Evolving ω-amine transaminase AtATA guided by substrate-enzyme binding free energy for enhancing activity and stability against non-natural substrates

ABSTRACT

In the field of chiral amine synthesis, ω-amine transaminase (ω-ATA) is one of the most established enzymes capable of asymmetric amination under optimal conditions. However, the applicability of ω-ATA toward more non-natural complex molecules remains limited due to its low transamination activity, thermostability, and narrow substrate scope. Here, by employing a combined approach of computational virtual screening strategy and combinatorial active-site saturation test/iterative saturation mutagenesis strategy, we have constructed the best variant M14C3-V5 (M14C3-V62A-V116S-E117I-L118I-V147F) with improved ω-ATA from Aspergillus terreus (AtATA) activity and thermostability toward non-natural substrate 1-acetylnaphthalene, which is the ketone precursor for producing the intermediate (R)-(+)-1-(1-naphthyl)ethylamine [(R)-NEA] of cinacalcet hydrochloride, showing activity enhancement of up to 3.4-fold compared to parent enzyme M14C3 (AtATA-F115L-M150C-H210N-M280C-V149A-L182F-L187F). The computational tools YASARA, Discovery Studio, Amber, and FoldX were applied for predicting mutation hotspots based on substrate-enzyme binding free energies and to show the possible mechanism with features related to AtATA structure, catalytic activity, and stability in silico analyses. M14C3-V5 achieved 71.8% conversion toward 50 mM 1-acetylnaphthalene in a 50 mL preparative-scale reaction for preparing (R)-NEA. Moreover, M14C3-V5 expanded the substrate scope toward aromatic ketone compounds. The generated virtual screening strategy based on the changes in binding free energies has successfully predicted the AtATA activity toward 1-acetylnaphthalene and related substrates. Together with experimental data, these approaches can serve as a gateway to explore desirable performances, expand enzyme-substrate scope, and accelerate biocatalysis.

IMPORTANCE

Chiral amine is a crucial compound with many valuable applications. Their asymmetric synthesis employing ω-amine transaminases (ω-ATAs) is considered an attractive method. However, most ω-ATAs exhibit low activity and stability toward various non-natural substrates, which limits their industrial application. In this work, protein engineering strategy and computer-aided design are performed to evolve the activity and stability of ω-ATA from Aspergillus terreus toward non-natural substrates. After five rounds of mutations, the best variant, M14C3-V5, is obtained, showing better catalytic efficiency toward 1-acetylnaphthalene and higher thermostability than the original enzyme, M14C3. The robust combinational variant acquired displayed significant application value for pushing the asymmetric synthesis of aromatic chiral amines to a higher level.

INTRODUCTION

Asymmetric amination is an essential chemical transformation due to the convenience of chiral amine functionalities as drug intermediates (1–4). Chiral amines are also found in many products across the fine chemical and agrochemical industries (5–7). However, chemical asymmetric amination reactions typically involve elevated temperatures, high pressures, protecting groups, and toxic byproducts (2, 8, 9). In contrast, comparatively environment-friendly asymmetric amination by biocatalytic reactions can proceed with mild reaction conditions and greener benign reagents (10, 11). Biocatalytic reactions’ exquisite selectivity (excellent chemo-, regio-, and enantio-selectivity) also reduces purification requirements (12). One established enzyme capable of asymmetric amination is ω-amine transaminases (ω-ATAs) (13), which can catalyze asymmetric amination of prochiral ketones or aldehydes to chiral amines with strict stereoselectivity and 100% theoretical yield (14, 15). ω-ATAs are pyridoxal 5′-phosphate (PLP)-dependent enzymes with large and small binding pockets (16, 17). Although various ω-ATAs exhibit potential catalytic performance in synthesizing chiral amines, their industrial application is still hampered by inherent drawbacks such as unfavorable activity toward non-natural substrates and insufficient stabilities (18, 19). Therefore, acquiring the engineering robust ω-ATAs that can be applied in industrial applications must be solved urgently.

As a typical ω-ATA, ω-ATA from Aspergillus terreus (AtATA) has been found and identified to have high activity toward the natural substrate pyruvate (20, 21). Simultaneously, AtATA exhibits low activity toward non-natural substrate 1-acetylnaphthalene, which is the prochiral ketone to synthesize the intermediate of cinacalcet hydrochloride for treating secondary hyperthyroidism and hypercalcemia (22, 23). Figure 1 shows that the difference in activity between natural substrate pyruvate and non-natural substrate 1-acetylnaphthalene is due to variations in binding strength and conformation between the two substrates and AtATA. Hence, optimizing the binding conformation by protein engineering will enhance the transamination activity for (R)-NEA and other bulky chiral amine synthesis (24, 25).

Fig 1.

(A) Transamination reaction catalyzed by AtATA toward the native substrate pyruvate. (B) Artificial reaction catalyzed by AtATA converts the non-natural substrate 1-acetylnaphthalene.

Semi-rational design is an efficient method for improving enzyme catalytic activity and stability (26, 27). Constructing a small and smart library is crucial for screening key hotspots and yielding beneficial mutations, such as a combination method of multiple-sequence alignment and structural analysis focusing on the functional pocket (28). The strategy of combinatorial active-site saturation test combined with iterative saturation mutagenesis (CAST/ISM) is developed to enhance the activity and expand the substrate scope of enzymes by designing a focused mutation library around the binding pocket (29). A well-known study on substrate-binding pocket expansion used computational substrate walking-guided strategy to modify ω-ATA from Arthrobacter sp. to enlarge the small binding pocket for accommodating the bulky ketone to produce sitagliptin (30). With the improvement of computational power and theoretical calculation, efficient molecular simulation tools have been rapidly popularized, promoting the progress of protein engineering (31–36). Yeo and co-workers (37) applied computational modeling combined with directed evolution to identify the key residues of galactose oxidase for expanding the substrate specificity toward bulky benzylic and alkyl secondary alcohols. Cui et al. (38) employed binding energy analysis to evolve an aminotransferase for efficient synthesis of valienamine. A detailed theoretical simulation and experimental data make this combination approach feasible as a gateway to exploring desirable enzyme performances.

In our previous work, synthetic shuffling and pocket modification were performed to improve the thermostability and activity toward 1-acetylnaphthalene of AtATA, and seven key residues (F115L, M150C, H210N, M280C, V149A, L182F, and L187F) were selected (22, 39). However, both single and combinational variants have low catalytic activity toward 1-acetylnaphthalene. Seven-site combinational variant M14C3 (AtATA-F115L-M150C-H210N-M280C-V149A-L182F-L187F, GenBank accession no. PP754878) was constructed as the parent enzyme. Furthermore, we presented a combined approach of computational virtual screening strategy and CAST/ISM strategy to evolve the activity of M14C3 toward more non-natural complex molecules, along with higher thermostability. Computational tools such as YASARA, Discovery Studio, Amber, and FoldX have been applied to predict and enable in silico catalytic activity and stability determination. Moreover, our virtual screening strategy has yielded valuable insights that can aid in the exploration of enzyme activity toward non-natural substrates in silico, providing inroads to accelerate enzyme engineering development.

RESULTS AND DISCUSSION

In silico selection for binding adaptation sites

The main challenge in acquiring robust enzymes is the time-consuming, experiment-intensive enzyme engineering process (29). With the rapid development in computational power and bioinformatics, the construction of in silico models via structural models and molecular dynamics simulation could potentially predict and guide the activity modification of AtATA toward various non-natural substrates. Herein, we applied four complementary approaches to model the activity of AtATA-(1-acetylnaphthalene) pairs. The first approach involved molecular docking of individual substrate to model AtATA parent enzyme M14C3 via YASARA (version 16.4.6). According to the reaction mechanism (Fig. S1), AtATA-catalyzed asymmetric ammoniation follows two steps. In the first step, PLP is covalently bound as an imine structure to the ε-group of the lysine (active site). Amine groups from amino donor are transferred to PLP while the PLP is converted to pyridoxamine-5′-phosphate (PMP). In the second step, amine groups in PMP nucleophilically attack the amino acceptor (1-acetylnaphthalene) from the si-face to generate an (R)-chiral amine (17). By identifying the nucleophilic attack direction of the amino groups in PMP to 1-acetylnaphthalene and analyzing the binding free energy between M14C3 and 1-acetylnaphthalene, the optimal docking model was obtained with a binding energy of −5.96 kcal/mol from 10 docking models after the running the docking program 10 times (Fig. 2A and B). The second approach involved the construction of a mutation library and screening potential mutation sites. Compared with random mutagenesis, the strategy of CAST/ISM would significantly decrease the screening effort via constructing small and smart libraries near the substrate-enzyme binding pocket (29). Generally, two amino acid residues would form an interaction if the distance between their C_β atoms (C_α for glycine) is <8 Å (40). Consequently, 82 residues within a contact distance of <8 Å around the 1-acetylnaphthalene-M14C3 in the substrate-enzyme binding pocket of AtATA were identified and selected for further analysis. For such a large number of residues, random or site-directed saturation mutations via experiment would need time and effort. Thus, in silico theoretical evaluation was performed to identify these 82 residues that potentially improve 1-acetylnaphthalene-enzyme binding via a virtual saturation mutation strategy employing the Calculate Mutation Energy (Binding) tool of Discovery Studio 2019 Client based on the mutation binding free energy of 1-acetylnaphthalene-M14C3. Each residue among the 82 residues was virtually mutated to all 20 amino acids (contained the parent enzyme), and the corresponding mutation binding free energies (ΔG) and average mutation energies were calculated. ΔG_bind-M14C3 and ΔG_bind-var represented the binding free energy of M14C3 and variant, respectively. ΔΔG represented the difference in binding free energy between variant and M14C3 (ΔΔG = ΔG_bind-var – ΔG_bind-M14C3). Mutation with positive ΔG_bind-var would decrease the affinity and potentially destabilize the binding conformation of substrate to enzyme. On the contrary, mutation with predicted negative ΔG_bind-var would increase the affinity and stabilize the enzyme-substrate complex’s favorable binding conformation (38). As shown in Fig. 2C, the average mutation energies among 6 of 82 residues were positive with >0.5 kcal/mol, and 10 of 82 residues were negative with <−0.5 kcal/mol. Based on this, 10 of 82 residues with <−0.5 kcal/mol average mutation energies were more likely to obtain ideal variants, which exhibited increased affinities toward 1-acetylnaphthalene. To verify the accuracy of the virtual screening strategy using Calculate Mutation Energy (Binding) tool, we further analyzed the contributions of residues on enzyme-substrate affinity via molecular mechanics-generalized born surface area (MM-GBSA, the third approach) using Amber 18. MM-GBSA is a powerful method for calculating the binding energy and analyzing the contribution of each residue to the binding energy of the enzyme and substrate (31). As shown in Fig. 2D, binding free energy decomposition on M14C3 suggested that 10 residues (V62, R79, V116, E117, L118, V147, R169, S215, V238, and C273) displayed negative contributions to 1-acetylnaphthalene binding. Combining 10 residues with <−0.5 kcal/mol average mutation energies via the Calculate Mutation Energy (Binding) tool in Fig. 2C and 10 residues with negative contributions to 1-acetylnaphthalene binding in Fig. 2D, the intersection of potential sites by two approaches is shown in Fig. 2E. The fourth approach involved virtual site-directed saturation mutation using FoldX 4.0 on five sites (V62, V116, E117, L118, and V147), which were identified using both Calculate Mutation Energy and MM-GBSA tools. According to the ΔΔG landscape, 11 single-site variants (yellow star marked in Fig. 2F) with negative ΔΔG (range from –0.625 to –0.011 kcal/mol) were selected for experimental verification of their effectiveness.

Fig 2.

(A) Cartoon drawing of docking result with PMP and substrate 1-acetylnaphthalene modeled via YASARA. (B) Close-up view of the catalytic site, PMP and 1-acetylnaphthalene geometry. (C) Value distribution of mutation binding energies and the average mutation energy (represented by dots) of the candidate residues in M14C3. Residues with >0.5 and <-0.5 kcal/mol are labeled in red. Values were determined by virtual saturation mutagenesis of the 82 contacting residues within 8 Å (a typical protein inter-residue contact distance) of the unfavorable binding conformation of 1-acetylnaphthalene-PMP in the binding pocket of M14C3 using Discovery Studio 2019 Client. (D) The enzyme-substrate binding free energy decomposition per residue on M14C3 calculated by Amber 18. (E) Venn diagram of the hot-spot residues’ intersection obtained by two virtual screening methods of Discovery Studio 2019 Client and Amber 18. (F) The binding free energy (unit: kcal/mol) change of virtual saturation mutagenesis among five potential residues calculated by FoldX 4.0.

Characterization of the potential variants acquired via virtual screening

To verify the effectiveness of 11 potential variants selected by binding free energy-guided virtual screening, a series of single-site mutations were constructed based on M14C3 as a template in the first step. As shown in Fig. 3A, 8 of 11 single-site variants exhibited 2.2%–68.8% higher specific activity toward substrate 1-acetylnaphthalene than M14C3, and 3 of 11 variants displayed decreased specific activity. Five variants of M14C3-V62A, M14C3-V116S, M14C3-E117I, M14C3-L118I, and M14C3-V147F displayed 68.8%, 20.8%, 40.4%, 52.8%, and 30.9% higher specific activity than M14C3, respectively. Next, combinational mutagenesis was performed using five positive single variants to acquire 10 double variants ($C_5^2$), 10 triple variants ($C_5^3$), 5 quadruple variants ($C_5^4$), and 1 five-site variant ($C_5^5$). Then, the specific activities and half-life (t_1/2) values at 45°C of combinational variants toward 1-acetylnaphthalene were detected (Fig. 3B). Compared with M14C3, 7 of 26 combinational variants showed decreased t_1/2 value, one variant showed decreased specific activity, and 19 variants enhanced both higher activity and higher thermostability. Finally, five-site variant M14C3-V62A-V116S-E117I-L118I-V147F (M14C3-V5, GenBank accession no. PP754879) acquired the highest activity and thermostability, reaching 136.3 U/g specific activity and 650.7 min t_1/2, respectively. The strategy of virtual screening based on substrate-enzyme binding free energy-guided evolution of M14C3 for enhancing activity and stability toward the non-natural substrate 1-acetylnaphthalene was successfully applied in this work. With the evolution from M14C3 to M14C3-V5, the enzyme-specific activity toward the natural substrate pyruvate was decreased, and the specific activity toward the non-natural substrate 1-acetylnaphthalene was increased (Fig. S3). Apparent kinetic constants of M14C3 and the best variant M14C3-V5 were determined using 1-acetylnaphthalene and pyruvate as substrates. M14C3-V5 showed improved catalytic efficiency (k_cat/K_m), acquiring a 3.7-fold enhancement compared with M14C3. Instead, M14C3-V5 showed an increase in K_m and a decrease in k_cat/K_m toward pyruvate compared with M14C3. The shift of substrate specificity between two substrates was defined as specificity constant [the ratio between the catalytic efficiency of enzymes on 1-acetylnaphthalene and pyruvate, ${\displaystyle \frac{{\left(k_{\mathrm{c}\mathrm{a}\mathrm{t}}/K_m\right)}_{1-\mathrm{A}\mathrm{c}\mathrm{e}}}{{\left(k_{\mathrm{c}\mathrm{a}\mathrm{t}}/K_m\right)}_{\text{Pyr }}}}$]. M14C3-V5 displayed an 8.9-fold increase in specificity toward 1-acetylnaphthalene than M14C3 (Table 1). The thermostability data of melting temperature (T_m), t_1/2, and half-inactivation temperature (T₅₀¹⁰) showed that M14C3-V5 acquired a significant thermostability improvement than M14C3 (Table 2).

Fig 3.

The specific activity and thermostability of different variants. (A) The specific activity and e.e._p of potential mutation sites selected by virtual screening strategy. (B) The specific activity and t_1/2 at 45°C of combination variants. The reaction conditions: 5 mL scale system containing 1 mg·mL^–1 purified enzyme, 20 mmol·L^–1 1-(R)-PEA, 20 mmol·L^–1 1-acetylnaphthalene or pyruvate, 0.1 mmol·L^–1 PLP, and 20% (vol/vol) DMSO in 50 mmol·L^–1, pH 8.0 potassium phosphate buffer. The reaction mixtures were shaken at 30°C for 5 min. Data obtained from three biological replicates are shown as the mean ± standard deviation. “*” indicates that there is a significant difference between the two studied groups (*P < 0.05; **P < 0.01; and ***P < 0.001).

TABLE 1.

Kinetic parameters of M14C3 and M14C3-V5 toward 1-acetylnaphthalene and pyruvate

Enzyme	1-Acetylnaphthalene			Pyruvate			${\displaystyle \frac{{\left(k_{\mathrm{c}\mathrm{a}\mathrm{t}}/K_m\right)}_{1-\mathrm{A}\mathrm{c}\mathrm{e}}}{{\left(k_{\mathrm{c}\mathrm{a}\mathrm{t}}/K_m\right)}_{\text{Pyr }}}}$
	Km (mM)	kcat (s−1)	kcat/Km (s−1·mM−1)	Km (mM)	kcat (s−1)	kcat/Km (s−1·mM−1)
M14C3	1.6 ± 0.2	0.2 ± 0.02	0.13 ± 0.01	1.9 ± 0.1	4.4 ± 0.2	2.3 ± 0.1	0.057 ± 0.004
M14C3-V5	1.5 ± 0.1	0.9 ± 0.03	0.61 ± 0.02	2.1 ± 0.1	2.6 ± 0.1	1.2 ± 0.1	0.510 ± 0.013

TABLE 2.

Thermostabilities of M14C3 and M14C3-V5

Enzyme	Tm (°C)	t1/2 at 40°C	t1/2 at 45°C	T5010 (oC)
M14C3	56.3 ± 1.3	282.2 ± 1.5	75.5 ± 4.3	52.8 ± 1.6
M14C3-V5	62.8 ± 2.6	1,256.3 ± 15.7	650.7 ± 5.1	57.9 ± 2.2

The insight analysis of the best variant with enhanced activity and stability

In order to investigate the relative affinity of M14C3 and M14C3-V5 to the non-natural substrate 1-acetylnaphthalene, we evaluated the binding free energy by the method of MM-GBSA. The results are listed in Table S2. After mutation, the binding free energy between 1-acetylnaphthalene and M14C3-V5 decreased compared with M14C3, in which the ΔG_bind was decreased from −15.94 kcal/mol of M14C3 to −19.43 kcal/mol of M14C3-V5. We performed an energy decomposition of the binding free energy to elucidate further the affinity changes on M14C3 and M14C3-V5 with 1-acetylnaphthalene. As shown in Fig. 4A and B about key residue decomposition, the contribution of Ala62, Ser116, Ile117, Ile118, and Phe147 in M14C3-V5 to the substrate 1-acetylnaphthalene binding was significantly decreased after mutation from Val62, Val116, Glu117, Leu118, and Val147 in M14C3, indicating that the intramolecular interactions further stabilized the stability between the substrate 1-acetylnaphthalene or the surrounding residues in the middle binding region and five mutation site residues.

Fig 4.

The free energy decomposition per residue in M14C3 (A) and M14C3-V5 (B), respectively. Gibbs free energy (unit: kJ/mol) landscape of M14C3 (C) and M14C3-V5 (D) at 303 K. Hydrogen bonds in the active center loop (residues: 164–183) amino acids of M14C3 (E) and M14C3-V5 (F) in the molecular dynamic simulation trajectory. The ordinate in panel E represents 1, P171@O-A174@N-H; 2, N181@O-Q183@N-H; 3, P171@O-I175@N-H; 4, D176@O-V179@N-H; 5, V179@O-N181@N-H; 6, P177@O-V179@N-H; 7, R165@O-V167@N-H; 8, A164@O-V167@N-H; and 9, P172@O-A174@N-H. The ordinate in panel F represents 1, A174@O-D176@N-H; 2, P171@O-A174@N-H; 3, D176@O-V179@N-H; 4, P171@O-I175@N-H; 5, P172@O-A174@N-H; 6, N181@O-Q183@N-H; 7, R165@O-V167@N-H; 8, A164@O-V167@N-H; 9, V179@O-N181@N-H; 10, P177@O-V179@N-H; 11, D176@O-T178@N-H; and 12, T178@O-K180@N-H. The colored lines on the left of (E) and (F) represent the same hydrogen bond formed between key residues in M14C3 and M14C3-V5.

To investigate the possible mechanism for the increase in enzyme activity observed for M14C3-V5 toward 1-acetylnaphthalene, 100-ns MD simulations of M14C3 and the M14C3-V5 were performed at 303 K. Figure S4B shows the variations of the root-mean-square deviation (RMSD), which indicated that M14C3 and M14C3-V5 complexes were dynamically equilibrated after 40 and 35 ns, respectively. By analyzing the hydrophobicity of the substrate-binding pockets, we found that the hydrophobicity of the M14C3-V5 increased compared to M14C3, which was more conducive to the binding of non-natural organic substrates (Fig. S5A and B). CAVER Analyst 2.0 was performed to calculate the volume of substrate-binding pockets for M14C3 and M14C3-V5. Compared to the volume of M14C3 (1,039.1 ų), the volume of M14C3-V5 (1,606.7 ų) has increased by 567.6 ų (Fig. S5C and D). Furthermore, the root-mean-square fluctuation (RMSF) value of the active center loop (loop 164–185) in M14C3-V5 was increased after introducing V62A, V147F, V116S, E117I, and L118I, indicating that these five mutation sites were conducive to increasing the flexibility of the active center (Fig. S4C). The active center loop (loop 164–185) hydrogen bonds of M14C3 and M14C3-V5 in the simulated trajectories were counted (Fig. 4E and F). Compared with M14C3, the hydrogen bonds in the active center loop of M14C3-V5 were decreased in the 100-ns simulated trajectories enhancing the flexibility of loop 164–185. Considering these data collectively, increasing the flexibility of the active center would accelerate the entry of 1-acetylnaphthalene, thus enhancing the enzyme activity. The increase in volume after mutation might be related to the enhanced flexibility of the active center loop 164–185, leading to the loosening of the substrate-binding pocket. The enlarged substrate-binding pocket volume would facilitate the entry of large-size substrates. In addition, the nucleophilic attack distance between carbonyl carbon in 1-acetylnaphthalene and -NH₂ in PMP was shortened in M14C3-V5 (Fig. S4D), which was more conducive to the occurrence of transamination reaction.

To gain insight into the mechanism by which M14C3-V5 contributed to the thermal stability of AtATA, the Gibbs free energy landscape corresponding to the conformation was calculated based on the radius of gyration (Rg, Fig. S4A) and RMSD. The Rg indicates the global compactness of the enzyme conformation. The smaller the value of Rg, the more stable the enzyme structure is (41). The Rg values of M14C3-V5 were significantly lower than that of M14C3, indicating that the structure of M14C3-V5 was more compact than that of M14C3. At 303 K, the blue distribution of M14C3-V5 broadened, indicating an increase in conformational stability (Fig. 4D). In contrast, the overall intensity of the red region in M14C3 increased, while the high free energy region expanded (Fig. 4C). The minimum values of Rg and RMSD on M14C3-V5 were lower than those on M14C3, indicating a lower Gibbs free energy existed in M14C3-V5. This phenomenon implied that the introduced mutation sites enhanced the densification and robustness of AtATA, consequently contributing to heightened thermal stability.

At the same time, the secondary structure changes of M14C3 and M14C3-V5 were analyzed (Fig. 5A). The secondary structure was divided into three types, including helix (α-helix, 3–10 helix, and π-helix), sheet (β-sheet and anti-parallel β-sheet), and other structures (β-bend and β-turn). With the evolution from M14C3 to M14C3-V5, there was no significant change in the overall secondary structure motifs of the two enzyme molecules. We found that partially loose β-turns of residues 146–157 and residues 238–253 in M14C3 were converted to more stable α-helixes in M14C3-V5. The detailed probability data of helix, sheet, and coil in residues 146–157 and 238–253 of M14C3 and M14C3-V5 are shown in Table S3. The stable structure of α-helix increased, while the flexible structure of β-turn in the secondary structure decreased (Fig. 5B). Residues 146–157 and residues 238–253 were located in the surface region of AtATA (Fig. 5C). The enhancement of protein surface structure rigidity had a positive effect on enzyme thermostability. This result was also consistent with the average value of RMSF between M14C3 (average RMSF = 0.79 Å) and M14C3-V5 (average RMSF = 0.74 Å).

Fig 5.

(A) The detail secondary structure change diagram of M14C3 and M14C3-V5 during 100-ns MD simulations. (B) The probability of helix, sheet, coil, and no structure in M14C3 and M14C3-V5 during 100-ns MD simulations. (C) The locations of residues 146–157 and residues 238–253 on the surface of AtATA.

Dynamics cross-correlation map and interaction analysis

Dynamics cross-correlation map (DCCM) is a valuable tool that is applied to evaluate the motion of an enzymatic system over the simulation, especially on the C_α atom, which is an indicator of protein rigidity (42). Given the interaction (mainly including H-bond and hydrophobic interaction), network remodeling on the entire protein was caused by a single point or combinatorial mutation. DCCM was performed using the last 20 ns of the MD simulation trajectory of M14C3 and M14C3-V5. As seen from the spaces (excluding white areas) regions in Fig. 6A and B, the dynamics of most regions of M14C3-V5 were more highly correlated than M14C3. Therefore, M14C3-V5 exhibited more cohesive motions than M14C3, either positively or negatively cross-correlated. In M14C3-V5, when V62, V147, E117, L118, and V116 were replaced by alanine, phenylalanine, isoleucine, isoleucine, and serine, respectively, the positive cross-correlation became stronger (Fig. 6C through G). As shown in the images on the right side of Fig. 6C through G, the advent of positively correlated motions between the five mutation sites (V62A, V147F, E117I, L118I, and V116S) and the surrounding residues changes the protein’s intramolecular interactions. Compared with M14C3, the interactions between M14C3-V5 and the substrate 1-acetylnaphthalene became stronger since the extra pi-pi T-shape was formed between Trp184 and 1-acetylnaphthalene. By analyzing the interactions between the five mutation sites and surrounding amino acids, it could be concluded that extra hydrogen bonds and hydrophobic interactions formed or changed after mutation from V to A in site 62, V to S in site 116, E to I in site 117, L to I in site 118, and V to F in site 147. In our opinion, these results showed that the whole M14C3-V5 protein became more cohesive. The mutation at five sites resulted in changes in intramolecular interactions and involved changes in the structure and function of AtATA.

Fig 6.

DCCM analysis of M14C3 (A) and M14C3-V5 (B). Stacked area shows the correlation coefficients (Cij) between residue 62 and all the residues in the protein (C). Stacked area shows Cij between residue 116 and all the residues in the protein (D). Stacked area shows Cij between residue 117 and all the residues in the protein (E). Stacked area shows Cij between residue 118 and all the residues in the protein (F). Stacked area shows Cij between residue 147 and all the residues in the protein (G). Cij values for DCCM are shown in different colors. Cij values from 0 to 1 represent positive correlations, whereas Cij values from −1 to 0 represent negative correlations. The interactions between the substrate 1-acetylnaphthalene (H) or mutation sites (the images on the right side of panels C–G) and the surrounding residues in M14C3 and M14C3-V5. The blue dashed line represents H-bond, and the orange dashed line represents hydrophobic interaction.

Production of (R)-NEA and substrate specificity analysis

A 50 mL scale-up reaction of M14C3 and the best-performing variant M14C3-V5 was conducted under 30°C and 50 mM substrate loading, and the conversion of 1-acetylnaphthalene was assayed. Production of (R)-NEA was monitored for 24 h. The reaction time course curves of both M14C3 and M14C3-V5 showed that the accumulation of (R)-NEA increased in the first 12 h and remained stable over 12–24 h (Fig. 7A). At the end of the reaction, the conversion of 1-acetylnaphthalene for M14C3 and M14C3-V5 was 30.2% and 71.8%, respectively. M14C3 and M14C3-V5 showed no change in stereoselectivity toward 1-acetylnaphthalene, giving >99.5% e.e._p for (R)-NEA. The conversion toward 1-acetylnaphthalene of M14C3-V5 was 2.4-fold as high as that of M14C3. The total turnover number (TTN, the ratio of moles of product to moles of biocatalyst) of M14C3-V5 was 1,400.4 after five rounds of evolution, which was higher than M14C3 (TTN = 508.8). Furthermore, we investigated the effects of increasing reaction temperature and increasing reaction system on enzyme catalysis (Fig. 7B). When the reaction temperature rose to 40°C, the conversion of 20 mM 1-acetylnaphthalene catalyzed using M14C3-V5 and M14C3 was 70.6% and 6.9%, respectively, and the conversion of 50 mM 1-acetylnaphthalene catalyzed using M14C3-V5 and M14C3 was 44.4% and 1.3%, respectively. At 40°C, 200 mL scale-up, the conversion of 20 mM 1-acetylnaphthalene catalyzed using M14C3-V5 and M14C3 was 62.4% and 4.2%, respectively (Fig. 7C). M14C3-V5 exhibited a decrease in catalytic efficiency under 200 mL scale-up and high-temperature conditions, although the substrate loading was decreased. In summary, these results indicated that the enhancement of thermostability of M14C3-V5 promoted the catalysis process.

Fig 7.

(A) The time course of asymmetric synthesis of (R)-NEA catalyzed by M14C3 and M14C3-V5 in 50 mL scale-up, 50 mmol·L^–1 substrate loading at 30°C. (B) The time course of asymmetric synthesis of (R)-NEA catalyzed by M14C3 and M14C3-V5 in 50 mL scale-up, 20 or 50 mmol·L^–1 substrate loading at 40°C. (C) The time course of asymmetric synthesis of (R)-NEA catalyzed by M14C3 and M14C3-V5 in 200 mL scale-up, 20 mmol·L^–1 substrate loading at 40°C. Reaction mixture contained 5.0 g_DCW·L^–1 Escherichia coli expressing M14C3 or M14C3-V5, 20 or 50 mmol·L^–1 1-(R)-PEA, 20% (vol/vol) DMSO, and 0.1 mmol·L^–1 PLP in potassium phosphate buffer (50 mmol·L^–1, pH 8.0). The reaction was shaken at 30°C or 40°C, 500 rpm.

To demonstrate the expanded application of M14C3-V5 toward other non-natural substrates, we performed enzymatic reactions in buffer solutions with 20% (vol/vol) DMSO added as a solubility aid to enable catalytic activity detection of 10 aromatic ketone substrates (Table 3). Here, M14C3-V5 gave significantly higher conversion and TTN and accepted a more comprehensive range of aromatic ketone substrates than M14C3. Remarkably, 8 of the 10 substrates had >75% conversions (up to 100%) and TTN over 587 based on the assayed conditions. All the products obtained via M14C3 and M14C3-V5, except the product of phenylmethanamine in Entry 3, achieved e.e._p > 99.5%. Notably, the expended substrate scope of M14C3-V5 extends its utility to more industrially relevant building blocks, broadening its potential application to access important chiral amine compounds for manufacturing fine chemicals or complex pharmaceuticals.

TABLE 3.

Conversion, e.e._p, and TTN of M14C3 and M14C3-V5 toward different non-natural organic substrates into corresponding products^a

Ketone	Structure	Ketone name	Conversion (%)		TTN		e.e.p (%)
			M14C3	M14C3-V5	M14C3	M14C3-V5
1a		1-(4-methoxyphenyl)ethan-1-one	23.2 ± 0.9	64.7 ± 3.3	180.7 ± 33.5	464.7 ± 69.8	>99.5 R
2a		1-(4-nitrophenyl)ethan-1-one	71.9 ± 2.7	89.2 ± 4.1	560.8 ± 72.1	695.8 ± 20.2	>99.5 R
3a		Benzaldehyde	77.1 ± 3.8	85.8 ± 5.8	601.4 ± 76.3	669.2 ± 32.1	–b
4a		1-(4-fluorophenyl)ethan-1-one	31.2 ± 1.4	87.3 ± 3.9	243.4 ± 10.9	680.9 ± 66.4	>99.5 R
5a		1-(4-chlorophenyl)ethan-1-one	35.5 ± 1.1	88.2 ± 4.4	276.9 ± 21.2	687.9 ± 41.8	>99.5 R
6a		1-(4-bromophenyl)ethan-1-one	48.3 ± 2.1	78.2 ± 6.2	376.7 ± 30.2	609.9 ± 59.7	>99.5 R
7a		1-(p-tolyl)ethan-1-one	30.7 ± 1.9	75.3 ± 3.1	239.5 ± 21.3	587.4 ± 87.2	>99.5 R
8a		1-(4-(trifluoromethyl)phenyl)ethan-1-one	54.4 ± 1.6	80.4 ± 3.5	424.3 ± 33.9	627.1 ± 46.8	>99.5 R
9a		1-(naphthalen-2-yl)ethan-1-one	51.2 ± 2.7	79.6 ± 0.8	399.4 ± 41.5	620.9 ± 41.5	>99.5 R
10a		3,4-dihydronaphthalen-1(2H)-one	8.8 ± 0.6	39.1 ± 1.2	68.6 ± 9.3	305.5 ± 21.2	>99.5 R

^a Reaction conditions: 20 mmol·L^–1 substrate, 20 mmol·L^–1 1-(R)-PEA, 20% (vol/vol) DMSO, 0.1 mmol·L^–1 PLP, and 1.0 g·L⁻¹ purified enzyme in PBS (50 mmol·L^–1, pH 8.0) at 30°C for conversion, TTN, and e.e._p assay. Experiments were conducted in triplicate.

^b –, the product is not a chiral component and the value of e.e._p does not exist.

Conclusion

In this study, a virtual screening strategy based on substrate-enzyme binding free energy in silico combined with the CAST/ISM method in vitro was employed to evolve ω-transaminase AtATA for enhancing activity and stability against a non-natural substrate by enhancing the binding affinity of the large-size non-natural substrate 1-acetylnaphthalene in the substrate-binding pocket of M14C3. As a result, the interaction between the best variant M14C3-V5 and the 1-acetylnaphthalene-PMP intermediate was reconstructed, and both the activity and thermostability toward 1-acetylnaphthalene were all improved compared to that of M14C3. M14C3-V5 displayed higher conversion toward 1-acetylnaphthalene and TTN than M14C3. Besides model substrate 1-acetylnaphthalene, M14C3-V5 gave significantly higher conversion and TTN against the other 10 aromatic ketone substrates. This work shows that AtATA displays high plasticity for producing valuable (R)-NEA and other chiral amine drug intermediates from their readily available ketone precursors. The virtual screening strategy based on substrate-enzyme binding free energy applied here is a promising operating model for overcoming the challenge of getting non-natural large-size organic substrates of interest into desired positions in candidate enzymes with catalytic pockets for artificial reactions.

MATERIALS AND METHODS

Gene, vector, strain, and general materials

The ω-amine transaminase AtATA employed in this study was sourced from Aspergillus terreus. The expression vector pET-28a(+), competent cells of Escherichia coli BL21 (DE3), and the parent variant E. coli BL21(DE3)/pET-28a(+)-AtATA-M14C3 (designated as M14C3, GenBank accession no. PP754878) were stored in our lab. DNA polymerase for gene amplification and mutant library construction was supplied by Takara Bio Inc. (Beijing, China). All polymerase chain reaction (PCR) primers were synthesized by Beijing Tsingke Biotech Co., Ltd. (China) and listed in Table S1. Other chemicals were of reagent grade or analytical grade and obtained from standard commercial sources.

Enzyme evolution and characterization

Mutation

pET-28a(+)-AtATA-M14C3 containing seven mutation sites (F115L, M150C, H210N, M280C, V149A, L182F, and L187F) were constructed using a modified version of the QuikChange PCR method based on pET-28a(+)-AtATA (wild type) as a template. Then, all variants were prepared using a modified version of the QuikChange PCR method using pET-28a(+)-AtATA-M14C3 as a template. For each reaction, 12.5 µL of DNA polymerase mix, 2 µL of template plasmid, and 1 µL of each primer were diluted to 25 µL with double distilled H₂O. The amplification was performed as follows: 95°C, 3 min; 35 cycles: 95°C, 20 s; 55°C, 20 s; 72°C, 1.5 min; and 72°C, 7 min. The remaining nonmethylated purified DNA was transformed into E. coli BL21(DE3) competent cells after digestion by Dpn I at 37°C for 2 h.

Protein expression and purification

E. coli BL21(DE3)/pET-28a(+)-AtATA-M14C3 and E. coli BL21(DE3/pET-28a(+)-AtATA-mutants were inoculated in 200 mL of Luria-Bertani (LB) medium containing 50 µg·mL^–1 kanamycin at 37°C with shaking at 200 rpm until the optical density at 600 nm reached approximately 0.5. Protein expression was then induced with 0.5 mmol·L^–1 isopropylthio-β-galactoside at 25°C, 200 rpm for 20 h. The cells were harvested by centrifugation (4°C, 10 min, and 8,000 rpm) and disrupted by sonication in 50 mmol·L^–1 potassium phosphate buffer (pH 8.0). The soluble proteins were purified using nickel-affinity chromatography and eluted with 250 mmol·L^–1 imidazole in 50 mmol·L^–1 potassium phosphate buffer (pH 8.0). The expression of purified proteins was verified by SDS-PAGE analysis, and the concentration of purified proteins was determined by UV absorption on a NanoDrop Lite Spectrophotometer (Thermo Fisher Scientific Inc., USA).

Enzymatic assay for the natural and non-natural substrates

The activities of M14C3 and variants toward the natural substrate pyruvate (20 mmol·L^–1) and non-natural substrate 1-acetylnaphthalene (20 mmol·L^–1) were, respectively, assayed in 5-mL reaction mixture containing 20 mmol·L^–1 1-(R)-phenylethylamine [1-(R)-PEA] as the amino donor, 0.1 mmol·L^–1 PLP as a cofactor, and 1 mg/mL purified enzyme in 50 mmol·L^–1 potassium phosphate buffer (pH 8.0). The reaction was carried out at 30°C for 15 min, and a mixture without enzyme was used as the control. The reaction was terminated by adding one equal volume of 50% (vol/vol) acetonitrile, and the mixture was centrifuged at 8,000 rpm for 10 min before product detection by HPLC analysis. The supernatant was microfiltrated against 0.22 µm filter membrane.

Apparent kinetic parameter assay

Apparent kinetic parameters of the enzymes on pyruvate and 1-acetylnaphthalene were determined in a 100 µL reaction mixture [1 mg·mL^–1 purified enzyme, 20 mmol·L^–1 1-(R)-PEA, and 0.1 mmol·L^–1 PLP] at pH 8.0 and 30°C. Initial velocities were determined with pyruvate or 1-acetylnaphthalene concentration gradient from 0.5 to 15 mmol·L^–1. An aliquot of the mixture (100 µL) was diluted with 50% (vol/vol) acetonitrile and analyzed by HPLC. One unit of enzyme activity presents 1 µmol of (R)-NEA formed per minute under optimum conditions, and the apparent kinetic parameters K_m and k_cat were acquired by performing nonlinear regression of the Michaelis-Menten equation in Origin 2021 software (OriginLab, USA). The specific activity was expressed as units of activity per gram of purified enzyme (U·g^–1). All experiments were conducted in triplicate.

Determination of the thermostability of M14C3 and variants

Half-life (t_1/2), half-inactivation temperature (T₅₀¹⁰), and melting temperature (T_m) are three important parameters for thermostability. For half-inactivation temperature (T₅₀¹⁰) measurements, purified M14C3 and variants were incubated at different temperatures (20°C–60°C) for 10 min and then swiftly cooled in an ice bath for 5 min. The activities of M14C3 and variants were detected at 30°C and pH 8.0, where the original activity was designated as 100%. The t_1/2 of purified M14C3 and variants were assayed by incubating each purified protein (1.0 g·L⁻¹) at 45°C for an appropriate time, followed by measuring the residual activity using 1-acetylnaphthalene as a substrate. The t_1/2 values were calculated according to the following first-order deactivation functions:

$${\displaystyle \mathrm{I}\mathrm{n}(A/A_0)=-k_{d}t}$$ (1)

$${\displaystyle t_{1/2}=\mathrm{l}\mathrm{n}2/k_d}$$ (2)

where A₀ is the initial activity, A is the residual activity at time t during thermal deactivation, and k_d is the deactivation rate constant (h^–1).

T_m values were assayed using differential scanning fluorimetry (DSF) (22). Aliquots of each mixture (50 µL) in final concentrations of 1 µL of 1× SYPRO Ruby orange dye (dissolved in DMSO) and 0.1 g·L^–1 AtATAs were diluted with buffer (150 mmol·L^–1 NaCl, 50 mmol·L^–1 PBS, and pH 8.0) and pre-incubated at 4°C for 20 min. The DSF protocol involved an initial temperature of 20°C, after which the temperature was increased to 75°C at a rate of 1.4°C·min^–1. Fluorescence was measured using real-time PCR, and the data were collected in increments of 0.7°C. The excitation and emission wavelengths were 490 and 605 nm, respectively. All experiments were conducted in triplicate. The T_m was calculated using the formula shown below:

$$y=\text{UF}+\frac{\text{NF}-\text{UF}}{1+e^{\frac{T_0-x}{\alpha }}}$$

where UF and NF are the minimum and maximum emission fluorescence intensities, respectively, and α is the slope of the curve within T₀. Here, x, y, and e represent the set temperature, fluorescence intensity, and natural constant, respectively.

Substrate specificity and preparation for (R)-NEA

A substrate specificity assay was performed to measure the conversion and stereoselectivity toward various ketones using M14C3 and M14C3-V5. The biocatalytic synthesis of (R)-NEA by M14C3 and M14C3-V5 was performed in a 50 or 200 mL scaled-up system. The reaction conditions were as follows: 50 or 20 mmol·L^–1 1-acetonaphthone with 1-(R)-PEA (1.0 equivalent to 1-acetonaphthone), 50 mmol·L^–1 potassium phosphate buffer (pH 8.0), 20% (vol/vol) DMSO, 0.1 mmol·L^–1 PLP, and 5.0 g_{dry cell weight, DCW}·L^–1 M14C3 or M14C3-V5 harbored by E. coli whole cells. The reaction was carried out at 30°C or 40°C and 500 rpm. The substrate and product were detected using HPLC, and the conversion rates were calculated.

Product analysis

HPLC analysis of (R)-NEA

HPLC 1220 Infinity II system (Agilent Technologies) with a C-18 column (4.6 × 150 mm, 4 µm) was used to assay the contents of 1-acetylnaphthalene and (R)-NEA. The detector wavelength was set at 210 nm. The mobile phase was composed of 0.15% ethanolamine, acetonitrile, and ultra-pure water at a volumetric ratio of 1:39:60 (vol/vol/vol) and ran at a flow rate of 1.0 mL/min. The column temperature was maintained at 30°C. The retention times of (R)-NEA and 1-acetylnaphthalene were 4.3 and 9.2 min, respectively. The enantiomeric excess of product e.e._p was determined after derivatization using HPLC. A volume of 50 µL reaction solution mixed with 100 µL of 1% (mass/vol) Marfey’s reagent (1-fluoro-2,4-dinitrophenyl-5-L-alaninamide, FDAA) in acetone and 20 µL NaHCO₃ solution (100 mM, pH 9.8) was incubated at 40°C for 2 h. Then, 20 µL of HCl (2 M) was added to quench the reaction. The samples were extracted with three times volume of dichloromethane and evaporated at room temperature. Samples were dissolved in 50% (vol/vol) acetonitrile aqueous solution. An HPLC 1220 Infinity II system (Agilent Technologies) with an EC-C18 column (4.6 × 150 mm, 4 µm) was used at 30°C. The detector wavelength was set at 340 nm. The mobile phase was composed of acetonitrile and ultra-pure water at a volumetric ratio of 60: 40 (vol/vol) and ran at a flow rate of 1 mL/min for 8 min. The retention times of (R)-NEA and (S)-NEA were 4.9 and 5.5 min, respectively. Enantiomeric excess (e.e._p) was calculated using the equation: e.e._p (%) = ([R-NEA] − [S-NEA])/([R-NEA] + [S-NEA]) × 100%, where [R-NEA] and [S-NEA] were the mole fractions of generated (R)-NEA and (S)-NEA during the enzymatic reaction.

HPLC analysis of alanine

The contents of generated (D)- or (L)-alanine were determined after derivatization by HPLC. A volume of 1 mL of sample with 2 mL derivative reagent containing absolute ethanol solution, 15 mM O-phthalaldehyde, 15 mM N-acetyl-L-cysteine, and borate buffer (0.1 M, pH 9.8) was incubated at 30°C for 15 min (5). Then, the derivatization products were detected by an HPLC 1220 Infinity II system (Agilent Technologies) with an EC-C18 column (4.6 × 150 mm, 4 µm) at 35°C. The mobile phase was composed of 50 mM sodium acetate and methanol at a volumetric ratio of 90: 10 (vol/vol) and ran at a flow rate of 1 mL/min. The detector wavelength was set at 254 nm. The retention times of derivatization (L)- and (D)-alanine were 14.1 and 12.3 min, respectively. Enantiomeric excess (e.e._p) was calculated using the equation: e.e._p (%) = ([D] − [L])/([D] + [L]) × 100%, where [D] and [L] were the mole fractions of generated (D)- and (L)-alanine during the enzymatic reaction. Each measurement was conducted at least three times with a standard deviation of less than 5%. Experiments were performed in triplicate, and data were provided as mean ± standard deviation. Statistical analysis was performed using a statistical software version SPSS 18.0. The data were subjected to variance analysis, and the means from at least three biological replicates (n ≥ 3) were tested by the least significant difference at P < 0.05.

Bioinformatic analysis

Molecular docking and molecular dynamics simulation

The crystal structure of wild-type AtATA was resolved (PDB: 4CE5) and used as a template to construct variants M14C3 and M14C3-V5 (20). Based on this model, M14C3 and M14C3-V5 were built by energy minimization after mutation using FoldX 4.0 software (43). Molecular docking was performed using YASARA (version 16.4.6) (44). The most stable docking poses showing the spatial orientation of the substrate were selected. Then, the substrate and enzyme were optimized using AMBER GAFF and AMBER 99SB force fields, respectively. The whole model is immersed in a rectangular TIP3P water box, and counterions were added into the system to balance the charge of the model. The initial coordinates and topology parameters were prepared for MD simulation using AMBER 18 software (31). The initial models built were optimized by two-step energy minimization to adjust poor interatomic interactions (32). The first step was to minimize the energy of the solvent while limiting the protein and substrate. The system was completely relaxed without any restriction in the second step. Then, the system’s temperature was heated from 0 to 303 K for 300 ps. The 100 ps of density equilibration was performed under the NPT ensemble at the target temperature of 303 K and pressure of 1.0 atm to relax the system. Finally, based on the periodic boundary conditions, a 100-ns MD simulation was performed under the NPT ensemble with a time step of 1 fs (32). Visualization and alignment of AtATA-PMP-substrate complex structures and calculation of distance were performed using PyMol (v 2.5.6). Intermolecular interaction analysis was performed using Discovery Studio 2019 Client.

Mutation binding energy

Virtual saturation mutations of residues within 8 Å of the binding position around 1-acetylnaphthalene-pyridoxamine-5′-phosphate in M14C3 were performed, and the mutation binding energies (ΔΔG) of each substitution were calculated using the Calculate Mutation Energy (Binding) module in Discovery Studio 2019 Client in pH-independent mode under the CHARMM forcefield.

Calculation of binding free energy using MM/GBSA

Molecular mechanics-generalized born surface area was used to calculate the binding free energy (ΔG_bind) between substrate and enzyme (45).

Dynamics cross-correlation maps

DCCMs were computed from the equilibrium state of M14C3 or M14C3-V5 after MD simulation using the R package Bio3D (46). The last 10 ns trajectory was saved every 10 ps and converted to a Bio3D input file using the trajectory processing program CatDCD. The C_α atoms were selected to calculate the correlation coefficients (47). DCCMs matrices were visualized using Origin 2021.

DATA AVAILABILITY

All data and materials are available upon reasonable request.

ETHICS APPROVAL

This article does not contain any studies with human participants or animals performed by any of the authors.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/aem.00543-24.

Supplemental material. aem.00543-24-s0001.docx.

Tables S1 to S5; Figures S1 to S8.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.