than free lysine at all time points being around 40-45 µmol/g. There were no significant differences between strains or time points for total lysine, reflecting that while free lysine can be readily increased, it only represents a fraction of the total lysine (Fig. 5d). Riboswitch secondary structure comparison As lysine production is under riboswitch control in Bacillus and Priestia we subsequently aimed to model differences in riboswitch structures among wild-types and AEC-resistant mutants to better understand how the mutations affect riboswitch structure. A total of 184 lysine riboswitches were identified in 78 wild-types and 3 mutants, consisting of 51 unique sequences. Three riboswitch sequences of P. flexa PRO116 were excluded due to their substantial divergence from the other sequences – averaging only 40% sequence identity with the others – which skewed the consensus secondary structure (Supplementary Fig. 4). After exclusion, 48 unique sequences remained. To evaluate structural differences among these riboswitches, we predicted the secondary and tertiary structures using the workflow illustrated in Fig. 6. A structure-aware multiple sequence alignment was performed with foldalignM25, followed by consensus secondary structure prediction with PETfold26. To visualize consensus secondary structures, unique sequences (n = 48) were divided into riboswitches controlling transporter genes (n = 18) or biosynthesis genes (n = 30).Fig. 6Bioinformatic workflow for the prediction of riboswitch secondary and tertiary structures.Names of tools in bold, with primary sequences in white, secondary structure predictions in blue, and tertiary structures in red. Abbreviations: MSA multiple sequence alignment, WT wildtypes, RMSD root-mean square distribution, INF Interaction Network Fidelity. Bioinformatic workflow for the prediction of riboswitch secondary and tertiary structures. Names of tools in bold, with primary sequences in white, secondary structure predictions in blue, and tertiary structures in red. Abbreviations: MSA multiple sequence alignment, WT wildtypes, RMSD root-mean square distribution, INF Interaction Network Fidelity. Structural differences were observed between riboswitches controlling transport (arcD, ydgI, potE, lysP) and controlling lysine biosynthesis (lysC, dapA, lysA) (Fig. 7). Transport-regulating riboswitches (Fig. 7a) were more conserved in sequence, exhibited a shorter P1 stem and included an additional pyrimidine nucleotide within the junction between helices P1 and P2, resulting in an expanded central junction. Greater sequence conservation was observed in lysine biosynthesis-regulating riboswitches, especially in the P3 and P4 helices (Fig. 7). Comparison of base-pair distances between RNAfold-predicted structures showed that transporter- and lysine biosynthesis-controlling riboswitches tended to group separately (Supplementary Fig. 5).Fig. 7Consensus secondary structure models of Bacillus and Priestia lysine riboswitches based on aligned (foldalignM) sequences with consensus secondary structure predicted by Petfold, visualized by R2R.Kissing-loop interactions (L2-L3) were not possible to predict using this method, despite evidence for their role in B. subtilis riboswitches76. a lysine riboswitches controlling transporter genes arcD, lysP, potE and ydgI (n = 18), b lysine riboswitches controlling biosynthesis genes lysC, dapA and lysA (n = 30). Nucleotides predicted to be involved in lysine binding27 are boxed, and the mutations of AEC-resistant lysC riboswitches from this study are circled. Consensus secondary structure models of Bacillus and Priestia lysine riboswitches based on aligned (foldalignM) sequences with consensus secondary structure predicted by Petfold, visualized by R2R. Kissing-loop interactions (L2-L3) were not possible to predict using this method, despite evidence for their role in B. subtilis riboswitches76. a lysine riboswitches controlling transporter genes arcD, lysP, potE and ydgI (n = 18), b lysine riboswitches controlling biosynthesis genes lysC, dapA and lysA (n = 30). Nucleotides predicted to be involved in lysine binding27 are boxed, and the mutations of AEC-resistant lysC riboswitches from this study are circled. AEC-induced mutagenesis only affected lysC riboswitches, none of the riboswitches controlling transporter genes or lysA/dapA were altered (Fig. 7). Two mutants (20-G9 and 77-E7) had single nucleotide polymorphisms (SNPs), and one mutant (27-H10) had a deletion of 10 bases in the lysC riboswitch (Fig. 7b). In mutant 27-H10, three nucleotides located within the J2–J3 junction were deleted, which were previously identified as important for lysine binding in Thermotoga maritima27. In mutant 20-G9, the G11A substitution was positioned near known lysine-binding nucleotides27. SNP effects (20-G9, 77-E7) on local secondary structure were further assessed with RNAsnp. For the 77-E7 and 77-WT pair, both predicted structures deviated substantially from the RFAM family (RF00168) consensus secondary structure (Supplementary Fig. 6). RNAsnp analysis of wild-type 20-WT and 20-G9 mutant yielded structures more resemblant to the RFAM RF00168 structure (Supplementary Fig. 6). In the predicted wild-type structure, G11 forms part of the lower stem of helix P1, while in the predicted mutant structure, the G11A substitution disrupted base-pairing at the base of P1, resulting in a partial unwinding of the P1 helix. Riboswitch tertiary structure comparison For tertiary structure prediction, cluster representatives (n = 24) were selected based on 95% ClustalΩ pairwise sequence identity (Fig. 6). After multiple sequence alignment with foldalignM, evolutionary constraints derived from PETfold were used to guide individual foldings. Tertiary structure prediction tools were compared by benchmarking with the T. maritima 3diz PDB structure (Supplementary Fig. 7). Structures of the best performing tool were selected for molecular docking (rDOCK) to predict riboswitch-lysine binding energies (Supplementary Fig. 8). Tertiary structures were modelled using six tools and converted to secondary structures for comparison. In Fig. 8, as an example we compare the 20-WT lysC riboswitch and the 3diz PDB structure. Prediction accuracy was evaluated by aligning each modelled structure to the 3diz PDB structure and computing RMSD and INF values. All 24 modeRNA-predicted Bacillus/Priestia structures aligned closely to 3diz (average RMSD: 1.8 ± 0.5 Å), but when parsed back to secondary structure (e.g. Figure 8e), the structures diverged markedly from 3diz (Fig. 8g). Deep learning tools resulted in the best INF and RMSD scores, including trRosettaRNA (INF 0.91, RMSD 1.3 Å) and AlphaFold3 (INF 0.88, 1.4 Å), while template-based tools resulted in lower INF ( ≤ 0.73) and higher RMSD values (≥13.2 Å) (Supplementary Fig. 7). Adding secondary structure constraints to the tertiary structure predictions affected performance differently across tools: for trRosettaRNA, similarity to the PDB structure decreased (RMSD increased from 1.3 Å to 3.7 Å; INF decreased from 0.91 to 0.78), for simRNA and RNAcomposer, both INF and RMSD scores increased; and for Vfold, both scores decreased (see Supplementary Fig. 7). Only deep learning tools predicted the P2–P3 kissing-loop pseudoknot: AlphaFold3 in all sequences, trRosettaRNA in 19/24 sequences. AlphaFold3 structures had acceptable global confidence levels (ptm > 0.5) in 14 riboswitches, 10 of which had secondary structures clustering closely with 3diz based on base-pair distance (Supplementary Fig. 8). Riboswitches with dissimilar secondary structures to 3diz had low ptm scores and were lysC riboswitches (incl. 20-WT lysCriboswitch, Fig. 8 g). AlphaFold3 uniquely predicted pseudoknots across all 24 Bacillus/Priestia riboswitches while also achieving low RMSD and high INF values against 3diz. AlphaFold3 structures were aligned and clustered by RMSD, revealing two groups: one with high ptm (> 0.5), the other with low ptm ( < 0.5) (Supplementary Fig. 5 and 8). Lysine-riboswitch binding energy prediction by rDOCK (INTER scores) did not result in any clear trends between lysC riboswitch mutants and their wildtypes. Three riboswitches (lysC riboswitches of 20-WT, 77-E7 and 19-WT) exhibited positive INTER scores, suggesting unfavourable docking interactions; all others had negative INTER scores.Fig. 8Secondary structures parsed from tertiary structures visualized by forna.a−f 20-WT (DSM3228) lysC riboswitch structures predicted by different programs: a simRNA62 with secondary structure constraints, b RNAcomposer63 with secondary structure constraints, c Vfold65 with secondary structure constraints, d modeRNA64 with 3diz as 3D template, e trRosettaRNA66 without secondary structure constraints, f Alphafold367 without secondary structure constraints, and g 3diz PDB structure as a reference. Secondary structures parsed from tertiary structures visualized by forna. a−f 20-WT (DSM3228) lysC riboswitch structures predicted by different programs: a simRNA62 with secondary structure constraints, b RNAcomposer63 with secondary structure constraints, c Vfold65 with secondary structure constraints, d modeRNA64 with 3diz as 3D template, e trRosettaRNA66 without secondary structure constraints, f Alphafold367 without secondary structure constraints, and g 3diz PDB structure as a reference. Discussion Our comparative genomic analysis of 78 Bacillus and Priestia strains confirmed a conserved reliance on the acetylase route for lysine biosynthesis, with universal presence of dapH, dapX, and ykuR. The detection of dapL in three strains suggests a strain-specific utilisation of the aminotransferase route, previously described in archaea and select eubacteria11 but not previously reported in Bacillus or Priestia species. We identified eight genes under lysine riboswitch control, including biosynthetic genes (lysC, lysA, dapA) and transporters (lysP, arcD, ydgI, potE). Riboswitch regulation of biosynthetic genes lysA, dapA and transporter gene lysP were exclusive to Priestia, a genus for which lysine riboswitch distributions remain largely unexplored. In contrast, regulation of transporter genes arcD, ydgI, and potE was unique to Bacillus. Mukherjee et al. (2017) show that B. amyloliquefaciens and B. licheniformis harbor lysine riboswitches upstream of lysC and yvsH28, the latter functionally analogous to arcD and ydgI. Geissler et al. (2021) report that yvsH is annotated with a lysine riboswitch in B. subtilis29, supporting our inference that arcD, as an yvsH homolog, is riboswitch-regulated. Notably, riboswitch control of potE has not been previously described for B. subtilis. Most strains harboured one transporter gene under riboswitch regulation, consistent with the regulatory patterns observed in Bacillales28. Riboswitch control of lysC, encoding aspartokinase, was conserved across all strains and species. While 13 strains contained a single lysC copy, the majority (n = 65) possessed multiple copies in two distinct operons. Riboswitch-controlled lysC variants formed a separate phylogenetic clade from non-regulated copies, suggesting functional divergence. Given the central role of lysC in lysine biosynthesis and its regulation by lysine-responsive riboswitches15, we targeted lysC riboswitches for mutagenesis using a sub-lethal concentration (0.4 mM) of lysine analogue AEC. Most mutants (17/21) were isolated from the medium containing EAA. All mutants were able to grow on a defined medium without EAA, except 109E11A and 109E11B which resembled their wild-type, indicating AEC mutagenesis did not induce additional essential amino acid auxotrophies, unlike other reported AEC mutants5. We hypothesised that AEC exposure would primarily induce mutations in the lysC riboswitch, as previously reported for Bacillus subtilis15,17. While three mutants did carry lysC riboswitch mutations, our findings reveal a broader mutation landscape. Among 21 AEC-resistant mutants, transporter gene mutations were most prevalent, particularly in arcD (n = 8), ydgI (n = 4), and lysP (n = 2), indicating strong selective pressure on lysine transport systems. Additionally, two mutants harboured mutations in lysine biosynthesis genes—dapF in both, and dapA in one, consistent with a previous report linking dapF mutations to AEC resistance and intracellular lysine levels in Escherichia coli30. To our knowledge, this is the first report of AEC-induced mutations in arcD, lysP, ydgI, dapA, and dapF in Bacillus and Priestia strains. These findings support three known AEC resistance mechanisms (1) reduced uptake of AEC via transporter mutations, (2) biosynthetic rewiring through DAP pathway gene mutations, (3) regulatory changes, e.g. via lysC riboswitch mutations that decrease AEC binding and increase lysine production30. We did not observe mutations in lysyl-tRNA ligases or lysine degradation genes, which represent additional resistance mechanisms described in other species30,31. Transporter mutations likely confer resistance by reducing AEC uptake or enhancing its efflux. The lysP gene encodes a lysine-specific permease, while arcD and ydgI, annotated as arginine/ornithine antiporters (yvsH), may also facilitate lysine transport. This is supported by evidence of competitive uptake among lysine, arginine and ornithine and reversal of AEC toxicity by arginine and ornithine32, as well as structural similarities in arginine and AEC binding to transporters33. In two B. subtilis mutants, arcD rearrangements led to the emergence of potE, a bidirectional putrescine:ornithine antiporter34, potentially altering AEC transport dynamics. The riboswitch-regulated expression of potE in strain PRO118 further supports its involvement in lysine metabolism and possibly AEC transport. Transporter-based resistance mechanisms have also been reported in E. coli, Thermus thermophilus, and Corynebacterium glutamicum, involving mutations in lysine and arginine transporters such as lysP, cadB, argT, and lysI30,32,33,35. In our dataset, each mutant carried a single mutation in a transporter gene. Notably, mutant 109-E11A harboured an 81 bases deletion in arcD, and mutant 122-E5 had a single nucleotide deletion in ydgI, with no other detectable mutations. These cases strongly implicate arcD and ydgI as direct mediators of AEC resistance in Bacillus and Priestia. Seven AEC-resistant mutants were selected for fermentation profiling. All showed increased lysine production in lysine-supplemented media, and three (27-H10, 20-G9, 19-F3) also produced more lysine in lysine-free media. Strain 109-E11B contained only two mutations: an 81 bases deletion in ydgI and a single insertion in a hypothetical protein, making it likely that the transporter gene mutation was directly responsible for the elevated lysine concentration. This link is further supported by strain 109-E11A, which carried the same 81 bases deletion in ydgI and had no other detectable mutations. The improved lysine yields in strains with transporter gene mutations are likely due to reduced intracellular lysine levels, which may alleviate negative feedback on lysine biosynthesis. Wild-type strain P. megaterium 25-WT spontaneously acquired AEC resistance during cultivation on AEC medium, matching its mutant (25-H8) in lysine yield and underscoring the strong selective pressure exerted by AEC. To assess the stability of AEC resistance and lysine production, we compared pre-cultivation on nutrient-rich BHI medium versus selective AEC medium for three riboswitch mutants. While mutant 20-G9 retained resistance across conditions, 77-E7 and 27-H10 showed replicate-specific loss of resistance following BHI pre-cultivation. Despite the loss of AEC resistance, lysine production remained high in 77-E7 and 27-H10 in lysine-supplemented medium compared to their wildtypes. Whole-genome sequencing confirmed that lysC riboswitch mutations were retained in mutants 27-H10 and 20-G9 regardless of pre-cultivation medium, while mutant 77-E7 reverted to the wild-type riboswitch sequence. The ydgI mutation in 77-E7 remained stable, suggesting that transporter mutations may be more genetically robust than riboswitch variants under nutrient-rich conditions. The loss of AEC resistance in 27-H10 despite retention of the riboswitch mutation suggests that lysC riboswitch mutations may not be crucial to confer resistance to AEC. This is further supported by the presence of an additional mutation in a hypothetical protein in this strain. The transient loss of resistance in 27-H10 may also reflect regulatory compensation, including shifts in global metabolic control, altered expression of transport or biosynthesis genes, or epigenetic-like mechanisms that temporarily change lysine pathway regulation and AEC sensitivity. While our findings align with previous work showing that riboswitch mutations can mitigate AEC-induced lysine starvation by maintaining lysC transcription15,17, they challenge the notion that such mutations alone are the primary drivers of AEC resistance. Our findings suggest that riboswitch-targeted AEC mutagenesis can yield high-lysine-producing strains, even when resistance is transient. Moreover, pre-cultivation on AEC medium further enhanced lysine production under some conditions, supporting its use in strain optimization strategies. Among the AEC-resistant mutants, 20-G9 emerged as a top performer, exhibiting consistently high lysine production across all tested conditions. Compared to its wild-type, 20-G9 produced 156-fold more lysine in lysine-free medium and 4.6-fold more in lysine-supplemented medium. This phenotype was confirmed in both liquid fermentation and solid-state fermentation of oat. In liquid cultures, 20-G9 accumulated significantly higher intracellular lysine at 22 h and secreted markedly more extracellular lysine across all media. The substantial extracellular accumulation suggests active lysine export, potentially mediated by the lysine secretion gene argO, which was identified in 20-G9 and has previously been implicated in lysine efflux in Bacillus species36. In oat fermentation with 20-G9, free lysine levels peaked at 8 and 22 h, significantly exceeding both initial oat lysine content and levels produced by 20-WT. Although free lysine had declined after 46 h, it remained elevated compared to the wildtype. The decline in free lysine likely reflects microbial consumption during prolonged fermentation as well as possible degradation or incorporation into newly synthesized proteins. Total lysine, however, remained unchanged across strains and time points, indicating that bacterial contribution to total lysine is modest due to limited biomass. For lysine-poor food matrices such as oats, rice, and wheat, even modest increases in total lysine would provide a meaningful nutritional benefit by helping alleviate the limiting-amino-acid constraint in plant-based diets. Nonetheless, the increase in free lysine provides a compelling proof of concept for in situ lysine fortification of food matrices. Free lysine is nutritionally relevant in terms of bioavailability but may also contribute to flavour enhancement37 and facilitate microbial cross-feeding38. These findings position 20-G9 as a promising candidate for food biotechnology applications. Lysine yields may be further improved by optimizing fermentation conditions, such as nutrient supplementation, moisture control, and temperatures around 37–40 °C, as shown in previous studies on Bacillus and Priestia mutants20,21,23. To better understand the regulatory architecture underlying lysine production, we developed a bioinformatics pipeline to compare riboswitch structures across wild-type and AEC-resistant strains. The pipeline integrates secondary structure prediction based on thermodynamic and evolutionary constraints, tertiary modelling and ligand docking. Comparative analysis revealed distinct structural differences between riboswitches regulating lysine biosynthesis (e.g., lysC, dapA, lysA) and those controlling transport (e.g., arcD, ydgI, potE, lysP). Biosynthesis-associated riboswitches showed greater sequence variability and divergence in helices P3 and P4, suggesting functional specialization and potential differences in ligand responsiveness. We identified novel AEC-induced mutations in lysC riboswitches, including a 10 bases deletion in 27-H10 affecting lysine-binding nucleotides and SNPs near the lysine binding pocket in 77-E7 and 20-G9. These mutations differ from previously characterized variants in literature13,15,39–42, expanding the known spectrum of lysC riboswitch alterations. The 77-E7 SNP falls within regions shown to undergo ligand-induced structural modulation and reduced spontaneous cleavage, suggesting a role in lysine recognition15,17. The 27-H10 deletion and 20-G9 SNPs overlap with conserved structural elements critical for regulatory function, as demonstrated by mutational analyses disrupting antiterminator formation and ligand responsiveness41. These findings support the hypothesis that the mutations may influence riboswitch activity through altered ligand binding or RNA folding dynamics. To investigate whether the identified riboswitch mutations affect lysine binding, we performed tertiary structure modelling and ligand docking. AlphaFold3 was the only tool to consistently predict the P2–P3 kissing-loop pseudoknot, a key structural feature for lysine interaction, likely because its model was trained on lysine riboswitch structures. Alphafold3 was therefore used for docking simulations with rDOCK despite low model confidence scores (ptm < 0.5) in most riboswitches. However, these simulations did not reveal consistent differences in binding energies between wild-type and mutant sequences, underscoring current limitations in RNA structure prediction for riboswitches with complex, ligand-responsive folding. While these challenges prevented conclusive assessment of binding alterations, our findings nonetheless highlight the value of riboswitch-informed screening strategies. Secondary structure predictions offer useful insights into the location and structural impact of mutations relative to ligand-binding sites, but they do not always capture specific structural motifs like pseudoknots. Tertiary structure predictions can inform on the global 3D shape, but current tools still struggle to accurately model local structural elements, which limits the reliability of docking simulations. Accordingly, the absence of clear docking trends underscores that these computational approaches are hypothesis-generating rather than definitive and represents a key limitation when inferring mechanistic effects of the mutations. Future work should combine improved structural prediction tools with experimental validation to clarify the effects of the mutations identified in this study. Genetic engineering approaches that combine mutations, such as introducing the 20 G9 riboswitch variant alongside transporter mutations in arcD, lysP, or ydgI, could reveal whether these changes have cumulative effects on lysine production. Functional assays should quantify AEC and lysine uptake and efflux in transporter mutants to determine whether altered transport underlies their resistance phenotypes, and in vitro ligand binding assays should assess changes in lysine or AEC affinity in riboswitch variants. Together, these studies will help identify rational combinations of mutations in transport, biosynthesis, and riboswitch elements to engineer next-generation lysine overproducing strains. This study elucidates the genetic and regulatory mechanisms underlying AEC-induced lysine overproduction in Bacillus and Priestia species. We report the first identification of the aminotransferase gene dapL in Priestia megaterium and Bacillus licheniformis, revealing previously unrecognized DAP pathway diversity. We generated 21 AEC-resistant mutants, of which transporter gene mutations - arcD, ydgI, and lysP were most frequent. Seven mutants exhibited increased lysine production. Although resistance to AEC varied in stability, riboswitch mutations consistently enhanced lysine biosynthesis, particularly in strain 20-G9, which also elevated free lysine levels during oat fermentation. Structural and bioinformatic analyses revealed distinct features between riboswitches regulating biosynthesis and transport, highlighting functional specialization. Together, these findings establish a multi-pathway framework for lysine enhancement and provide novel genetic targets, including transporter, biosynthetic, and regulatory elements, for microbial strain development aimed at nutritional fortification of plant-based foods. Methods Re-sequencing of 78 Bacillus and Priestia strains To improve genome qualities, the collection of 78 Bacillus and Priestia strains described by Wiedenbein et al. (2023)24 was re-sequenced. DNA repair was performed using a modified version of the Oxford Nanopore Technologies (ONT) ligation protocol (SQK-LSK114). Specifically, 10 µL of DNA extract (Wiedenbein et al., 202324) was combined with 3.5 µL NEBNext FPE DNA Repair Buffer, 2 µL NEBNext FFPE DNA Repair Mix, and 37 µL MilliQ water, followed by purification with AMPure XP beads. Sequencing libraries were prepared using the R10 Rapid kit (SQK-RBK114.96) and sequenced on the PromethION 2 Solo (ONT) on a R10.4.1 flow cell for 72 h under standard conditions. Bacterial strains and pre-cultivation on BHI From the collection of 78 Bacillus and Priestia strains, eleven strains were selected for mutagenesis (Table 1) based on a diversity of genomic characteristics. Strains were pre-cultivated twice in Brain Heart Infusion (BHI) medium (Himedia, Germany). The first preculture was initiated by inoculating a single colony from a freshly streaked BHI agar plate (2% agar) into 10 mL BHI broth in 50 mL Falcon tubes with loosely secured lids, incubated at 30 °C for 16 h with shaking at 200 rpm. For the second preculture, 2% (v/v) of the first culture was transferred to fresh 10 mL BHI broth and incubated under the same conditions for 4 hours. Cells were harvested by centrifugation (4600 g, 10 min, Eppendorf centrifuge 5290 R), washed twice with 0.9% NaCl, and resuspended in 1 mL of 0.9% NaCl. Chemically defined media Different media variants were developed based on the Pafoba medium, a chemically defined medium optimised for the growth of food-relevant Bacillus species43. The standard Pafoba medium with lysine was prepared accordingly. A lysine-free version was made by substituting the Gibco MEM amino acid solution with an equivalent essential amino acid mix in 0.1 M HCl, excluding lysine. For AEC-supplemented medium, the lysine-free version was supplemented with 0.4 mM S-(2-Aminoethyl)-L-cysteine hydrochloride (Sigma Aldrich, A2626, lot BCCJ3020, CAS 4099-35-8) added from a 100× stock in demineralized water. AEC-containing plates including 2% (w/v) agar were prepared by cooling an autoclaved fraction of Pafoba base ingredients with agar to approximately 45 °C, followed by aseptic mixing with the remaining sterile ingredients. AEC medium without essential amino acids (EAA) was prepared similarly to the AEC-supplemented medium, but by omitting the addition of any essential amino acids. Pafoba without EAA was prepared similarly to the standard Pafoba, but by omitting the Gibco MEM EAA solution. AEC-induced mutagenesis Bacterial cells were subjected to chemical mutagenesis to select for increased AEC resistance. After pre-cultivation on BHI, two inocula were prepared: (1) OD-adjusted cultures (OD600 = 0.64) and (2) 100× dilutions of these cultures. For each inoculum, 10 µL was added to 190 µL of either Pafoba medium with AEC or Pafoba with AEC without EAA in a 96-well plate. Two biological replicates per strain were included. Plates were sealed with an oxygen-permeable film (Breath-Easy, Sigma Aldrich), incubated at 30°C with shaking (900 rpm) for 37 h. Aliquots (60 µL) from wells showing visible growth were spread-plated on Pafoba with AEC agar. Single colonies were purified by re-streaking, cultivated in BHI for 16 h at 30 °C, and cryopreserved in 20% glycerol. In addition, 1 mL was centrifuged (10,000 g, 3 min) to harvest the cell pellets, stored at −20 °C for DNA extraction. Liquid fermentation in microtitre plate All AEC mutants (n = 21) were tested for growth on Pafoba with and without EAA. After pre-cultivation and washing, cell densities were adjusted to OD600 = 0.04. Inocula (10 µL) were added to 190 µL medium in 96-well plates, sealed with Breath-Easy film (Diversified Biotech BEM-1). OD620 was measured every 30 min for 42 hours at 30 °C using a Multiskan FC Quick (Thermo Fisher) with continuous shaking. Each condition included three biological replicates. Additionally, seven AEC-resistant mutants and their wildtypes (Table 1) were selected for growth profiling in microtitre plates and lysine quantification after 22 hours. Inocula preparation followed the same protocol. Each strain was inoculated (10 µL) into three media (190 µL each): Pafoba, Pafoba without lysine, and Pafoba with AEC, and incubated at 30 °C for 22 h. Each condition included three biological replicates (independent cultures) and 3-4 technical replicates (repeated cultures from the same biological replicate). After confirming similar growth profiles, technical replicates were pooled per biological replicate (600–800 µL total), to ensure sufficient volume for cell lysis and lysine quantification. Effect of pre-cultivation medium on mutants For mutants 77-E7, 27-H10, and 20-G9, the impact of pre-cultivation on growth, lysine production, and mutation stability was assessed. In addition to BHI, these strains were also pre-cultivated on AEC-containing media. This followed the same procedure as BHI pre-cultivation, except colonies were streaked on Pafoba with AEC agar and transferred to Pafoba with AEC broth for the first preculture. After the first pre-culture step, cells were harvested by centrifugation (10,000 g, 3 min) and stored at −20 °C for DNA extraction. The second 4-hour preculture was performed in BHI to obtain sufficient biomass. Inocula were then prepared from the second preculture by washing the cells, resuspending them in 0.9% NaCl, and adjusting to OD600 = 0.04, as previously described. These inocula were used for microtitre plate fermentation to assess growth and free lysine production as described above. Liquid fermentation in shaking culture Mutant 20-G9 and its wildtype 20-WT were tested in shaking culture fermentation. Inocula were prepared by pre-cultivation on BHI, followed by washing and adjustment to OD600 = 0.4. Fermentations were initiated by mixing 100 µL of inoculum with 10 mL of medium in 50 mL Falcon tubes, using three different media: Pafoba, Pafoba without lysine, and Pafoba with AEC. Tubes were loosely capped and incubated at 30 °C with shaking at 200 rpm. Samples were collected at 0, 8, and 22 hours for growth measurement (all time-points) and lysine quantification (8 h and 22 h). Cell counts were determined by spot-plating 5 × 10 µL serial dilutions on BHI agar and incubating at 30 °C for 24 h. Dry cell weight at 22 hours was measured using pre-weighed 0.2 µm membrane filters (Whatman ME 25, 47 mm, Sigma Aldrich). Filters were dried in a microwave (100 W, 30 min), stored in a desiccator, and used for vacuum filtration of 3 mL sample, followed by rinsing with 6 mL sterile MilliQ water. Filters were dried again, cooled, and reweighed to determine biomass. Solid-state fermentation of oat Whole oats (Aurion Glutenfri Hel Havre Økologisk Demeter, LOT#0073) were soaked in sterile MilliQ water (1:2 w/v) for 16 hours at 4 °C, drained, resoaked with new milliQ water (1:2 w/v), and steamed in an autoclave at 100 °C for 5 min. Inocula of 20-G9 and 20-WT were prepared by pre-cultivation on BHI, cell washing and adjustment to 0.4 OD600. The oats were inoculated with 1 mL of OD600-adjusted inoculum per 100 g of heat-treated oat, mixed and portioned into four sterile 9 cm petri dishes ( ~ 30 g each). Petri dishes were taped and incubated at 30 °C. At 0, 8, 22, 46 h of fermentation, individual petri dishes were sampled for cell count, pH and lysine quantification. Cell counts were obtained by stomaching 5.0 g oat with 50 mL 0.9% NaCl (1:10 w/v) and spot-plating serial dilutions. The pH of 5.0 g of oat sample mixed with 5 ml of sterile milliQ water was determined. For lysine quantification, 15.0 g of oat was freeze-dried, and ground with 5 g dry ice using an IKA A11 analytical mill for 30 seconds. Cell lysis To quantify free lysine (sum of extra- and intracellular content), samples from microtitre plates and shaking cultures were lysed. Additionally, for shaking culture samples, 1 mL was centrifuged (10,000 g, 10 min) to separate supernatant (extracellular) and pellet (intracellular) fractions. Pellets were resuspended in 1 mL sterile milliQ water prior to lysis. Pooled microtitre samples and resuspended pellets were transferred to 2 mL MBio lysis tubes containing 0.1 mm silica beads and lysed using a FastPrep-24 bead beater (MP Biomedicals) for three 60-second cycles at 4.5 m/s, with 2-min cooling intervals on ice. Lysates were collected after centrifugation (10,000 g, 5 min). Protein removal for free lysine determination Proteins were precipitated by mixing 200 µL of lysate or supernatant with 200 µL of 12.5% ice-cold trichloroacetic acid (TCA). For fermented oat samples, 100 mg of freeze-dried ground oat was mixed with 1 mL of 0.1 M HCl, ultrasonicated for 10 minutes, and 200 µL of the extract was treated with 200 µL of 12.5% ice-cold TCA. All mixtures were vortexed and incubated at 4 °C for 1.5 h, then centrifuged (10,000 g, 20 min). Supernatants were collected and neutralized with freshly prepared 1 M NaOH. Protein hydrolysis for total lysine determination Total lysine (sum of free and protein-bound content) in fermented oat samples was quantified. Protein content of oats was determined via nitrogen analysis using a CNS macro analyzer (vario MACRO cube, Elementar), with nitrogen converted to protein using an N-factor of 5.3444. Microwave-assisted hydrolysis was performed following Poojary et al. (2020)45: oat samples equivalent to 5 mg protein were mixed with 3 mL of 4 M methanesulfonic acid (MSA) containing 0.2% (w/v) tryptamine, flushed with nitrogen, sealed, and heated using a Biotage Initiator+ microwave synthesizer at 150 °C for 1 min and 165 °C for 10 min. Hydrolysates were neutralized with 4 M NaOH. Quantification of free and total lysine Samples were mixed 1:1 with 50 µM 6-aminocaproic acid (internal standard) and filtered through 0.2 µm syringe filters (Phenomenex) into HPLC vials. Eight concentrations of mixed amino acid standards were prepared, according to Bayati & Poojary (2025)46. Analysis was performed using UHPLC with fluorescence detection (Ultimate 3000RS, Thermo Fisher Scientific) and an Agilent AdvanceBio-AAA column (100 × 3.0 mm, 2.7 µm). Derivatization was done in-needle using o-phthalaldehyde. Chromatography conditions were equivalent to Zainudin et al. (2019)47. Whole-genome sequencing of mutants and corresponding wildtypes To ensure that no additional mutations had accumulated in the wild-type strains between their initial sequencing and the mutant selection process, we re-sequenced both mutants and the corresponding wild-type replicates (Table 1) used to generate the mutants, a total of 14 replicates from 11 wild-type strains. Genomic DNA of wildtypes and newly generated mutants was extracted from bacterial pellets using the bead-beat micro–AX Gravity kit (A&A Biotechnology, Poland, Catalogue nr. 106-100). Libraries were prepared without size selection or shearing using the ONT native barcoding kit (SQK-NBD114.96) and protocol NBE_9171_v114_revP_15Sep2022. Sequencing was performed on a PromethION 2 Solo on a R10.4.1 flow cell for 72 h using standard settings. Assembly and annotation of all genomes Basecalling was performed using Dorado v0.8.1 in super accuracy (sup) mode. Genome assembly was conducted with Hybracter v0.11.248 using the following parameters: quality score ≥15, read length ≥1000 bp, assembly coverage of 100×, plasmid assembly with Plassembler, and no polishing with Medaka. If assemblies were incomplete, the assembly coverage was increased to 110× or 150×. All genomes were annotated using Prokka v1.14.6 with the --rfam flag to include non-coding RNA annotations49. Phylogenetic trees and riboswitch regulation of lysine pathway genes A phylogenetic tree of the 78 Bacillus and Priestia strains was generated by concatenation of 92 conserved single-copy bacterial core genes according to Na et al. (2018)50. Maximum-likelihood trees were generated with FastTree51, using Halobacillus halophilus DSM 2266 (GenBank: GCA_000284515.1) as outgroup. Annotated genomes were screened for lysine biosynthesis and lysine transport genes using gene names, EC numbers, and COG annotations. Maximum likelihood phylogenetic trees of the lysC gene were constructed based on protein sequence alignments performed using MUSCLE52. Tree inference was carried out with IQTREE using default parameters, with 1000 bootstrap replicates53. Lysine riboswitches were identified in all genomes, and the first downstream gene on the same strand was recorded. Operon structures were predicted using Operon-mapper54 and matched with riboswitch locations to infer the riboswitch control of genes. Phylogenetic trees were visualized using iTOL55. Variant calling and functional annotation Variant analysis was performed using a custom pipeline available at https://github.com/reillyeo/Cerebro. Briefly, nucmer (MUMmer v3.1) was used to generate a pairwise alignment of each mutant genome with its corresponding wild-type genome, and a list of all point mutations and their genomic coordinates was extracted from the resulting alignment using nucmer’s show-snps utility. This data was converted to BED format and intersected with Prokka-annotated features of the wild-type genome (GFF3) using bedtools intersect (v2.30.0), linking each mutation to its corresponding gene. Consecutive insertions and deletions were merged into single events. Prokka-generated protein sequences of the affected genes were queried using EggNOG-mapper web (v2.1.12) to assign COG classifications and other functional annotations56. Gene clusters from wild-type and mutant strains were compared using Clinker57. Riboswitch secondary structure prediction Riboswitch secondary structures were predicted based on wild-type strains and AEC-resistant mutants. Unique riboswitch sequences were preselected, followed by further selection based on <95% ClustalΩ pairwise sequence identity58. Cluster representatives were chosen by including the sequence representing the largest number of sequences inside the cluster, while ensuring inclusion of all unique AEC-resistant mutants with their wild-types. Multiple sequence alignment (MSA) of riboswitch sequences was performed through foldalignM version 2.1 BETA, sub revision 7025. A consensus RNA secondary structure was predicted from the MSA using PETfold26. Aligned sequences of foldAlignM and the consensus structure of PETfold were combined to visualize consensus secondary structures with R2R59. PETfold base-pair reliabilities were used to determine the dot-bracket notations as constraints for tertiary structure predictions. Bound base pairs were maintained if the reliability was above a 0.8 threshold. A base-pair distance matrix was generated by obtaining individual secondary structures with RNAfold and determination of base-pair distances with RNAdistance, both programs from the ViennaRNA tools package (v2.5.1)60. The RNAsnp webserver was used in Mode 1 with 150 SNP flanking regions61 to predict the mutant SNP effect (20-G9 and 77-E7) on the secondary structure relative to their corresponding wild-types (20-WT and 77-WT). Riboswitch tertiary structure prediction and molecular docking Tertiary structures of the lysine riboswitches were predicted using six programs, including the ab initio method simRNA62, the template-based programs RNAComposer63, modeRNA64 and Vfold pipeline65 and deep-learning programs trRosettaRNA66 and Alphafold367. All programs were run via their webservers with and without PETfold secondary structure constraints, except Alphafold3, where no secondary structure constraints were possible and modeRNA, which was run with Protein Databank (PDB) T. maritima tertiary structure 3diz as template. The tools were benchmarked by aligning the experimentally determined 3diz PDB structure with the predicted structures and determining Root Mean Square Deviation (RMSD) and Interaction Network Fidelity (INF) ALL function scores with RNAQUA version 1.168. Tertiary structure predictions were parsed back to secondary structures using DSSR (version v2.5.4-2025jun04)69, which were then visualized with forna70. Riboswitch-lysine binding energies were predicted based on tertiary structures generated by Alphafold3 using rDOCK71 with default parameters. To place the lysine under the correct initial position, each individual riboswitch was aligned against the PDB 3diz reference with PyMol72, and the lysine structure was extracted to obtain the right coordinates for ligand binding analysis. Statistics and data analyses Chromatographic data were pre-processed with Chromeleon version 7.2.7 (Thermo Scientific, MA, USA) and exported to Microsoft Excel. All subsequent statistical analyses were conducted in R Studio (version 4.4.1). Residuals from parametric models (ANOVA and linear models) were inspected visually and statistically to confirm normality and homoscedasticity. Log transformation was applied where necessary to meet model assumptions. For microtitre plate fermentations, two-way aligned rank transform (ART) ANOVAs (R package ARTool73) were performed separately for each mutant–wildtype pair (n = 7) to assess the effects of strain, cultivation medium, and their interaction on free lysine concentrations. ART was selected due to persistent violations of normality assumptions. These analyses included only samples pre-cultivated in BHI medium. To investigate the effect of pre-cultivation medium (BHI vs. AEC) on lysine production in selected mutants (20-G9, 77-E7, 27-H10), lysine concentrations were log-transformed to meet normality assumptions. The data were analysed using two-way ANOVAs with pre-cultivation and cultivation medium as fixed factors. For both microtitre datasets, post hoc comparisons were conducted using estimated marginal means (R package emmeans74) with Holm correction applied within each medium and across all comparisons. For the shaking culture fermentations, bacterial growth (log CFU/mL), extracellular lysine and intracellular lysine (both log-transformed) were analyzed using linear mixed models (R package lme475) with strain, medium, and time as fixed effects, and biological replicate as a random effect. Lysine yield at 22 h was analyzed using a linear model with strain and medium as fixed effects. For the oat fermentations, free and total lysine levels (log-transformed), growth (log CFU/g), and pH were analyzed using linear mixed models with strain and time as fixed effects, and biological replicate as a random effect. Post hoc comparisons were performed using emmeans, with Holm-adjusted p-values. Pairwise comparisons focused on mutants versus wildtype strains and additionally included comparisons between media for growth and pH, and between timepoints for lysine levels. All comparisons were conducted in the context of significant interactions where applicable. One sample from the shaking culture fermentations (20-WT, time 8 h, lysine-supplemented medium, extracellular lysine) was excluded due to consistent and substantial deviation across multiple amino acid measurements, likely reflecting a technical artifact. No other replicates showed similar behaviour. Following this exclusion, two biological replicates were retained for the affected condition; all other conditions included three biological replicates. Exclusion did not alter statistical conclusions. All analytical measurements were repeated once. All statistical tests were two-tailed, with significance assessed at α = 0.05. Exact F values, degrees of freedom and P values for all main statistical tests are provided in Supplementary Data 4. Supplementary information Supplementary Materials Supplementary Data1 Supplementary Data2 Supplementary Data3 Supplementary Data4 Supplementary Materials Supplementary Data1 Supplementary Data2 Supplementary Data3 Supplementary Data4