GABA-producing Lb. brevis 145 and a reference strain, Lb. plantarum WCFS1, were cultivated in Lactobacilli MRS medium (BD Company, Franklin Lakes, NJ). Unless otherwise stated, Lactobacillus strains used in this study were anaerobically cultivated in the above medium statically. For cloning and protein expression purposes, E. coli strains as indicated in Table 1 were cultivated in Luria-Bertani (LB) medium with or without antibiotics supplemented to the medium. The model strain of high GABA-producing Lb. brevis 145 was completely sequenced in this study (please refer to Supplementary Materials and Methods for details). The NCBI accession number for the chromosome of Lb. brevis NPS-QW-145 is CP015398.

Currently, there are more than a thousand of strains of LAB and bifidobacteria that have been sequenced and deposited in NCBI-GenBank database. Thus, genomic survey was carried out for the presence or absence of gad operon and genes encoding glutamate decarboxylase in most of the sequenced strains (all assembly levels; 890 strains) of Bifidobacterium and LAB including Lactobacillus, Lactococcus, Leuconostoc, Pedicoccus, Oenococcus, Weissella, and Streptococcus thermophilus released in NCBI-GenBank genome database. Only the species having at least two strains sequenced were included in this genomic survey. In addition, the distribution of acid resistance systems including F₀F₁-ATPase system, amino acid/cation:proton antiporters, glutamic acid decarboxylase (GAD) system, tyrosine decarboxylase (TDC) system, agmatine deiminase (AgDI) system, arginine deiminase (ADI) system and urea system in Lb. brevis was also surveyed in terms of presence or absence of above AR systems in the sequenced strains.

Extracellular pH (pH_ex) was measured directly with the pH meter. The fluorescent probe–5(6)-carboxyfluorescein diacetate N-succinimidyl ester (cFDA-SE; Thermo Fisher Scientific) was used to label bacterial cells for intracellular pH (pH_in) measurement. There are several reports on the pH_in measurement of bacteria suspended in citrate-based buffer (Siegumfeldt et al., 2000; Teixeira et al., 2014); however, this buffer did not allow the discrimination between cells equilibrated to pH between 3.5 and 5.0 thus limiting the detection range of cFDA-SE probe (Hansen et al., 2016). Thus, phosphate-buffered saline (PBS) was used to re-suspend bacterial cells for pH_in measurement in the range of pH 3.5–7.0 in this study according to the previous studies (Breeuwer et al., 1996; Hansen et al., 2016). Briefly, the pH_ex of cultures was measured first, and bacterial cells were centrifuged, washed and re-suspended in PBS where its pH level was adjusted to the pH_ex of the cultures. The cell density was adjusted to an optical density (λ = 600) of 0.6–0.7, followed by the addition of both cFDA-SE and glucose to the final concentrations of 10 μM and 10 mM, respectively. The mixture was incubated in the dark at 37°C for 30 min. The stained cells were later harvested by centrifugation at 12,000 × g and 4°C for 5 min, and were re-suspended in the PBS buffer with the same pH_ex prior to the measurement.

Fluorescence intensities of the stained cells were measured in the Fluorescence Spectrophotometer F-7000 (Hitachi High-technologies, Shenzhen, China) at the excitation wavelengths of 488 nm (pH-sensitive) and 435 nm (pH-insensitive) by rapidly altering the monochromator between both wavelengths. The emission wavelength was started from 400 to 650 nm. Both excitation and emission slit width was 5 nm. For the calibration curve (pH 3.5–7.0) of Lb. brevis, the stained cells were suspended in PBS buffer having different pH values adjusted by hydrochloric acid (HCl). Valinomycin (0.2 mM in methanol; Sigma) and nigericin (0.2 mM in methanol; Sigma) were added to the final concentrations of 5 μM followed by incubation at 37°C for 10 min resulting in the equilibration of both potassium and proton ions across cell membrane. The equilibrated cells were later measure the same as described above.

Acid resistance assay was carried out based on previously method described (Seo et al., 2015). Briefly, cells of Lb. brevis 145 grown for 12 h (early stationary phase) in Lactobacilli MRS broth were inoculated into Lactobacilli MRS broth (pH 2.5) supplemented with or without 1 g/L of MSG, arginine, agmatine sulfate or tyrosine. The initial cell density after inoculation for acid challenge was 1.4 × 10⁸ CFU/mL. Cells were incubated at 37°C for 2 h statically. Plate count method was then applied to assess the survival rates of Lb. brevis cells under above conditions. For acid challenge assay, the 3-h Lb. brevis cultures (lag phase, not acid-adapted cells) were centrifuged, washed and suspended in PBS buffer (pH 7.0), followed by cFDA-SE staining as described above. The stained cells were then placed in fluorescence spectrophotometer and their fluorescence emission intensities were measured every 5 min at 37°C. Acid challenge was achieved at the point of 5 min by adding hydrochloric acid that changed pH_ex from 7.0 to 3.5. After another 5 min, the stock solution (100 mM) of individual substrate (glutamate, arginine, agmatine, and tyrosine) was added to the final concentration of 10 mM in the solution. The changes in pH_in were recorded accordingly as per the above description.

Cells of Lb. brevis 145 grown for 12 h (early stationary phase) in Lactobacilli MRS broth were inoculated into Lactobacilli MRS broth (pH 6.5) supplemented with or without the mixture of MSG, arginine, agmatine sulfate salt and tyrosine (1 g/L each; Sigma). However, tyrosine was partially dissolved in MRS broth (pH 6.5), and the insoluble tyrosine (solid form in the broth) was removed by membrane filtering (0.45 μm). The initial cell density after inoculation for acid challenge was 1.0 × 10⁸ CFU/mL and the cells were incubated at 37°C for 36 h statically. Samples were collected at the time point of 0, 3, 6, 9, 12, 18, 24, and 36 h. Subsequently, measurements of acids, amino acids, amines and gene expression were carried out as discussed below. Plate count method was used to assess the cell viability, and pH_in and pH_ex measurements were carried out as stated above.

Acids in the culture broth were analyzed by HPLC (Model Shimadzu LC-2010A, Shimadzu Corporation, Kyoto, Japan) equipped with Aminex HPX-87H column (300 × 7.8 mm; Bio-Rad). Briefly, cell-free supernatants (diluted if necessary) after removal of bacterial cells and passing through Acrodisc® syringe filter with 0.20 μm Supor® membrane (Pall, Ann Arbor, MI, USA) were injecting into the HPLC system. Acids were eluted with 5 mM sulfuric acid with a flow of 0.60 mL/min. The temperature of the column was maintained at 50°C and the absorbance of the detector was set to 210 nm for detection.

Amino acids including glutamic acid, GABA, arginine, ornithine and tyrosine were separated and quantified as per the method we have previously developed (Wu and Shah, 2016). Briefly, cell-free supernatant was collected after centrifugation of fresh cultures. The derivatization was achieved by adding 200 μL acetonitrile, 200 μL NaHCO₃ (1 M, pH 9.8 adjusted with NaOH), 300 μL distilled water and 100 μL dansyl chloride solution (40 g/L; dissolved in acetonitrile) to 100 μL supernatant (diluted if necessary), then the mixture was kept in an oven (40°C) for 60 min. After derivatization, 100 μL of 20% (v/v) acetic acid was added to stop the reaction and centrifuged at 10,000 × g and 20°C for 2 min. The supernatant was passed through a 0.22 μm membrane filter before HPLC analysis.

Amines including agmatine, putrescine, and tyramine in the supernatants were analyzed as previously described (Dugo et al., 2006). Briefly, 1.6 mL of 10 g/L dansyl chloride solution (dissolved in acetone) was added to 1.5 mL of the supernatant (diluted if necessary). The pH of the mixture was adjusted to pH 8.2–8.3 with 40 g/L of NaCO₃ solution, then the mixture was heated in a water-bath (40°C) for 60 min in dark. After derivatization, acetone was removed under a stream of nitrogen and then the volume was made up to 5 mL with acetonitrile and centrifuged at 1000 × g and 20°C for 2 min. The supernatant was passed through a 0.22 μm membrane filter before HPLC analysis.

Dansyl amino acids and dansyl amines were separated and detected using a previously developed gradient elution mode (Wu et al., 2015) by HPLC (Model Shimadzu LC-2010A, Shimadzu) equipped with a Kromasil 5 μ 100A C18 column (250 × 4.6 mm; Phenomenex). All the standards of amino acids and amines were derivatized and analyzed as the same in the above procedures. The absorbance for dansyl-amino acids and dansyl-amines were recorded at 275 nm and 254 nm, respectively.

Total RNA extraction using hot SDS/phenol method, DNase I treatment, cDNA synthesis, qPCR assay were carried out as per previously described (Wu et al., 2015). The expression of target genes (gadA, gadB, aguB, arcB, and tyrDC in Lb. brevis 145; gadB in Lb. plantarum WCFS1) listed in Table 1 was quantified by real-time qPCR assay. Reference genes including tuf in Lb. brevis and ldhD in Lb. plantarum listed in Table 1 were used to normalize the expression of target genes in both strains. The efficacy of qPCR amplification using each pair of primers was in the range of 90–110%. Comparative critical threshold method (2^−ΔΔCt) was used to calculate the relative gene expression. The qPCR assay was performed in duplicates for each sample and independent experiments were carried out in triplicates.

For the procedure of cloning, hetero-expression and purification of GadA and GadB from Lb. brevis 145 and GadB from Lb. plantarum WCFS1, please refer to Supplementary Materials and Methods for details. The effects of pH (3.0–6.6) and temperature (30–90°C) on the purified GADs including Lb. plantarum GadB, Lb. brevis GadB, and Lb. brevis GadA and its mutants (Lb. brevis GadA Δ5 and Lb. brevis GadA +GSHM) were carried out. Low concentrations of PLP such as 20 μM is sufficient to activate the activities of both Lb. plantarum GadB and Lb. brevis GadB (Fan et al., 2012; Yu et al., 2012; Shin et al., 2014), thus the reaction mixture consisted of 5 μL of 1 M monosodium glutamate (MSG; Sigma), 5 μL of 2 mM pyridoxal 5′-phosphate (PLP; Sigma), 500 μL of McIlvaine (citrate-phosphate) buffer, and 10 μg of GAD. Previous studies have indicated the PLP- and sulfate ion-dependent activation of Lb. brevis GadA and its homologous Lb. zymae GadB (Ueno et al., 1997; Park et al., 2014); thus, for Lb. brevis GadA and its mutants, the reaction mixture contained 5 μL of 1 M MSG, 5 μL of 20 mM PLP, 500 μL of sulfate buffer (0.8 M sodium sulfate and 50 mM sodium acetate; different pH), and 10 μg of GAD. The kinetics of these enzymes was determined under their optimal pH and temperature, and the concentrations of substrate (glutamate) ranged from 0.9 to 36 mM.

All presented data in the bar charts and tables correspond to means ± standard deviation. Significant difference was carried out by using IBM SPSS Statistics 20.0 version.

Currently, there have been two major pathways reported for GABA production either through the degradation of putrescine (Puu and ADC pathways) or via the decarboxylation of glutamate (GAD pathway) in bacteria (Figure 1A). However, GABA production from putrescine degradation is not common in LAB and bifidobacteria due to the absence of Puu and ADC pathways. Interestingly, genes (gadA and gadB) encoding glutamate decarboxylases (GadA and GadB) and an intact gad operon including gadR (regulator), gadA and gadC (Glu/GABA antiporter) have been identified in all of the sequenced strains of Lb. brevis having varying origins excluding the one of strain WK12 with incomplete (contig levels) genome sequence (Figure 1C). The location of gadB is far away from gad operon in Lb. brevis suggesting there may be two different gene regulations for gadA and gadB in this species. However, arrangement of gadA and gadC close to each other in the genome could ensure the timely co-regulation of transcription and translation of gadA and gadC for GABA production (De Biase and Pennacchietti, 2012). Moreover, phylogenetic analysis of four gene components (gadR, gadC, gadA, and gadB) indicated the presence of two isoforms of glutamate decarboxylases, GadA and GadB, in Lb. brevis; all of the four components from varying Lb. brevis strains are highly conserved in this species (Figure 1B). This implies that co-existence of gadA and gadB in Lb. brevis contributes to its GABA synthesizing capacity.

In order to gain a global view on the distribution of gad operon and genes encoding glutamate decarboxylases in LAB and bifidobacteria, a genomic survey was carried out on the sequenced LAB and bifidobacteria species, of which at least two strains have been sequenced and deposited in GenBank database (accessed on 10 January 2016). As shown in Table 2, many strains carry gadA or gadB, but gadC is only present in the genomes of limited strains indicating their strain-specific characteristic of GABA production. Among gad operon-positive strains, Lb. brevis is the only one species that has close to 100% probability to carry the intact gad operon in their chromosomes. All the sequenced strains of B. dentium possessed gadBC operon in their chromosomes. However, there is very little information on GABA production from B. dentium with only gadB gene, whereas there are several reports on high GABA production from Lb. brevis with two gad genes (Li and Cao, 2010; Dhakal et al., 2012; Wu and Shah, 2016). Thus, the present study focused on only Lb. brevis for its efficient GABA synthesis. Moreover, it was found that other common reported GABA-producing species such as Lb. reuteri and Lc. lactis showed strain-specific GABA biosynthesis at the genetic level.

Although, there are several reports on GABA-producing Lb. plantarum without sequencing their genomes (Li and Cao, 2010; Wu and Shah, 2016), in this study we found that gadC is absent in all the sequenced strains of Lb. plantarum (Table 2). Hence, a fermentation study was carried out using the sequenced Lb. plantarum WCFS1 as a model organism. Although, there was an incomplete pathway for GABA metabolism in this organism based on KEGG pathway (Figures S1A,B), its viable counts, gadB transcripts level, acid profiles and GABA production did not significantly change after glutamate supplementation to the medium (Figures S1C–H). Although gadB transcript was increased in the stationary phase suggesting an enhanced GABA synthesis, gadC in LAB is required and specific for exchanging extracellular glutamate with intracellular GABA (Figure S1D). Our cultivation experiment on Lb. plantarum WCFS1 complied with previous gene knock-out studies in other bacteria with intact gad operon that GadC is an important element for GABA production (Cotter et al., 2001; Lu et al., 2013).

A genomic survey on the distribution of amino acid-dependent AR systems in all the sequenced strains of Lb. brevis was carried out. Except for universal amino acid-dependent AR systems including F₀F₁-ATPase system and amino acid/cation:proton antiporter, extra amino acid-dependent AR systems including GAD system, tyrosine decarboxylase (TDC) system, agmatine deiminase (AgDI) system and arginine deiminase (ADI) system are commonly present in all sequenced strains of Lb. brevis (Figure 2A and Table 3).

Although, the contribution of AR systems for acid resistance in other LAB such as Lb. reuteri has been characterized (Teixeira et al., 2014), individual contribution of above AR systems to resist exogenous acids in Lb. brevis has not been well demonstrated. In this study, it was found that both glutamate and arginine were more effective in increasing the survivability of Lb. brevis significantly (p < 0.05) when exposed to acidic condition (pH 2.5; HCl-adjusted) for 2 h than that by agmatine and tyrosine (Figure 2B). Moreover, both glutamate and arginine were also able to increase the intracellular pH of Lb. brevis significantly (p < 0.05) than agmatine and tyrosine after challenging cells with hydrochloric acid (Figure 2C). This suggests that glutamate and arginine increased cell viability of Lb. brevis by increasing intracellular pH of the cell thus maintaining its metabolic activity. However, the increased level of intracellular pH of Lb. brevis cells by arginine and glutamate differed greatly. This may be due to the stages of cells selected for acid resistance experiment (12-h culture; acid-adapted cells) and acid challenge assay (3-h cultures; non-acid-adapted cells). The stages of cells may affect the contents of key cytosolic enzymes including GadA, GadB, ADI, and OTC from the four amino acid-dependent AR systems (Figure 3D). In addition, the increase in the intracellular pH by glutamate may be explained by GABA synthesis that consumed cytosolic protons in Lb. brevis upon acid challenge. Since the function of GadC is activated under acidic conditions (Ma et al., 2012), GAD system in Lb. brevis may be activated by acid challenge.

To understand the association between GAD system for endogenous acid (lactic acid) resistance and GABA production capacity of Lb. brevis, a 24-h course of cultivation experiment on Lb. brevis 145 in Lactobacilli MRS broth supplemented with four amino acid-dependent AR's substrates (glutamate, arginine, agmatine and tyrosine) was carried out. Firstly, it was observed that the AR's substrates were able to increase the intracellular pH of Lb. brevis cells by about 0.5 compared to that in Lb. brevis without the AR's substrates supplementation (Figures 3A–C). In addition, results indicate that Lb. brevis maintained its intracellular pH level in the near-neutral range (pH 5.5–6.5) or close to neutral range (pH 6.5–7.0) when extracellular pH decreased spontaneously by secreting protons during its lactic acid production (Figures 3B,C). Moreover, gene transcription of gadB and aguB increased by at least 2 fold whereas gadA mRNA was highly up-regulated by about 55 fold at the point of 18 h (stationary phase) when compared to that at 6 h (logarithmic phase) (Figure 3D). In combination with the higher rate of GABA production at 18 h than that at 6 h (Figure 3G), it appears that GAD system was more active in stationary phase than that in lag and logarithmic phases. In general, changes in the end metabolites of four amino acid-dependent AR systems also demonstrated that utilization of the AR's substrates varied with phases. Based on the metabolic data in Figures 3E–H and gene expression data in Figure 3D, TDC and ADI pathways were active in early cultivation (Figures 3E,F) but were down-regulated in late cultivation (Figure 3D), while GAD and AgDI pathways were more active in stationary phase (Figures 3G,H). Although a gene encoding lysine decarboxylase was found in Lb. brevis 145, the lysine content little changed suggesting an inactive state of this AR system (Figure S2).

The acid resistance and challenge assays may not be able to indicate actual contributions of AgDI and TDC pathways for resisting acids (Figures 2B,C); however, changes in the content of metabolites during different cultivation stages evidenced that all the four amino acid-dependent AR systems contributed to intracellular pH homeostasis (Figure 3C). Since GABA production from Lb. brevis was increased in the stationary and late stages, its GAD system played an important role in bacterial acid resistance because its lactic acid accumulated largely during late cultivation (Figure 3B).

Two Gads, GadA and GadB, from Lb. brevis and a reference type I GadB from Lb. plantarum were hetero-expressed and purified for enzyme assay (Figure S3). Two mutants of Lb. brevis GadA (wild-type; WT), Lb. brevis GadA +GSHM (extra GSHM to N-terminus) and Lb. brevis GadA Δ5 (5 amino acids deleted from N-terminus), were constructed in this study. As shown in Figure 4A, Lb. brevis GadA (WT) exhibited a narrow activity spectrum toward the change of temperature, but still retained about 50% of the highest activity at 37°C, whereas Lb. brevis GadB and Lb. plantarum GadB were heat-stable enzymes. More interestingly, Lb. brevis GadB and Lb. plantarum GadB retained similar catalytic spectrum in the range of pH 3.0–5.5 while Lb. brevis GadA (WT) exhibited activity toward a weak acidity range (pH 5.5–6.6) (Figure 4B). The in vitro enzyme kinetic assay (Figure 4C and Table 4) demonstrated that catalytic efficiency of Lb. brevis GadA (WT) was largely lower than that of two GadB enzymes. However, these three Gads had a very similar K_m value suggesting a similar level of substrate saturation (Table 4). Moreover, Lb. brevis GadA (WT) exhibited a significant (p < 0.05) activity than its two mutants with modifications to its N-terminus (Figure 4D). Previous gel filtration studies have indicated the tetramer state of Lb. brevis GadA (Hiraga et al., 2008), the monomer state of Lb. brevis GadB (Yu et al., 2012; Shi et al., 2014) while Lb. plantarum GadB function as dimers (Shin et al., 2014). This suggests that N-terminus of Lb. brevis GadA may be critical to the tetramer formation via incorporation to its N-terminus. The phylogenetic analysis of representative Gads from LAB and Bifidobacterium highlighted that Lb. brevis GadA was a type III group of Gads which differed largely from Lb. brevis GadB (sub-group 2) and Lb. plantarum GadB (sub-group 1) which were classified in type I Gads (Figure 4E). Overall, Lb. brevis, as a cell factory, exhibited GABA biosynthesizing capacity in a board range of acidity, especially in near-neutral pH range (pH 5.5–6.5).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.