Complex Structure of Bacillus subtilis RibG

Bacterial RibG is a potent target for antimicrobial agents, because it catalyzes consecutive deamination and reduction steps in the riboflavin biosynthesis. In the N-terminal deaminase domain of Bacillus subtilis RibG, structure-based mutational analyses demonstrated that Glu51 and Lys79 are essential for the deaminase activity. In the C-terminal reductase domain, the complex structure with the substrate at 2.56-Å resolution unexpectedly showed a ribitylimino intermediate bound at the active site, and hence suggested that the ribosyl reduction occurs through a Schiff base pathway. Lys151 seems to have evolved to ensure specific recognition of the deaminase product rather than the substrate. Glu290, instead of the previously proposed Asp199, would seem to assist in the proton transfer in the reduction reaction. A detailed comparison reveals that the reductase and the pharmaceutically important enzyme, dihydrofolate reductase involved in the riboflavin and folate biosyntheses, share strong conservation of the core structure, cofactor binding, catalytic mechanism, even the substrate binding architecture.

Bacterial RibG is a potent target for antimicrobial agents, because it catalyzes consecutive deamination and reduction steps in the riboflavin biosynthesis. In the N-terminal deaminase domain of Bacillus subtilis RibG, structure-based mutational analyses demonstrated that Glu 51 and Lys 79 are essential for the deaminase activity. In the C-terminal reductase domain, the complex structure with the substrate at 2.56-Å resolution unexpectedly showed a ribitylimino intermediate bound at the active site, and hence suggested that the ribosyl reduction occurs through a Schiff base pathway. Lys 151 seems to have evolved to ensure specific recognition of the deaminase product rather than the substrate. Glu 290 , instead of the previously proposed Asp 199 , would seem to assist in the proton transfer in the reduction reaction. A detailed comparison reveals that the reductase and the pharmaceutically important enzyme, dihydrofolate reductase involved in the riboflavin and folate biosyntheses, share strong conservation of the core structure, cofactor binding, catalytic mechanism, even the substrate binding architecture.
Recently, we have solved the tetrameric ring-like structure of BsRibG (10). The D domain belongs to the cytidine deaminase superfamily, in which the members catalyze the hydrolytic deamination of the base moiety of a variety of nucleotides including RNA, DNA, mononucleotides, and several therapeutically useful analogues (11). Despite no significant sequence similarity, the R domain is the only protein so far to share high structural homology with dihydrofolate reductase (DHFR). DHFR catalyzes the reduction of dihydrofolate into tetrahydrofolate, and many inhibitors of this enzyme have long been used clinically for the treatment of cancer, inflammation, and microbial infection (12)(13).
The reduction mechanism of the ribose ring into a ribityl group is still inconclusive. Accumulation of an Amadori derivative in riboflavin-requiring B. subtilis mutants suggested that in bacteria, the reaction involves an Amadori rearrangement initiated by proton abstraction from C 2Ј , followed by reduction of the C 2Ј carbonyl group (14). In contrast, feeding experiments using deuterium incorporation in the yeast Ashbya gossypii, suggested that the reduction occurs by hydride transfer to C 1Ј , ruling out an Amadori mechanism (15). Recently, kinetic isotope studies of Escherichia coli RibD suggested a direct hydride transfer from the C 4 -pro-R hydrogen of NADPH to C 1Ј , and that this is the rate-limiting step in the overall deaminationreduction reaction (16). To gain structural insights into the substrate specificity, catalytic mechanisms, and inhibitor design, we have determined the structure of BsRibG as a complex with AROPP at 2.56-Å resolution. Together with the previous NADPH complex structure (10), a reduction mechanism is proposed. In the D domain, mutational analysis of the predicted substrate binding residues was also carried out.

EXPERIMENTAL PROCEDURES
Preparation of Protein and Substrates-Site-directed mutagenesis was carried out using a QuikChange mutagenesis kit (Stratagene). Protein expression, purification, enzyme activity assay, and crystallization of the N-terminally His-tagged recombinant BsRibG were performed as previously described (10). The E51A mutant and wild-type crystals were grown in 28% polyethylene glycol 400, 200 mM CaCl 2 , and 100 mM HEPES (pH 7.5) with a protein concentration of 20 -25 mg/ml. The DAROPP compound was prepared from a reaction solution containing 5 mg/ml of recombinant E. coli cyclohydrolase II, 10 -30 mM GTP, 8 mM MgCl 2 , and 20 mM Tris-HCl (pH 8.0). The reaction was complete at around 10 min, and then 2 mg of BsRibG was added for production of AROPP. The reactions were monitored by UV absorption spectra recorded on a Beckman DU640B spectrophotometer.
Activity Assay-For measurement of the relative deaminase activity of the mutants, 20 g BsRibG was incubated into the 1-ml reaction solution containing 0.1 M Tris-HCl (pH 7.5), 100 mM NaCl, 10 mM dithiothreitol, and ϳ0.2 mM DAROPP for 20 min at room temperature. The deaminase activity was measured by monitoring change in absorbance at 292 nm. For the reductase assay, after the deamination was complete, 1 mM NADPH was added into the reaction mixture for 15-60 min at 37°C. Diacetyl was added with a final concentration of 1% (v/v), and the mixture was incubated at 95°C for 30 min. After centrifugation to remove protein precipitant, the fluorescence spectrum was recorded with an excitation wavelength of 408 nm. The relative reductase activity was estimated by the emission intensity at 485 nm.
Structure Determination-The reaction solution was added to the crystal drops for 30 -60 min with a final concentration of ϳ10 -15 mM DAROPP or AROPP. X-ray diffraction data from several crystals between 2.56 -2.68 Å resolution were collected and processed at the beamlines BL13B1 at NSRRC (Hsinchu, Taiwan), and NW12 at the Photon Factory (Tsukuba, Japan). The crystals belong to the P2 1 2 1 2 1 space group with one tetramer per asymmetric unit. The complex structure was determined using the molecular replacement method and refined using CNS (17). The statistics for the best data set are summarized in Table 1

RESULTS AND DISCUSSION
Substrate Preparation-Production of DAROPP and AROPP was monitored by UV-VIS absorption spectra because the distinct nucleobases of GTP, DAROPP and AROPP result in different spectra with different maximum wavelengths of 254, 293, and 284 nm, respectively. The DAROPP concentration was estimated using an absorption coefficient ⑀ 293 of 9600 M Ϫ1 cm Ϫ1 (20,21), because GTP has hardly any absorption at this wavelength. The yields were consistent with previous studies, in which DAROPP formation accounts for ϳ90% of the E. coli cyclohydrolase II products (21). The AROPP concentration was estimated using NADPH consumption catalyzed by the R domain in term of the decrease in absorbance at 340 nm. The D domain of BsRibG produces AROPP as its main product at Ͼ80% of total conversion. In the absence of dithiothreitol, the reaction solution gradually turned yellow in a couple of hours with the maximum wavelength shifted from 284 to 245 nm and with a shoulder at 270 nm. Addition of 30 mM dithiothreitol into the reaction solution prevented the formation of the yellow color and the shift of the UV spectra.
Mutational Analysis in the D Domain-Asn 23 and His 42 , His 76 and Lys 79 , and Asp 101 and Asn 103 were previously predicted to interact with the pyrimidine, phosphate and ribose of the substrate DAROPP, respectively (10). These six residues as well as the invariant glutamate in the family signature HXE, were substituted with alanine. No detectable deaminase activity was observed in the E51A and K79A mutants, whereas the other five mutants showed a decreased relative activity by 2-3fold compared with the wild-type enzyme. As with other family members (22)(23)(24)(25)(26), the invariant glutamate in BsRibG, Glu 51 , may be also involved in the proton shuttle during the deamination process (Fig. 1), and hence substitution with alanine eliminates the deaminase activity. Previous studies showed that the dephosphorylated form cannot serve as a substrate (27). Thus as predicted, Lys 79 may interact with the phosphate group and such interactions would seem to be essential for the deaminase activity. In addition, in the soaking experiments, neither did AROPP bind to the D domain of the wild-type BsRibG, nor did DAROPP bind to the E51A mutant. In these crystal structures, the RibG unique loop containing His 76 and Lys 79 is flexible and swings away from the active site, and hence a substrate binding cavity may not be fully formed.
The Substrate Binding Site in the R Domain-BsRibG exists as a tetramer in solution (10). The four subunits (Molecules A-D) in the crystal asymmetric unit form a ring-like tetramer (Fig. 2). Significant strong electron density for the ligand was observed only in the R domain of Molecule C. The substrate AROPP was initially modeled into the electron density, but the pyrimidine and the phosphate groups could not be modeled simultaneously into their respective electron dense areas (Fig. 3A). In addition, the carboxyl group of Glu 290 would seem to form unfavorable interactions with the ribosyl O 4Ј , and the ribosyl geometry is distorted during the structure refinement. Feeding experiments with A. gossypii suggested that the reduction occurs at C 1Ј , and the authors proposed a formation of a Schiff base intermediate (15). Interestingly, this proposed ribitylimino intermediate could be modeled nicely into the electron density (Fig. 3A) (Fig. 3B) Superposition of the four R domains in the tetramer showed multiple conformations that covered two regions, residues 167-171 and 202-212 (Fig. 3C). In Molecule C, upon AROPP binding, residues 167-171 and 202-212 moved a small amount to allow interaction with the pyrimidine ring and the phosphate group, respectively, particularly, the backbones of Ser 167 , Ile 170 , Thr 171 , Ser 202 , and Leu 203 . Similarly, when 10 mM NADPH was added into the BsRibG crystals for 4 days, strong density was also detected only in the R domain of Molecule C (10). Upon NADPH binding, residues 156 -170 moved toward the cofactor and hence some conformational changes were induced, particularly affecting the backbones of Leu 160 and Gly 165 . The backbones of residues 193-195 also shifted toward the pyrophosphate of the cofactor.
Structure Comparison within RibG, and Rib7-Several other crystal structures of R domain homologs have also been determined: E. coli RibD (EcRibD, PDB 2G6V for apo; 2OBC for binary ribose 5-phosphate; and 2O7P for binary NADP ϩ ) (28), Thermotoga maritima RibG (TmRibG, PDB 2HXV for binary NADPH), Methanocaldococcus jannaschii Rib7 (MjRib7, PDB 2AZN for binary NADPH) (29) and the putative Rib7 from Corynebacterium diphtheriae (PDB 2P4G). The R domains and Rib7s as well as DHFR, share diverse loop conformations but a virtually identical structure core including the central eight ␤-strands (␤A-␤H) and the four flanking helices (␣B, ␣C, ␣E, and ␣F), with the secondary elements numbered as found in DHFR (30) (Fig. 4). The N-terminal half constitutes the majority of the active site and hence is more conserved with a sequence identity of 30 -50%, whereas the C-terminal half is more diverse, with 15-30% identity (Fig. 4A). The most diverse region is the RibG/Rib7-unique insertion between the ␤D and ␣E, which has various lengths and a low sequence homology. This insertion forms a helix (␣DЈ) and a ␤-strand (␤DЈ) in BsRibG and MjRib7, but it is only a strand and a long loop in TmRibG and EcRibD (Fig. 4B). The second diverse section is the C-terminal tail including the ␤G and ␤H strands and L ␤F-␤G , which together with the first loop, L ␤A-␣B , and the ␤F strand, constitute the R-R interfaces. Thus the diverse C-terminal half of the molecule results in the distinct R-R interfaces, and may contribute to the different oligomerization states, a tetramer with BsRibG but a dimer with EcRibD and TmRibG.    (Fig. 4C). In contrast, the remaining parts of the NADPH cofactor occupy similar spatial positions and form similar interaction networks, particularly with the protein backbones. The binding architectures of the reactive nicotinamide rings in BsRibG, TmRibG, and MjRib7, and even in the DHFRs, are virtually identical.
Superposition of the AROPP intermediate in BsRibG against the ribose 5-phosphate in EcRibD reveals that the phosphate groups occupy a similar position and form virtually identical interactions with two conserved arginines and two backbone NHs (Fig. 4C). However, the sugar moieties point in opposite directions and hence distinct residues have been proposed to be involved in the proton transfer, namely Asp 200 in EcRibD (28) and Glu 290 in BsRibG. Ribose-5-phosphate may not be a suitable substrate analogue because it cannot mimic substrate binding due to the lack of the molecule pyrimidine ring and a ribitylimino intermediate. Moreover, the substrate AROPP could not be docked into the current EcRibD crystal structures (28), because the active site cavity is partially blocked by L ␤A-␣B and hence there is no space to accommodate the pyrimidine ring (Fig. 4B).
The observed multiple "inactive" conformations of EcRibD may be due to the crystal packing effects. Unlike BsRibG and TmRibG, Molecules A and B in the dimeric EcRibD display very distinct conformations for L ␤A-␣B (residues 159 -173) and L ␤F-␤G (residues 322-348) (Fig. 4B). These two loops form several hydrogen bonds and constitute a The R-R interface in EcRibD is much less extensive than those present in BsRibG and TmRibG, with the buried areas of ϳ2600 Å 2 versus ϳ4000 Å 2 and ϳ3800 Å 2 , respectively, calculated by the PISA server (31). When NADP ϩ was soaked into the EcRibD crystal, the cofactor binding induced L ␤A-␣B A to move toward the cofactor allowing interaction with the nicotinamide moiety via Thr 161 , Ala 164 , and Trp 170 (Fig. 4B); this consequently caused conformational shifts of L ␤F-␤G A , L ␤F-␤G B , and L ␤A-␣B B . In BsRibG, only subtle structural shifts in L ␤A-␣B were observed on binding of AROPP and NADPH. Furthermore, the current MjRib7 structure is also affected by crystal packing; MjRib7 exists as a dimer in solution, whereas a trimer is observed in the crystal (29).
Substrate Specificity-During riboflavin biosynthesis, the deamination and reduction steps proceed in the opposite order when eubacteria and yeast are compared. Their substrates are distinct (Fig. 1), namely DAROPP with a cyclic ribose binding the D domain of RibGs versus DARIPP with a linear ribitol binding yeast Rib2; this has lead to the two deaminases evolving different residues for substrate recognition (10). In contrast, the reductase substrates, AROPP for the R domain and DAROPP for Rib7, display similar structures with the only difference being a carbonyl versus an amino group; in this case, these two reductases share several conserved residues for substrate binding (Fig.  4A). Particularly, Thr 171 (or Ser), Arg 183 , Asp 199 , Arg 206 , and Glu 290 in BsRibG are all conserved across plants, eubacteria, archea, and fungi, indicating that RibG and Rib7 share a similar reduction mechanism through the ribitylimino intermediate. However, the R domain has to discriminate the deaminase product AROPP from the substrate DAROPP. The complex structure here clearly demonstrates that an invariant lysine (Lys 151 ) in eubacteria and plants but not in fungi and archaea, has evolved to ensure substrate specificity through favorable interactions of its amino side chain with the AROPP O 2 , and The protein length is listed in brackets and the PDB accession code is in parentheses. The extra 13-residue insertion in EcRibD is indicated as #. Residues for the conserved hydrophobic core are shaded in yellow, while the residues involved with NADPH are in cyan. For BsRibG and DHFR, the residues interacting with the nucleobases, the middle parts, and the negatively charged moieties of their substrates are shaded in blue, green, and red, respectively. B, stereo view of the R domain superposition of BsRibG (red), EcRibD (green), and TmRibG (blue). Four distinct conformations were observed for L ␤A-␣B and L ␤F-␤G in the available EcRibD structures (28). In the apo structure, crystal contacts may result in the observed L ␤A-␣B B conformation (yellow), and hence affect L ␤F-␤G B (yellow) and even L ␤F-␤G A and L ␤A-␣B A (magenta). Both L ␤A-␣B A and L ␤F-␤G A are quite mobile, and several residues (164 -166 and 339 -345) are even disordered. The NADP ϩ binding induced L ␤A-␣B A to move toward the cofactor (green); there is a consequent shifting of L ␤F-␤G A with residues 338 -346 disordered (green), and of L ␤F-␤G B and L ␤A-␣B B with residues 162-169 disordered (cyan). C, comparison of the substrate and cofactor binding in BsRibG (magenta), EcRibD (green), and TmRibG (cyan). The nicotinamide C 4 and the reactive C 1Ј are highlighted in black.
unfavorable interactions with the DAROPP NH 2 2 (Fig. 3B). Furthermore, MjRib7 can only recognize DAROPP but not AROPP as substrate (6). Based on the structural comparison described above, DAROPP was modeled into the active site of MjRib7. This predicted complex structure revealed a short distance of ϳ3 Å between Asp 37 O ␦2 and the NH 2 2 , suggesting that this aspartate residue, conserved in archaea but not fungi, may be responsible for the substrate recognition. Further mutational analyses are under way.
A Similar Substrate Binding Architecture in RibG and DHFR-The R domain and DHFR share high homology not only in core structure but also in active-site architecture. Even the R-R structure is similar to that of the dimeric DHFR from T. maritima (TmDHFR) (32), due to the presence of similar regions for subunit association. In addition to the four conserved regions for NADPH binding, interestingly, even though their substrates are very distinct, AROPP versus dihydrofolate, the binding residues for the individual parts are located in similar spatial positions (Fig. 4A). The residues responsible for the nucleobase are located between the ␤A strand and the ␣B helix, for example, Lys 151 and residues 167-171 for the pyrimidine ring in BsRibG are comparable to Ile 5 , Met 20 , Asp 27 , Phe 31 and Ile 94 for the 6-methylpterin in the E. coli DHFR (EcDHFR) ternary complex with folate and NADP ϩ (PDB 7DFR) (33). Residues for the middle part are located at the C terminus of the ␣C helix; these are Asp 199 for the ribitol in BsRibG and Ile 50 for the p-aminobenzoate in EcDHFR. Interestingly, both reductases possess a positively charged groove for neutralization of the negatively charged group of the substrate: Two or three conserved arginines (or lysine) located at ␣B and L ␣C-␤C , Arg 183 and Arg 206 in BsRibG and Arg 32 , Arg 53 , and Arg 58 in EcDHFR, form salt bridges with the phosphate group and the glutamyl carboxylate moiety, respectively. These salt bridge interactions are essential for the reduction activity because BsRibG cannot utilize the dephosphorylated form as substrate (27).
The Reduction Mechanism in RibG-For BsRibG, the substrate and cofactor binary complex structures were superimposed to mimic the ternary complex to obtain a geometry estimation of hydride transfer (Fig. 5). This ternary complex was then compared with those in the DHFRs where the enzymatic mechanism has been chemically and structurally studied in detail (30). In addition to the virtually identical binding of the cofactor nicotinamide ring, surprisingly, the reactive carbon atoms of the different substrates occupy a similar position and share a virtually identical orientation toward the nicotinamide C 4 (Fig. 5). In BsRibG, the pyrimidine is parallel to the nicotinamide ring with a distance of 3.3 Å between the C 4 donor and the C 1Ј acceptor with a N 1Ј -C 4 -C 1Ј angle of 115 degrees. The corresponding distance and angle are 3.3 Å and 114 degrees for the EcDHFR ternary complex (33), and 3.2 Å and 116 degrees in the DHFR complex from Pneumocystis carinii (34). Therefore, as well as DHFR, RibG would seem to catalyze the ribosyl reduction by C 4 -pro-R hydride transfer from NAD(P)H to the C 1Ј , which is consist with the isotope studies (16).
On the basis of the previous studies on DHFR and the complex BsRibG structures described above, a reduction mechanism for the R domain can be proposed as outlined in Fig. 1. The closest ionizable residue in the vicinity of the NH 1Ј group of AROPP is the strictly conserved Glu 290 , with a distance of ϳ4.6 Å between N 1Ј and Glu 290 O ⑀1 . A water molecule could be placed to mediate hydrogen bonds between Glu 290 and the NH 1Ј group, and thereby together with Glu 290 , may assist in proton transfer, abstracting a proton from the NH 1Ј group on the one hand and on the other hand, protonating the O 4Ј to yield the ribitylimino intermediate. In addition, Glu 290 would seem to repel the ribose because of a short distance between its carboxyl group and the ribosyl O 4Ј (ϳ3 Å), and hence may facilitate the formation of the Schiff base intermediate. Analogous with the DHFR reaction, formation of the product ARIPP occurs by hydride transfer of the nicotinamide H 4Ј to the C 1Ј of the intermediate and protonation of the N 1Ј by a water molecule.
The strong conservation between the R domain and DHFR suggests that these two reductases involved in riboflavin and folate biosyntheses might have evolved by gene duplication with conservation of the core structure, catalytic mechanism, and cofactor binding, but with subsequent divergence in which the substrate binding residues were changed, although with some aspects of their recognition properties being retained.