Main

Epoxides are highly versatile building blocks for the synthesis of many organic molecules. In the presence of strong Lewis or Brønsted acids, epoxides isomerize to carbonyl compounds by Meinwald rearrangement1, which generates multifunctional active aldehyde and ketone intermediates that are widely applied in fine chemical and pharmaceutical syntheses. However, this chemical reaction usually requires the use of corrosive acids as catalysts and protective anhydrous and inert atmospheric conditions, as the substrates are moisture- and air-sensitive2. Furthermore, the Meinwald rearrangement suffers from low yields due to side reactions and poor regio-selectivity and stereo-specificity, typically resulting in the formation of a mixture of compounds3.

In contrast, styrene oxide isomerase (SOI), an integral membrane protein found in the styrene-degradation pathway of microbes4,5,6,7, catalyses the isomerization of aryl epoxides such as styrene oxide derivatives to carbonyl compounds under physiological conditions by Meinwald rearrangement. SOI possesses several characteristics that make it an attractive enzyme for biocatalytic applications and an alternative to chemical synthesis. First, there are very few unique enzymes that can catalyse the isomerization of an epoxide to an aldehyde. Apart from SOI8,9 and quinolone epoxide rearrangement protein (PenF)10 from the penigequinolone synthesis pathway, no other enzymes have been reported to catalyse Meinwald rearrangements. Second, SOI has a broad substrate range and may therefore be used for the production of a variety of organic molecules, including new compounds (Extended Data Fig. 1a) for synthetic applications5). The R enantiomer has been shown to react more quickly, but the difference is too small to be rationalized by static structural considerations. Similarly, the ring-opened sp2-carbocations of both enantiomers, assumed to be stabilized by co-planarity with the aryl ring and with their oxygens still ligated to the haem iron, can be modelled without causing substantial repulsive interactions and maintaining the approximate mirror relationship. Fixed in this relatively rigid conformation, only one of the two hydrogens of Cβ of the epoxide is in a favourable position to shift to Cα, explaining the high specificity of the reaction and the conserved chirality at Cα. At the same time, it explains why internal epoxides methylated at Cβ, such as (1R,2R)-2-methyl-3-phenyloxirane, gave only one diastereomer by shifting the methyl group in the trans-position, but not the other product by transferring the hydrogen in the cis-positionIdentification of the ferric haem b prosthetic group

XANES data for SOI was collected at the Swiss Light Source at beamline SuperXAS, using a Si(111) monochromator, Si mirrors for collimation and harmonic rejection, a toroidal Rh-coated mirror for focusing, ionization chambers for incident intensity detection and a five-element silicon drift detector (SDD) for X-ray fluorescence measurements. A 200–300-μl sample of 3 mg ml−1 SOI was filled into a 4–5-cm kapton capillary (diameter of 2 mm) using a Hamilton syringe. The following measurement strategy was used. First, a series of XANES spectra were measured to check the influence of X-ray-induced damage at room temperature After that, the sample was translated to fresh sections multiple times, and scans were performed during the time when the X-ray-induced damage was small. Measured XANES spectra were compared with theoretical XANES calculations performed using finite-difference method near-edge structure (FDMNES) code using the full multiple scattering method38. Our data show a mixture of 5- and 6-coordinated SOI at room temperature in solution (Extended Data Fig. 2). The fifth axial ligand of iron is a nitrogen atom of histidine with ~2.3 Å and the sixth distal ligand of iron can be an oxygen or nitrogen atom.

EPR spectroscopy

The EPR samples contained 305 µM SOI, 545 µM SOI–NB complex or 309 µM SOI Y103A in 200 µl of buffer consisting of 20 mM HEPES, 150 mM NaCl and 0.03% DDM pH 7.5, unless stated otherwise. Reduced samples were prepared by addition of 10 mM sodium dithionite in an anaerobic glovebox. Stock solutions of 200 mM styrene oxide and phenylacetaldehyde were prepared in methanol. A 200 mM benzylamine stock solution was prepared in water. Styrene oxide, phenylacetaldehyde and benzylamine were added to a final concentration of 10 mM to the three different SOI samples, and the EPR samples were frozen in liquid nitrogen. EPR spectra were recorded on a Bruker EMXplus spectrometer with a helium-flow cryostat at 17 K (refs. 39,40) using the following EPR parameters: microwave frequency of 9.410 GHz, microwave power of 20 mW, modulation frequency of 100 kHz and a modulation amplitude of 20 G. The magnetic field was calibrated using the Bruker BDPA (1,3-bis(diphenylene)-2-phenylallyl radical) standard with a g-value of 2.00254 ± 0.00003.

Functional characterization

For isomerization activity measurements, the following coupled enzyme assay was carried out (Extended Data Fig. 5c). In a 1.5-ml cuvette, a 1-ml reaction for (S)-styrene oxide was performed at 25 °C in reaction buffer (0.05 M potassium phosphate buffer pH 8, 0.01% DDM) containing 2 mM NAD+ and 6 U of EcALDH. The change in absorbance at 340 nm was recorded at regular time intervals using a Hitachi U2900 spectrophotometer, and absorbance units were converted to concentration using the extinction coefficient at 340 nm (ɛ340 nm) value of NADH (6.22 Abs mM−1 cm−1). The slope of the concentration–time plot for the first 15 s was calculated to determine activity, with 1 U defined as the activity of SOI that gives 1 µmol of product in 1 min under assay conditions. To determine the kinetic parameters KM and kcat, the isomerization activity was determined with (S)-styrene oxide concentrations of 0.1–5.0 mM. SOI (0.14 μg, 7.26 pmol) or 0.07 μg (3.63 pmol) of SOI–NB complex (containing an equimolar mixture of SOI and nanobody) was added to the assay mixture at 25 °C in reaction buffer (0.05 M potassium phosphate buffer, pH 8, 0.01% DDM) containing 2 mM NAD+, 6 U EcALDH, to start the reaction. The slope of the concentration–time plot for the first 15 s was used to determine the initial rate of each reaction in mM min−1, to obtain a relationship between initial activity and concentration. Each reaction was performed in triplicate.

AlphaFold structure prediction

AlphaFold2 (ref. 22) was implemented by locally running an adapted code written by ColabFold41. All runs used only the sequence of SOI, with no templates or Amber relaxation. We assessed monomeric to tetrameric models.

Reconstitution of SOI in a lipid nanodisc

The lipid nanodisc reconstitution of SOI was performed using E. coli polar lipids and an MSP1D1 nanodisc. The concentrations of SOI for nanodisc reconstitution were in the 300–400 μM range, with a molar ratio of SOI to MSP1D1 to lipid of 3:2:75. Initially, 5 mg of E. coli polar lipid (EPL, Avanti) was dissolved in 100 μl of chloroform and dried under a stream of N2. The resulting lipid film was then mixed with 600 μl of buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl and 1% DDM), and sonicated in a bath sonicator (Bandelin SONOREX SUPER) until the mixture turned translucent. The detergent-solubilized EPL extract was combined with freshly purified SOI (in a 1:25 molar ratio), then gently mixed for 30 min at room temperature. Following this incubation, MSP1D1 was introduced to the protein–lipid mixture and incubated at room temperature for another 30 min. Incorporation of SOI in nanodisc was triggered by adding 300 mg of wet Bio-beads (precleaned with 100% methanol and Milli-Q water). The reconstitution solution was then incubated overnight at 4 °C with gentle mixing. The supernatant was cleared of beads, and the sample was spun before loading onto a Superdex 200 Increase 10/300 GL column (GE Healthcare) equilibrated in 20 mM Tris pH 8.0 and 150 mM NaCl. The peak fractions corresponding to SOI in MSP1D1 (elution volume, 10.3–11.8 ml) were collected, concentrated with a 100-kDa-cutoff Amicon concentrator (Millipore) and used for cryo-EM grid preparation.

Cryo-EM sample preparation and data collection

The cryo-EM samples of SOI–NB were prepared using freshly reconstituted SOI–EPL–MSP1D1 nanodisc complex. SOI–benzylamine was prepared by adding 1 mM benzylamine (100 mM stock concentration in 100% DMSO) to SOI–NB, then incubating the samples on ice for 10–15 min. The final concentration of SOI–NB and SOI–NB–BA complexes used for freezing grids was ~5–7 mg ml−1. The concentration of protein was estimated based on absorbance at 280 nm and a normalized extinction coefficient of 154,685 M−1 cm−1 (considering one molecule of MSP1D1 per three molecules of SOI and nanobody molecule, respectively). All the grids used for cryo-EM grid preparation were freshly glow-discharged in a PELCO easiGlow (Ted Pella) glow-discharge cleaning system for 25 s at 30 mA in air, then 3.5 μl of protein solution was applied to Quantifoil 1.2/1.3 grids (300 mesh), blotted for 3 s with blot force 20, and plunged into liquid ethane using a Vitrobot Mark IV system (Thermo Fisher Scientific) with 100% humidity and an ambient temperature of 4 °C. The frozen grids were stored in liquid nitrogen for subsequent cryo-EM data collection.

Cryo-EM data collection and processing

All the cryo-EM datasets of SOI were collected using EPU software on a 300-kV Titan Krios system (Thermo Fisher Scientific) equipped with a Gatan K3 direct electron detector and a Gatan Quantum-LS GIF, at ScopeM, ETH Zurich. All movies were acquired in super-resolution mode with a defocus range of −0.5 to −3 μm and were binned twofold after acquisition in EPU. The dataset of SOI apo was composed of 8,942 movies with an average dose of 65 e−/Å2 and final pixel size of 0.66 Å. The cryo-EM processing was performed in Relion (version 3.1.3 and 4.0.0)42,43. A flow chart of cryo-EM processing of the SOI–NB complex is provided in Extended Data Fig. 6. In brief, all movie stacks were motion-corrected using MotionCorr2 (version 1.4.0)44, then CTF-corrected using Gctf (version 1.0.6)45. A total of 2,715,264 particles were autopicked, then subjected to several rounds of 2D classifications, yielding 955,410 particles. The best set of 2D classes was then used to generate an initial model to be used in 3D classification. After multiple rounds of 2D and 3D clarifications, a set of 396,760 particles were selected for masked 3D refinement (using masks excluding the nanodisc density), resulting in a 3D reconstruction at a resolution of 2.54 Å. These refined particles were then subjected to CTF refinement and particle polishing, and another round of 3D classification without alignment and with masking of the nanodisc. The particles from the best 3D class were subjected to several iterative cycles of 3D refinement, CTF refinement and particle polishing, yielding a final post-processed density map at a resolution of 2.05 Å.

For the SOI–NB–BA complex, 11,905 movies with 56 e−/Å2 were collected. The data-collection strategy and image processing of the SOI–NB–BA complex were similar to those used for the SOI–NB complex. The 3D projections from the best 3D class from SOI apo were used as templates for autopicking and 3D classification jobs. The detailed steps for processing of the SOI–NB–BA complex are shown in Extended Data Fig. 7. The cryo-EM density features of the transmembrane helices of SOI, nanobodies and ferric haem b are shown in Extended Data Fig. 8. The resolutions of the cryo-EM maps of the SOI–NB complex and SOI–NB–BA complex are shown by Fourier shell correlation (FSC) curves (Supplementary Fig. 1).

Details of cryo-EM data collection and the statistical analysis are shown in Table 1.

Table 1 Cryo-EM data collection and analysis statistics
Full size table

Model building and refinement

As starting models, the AlphaFold22 predicated model of SOI and a homology model of the nanobody generated using SwissModel46 (with PDB 5VAK as template) were docked on the final post-processed density map and manually refined using Coot47. The ligands, the ferric haem b prosthetic group and benzylamine were generated from SMILE codes using the eLBOW program in Phenix48. The structures were finally refined using phenix.real_space_refine48. The quality of the final models was assessed using MolProbity49. Local resolution maps were calculated by ResMap50 implemented in Relion 4.0.0 (ref. 42). All figures were generated using PyMOL 2.5.2 (ref. 51) and ChimeraX52.

Reporting summary

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