Introduction

The Baldwin cyclization rules were originally proposed with the aim of systematizing a huge set of previous empirical observations1,2. Baldwin proposed that maximum orbital overlap as the crucial element of stereoelectronic effects is the reason for the often observed regioselectivity leading to differently sized cyclic products. In seminal prior work, Eschenmoser et al. studied ring closures involving SN2 reactions at sp3 hybridized C-centers, demonstrating that they occur most rapidly when an angle of 180o between the three interacting atoms is maintained3. They also surmised that when the intramolecular substitution reaction occurs at the C-atoms of epoxides, the extent of the stereoelectronic effect may be somewhat relaxed. These developments inspired Janda, Lerner and coworkers to generate catalytic antibodies which catalyze a disfavored ring closure reaction (anti-Baldwin) of a specifically designed epoxy alcohol4,5. A chiral N-oxide antigen was coupled to the keyhole limpet hemocyanin, the conjugate then being used to immunize 129 mice for monoclonal antibody production. It was found that the 6-endo-tet cyclization mode with formation of the anti-Baldwin tetrahydropyran product dominates. In the absence of the antibody, cyclization to the normal product occurred, a background reaction in the designed system. Nicolaou et al. described an unusual anti-Baldwin cyclization in the synthesis of brevetoxin B6. More recently, Alabugin et al. have analyzed the Baldwin rules more closely and extended them accordingly7,8. They pointed out, inter alia, that in both cyclization modes of the Janda/Lerner-study, the “breaking bond is outside of the formed ring”, calling for an extension of the original Baldwin rules in which spiro and fused transition states, respectively, are compared8. Relevant is the literature describing polyether biosynthesis in which both favored and disfavored epoxy-alcohol ring closures occur. Hotta, Chen, Houk and coworkers have studied these enzymatic processes by X-ray structures and quantum mechanical (QM) computations9. In this context, other Baldwin cyclization studies deserve mention10,11.

In the present study, we devised a rational biocatalytic mutagenesis strategy using an epoxide hydrolase as the enzyme and hydroxyl/amino-substituted epoxides and oxetanes as substrates, the aim being to induce Baldwin cyclization rather than hydrolysis leading to the respective natural hydrolytic 1,2-diol products (Fig. 1a). Control of Baldwin and anti-Baldwin cyclization modes was part of the challenge. Originally, we chose 1-epoxy-1-pentanol as the model substrate, structurally similar to the substrate in the Janda/Lerner study, but soon discovered that even in the absence of an enzyme this compound undergoes a rapid background reaction leading to the respective tetrahydrofuran derivative. Rather than subtracting the effect of the background reaction from the experimentally observed result4, we opted for the next higher homologous substrate 4-(oxiran-2-yl)butan-1-ol with formation of tetrahydropyran-2-methanol as Baldwin product and oxepan-3-ol as anti-Baldwin product (Fig. 1b). Limonene epoxide hydrolase (LEH) was chosen as the enzyme, its crystal structure and mechanism being essential for our endeavor. We were especially motivated by the possibility to access a broad range of N- and O-heterocyclic building blocks needed for the synthesis of chiral therapeutic drugs, including anti-cancer compounds, antibiotics and immunosupressants (Supplementary Fig. 1)12,13.

Fig. 1: Schematic representation of LEH-catalyzed reactions.
figure 1

a Natural hydrolytic reaction catalyzed by wild-type (WT) ReLEH (top) and its known catalytic mechanism (bottom)23; D, Aspartic acid; N, Asparagine; Y, Tyrosine. b The proposed approach to biocatalytic routes using rationally engineered ReLEH mutants in this work.

Epoxides and oxetanes have been used as starting materials for chemical syntheses of chiral heterocycles by employing chiral transition metal catalysts or reagents14,15,16,17. In this realm, the Jacobsen group14 reported the intramolecular enantioselective ring opening of epoxides for the preparation of chiral O-heterocycles synthesis. Intramolecular desymmetrization of oxetanes has also been established as a synthetically useful strategy16. Nevertheless, most of these approaches require either transition metals, which in the workup need to be eliminated, or stoichiometric amounts of organic reagents. It inevitably leads to extra operational labor or stoichiometric waste generation, so the development of economically and ecologically viable catalytic approaches remains in high demand. Enzymes are alternatives, which can catalyze reactions with unrivaled rate and exceptional selectivity18,19,20,21.

Nature has created a vast repertoire of enzymes that synthesize chiral N- and O-heterocycles through ring-opening and intramolecular transformations9,22, including epoxidation-cyclization, chlorination-cyclization and others, but the respective substrate scopes are limited, as in polyether epoxide hydrolases/cyclases and isomerases/cyclases23,24. Consequently, we wished to develop an alternative biocatalytic approach, in which a single catalyst platform would provide direct access to diverse chiral heterocycles from appropriate epoxide/oxetane precursors as starting materials. It should be noted that oxetanes have never been subjected to any epoxide hydrolase-mediated transformations.

We turned to cofactor-independent limonene epoxide hydrolases (LEHs) that normally catalyze direct hydrolysis of epoxides by positioning and activating water at the active site. Taking the LEH from Rhodococcus erythropolis (ReLEH) as an example, it is highly selective to the hydrolysis of the epoxide substrates with the formation of the respective chiral vicinal diols, which in our endeavor had to be shut down, as will be shown. On the basis of the well-documented mechanism of ReLEH, a water molecular is positioned by H-bonds to N55 and Y53 while D132 abstracts a proton from the water, enabling a smooth nucleophilic attack on one of the substrate’s epoxide carbon center (Fig. 1a)25. In addition, previous studies have revealed that some EHs convert bis-epoxy polyketide intermediates into the corresponding furan-/pyran-cyclized natural products, and the stereochemical courses of the cyclization were solely governed by Baldwin’s rules10,26,27,28,29,30. In contrast, the enzymatic processes in Fig. 1b catalyzed by LEH have not been reported to date. As a first step in achieving this reprogrammed reactivity, the natural hydrolysis activity of wildtype (WT) ReLEH must be inhibited, which affords mainly the respective 1,2-diol and about 10% of the 6-membered Baldwin cyclization product. Here we present a rational mutagenesis-based design of ReLEH that selectively catalyzes the transformation of a range of epoxides and oxetanes to their cyclization-products, in which the distal HO- and RHN-groups serve as nucleophiles (Fig. 1b).

In this work, the genetically modified enzymes described herein perform the enantioselective synthesis of structurally diverse analogs of heterocycles accessible in fewer steps and in an ecologically and economically viable manner, while in conventional chemistry such syntheses are often lengthy and generally require the use of protecting groups and/or transition metals.

Result and discussion

Engineering ReLEH towards Baldwin cyclizations

Figure 1a displays the conventional catalytic mechanism of LEH, residues Y53 and N55 positioning water for nucleophilic attack, while D132 assists nucleophilic substitution and D101 ensures stabilization of the incipient negatively charged O-atom of the epoxy function31. We set out to disrupt water activation by appropriate mutagenesis, while keeping D101. Consequently, our focus was on Y53 and N55, but not on D101. In order to reprogram the enzyme catalytic properties, hydrophobic amino acids (Gly, Ala, Val, Pro, Leu, Ile, and Phe) in addition to WT were employed as building blocks at the residues Y53 and N55 with the aim to break the hydrogen bonding network (Fig. 2a). Degenerate codons GBA, WTK and Pro were rationally employed for library construction (Supplementary Figs. 2 and 3, and Supplementary Tables 1 and 2). To ensure 95% library coverage, ~180 clones were then screened in cell lysates toward the transformation of 5 mM rac-1 (Fig. 2b). Several hits with essentially eliminated WT hydrolytic activity were identified, including variants SZ532 (Y53L/N55P), SZ534 (Y53V/N55P), SZ609 (Y53I/N55A) and SZ611 (Y53F/N55A). Indeed, they enable Baldwin-type cyclization with little formation of undesired 1,2-diols (Fig. 2c). In the case of the double mutant SZ611 (Y53F/N55A), the formation of Baldwin product 2 is improved from 10% (WT) up to 78% (2:3 = 7.8) (Fig. 2d). In parallel, to double check the drop-off of hydrolytic property, the WT enzyme and variants were also employed in the transformation of cyclohexene oxide, which lacks a hydroxyl group for possible intramolecular processes. Compared to WT, the hydrolysis conversion is largely decreased from 99 to 46% (Supplementary Table 3). These results suggest that the substitutions Y53F and N55A of ReLEH can effectively break the water binding network, thereby endowing Baldwin cyclization reactivity. This was our primary goal, to be followed by engineering notable regioselectivity (Baldwin vs. anti-Baldwin) and high enantioselectivity.

Fig. 2: Reprogramming the catalytic repertoire of ReLEH towards the Baldwin-type reactions.
figure 2

a Design strategy of breaking the water binding network in ReLEH. b ReLEH-catalyzed model reaction leading to three distinct products. c Screening results of the first round of mutagenesis performed on residues 53 and 55. d GC profiles of the mutant SZ611. e Screening results of the second round of mutagenesis targeting residues 80, 83, 114 and 116. f Conversion and e.r. values of the mutants. M1 and M2 indicate the first and the second rounds of mutagenesis. Reaction conditions: LEH (0.1 g/mL wet cell), rac-1 (5 mM), PBK buffer (50 mM, pH 7.4), Lysozyme (1 mg/mL), DNase I (6 U/mL), 0.4 mL total volume, 30 °C, 1000 rpm, 24 h.

In order to exert greater evolutionary pressure, a second round of genetic manipulation based on rational saturation mutagenesis using mutant SZ611 (Y53F/N55A) as template was performed at the key active site residues Ile80, Val83, Leu114, and Ile116 surrounding the catalytic center D101 (Fig. 2a). We were guided by the enzyme mechanism and realization that hydrophobic amino acids should be chosen for exchange events at these hotspots. Moreover, based on earlier mutagenesis studies of LEH, QM cluster techniques had supported the mechanistic role of D10132. In the present study, we chose the triple code Val-Phe-Tyr which encodes hydrophobic amino acids, the double mutant Y53F/N55A serving as the template. After screening 576 transformants with automated gas chromatography (GC) for 95% library coverage, a distinctly improved mutant SZ616 (Y53F/N55A/I116V) was obtained, enabling a 2:3 ratio of 6.7 favoring the Baldwin reaction mode, although conversion to the anti-Baldwin product 3 was only 13%, the rest of complete conversion being 2. Stereoselectivity varied, reaching a maximum of e.r. = 90:10. In curiosity-driven experiments, the mutants SZ612 (Y53F/N55A/I80F/L114V/I116V) and SZ621 (Y53F/N55A/I80F/I116V) produced only the Baldwin product 2 with none of 3 nor of the vicinal diol 4 as side-products (Fig. 2e, Supplementary Fig. 4).

Expansion of substrate scope of ReLEH variants following Baldwin cyclization reactivity

In order to evaluate the substrate scope of the engineered ReLEH mutants for the synthesis of heterocycles, a series of epoxy alcohols were prepared and tested. Most of the respective Baldwin products were obtained with good to excellent conversions and high cyclization mode selectivity without the need for additional mutagenesis, while diastereo- and enantioselectivity varied (Fig. 3). For example, variant SZ616 (Y53F/N55A/I80F/I116V) catalyzes the cyclization of the 5,6-epoxy alcohol 5 with sole formation of the Baldwin tetrahydropyran product 6 (99% conversion), and the diastereomeric ratio is 86:14.

Fig. 3: Substrate scope of Baldwin reactions catalyzed by LEH mutants.
figure 3

Reaction conditions: LEH (0.1 g/mL wet cell), Substrates (5 mM), PBK buffer (50 mM, pH 7.4), Lysozyme (1 mg/mL), DNase I (6 U/mL), 0.5 mL total volume, 30 °C, 1000 rpm, 24 h.

Selective desymmetrization reactions of oxetanes

Natural enzymes that catalyze oxetane-opening reactions have rarely been reported33,34. We were all the more interested to learn how our engineered LEH mutants would perform using such substrates (Fig. 4). Initially, we explored the intramolecular desymmetrization reaction of substrate 9, WT not showing any reaction. It is of note that WT still maintains hydrolysis ability towards cyclohexene oxide when simultaneously adding substrate 9 in the reaction system, suggesting that substrate 9 does not function as an inhibitor of WT ReLEH (Supplementary Fig. 5). We were pleased to discover that when using triple mutant SZ616 (Y53F/N55A/I116V), the tetrahydrofuran product 10 was obtained in 96% conversion with a 72:28 enantioselectivity ratio (e.r.). Upon testing variant SZ621 (Y53F/N55A/I80F/I116V) having the additional I80F mutation, stereoselectivity of the target product 10 was improved to 87:13 e.r. (Fig. 4, Supplementary Table 4). All other variants have no increased catalytic activity in this reaction. We therefore employed variants SZ616 and SZ621 as catalysts in the reaction of a range of structurally different oxetanes. Substrate scope and selectivities proved to be good to excellent in numerous cases (Fig. 4). A series of oxetanes bearing electron-donating and electron-withdrawing groups on the phenyl rings also underwent cyclization to afford the respective products (10, 12–32) in high conversion with moderate stereoselectivity. Notably, substrates bearing dimethyl-substituted phenyl (15) and bicyclic naphthalene rings (23) were well tolerated under standard conditions, affording the cyclization products in conversion of 85% and 95%, respectively. Oxetanes 33, 35, 37, and 39 with various alkyl groups or hydrogen atom, were accepted as substrates, and the corresponding products (34, 36, 38, and 40) were obtained with 20–96% conversion and up to 96:4 e.r. In addition, even substrates 41 and 43 reacted to give chiral tetrahydropyran products. To demonstrate the synthetic practicality of this methodology, several preparative up-scaled reactions were performed for oxetanes 9 and 11-35, employing the best mutant SZ616 at 50 mM substrate loading, resulting in moderate to good yields (53–88%) with moderate to excellent stereoselectivity.

Fig. 4: Substrate scopes of ReLEH mutants in the synthesis of chiral tetrahydrofurans and tetrahydropyrans.
figure 4

Reaction conditions: ReLEH (0.1 g/mL wet cell), Substrates (5 mM), PBK buffer (50 mM, pH 7.4), Lysozyme (1 mg/mL), DNase I (6 U/mL), 0.5 mL total volume, 30 °C, 1000 rpm, 24 h. *: Isolated yields of products with a substrate loading of 50 mM in 20 mL volume. c: Conversion (%). n.d.: Not detected.

Due to the widespread presence of N-heterocyclic compounds in natural products and pharmaceuticals, we extended our synthetic targets to obtain such compounds using aforementioned mutants (Fig. 5). This prompted us to examine the intramolecular opening of oxetanes with N-centered nucleophiles. To our delight, a wide variety of oxetanes underwent cyclization-opening with very high enantioselectivity and conversion. Change of substituents on the amide moiety was well tolerated (substrates 45, 4757). Noteworthy is that compounds bearing N-aromatic substituents with electron-donating or withdrawing groups were readily converted into pyrrolidines (products 54, 56, and 58) in high conversion with excellent enantioselectivity. Further examples of wide structural variation are shown in Fig. 5. Furthermore, cyclizations into the next larger ring to provide chiral piperidines with up to 98% conversion and an enantiomeric ratio of 99:1 were achieved, thereby establishing a simple method to synthesize chiral piperidine drug candidates. The absolute configuration of the product was determined by X-ray crystal structure of a derivative of 54 (Supplementary Table 5). In further work using purely synthetic reagents, many of the N-heterocyclic compounds presented herein were transformed by chemical means into other derivatives of potential application in the production of chiral pharmaceuticals (Supplementary Fig. 6). Intriguingly, the enzymatic product 46 can be utilized to synthesize the natural product (R)-Coerulescine that has analgesic effects as reported35.

Fig. 5: Substrate scope of ReLEH variants in the synthesis of chiral pyrrolidines.
figure 5

Reaction conditions: ReLEH (0.1 g/mL wet cell), substrate (5 mM), PBK buffer (50 mM, pH 7.4), Lysozyme (1 mg/mL), DNase I (6 U/mL), 0.5 mL total volume, 30 °C, 1000 rpm, 24 h. *: Isolated yields of products with a substrate loading of 50 mM in 20 mL volume. c: Conversion (%). n.d.: Not detected.

Shedding light on the catalytic mechanism of Baldwin ring closure and desymmetrization

In order to gain insights on the active substrate-LEH interactions, the X-ray structures of the key variants SZ612 and SZ616 were analyzed (Supplementary Table 6). Consequently, the representative substrates in complex with LEH mutants, including the models SZ612‒(R)-1 and SZ616‒(R)-1, were constructed by molecular docking. Following the docking analysis, the epoxy ring O-atom of the substrate consistently establishes a hydrogen-bond interaction with the protonated catalytically active D101; the aforementioned hydrogen-bond network with water molecular is terminated by substitutions Y53F and N55A, enabling the substrate distal hydroxyl moiety to form a hydrogen-bond interaction with the deprotonated D132 (Fig. 6a).

Fig. 6: Computational analysis of the designed mutants SZ612 and SZ616.
figure 6

a Proposed mechanism of LEH-catalyzed cyclization of the model substrate (R)-1. b Schematic representation of the four possible products. c Model SZ612‒(R)-1 and its free energy landscapes (kcal/mol) as a function of the dihedral angle θ (ON-C1-C2-O). d Model SZ616‒(R)-1 and its free energy landscapes (kcal/mol) as a function of the dihedral angle θ (ON-C1-C2-O).

In the scenario of evolved Baldwin/anti-Baldwin selection and stereoselectivity, two regioselective products are possible via Baldwin or anti-Baldwin selection, each in two enantiomeric forms. Thus, four different products can be formed (Fig. 6b). In detail, if the distal deprotonated O-atom (dubbed ON) attacks the reactive carbon (C1 or C2) from the Si-face, the enantiomers (S)-2 or (S)-3 are generated. In contract, the enantiomers (R)-2 or (R)-3 are formed when attack occurs from the Re-face (Fig. 6b). In order to probe the conformational bias for the substrate (R)-1 in variants SZ612 and SZ616, two criteria were employed to analyze the MD trajectories of models SZ612‒(R)-1 and SZ616‒(R)-1: (I) The (S)- or (R)-selectivity caused when the ON attack occurs from the Si- or Re-face, by estimating the dihedral angle θ between the four atoms ON-C1-C2-O. If 0o < θ < 180o, then ON attacks from the Si-face, generating (S)-configurated products; or if −180o < θ < 0o, then ON attacks from the Re-face, to form (R)-configurated products. (II) The Baldwin or anti-Baldwin selectivity caused when the ON attack occurs at C1 or C2, respectively, by determining the difference in attacking distances d1 (from ON to C1) and d2 (from ON to C2). If d1 < d2, then C1 is attacked preferentially, yielding Baldwin products. If d1 > d2, then C2 will be attacked, leading to anti-Baldwin products. As such, 500-ns adaptively biased molecular dynamics (MD) simulations were performed to investigate the free energy landscapes using the dihedral angle θ and the distance d1 as collective variables. In model SZ612‒(R)-1, the dihedral angle θ was mainly gathered between 0o and 180o during the MD simulations (Fig. 6c), suggesting that attack from the Re-face is unsupportive for SZ612. In contrast, the model SZ616‒(R)-1 unveiled two peaks in terms of the dihedral angle θ (Fig. 6d), revealing that the attack from Si- or Re-face has no preference in the case of SZ616. Moreover, for both of the SZ612 and SZ616 mutants, the slightly shorter distance d1 was observed during the simulations (Supplementary Figs. 7a and 7b), indicating that the ON atom preferentially attacks C1. Furthermore, to better illustrate the bond forming and breaking in the Baldwin/anti-Baldwin selection pathways (Supplementary Movies 14), the hybrid QM/MM scheme was employed, enabling the calculation the activation energy barriers. It is of interest to note that the shorter distance d2 allows a relatively low activation energy (Supplementary Figs. 7c and 7d), and hence the formation of Baldwin product is favored for the two variants. These observations are also in good agreement with the experiment results.

For the desymmetric oxetane reactions, we propose a putative catalytic pathway, which mainly includes the following crucial processes: (1) deprotonation of the amine group; (2) nucleophilic attack of carbon atom (C2) of the oxetane; (3) protonation of oxygen atom of oxetane (Fig. 7a). To further elucidate the mechanism, crystallographic experiments were performed. We obtained the complex crystals of SZ611-substrate 59 and SZ611-product 46 using the soaking methods as described in Material and Methods. The complex structures of SZ611-substrate 59 (PDB code: 7XEE) and SZ611-product 46 (PDB code: 7XEF) were determined to 1.88 Å and 1.82 Å, respectively (Supplementary Table 6). SZ611 assembles into a homodimer structure with four α-helices and five-strand β-sheets in each monomer, forming a cavity that serves as a deep substrate-binding pocket, in which a substituent of substrate 59 or of product 46 can occur without steric repulsion (Fig. 7b). The electron density map of such ligands and surrounding key residues were defined with good quality (Fig. 7c, d). A major portion of substrate 59 was clearly defined in the electron density map, apart from a minor portion of the phenyl group. In contrast, product 46 was well-defined in the electron density map, including the phenyl group corresponding to that of substrate 59. The phenyl group is located in the hydrophobic local space formed by the hydrophobic residues of L71, L74 and F75, which were clearly defined in the electron density map. Two other essential mutations, Y53F and N55A, have well-defined electron density maps that together expand the hydrophobic local space, thereby leading to the preference of intramolecular cyclization reactions. The molecular architecture of substrate 46 in the binding pocket was mainly attributed to its multiple interactions with catalytic residues D101 and D132 (Fig. 7e). The N-atom of substrate 59 is hydrogen-bonded to D132, which commonly functioned as a proton receptor for the deprotonation of an amine group, or for deprotonation of water molecules in WT LEH and other LEH mutants. The O-atom of oxetane is within the hydrogen-bonding distance to D101, which acts as a proton donor for the protonation of the ring-opening O-atom, in agreement with an earlier study featuring the X-ray structure of WT LEH36. Product 46 shares a similar conformation with substrate 59 considering the hydrogen-bonding networks of amines and hydroxyl groups with D132 and D101 (Fig. 7f), which together predicts a pathway for the LEH-catalyzed selective desymmetrization reactions of oxetanes. In the catalytic reactions, residue D132 acts as a proton acceptor to promote the deprotonation of the amino group. The deprotonated N-atom then nucleophilically attacks the oxetane moiety, resulting in ring opening, followed by the protonation of the O-atom of the oxetane precursor with D101 as the proton donor.

Fig. 7: Structural analysis of the designed mutant SZ611.
figure 7

a Proposed mechanism of LEH-catalyzed selective desymmetrization reactions of oxetanes. b Overall structure of LEH bound to substrate 59 and product 46 (yellow). c Electron density for substrate 59. d Electron density for substrate product 46, Fo-Fc map contoured at 3.0σ, green mesh and protein side chains shown in blue mesh, 2Fo-Fc map contoured at 2.0σ. e Hydrogen-bonding interactions between substrate 59 and protein side chains. f Hydrogen-bonding interactions between product 46 and protein side chains. Dash: hydrogen-bonding interactions.

We have rationally engineered the epoxide hydrolase ReLEH with construction of a general platform for Baldwin cyclization reactivity of hydroxy- and amino-epoxides and oxetanes. The genetic reprogramming of the enzyme enables an efficient biocatalytic route to chiral N- and O-heterocycles of considerable significance in the pharmaceutical field. Structurally diverse chiral heterocycles can now be accessed from appropriate epoxides and even oxetanes with high reactivity and selectivity. Considering the fundamental roles of chiral heterocycles in the biological world and the pharmaceutical industry, we anticipate that these fully genetically encoded enzymes will provide a starting point to further extend the toolbox of ring-expansion catalysts. The rational enzyme design strategy presented in this study also provides a guidance for engineering other natural enzymes with the aim to access hitherto never observed functions.

Methods

General

NMR spectra were recorded on Bruker AV 400 MHz instrument (400 MHz for 1H, 100 MHz for 13C, and 376 MHz for 19F). Chemical shifts were reported in ppm down field from internal Me4Si and external CCl3F, respectively. Multiplicity was indicated as follows: s (singlet), d (doublet), t (triplet), q (quartet), m (multiplet), dd (doublet of doublet), br (broad). A waters G2-XS qtof instrument was used to generate the high-resolution mass spectrometry (HRMS) spectra data. Small molecular X-ray structural analysis was carried out using Bruker APEX-II CCD instrument. SHIMADZU 2030 HPLC and GC systems were used for HPLC and GC analysis, respectively. KOD Hot Start DNA Polymerase (Novagen) and restriction enzyme Dpn I (NEB) were used in this work. The oligonucleotides were synthesized by Life Technologies. Plasmid preparation kit was ordered from Zymo Research, and PCR gel extraction kit was bought from QIAGEN. DNA sequencing was conducted by GATC Biotech. Tetrahydrofuran (THF), diethyl ether, and toluene were distilled from sodium/benzophenone prior to use; CH2Cl2 was distilled from CaH2; All purchased reagents were used without further purification. Analytical thin layer chromatography (TLC) was performed on 0.20 mm silica gel plates, while silica gel (200‒300 mesh) was used for flash chromatography. All the silica gels were purchased from Qingdao Haiyang Chem. Company, Ltd.

PCR-based methods for library construction

KOD Hot Start polymerase was used to construct mutant libraries via the megaprimer approach. 30 µL water, 5 µL KOD hot start polymerase buffer (10×), 3 µL 25 mM MgSO4, 5 µL 2 mM dNTPs, 2.5 µL DMSO, 0.5 µL (50~100 ng) template DNA, 100 µM primers mix 0.5 µL each and 1 µL KOD hot start polymerase were added as a 50 µL reaction mixtures. The PCR cycles used for generating short fragment: 95 °C 3 min, (95 °C 30 s, 56 °C 30 s, 68 °C 40 s) × 32 cycles, 68 °C 120 s, 16 °C 30 min. While in mega-PCR: 95 °C 3 min, (95 °C 30 s, 60 °C 30 s, 68 °C 5 min 30 s) × 24 cycles, 68 °C 10 min, 16 °C 30 min. The PCR products were analyzed on agarose gel by electrophoresis. The digestion was carried out at 37 °C for more than 3 h by adding 2 µL NEB CutSmart™ Buffer and 2 µL Dpn I in 50 µL PCR reaction mixture. After digestion, 1 µL PCR products were directly transformed into electro-competent E. coli BL21(DE3) to create the final library for Quick Quality Control and screening. Primers used in this work are listed in Supplementary Tables 1 and 2.

Screening procedures

Colonies were picked and transferred into deep-well plates, which contain 300 µL LB medium with 50 µg/ml carbenicillin in each well. After culturing overnight at 37 °C with shaking, 120 µL was transferred to glycerol stock plate and stored at −80 °C. Then, 800 µL TB medium with 0.5% (m/v) lactose and 50 µg/mL carbenicillin was added directly to the culture plate, shaking at 28 °C for protein expression. After 8 h, the cell pellets were harvested and washed using 400 µL 50 mM pH 7.4 potassium phosphate buffer and centrifuged for 10 min 1100 × g at 4 °C. Then, the pellets were resuspended in 400 µL of the same buffer with 6 U DNase I and 1 mg/mL lysozyme for breaking the cell at 30 °C for 1 h with shaking. The crude lysate was centrifuged at 1100 × g and 4 °C for 30 min. Afterwards, 300 µL supernatant of the lysate was transferred into new deep-well plates, which preloaded 100 µL 5 mM substrate rac-1 and 5% acetonitrile as co-solvent. After reaction at 30 °C and 800 rpm for 14~16 h, the product and remaining substrate were extracted using equal volumes of ethyl acetate (EtOAc) for GC analysis by chiral column Hydrodex-β-TBDAc, 25 m × 0.25 mm ID as the follows condition: 110 °C, 5 °C/min, 135 °C, 20 °C/min, 220 °C hold 2 min. N2: 1.5 bar.

X-ray structure determination

The proteins, SZ611 at 10 mg/mL, SZ612 at 10 mg/mL and SZ616 at 20 mg/mL, containing solution (25 mM HEPES pH 7.5, 150 mM NaCl) were mixed in a 1:1 ratio with the reservoir solution in a final volume of 2 μL and equilibrated with the reservoir solution. The SZ611 crystals were grown in 0.2 M NaCl, 0.1 M Bis-Tris pH 5.5, 30% (w/v) PEG 3350, and supplemented with additives (0.5% (w/v) casein, 0.5% (w/v) hemoglobin, 0.005% (w/v) pepsin, 0.005% (w/v) protease, 0.005% (w/v) proteinase K, 0.005% (w/v) trypsin, and 0.02 M HEPES sodium pH 6.8) at a ratio of 10:1, using the sitting-drop vapor diffusion method at 20 oC. The complex crystals of SZ611-substrate 59 and SZ611-product 46 were made by soaking SZ611 crystals in growth buffer containing 32 mM substrate 59 and substrate 45 for 10 min, respectively. The SZ612 crystals were grown in 0.2 M ammonium acetate, 25% (w/v) PEG 3350, and the SZ616 crystals were grown in 0.2 M ammonium formate, 15% (w/v) PEG 3350, both of which used the sitting-drop vapor diffusion method at 4 oC. All crystals were mounted in a cryoloop and frozen in liquid nitrogen after immersion in a well solution containing 10–25% glycerol. All data were collected at 100 K. Diffraction data of the SZ611 and SZ612 were measured at the wavelength of 1.03312 Å and SZ616 at 0.97853 Å at BL17B1 beamline of Shanghai Synchrotron Radiation Facility. Diffraction data of both SZ611-substrate 59 and SZ611-product 46 were measured at the wavelength of 1.5418 Å on an R-AXIS IV++ imaging plate detector. All data sets were processed by using the HKL-2000 package37. The crystal structures were solved by molecular replacement method using the program PHASER (version 2.7.16)38 using the structure of wild-type LEH (PDB code 1NWW) as a search model. Further refinement was performed with PHENIX (1.11.1_2575)39, and the models were rebuilt manually with Coot (0.9.8.1)40. Data collection and refinement statistics were summarized in Supplementary Table 6.

Model preparation and model generation and substrate docking

Schrödinger Maestro software was used to prepare protein and ligand structures41. The X-ray structures of variants solved in this study (PDB IDs 7VWD, 7VX2, and 7VWM) were used as templates. The substrate (R)-1 was separately docked into 7VX2 and 7VWM and generated the models SZ612‒(R)-1 and SZ616‒(R)-1. The docking of substrate to the substrate binding pocket was carried out by Glide42 and Induced fit docking (IFD)43.

MD simulations and QM/MM calculations

All of the MD runs were performed by using the GPU version of AMBER 201644. Apo-proteins was protonated at pH 7.4 (to be consistent with the experimental conditions) by H++45. The Amber ff14SB force field46 and the TIP3P water model47 were employed, and the solvent layer was set as 10 Å in MD runs. The parameters and charges for substrates (R)-1 were generated with the antechamber module using the AM1-BCC charge model and the general AMBER force field (GAFF)48. After proper parameterizations and setup, the resultant system’s geometries were minimized in two-stage, the first stage only focused on the positions of solvent molecules and ions, and the second stage is an unrestrained minimization of all the atoms in the system. And then the system was heated from 0 to 303 K under the NVT ensemble for 50 ps. Subsequently, the systems were maintained for 50 ps of NPT equilibration at 303 K and 1.0 atm, using Langevin-thermostat (ntt=3) with a collision frequency of 2 ps−1 and pressure relaxation time of 1 ps. A weak restraint of 10 kcal mol−1 Å−2 was performed on the protein residues during heat and density equilibrations. The systems were further equilibrated for 1 ns without restraints. For models SZ612‒(R)-1 and SZ616‒(R)-1, the adapted biased 500-ns MD simulations were carried out, the free energy landscapes were calculated at the temperature 1200 K, following the procedures reported in the previous work49. During all MD simulations, the covalent bonds containing hydrogen were constrained using SHAKE algorithm50. The hybrid QM/MM calculations were performed by using GTKDynamo tool (version 1.9.0)51, where QM part is treated using the PM6 semiempirical method52 and the MM part using the AMBER force field. Residues 53, 55, 80, 99, 101, 114, 116, and 132 along with the substrate were chosen as QM region53. The catalytic processes of the two models SZ612‒(R)-1 and SZ616‒(R)-1 to generate Baldwin and anti-Baldwin products were visualized in Supplementary Movies 14.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.