Design of a tri-enzymatic cascade pathway to synthesize SAA in vivo
As illustrated in Fig. 2a, SAA synthesis from L-DOPA comprises two steps: first, L-DOPA undergoes oxidative deamination by L-amino acid deaminase (LAAD, EC 1.4.3.2) to produce the prochiral intermediate 3,4-Dihydroxyphenylpyruvic acid (DHPPA); second, DHPPA undergoes reduction by NADH-dependent α-keto acid reductase to generate SAA, with coupling of formate dehydrogenase (FDH, EC 1.2.1.2) for NADH regeneration. For this process, three genes were amplified, overexpressed, and purified (Additional file 1: Fig. S1a). A previously reported highly active LAAD mutant from Proteus mirabilis (PmLAADM2, H295S/V437S) (Wu et al. 2021) and phenylpyruvate reductase from Lactobacillus sp. CGMCC 9967 (LaPPR) (Xu et al. 2016) were selected based on its specific activity (Additional file 1: Table S3-S4); and the FDH from Candida boidinii (CbFDH)was used to regenerate NADH. To assess the feasibility of the cascade pathway in vitro, theses three enzymes were combined in an equimolar ratio and incubated with 1 g L− 1 L-DOPA, 0.5 mM NAD+, and 1 g L− 1 sodium formate for 1 h, after which 0.62 ± 0.07 g L− 1 SAA was detected (Additional file 1: Fig. S2). The identity of the final product was confirmed using liquid chromatography-mass spectrometry (LC-MS) (Additional file 1: Fig. S3). This result demonstrates that the designed cascade composed of PmLAADM2, LaPPR, and CbFDH successfully converts L-DOPA to SAA. The effect of the ratio of CbFDH: LaPPR (from 0.1:1 to 3:1) on the SAA titer was further investigated using 15 g L− 1 DHPPA and LaPPR activity fixed in 5 U mL− 1. As illustrated in Fig. 2b, when CbFDH: LaPPR ratio reached 0.3:1, the SAA titer increased to 13.9 ± 0.4 g L− 1, with 91.7% conversion rate. Similarly, when the PmLAADM2: LaPPR ratio was set to 0.6:1, the SAA titer was 12.9 ± 0.4 g L− 1, with 85.6% conversion rate (Fig. 2c). Therefore, the optimal PmLAADM2: LaPPR: CbFDH ratio was defined to 0.6:1:0.3.
To construct a highly efficient conversion system for industrial application, three enzymes were co-expressed in one host strain. Then, the genes PmLAADM2, LaPPR, and CbFDH were inserted into the plasmid pRSFDuet-1 (Fig. 3a) and transformed in Escherichia coli BL21 (DE3), resulting in strain E. coli YJH01. The expression of these three enzymes was verified using SDS-PAGE (Additional file 1: Fig. S1b). The conversion performance of E. coli YJH01 was investigated with 20 g L− 1 wet cells (Fig. 3b). When L-DOPA concentration was increased from 15 g L− 1 to 50 g L− 1, the SAA titer increased from 12.75 ± 0.36 g L− 1 to 26.88 ± 0.85 g L− 1, while the conversion rate decreased from 91.2–53.6%, respectively. Concurrently, the intermediate DHPPA increased from 1.35 ± 0.05 g L− 1 to 16.28 ± 0.61 g L− 1, suggesting insufficient LaPPR activity. The comparison between the in vivo activities of PmLAADM2 (145.21 ± 3.72 U mL− 1), LaPPR (36.21 ± 1.06 U mL− 1) and CbFDH (357.1 ± 10.21 U mL− 1) revealed LaPPR as the bottleneck in this pathway, at the ratio of 4.0:1:9.9 (Additional file 1: Table. S6). Therefore, improving the enzymatic activity of LaPPR is crucial for enhancing SAA production.
Crystal structure and catalytic mechanism of LaPPR
The catalytic mechanism of LaPPR was investigated to improve its enzymatic activity. First, we solved the crystal structure of apo-LaPPR (PDB ID: 8HPG) (Fig. 4a and Additional file 1: Fig. S5). It exists as a homodimer, and each monomer folds into two distinct domains: the substrate-binding domain (SBD) and the nucleotide-binding domain (NBD). The SBD is mostly formed by 88 N-terminal residues, which folds into four α-helices and five β-strands, as well as the corresponding connecting loops. The NBD is formed by intervening residues (89–275), which fold into six central parallel β-strands surrounded by seven α-helices. The active site is located at the base of the interface between the two domains, and the two active sites of the dimer are ~ 35 Å apart. We conducted several attempts to obtain the X-ray crystal structure of LaPPR in complex with NADH and DHPPA, through optimization of the protein concentration, pH value, temperature, and precipitant concentration, and even attempted rescreening for new crystal conditions; however, we were not successful. Alternatively, we constructed the LaPPR-NADH binary complex by overlapping with the crystal structure of the complex (PDB ID: 2EKL) (Singh et al. 2014). Subsequently, DHPPA was docked into the active site of the LaPPR-NADH binary complex to obtain a kinetically stable LaPPR-NADH-DHPPA ternary complex by molecular dynamics (MD) simulations (Fig. 4b). The carboxyl group of DHPPA forms hydrogen bonds with the side chain guanidine group of R224 and with the main chain amide groups of A65 and G66. The two hydroxyl groups of DHPPA also form hydrogen bonds with G275 and T276. The conserved residues R224, E253 and H272 (Additional file 1: Fig. S7) may work as catalytic triad; H272 is suggested as a general acid to protonate the carbonyl carbon of DHPPA, and E253 could form a charge relay system with H272 for proton transfer. Based on the reported mechanism underlying the 2-hydroxy acid dehydrogenase subfamily (Jia et al. 2018; J Zhou et al. 2020), we proposed a catalytic mechanism for LaPPR: a hydride is transferred from the nicotinamide moiety carbon C4H of NADH to the carbonyl carbon atom of DHPPA, while a proton is transferred from H272 to the carbonyl oxygen atom of DHPPA to produce SAA (Fig. 4c). To verify the functions of the above residues, we site-directed mutated G66, R224, E253, H272, G275 and T276 to alanine (Fig. 4d). R224A, E253A and H272A completely abolished activity, whereas G66A, G275A and T276 decreased the activity by > 50%.
According to the mechanism of LaPPR, two key distances were defined: the hydride transfer d1 describing the distance between the carbonyl carbon atom of DHPPA and the C4H hydrogen atom of NADH, and the proton transfer d2 as the distance between the carbonyl oxygen atom of DHPPA and the imidazole ring HE2 hydrogen atom of H272. However, MD simulations on LaPPR-NADH-DHPPA ternary complex showed that, among 25000 snapshots, only 42 frames were in catalytically active conformation, with both d1 and d2 smaller than 3.0 Å (Fig. 5a). Furthermore, d1 was of 3.97 ± 0.74 Å and d2 of 4.12 ± 0.46 Å (Additional file 1: Fig. S8), suggesting that longer d1 and d2 difficult hydride and proton transfer, ultimately resulting in lower enzyme activity. Therefore, shorter d1 and d2 values are expected to improve the activity of LaPPR.
Protein engineering to enhance the activity of LaPPR
To shorten d1 and d2, 24 residues near the active site were selected as potential mutation sites. First, 16 residues near R224 and H272 were selected for NNK site saturation mutagenesis (SSM). Among those mutants, the SAA titer of the two single mutants H89M and H143D increases from 11.26 g/L (wild type) to 17.76 g/L and 18.64 g/L, respectively (Fig. 5b). When combining these two single mutants to obtain the double mutant LaPPRMu1 (H89M/H143D), the SAA titer increased to 19.48 g/L. (Fig. 5b) Furthermore, MD simulations showed that the flexible regions of loop-7 (residues 249–257) tended to approach the active site. Eight residues in the loop-7 were selected to further improve the activity of LaPPRMu1 (H89M/H143D). The mutant LaPPRMu2 (H89M/H143D/P256C) exhibited 2.5-fold SAA titer that of WT (Fig. 5b). The kinetic parameters of LaPPR and its mutants are summarized in Table 1. The specific activity, Km, kcat/Km of LaPPRMu2 were 3.8 times, 0.6-fold, 10.3 folds the corresponding values of WT.
Table 1
Kinetic parameters of the purified LaPPR WT and its mutants
Mutants | Specific activity (U·mg− 1) | Km (mM) | kcat (S− 1) | kcat/Km (S− 1·mM− 1) |
LaPPRWT | 5.79 ± 0.16(1) | 6.20 ± 0.15 | 1.21 ± 0.03 | 0.20(1) |
LaPPRH89M | 9.07 ± 0.28(1.6) | 5.74 ± 0.08 | 1.37 ± 0.06 | 0.24(1.2) |
LaPPRH143D | 9.33 ± 0.26(1.6) | 5.03 ± 0.14 | 2.08 ± 0.11 | 0.41(2.1) |
LaPPRMu1 | 11.35 ± 0.32(2.0) | 3.63 ± 0.17 | 2.35 ± 0.12 | 0.65(3.3) |
LaPPRMu2 | 21.77 ± 0.67(3.8) | 2.52 ± 0.27 | 5.74 ± 0.24 | 2.05(10.3) |
To elucidate the molecular basis of higher activity of LaPPRMu2, we conducted MD simulations on the LaPPRMu2-NADH-DHPPA ternary complex. As expected, the proportion of catalytically active conformations in the mutant increased (610/25000 vs 42/25000, Fig. 5a). Furthermore, the hydride transfer d1 and proton transfer d2 shortened from 3.97 ± 0.74 Å to 3.51 ± 0.61 Å and from 4.12 ± 0.46 Å to 4.01 ± 0.56 Å, respectively (Additional file 1: Fig. S8), suggesting that hydride and proton transfer are more likely to occur, which is consistent with the measured increase in kcat. Upon the mutation of H143D, anticorrelated motions from the increased interactions between the phosphate group in NADH and the nearby residues G142 and T141 have a pushing motion, which pushes the C4H of NADH niacinamide ring toward the Cα atom of DHPPA (Fig. 5c-d). When H89 was mutated to methionine, the adverse interaction of nearby residues around the substrate was reduced, allowing the Cα atom of DHPPA to be closer to NADH and DHPPA. Additionally, the P256C mutation contributed to the formation of a larger catalytic pocket (Additional file 1: Fig. S9) and a more flexible conformation (Additional file 1: Fig. S8), which modified the binding model of DHPPA, thus making DHPPA closer to NADH and His272.
One-pot production of SAA in vivo
We imported the best triple mutant LaPPRMu2 into the cascade pathway in vivo to generate the resulting strain E. coli YJH02 (Fig. 6a), and it produced 35.75 g L− 1 SAA from 50 g L− 1 L-DOPA (Fig. 6b). However, 1.75 g L− 1 L-DOPA remained unmodified and 9.54 g L− 1 DHPPA accumulated. This could derive from the specific activity of PmLAADM2, LaPPRMu2 and CbFDH in E. coli YJH02, being 133.51 ± 3.51 U mL− 1, 63.15 ± 2.56 U mL− 1, and 309.58 ± 8.71 U mL− 1, respectively(Additional file 1: Table. S5). These values correspond to an enzymatic activity ratio of 2.1:1:4.9, suggesting that high CbFDH activity affected the expression of PmLAADM2 and LaPPRMu2. To overcome this issue, the recombinant strains were constructed using different combinations with different copy number plasmids (pCDFDuet-1, pETDuet-1, and pRSFDuet-1 plasmids with copy numbers of 20, 40, and 100, respectively) (Fig. 6a). Among them, E. coli YJH05 was found to be best for SAA production (Fig. 6b). In this strain, the activities of CbFDH and PmLAADM2 were decreased by 42.3% (178.56 ± 4.32 U mL− 1) and 4.4% (127.63 ± 3.15 U mL− 1), while the activity of LaPPRMu2 was increased by 38.6% (87.55 ± 2.82 U mL− 1) in E. coli YJH05 compared to the E. coli YJH02 (Additional file 1: Table. S5). Whole-cell catalysis in E. coli YJH05 produced 41.67 g L− 1 SAA from 50 g L− 1 L-DOPA, corresponding to 83.1% conversion rate, and 5.62 g L− 1 DHPPA was detected (Fig. 6b). Subsequently, we applied ribosome-binding site (RBS) regulation to enhance the protein expression (Additional file 1: Table. S6). Among the RBS strains, E. coli YJH12 with RBS6 produced a SAA titer of 45.84 g
L− 1, with a conversion rate of 91.2%, minimal DHPPA accumulation (0.86 g L− 1) and no detectable remaining L-DOPA (Fig. 6c). The specific activities of PmLAADM2 and CbFDH in this strain were 94.38 U mL− 1 and 139.32 U mL− 1, respectively, whereas the activity of LaPPRMu2 was 112.36 U mL− 1, resulting in a ratio of 0.8:1:1.2 (PmLAADM2: LaPPR: CbFDH) (Additional file 1:Table. S5), suggested that the balance between PmLAADM2, LaPPRMu2 and CbFDH was unobstructed. To further improve the production efficiency of SAA, the transformation conditions for E. coli YJH12 (Additional file 1: Fig. S10). The effect of pH (6.5 to 9.0), temperature (20 to 40°C), wet-cell concentration (10 to 40 g L− 1) and NAD+ content (0.2 mM to 1.6 mM) on the SAA titer were investigated at a 10 mL scale. Under the optimal conditions (pH 7.0, 30°C, 15 g L− 1 wet cells, 0.6 mM NAD+), 47.14 g L− 1 SAA can be produced from 50 g L− 1 L-DOPA with a conversion rate of up to 93.8%.
Finally, we explored the scale-up transformation of SAA and the effect of substrate feeding modes on its titer. Among them, the batch feeding has the highest titer: the initial concentrations of L-DOPA was 15 g L− 1 and the ratio of L-DOPA to sodium formate was set 1:1.2, then L-DOPA was added hourly to increase its concentration by 6.25 g L− 1 for the duration of 12 batches (Fig. 6d). Finally, from 90 g L− 1 L-DOPA (the optimal batch feeding) in 12 h under the optimal transformation conditions, we obtained 82.55 g L− 1 SAA using 30 g L− 1 E. coli YJH12 (wet cells), with a conversion rate of 91.3%, productivity of 6.88 g L− 1 h− 1 and excellent e.e value (99%). These results demonstrate the potential of the engineered strain E. coli YJH12, for the industrial production of SAA from L-DOPA.