Structure and Mechanism of the Diiron Benzoyl-Coenzyme A Epoxidase BoxB

Abstract

The coenzyme A (CoA)-dependent aerobic benzoate metabolic pathway uses an unprecedented chemical strategy to overcome the high aromatic resonance energy by forming the non-aromatic 2,3-epoxybenzoyl-CoA. The crucial dearomatizing reaction is catalyzed by three enzymes, BoxABC, where BoxA is an NADPH-dependent reductase, BoxB is a benzoyl-CoA 2,3-epoxidase, and BoxC is an epoxide ring hydrolase. We characterized the key enzyme BoxB from Azoarcus evansii by structural and Mössbauer spectroscopic methods as a new member of class I diiron enzymes. Several family members were structurally studied with respect to the diiron center architecture, but no structure of an intact diiron enzyme with its natural substrate has been reported. X-ray structures between 1.9 and 2.5 Å resolution were determined for BoxB in the diferric state and with bound substrate benzoyl-CoA in the reduced state. The substrate-bound reduced state is distinguished from the diferric state by increased iron-ligand distances and the absence of directly bridging groups between them. The position of benzoyl-CoA inside a 20 Å long channel and the position of the phenyl ring relative to the diiron center are accurately defined. The C2 and C3 atoms of the phenyl ring are closer to one of the irons. Therefore, one oxygen of activated O₂ must be ligated predominantly to this proximate iron to be in a geometrically suitable position to attack the phenyl ring. Consistent with the observed iron/phenyl geometry, BoxB stereoselectively should form the 2S,3R-epoxide. We postulate a reaction cycle that allows a charge delocalization because of the phenyl ring and the electron-withdrawing CoA thioester.

Introduction

The metabolism of aromatic compounds has attracted broad attention due to its importance in the biogeochemical carbon cycle (includes 10–20% of the biomass) and because of the chemistry necessary to overcome the high resonance energy of the conjugated cyclic ring. A central intermediate of the aromatic metabolism is benzoate, which can be degraded by microorganisms using three different strategies. The first strategy in the presence of oxygen uses mono- and dioxygenases to hydroxylate benzoate to catechol or protocatechuate. Central ring-cleaving dioxygenases with mononuclear iron centers subsequently cleave the aromatic ring either between the two hydroxyl groups (ortho cleavage, β-ketoadipate pathway) or next to one of the hydroxyl groups (meta cleavage) (1). The second strategy under anoxic conditions involves benzoyl-CoA, which is reduced by two electrons to a non-aromatic cyclic diene followed by hydrolytic ring opening. Benzoyl-CoA reduction is accomplished by two completely different enzyme systems that may catalyze a Birch-like reduction (2).

The third, semi-aerobic strategy to metabolize benzoate shares characteristic features of both of these options (3). Oxygen is still required for attacking the ring, as in the aerobic metabolism, whereas all metabolites are activated to CoA thioesters and ring cleavage is performed hydrolytically, as in the anaerobic pathway (Fig. 1). According to genome analysis, the pathway is present in 5% of all bacteria (4). Moreover, this strategy not only applies to the metabolism of benzoate but also of phenylacetate (5).

FIGURE 1.

Coenzyme A-dependent aerobic benzoate pathway. This pathway was mainly studied in the β-proteobacterium A. evansii. The enzymatic apparatus for benzoate oxidation is coded by 15 clustered box genes that may form an operon (5).

After benzoate is converted to benzoyl-CoA, the three enzymes BoxABC³ (benzoyl-CoA oxidizing) catalyze the crucial dearomatization reaction by epoxidation and following ring hydrolysis (Fig. 1). The key enzyme is BoxB. This 55-kDa monomeric protein catalyzes the transformation of benzoyl-CoA and oxygen with two electrons and two protons to the non-aromatic 2,3-epoxybenzoyl-CoA and water (4). BoxB is found with a sequence identity higher than 53% mainly in α- and β-proteobacteria. Very low amino acid sequence identities are found to other proteins; the closest relative to BoxB in the database is subunit PaaA of phenylacetyl-CoA epoxidase. This 30-kDa PaaA subunit belongs to a related multicomponent epoxidase system PaaABCDE (6), which was found in 16% of bacteria and converts phenylacetyl-CoA to 1,2-epoxyphenylacetyl-CoA. The crystal structure of the inactive heterotetrameric PaaAC complex has been recently reported in complex with its substrate (PDB code 3PW1) but in absence of the diiron center (7). BoxB is considered as a member of the class I diiron protein family (8) as they contain two characteristic EXXH motifs required for diiron ligation: E-X₂₉-¹⁴⁹EEGRH¹⁵³ and D-X₂₈-²³⁹EEAHH²⁴³. Amino acids in bold letters indicate the putative ligands of the diiron center (9), which is embedded in a central, ferritin-like four-helix bundle. Class I diiron centers appear to share a common O₂ activation mechanism (10) but catalyze diverse oxygen-dependent reactions. Such reactions are hydroxylation (soluble methane diiron monooxygenase and toluene/o-xylene diiron monooxygenase), aryl amine oxidation (p-aminobenzoate N-oxygenase), radical cofactor generation (ribonucleotide reductase R2), double bond formation (stearyl-ACP Δ9 desaturase), and epoxidation.

BoxA is a homodimeric reductase (46-kDa subunit), which hosts one FAD and two [4Fe-4S] clusters per subunit and transfers two electrons from NADPH to the diiron center of BoxB (9, 11). Although the BoxAB genes are clustered and cotranscribed (5), BoxA does not copurify with BoxB and is present in the cell in much lower concentration than BoxB. Only substoichiometric amounts of BoxA are required for benzoyl-CoA epoxidation. These findings argue for a transient interaction between BoxA and BoxB and the existence of docking sites on both proteins to position the distal [4Fe-4S] center of BoxA in close vicinity to the diiron center of BoxB. Homodimeric BoxC (61-kDa subunit) is a dihydrolase. It catalyzes the transformation of 2,3-epoxybenzoyl-CoA with two molecules of water to the non-cyclic 3,4-dehydroadipyl-CoA semialdehyde and formic acid (Fig. 1) (4). The BoxB-catalyzed epoxidation is slow and incomplete due to product inactivation. The addition of BoxC stimulates the epoxidation rate 7-fold and prevents inactivation of BoxB (12). This strong effect of BoxC on the catalytic performance of BoxB and copurification of BoxC with BoxB strongly suggest a BoxB-BoxC interaction in vivo.

BoxB is a widespread bacterial enzyme in a new type of coenzyme A-linked aerobic aromatic metabolism. Our goal was to shed light onto the dearomatization and epoxidation reaction on an atomic basis. Here we describe the crystal structure of BoxB at high resolution including its diiron center and the binding of benzoyl-CoA. The observed redox- and substrate-dependent changes of the ligand pattern and the spatial arrangement of the substrate relative to the diiron center provide new insights into the reaction of BoxB and of diiron enzymes in general.

EXPERIMENTAL PROCEDURES

Materials

⁵⁷Fe metal powder (95.38% isotopic enrichment) was obtained from Campro Scientific (Berlin, Germany). Benzoyl-CoA was prepared according to published procedures (3), purified by reverse-phase HPLC in 40 mm ammonium acetate, pH 6.8 (4), and freeze-dried.

Cultivation of Bacteria

Azoarcus evansii KB740 (DSMZ6869) harboring chromosomally a C-terminal streptavidin affinity-tagged BoxB was grown aerobically with benzoate as the sole source of cell carbon and energy (3). The yield was 200 g of cells (wet mass) mol⁻¹ benzoate. Cells for Mössbauer measurements were grown with 200 mg of ⁵⁷Fe to an optical density at 578 nm of 2, which corresponded to 0.5 g of cells (dry mass) liter⁻¹.

Protein Purification

Cell extracts from 100 g of cells were prepared according to Rather et al. (4). BoxA was purified and assayed as reported previously (11). Streptavidin affinity-tagged BoxB was purified in an anaerobic glove box under N₂/H₂ atmosphere without reducing agents (BoxB_ox) by affinity chromatography and assayed according to Rather et al. (4). BoxB and BoxC copurify. To quantitatively remove BoxC, BoxB_ox was further purified under anaerobic conditions by gel permeation chromatography (Amersham Biosciences, HiLoad 26/60, 320 ml, 2 ml min⁻¹) and stored at −70 °C in 10 mm Tris/HCl, 100 mm KCl, pH 8.0, and 10% (v/v) glycerol (0.7 ml, 22 mg ml⁻¹) for crystallization. Protein concentration and purity were determined by standard methods. Protein activity was determined according to Rather et al. (4).

Determination of Metal Content

4.4 mg of BoxB_ox in 1.2 ml of 10 mm Tris/HCl, 150 mm KCl, pH 8.0, without glycerol was freeze-dried and analyzed via inductively coupled plasma-optical emission spectrometry (ICP-OES) at the Chemical Analysis Laboratory, University of Georgia, Athens, GA. As control, 1.2 ml of 10 mm Tris/HCl, 150 mm KCl, pH 8.0, was treated the same way and analyzed.

Mössbauer Measurements

⁵⁷Fe-labeled protein was purified only by affinity chromatography. Sample preparation was performed at 4 °C under anaerobic conditions. A sample of 600 μl of BoxB_ox (23.5 mg ml⁻¹, 0.427 mm in 10 mm Tris/HCl, pH 8.0, 2.5 mm desthiobiotin, and 10% (v/v) glycerol) was transferred into a Mössbauer cup, immediately frozen, and stored in liquid nitrogen. Mössbauer data were recorded with a conventional spectrometer with alternating constant acceleration of the source. The minimum line width was 0.24 mm s⁻¹ (full width at half-height). The sample temperature was maintained constant with an Oxford Instruments Variox cryostat. The γ-source (⁵⁷Co/Rh, 1.8 GBq) was kept at room temperature, and isomer shifts are quoted relative to the iron metal at 300 K.

Crystallization

Prior to crystallization, the protein was concentrated to about 30 mg ml⁻¹ in a Vivaspin concentrator and washed with 10 mm Tris, pH 8.0, 0.1 m KCl or 10 mm PIPES, pH 7.0, 0.1 m KCl. Crystallization attempts were performed with the hanging drop method at 18 °C using a ratio of precipitant to protein solution of 1 μl:1 μl. BoxB_ox-phosphate, BoxB_ox-malonate, and BoxB_ox-peg crystals were grown aerobically. For co-crystallization experiments, benzoyl-CoA was added to the protein solution prior to crystallization to a final concentration of 5 mm. BoxB:BCA_red crystals (BoxB with the substrate benzoyl-CoA under reducing conditions) were grown in an anaerobic glove box under N₂/H₂ atmosphere. For reduction, BoxA, NADPH, and benzoyl-CoA were supplemented to the BoxB solution to a final concentration of 0.1 mg ml⁻¹, 1 mm, and 5 mm, respectively. The residual oxygen in the protein solutions of BoxA and BoxB was removed by incubation in the glove box for 3 weeks. NADPH and benzoyl-CoA solutions were freshly prepared with anaerobic PIPES buffer. The crystallization and cryoprotectant conditions are given in supplemental Table S1.

X-ray Structure Analysis

Data were collected at the Swiss Light Source (SLS) beamline PXII using an MAR CCD225 detector and processed with XDS (13) and the CCP4 program suite (14). The data quality was examined with PHENIX (15). The initial structure was determined with a BoxB_ox-phosphate crystal using the single wavelength anomalous dispersion method based on the two intrinsic iron ions (see Table 1). The positions of the iron atoms were detected with SHELXD (16), and the phases were subsequently calculated with SHARP (17). After solvent flattening (18) of the electron density, an initial model was automatically built with BUCCANEER (19), which could be manually completed with COOT (20). The refinement was carried out using REFMAC5 (21) and PHENIX (15). The structures of the other crystal forms were solved with the molecular replacement method using PHASER (22), and the refinement was carried out with PHENIX and REFMAC5. The optimal segmentation for TLS refinement of the different structures with REFMAC5 and PHENIX was obtained using the TLSMD server (23). The BoxB structures were nearly identical, and the root mean square deviations varied between 0.3 and 0.4 Å but revealed differences in the active site region. The homology model of the [4Fe-4S]-containing domain of BoxA and the homology model of the BoxC dimer were obtained using Protein Data Bank (PDB) entries 2FDN and 2W3P, respectively, and the SWISS-MODEL server (24). The structures were analyzed using MolProbity (25). The different crystal contact regions within the different crystal forms and the small contact area confirmed the presence of BoxB as a monomer. Fig. 2 and supplemental Figs. S1, B and C, S2A, S3, and S4 were prepared with PyMOL (Schrödinger, LLC). Supplemental Fig. S2B was prepared with CHIMERA (26). Crystal parameters, data collection, and refinement statistics are given in Table 1.

TABLE 1.

Data collection, and refinement statistics

The values of the highest resolution shell are given in parentheses.

	BoxBox-phosphate peak	BoxBox-phosphate native	BoxBox-malonate	BoxBox-peg	BoxB:BCAred
Data collection statistics
Space group	C2221	C2221	C2221	C2	C2
Wavelength (Å)	1.738	0.9999	1.000	1.000	1.2137
Resolution (Å)	50.0-2.7 (2.8-2.7)	50.0-2.1 (2.2-2.1)	50.0-1.9 (2.0-1.9)	50.0-2.5 (2.6-2.5)	50.0-2.3 (2.5-2.3)
No. of unique reflections	88,005 (8511)	96,760 (12,375)	124,905 (15,532)	76,352 (8215)	98,025 (21,689)
Redundancy	4.0 (3.5)	3.5 (3.2)	5.2 (4.3)	3.8 (3.7)	3.6 (3.8)
Unit cell
a, b, c [Å]	113.5, 214.6, 139.0	112.3, 213.9, 139.0	112.8, 213.5 137.2	209.7, 76.8, 147.8	208.2, 76.6, 148.2
β [°]				109.4	108.3
Number of molecules in the asymmetric unit	2	2	2	4	4
Completeness (%)	97.7 (91.8)	99.2 (98.3)	96.2 (84.7)	99.0 (96.3)	99.2 (99.9)
Rsym [%]	11.0 (68.3)	7.5 (67.7)	9.4 (71.2)	9.4 (71.2)	6.4 (40.7)
I/σ(I)	10.8 (2.7)	11.8 (2.3)	13.3 (2.7)	8.6 (1.8)	12.1 (3.4)
Refinement statistics
Ligands per monomer		2 Fe3+, 2 PO43−, 4 glycerol	2 Fe3+, 1 malonate, 0.5 Cl−	2 Fe3+, 1 Cl−, 1 SO42−	Fe2+, Fe2+/3+, 1 Cl−1 benzoyl-CoA, 2 SO42−, 0.5 glycerol, 1 propanediol, 0.25 tetraethylene glycol
Rfree/Rcryst (%)		20.3 (35.9)/17.2 (29.3)	21.7 (48.6)/18.2 (46.1)	23.5 (42.0)/20.6 (40.0)	22.4 (33.0)/17.4 (25.0)
Ramachandran favored/allowed/disallowed (%)		90.9/100/0	91.9/100/0	91.5/100/0	91.1/100/0
Average B-factor protein/solvent/iron/substrate [Å2]		43.2/46.6/34.7/—	28.7/41.0/22.2/—	52.7/44.2/45.2/—	52.6/47.6/71.1/68.4
r.m.s.d.a
Bond length		0.011	0.012	0.015	0.020
Bond angle		1.096	1.205	1.317	1.724

^a r.m.s.d., root mean square deviation.

FIGURE 2.

Structural data for BoxB. A, overall fold. The central feature of the essentially helical BoxB structure is the strictly conserved four-helix bundle (in red) built up by two pairs of anti-parallel helices B + C and E + F. The moderately conserved helices are shown in orange, and the variable parts are shown in blue. C-term, C terminus; N-term, N terminus. B, the structure of the diiron center of the BoxB_ox-malonate state (carbons in yellow) and superimposed BoxB:BCA_red state (carbons in gray). The carboxylate oxygens of Glu-150 bridge the two iron atoms in the oxidized but not in the reduced state. C, the substrate binding site shown as surface representation. The contour level of the electron density is 1.0 σ. Despite the considerable size of benzoyl-CoA and multiple protein-substrate interactions, only small substrate-induced conformational changes of the protein matrix occur. D, architecture of the active site in BoxB:BCA_red. O₂ and the epoxide are modeled in gray. The O₂ binding site in BoxB is surrounded by the non-polar atoms of Glu-120, Glu-150, Asp-211, and Glu-240 and by the phenyl group of benzoyl-CoA after its binding.

RESULTS AND DISCUSSION

Spectral Characterization of the Oxidized Diiron Center of BoxB

Recombinant BoxB of A. evansii was produced as a C-terminal streptavidin affinity-tagged variant and purified under anaerobic conditions without reducing agents (BoxB_ox) (4). The metal content of BoxB_ox was determined by inductively coupled plasma-optical emission spectrometric measurements, revealing two irons per BoxB monomer (2.1 ± 0.1). Because the oxidized protein is colorless (9), a Rieske-type [2Fe-2S] cluster or heme irons could be excluded. Zero-field Mössbauer studies of ⁵⁷Fe-labeled BoxB_ox showed a Lorentzian quadrupole doublet with isomer shift δ = 0.49 mm s⁻¹, electric quadrupole splitting ΔE_Q = 0.69 mm s⁻¹, and line width Γ_FWHM = 0.65 mm s⁻¹. These values are typical of high spin Fe^III in an octahedral coordination shell with hard ligands such as nitrogen or oxygen (from carboxylates or water). Similar values were found for other non-heme diiron enzymes in the oxidized state (27). Interestingly, the low electric quadrupole splitting, being in the normal range for ferric compounds, is consistent with the presence of one or more μ-hydroxo groups coordinated to the diiron center, but it particularly rules out a bridging μ-oxo group. In contrast, complexes with the corresponding Fe^III–O–Fe^III core show much larger quadrupole splitting in the range 0.9–2.4 mm s⁻¹ (except diferric porphyrins and other systems with macrocyclic and highly covalent ligand systems). This is due to the pronounced charge anisotropy caused by the uniquely short iron-oxo bond (28, 29).

Structural Basis

BoxB was crystallized aerobically in the diferric oxidation state BoxB_ox. Three different crystal structures were obtained and termed BoxB_ox-phosphate, BoxB_ox-malonate, and BoxB_ox-peg based on the precipitants phosphate, malonate, and partly polyethylene glycol (PEG) contributive in iron ligation (supplemental Table S1). In addition, BoxB was co-crystallized with the substrate benzoyl-CoA under anaerobic and reducing conditions by supplementing the enzyme solution with NADPH and catalytic amounts of BoxA inside an anaerobic glove box. EPR spectroscopic measurements also using the physiological NADPH/BoxA system for reduction resulted in a semireduced Fe^IIFe^III center.⁴ In vitro the fully reduced Fe^IIFe^II center was only achieved with the stronger reducing agent dithionite in the presence of methyl viologen.⁴ Because of the deviating conditions concerning solution composition, time, and radiation used for x-ray and EPR studies, we cannot specify whether the diiron center in the crystal is present in the semireduced Fe^IIFe^III or the completely reduced Fe^IIFe^II state. We named the corresponding structure BoxB:BCA_red. Interestingly, BoxB in the diferric oxidation state (BoxB_ox-peg) did not co-crystallize with benzoyl-CoA under non-reducing conditions despite similar crystallization conditions as used for growing the BoxB:BCA_red crystals (supplemental Table S1). This suggests that benzoyl-CoA can only be bound to the reduced state, as reported for the substrate methane and the oxygenase component of methane monooxygenase (30). The structure of BoxB was determined in the BoxB_ox-phosphate state using the single wavelength anomalous dispersion method based on the anomalous signal of the two intrinsic iron atoms. The BoxB structures of the other crystal forms were obtained by the molecular replacement method. The BoxB_ox-phosphate, BoxB_ox-malonate, BoxB_ox-peg, and BoxB:BCA_red structures were refined at 2.1, 1.9, 2.5, and 2.3 Å, respectively. The detailed composition of the individual states and the corresponding crystallographic parameters are summarized in Table 1. The extracted structural information is based on average values of the different molecules in the asymmetric unit.

Overall Structure

BoxB is present in solution (9) and in the crystal as a monomer (Fig. 2A). In comparison, other family members such as stearyl-ACP Δ9 desaturase, ribonucleotide reductase R2, and methane monooxygenase are homodimers, heterotetramers, and heterohexamers, respectively (8, 31, 32). BoxB is architecturally composed of eight largely parallel α-helices complemented by several short helices, irregular elements, and three small sheets (Fig. 2A and supplemental Fig. S1A). The core of the enzyme is formed by the canonical four-helix bundle (helices B, C, E, and F), into which the catalytic diiron center is embedded. Only a small surface region (17 × 25 Å) of the four-helix bundle in front of helix F is directly accessible to the bulk solvent. We postulate this region as docking site for the iron-sulfur flavoprotein BoxA to transfer electrons from NADPH to the diiron center of BoxB (see supplemental Fig. S2). The residual surface of the four-helix bundle is encircled by helices A, D, G, and H, an N-terminal β-hairpin, and an irregular stretch proceeding helix B.

Comparison with Structures of Other Diiron Proteins

A superposition between BoxB and other enzymes of the class I diiron protein family indicated a high structural relationship of the four-helix bundles despite very low sequence similarity. According to the Dali server (33), the root mean square deviations between BoxB (106:163, 201:264) and methane monooxygenase (1FZ1, 100:157, 198:262), toluene/o-xylene diiron monooxygenase (1T0Q, 90:147, 186:250), stearyl-ACP Δ9 desaturase (1AFR, 91:156, 186:249), ribonucleotide reductase (1SYY, 75:133, 182:248), p-aminobenzoate N-oxygenase (3CHH, 90:149,188:248), and PaaA (3PVT, 28:85, 116:175) were 2.4, 2.4, 3.1, 2.4, 3.4, and 1.7 Å, respectively. The structure-based sequence identity was 12, 17, 16, 12, 12, and 19%, respectively. Helices D, G, and H flanking the four-helix bundle are, in principle, found in all family members, although their length and orientation differ (Fig. 2A and supplemental Fig. S1A).

The Oxidized Diiron Center

In the diferric BoxB_ox-malonate, BoxB_ox-phosphate, and BoxB_ox-peg structures, both iron sites are highly occupied, as reflected by the low temperature factor of the irons (Table 1), which are comparable with those of the surrounding amino acids. The diiron center is anchored to the conserved residues of the four-helix bundle, as illustrated in Fig. 2B and supplemental Fig. S3 (with electron density) for first sphere interactions and in supplemental Fig. S1B for second sphere interactions.

In the oxidized substrate-free BoxB structures, the two irons of the diiron center have a distance of ∼3.5 Å and are octahedrally ligated. Fe1 is coordinated to Glu-120-O_ϵ1 and His-153-N_δ1, and Fe2 is coordinated to Asp-211-O_δ2 and His-243-N_δ1 in a nearly symmetric manner. The third ligand of Fe1 in the BoxB_ox-malonate state, and presumably in all oxidized states, is a solvent molecule, which was assigned as H₂O. H₂O appears to be more suitable than a hydroxo group for forming hydrogen bonds to the adjacent carboxylate oxygens of Glu-120 and malonate. The equivalent position at Fe2 is occupied by Glu-240-O_ϵ1, providing a clear asymmetry between the two iron ligands. In the oxidized states, the conformation of the side chain of Glu-240 is more variable than that of the other iron ligands; this variability is reflected in an increased temperature factor and the absent interaction between its carboxylate group and the polypeptide scaffold (supplemental Fig. S1B). Moreover, both irons are bridged by an oxo/hydroxo group, two Glu-150 carboxylate oxygens, and oxygens of the precipitant/solvent. A bridging hydroxo group is more likely than an oxo group, as indicated by the microenvironment (Glu-150, Glu-240, and malonate carboxylates), by the distance of ∼2.0 Å to the irons, best observable in the BoxB_ox-malonate structure (Fig. 2B), and by the Mössbauer spectroscopic data.

In the BoxB_ox-malonate and BoxB_ox-phosphate structures, the two malonate carboxylate/phosphate oxygens of the precipitants bridge the two irons in a bidentate manner and complete the distorted octahedral coordination shell of both irons (Fig. 2B and supplemental Fig. S3). We assume that O₂ binds between the two bridging oxygens (Fig. 2D). Correspondingly, in the BoxB_ox-peg structure, the potential O₂ ligation site was partly occupied. The density was tentatively interpreted as a PEG fragment and/or a hydroxo group.

The Reduced Diiron Center with Bound Benzoyl-CoA

In the BoxB:BCA_red structure, the irons are also ligated to Glu-120-O_ϵ1, His-153-N_δ1, Asp-211-O_δ2, and His-243-N_δ1, but the iron-ligand distances are significantly longer when compared with those found in the oxidized BoxB structures (supplemental Table S2). The carboxylate of Glu-240 appears to be ligated to Fe2 in a bidentate manner. The solvent molecule connected to Fe1 is not visible, but its presence in the BoxB:BCA_red state cannot be completely excluded at 2.3 Å resolution. Despite the similar distance of 3.5 Å between the two irons of the oxidized and reduced states, the irons do not superimpose exactly; in particular, Fe2 is displaced by ∼0.7 Å (Fig. 2B). In the BoxB:BCA_red state, the irons have high temperature factors (71 Ų) relative to the surrounding protein (40 Ų). This could be due to increased iron-ligand distances, a possible heterogeneity effect, and the surprising absence of ligands bridging the irons.

In the oxidized states, the non-polar C_β and C_γ atoms of Glu-150 contact the Val-146 peptide oxygen and a solvent molecule (supplemental Fig. S1B). Induced by reduction and substrate binding, the side chain of Glu-150 is rotated away from the diiron center. Thus, the Glu-150 carboxylate group in the BoxB:BCA_red state predominantly occupies the same polar binding site as the solvent molecule in the oxidized states and is hydrogen-bonded to Ser-123, Gln-127, and Asn-147. Because weak residual electron density (20–40% occupancy) is present at the same position as in the oxidized states, Glu-150 seems to occupy a split conformation in the BoxB:BCA_red structure (supplemental Fig. S3C).

Comparison of Diiron Ligation

The diiron ligation structure of BoxB is well conserved in all class I diiron proteins, implying a related oxygen activation mechanism (34). BoxB_ox-malonate and methane monooxygenase_ox (35) (and toluene/o-xylene diiron monooxygenase_ox (36)) only reveal minor differences in the ligation pattern. The carboxylate chelated to Fe2 originates from an aspartate in BoxB (Asp-211) and PaaAC (7) and from a glutamate in the other family members. This change might be correlated with the different π-helix character of helix E (one π-helix turn in BoxB and PaaA and two in methane monooxygenase (32)). The variable character of Glu-240 described for BoxB was also found in other family members. For example, one carboxylate oxygen bridges the two irons in the reduced state of methane monooxygenase, in contrast to two carboxylate oxygens in stearyl-ACP Δ9 desaturase (32, 37). The different side chain arrangement of Glu-150 in the oxidized and reduced substrate-bound state is unique in class I diiron center proteins. The Fe–Fe distances are rather variable among the family members; the value for BoxB is in this range.

The Substrate Binding Site

The approximate position of the benzoyl-CoA binding site can be assigned on the basis of a protein surface analysis as a 20 Å long channel extends from the protein surface to the diiron center in the protein interior (Fig. 2C). The overall architecture of the channel is formed by helices B and E, by the C-terminal end of helix G, by the segment following helix D, and by the small helix 303:313. It most resembles that of PaaA (7) and next that of stearyl-ACP Δ9 desaturase (37). Because of the longer stearyl chain of the latter, the substrate channel protrudes deeper into the protein than in the case of BoxB. The channel is shortened by the side chains of Ser-123 and Phe-206 in BoxB, and its exit is displaced by a positional change of helix G and the following segment when compared with stearyl-ACP Δ9 desaturase. Interestingly, the channel entrance between helices B and E is smaller in BoxB, PaaA, and stearyl-ACP Δ9 desaturase than in methane monooxygenase or toluene/o-xylene diiron monooxygenase, although BoxB, PaaA, and stearyl-ACP Δ9 desaturase use larger substrates. In methane monooxygenase and toluene/o-xylene monooxygenase, longer and more hydrophobic side chains define the channel entrance (34). In agreement with the polar character of CoA, the channels of BoxB and PaaA are the most hydrophilic ones.

The benzoyl-CoA binding mode was accurately established based on the 2.3 Å BoxB:BCA_red structure (Fig. 2C). The benzoyl-CoA site is highly occupied and clearly defined in the electron density except the phosphorylated adenosine moiety of CoA, which is partially disordered. This moiety positioned at the solvent-accessible entrance of the channel might be influenced in the current crystal form by contacts to a neighboring molecule (supplemental Fig. S1C). Nevertheless, benzoyl-CoA adopts the characteristic J-shaped conformation with a bent adenosine and a small kink after the thioester group. The catalytic relevant phenyl ring is accommodated by a rather hydrophobic pocket, which is built up of Thr-119, Ser-123, Phe-193, Phe-206, and Asp-211 (supplemental Fig. S1C). Substrate binding in BoxB and PaaA (7) is related, although the contacting residues are only moderately conserved. In BoxB, the phenyl ring is buried less deeply inside the channel than in PaaA, perhaps due to the presence of the diiron center.

Although a few diiron center protein-substrate (analog) complexes were structurally characterized (7, 36, 38), the BoxB:BCA_red structure reveals the first one where the substrate is bound to its binding site in a catalytically relevant orientation relative to the diiron center. The shortest distance between the phenyl ring of benzoyl-CoA and the diiron center is 3.5 Å between Fe1 and C2, whereas the corresponding distance to Fe2 is 4.9 Å (Fig. 2D). The residue Gln-116, which is also involved in malonate binding, is hydrogen-bonded to the thioester oxygen and cysteamine nitrogen of benzoyl-CoA and thereby determines the position of the phenyl ring. The benzoyl thioester group snugly fits into the narrowest segment of the channel with the smallest diameter of ∼6.5 Å between Thr-119 and Gly-214 (Fig. 2C and supplemental Fig. S1C). Therefore, the chemical reaction is completely shielded from bulk solvent.

Proposed Association of BoxB with BoxA and BoxC

The catalytic machinery that epoxidizes benzoyl-CoA and hydrolyzes the epoxide consists of the three proteins BoxA, BoxB, and BoxC, which somehow must be associated in vivo. Our postulated molecular architecture is reminiscent of other multicomponent monooxygenases regarding basic principles of electron supply, catalysis, and stimulation, with the exception that BoxC also catalyzes the subsequent ring cleavage of the epoxide (Fig. 1). For creating a complete picture of the BoxABC system, we tentatively modeled the transient BoxAB and BoxBC associations. In the BoxAB model, the distance between the distal [4Fe-4S] center of BoxA and the diiron center of BoxB is 13.5 Å and thus allows electron transfer at physiological rates (supplemental Fig. S2). In the BoxBC model, the substrate binding site of one BoxC monomer is close to that of BoxB. The association between BoxB and the ring-cleaving dihydrolase BoxC (39) may prevent an undesirable release of the reactive epoxide and may guarantee its quick removal by dihydration (supplemental Fig. S4).

Proposed Catalytic Cycle

The combination of structural and spectroscopic data provides new insights into the catalytic mechanism of BoxB outlined in Fig. 3. The resting state (Box_ox) is defined as ferric diiron complex bridged by two μ-hydroxo groups (Fe1^III–(OH)₂–Fe2^III) as found also for other multicomponent monooxygenases. In most BoxB_ox structures, one bridging hydroxyl group is replaced by a ligand from the crystallization solution. According to EPR data,⁴ the reaction cycle is initiated by reducing BoxB with BoxA/NADPH to a mixed-valent Fe1^IIFe2^III state, which is in agreement with theoretical studies on methane monooxygenase (40). In parallel, one or both bridging hydroxo groups are released as H₂O. Upon reduction, the interactions between the diiron center and the surrounding polypeptide are decreased, and the (positional) mobility of the irons is increased, suggesting a fluctuating Fe–Fe distance during the catalytic cycle. Benzoyl-CoA binding (BoxB:BCA_red) and the weaker iron-ligand interactions in the reduced state induce a rotation of Glu-150 away from the hydrophobic phenyl group to a more polar position (Fig. 2, B and C). We suggest that benzoyl-CoA binding and the displacement of Glu-150 allows O₂ to move to a position that is favorable for activation (Fig. 2D). O₂ activation induced by transferring the second electron would be directly coupled to the reaction, which is useful as the production of reactive oxygen species and the waste of reducing power are thus prevented.

FIGURE 3.

Proposed mechanism for the BoxB reaction. The reductase BoxA transfers electrons from NADPH to the resting state of BoxB (Box_ox). Upon reduction, Glu-150 rotates and allows BoxB to bind its substrate benzoyl-CoA (BoxB:BCA_red). The diiron center might be in a semireduced Fe^IIFe^III or a completely reduced Fe^IIFe^II state. As proposed for other diferric monooxygenases, O₂ might be activated via the μ-peroxo-diferric state (P_*) and the diamond core with substrate (QS) (42, 48). The terminal state (T) with the epoxide adduct is formed either via an open core radical intermediate (R) in a radical rebound mechanism or via an alternative, concerted mechanism.

Although the exact geometry of the active site prior to the oxygen attack remains elusive, the spatial relationship between the irons and the substrate, so far not reported for other class I diiron enzymes, provides several mechanistic clues. Thus, the structural data suggest that the attacking oxygen atom must be approximately located both between and in contact to the C2 and C3 atoms of benzoyl-CoA and in ligation to the diiron center. The finding that C2 and C3 are closer to Fe1 than to Fe2 suggests that the symmetrical peroxo Fe1^III–O–O–Fe2^III species (P*) is converted to the two μ-oxo-bridged diferryl irons (Fe1^IV(μ–O)₂Fe2^IV) intermediate QS (41) where the attacking oxygen atom of activated O₂ is predominantly bound to Fe1 before and during epoxide formation (Fig. 2D). This is in agreement with the “diamond core” intermediate QS of methane monooxygenase, which has one short (∼1.77 Å) and one long (∼2.05 Å) Fe–μ–O bond (42). QS can also be viewed as a head-to-tail dimer of Fe^IV=O units (Fe^IV=O)₂. Furthermore, biomimetic bis-μ-oxo diiron complexes can also open to an “open core” O=Fe1^IV–O–Fe2^III intermediate (R), for which the terminal oxo group is more reactive than the bridging oxo group (43, 44). Therefore, we suggest that the terminal oxo group attacks the C2 atom of the phenyl ring.

This attack may proceed via a radical mechanism or alternatively via a concerted non-radical mechanism (45) (Fig. 3). An electrophilic addition would be destabilized by the CoA thioester substituent, and a nucleophilic addition to an electron-rich aromatic ring is rather unusual. We favor a radical mechanism as the Fe^IV/III redox potential is higher than +2 V versus the normal hydrogen electrode (46), and the transfer of one electron to the aromatic ring of benzoyl-CoA would require −1.9 V (47). Furthermore, the electron-withdrawing thioester group allows extended charge delocalization, which facilitates the oxygen attack to the aromatic ring and stabilizes the intermediate.

The epoxide ring is subsequently closed to the 2,3-epoxybenzoyl-CoA adduct (T). The relative orientation of benzoyl-CoA and the diiron center clearly determine the side of the phenyl ring to which the generated oxygen must point (Fig. 2D). According to these geometric considerations, BoxB should stereoselectively produce the 2S,3R-epoxide. Finally, the enzyme releases the product and returns to the resting state (BoxB_ox).