Crystal Structure of 3-Hydroxybenzoate 6-Hydroxylase Uncovers Lipid-assisted Flavoprotein Strategy for Regioselective Aromatic Hydroxylation*

Background: 3-Hydroxybenzoate 6-hydroxylase (3HB6H) is a flavoprotein monooxygenase involved in the catabolism of aromatic compounds in soil microorganisms. Results: The enzyme crystal structure features natively bound phospholipids and a Tyr-His pair for substrate binding and catalysis. Conclusion: 3HB6H has a peculiar substrate-binding site that uses a bound lipid to help to discriminate between ortho- and para-hydroxylation. Significance: 3HB6H structure uncovers new flavoprotein strategy for regioselective aromatic hydroxylation. 3-Hydroxybenzoate 6-hydroxylase (3HB6H) from Rhodococcus jostii RHA1 is a dimeric flavoprotein that catalyzes the NADH- and oxygen-dependent para-hydroxylation of 3-hydroxybenzoate to 2,5-dihydroxybenzoate. In this study, we report the crystal structure of 3HB6H as expressed in Escherichia coli. The overall fold of 3HB6H is similar to that of p-hydroxybenzoate hydroxylase and other flavoprotein aromatic hydroxylases. Unexpectedly, a lipid ligand is bound to each 3HB6H monomer. Mass spectral analysis identified the ligand as a mixture of phosphatidylglycerol and phosphatidylethanolamine. The fatty acid chains occupy hydrophobic channels that deeply penetrate into the interior of the substrate-binding domain of each subunit, whereas the hydrophilic part is exposed on the protein surface, connecting the dimerization domains via a few interactions. Most remarkably, the terminal part of a phospholipid acyl chain is directly involved in the substrate-binding site. Co-crystallized chloride ion and the crystal structure of the H213S variant with bound 3-hydroxybenzoate provide hints about oxygen activation and substrate hydroxylation. Essential roles are played by His-213 in catalysis and Tyr-105 in substrate binding. This phospholipid-assisted strategy to control regioselective aromatic hydroxylation is of relevance for optimization of flavin-dependent biocatalysts.

3-Hydroxybenzoate 6-hydroxylase (3HB6H) from R. jostii RHA1 participates in the gentisate pathway where it catalyzes the para-hydroxylation of 3-hydroxybenzoate to gentisate (Fig. 1). We recently established that 3HB6H is a dimeric flavoprotein with each subunit containing a tightly but non-covalently bound FAD.
3HB6H uses NADH as an electron donor and features narrow substrate specificity. Besides 3-hydroxybenzoate, the enzyme converts a limited number of 3-hydroxybenzoate analogs to the corresponding gentisates (14).
Here we present the crystal structure of 3HB6H refined at 1.51-Å resolution, which reveals that the protein harbors natively bound phospholipids. We found that 3HB6H shares a common fold with other flavoprotein hydroxylases and that the domain architecture and topology is closest to family members that catalyze para-hydroxylation reactions. To gain insight into the structural determinants of the regioselectivity of flavoproteindependent 3-hydroxybenzoate hydroxylation, we addressed the mode of substrate binding in 3HB6H and the role of putative active site residues in catalysis. The results uncover a strategy in which a flavoprotein monooxygenase uses a phospholipid chain and properly positioned residues to achieve regioselective para-hydroxylation.

EXPERIMENTAL PROCEDURES
Chemicals-Pfu DNA polymerase, DpnI, and dNTPs were purchased from Invitrogen. Oligonucleotides were synthesized by Eurogentec (Liege, Belgium). Escherichia coli TOP10 was from Invitrogen. Aromatic compounds were purchased from Sigma-Aldrich and Acros Organics (Fair Lawn, NJ). Crystallization kits were purchased from New Hampton (Aliso Viejo, CA). All other chemicals were from commercial sources and of the purest grade available.
Protein Production-3HB6H from R. jostii RHA1 and 3HB6H variants were expressed in E. coli and purified as C-terminal His 6 -tagged protein as described previously (14).
Activity Measurements-3HB6H activity was routinely assayed by following the decrease in absorbance of NADH at 360 nm at 25°C on a Hewlett-Packard 8453 diode array spectrophotometer. Reactions were performed in 50 mM Tris-SO 4 (pH 8.0) and 10 M FAD and started by addition of 46 nM enzyme. Initial rates were calculated using a molar absorption coefficient (⑀ 360 ) of 4.31 mM Ϫ1 cm Ϫ1 (14). One unit of enzyme activity is defined as the amount of enzyme that consumes 1 mol of NADH/min.
Oxygen Consumption-An OxyTherm Clark-type oxygen electrode system (Hansatech, Norfolk, UK) was used to determine the hydroxylation efficiency of 3HB6H variants toward different aromatic substrates. The assay solution (final volume, 1.0 ml) contained 350 M aromatic substrate, 250 M NADH, and 1-10 M enzyme in 50 mM air-saturated Tris-SO 4 (pH 8.0) at 25°C. At the end of the reaction, the amount of hydrogen peroxide produced was determined by adding 20 g of catalase Substrate Binding-The interaction of 3HB6H (25 M) with substrate analogs was studied in 50 mM Tris-SO 4 (pH 8.0). Dissociation constants (K d ) of enzyme-substrate complexes were determined from flavin absorption difference spectra essentially as described elsewhere (14).
Lipid Identification-Identification of protein-bound lipids was performed as follows. Extraction of lipids was achieved by mixing 100 l of buffer-exchanged protein solution (50 mM ammonium acetate (pH 6.8)) with 267 l of chloroform and 133 l of methanol in a glass reaction tube. Once phase separation  was reached, the chloroform phase was extracted using a Pasteur pipette and collected in a new reaction tube. Evaporation of the organic phase was done by flushing with nitrogen gas, and the dry sample was dissolved in 15 l of isopropanol and analyzed. Electrospray ionization-MS or -MS/MS analysis of lipid identification was performed on a Quattro Ultima nanoflow triple quadrupole mass spectrometer (Micromass, Manchester, UK) equipped with a Z-spray nanoelectrospray ionization source. All measurements were performed by operating in the positive and negative ion modes using gold-coated needles made with borosilicate glass capillaries (Kwik-Fill, World Precision Instruments, Sarasota, FL) on a P-97 puller (from Sutter Instruments, Novato, CA). The needles were coated with a gold  Fig. 2). Residues involved in direct FAD contact (yellow circles), chloride contact (green circles), and lipid contact (blue circles) are indicated. Conservation of amino acid residues in fingerprint sequences: identical residues (*), conserved substitutions (:), semi-conserved substitutions (.). 3HB6H and PHBH residues involved in dimerization are in regular font. Shading of amino acid residues depends on their chemical characteristics. Structural alignment was done by STRAP (52) and edited in ClustalW (53). layer, which was performed by an Edwards Scancoat Six Pirani 501 sputter coater (Edwards Laboratories, Milpitas, CA). Mass spectra were recorded with a capillary voltage of 1.3 kV and a cone voltage of 150 V. For MS/MS, argon was supplied in the collision cell (2.0 ϫ 10 Ϫ3 mbar). Collision energy was adjusted to gain optimal fragmentation.

RESULTS
Overall Structure of 3HB6H-The structure of 3HB6H was solved at 1.51-Å resolution using the single wavelength anomalous dispersion method ( Table 1). The final model of 3HB6H monomer (Fig. 2) contains one molecule of FAD, one chloride ion, and a phospholipid ligand. All protein residues (except the N-terminal 1-2 residues and C-terminal 25 residues containing a spacer and the His 6 tag) have well defined ordered conformations. The three-dimensional structure of 3HB6H is similar to that of PHBH and other flavoprotein hydroxylases (1,2). A DALI structural homolog search (Table 2) revealed that despite a relatively low sequence identity (14 -26%) most parts of the protein could be structurally aligned with related flavoprotein hydroxylases (Fig. 3).
FAD-binding Domain-The FAD-binding domain contains a Rossmann fold for binding the ADP moiety of the cofactor. The isoalloxazine ring of FAD is located in the interior of the protein at the interface between the FAD-and substrate-binding domains. FAD contacts around 30 residues along the polypeptide chain (Figs. 3 and 4A), resulting in a tight protein-cofactor interaction. One-third of the protein-FAD contacts are strictly conserved among flavoprotein hydroxylases and cluster around the three fingerprint sequences (29).
For PHBH, it was observed that the isoalloxazine ring of FAD occupies different conformations: in, out, and open (30,31). The out conformation enables flavin reduction, the in conformation ensures substrate hydroxylation during the oxidative half-reaction, and the open conformation allows substrate/ product binding and release. In 3HB6H, the flavin ring has a well defined electron density and was refined with full occupancy in the in conformation (Fig. 4A), similarly to the FAD conformation of the substrate-free forms of 2-methyl-3-hydroxypyridine-5-carboxylic acid oxygenase and 2,6-dihydroxypyridine 3-hydroxylase. Earlier studies with PHBH have indicated that the FAD domain plays a crucial role in NADPH binding (32,33) and that a small helix near the protein surface is the key determinant for the coenzyme specificity (34). The absence of this surface helix in 3HB6H does not explain the NADH preference of the enzyme.
Substrate-binding Domain-The active site of 3HB6H is formed by residues of the FAD-and substrate-binding domains. Although wild-type 3HB6H crystals were prepared with excess 3-hydroxybenzoate in the crystallization drop, no electron density for the aromatic substrate was found in the active site (Fig. 4A). Crystals contain a chloride ion interacting with Gln-301, Ala-304, and Gly-306 at 4-Å distance from flavin C4a.
Dimerization Domain-3HB6H is a dimer both in solution (14) and in the crystal form (Fig. 2B). Among structurally related proteins, PHBH shows similar dimer association. PHBH uses three helices to form a hydrophobic dimerization interface with a few salt bridges (35,36). Compared with PHBH, in 3HB6H, there are fewer intermonomer contacts, and dimer interactions occur mainly via one helix (H12H13* and H13H13*) (Figs. 2B and 3).
Lipid Binding and Identification-Electron density of a lipidlike ligand was observed in all 3HB6H structures (Fig. 4, B and  C). Tentatively, a phospholipid with two chains of 16 carbons was modeled on the basis of the electron density in the difference Fourier map (Fig. 4B). The lipid is inserted in a tunnel, which runs from the protein surface to the active site (Fig. 4C). In the tunnel, the lipid contacts hydrophobic and aliphatic residues. Some interaction occurs between the phospholipid and the opposite monomer (Fig. 2B). From the crystal structure, the exact identity of the phospholipid was difficult to assign as the density of the headgroup is poorly defined. Detailed assignment of the phospholipid was achieved by mass spectral analysis of the low molecular weight components extracted from denatured 3HB6H. The mass spectrum (  (Table 3). Phosphatidylglycerols were visible both in negative and positive modes as complexes with two sodium molecules, whereas phosphatidylethanolamines in the positive mode mostly were found in complex with one sodium ion.
Investigation of Substrate Binding and Catalytic Residues-Putative active site residues and orientation of the substrate were initially inferred from comparison of the native 3HB6H structure with those of other flavoprotein hydroxylases (Fig. 6). Based on these structural comparisons, we hypothesized that residues Tyr-105, His-213, and Gln-301 could be predictably involved in binding of the substrate, especially in the interactions with its carboxylate and hydroxyl groups. To probe the role of these catalytic residues, the enzyme variants Y105F, H213A, H213S, and Q301E were expressed and purified. The mutant proteins were analyzed for their catalytic and binding properties with 3-hydroxybenzoate (Table 4 and Fig. 7). Compared with wild-type enzyme, Y105F showed less affinity for the aromatic substrate with a 14-fold increase in K d (Fig. 7). Weaker substrate binding does not affect the hydroxylation efficiency (Ͼ90% coupling) and turnover rate. Due to a 4-fold higher apparent K m , Y105F has a lower specificity constant (k cat /K m ) than wild-type enzyme. H213A turned out to be inactive. No NADH oxidation was observed under standard assay conditions, although at very high (10 M) enzyme concentrations, NADH was slowly consumed. Intriguingly, the strength of substrate binding is not affected by the His-213 3 Ala replacement ( Fig. 7B and Table 4). H213S and Q301E were also inactive ( Table 4). With these variants, no characteristic perturbations in the absorption spectrum of protein-bound FAD occurred in the presence of 3-hydroxybenzoate (Fig. 7A).
Crystal structures were obtained of Y105F, H213S, and Q301E ( Table 1). The H213S mutant shows a small modification in the active site structure with the side chain of Gln-301 rotating away from the conformation present in the wild-type  Table 4. a Peaks 1* and 3* contain one sodium ion; the other peaks contain two sodium ions. Parenthetical value represents mass of peak 1* without sodium ion. enzyme (Fig. 6, B and C). Furthermore, although all variants were crystallized in the presence of 3-hydroxybenzoate, H213S crystals were the only ones to contain a bound substrate (Fig.   6C). The ligand carboxylate group is H-bonded to Gln-49 and Tyr-217, whereas the phenyl ring interacts with the C4a-N5 locus of the flavin, the terminal part of a phospholipid aliphatic chain, and Ser-213. As gathered from the electron density, the substrate binds in two conformations flipped by 180°around the C1-C4 axis of the aromatic ring (Fig. 6C). The double orientation is likely to be a consequence of the disruption of the anchoring point for the substrate 3-hydroxyl group caused by the H213S mutation and the conformational change of Gln-301 (Fig. 6, B-D). This feature may also explain why this mutant exhibits no spectroscopic perturbation upon substrate binding, although, at least in the crystallization conditions, substrate affinity is sufficiently tight to permit crystallographic analysis.  The Y105F and Q301E mutant structures are virtually identical to the native protein except for the side chain substitution. The network shown in Fig. 6D shows Tyr-105 hydrogen bonding to the carbonyl of Gln-49 whose amide hydrogen bonds to the carboxylate of the substrate. Eliminating the Tyr-Gln hydrogen bond could either make the amide less polarized, decreasing the partial positive charge on the nitrogen and causing a weaker hydrogen bond, or allow the amide to rotate 180°s o that the carbonyl of the Gln faces the carboxylate, preventing the hydrogen bond. Collectively, these findings provide the foundation for defining a structural framework to substrate binding and hydroxylation.

DISCUSSION
Chloride Binding-A chloride ion was found to bind in front of the flavin ring in direct contact with the pyrimidine moiety of the cofactor and the NH backbone atoms of residues 304 and 305 (Fig. 4A). This binding site corresponds to the niche predicted to host the oxygen atoms of the flavin-hydroperoxide adduct formed upon reaction of the reduced flavin with oxygen. In this context, it is noticeable that many flavoprotein hydroxylases are inhibited by chloride ions (41,42). For instance, studies with PHBH showed a competitive mode of inhibition with respect to NADPH and a mixed-type inhibition with respect to the physiological substrate 4-hydroxybenzoate (43). 3HB6H is no exception because it also exhibits similar mixed-type inhibition with chloride ions (Fig. 8). Collectively, these findings indicate that the negatively charged chlorine can both interfere with the binding and/or reactivity of the enzyme with oxygen, NADH, or both.
Lipid Binding-A non-covalently bound phospholipid molecule was found in each monomer of 3HB6H in the crystallographic model. The fatty acid chains occupy hydrophobic channels that deeply penetrate into the interior of the substrate-binding domain, whereas the hydrophilic portion of the molecule protrudes out of the protein surface, connecting the dimerization domains via a few interactions. Any attempt to natively remove the lipid from the protein by using beads (44), detergents, or reversible unfolding failed, indicating that the binding of the phospholipid is very strong. Interestingly, the flavoenzyme 6-hydroxy-L-nicotine oxidase from Arthrobacter nicotinovorans (45) also contains a similarly bound phospholipid molecule, which remains to be functionally characterized. In the case of 3HB6H, it is interesting to observe that the enzyme presents a different dimerization interface with respect to that of PHBH despite similar secondary structure organiza- tion. Most of the hydrophobic residues involved in PHBH dimerization are not conserved in 3HB6H, and the few proteinprotein interactions are primarily made by helix H13 contacts. We hypothesize that phospholipids in 3HB6H function as glue, promoting and stabilizing dimer formation. Recently, Clarke (46) and Ozeir et al. (47) have characterized a flavoprotein hydroxylase in Saccharomyces cerevisiae (Coq6) that is involved in coenzyme Q biosynthesis and is homologous to the E. coli UbiF, UbiH, and UbiI enzymes (48 -50). Coq6, UbiF, and UbiI catalyze the ortho-hydroxylation of 4-hydroxybenzoic analogs containing a polyprenyl chain. UbiH catalyzes the para-hydroxylation of a similar lipophilic aromatic substrate. Flavoprotein hydroxylases acting on phenolic acids and aromatic redoxactive lipids such as coenzyme Q could clearly derive from a common ancestor based on their evident structural homology and mechanistic similarity. A hypothesis is that the ability to bind aliphatic chains such as phospholipids and prenyl chains has been maintained (with different affinity) by 3HB6H, UbiF/ H/I, and Coq6 and was lost by most other FAD-dependent hydroxylases. Alternatively, the capacity to bind aliphatic ligands arose independently during evolution of flavoprotein hydroxylases.
Substrate Binding-Elucidation of the interaction between 3HB6H and its physiological substrate is particularly interesting in relation to the regioselectivity of the hydroxylation reaction. One of two observed substrate orientations observed in the H213S structure most likely corresponds to the catalytically competent binding mode with the substrate C6 carbon (the site of hydroxylation) close to the flavin C4a (Fig. 6D). The carboxylate group of 3-hydroxybenzoate interacts with Gln-49 and Tyr-105, whereas its 3-hydroxyl group is engaged in H-bonds with His-213 and Gln-301 side chains. These features are fully consistent with the mutagenesis data. The properties of the Y105F variant indicate that disruption of the H-bond network involving Gln-49 can mildly perturb binding affinity without, however, affecting activation of the substrate (Table 4). Conversely, substitutions of Gln-301 with Glu and of His-213 with Ala or Ser render the enzyme inactive. These two side chains have a primary role in catalysis, possibly facilitating deprotonation of the substrate phenol. Indeed, a comparable mechanism of substrate activation occurs in PHBH where a hydrogen bond network that extends to the protein surface is involved in ionization of the substrate hydroxyl moiety (31).
Analysis of substrate binding highlights a most remarkable feature. The aromatic ring is in van der Waals contact with the terminal part of the fatty chain of the bound phospholipid, which is, therefore, an integral element of the substrate-binding site. Thus, 3-hydroxybenzoate binds through a three-point anchoring mechanism that relies on the His-213/Gln-301 pair for providing the substrate-activating H-bond partner to the 3-hydroxyl, on the Gln-49/Tyr-105 pair for interacting with the carboxylate group, and on the lipid tail for creating a solventexcluded environment that embeds the more hydrophobic aromatic group of the ligand. Consistently, an additional OH group on the lipid-facing edge of the substrate ring makes the binding weaker (3,5-dihydroxybenzoate) or very poor (3,4-dihydroxybenzoate). Only the 2-hydroxyl group, which will bind opposite the lipid wall, can be fully tolerated with 2,3-dihydroxybenzoate, which is essentially as good a substrate as 3-hydroxybenzoate (14). This mode of affording selectivity is nicely illustrated by the comparison of 3HB6H with the substrate-bound structure of 3HB4H (Protein Data Bank code 2DKH). These two enzymes act on the same substrate but with different regioselectivities (6versus 4-hydroxylation). The structural superposition (Fig. 6D) shows that the bound 3-hydroxybenzoate molecules are rotated with respect to each other around an axis orthogonal to the aromatic ring. The substrates are coplanar, but their differing orientations make C6 of the 3HB6H-bound ligand perfectly overlap with C4 of the 3HB4H-bound substrate. In this way, the respective sites of hydroxylation are at the proper distance and orientation from the oxygen-activating C4a atom of the flavin. This is a beautiful example of how enzymes can vary the regioselectivity of their reaction with the peculiarity that a tightly bound phospholipid is utilized to attain this effect.