Crystal Structures of NADH:FMN Oxidoreductase (EmoB) at Different Stages of Catalysis*♦

EDTA has become a major organic pollutant in the environment because of its extreme usage and resistance to biodegradation. Recently, two critical enzymes, EDTA monooxygenase (EmoA) and NADH:FMN oxidoreductase (EmoB), belonging to the newly established two-component flavin-diffusible monooxygenase family, were identified in the EDTA degradation pathway in Mesorhizobium sp. BNC1. EmoA is an FMNH2-dependent enzyme that requires EmoB to provide FMNH2 for the conversion of EDTA to ethylenediaminediacetate. To understand the molecular basis of this FMN-mediated reaction, the crystal structures of the apo-form, FMN·FMN complex, and FMN·NADH complex of EmoB were determined at 2.5Å resolution. The structure of EmoB is a homotetramer consisting of four α/β-single-domain monomers of five parallel β-strands flanked by five α-helices, which is quite different from those of other known two-component flavin-diffusible monooxygenase family members, such as PheA2 and HpaC, in terms of both tertiary and quaternary structures. For the first time, the crystal structures of both the FMN·FMN and FMN·NADH complexes of an NADH:FMN oxidoreductase were determined. Two stacked isoalloxazine rings and nicotinamide/isoalloxazine rings were at a proper distance for hydride transfer. The structures indicated a ping-pong reaction mechanism, which was confirmed by activity assays. Thus, the structural data offer detailed mechanistic information for hydride transfer between NADH to an enzyme-bound FMN and between the bound FMNH2 and a diffusible FMN.

EDTA has quietly become a major organic pollutant, currently present in the environment at higher concentrations than any other organic pollutant (1). A high level of EDTA in natural waters is due to its extensive usage, such as in industrial cleaning to remove calcium deposits, in detergent as a sequestering agent, in phytoremediation to mobilize heavy metals, and in scientific laboratories as a chelating agent (2,3). EDTA is recalcitrant to biodegradation and exists mainly in metal⅐EDTA complexes, many of which are toxic (4,5). In addition, the codisposal of EDTA with radionuclides has led to the enhanced mobilization of radionuclides in groundwater, rapidly spreading radioactive contamination (3, 6 -8). Concerns over EDTA recalcitrance and the potential mobilization of heavy metals and radionuclides have led the European Union, Australia, and some parts of the United States to ban EDTA in detergent. It is now also being carefully controlled in many other products to reduce contamination of water resources.
EmoA and EmoB are members of the recently discovered two-component flavin-diffusible monooxygenase (TC-FDM) 3 family, consisting of a large monooxygenase component and a small NAD(P)H:flavin oxidoreductase component. A demonstrable physiological function of the small component of TC-FDM is to provide reduced flavins to the large component. Bacterial luciferase of Vibrio fischeri is the first reported FMNH 2 -dependent monooxygenase (15). Recently, more FMNH 2 -dependent monooxygenases have been discovered, including pristinamycin IIA synthase of Streptomyces pristinaespiralis (16), two monooxygenases involved in desulfurization of dibenzothiophene in Rhodococcus sp. strain IGTS8 (17), organosulfur monooxygenases (SsuD) in Pseudomonas putida S-313 (18), and EDTA monooxygenase in Mesorhizobium sp. bacterium BNC1 (6). Furthermore, reduced flavin adenine dinucleotide (FADH 2 )-dependent monooxygenases have been   (20), and 2,4,5trichlorophenol 4-monooxygenase of Burkholderia cepacia AC1100 (21). All of these monooxygenases have a small-component NAD(P)H:flavin oxidoreductase as a partner, and the genes coding for the two components are normally physically linked and often organized in the same operon. Some of the small-component NAD(P)H:flavin oxidoreductases, which have tightly bound FAD or FMN as a prosthetic group, follow a Ping Pong Bi Bi mechanism; however, the ones that do not have tightly bound flavins seem to follow an ordered sequential mechanism (22)(23)(24). For example, PheA2 is a smallcomponent reductase that supplies FADH 2 to PheA1, a largecomponent monooxygenase for phenol hydroxylation. PheA2 has a tightly bound FAD prosthetic group (K d ϭ 10 nM), and its reaction follows a Ping Pong Bi Bi mechanism. HpaC St , the small-component reductase for 4-hydroxyphenylacetate 3-monooxygenase from Sulfolobus tokodaii, has a tightly bound FMN, probably also following a Ping Pong Bi Bi mechanism. However, another HpaC from Thermus thermophilus HB8 and EmoB from Mesorhizobium sp. BCN1 have weakly bound FAD and FMN, respectively (6,25), but their reaction mechanisms are unknown so far. Crystal structures for some of these smallcomponent reductases have been reported, such as PheA2 (26), HpaC St (27), and HpaC Tt (25). In addition to their structural similarities, the crystal structures of HpaC St and PheA2 reveal tightly bound FAD and FMN, respectively. The crystal structure of HpaC Tt shows FAD at low occupancy, as expected.
Interest in the structures and catalytic mechanisms of EmoB, which supplies FMNH 2 to EmoA for EDTA degradation, is due to its important physiological and future environmental roles (6). Purified recombinant EmoB does not contain a bound flavin, reflecting its loose binding of FMN. Here, we demonstrate that EmoB reduces FMN following a Ping Pong Bi Bi mechanism and have determined the corresponding crystal structures for two separate stages of catalysis. Therefore, we are able to provide complete structural features and explain the reaction mechanism. In addition, systematic investigations using light scattering and isothermal calorimetry offer critical information about EmoB in the EDTA degradation pathway.

Expression and Purification of EmoB-The
EmoB open reading frame was cloned from the Gram-negative bacterium BNC1 into the pET30 vector for overexpression of the protein as described previously (6). For EmoB expression, 100 ml of LB medium supplemented with 30 g/ml kanamycin was inocu-lated from a freezer stock of pET30EmoB in BL21(DE3) cells. After growing overnight at 37°C, this culture was used to inoculate 1 liter of LB medium in a 4-liter flask. Cells were grown at 37°C with constant shaking until the absorbance at 600 nm reached 0.6; the temperature of the shaker was then reduced to 20°C, and 0.5 mM isopropyl ␤-D-thiogalactopyranoside was added to induce EmoB expression. After incubation for an additional 10 h, the induced cells were harvested by centrifugation. The resulting cell pellet was resuspended in 40 ml of lysis buffer (50 mM sodium phosphate (pH 8.0), 300 mM NaCl, and 20 mM imidazole) and sonicated five times for 10 s each (Model 450 Sonifier, Branson Ultrasonics Corp.), and the lysates were cleared by centrifugation at 38,000 ϫ g for 45 min.
The cleared lysate was applied to a nickel-nitrilotriacetate column, and the column was washed extensively with lysis buffer. Protein was eluted from the column with lysis buffer containing 300 mM imidazole. Fractions containing EmoB were pooled, concentrated, and exchanged into 20 mM Tris buffer (pH 7.5) and then applied to an ion exchange column (Toyopearl DEAE-650M). Protein was eluted with a linear NaCl gradient.
The selenomethionine derivative of EmoB was made by transforming pET30EmoBL142M into the methionine auxotroph E. coli strain B834(DE3) (Novagen) and was cultured in minimal medium containing 30 mg/liter selenomethionine. Purification of selenomethionyl-EmoB was performed following the protocol for native EmoB except for the addition of 1 mM dithiothreitol in every buffer to prevent oxidation of the selenomethionine.
All the purification steps for both native EmoB and the selenomethionine derivative were analyzed by electrophoresis on a 12% Tris/glycine/SDS-polyacrylamide gel stained with Coomassie Blue. Protein concentrations were determined by the Bradford assay using bovine serum albumin as a standard.
Molecular Mass Determination-The weight average molecular mass of EmoB was measured by combined size exclusion chromatography and multiangle laser light scattering (MALLS) as described previously (28). In brief, 100 g of EmoB was loaded onto a BioSep-SEC-S-2000 column (Phenomenex) and eluted with phosphate-buffered saline. The eluate was passed through a tandem UV detector (Gilson), an interferometric refractometer (Optilab DSP, Wyatt Technology Corp.), and a laser light-scattering detector (Dawn EOS, Wyatt Technology Corp.). The light scattering data were analyzed with Astra software (Wyatt Technology Corp.) using the Zimm fitting method.
Isothermal Titration Calorimetry (ITC)-The interactions between EmoB and FMN, riboflavin, and NAD ϩ were measured using a MicroCal VP-ITC instrument at 25°C following standard procedures. The protein was dialyzed into 20 mM sodium phosphate buffer (pH 7.5) containing 150 mM NaCl. The protein was added to the calorimetric reaction cell at a concentration of 0.01 mM and titrated with 0.1 mM FMN, NAD ϩ , or riboflavin in the same buffer. Enzyme and ligand solutions were degassed prior to use. Each titration experiment was performed with 29 injections of 10 l at 300-s equilibration intervals. Heats of dilution for an individual ligand were determined by titrating ligand into the same buffer without protein  and were used to correct the protein titration. Data were fit to a single-site binding model by nonlinear least-squares regression with the Origin software package. The fit of data yields the binding affinity, enthalpy change, entropy change, and binding stoichiometry for the titration.
Kinetics of EmoB Reduction-Enzyme assays were conducted in 20 mM potassium phosphate (pH 7.0) at 25°C. FMN and NADH concentrations were varied from 0.5 to 5 M and 19 to 78 M, respectively. The reaction was initiated by the addition of 6.3 nM EmoB. NADH:FMN oxidoreductase activity was monitored by following the decrease in NADH absorbance at 340 nm (⑀ 340 ϭ 6220 M Ϫ1 cm Ϫ1 ) using an Ultrospec 4000 spectrophotometer (GE Healthcare).
Crystallization and Data Collection-Apo-form crystals of recombinant EmoB were grown at 4°C using the hanging drop vapor diffusion method. For apo-form crystallization, the solution of purified EmoB (10 mg/ml) in 20 mM sodium acetate (pH 5), 1 mM EDTA, and 1 mM dithiothreitol was mixed with an equal volume of the reservoir solution (0.1 M MES (pH 6.5), 1.6 M (NH 4 ) 2 SO 4 , and 10% dioxane) and equilibrated against the reservoir. Diffraction quality crystals appeared after 10 days. Crystals of selenomethionyl-EmoB were prepared in the identical fashion except with 1 mM dithiothreitol.
The apo-form crystals of EmoB belong to the hexagonal crystal system with unit cell dimensions of a ϭ b ϭ 101.59 Å and c ϭ 130.16 Å. The space group of these crystals was later determined to be P6 4 22 after disproving P6 2 22. There is one EmoB molecule in the asymmetric unit. Diffraction data up to 2.9 and 2.5 Å resolution were collected with a CCD detector/rotating anode x-ray generator (Saturn 92/MicroMax-007 x-ray generator, Rigaku/MSC) and at the Berkeley Advanced Light Source (beam line 8.2.1), respectively. For data collection at 100 K, crystals were transferred stepwise to a cryoprotection solution containing all components of the reservoir solution with 25% glycerol. The selenomethionyl-EmoB crystals were grown under the same conditions and found to be the same space group as the native crystals. To use the multiwavelength anomalous diffraction (MAD) method for phasing, data to 2.5 Å resolution were collected at 100 K at the Berkeley Advanced Light Source (beam line 8.2.1). The x-ray fluorescence spectrum was recorded and used to select wavelengths for subsequent MAD data collection. Data were collected at the selenium absorption peak ϭ 0.97925 Å (12,660.73 eV), the absorption inflection ϭ 0.97942 Å (12,658.53 eV), and the remote reference wavelength ϭ 0.91162 Å (13,600 eV). For complex crystals with FMN, riboflavin, FMN⅐NADH, and riboflavin⅐NADH, the soaking method was used. For the FMN or riboflavin complex, the apo-form EmoB crystals were incubated for 2 h at 4°C in a crystallization mother liquor solution containing either 1 mM FMN or riboflavin. For the FMN⅐NADH or riboflavin⅐NADH complex, the apo-form crystals were soaked for 2 h in the mother liquor solution containing either 1 mM FMN or riboflavin and then moved to the mother liquor solution containing 1 mM NADH. The corresponding complex data were collected at the Berkeley Advanced Light Source (beam line 8. for FMN⅐NADH-and riboflavin⅐NADH-soaked crystals were a ϭ 101.18 and c ϭ 129.71 Å and a ϭ 102.11 and c ϭ 130.41 Å, respectively. Diffraction data were collected up to 2.5 Å for the complex crystals. All diffraction data were processed and scaled with the HKL2000 package (29) and CrystalClear 1.3.6 (Rigaku/ MSC). The statistics for the diffraction data are listed in Table 1.
Phasing and Refinement-Initial phases of the EmoB crystal structure were determined by the MAD phasing method (30) using the software SOLVE (31) after prior approaches with the molecular replacement method. Data collected at the remote wavelength were treated as the reference data set, and resolution limits of 40-2.7 Å were imposed. Experimental values of fЈ and fЉ estimated from fluorescence spectra were used. The selenium site was located, and the resulting phases had a figure of merit of 0.51. A density modification process using the maximum likelihood method was performed with the software RESOLVE (32), which eventually resulted in a clearly interpretable electron density map with many well defined secondary structural elements. The corresponding amino acids were assigned and manually fitted into this map using the software O (33). The resulting rough coordinates of the EmoB structure were refined using X-PLOR (34) with the simulated annealing protocol, resulting in a crystallographic R value of 28%. Several rounds of manual adjustment followed cycles of refinement, and picking solvent molecules led to an R factor of 20.2% (R free ϭ 23.6% for the random 5% data). The structures of the complexes of EmoB were then solved using the refined coordinates of the apo-form EmoB. The final R factors for the FMN⅐FMN and FMN⅐NADH complexes of EmoB were 20.3% (R free ϭ 23.2% for the random 5% data) and 20.2% (R free ϭ 23.8%), respectively (see Table 1). The number of reflections above the 2 level for the apo-form and FMN⅐FMN and FMN⅐NADH complexes was 12,415 (89% completeness), 13,826 (99%), and 13,264 (96%) between 12.0 and 2.5 Å resolution, respectively. The root mean square deviations from ideal geometry of the final coordinates corresponding to the apo-form and complexes were 0.014, 0.016, and 0.016 Å for bonds and 3.18, 3.85, and 3.82°for angles, respectively. All EmoB coordinates have been deposited in the Protein Data Bank: 2VZF (apo-form), 2VZH (FMN⅐FMN complex), and 2VZJ (FMN⅐NADH complex).

RESULTS
Recombinant EmoB from Mesorhizobium sp. BNC1 was purified, and its corresponding structures in the apo-form and FMN⅐FMN and FMN⅐NADH complexes were determined at 2.5 Å resolution. The apo-forms of both native and selenomethionine-substituted enzymes were crystallized in the same hexagonal space group, P6 4 22, with one molecule/asymmetric unit ( Table 1). The structure of the apo-form EmoB was determined by the selenomethionyl MAD method (30). Most of the backbone and side chain residues were fitted using the selenomethionyl MAD map, but the electron density for the C-terminal seven residues and most of the side chains of residues 145-156 was not visible due to disorder. The final R factor for the apo-form EmoB was 20.2% (R free ϭ 23.6%) for 12,415 (12-2.5 Å) unique reflections. In turn, both complex structures were determined using the coordinates of the deduced apo-form EmoB structure. Detailed crystallographic data are reported in Table 1.
Global Structure-The structure of the EmoB molecule showed that it is a single-domain enzyme and belongs to a globular ␣/␤-structure similar to that of flavodoxin. It consists of a central five-stranded parallel ␤-strand flanked by either two (␣1 and ␣5) or four (␣2A, ␣2B, ␣3, and ␣4) ␣-helices on both sides of the sheet (Fig. 2A). The twisted central ␤-sheet constituting the core of the enzyme is arranged in the order ␤2-␤1-␤3-␤4-␤5. Among the five ␤-strands, ␤5 uniquely has a two-residue bulge at Val 137 and Gln 138 , which is observed in every compared NAD(P)H:flavin oxidoreductase mentioned below, even though the size of each bulge is somewhat different. In fact, based on the presence of this ϳ20-residue loop inserted at this bulged site, flavodoxins are classified into two groups, long and short flavodoxins, even though the corresponding functions for most of them are not clearly known (35). Therefore, EmoB could belong to a short flavodoxin subfamily in terms of its core structure. Distinctively, most of the surface residues of the central ␤-sheet that constitutes the core of EmoB have a hydrophobic nature. Those hydrophobic residues interact with the hydrophobic side of the amphiphilic ␣-helices, the hydrophilic faces of which, together with hydrophilic loops, are exposed to solvent. Therefore, the surface residues of EmoB are very hydrophilic, explaining its high solubility reaching ϳ0.5 mM in phosphate-buffered saline. The overall structures of the apoform and complex structures show no major differences in terms of their backbone structures. The C-␣ carbons among the apo-form and complexes are superimposable, with root mean square deviations values of 0.2-0.3 Å without including the flexible area (the residues between positions 144 and 156). However, the side chains of several residues show different conformations to accommodate FMN and NADH binding.
As shown clearly in Fig. 3, most of the loop regions have slightly elevated temperature factors, especially the loop connecting ␣5 and ␤5, which show very high values and contain disordered side chains in the middle. However, temperature factors of the residues corresponding to two loops, between ␤3 and ␣3 and between ␣3 and ␤4, are similar to those of the secondary structures of the ␤-sheet core due to the existence of many hydrogen bonds. As mentioned in detail below, one of these areas, a loop connecting ␤3 and ␣3 (residues 78 -86), constitutes the bottom of the FMN-binding pocket. Notably, the corresponding temperature factors for residues 112-114 are reduced significantly upon complex formation.
Oligomer State-A crystallographic symmetry operation assembled one EmoB molecule in the asymmetric unit into a tightly associated tetrameric unit in the crystal lattice (Fig. 2B). This oligomeric status of EmoB as a tetramer was verified in solution by an MALLS experiment. MALLS analyses of solutions of both apo-EmoB and the FMN complex were done, and both were shown to be a tetramer (supplemental figure). The observed tetramer interface has an extensive network of symmetrically oriented intersubunit hydrogen bonds. These intersubunit interactions are mainly between residues of ␣4 and its flanking loops, i.e. hydrogen bonds between the hydroxyl group of Tyr 122 and the backbone amide nitrogen of Ala 82 , the hydroxyl groups of Ser 83 and Tyr 84 , the side chain amide group of Lys 81 and the backbone carbonyl oxygen of Leu 95 , the carboxyl side chain of Asp 93 and the backbone amide nitrogen of Gly 86 , and the side chain amide group of Lys 89 and the backbone carbonyl oxygens of Tyr 80 and Tyr 84 . As mentioned below, the backbone of Ala 82 , which participates in the intersubunit interaction, also coordinates the O-4 atom of the isoalloxazine ring; thus, the observed tetramer interaction is also associated with FMN binding.
FMN-and NADH-binding Site-The F o Ϫ F c maps of the FMN-soaked crystal data clearly show the corresponding electron density for two bound FMN molecules at the carboxyl edge of the ␤-sheet (Fig. 4A). They are stacked through the si-faces of the isoalloxazine rings. However, the same area calculated from the riboflavin-soaked crystal data does not have any density. In the FMN complex structure, two FMN molecules have similar conformations, and the ribityl moieties of both FMN molecules adopt an extended conformation. For clarity, the deeply inserted FMN molecule is referred to as the first FMN throughout this study. The first FMN is in the crevice located at the topological switch point (or chain reversal point), as observed in many flavoproteins (36). The phosphoribityl group of this first FMN molecule is in the relatively shallow pocket formed by the two loops between ␤1 and ␣1 and between ␤5 and ␣5, and the edge of its isoalloxazine ring is surrounded by parts of the loop connecting ␤3 and ␣3 and the loop connecting ␤4 and ␣4.
In detail, the phosphate group of the first FMN molecule is within hydrogen bond distance of the side chains of Ser 11 , Ser 16 , and Thr 18 and the backbone nitrogen atoms of Thr 17 and Thr 18 , which together constitute the hydrophilic bottom of the pocket (Fig. 4). The apo-form has a phosphate ion at the same position as the phosphate group of the first FMN. The ribityl O-2* and O-3* are hydrogen-bonded to the backbone nitrogen atom of Ile-79 and a solvent molecule that is connected to the phenolic side chain of Tyr 80 . The O-2 and O4 atoms in isoalloxazine form hydrogen bond with the backbone nitrogen atoms of Ala 82 , Gly 112 , and Ser 113 . Consequently, their observed temperature factors are reduced upon complex formation (Fig. 3). The phenol ring of Tyr 80 is also in a stacked position with the isoalloxazine ring of the first FMN. In contrast to this strongly coordinated interaction for the first inserted FMN molecule, the other FMN molecule that is stacked with the FMN (referred to as the second FMN for clarity) is located at the surface of the enzyme exposed partially to solvent with fewer interactions with EmoB, i.e. only the O-4 atom of its isoalloxazine ring is hydrogen-bonded to the side chain of Lys 81 in addition to its phosphate group being hydrogen-bonded to a solvent molecule, which is coordinated in turn by the phenol side chain of Tyr 80 and the phosphate group of the first FMN.
On the other hand, the F o Ϫ F c maps generated with the data collected from the crystal soaked with both FMN and NADH clearly show the corresponding electron density for both FMN and NADH molecules stacked through their isoalloxazine and nicotinamide rings (Fig. 4B). The position of the FMN molecule in this FMN⅐NADH complex is superimposable with the first FMN molecule in the FMN⅐FMN complex structure. The amine group of Lys 81 is also within hydrogen bond distance of the amide group of NADH, the re-face (A-side) of which faces the isoalloxazine ring. The corresponding density for the adenosine part of NADH is not visible, indicating its disordered character. In addition to Lys 81 , the backbone carbonyl group of Gly 112 and the ribityl oxygen of the first FMN are within hydrogen bond distance of the phosphate groups of the NAD(H) molecule. However, the F o Ϫ F c maps generated from the crystals soaked in the mother liquor solution containing both 1 mM riboflavin and 1 mM NADH for various times (2-48 h) show neither riboflavin nor NADH.
ITC-To confirm the differential binding affinities between FMN, riboflavin, and NAD(H) with the EmoB protein, thermodynamic characterizations were done with ITC. As shown in Fig. 5, a significant amount of heat was released when EmoB associated with FMN, indicating that the binding interactions had significant enthalpic contributions (⌬H ϭ Ϫ15.3 kcal/mol). Analysis of the ITC data revealed a slightly unfavorable entropic contribution (⌬S ϭ Ϫ24.7 cal/mol/degree), possibly indicating that the EmoB structure was slightly stabilized upon binding to FMN and that very few solvent molecules were freed from the pocket. These were consistent with structural data, as indicated by the significant reduction of the B values for two loops constituting the binding pocket upon formation of the FMN binary complex. There were few solvent molecules in the FMN-binding pocket. The calculated K d for FMN from ITC data analysis was ϳ0.42 M. Neither riboflavin nor NADH independently showed any significant binding to EmoB (Fig. 5).

DISCUSSION
NAD(P)H:flavin oxidoreductases, to which EmoB belongs, catalyze a wide range of electron transfer reactions. The exact biological function and electron transfer mechanism of most of these enzymes are not known (37). We have investigated the structure of EmoB to shed light on its biological function, reac-tion mechanism, and structural relationship to other members in this functional class.
The structure of EmoB, a flavodoxin-like tetramer, is quite different from those of other TC-FDM family members, such as PheA2, HpaC St , and HpaC Tt . These family members all have a dimeric character and a ␤-barrel-like structure similar to the structures of ferric reductases or FAD-binding domains of ferredoxin reductases (25,26). To establish the proper structural classification for EmoB and to identify its structural homologs, a detailed comparison with available structures in the Protein Data Base was carried out using a Dali search (38). The results showed that the most similar structure is a putative flavin-binding arsenic-resistance protein from Shigella flexneri (code 2FZV) with a high Z score of 20.1, followed by azobenzene reductase from Bacillus subtilis (code 1NNI) with a Z score of 18.2 and a putative NADH-dependent reductase from Pseudomonas aeruginosa PA01 (code 1RTT) with a Z score of 17.7 (Fig. 7). Despite their unclear functionality, these proteins all belong to the flavodoxin-like flavoprotein family or have at least known flavin-binding ability.
Comparison of sequence similarity to known structures suggested the structural classification of EmoB. A search for similar amino acid sequences in the Protein Data Bank using BLAST (39) revealed that FMN-dependent azoreductase from Enterococcus faecalis (code 2HPV) shows the highest score (31.6 bits) with 23% identity among matched amino acids to EmoB, followed by a putative NADH-dependent reductase from P. aeruginosa PA01 (code 1RTT; 31.2 bits, 26% identity), Trp repressor-binding protein from B. subtilis (code 1RLI; 30.8 bits, 32% identity), a probable short chain dehydrogenase from P. aeruginosa (code 2NWQ; 29.3 bits, 28% identity), and ArsH from Sinorhizobium meliloti (code 2Q62; 29.3 bits, 27% identity) (Fig. 7). Overall, however, the level of sequence identity is low, and most of the BLAST alignments are rather random. A meaningful sequence alignment was possible only by manual alignment with a structural guide (Fig. 8); however, even that approach was not possible for 2HPV and 2NWQ. The conserved residues are distributed sporadically mainly through the entire secondary structural elements (Fig. 8). In general, the enzymes with the highest similarity scores have a mixed ␤-sheet at the core of the enzyme with the same topological order as observed in EmoB (Fig. 7), but a detailed visual inspection revealed a great deal of structural heterogeneity in terms of the number and size of helices in the peripheral regions, which could correlate with the great functional diversity among them.
Significantly, none of the enzymes belonging to the TC-FDM family were found in our Dali and BLAST searches. As shown in Fig. 9, EmoB is somewhat distant from the other known small components of the TC-FDM family members in terms of their primary sequence. This is consistent with the differences found in the secondary, tertiary, and quaternary structures of EmoB compared with PheA2 and HpaC. This is probably due to the fact that a small component of TC-FDM, the sole function of which is to provide reducing power (through FADH 2 or FMNH 2 ) to its partner monooxygenases, does not have to evolve in a parallel way.
The primary and tertiary structural comparison of the abovelisted oxidoreductases of high similarity revealed that EmoB  does not have the well known canonical flavodoxin motif. The motif, (T/S)XTGXT, is strongly conserved among classical flavodoxins and is located in the loop area connecting ␣1 and ␤1, which is involved in coordinating the phosphate group of FMN. EmoB uses loop 11 SPSRNSTT 18 at this same location. As reported recently for other FMN reductases and flavoproteins (34,40), this area is not well conserved among the structurally related proteins (Fig. 8). However, despite a lack of sequence similarity, the backbone and side chains of the corresponding loop regions are all engaged in phosphate coordination in all of the proteins with available complex structures. In addition, Tyr 80 and Pro 78 of EmoB also coordinate the phosphate moiety, and these two residues are more conserved than the abovementioned canonical flavodoxin motif. In addition, Gly 111 , Gly 112 , and Ser 113 in EmoB are highly conserved among flavodoxins and related proteins, and the backbones of Gly 112 and Ser 113 are hydrogen-bonded to the N-1 and O-2 atoms of the isoalloxazine ring. Therefore, the canonical motif, (T/S)XT-GXT, might be unique to the classical flavodoxins, and the general peptide motifs for flavin binding should be revisited and must include Tyr 80 , Pro 78 , Gly 111 , Gly 112 , and Ser 113 .
The oligomeric structure of EmoB is also suggestive of its structural classification. Most of the NAD(P)H:FMN oxidoreductases and flavodoxin-related proteins with known structures exist in ether dimeric or monomeric forms. Especially all of the reported NAD(P)H:flavin oxidoreductases in the TC-FDM family (HpaC St , HpaC Tt , and PheA2) exist as dimers. However, both our crystal structure and MALLS data indicate that EmoB is a tetramer in solution, as is the case for some of the structurally related proteins, such as the flavin-binding protein from S. flexneri (Protein Data Bank code 2FZV), the flavoprotein from S. meliloti (code 2Q62), and the Trp repressor-binding protein from B. subtilis (code 1RLI). Therefore, it is tempting to speculate that these tetrameric flavoenzymes could constitute a unique branch in NAD(P)H:flavin oxidoreductases with EmoB. However, our MALLS analysis indicated that the apo-form of EmoB still maintains its tetrameric character, which is different from the FMN-dependent dimer-tetramer transition observed in other bacterial flavodoxin-like proteins, such as the Trp repressor-binding protein and azoreductase (41,42). In all of these tetrameric enzymes, a typical dimer configuration among many FMN reductases is still maintained in their tetramer interface (e.g. the yellow/green or blue/red pair in Fig. 2B).
Active Site and Catalytic Mechanism-Our results provide detailed information about the mechanism by which EmoB performs its function. In EmoB, four equivalent redox sites are located near the tetramer interface. At each of these sites in the complex form of EmoB, either pairs of stacked FMN molecules or stacked FMN and NADH are observed. Like other tetrameric FMN reductases, such as the above-mentioned flavin-binding protein from S. flexneri (Protein Data Bank code 2FZV), the flavoprotein from S. meliloti (code 2Q62), and the Trp repressor-binding protein from B. subtilis (code 1RLI), the EmoB tetramer has a somewhat exposed entry site for the FMN(H 2 ) and NAD(H) molecules. This is consistent with the fact that the apo-form and the complex form of EmoB do not show any significant structural differences. For example, the FMN-binding pocket of NAD(P)H:FMN oxidoreductase from Saccharomyces cerevisiae (code 1T0I) has a very limited accessibility due to blockage by the extra residues between ␤2 and ␣2 (Fig. 8) contributed from the other subunit (43); thus, it is very unlikely to bind an FMN molecule in this dimeric enzyme without a substantial conformational change. Among NAD(P)H:flavin oxidoreductases and flavodoxin-like proteins (Figs. 7 and 8), the entry sites for the flavin molecule have differently sized and positioned secondary structural elements, probably dictating the substrate preference and activities for an individual class of enzymes. Our results provide essential information for determining the catalytic mechanism of EmoB. Especially the complex structure for two FMN molecules in their redox position has been established for the first time. In the FMN complex form of EmoB (oxidation state form), the two stacked FMN molecules are oriented in opposite directions (e.g. stacked through the si-faces of their isoalloxazine rings), making the distance between the N-5 atoms of the isoalloxazine rings 3.4 Å, which is a proper distance for hydride transfer (Fig. 4A). Compared with the one significant interaction (with Lys 81 ) of the loosely bound second FMN molecule, the first FMN molecule has a tight interaction with the enzyme. The experimental results that (i) a riboflavin molecule is not visible in the riboflavin-soaked crystals, (ii) there is a very weak heat of binding for riboflavin in the ITC data, and (iii) the apo-form structure has a phosphate ion at the same location for the phosphate group of the first FMN reflect the importance of the phosphate group of FMN in its affinity. This is consistent with the crystal structure showing that the phosphate group of the first FMN forms a network of interactions with Ser 11 , Thr 17 , Thr 18 , and Tyr 80 (Fig. 4C). In addition, the corresponding structure for the other half-redox reaction has been established. In our FMN⅐NADH complex form of EmoB (reduction state form), it is likely that the FMN molecules are in their reduced state, as judged by the pale yellow color of the soaked crystal in contrast to the intense yellow color of the FMN⅐FMN complex crystal. The stacked isoalloxazine and nicotinamide rings are also properly oriented at a distance of 3.5 Å between C-4 of NAD(H) and N-5 of FMN(H 2 ).
Compared with the loosely bound second FMN molecule of the FMN⅐FMN complex, the NAD(H) molecule has more interactions with the enzyme (with Lys 81 and Gly 112 ) and the first FMN, indicating its higher affinity. This matched well with the observation that the crystal soaked with both FMN and NADH has FMN(H 2 )⅐NAD(H) molecules instead of two FMN molecules in the binding sites. However, the NAD(H) molecule by itself has very weak affinity, as indicated by the ITC results and the lack of NADH density in the soaked crystal with riboflavin and NADH. Therefore, an affinity of NADH for EmoB exists only after the first FMN occupies its site.
The complex structures of EmoB also suggest its potential catalytic mechanism. Significantly, the N-1 atom of the isoalloxazine ring of the first FMN is within hydrogen bond distance of the backbone amide nitrogen of Gly 112 , which could serve as a general acid/base catalyst for protonation/deproto- FIGURE 8. Amino acid sequence alignment of EmoB with other flavin-dependent oxidoreductases. Secondary structural elements are highlighted in green for the ␤-strands and in red for the ␣-helices. Code 2FZV is the flavin-binding protein from S. flexneri; code 2Q62 is the flavoprotein ArsH from S. meliloti; code 1NNI is the azobenzene reductase from B. subtilis; code 1RTT is the putative reductase from P. aeruginosa PA01; code 1T0I is the NAD(P)H:FMN oxidoreductase from S. cerevisiae; code 1RLI is the Trp repressor-binding protein from B. subtilis; and code 1QRD is the quinone reductase from rat.  Left, reduction process of the first FMN by the stacked NADH; right, oxidation process of the bound reduced FMN by the second FMN. The backbone amide nitrogen atom of Gly 112 is within hydrogen bond distance (2.8 Å), which can either stabilize the negative charge of the semiquinone form or provide a proton to it. The arrows indicate the movement of an electron. The figures were made using ChemBioDraw Ultra 11.0 (Cambridge Corp.).