Three-dimensional Structure of AzoR from Escherichia coli

The crystal structure of AzoR (azoreductase) has been determined in complex with FMN for two different crystal forms at 1.8 and 2.2Å resolution. AzoR is an oxidoreductase isolated from Escherichia coli as a protein responsible for the degradation of azo compounds. This enzyme is an FMN-dependent NADH-azoreductase and catalyzes the reductive cleavage of azo groups by a ping-pong mechanism. The structure suggests that AzoR acts in a homodimeric state forming the two identical catalytic sites to which both monomers contribute. The structure revealed that each monomer of AzoR has a flavodoxin-like structure, without the explicit overall amino acid sequence homology. Superposition of the structures from the two different crystal forms revealed the conformational change and suggested a mechanism for accommodating substrates of different size. Furthermore, comparison of the active site structure with that of NQO1 complexed with substrates provides clues to the possible substrate-binding mechanism of AzoR.

other hand, biological degradation using microorganisms can decompose azo dyes under mild conditions, without the problems described above (4,5). To facilitate the development of biodegradation systems, it is essential to understand the detailed mechanisms of dye-degrading enzymes.
AzoR is an oxidoreductase isolated from Escherichia coli (6) as a protein responsible for the reduction of azo compounds. Biochemical studies have revealed that AzoR catalyzes the reductive cleavage of azo groups (-NϭN-) utilizing NADH but not NADPH as an electron donor by means of the flavin cofactor FMN. The reaction follows a ping-pong mechanism requiring 2 mol of NADH to reduce 1 mol of methyl red (4Ј-dimethylaminoazobenzene-2-carboxylic acid), a typical azo dye, into 2-aminobenzoic acid and N,NЈ-dimethyl-p-phenylenediamine. Therefore, it is thought that two cycles of the ping-pong mechanism were required for the reductive cleavage of azo groups. However, details about the molecular mechanism of the catalysis remain unknown.
Many kinds of bacterial azoreductases have been isolated and characterized for the progress of bioremediation (7)(8)(9)(10)(11)(12)(13)(14)(15). However, AzoR is different from other azoreductases reported thus far with respect to its requirements for cofactors, electron donors, and substrate specificity and its amino acid sequence. In addition, although the physiological function of AzoR remains unknown, its importance has been deduced from genome projects that have revealed the wide distribution of highly homologous genes in many microorganisms. For example, these genes are found in Yersinia pseudotuberculosis (16), Salmonella enterica (17), Photorhabdus luminescens (18), Vibrio vulnificus (19), Haemophilus influenzae (20), Pseudomonas putida (21), and so forth. On the other hand, AzoR has the ability to reduce menadione (vitamin K 3 , 2-methyl-1,4-naphthoquinone) (6). Quinone compounds are biologically active molecules that function as lipid electron carriers. Especially, a derivative of vitamin K is predominantly employed during anaerobic respiration in E. coli (22). These facts imply that AzoR plays an essential role in electron transport or metabolism during redox reaction or other processes.
AzoR is representative of a poorly characterized family of azo dye reductases. The structural determination of AzoR would be a first step toward elucidating its molecular mechanism of function, as well as those of highly homologous proteins of AzoR in other species, and the development of the biodegradation system. Here we report the crystal structures of AzoR complexed with FMN in two different crystal forms at 1.8 and 2.2 Å resolution. The structures revealed that AzoR has a flavodoxin-like structure without the explicit overall amino acid sequence homology and would likely act as a homodimeric FMN-containing enzyme. Superposition of the two crystal structures revealed the regions that participate in the conformational change of the active site, which would be a mechanism for accommodating substrates of different sizes. Furthermore, structural comparison of the active site with that of NQO1 revealed the conservation of amino acid residues, suggesting the substrate binding mode of AzoR. This is the first report of the structure of AzoR orthologues and of FMN-dependent NADH-azoreductase.

EXPERIMENTAL PROCEDURES
Protein Preparation and Crystallization-The recombinant E. coli AzoR used in this paper was expressed and purified as described previously (6). Two crystal forms, P4 2 2 1 2 and P42 1 2, of the AzoR were obtained under different crystallization conditions that contained FMN. The P4 2 2 1 2 crystals were obtained according to the published method (23). The P42 1 2 crystals were obtained from a drop made by mixing a solution containing 23 mg/ml protein in 10 mM Tris-HCl (pH 8.0), 1 mM FMN, and an equal volume of reservoir solution containing 200 mM NaCl, 100 mM CAPS 2 (pH 10.5), 20% (w/v) polyethylene glycol 8000, 20% (v/v) 1,4-dioxane, and 4 mM menadione. The drop was equilibrated over 500 l of the reservoir solution by the hanging drop vapor diffusion method at 293 K. The crystals grew to full size (0.08 ϫ 0.6 ϫ 0.6 mM) within 2 weeks.
Preparation of Heavy Atom Derivative-To prepare the heavy atom derivative, the P4 2 2 1 2 crystals were soaked for 23 days in a solution prepared by diluting 1 part K 2 PtCl 4 -saturated reservoir solution with 10 parts reservoir solution at 288 K. The details of the screening of heavy atom derivatives were described in a previous report (23).
Data Collection and Processing-Prior to data collection, P4 2 2 1 2 crystals and the K 2 PtCl 4 derivative were soaked in a reservoir solution containing 30% (v/v) ethylene glycol as a cryoprotectant. The P42 1 2 crystals were soaked in a reservoir solution containing 25% (v/v) glycerol as a cryoprotectant. The native data of the P4 2 2 1 2 and P42 1 2 crystals were measured at BL6A of the Photon Factory, KEK (Tsukuba, Japan) using an ADSC Quantum 4 CCD detector, and the K 2 PtCl 4 derivative data were collected at the BL41XU of SPring-8 (Harima, Japan) using a Mar Research 165-mm CCD detector. All of the diffraction data were collected under cryogenic conditions at 100 K. The data were reduced with MOSFLM, SCALA, and TRUNCATE from the CCP4 program suite (24).
Structure Solution and Phasing-The initial structure in the P4 2 2 1 2 crystal was determined by the SIRAS method using the K 2 PtCl 4 derivative. The heavy atom parameters and phases were calculated with SOLVE (25). Subsequently, maximum likelihood density modification was performed with RESOLVE (26), and the phases were further improved and extended to a resolution of 65.2-1.8 Å with ARP/wARP (27).
The initial structure in the P42 1 2 crystal was determined by molecular replacement. The calculation of the molecular replacement was carried out with MOLREP (28) using the structure of the P4 2 2 1 2 crystal as a search model. The high resolution limit of the diffraction data were set to 3.0 Å, and the low resolution limit was selected according to the standard protocol of MOLREP. The solution was then improved by rigid body refinement with REFMAC5 (29). The phases were improved and extended to a resolution of 67.4 -2.2 Å with ARP/wARP.
Model Building and Refinement-ARP/wARP was used for automatic model building followed by iterative manual model building with XtalView (30). 2mF o Ϫ DF c and mF o Ϫ DF c SIGMAA-weighted electron density maps (31) were used as references. All stages of maximum likelihood refinement were carried out with REFMAC5.
The P4 2 2 1 2 crystal structure consisted of 197 residues (1-59 and 63-200) of a possible 200 residues, one FMN, one ethylene glycol, one 2-propanol, and a total of 140 water molecules in an asymmetric unit. Arg-58 was modulated to Ala, and loop residues 60 -62 were not built because of their poor electron density. The structure from the P42 1 2 crystal included all possible amino acid residues, one FMN, two glycerols, and a total of 66 water molecules in an asymmetric unit. Although menadione was included in the crystallization solution for the P42 1 2 crystals, its electron density was not observed.
Model Analysis-The quality of the model was checked using PROCHECK (32). The superposition of molecules and calculation of the r.m.s. deviation between pairs of equivalent C␣ atoms and all atoms of proteins were executed using LSQKAB (33). Secondary structures were assigned using DSSP (34).

RESULTS
Overall Structure-Both forms of the AzoR crystal contain one 23-kDa monomer in an asymmetric unit, and the structures in both crystal forms are similar except with respect to a few regions (see below). Each monomer in both crystal forms constitutes a homodimer by a crystallographic symmetry operation in the same manner. The redox center FMN is found at the dimer interface, and both monomers contribute to form the two identical catalytic sites by the distance in ϳ25 Å away. These facts are consistent with the previous report that AzoR exists as a homodimer in solution, as shown by analytical gel filtration (6). Throughout this report, positions in the two monomers are distinguished by primed and nonprimed characters.
The structure from the P42 1 2 crystal includes all possible amino acid residues, giving the overall structure shown in Fig. 1 with the secondary structure assignment. The structure revealed that each monomer of AzoR has a flavodoxin-like structure, although a marginal level of sequence homology has been found only in the active site region. AzoR has a structure in which five parallel ␤-strands (␤1, ␤2, ␤3, ␤8, and ␤9) form an open twisted central ␤-sheet surrounded on both sides by a total of six helices (␣1-␣6). Helices ␣1 and ␣6 are on one side of the ␤-sheet, and helices ␣2, ␣3, ␣4, and ␣5 are located on the opposite side. The dimerization occurs mainly via anti-parallel side-to-side packing of the loop L7-helix ␣4 region and loop L11-helix ␣5 region of each monomer. The loop L3-helix ␣2  JULY 21, 2006 • VOLUME 281 • NUMBER 29 region also participates in the dimerization and interacts with the loop L3Ј-helix ␣2Ј region of the other monomer. A summary of the data collection and SIRAS phasing statistics is presented in Table 1. The final refinement statistics are summarized in Table 2.

Structure of AzoR (Azoreductase) from E. coli
FMN Binding Site-The FMN prosthetic groups bind on the C-terminal end of the central ␤-sheet at the dimer interface. Each FMN cofactor binds to both monomers; 15 hydrogen bonds are formed with one monomer, whereas hydrophobic contacts are made between both monomers (Fig. 2a). These interactions are conserved in both crystal forms. The isoalloxazine moiety of FMN interacts with residues involved in loops L7 and L11 of one monomer and loop L3Ј of the other monomer. The two oxygen atoms of the isoalloxazine ring form hydrogen bonds with the main chain NH groups of the polypeptide moiety: O-2 with Gly-142, and O-4 with Asn-97 and Phe-98. O-4 also forms a hydrogen bond with N⑀-2 of His-144. The nitrogen atoms of the isoalloxazine ring form hydrogen bonds with NH groups of the peptide moiety: N-1 with Gly-141, and N-5, which is thought to be the site of hydride transfer, with Asn-97. Although the redox potential of AzoR is unknown, the existence of the hydrogen bond donor to N5 of the isoalloxazine is attractive from the point of view of protein engineering, because this bond has been thought to directly affect the resonance characteristics of the pyridine ring that cause the alteration of redox properties in various flavoproteins (38). C7M of FMN makes hydrophobic interaction with Leu-50Ј. The isoalloxazine ring of FMN is close to planar (Fig. 2b). During data collection, the color of the crystal was yellow; therefore, this conformation is thought to be a fully oxidized one. The ribityl moiety interacts with residues involved in the ␤-strands ␤3 and ␤8 and loop L11. The OH groups of the ribityl moiety form hydrogen bonds with the polypeptide moiety. O-2* bonds the NH group of Gly-141 and the main chain carbonyl of Met-95. O-5 * also forms a hydrogen bond with the side chain OH group of Ser-139. The phosphate group of FMN is anchored in a pocket formed by loop L1, the N-terminal region of helix ␣1, and ␤-strand ␤4. This phosphate group makes several hydrogen bonds with the polypeptide moiety. O1P bonds with the side chain OH groups of Ser-15 and Ser-17. The main chain NH group of Ser-17 also forms a hydrogen bond with O1P. O2P forms hydrogen bonds with the side chain OH groups of Ser-9 and Tyr-96, and O3P forms a hydrogen bond with the main chain NH group of Gln-16. The phosphate group

Structure of AzoR (Azoreductase) from E. coli
of FMN is anchored in the pocket by hydrogen bonds rather than electrostatic interaction (Fig. 2c).
Superposition of the Two Structures Obtained from the Different Crystal Forms-To investigate the conformational changes of the molecule, we superposed the two structures obtained from the different crystal forms (Fig. 3). The superposition demonstrated that the fold of the polypeptide chains and the environment of the cofactor FMN are extremely similar in each of the two structures, as mentioned above. The r.m.s. deviations for the positions of equivalent 197 C␣ atoms and residues including side chains are 0.562 and 0.793 Å, respectively. However, the prominent conformational changes are found in three regions, which would mainly be a result of different crystal packing. The conformation of loop L4 is different between the P4 2 2 1 2 and P42 1 2 crystal structures. 3 The flexibility of this region is deduced from the high average thermal factor (B factor) in both crystal forms. Residues 60 -62 are particularly disordered in the P4 2 2 1 2 crystal, and thus we could not build a model of this region. The second prominent conformational change is in loop L9, which covers the active site cavity. The largest shift of C␣ in loop L9 is Asn-123, resulting in a 1.44 Å difference in position. In addition to loop L4, the flexibility of loop L9 is also deduced from the high B factor of the P42 1 2 crystal structure. Although the B factor of loop L9 is relatively low in the P4 2 2 1 2 crystal, it is attributed to the crystal packing in which loop L13 of the crystallographically related neighboring molecule penetrates the active site and tightly interacts with L9 (data not shown). The third important change is in the region that includes loop L13 and the N-terminal part of helix ␣6 (L13-N␣6). Pro-180, a residue included in the L13-N␣6 region, shows the largest change in the whole structure, resulting in a 3.05 Å difference in the C␣ position.

DISCUSSION
Structure Comparison with Other Flavoproteins-A structural similarity search in the Protein Data Bank with DALI (39) resulted in a number of matches to other flavoproteins, although they did not exhibit significant overall amino acid sequence similarity with AzoR. Proteins with a higher Z score than the value obtained for a typical prokaryote flavodoxin are as follows: the human FAD-dependent NAD(P)H:quinine oxidoreductase 1 (NQO1, originally called DT-diaphorase) (accession code 1D4A; Z score ϭ 17.3 and r.m.s. deviation ϭ 2.7 Å for FIGURE 2. Interactions between FMN and amino acid residues in the active site. a, schematic diagram showing contacts of the FMN cofactor to amino acid residues. Hydrogen bonds are shown as broken green lines (red residue numbers) and van der Waals' interactions by red shading (black residue numbers). Each atom element is represented by a sphere of different colors with a chemical symbol. b, SIGMAA-weighted 2mF o Ϫ DF c electron density maps surrounding the FMN. The map was calculated using the data of P4 2 2 1 2 crystal structure and is contoured at 1.2 . c, the electrostatic potential of AzoR is mapped onto the solvent-accessible surface, as calculated with GRASP. The FMN molecules shown in a and b are represented by a stick model, with color coding identical to that in Fig. 1. JULY 21, 2006 • VOLUME 281 • NUMBER 29 ent rubredoxin:oxygen oxidoreductase (ROO) from Desulfovibrio gigas (accession code 1E5D; Z score ϭ 13.2 and r.m.s. deviation ϭ 2.3 Å for 135 residues of a total 401 residues) (41). Bacillus subtilis Yhda protein (accession code 1NNI; Z score ϭ 12.9 and r.m.s. deviation ϭ 2.4 Å for 142 residues of a total 165 residues), although its function has not yet been documented, has a structure similar to that of yeast YLR011wp (accession code 1T0I; Z score ϭ 19.4 and r.m.s. deviation ϭ 2.3 Å for 156 residues of a total 191 residues, compared with 1NNI), and YLR001wp was characterized as an FMN-dependent NAD(P)H:ferric iron oxidoreductase (42). All of these flavoen-zymes exist as homodimers. Fig. 4 shows a comparison of the topology and folding of these flavoenzymes with a ribbon diagram. In the structure of these enzymes, a core region, where a central five-stranded parallel ␤-sheet (Fig. 4, cyan) is flanked on each side by two and three helices (Fig. 4, purple), is well conserved (loops of this conserved region are drawn in green). However, there are some prominent differences in other parts. First, there is a difference with respect to the lengths of the regions corresponding to between ␤-strand ␤2 and helix ␣3 of AzoR, as well as the variance of the number of helices (Fig. 4,  pink). These regions contribute to the formation of the active site cavities in all of these flavoproteins except ROO. Second, additional small C-terminal and N-terminal domains are found in NQO1 and ROO, respectively (Fig.  4, gray). The small C-terminal domain of NQO1 is responsible for the binding of the ADP moiety of NADP ϩ , an analogue for electron donor NADPH (43). The N-terminal domain of ROO, which has structural similarity with ␤-lactamase, reduces dioxygen to waters at the di-iron center, taking electrons from the FMN-containing, flavodoxin-like domain. Third, the structural functions of the regions corresponding to between helix ␣4 and ␤-strand ␤8 of AzoR are not equivalent (Fig. 4, yellow). This loop participates in the formation of the active site in AzoR and NQO1. On the other hand, this loop contributes to the stabilization of the additional N-terminal domain in ROO and contributes to the stabilization of the loop that is the component of the active site cavity in YLR011wp. Interestingly, these differences are found in the vicinity of the active site and probably result in the variety of the biochemical properties of these enzymes.

Structure of AzoR (Azoreductase) from E. coli
Our structural similarity search revealed that the flavoproteins, which share a similar structure with AzoR, are involved in various biological reactions. Among these enzymes, NQO1 is interesting because it has a function similar to that of AzoR regarding its substrate specificity. For example, both AzoR and NQO1 can exploit both methyl red and menadione as electron acceptors and NADH as electron donors (6,44). NQO1 plays an important role in detoxification and exerts a protective effect against mutagenicity, carcinogenicity, and other toxicities via its reduction activity in the cytosol, which prevents the production of semiquinone intermediates (45)(46)(47). These facts imply that AzoR may also have an important role in detoxification in E. coli.
Active Site and Conformational Changes-In the active site cavity, the re-face of FMN is fully occluded by a polypeptide moiety, and thus the reactions must occur at the si-face. In addition, there is not sufficient space near FMN to contact both the electron donor and acceptor simultaneously, which implies that only one substrate can bind to the enzyme at a time (Fig. 2c). This view is consistent with the ping-pong bi-bi mechanism for AzoR as has been proposed previously (6). Furthermore, the low average crystallographic B factors for FMN (Table 2), which are caused by the numerous contacts with amino acid residues mentioned above, suggest that FMN remains bound to the enzyme throughout the entire catalytic cycle.
Superposition of the two crystal structures revealed the prominent conformational changes in the molecule. Indeed, different crystal packing often affects the structure of inherently flexible region of a protein. In AzoR, loop L9Ј is positioned over the isoalloxazine ring moiety of FMN. The L13-N␣6 also participates in the formation of the active site cavity. Therefore, it appears that these conformational changes are important for substrate binding or recognition, which can rearrange the structure of the active site cavity to accommodate substrates of different size. Loop L4Ј is ϳ15 Å away from the catalytic center (isoalloxazine ring) of the cofactor FMN. However, if the nicotinamide ring of NADH binds to the active site in an extended conformation, this loop could be close to the adenosine moiety of NADH, and thus loop L4Ј also may participate in substrate binding or recognition.
Implications for a Substrate Binding and Electron Transfer-To obtain direct information about the substrate binding, we tried to provide the crystal of AzoR that is complexed with substrates, substrate analogues, and competitive inhibitors by the soaking method. However, the electron density of these molecules could not be found, which is probably attributed to the crystal packing in which the enzymes stacked each other around the active site in both crystal forms, as mentioned above. We have also tried cocrystallization experiments and obtained some crystals, but it is not clear whether or not these enzymes are complexed with any compounds in these crystals at present. Instead, our results revealed that the locations of some amino acid residues around the isoalloxazine ring are conserved between AzoR and NQO1 (Fig. 5). Phe-98, Gly-141, Gly-142, Tyr-120Ј, and Phe-162Ј of AzoR are found in the positions corresponding to Phe-106, Gly-149, Gly-150, Tyr-128Ј, and Phe-178Ј in NQO1, respectively. Among these amino acid residues, Gly-142, Tyr-120Ј, and Phe-162Ј of AzoR are attractive because the function of the corresponding amino acid residues in NQO1 have been clarified (40,43). That is, the main chain NH group of Gly-150, which makes a hydrogen bond with O-4 of the isoalloxazine ring, forms a hydrogen bond with O-4Ј of the ribose adjacent to the nicotinamide of NADP ϩ , a substrate analogue for the first half of the reaction of the ping-pong mechanism (electron donor). 4 The side chain of Tyr-128Ј can adopt many different conformations. The different conformations result in opening and closing of the binding site, as well as in interactions with substrates of different size. Phe-178Ј stacks onto the nicotinamide ring of NADP ϩ , resulting in sandwiching of the nicotinamide ring between the isoalloxazine ring and Phe-178Ј. Duroquinone (2,3,5,6-tetramethyl-quinone), a substrate for the second half of the reaction of the ping-pong mechanism (electron acceptor), is also sandwiched between the isoalloxazine ring and Phe-178Ј. If Phe-162Ј of AzoR stacks substrates in the manner of Phe-178Ј of NQO1, the configuration of a substrate and the isoalloxazine ring of a flavin cofactor would be ideal for a -stacking system (48) and direct hydride transfer (43). Menadione, which has also been reported as an electron acceptor for AzoR, occupies an area corresponding to between Phe-162Ј and the isoalloxazine ring of AzoR in QR2 (quinone reductase type 2), the homologous protein of NQO1 in mammals (49). These observations suggest that Gly-142, Tyr-120Ј, and Phe-162Ј of AzoR may play similar roles in the catalysis. Furthermore, the conservation of these amino acid residues in orthologues of the AzoR protein lends further support to our propositions. Indeed, binding of substrates by means of a -stacking system using phenylalanine in the vicinity of an isoalloxazine FIGURE 5. Comparison of the active site of AzoR with that of NQO1. One active site is shown for each of AzoR (a) and rat NQO1 (b) (Protein Data Bank accession code 1QRD). The flavin cofactors, duroquinone, and the side chains of conserved amino acid residues in both enzymes are represented by a stick model, with carbon atoms in gray, oxygen atoms in red, nitrogen atoms in blue, and phosphorous atoms in orange. The polypeptide moieties of one subunit for each enzyme are drawn as C␣ traces in yellow and green.
ring is often observed in the crystal structures of flavoproteins (50 -55). Hence, the aromatic ring of azo compounds may also be stacked on Phe-162Ј of AzoR using a -stacking system, with their azo group laid almost on top of N-5 of the isoalloxazine ring for an effective hydride transfer.
The C-terminal domain of NQO1 is responsible for the binding of the ADP moiety of NADP ϩ (43, 56). As mentioned above, the part corresponding to the C-terminal domain of NQO1 is lacking in AzoR. In addition, AzoR does not contain an NADHbinding motif (GXGXXG). These facts strongly suggest that NADH binds to the AzoR in a different manner than NQO1 and that AzoR may have a novel NADH-binding mechanism.
Generally, two electron-reduced FMN is protonated at N-1 of isoalloxazine ring to compensate for the negative charge built by hydride transfer. In AzoR, however, no functional groups can donate a proton to the N-1 position. Therefore, an enol tautomer at the O-2 position of isoalloxazine ring is proposed as the alternative site of protonation, in which proton is donated from N⑀-2 of His-144 to the O-2 of isoalloxazine (Fig. 2a). The other possibility is that the negative charge is delocalized at the N-1-C-2ϭO-2 locus, and it is stabilized by the positive charge of imidazole ring of His-144. These states of isoalloxazine are feasible from a standpoint on a research of molecular orbital (57).
In conclusion, we have determined the three-dimensional structure of AzoR from E. coli in two different crystal forms. AzoR would likely act as a homodimeric FMN-containing enzyme and has a flavodoxin-like structure without the explicit overall amino acid sequence homology. Superposition of the two structures from different crystal forms revealed the regions that allow conformational change near the active site. Comparison of the active site with NQO1 suggests some of the amino acid residues that may be involved in substrate binding. These amino acid residues are conserved in orthologues of the AzoR protein. It is also suggested that AzoR has a novel NADH-binding mechanism. The structure of AzoR will provide essential clues for the elucidation of the molecular mechanism of its function, as well as that of AzoR orthologues and the development of the biodegradation system.