Crystal Structure and Functional Characterization of Yeast YLR011wp, an Enzyme with NAD(P)H-FMN and Ferric Iron Reductase Activities*

Flavodoxins are involved in a variety of electron transfer reactions that are essential for life. Although FMN-binding proteins are well characterized in pro-karyotic organisms, information is scarce for eukaryotic flavodoxins. We describe the 2.0-Å resolution crystal structure of the Saccharomyces cerevisiae YLR011w gene product, a predicted flavoprotein. YLR011wp indeed adopts a flavodoxin fold, binds the FMN cofactor, and self-associates as a homodimer. Despite the absence of the flavodoxin key fingerprint motif involved in FMN binding, YLR011wp binds this cofactor in a manner very analogous to classical flavodoxins. YLR011wp closest structural homologue is the homodimeric Bacillus subtilis

The most frequently used cofactors in enzymatic redox reactions are the pyridine (NAD and NADP) and the flavin (FAD and FMN) nucleotides. Although NAD and NADP are soluble cofactors used by dehydrogenases, FAD and FMN usually work as prosthetic groups in flavoproteins to which they are tightly bound. Flavodoxins are small monomeric flavoproteins (15)(16)(17)(18)(19)(20)(21)(22) kDa) that noncovalently bind a single FMN molecule, acting as a redox center (1). The flavodoxin scaffold contributes to the mechanism of electron transfer by stabilizing the FMN molecule in an environment that promotes the highly negative redox potential required for its biochemical activity. This is accomplished by sandwiching the FMN ring between two hy-drophobic side chains. Flavodoxins are involved in a variety of electron transfer reactions that are essential in the metabolism of pyruvate (2), nitrogen, and pyridine nucleotides (3). These enzymes exist under three redox states: oxidized; partially reduced (semiquinone); and fully reduced (hydroquinone). Until now, flavodoxins have been identified in many prokaryotes and also in some eukaryotic algae (4 -6). In mammalian systems, flavodoxins have only been found as domains inserted in larger redox proteins such as cytochrome P450 reductase where they act as the electron transport intermediate between FAD and the P450 iron center (7). Flavodoxin-like domains are also found in mammalian nitric-oxide synthase (8) and in human erythrocyte NADPH-flavin reductase (9).
The Saccharomyces cerevisiae YLR011w open reading frame codes for a protein of unknown function that has been found to be up-regulated in response to low temperature exposure (10). The sequence belongs to a COG4530 (Cluster of Orthologous Genes) predicted to generally contain flavoproteins. Hence, the YLR011wp 21-kDa protein was proposed to be composed of a single domain present in NADPH-dependent FMN reductases. We have solved the 2-Å resolution crystal structure, confirming that it adopts a flavodoxin fold and binds the FMN cofactor. YLR011wp exists as a dimer in the crystal as well as in solution. The YLR011wp closest structural homologue is the Bacillus subtilis Yhda protein (Protein Data Bank code 1NNI) (11), whose function is not documented but has 53% sequence identity with the Bacillus sp. OY1-2 azoreductase. These enzymes are responsible for the reductive cleavage of azo dyes. Organisms producing azoreductase present a biotechnological interest for the early step in detoxification of industrial dyes. This type of activity has not been reported before in S. cerevisiae. Therefore, we have attempted to characterize YLR011wp reductase activity by testing different azo dyes and nitrocompounds but also ferricyanide as electron acceptors.

EXPERIMENTAL PROCEDURES
Cloning and Purification-The YLR011w open reading frame DNA was amplified by PCR using oligodeoxynucleotides synthesized by MWG Biotech and cloned between the NdeI and NotI sites of pET29 vector from Novagen. The expression strain Rosetta was obtained from Novagen. 2ϫ YT medium (5 g of NaCl, 16 g of Bactotryptone; 10 g of yeast extract) was purchased from BIO 101, Inc., and isopropyl-␤-Dthiogalactopyranoside was purchased from Sigma. Transformed Rosetta cells were grown at 37°C up to an A 600 nm of 1. Expression was induced with 0.5 mM isopropyl-␤-D-thiogalactopyranoside, and the cells were grown for a further 12 h at 15°C. Cells were collected by centrifugation, washed, and resuspended in lysis buffer (20 mM Tris-HCl, pH 8, 300 mM NaCl, 10 mM ␤-mercaptoethanol) at 4°C. Cells were lysed by 3 cycles of freeze-thawing and 30 s of sonication and then were centrifuged. The His-tagged proteins were purified using a Ni 2ϩ affinity column (Qiagen Inc.) according to the standard protocols. Eluted protein was further purified by gel filtration using a Superdex TM 75 column (Amersham Biosciences) equilibrated against the buffer containing 20 mM Tris-HCl, pH 8, 100 mM NaCl, and 20 mM ␤-mercaptoethanol. The purity of the pooled fractions was checked by SDS-PAGE, and the integrity of the protein samples was checked by mass spectrometry. Selenium Met (Se-Met) 1 -labeled proteins were prepared as follows. Transformed cells were grown in M63 medium at 37°C up to an A 600 nm of 1. The culture then was complemented with a mixture of amino acids (L-Lys, L-Phe, and L-Thr at 100 mg/liter and L-Ile, L-Leu, L-Val, and L-Se-Met at 50 mg/liter) within 30 min, and the recombinant protein was overexpressed by the addition of 0.3 mM isopropyl-␤-D-thiogalactopyranoside within 12 h at 15°C. The protein was purified as described above. Seleno-methionine incorporation into the protein was Ͼ95% as assayed by mass spectrometry.
Crystallization and Resolution of the Structure-Native and Se-Met labeled protein samples were stored in 20 mM Tris-HCl, pH 8, 100 mM NaCl, and 20 mM ␤-mercaptoethanol. Prior to crystallization screenings, 20 mM CaCl 2 and 2 mM FMN were added to the protein solutions. Crystals of both the native and Se-Met-labeled protein were grown from a 1:1-l mixture of protein (6 mg/ml) with 10 -15% polyethylene glycol 3000, 20% polyethylene glycol 400, and 0.1 M Tris-HCl, pH 8.5 (reservoir solution). Crystals with a maximal size of 100 -200 m appeared within 3 days. For data collection, the crystals were frozen in liquid nitrogen after soaking in cryoprotectant buffer containing 10% polyethylene glycol 3000, 30% polyethylene glycol 400, and 0.1 M Tris-HCl, pH 8.5. Crystals of the native and Se-Met labeled proteins diffracted to 1.9 and 2.3 Å, respectively (beamline BM30A at the European Synchrotron Radiation Facility, Grenoble, France) (12).
The structure was determined using single wavelength anomalous dispersion x-ray diffraction data collected from the Se-Met derivative crystal. Data were processed using the HKL package (13). The space group was P2 1 2 1 2 with two molecules per asymmetric unit. Two selenium atom sites of three were found for each monomer with the program SOLVE in the 80 -2.8-Å resolution range (14). After solvent flattening and phase extension to a 2.3-Å resolution with the program RESOLVE (15), the quality of the electron density map allowed the manual construction of most of the secondary structure elements with the TURBO molecular modeling program (16). The model was completed using iterative cycles of model building using TURBO and refinement with the REFMAC program (17). The final model contains residues 1-185 from monomer A, residues 2-30 and 39 -186 from monomer B, two FMN molecules, two calcium ions, and 111 water molecules. All of these residues have well defined electron density and fall within the allowed regions of the Ramachandran plot as defined by the program PROCHECK (18). Data collection and refinement statistics of the structure are summarized in Table I.
Enzyme Assays-YLR011wp was assayed for FMN reductase activity in a reaction mixture (1 ml of final volume) consisting of 150 M NADH or NADPH, 100 M FMN, and 0.24 M enzyme in 25 mM Tris-HCl, pH 7.5. NAD(P)H oxidization was followed at 30°C by measuring the decrease in absorbance at 340 nm (⑀ M ϭ 6.22 mM Ϫ1 cm Ϫ1 ).
Azoreductase assay was performed in a reaction mixture (1 ml) containing 35 M azo dye, 100 M NADH or NADPH, 20 M FMN, and 2.4 M of enzyme in 25 mM Tris-HCl, pH 7.5. The reaction was run at 30°C, and the initial rate was determined by monitoring the decrease in absorbance at optimal wavelength for each azo dye tested as a substrate: 430 nm for Methyl Red (4-(dimethylamino)azobenzene-2Јcarboxylic acid); 450 nm for Ethyl Red (4-(diethylamino)azobenzene-2Јcarboxylic acid); 485 nm for Orange II (4-(2-hydroxy-1-naphthylazo)benzenesulfonic acid); and 597 nm for Reactive Black 5. NAD(P)H-ferricyanide reductase assay was performed in a reaction mixture (1 ml) consisting of 150 M NADH or NADPH, 100 M ferricyanide (potassium hexacyanoferrate(III), K 3 Fe(CN) 6 ) and 9 nM of enzyme in 25 mM Tris-HCl, pH 7.5. Ferricyanide reduction was followed at 30°C by measuring the decrease in absorbance at 420 nm (⑀ M ϭ 1 mM Ϫ1 cm Ϫ1 ). The effect of FMN on YFL011wp NAD(P)H-ferricyanide reductase activity was tested by the addition of 20 M FMN to the reaction mixture.
YLR011wp was also assayed for nitroreductase activity. The reaction mixture and protocol were identical as for the ferricyanide reductase assay with the exception that ferricyanide was replaced by nitrofurazone and activity was monitored by measuring the decrease in absorbance at 400 nm (⑀ M ϭ 12.96 mM Ϫ1 cm Ϫ1 ).

RESULTS AND DISCUSSION
Overall Structure-The crystal structure of the 21-kDa YLR011w protein was solved by single wavelength anomalous dispersion phasing using Se-Met-substituted protein crystals. The asymmetric unit of the crystal contains two YLR011wp molecules with identical structures (the r.m.s.d. for 229 C␣ positions is 0.44 Å). Refinement of the structure against 2-Å resolution native data yielded R and R free factors of 20.4 and 26%, respectively. All of the residues of the final model are well defined in the 2F o Ϫ F c electron density map and occupy favorable regions of the Ramachandran plot as defined by PRO-CHECK (18). All of the refinement data are gathered in Table I.
Flavodoxins have been classified into two subfamilies: 1) short-chain flavodoxins, which are roughly 150 residues long, and 2) long-chain flavodoxins with an insertion of ϳ20 amino acid residues in the middle of the last strand of the ␤-sheet (20). Although YLR011wp is made up of 191 residues, its final ␤-strand (␤5 in Fig. 1a) has no insert and hence this protein belongs to the short-chain flavodoxin subfamily. The extra 40 amino acid residues of YLR011wp are part of two other insertions relative to classical short-chain flavodoxins. The first insertion (Leu 45 -Leu 85 ) is found in the segment connecting strands ␤2 and ␤3 and containing helix ␣2. This region is longer by 30 amino acids compared with classical flavodoxins. The N-terminal part of this segment is involved in homodimer formation (see below). The second 10-residue insertion (Pro 156 -Ile 166 ) is found in the linker between strand ␤5 and the Cterminal helix ␣5. These two insertions are located at each extremity of the YLR011wp monomer. Because of the head-totail organization of the homodimer (see below), the first insertion from monomer A is facing the second insertion from monomer B (Fig. 2a). This dimer contact forms a closed lid that completely hides the dimethyl benzene part of the isoalloxazine moiety of the FMN molecule (see below).
FMN Binding-Because YLR011wp sequence analysis suggests that it is entirely made up by a FMN binding domain as found in the NADPH-dependent FMN reductases, we incubated the protein with 2 mM FMN prior to the initial crystallization screening. During refinement of the structure, inspection of the electron density maps clearly revealed the presence of a bound FMN cofactor in each monomer. As observed for other flavodoxins, the cofactor lies on top of the C-terminal end of the YLR011wp central ␤-sheet inside a positively charged pocket (Fig. 1a). The FMN phosphoribityl moiety is deeply buried in the protein, whereas the isoalloxazine ring remains partially accessible to the solvent. Each FMN is in contact with both monomers, and only 10% of its surface is solvent-accessible. The main contribution of YLR011wp to FMN binding comes from three loops. The linker that is connecting strand ␤1 to helix ␣1 is in contact with the phosphoribityl moiety, whereas the loops between strand ␤3 and helix ␣3 and between strand ␤4 and helix ␣4 are interacting with both FMN parts. In total, each FMN molecule forms 12 direct hydrogen bonds (Table II) and shields 570 Å 2 of its surface from the solvent upon complex formation with the homodimer.
The isoalloxazine ring is mainly in contact with residues from the loops connecting strand ␤3 to helix ␣3 (Pro 93 -Trp 97 ) and strand ␤4 to helix ␣4 (Gly 125 -Gly 129 ), completed by contacts with Ile 60 and Asp 108 from the other monomer. It is stabilized by five direct hydrogen bonds with the protein and has three water molecules bound. The pyrimidine end of the isoalloxazine is hydrogen-bonded to the Gln 94 side chain (N 3 atom from FMN) and to three main chain amide groups (Asn 96 H-bonds to FMN N 5 and O 4 atoms, whereas Trp 97 and Gly 126 H-bond to FMN O 4 and O 2 atoms, respectively). The re-face of the isoalloxazine ring lies against the hydrophobic side chain from Tyr 95 , whereas its partly solvent-exposed si-face is in contact with Ile 60 from the neighbor monomer.
The phosphoribityl moiety of the molecule is mainly stabilized by electrostatic interactions (Fig. 1b) The FMN phosphate group forms a bidentate salt bridge with Arg 11 and five hydrogen bonds (Fig. 1, b and c, and Table  II) with the Ser 9 , Tyr 95 , and Tyr 124 hydroxyl groups as well as with the Val 15 and Cys 16 amides from the ␤1-␣1 loop. This latter loop is the most invariant feature of the flavodoxin fold. It bears the flavodoxin key fingerprint motif ((T/S)XTGXT), which is always involved in interaction with the FMN phosphate moiety (5). Interestingly, the S. cerevisiae YLR011wp FMN phosphate binding loop sequence (SVRAKRVC) (Fig. 3) does not match the flavodoxin signature and is longer by 2 residues. This results in a clearly different conformation from that of the canonical loop (r.m.s.d of the C␣ atoms of the key fingerprint motif around 3 Å between YLR011wp and flavodoxins), whereas this region varies by Ͻ1 Å within the more classical flavodoxins. Despite these differences, the YLR011wp FMN phosphate binding pocket is reminiscent of the corresponding pocket in the typical flavodoxin proteins (Table II). In flaxodoxins, the FMN phosphate oxygens are in interaction with the amide groups of the five residues (XTGXT) from the loop motif as well as with the O␥1 and N⑀1 atoms of Thr 14 and Trp 57 , two invariant residues in bacterial flavodoxins (Fig. 1d). In YLR011wp, only two hydrogen bonds are made with the amide groups (Val 15 and Cys 16 in place of Val 13 and Thr 14 ) but Arg 11 N⑀ and N1 atoms replace the amide groups from residues at positions 10 and 11 (Gln and Thr in the case of Anacystis nidulans) (5) (Table II and Fig. 1, c and d). Similarly, residues Thr 14 and Trp 57 , which are invariant in classical flavodoxins but not conserved in YLR011wp (Fig. 3), are substituted by Tyr 124 and Tyr 95 (this Tyr 95 is invariant in YLR011wp orthologues) hydroxyl groups as hydrogen bond donor.
The vast majority of flavodoxin domains whose structure is known have the six-residue-long flavodoxin key fingerprint motif ((T/S)XTGXT) in common. The sequence of the fingerprint motif of rubredoxin-oxygen reductase from Desulfovibrio gigas deviates from the canonical motif, but this does not affect the FMN binding mode (21). The structure of YLR011wp presents a yet unobserved variation of the flavodoxin loop in FMNbinding proteins. Although the flavodoxin loop of yeast YLR011wp and B. subtilis Yhda is longer by two residues, both bind the FMN phosphate in a manner similar to other flavodoxins. A sequence motif search for this YLR011wp loop only revealed its presence in a Schizosaccharomyces pombe orthologue (46% sequence identity with YLR011wp) (Fig. 3) and putative reductases from Streptomyces and Bordetella (having an ϳ36% sequence identity).
Oligomerization State-Analytical gel-filtration chromatography revealed that YLR011wp exists as a dimer in solution (data not shown). The asymmetric unit of the crystal contains two copies of the molecule related by a 2-fold symmetry axis. The contact area between these two molecules in the asymmetric unit is 2900 Å 2 , strongly suggesting that this configuration corresponds to the biologically relevant dimer (22). This obser- vation is unexpected as most flavodoxin-related proteins, whose structures have been solved, are monomeric (6,23,24). The homodimer has a half-cylinder-like shape with approximate dimensions of 75 ϫ 45 ϫ 40 Å 3 and a large flat solventexposed surface. Dimerization occurs mainly via anti-parallel side to side packing of helices ␣3 and ␣4 of each monomer (Fig.  2a). 21 residues from each monomer are involved in dimer formation. These residues belong to the linkers connecting strand ␤2 to helix ␣2 (insertion I: Leu 45 ; Ala 49 -Tyr 53 ; and Ile 60 ) and strand ␤3 to helix ␣3 (Tyr 95 -Tyr 99 ) but also to helices ␣3 (Ala 101 , Ala 102 , Lys 104 , Asn 105 , and Asp 108 ) and ␣4 (Gln 135 , Glu 138 , Val 139 , Gly 142 , and Leu 143 ). This interface is mainly formed by the packing of hydrophobic residues supplemented by fourteen hydrogen bonds (Table III).
Most of these residues are either absent or not conserved in the monomeric bacterial flavodoxin sequences. However, some other FMN-binding proteins do exist as homodimers. This is the case for Vibrio harveyi flavin reductase P (25), Escherichia coli nitroreductase NfsA (33), D. gigas rubredoxin-oxygen reductase (21), B. subtilis Yhda (Protein Data Bank code 1NNI), and probably also for the E. coli Trp repressor-binding protein WrbA (26) and S. cerevisiae Ycp4 (27), whose structures are still unknown. A comparison of these dimer structures revealed that E. coli NfsA and V. harveyi flavin reductase P, which share a 52% sequence identity, can be superimposed with an r.m.s.d. as low as 1 Å. As expected from the lack of significant structural resemblance between NfsA (or flavin reductase P) and YLR011wp monomers, the NfsA (or flavin reductase P) and YLR011wp homodimers do not match at all. For the D. gigas rubredoxin-oxygen reductase enzyme homodimer, no significant structural alignment can be found with other homodimeric FMN-binding proteins. Among these proteins, the bacterial Yhda is the only one to adopt the same quaternary structure as YLR011wp (r.m.s.d. Color coding in this figure is the same as in Fig. 1a. b, superimposition of the B. subtilis Ydha (yellow) and the YLR011wp (monomers A and B are in green and blue, respectively) homodimers. Bound FMNs are shown as sticks. c, conservation of the residues among the azoreductase enzymes mentioned in Fig.  3. Coloring is from purple (highly conserved) to gray (low conservation).

Structure of a New Yeast NAD(P)H-dependent FMN Reductase
(aligned dimer structures are represented in Fig. 2b). This is supported by a high degree of sequence conservation for most of the residues involved in homodimer formation (Fig. 3). Hence, all of the Yhda homologous proteins described in Bacillus subspecies should adopt the same dimer architecture as Yhda and S. cerevisiae YLR011wp.
YLR011wp Exhibits Various NAD(P)H-dependent Reductase Activities-Sequence analysis of YLR011wp, whose function is not annotated, assigned this protein to the NADPH-dependent FMN reductase family. The present crystal structure of YLR011wp with bound FMN indeed confirms that it belongs to the flavodoxin family. To our view, YLR011wp forms an interesting test case for deriving biochemical function from the three-dimensional structure.
We first have assayed the enzyme for NAD(P)H-dependent FMN reductase activity by following NAD(P)H oxidation at 340 nm. YLR011wp was able to oxidize NADH and NADPH with a slight preference for the latter. The reaction was strictly dependent on exogenously added FMN (Fig. 4). The specific ac-tivity of 0.9 mol of NADPH oxidized/min/mg protein was measured, a value coherent with FMN reductase-specific activities described for Salmonella typhimurium nitroreductase (28) and E. coli NfsA and NfsB nitroreductases (29). In many enzymes with NAD(P)H-dependent activity, the FMN reduction step is a prerequisite to additional reductive reaction in which FMNH 2 acts as an electron donor (Fig. 5a). Therefore, if YLR011wp has a natural acceptor for electron transfer via FMN, the FMN reductase activity detected reflects either the reduction of free oxidized FMN via enzyme-bound FMN (Fig.  5b) or the rapid exchange reaction between the reduced and oxidized forms of FMN at the protein FMN binding site (Fig.  5c). This latter exchange reaction might not occur when a suitable downstream electron acceptor is present.
We have screened several compounds as electron acceptor candidates for NAD(P)H-and FMN-dependent reduction. We first investigated the YLR011wp capacity to reduce azo dyes in vitro. This assumption was based on the close structural resemblance between YLR011wp and B. subtilis Yhda that has a 52.8% amino  Ethyl Red reductase activity was almost zero in the absence of exogenously added FMN but could be totally restored by the addition of FMN during the assay (Fig. 6). These data strongly suggest that FMN is necessary for YLR011wp azoreductase activity and that the electron transfer from NAD(P)H to Ethyl Red occurs via FMN. The YLR011wp azoreductase-specific activity measured from our experiments (2-3 nmol of Ethyl Red reduced/ min/mg) is several orders of magnitude below the specific activities described for bacterial azoreductases (from a few tens of nmol of azo dye reduced/min/mg up to mmol/min/mg depending on enzyme and substrate) (11, 30 -32). Therefore, Ethyl Red, although reduced specifically by YLR011wp, probably does not represent the physiological substrate for YLR011wp. E. coli NfsA and NfsB nitroreductases are two other exam-ples of enzymes exhibiting weak azo-dependent and standard NAD(P)H-dependent FMN reductase activities but strong ferricyanide reductase activity (29). Therefore, we have tested YLR011wp for ferricyanide and nitroreductase activities. As illustrated in Fig. 7, ferricyanide reductase activity was unambiguously detected and, in our assay, NADPH was a more efficient donor than NADH. The presence of 20 M FMN during the assay increased sensibly the rate of the reaction. Therefore, as already mentioned for YLR011wp azoreductase activity, the electron transfer from NAD(P)H to the ferricyanide acceptor occurs via FMN. Ferricyanide proved to be an excellent acceptor as a specific activity of 625 mol of ferricyanide reduced/ min/mg was calculated from the assay with NADPH as a donor and 20 M FMN (Fig. 7). This level of ferricyanide reductase activity is comparable with the levels described for NfsA and NfsB nitroreductases (29). However, only very weak nitroreductase activity could be detected with YLR011wp using nitrofurazone as an electron acceptor compared with that of NfsA and NfsB (10 -20 nmol/min/mg for YLR011wp and 73 and 13 mol/min/mg for NfsA and NfsB, respectively). From these enzymatic data, it is still difficult to conclude on the precise biochemical function of YLR011wp. Although several specific reductase activities have been characterized using NAD(P)H as an electron donor and FMN as a cofactor, only ferricyanide reductase assays gave high specific activity. We propose that ferric iron, present in the ferricyanide complex, may be the natural substrate for the reduction catalyzed by this protein. Ferric iron reduction is the central reaction in cellular iron metabolism. Microbial cells have developed different systems for iron uptake (33). In budding yeast, Arn1-4 permeases import extracellular ferric chelates and iron siderophores, whereas Frt1 plasma membrane transporter is specific for free ferric iron. In both cases, the cell indeed has to convert cytosolic ferric iron (free or chelated) into ferrous iron before its incorporation into heme-and nonheme iron-containing proteins. The cytosolic protein responsible for ferric iron reduction is still unknown despite previous biochemical studies (34). From our data, YLR011wp is a good candidate for cytosolic ferric iron reductase activity in yeast and further experiments are needed to reinforce this hypothesis.
In conclusion, we clearly show that YLR011wp is a flavodoxin that binds an FMN cofactor despite the absence of the flavodoxin key fingerprint motif ((T/S)XTGXT) usually involved in FMN phosphate binding. The protein exists as a homodimer in the crystal as well as in solution. The biochemical studies aimed to identify YLR011wp in vivo activity showed that the protein bears several reductase activities that are NAD(P)H-dependent and involve FMN as a cofactor. Ferricyanide was by far the best substrate for reduction, strongly suggesting that this protein could be involved in ferric iron assimilation.