DNA manipulations were performed according to the method of Sambrook, etc. (Sambrook et al., 2006). E.coli DH5α was used for DNA manipulation, and E.coli BL21 (DE3) was used as a host strain for gene expression. Escherichia coli BL21 (DE3) genomic DNA was used as the source for adiA gene, which encodes protein ADCes (GenBank accession: CAQ34466). To express C-terminal His-tagged fusion ADCes, the coding region of adiA was amplified by PCR using forward primer 5′- TAT​AATCAT​ATGATG​AAA​GTA​TTA​ATT​GTT​GAA -3′ and reverse primer 5′- ATA​TACTC​GAGCGC​TTT​CAC​GCA​CAT -3’. The PCR product was digested with restriction enzymes (NdeI/XhoI) and inserted into the corresponding restriction enzyme sites in pET24ma vector (Lee et al., 2010) as reported previously (Hong, 2017).

ADCes PDB structure (2VYC) was analyzed using DbD v2.12, which predicts the appropriate pair of residues capable of disulfide bonding (Craig and Dombkowski, 2013). Predicted pairs were identified using Pymol, and mutation was performed by only selecting disulfide residue pairs in the interfacial region. The primer sequences for the three disulfide-linked mutants W16C/D42C, W16C/D43C, and W16C/D42C/D43C are shown in Supplementary Table S1.

To screen for positive mutants in selected residues of ADCes, NNK codons were used for site-saturated mutagenesis (Hong et al., 2018). The primers corresponding to Supplementary Table S1 were used for PCR to perform alanine scanning and site-saturated mutation. The gene libraries were treated with DpnI for 1 h at 37°C, and 10 μl of the library plasmids were transformed into E. coli BL21 (DE3). To screen mutants from a single residue, ∼200 colonies were examined to ensure a mutation coverage of 95% or greater. The mutant libraries were cultured in 96 deep-well plates for 20 h, and after centrifuging the cultured cells, the supernatant was discarded, and the relative activity of mutants was measured.

According to the previously reported method, the expression and purification of ADCes variants were carried out (Hong, 2017). IPTG (0.2 mM) was added to the culture when the OD₆₀₀ reached 0.8, and the cells were harvested after 18 h induction at 18°C. His₆-tagged fusion proteins were purified at 4°C with Ni-NTA agarose resin from Qiagen (Hilden, Germany). Elution containing the purified His₆-tagged ADCes variants was concentrated by ultrafiltration using Centriplus YM-30 (Millipore, Bedford, MA) with a molecular weight cut-off of 100 kDa.

The protein melting temperature (T _m ) of the variants were compared by SYPRO Orange excitation/emission wavelength profile (Niesen et al., 2007). 1000X SYPRO Orange was dissolved in Tris-HCl (50 mM, pH 8.0) buffer, and 20 μl solution was added to the 0.5 μg purified enzyme. Light cycler 480 II System (Roche Applied Science) was used for the detection of fluorescence changes at specific time points. By the excitation/emission curve, Δ T _m was calculated (Hong, 2017).

The thermal activity of ADC was evaluated by measuring the enzyme activity at various temperatures. The purified enzyme variants were incubated for 1 h at 33 to 65°C, and then cooled on ice for 10 min. The residual enzyme activity was then measured.

To confirm thermal stability under acidic or alkaline conditions, the purified enzyme was placed in 50 mM citrate buffer, pH 5.6, or 50 mM sodium phosphate buffer, pH 8.0, respectively. The samples were then placed in 60°C water bath. Residual activity was calculated based on the ADC activities of non-incubated enzymes. All reactions were done in triplicate.

The relative reaction rates of the mutant libraries were quantitatively measured by a colorimetric method using cresol red (pKa = 8.3) (Hong, 2017). The decarboxylase reaction was measured using 5 mM pH8.0 sodium phosphate buffer, 4 mM L-Arg, 0.2 mM PLP and 0.2 mM indicator. For high-throughput library screening, the same method was applied, and the cultured mutants were examined in 96-well plates. The increase in absorbance at 574 nm was recorded using a UV spectrophotometer (BMG LABTECH, Germany), and the activity was determined via the ΔOD on the basis of the solution containing all substances other than the enzyme.

To determine the k _cat and K_M values, 200 μl reaction was performed in 96-well plates using purified enzymes. The reactivity was confirmed at 45°C by increasing Arg concentrations from 0.5 to 20 mM in pH 8.0 potassium phosphate buffer. k _cat and K_M were calculated with the nonlinear regression analysis of the Michaelis-Menten equation by GraphPad Prism 7 (https://www.graphpad.com/scientific-software/prism/) software.

To determine enzyme activity at various pH, a reaction solution was obtained using the following 60 mM buffers: sodium citrate (pH 5.0–6.0); sodium phosphate (pH 7.0–8.0); and Tris-HCl (pH 8.8). 30 mM L-Arg, 0.1 mM PLP, and 7.5 μg of purified enzyme were added to the reaction mixture. Enzyme activity was measured using HPLC.

To compare agmatine production over time, 150 μg of the purified enzyme was added to the 500 μl reaction mixture of 100 mM Arg and 0.2 mM PLP. Reactions were carried out at 45°C and quantitated by HPLC after certain time intervals. All the reactions were done in triplicate.

Quantitative analysis of Arg and agmatine was performed using Younglin HPLC system with IonoSpher 5 C, G100x3 Column (Agilent, United States). The flow rate of eluant consisting of 0.6 g L⁻¹ citrate, 4 g L⁻¹ tartaric acid, 1.4 g L⁻¹ ethylene diamine, 5% methanol in 95% water was 0.9 ml min⁻¹ at 40°C. Refractive index (RI) detection was used for the detection of substrate and product. Retention times for Arg and agmatine were 1.5 and 4.1 min, respectively.

The ADC mutants were modeled by introducing the mutant residues into the reported wild-type structure (PDB 2VYC) using SWISS-MODEL (Bordoli et al., 2009). The template and W16C/D43C/I258A ADC mutant had three amino acid differences, resulting in a sequence identity of nearly 100%. The docking model of substrate Arg into the ADC active site was generated around 5 Å of the PLP site using AutoDock Vina 1.1.2 (Trott and Olson, 2010).

ADCes have been reported to be activated depending on pH due to its conformation changes (Andrell et al., 2009). To confirm the pH dependent activity of ADCes, the specific activity was measured in a pH range between 5.0 and 8.0. The highest specific activity of the ADCes was observed at pH 5.0, and the activity decreased sharply when the pH increased. Little activity remaining at pH above 7.0, which is in good agreement with the reported activity profile (Kanjee et al., 2011) (Supplementary Figure S1).

To rationally create a focused, smart library for selecting ADCes mutants with enhanced structural stability at alkaline pH, insertion of disulfide bonds at the decameric interface was attempted. A web-based tool, “Disulfide by Design (DbD) v2.12”, was used for in silico design of new disulfide bridges using the 3D structure of ADCes (PDB 2VYC). The computational algorithm predicted the distance between the two C_βs of a Cys pair when a particular target residue is mutated to cysteine (Cys) and verified possible disulfide bonds (Craig and Dombkowski, 2013). Out of 83 disulfide pair candidates, five pairs were situated on the decamer interface region. Among the five putative disulfide bond pairs, we narrowed down the putative mutation targets to the residues involved in the pentameric inner ring formation. Thus, the N-terminal wing domain interface in helix α′ 1 and helix α′ 2 regions were selected and counted as ‘hot-spot’ (Andrell et al., 2009) (Figure 2A). From the two putative disulfide pair candidates in the hot-spot regions, W16C was predicted to form a disulfide bond with D42C or D43C, respectively. W16 is located in helix α′ 1, whereas D42 and D43 residues are in helix α′ 2 of another homodimer subunit (Figure 2B). To examine the activity and stability changes of ADCes through Cys substitution, two mutants were generated: W16C/D42C and W16C/D43C.

The enzyme activities of the two resulting mutants were measured at pH 8.0 using their cell extracts. As given in Supplementary Figure S2, the relative activities of the variants were measured using ADCes activity as control. Among the two mutants examined, only W16C/D43C mutation was active towards the substrate. When protein expression levels were compared, the protein expression level of the W16C/D42C mutant was similar to that of the wild-type, but this mutant lost activity (Supplementary Figure S3); thus, only W16C/D43C mutant was chosen for further enzyme characterization.

To better understand and precisely compare the W16C/D43C mutant activity, enzyme reactions were conducted in different alkaline pH buffers using the purified enzymes. At pH 8.0, the specific activity of ADCes and W16C/D43C mutant were 3.01 ± 0.10 and 6.28 ± 0.12 μmol/min/mg, respectively, whereas at pH 8.8, the specific activity of ADCes and W16C/D43C mutant were 2.73 ± 0.11 and 5.05 ± 0.6 μmol/min/mg, respectively. One thing to note is that W16C/D43C mutant showed almost two-fold enhanced specific activity (i.e., 2.1 and 1.9 folds at pH 8.0 and 8.8, respectively) at alkaline conditions compared to the wild-type (Figure 3). The purified enzyme reaction at pH 8.0 showed higher activity enhancement than the cell extract (Supplementary Figure S2).

The thermostability of wild-type and W16C/D43C was compared by incubating the purified enzymes at 60°C, as explained in the Material and Methods section. To examine the thermostability at different pH, the samples were incubated at pH 5.6 (Figure 4A) and pH 8.0 (Figure 4B), respectively. The wild-type activity dropped to almost zero within 10 min at both pH 5.6 and 8.0 conditions; However, W16C/D43C maintained activity compared to wild-type at both pHs. When the enzymes were heated at pH 5.6, W16C/D43C was exceptionally stable, and nearly 97% of the initial activity remained after 10 min incubation. The dramatic thermostability enhancement was calculated by the half-life (t _1/2-pH5.6 ) as shown in Supplementary Figure S4a. The t _1/2-pH5.6 of W16C/D43C mutant was 2100 min, 1050-fold as that of ADCes at 60°C. In the case of pH 8.0 (Supplementary Figure S4b), the t _1/2-pH8.0 of W16C/D43C mutant was 560 min, almost 280-fold as that of the wild-type indicating that the dramatic thermostability enhancement of W16C/D43C mutant was maintained under both acidic and alkaline conditions.

T _50, the temperature at which any enzyme loses 50% of its activity after 1 h incubation, was also measured (Figure 4C), and the W16C/D43C mutant displayed 12.7 ± 3.2°C higher T ₅₀ than ADCes. All of the examined thermostability values were summarized in Table 1, verifying that the disulfide bond mutation substantially enhanced the ADC’s stability.

In order to further increase the activity of W16C/D43C disulfide bond mutant, single SSMs at the active site residues of ADCes were subsequently performed. Candidate residues for mutagenesis were selected based on ADCes -Arg docking simulations and the previously reported agmatine aldimine structure at ADCes active site (Andrell et al., 2009). Among the residues directly contacting the cofactor PLP or the substrate Arg, the residues interacting with the phosphate group of PLP and carboxylic group of Arg were carefully examined. Among 13 residues, seven residues near the phosphate group of PLP and six residues near the carboxylic group of Arg were chosen (Figure 5A). To minimize the experiments for SSM, alanine scanning was performed in these residues before SSM. When the corresponding amino acids of the selected residues were substituted to Ala, only the two mutants, C254A and I258A, maintained more than 60% of their original activities towards Arg (Figures 5B,C). The activity loss of most of Ala substituted mutants was somewhat expected because the examined residues were highly conserved based on the conservation analysis using HotSpot Wizard 2.0 (Bendl et al., 2016) (Figure 5A).

C254 and I258 were subjected to SSM using the disulfide bond mutant (W16C/D43C) template. Degenerate NNK codons were used for the library construction, and at least 172 mutant clones were examined to ensure a mutation coverage of 95% or more for a single SSM (Nov 2013). 200 colonies were selected from each residue library from the agar plates, and, in total, 400 clone activities were compared through whole-cell cultures in 96 well plates at pH 8.0. Among the mutants, 12 colonies with higher activity than the template W16C/D43C were identified. However, only one mutant displayed more than 1.4-fold increased activity, and its mutation site was identified as W16C/D43C/I258A. Interestingly, the best mutant from the SSM library at position I258 was the same mutant created by alanine scanning.

To investigate the mutational effect on the kinetic values, the kinetic parameters of W16C/D43C, I258A, and the final mutant W16C/D43C/I258A of ADC were determined after purification at pH 8.0, as shown in Table 2. Compared with ADCes, all the three variants displayed an increase in k _cat and a decrease in K _M value, resulting in enhanced catalytic efficiency (k _cat /K _M). In particular, the final mutant W16C/D43C/I258A showed enhanced k _cat value for Arg (2.1 fold increase) and showed decreased K_M value (2.0 fold decrease), resulting in a significant k _cat /K _M value (4.4 fold increase) compared to that of the wild-type.

In order to confirm the influence of the additional active site mutation (W16C/D43C/I258A) on the disulfide mutant (W16C/D43C), the melting temperatures (Tm) of ADCes variants were measured using differential scanning fluorometry (DSF) (Supplementary Figure S5). The disulfide bond W16C/D43C mutant exhibited 18°C higher Tm than that of ADCes. However, I258A mutation alone harmed the stability of ADCes and displayed 2.4°C lower Tm value. Nonetheless, when the single mutation (I258A) was combined with the disulfide mutation (W16C/D43C), the melting temperature of the final mutant (W16C/D43C/I258A) was 81.7 ± 0.01°C, which was 16.0°C higher than the wild-type. The final mutant W16C/D43C/I258A provided dramatic structural stability against the single I258A mutation.

Finally, the time course of agmatine production was monitored by HPLC with RI detector using the purified enzyme variants (150 μg) together with 100 mM Arg and 0.2 mM PLP. The starting pH of the batch reaction for agmatine production using Arg monohydrochloride as substrate was ca. 6.7. As shown in Figure 6, the examined three mutants were identified to be superior to ADCes in converting Arg into agmatine throughout the whole reaction. I258A and W16C/D43C/I258A mutants exhibited an initial conversion rate of up to 1.4- and 1.5-fold, respectively, within 20 min compared to the wild type. In terms of the final yield, W16C/D43C and W16C/D43C/I258A mutants with the disulfide bond mutation showed 1.8- and 2-fold increases of agmatine yield compared to the wild-type within 5 h, respectively. As a result, due to the improved stability and the activity, W16C/D43C/I258A mutant showed the highest agmatine yield, which is ca. 2-fold improvement (>89% yield within 5 h) in total compared to that of the wild-type (>43% yield within 5 h).