Crystal Structure of 1-Deoxy-d-xylulose 5-Phosphate Synthase, a Crucial Enzyme for Isoprenoids Biosynthesis*

Isopentenyl pyrophosphate (IPP) is a common precursor for the synthesis of all isoprenoids, which have important functions in living organisms. IPP is produced by the mevalonate pathway in archaea, fungi, and animals. In contrast, IPP is synthesized by a mevalonate-independent pathway in most bacteria, algae, and plant plastids. 1-Deoxy-d-xylulose 5-phosphate synthase (DXS) catalyzes the first and the rate-limiting step of the mevalonate-independent pathway and is an attractive target for the development of novel antibiotics, antimalarials, and herbicides. We report here the first structural information on DXS, from Escherichia coli and Deinococcus radiodurans, in complex with the coenzyme thiamine pyrophosphate (TPP). The structure contains three domains (I, II, and III), each of which bears homology to the equivalent domains in transketolase and the E1 subunit of pyruvate dehydrogenase. However, DXS has a novel arrangement of these domains as compared with the other enzymes, such that the active site of DXS is located at the interface of domains I and II in the same monomer, whereas that of transketolase is located at the interface of the dimer. The coenzyme TPP is mostly buried in the complex, but the C-2 atom of its thiazolium ring is exposed to a pocket that is the substrate-binding site. The structures identify residues that may have important roles in catalysis, which have been confirmed by our mutagenesis studies.

Isoprenoids are an extensive class of extraordinarily diverse natural products and have important functions in all living organisms (1)(2)(3)(4). Isopentenyl pyrophosphate (IPP) 2 is a common precursor for the synthesis of all isoprenoids. Although it has long been known that IPP can be generated from the mevalonate pathway, recent studies have revealed a mevalonate-independent pathway for IPP biosynthesis in most bacteria, algae, and plant chloroplasts (2)(3)(4)(5)(6)(7)(8)(9). This pathway is also called the MEP pathway because 2-C-methyl-D-erythritol 4-phosphate (MEP) is its first committed precursor. Because the mevalonate-independent pathway is absent in animals, it represents a promising target for the development of novel antibiotics, antimalarials, herbicides, and other drugs. The herbicide fosmidomycin functions by inhibiting an enzyme in this pathway (10), and it also has activity against malarial infection in an animal model (11).
DXS is highly conserved in plants and bacteria (Fig. 1B). Weak sequence homology (about 20% identity) has also been identified with transketolase and pyruvate dehydrogenase E1 subunit (5-8, 18, 20). These enzymes catalyze similar biochemical reactions, and they all require the coenzyme thiamine pyrophosphate (TPP). However, DXS is distinct from these other enzymes and represents a novel class of transketolase-like proteins (5,6). We report here the first crystal structures of DXS, from E. coli and Deinococcus radiodurans, in complex with the coenzyme TPP.

MATERIALS AND METHODS
Protein Expression and Purification-E. coli and D. radiodurans dxs were amplified from genomic DNA, inserted into vectors pET26b and pET28a, respectively, and overexpressed in E. coli at 20°C. The recombinant proteins were purified by nickel-agarose affinity chromatography, anion exchange, and gel filtration chromatography. E. coli DXS, with a C-terminal His tag, was concentrated to 20 mg/ml in a buffer containing 20 mM Tris (pH 7.5), 250 mM NaCl, 10 mM dithiothreitol, and 5% (v/v) glycerol. D. radiodurans DXS, with an N-terminal His tag, was concentrated to 20 mg/ml in a buffer containing 20 mM Tris (pH 7.5), 50 mM NaCl, 10 mM dithiothreitol, and 5% (v/v) glycerol. The protein samples were frozen in liquid nitrogen and stored at Ϫ80°C. Selenomethionine-labeled E. coli DXS was produced in E. coli B834 cells (Novagen) grown in defined LeMaster medium (21) and purified following the same protocol as that for the native enzyme.
Protein Crystallization-Crystals were obtained at room temperature with the sitting drop vapor diffusion method. Prior to crystallization, the DXS proteins were supplemented with 1 mM thiamine pyrophosphate and 5 mM MgCl 2 . It was discovered that the E. coli enzyme crystallized only after in situ proteolysis by a fungal protease, and SDS gels of the crystals showed two bands (at 20 and 40 kDa). The reservoir solution, containing 20% (w/v) PEG3350 and 200 mM Na,K-tartrate, had been infected by a fungus, and this in situ proteolysis was also crucial in the crystallization of two other proteins in our laboratory (22,23). The crystallization condition was optimized using this solution for the drop, and the reservoir solution contained 30% (w/v) PEG3350 and 200 mM Na,K-tartrate. The crystals were cryo-protected by 25% (v/v) ethylene glycol and flash frozen in liquid nitrogen for data collection at 100 K. They belong to space group P2 1 with cell parameters of a ϭ 86.8 Å, b ϭ 171.2 Å, c ϭ 94.8 Å, and ␤ ϭ 107.2°. There are four molecules of DXS in the asymmetric unit.
D. radiodurans DXS was crystallized against a reservoir solution containing 100 mM Tris (pH 7.5), 200 mM ammonium acetate, 150 mM NaCl, and 20% (w/v) PEG6000 or PEG8000. The crystals were cryo-protected by 25% (v/v) ethylene glycol and flash frozen in liquid nitrogen. They belong to space group P2 1 with cell parameters of a ϭ 78.3 Å, b ϭ 154.1 Å, c ϭ 124.9 Å, and ␤ ϭ 98.9°. There are four molecules of DXS in the asymmetric unit.
Data Collection and Processing-A selenomethionyl singlewavelength anomalous diffraction data set to 2.4 Å resolution was collected at the X4A beamline of the National Synchrotron Light Source on a CCD detector. A native data set to 2.9 Å resolution was collected at the X4C beamline of National Synchrotron Light Source on a MarResearch image plate detector. The diffraction data were processed with the HKL package (see Table 1) (24).
Structure Determination and Refinement-A total of 84 selenium sites were expected for the four E. coli DXS molecules in the asymmetric unit. Two sites were identified by the program Solve in the first cycle (25). By feeding these two sites back to Solve, six sites were identified. Starting with these six sites, Solve was able to locate 76 selenium atoms, and Resolve could automatically trace about 80% of the asymmetric unit after phase improvement. The atomic model was completed manually using O (26), and the structure refinement was carried out with CNS (27).
The structure of D. radiodurans DXS was solved by the molecular replacement method with the program Molrep (28). The E. coli DXS structure, modified to have the same sequence as D. radiodurans DXS with the program Spdbv (29), was used as the search model. The refinement was carried out with Refmac (30), incorporating TLS refinements.
DXS Assay-DXS activity was determined using an end point assay. The pyruvate substrate remaining after the reaction was converted to lactate with lactate dehydrogenase, and the concomitant consumption of NADH was determined by fluorescence. The assays were performed in 384-well microtiter plates (Greiner), and each well contained 25 l of substrate solution (50 M NADH, 60 M pyruvate, 60 M GAP, 10 mM dithiothreitol, 5 mM MgCl 2 , 600 M TPP, 50 mM Tris (pH 7.5)), 5 l of water, and 20 l of enzyme in a buffer of 50 mM Tris (pH 7.5), 600 M TPP, 5 mM MgCl 2 , and 10 mM dithiothreitol. The reaction was allowed to proceed for 60 min at room temperature. Then 50 l of 5 units/ml lactate dehydrogenase was added, and NADH fluorescence was determined 5 min later.

RESULTS AND DISCUSSION
Structure Determination-The crystal structure of E. coli DXS was determined at 2.4 Å resolution by the selenomethionyl single-wavelength anomalous diffraction method (Table 1) (31). We discovered that in situ proteolysis by a fungal protease was essential for the crystallization of this protein (22,23), and two segments of the enzyme, residues 183-238 and 292-317, had no electron density. The remaining parts of the enzyme, 1-182, 239 -291, and 318 -620 are consistent with the 20-and 40-kDa species observed in SDS gels of the crystals (the 239 -291 segment is too small to be visible in the gels), confirming that the two missing segments were removed by the fungal protease. Unfortunately, the first segment is located near the active site, and the second segment is the linker between two domains of the enzyme. Consequently, there was ambiguity about the active site and the domain organization of DXS based on this structure (see below).
To determine the structure of unmodified DXS, we screened through six DXS proteins from other bacterial sources and found that the enzyme from D. radiodurans could be overexpressed in E. coli and purified. Moreover, this enzyme could be crystallized without the need for in situ proteolysis, which allowed us to determine the structure of full-length DXS (Table  1). Clear electron density was observed for the segment linking two domains in the structure (Fig. 2).
The refined atomic models have good agreement with the observed diffraction data and the expected bond lengths and bond angles ( Table 1). The majority of the residues are located in the most favored region of the Ramachandran plot. The DXS monomers in both crystals have similar conformations, with root mean square distance of 0.4 Å among their equivalent C␣ atoms. Moreover, the structures of the E. coli and D. radiodurans DXS enzymes are also similar, with root mean square distances of 0.7 Å among their equivalent C␣ atoms, consistent with their significant sequence conservation (46% amino acid sequence identity; Fig. 1B).
Structure of DXS-The structure of DXS monomer can be divided into three domains, I, II, and III (Fig. 3). All three domains have the ␣/␤ fold, with a central, mostly parallel ␤-sheet that is sandwiched by ␣-helices (supplemental Fig. S1). Domain I (residues 1-319) contains a five-stranded parallel ␤-sheet, and domain II (residues 320 -495) contains a sixstranded parallel ␤-sheet. Domain III (residues 496 -629) contains a five-stranded ␤-sheet, but the first strand is anti-parallel to the other four strands. The larger size of domain I is due to several extended surface segments (supplemental Fig. S1) at the N terminus (residues 1-49), after the first strand (residues 81-122), and in the connection between the fourth and fifth strands (residues 184 -250). In fact, this connection is one of the two segments in E. coli DXS that were removed by proteolysis during crystallization (Fig. 3B). In the structure of D. radio- durans DXS, residues 199 -242 in this connection are also disordered (Fig. 3A). However, the beginning of this connection is ordered in this structure, and some of these residues are located in the active site (especially residues Asn 183 and Met 185 ; see below). A tightly associated DXS dimer is revealed by the crystallographic analyses (Fig. 3), consistent with our solution light scattering studies 3 as well as earlier native gel and gel filtration results (32,33). Each monomer contributes more than 3,900 Å 2 of surface area to the dimer interface, which is mostly hydrophobic in nature. The two DXS monomers are arranged side-by-side in the dimer, such that each domain of one monomer is in contact with its equivalent in the other monomer (Fig. 3A). The structure of the E. coli DXS dimer (Fig. 3B) is similar to that of the D. radiodurans DXS dimer (Fig. 3A), with root mean square distances of 1.0 Å among their equivalent C␣ atoms.
However, the three domains in the DXS monomer are arranged differently as compared with TK and PDH. As a result, the organization of the dimer is also different. In the DXS dimer, domain I of one monomer is located directly above domains II and III of the same monomer (Fig. 3A). In contrast, domain I in one monomer is located above domains II and III of the other monomer in the dimers of TK (Fig. 4A) and PDH (Fig. 4B). An important consequence of the difference in dimer organization is that the active site in DXS is located within the same monomer, whereas that in TK and PDH is located at the interface between two monomers (see below).
One reason for the difference in domain organization may be the exceptionally long linker (95 residues) between domains I and II in TK (Fig. 4A) and PDH (Fig. 4B). In comparison, DXS contains only 20 residues in this linker (Fig.  3A), which is too short for DXS to assume the same domain organization as in TK and PDH. This linker was removed by proteolysis in the E. coli DXS structure (Fig. 3B). However, we observed clear electron density for this linker in the D. radiodurans DXS structure (Fig. 2), and there is no ambiguity in the connectivity between the two domains in DXS.
Residues 199 -243 in domain I of D. radiodurans DXS are not observed in the structure, very likely because of disorder. This missing segment is located close to the dimer interface (Fig.  3A). Therefore, the possibility of a domain-swapped connection in this segment between the two monomers, linking residue 198 in one domain with 244 in the other domain (supplemental Fig. S3), cannot be completely excluded based on the current structural information. With such a connection, the dimer organization of DXS would be similar to that in TK and PDH, although domain I would then have a novel, domainswapped organization.
The Active Site of DXS-The active site of DXS is located at the interface between domains I and II of the same monomer, with no direct contribution from residues in the other monomer of the dimer (Fig. 5A). The central parallel 3 S. Xiang and L. Tong, unpublished data. ␤-sheets of the two domains are oriented such that their C-terminal ends are pointed toward each other, and the TPP coenzyme is located at the bottom of a pocket in this interface (Fig. 5B). Residues in the active site are highly conserved among the DXS enzymes (Fig. 1B), and many of them are also conserved in TK and PDH (supplemental Table S1). This conservation is supported by observations that the dxs-null lethal phenotype in E. coli can be rescued by a PDH mutant (E636Q; see below) (38).
The TPP molecule is bound in a V conformation (Fig. 5A) (39), and this has been observed in all other TPP-dependent enzymes (35)(36)(37). However, the active site in TK and PDH is located at the dimer interface, between domain I of one monomer and domain II of the other monomer, because of the differences in dimer organization of these enzymes (Fig. 4). The amino-pyrimidine ring of TPP interacts with domain II, whereas the pyrophosphate group interacts with domain I. The TPP molecule is mostly buried in the structure; only the thiazolium ring, especially its C-2 atom, is accessible from the solvent (Fig. 5B). This is consistent with the catalytic mechanism where the C-2 atom participates in the reaction (see below).
The pyrophosphate moiety of TPP has numerous polar interactions with the enzyme (Fig. 5A). A magnesium ion is bound between the two phosphate groups, which is also coordinated by the side chains of Asp 154 , Asn 183 , and the main chain carbonyl of Met 185 . These ligands form a square pyramidal arrangement around the magnesium ion, and a water molecule as the sixth ligand would complete the octahedral coordination of this cation, as observed in the structures of TK and PDH (35,36). The Gly 153 -Asp-Gly 155 -Asn 183 sequence in DXS is consistent with the TPP binding motif of GDGX 25-30 N (40). The phosphates are also involved in direct hydrogen bonding interactions with the enzyme, and the Lys 289 side chain may help balance the extra negative charge on the terminal phosphate (Fig. 5A).
One face of the amino-pyrimidine ring of TPP has -stacking interactions with the side chain of Phe 398 , and the other face has van der Waals' interactions with the side chain of Ile 371 , which, together with Ile 187 , also helps to hold the thiazolium ring in place (Fig. 5A). The three nitrogen atoms in this ring are recognized specifically by hydrogen bonds from the enzyme. Especially, the hydrogen bond between the N-1 atom and the side chain of Glu 373 may help the deprotonation of the C-2 atom of the thiazolium ring for catalysis (41). The Glu 373 side chain is also ion-paired with the side chain of Arg 401 , at a distance of 4.5 Å (Fig. 5A).
Substrate Binding Modes-The pyruvate substrate is expected to form a covalent, semi-stable adduct with the C-2 atom of the TPP coenzyme (supplemental Fig. S4) (41), and such a TPP adduct has been observed for TK (42). We soaked crystals of D. radiodurans DXS with pyruvate and observed electron density near the C-2 atom (data not shown). However, the electron density could not be readily interpreted based on either the enamine or the carbanion intermediate (supplemental Fig. S4). We built a model for the enamine adduct between TPP and pyruvate (Fig. 5B), based on the structure of the enamine adduct in TK (42). The adduct can be incorporated into the active site without causing any steric clashes, and no conformational changes were observed in TK either upon the formation of this adduct (42).  (35). B, structure of E. coli pyruvate dehydrogenase E1 subunit dimer (36). The structures are viewed in the same orientation as that for DXS in Fig. 3. The images were produced with PyMol (44).
To obtain the binding mode of the GAP substrate, we soaked DXS with this compound at high concentrations (up to 6 mM), but so far have not been able to observe GAP in the active site. We have therefore produced a model for the binding mode of this substrate based on that of erythrose 4-phosphate in TK (43), which is supported by our mutagenesis studies (see below) and provides significant insight into the catalysis by this enzyme. GAP is located in the pocket (Fig. 5B) and could have interactions with the side chains of His 51 , His 304 , Tyr 395 , Arg 423 , Asp 430 , and Arg 480 (Fig. 5A). The phosphate group of GAP is expected to be located near the Arg 423 and Arg 480 side chains, but it is also exposed to the solvent in this model (Fig. 5B). Biochemical studies showed that DXS can use glyceraldehyde itself as the substrate (6).
Mutagenesis Studies-Our structural analyses have identified the residues in the active site of DXS ( Fig. 5A and supple-mental Table S1). Some of these residues have already been shown to be important for the catalysis by DXS or the related enzymes TK and PDH. For example, His 49 in E. coli DXS, equivalent to His 51 in D. radiodurans DXS (Fig. 5A), is essential for catalysis (20). This residue is involved in GAP binding. It is equivalent to His 30 in TK (supplemental Table S1), which has been proposed to play a role in proton transfer during that reaction (35).
The mutation in PDH that can rescue the dxs-null phenotype (38), E636Q, is also located in the active site. This residue is equivalent to Asp 430 in DXS, which is located in the GAP-binding site (Fig. 5B). Therefore, it may be possible that the E636Q mutation in PDH can alter the substrate specificity of the enzyme to allow it to compensate for the dxsnull mutation, although it remains to be determined whether this mutant can really produce 1-deoxy-D-xylulose 5-phosphate.
To assess the functional roles of the other residues in the active site, we have mutated several of them and characterized their effects on the catalysis of the E. coli enzyme (Fig. 5C). The mutation sites included Glu 370 (equivalent to Glu 373 in D. radiodurans DXS), Tyr 392 (Tyr 395 ), Arg 398 (Arg 401 ), His 431 (His 434 ), and Arg 478 (Arg 480 ). The data show that Glu 370 , Arg 398 , and Arg 478 are crucial for catalysis. In comparison, mutations of Tyr 392 and His 431 have minimal impact on the catalytic activity. These observations are fully consistent with our structural information. The Glu 370 and Arg 398 side chains interact with each other and with TPP (Fig. 5A) and may be important for the activation of this coenzyme for catalysis (41). The functional importance of Arg 478 supports our hypothesis that it helps recognize the GAP substrate (Fig. 5B). On the other hand, the side chains of Tyr 392 and His 431 probably do not participate in direct interactions with the substrate.
DXS is a crucial enzyme for the biosynthesis of IPP, thiamine, and pyridoxol and is an attractive target for the development of novel antibiotics, antimalarials, and herbicides. Our studies define the three-dimensional structures of E. coli and D. radiodurans DXS and the binding mode of the TPP coenzyme and reveal a novel domain organization in these enzymes. The structural information provides a foundation for developing and optimizing inhibitors against this important target.  (42,43), is shown in white and blue, respectively. His 51 , Tyr 395 , Asp 430 , and Arg 480 , which may be involved in binding the GAP molecule, are highlighted. C, relative catalytic activity of active site mutants of E. coli DXS. The activity of the wild-type enzyme is set at 100.