Catalytic mechanism revealed by the crystal structure of undecaprenyl pyrophosphate synthase in complex with sulfate, magnesium, and triton.

Undecaprenyl pyrophosphate synthase (UPPs) catalyzes chain elongation of farnesyl pyrophosphate (FPP) to undecaprenyl pyrophosphate (UPP) via condensation with eight isopentenyl pyrophosphates (IPP). UPPs from Escherichia coli is a dimer, and each subunit consists of 253 amino acid residues. The chain length of the product is modulated by a hydrophobic active site tunnel. In this paper, the crystal structure of E. coli UPPs was refined to 1.73 A resolution, which showed bound sulfate and magnesium ions as well as Triton X-100 molecules. The amino acid residues 72-82, which encompass an essential catalytic loop not seen in the previous apoenzyme structure (Ko, T.-P., Chen, Y. K., Robinson, H., Tsai, P. C., Gao, Y.-G., Chen, A. P.-C., Wang, A. H.-J., and Liang, P.-H. (2001) J. Biol. Chem. 276, 47474-47482), also became visible in one subunit. The sulfate ions suggest locations of the pyrophosphate groups of FPP and IPP in the active site. The Mg2+ is chelated by His-199 and Glu-213 from different subunits and possibly plays a structural rather than catalytic role. However, the metal ion is near the IPP-binding site, and double mutation of His-199 and Glu-213 to alanines showed a remarkable increase of Km value for IPP. Inside the tunnel, one Triton surrounds the top portion of the tunnel, and the other occupies the bottom part. These two Triton molecules may mimic the hydrocarbon moiety of the UPP product in the active site. Kinetic analysis indicated that a high concentration (>1%) of Triton inhibits the enzyme activity.

Undecaprenyl pyrophosphate synthase (UPPs) 1 catalyzes the consecutive condensation reactions of eight molecules of isopentenyl pyrophosphate (IPP) with farnesyl pyrophosphate (FPP) to form a lipid carrier to mediate bacterial peptidoglycan synthesis (1)(2)(3). This enzyme belongs to a family of prenyltransferases that make linear IPP condensation products with designate chain lengths (4). These enzymes are divided into trans-type and cis-type, which catalyze the trans-and cisdouble bond formation during each IPP condensation, respectively (5,6). Unlike trans-prenyltransferases, which tend to make short and medium chain-length products ranging from C 15 to C 50 , UPPs and other cis-prenyltransferases mostly generate ՆC 55 long-chain products. Significant sequence homology has been found within the cis-and trans-prenyltransferases, but the two groups of enzymes do not share sequence similarity (7,8). Among the trans-prenyltransferases, the crystal structure of avian farnesyl pyrophosphate synthase has been solved almost a decade ago, and the mechanism has been elucidated (9). However, only recently have the first crystal structures of cis-prenyltransferase (UPPs) become available (10,11), and they provide a template for modeling other cis-enzymes such as dehydrodolichyl pyrophosphate synthase from yeast and human and a polyprenyl pyrophosphate synthase discovered in Arabidopsis thaliana (12,13).
UPPs from Escherichia coli is a dimer of identical subunits of 253 amino acids. The three-dimensional structure of the apoenzyme reveals an elongated tunnel-shaped active site crevice surrounded by two ␣-helices and four ␤-strands (11). Previous site-directed mutagenesis studies suggested that the substrates FPP and IPP are bound on top of the tunnel, and the farnesyl moiety of FPP migrates toward its bottom during product chain elongation (14 -16). The tunnel is sealed at the bottom by the side chain of Leu-137 (11). In the crystal structure of UPPs from Micrococcus luteus, a sulfate ion bound to a conserved structural P-loop represents the location of the pyrophosphate moiety of FPP (10).
The previous E. coli UPPs crystal was grown using polyethylene glycol (PEG), and no sulfate ion was observed (11). Metal ion was not found in both apo-UPPs structures, although the enzyme requires Mg 2ϩ for activity. Two protein conformers of the E. coli UPPs were observed, i.e. one with a bound PEG fragment that adopts a narrower conformation than the other one with water molecules in the active site, indicating the possible open and close mechanism for substrate binding and product release. These two conformers have the most striking difference in the position of the ␣3 helix, which is connected to a loop containing amino acids 72-82. In the proposed catalytic model, the loop may play an essential role in pulling the ␣3 helix toward the active site (17). Perhaps because of the requirement of its flexibility in enzyme catalysis, this loop was invisible in the previous crystal structure of the apoenzyme. Neither was this loop observed in the crystal structure of UPPs from M. luteus, in which both subunits had the closed conformation (10).
To answer the questions regarding the location of substrate binding, the role of the metal ion, and the function of the flexible loop in enzyme catalysis, we report here the structure of UPPs with sulfates, Mg 2ϩ ions, and two molecules of Triton X-100 occupying the tunnel in conjunction with the kinetic results of a few relevant mutants. Triton at low concentration has been shown to increase the UPPs steady-state reaction rate (18). However, a high concentration of Triton in the crystallization condition resulted in the occupancy of Triton in the active site. Therefore, the dose dependence of Triton in altering the reaction velocity was also examined.

EXPERIMENTAL PROCEDURES
Materials-Radiolabeled [ 14 C]IPP (55 mCi/mmol) was purchased from Amersham Biosciences, and FPP was obtained from Sigma. Reversed-phase thin layer chromatography (TLC) plates were purchased from Merck. Taq DNA polymerase was obtained from Invitrogen. The plasmid mini-prep kit, DNA gel extraction kit, and a nickel-nitrilotriacetic acid resin were purchased from Qiagen. Potato acid phosphatase (2 units/mg) was purchased from Roche Applied Science. FXa and the protein expression kit (including the pET32Xa/LIC vector and competent JM109 and BL21 cells) were obtained from Novagen. All commercial buffers and reagents were of the highest grade.
Crystallographic Analysis-For purification of enzymes, our previously reported protocol of using nickel-nitrilotriacetic acid column was followed (18). The purified wild-type UPPs was crystallized using the hanging drop set-up from Hampton Research (Laguna Niguel, CA), while attempts were tried to incorporate FPP into the crystal. In the end, 2 l of mother liquid containing 0.01 M cobalt chloride, 0.1 M MES, and 1.8 M ammonium sulfate at pH 6.5 was mixed with 2 l of protein solution consisting of 10 mg/ml UPPs, 2% Triton X-100, 5 mM MgCl 2 , and 660 M FPP. The mixture was equilibrated against 200 l of the mother liquid at 25°C. Crystals started to appear within 10 days. This condition was different from the previous one for the Se-Met enzyme (11), but the crystals turned out to be isomorphous. Diffraction experiments on the UPPs crystal was carried out at Ϫ150°C on beam line 17B2 of the National Synchrotron Radiation Research Center in Hsinchu, Taiwan. Data were processed and scaled by employing the program HKL2000 (19). For computational refinement, manual modification, and analysis of the crystal structure, the programs CNS (20), O (21), and CCP4 (22) were used.
Using 2.0-Å resolution data and the apoenzyme model that we solved previously, but with all solvent and cofactor molecules removed, an initial R-value of 0.52 was calculated, and it was reduced immediately to 0.39 after rigid body refinement. The Se-Met residues in the original model were replaced with methionines, and subsequent energy minimization yielded an R-value of 0.29 before calculation of the first Fourier map. According to the density level of 1.5 , 327 water molecules were added to the model, and the amino acid side chains were adjusted. Strong densities in the active site tunnel of one subunit (monomer B) clearly showed two Triton molecules, but they were modeled as a series of solvent atoms. Possible densities for sulfate and metal ions were also modeled as water molecules.
Prior to subsequent refinement, 5% of randomly selected reflections were set aside to calculate R free values for monitoring progress of refinement (23). The model yielded R and R free values of 0.227 and 0.256 after simulated annealing and temperature factor refinement when the resolution was increased to 1.8 Å. Explicit models for the two Triton molecules, the sulfate, and the magnesium ions were constructed, and the polypeptide termini were also modified according to the Fourier maps. With the bound cofactors and 393 water molecules, the R and R free values were reduced to 0.200 and 0.231, respectively. At this point, densities for residues 72-82 of monomer B became interpretable in the map, and the corresponding fragment was modeled. Finally, the resolution was increased to 1.73 Å, and more water molecules were added according to 1.0 density level in the 2F o Ϫ F c map. Statistical numbers for the diffraction data set and the refined model are listed in Table I.
Site-directed Mutagenesis of UPPs-UPPs mutants were prepared by using PCR techniques in conjunction with the E. coli Bos-12 UPPs gene template in the pET32Xa/Lic vector. The mutagenic primers used were prepared by MdBio, Inc. The mutagenic oligonucleotides for performing site-directed mutagenesis are as follows: 5Ј-GCCTTTGGGGCTAAAG-CC-3Ј for H43A; 5Ј-GGGGAGGCTCGCATTAGT-3Ј for H199A; and 5Ј-GGGGAGGCTCGCATTAGT-3Ј and 5Ј-GCCTATGCCGCACTTTAC-3Ј for the H199A/E213A double mutant. Subsequently, the forward primer 5Ј-GGTATTGAGGGTCGCATGTTGTCTGCT-3Ј and the reverse primer 5Ј-AGAGGAGAGTTAGAGCCATCAGGCTGT-3Ј were used in combination with the PCR products obtained using the above mutagenic oligonucleotides to create the full-length mutant UPPs genes. The FXa cleavage site (IEGR) and the complementary sequences to the sticky ends of the linear vector pET-32Xa/LIC were included in these primers. Thirty cycles of PCR were performed using a thermocycler (Applied Biosystems) with the melting temperature at 95°C for 2 min, annealing temperature at 42°C for 2 min, and polymerization temperature at 68°C for 40 s. The PCR product was subjected to electrophoresis on 0.8% agarose gel in Tris-acetate-EDTA (TAE) buffer, and the gel was then stained with ethidium bromide. The part of the gel containing the band of the correct size was excised, and the DNA was recovered using a DNA elution kit. The constructed gene of a mutant enzyme was ligated to the vector by incubation for 1 h at 22°C. The recombinant

Structure/Mechanism of Undecaprenyl Pyrophosphate Synthase
UPPs plasmid was then used to transform E. coli JM109 competent cells that were streaked on a Luria-Bertani (LB) agar plate containing 100 g/ml ampicillin. Ampicillin-resistant colonies were selected from the agar plate and grown in 5 ml of LB culture containing 100 g/ml ampicillin overnight at 37°C. The mutation was confirmed by sequencing the entire UPPs mutant gene of the plasmid obtained from the overnight culture. The correct construct was subsequently transformed to E. coli BL21 for protein expression. The 5-ml overnight culture of a single transformant was used to inoculate 500 ml of fresh LB medium containing 100 g/ml ampicillin. The cells were grown to A 600 ϭ 0.6 and induced with 1 mM isopropyl-␤-thiogalactopyranoside. After 4 -5 h, the cells were harvested by centrifugation at 7,000 ϫ g for 15 min. To measure the initial rate, 40-l portions of the reaction mixture were periodically withdrawn within 10% substrate depletion and mixed with 10 mM EDTA for reaction termination. The radiolabeled products were then extracted with 1-butanol, and the radioactivities associated with aqueous and butanol phases were separately quantitated by using a Beckman LS6500 scintillation counter. Initial velocity data were fitted to Equation 1 to obtain K m and k cat values by non-linear regression using KaleidaGraph computer program. The k cat was calculated from where v 0 is the initial velocity, [E] is the enzyme concentration, [S] is the substrate concentration, V max is the maximum velocity; and K m is the Michaelis constant. Enzyme Activity Measurement under Different Concentrations of Triton X-100 -To measure the kinetic constant of E. coli UPPs in the presence of Triton X-100, the enzyme reaction was initiated by adding 0.1 M UPPs to a reaction mixture containing 0.001% Triton X-100, whereas to reaction mixtures containing 0.01, 0.05, 0.07, 0.1, 0.5, or 1.5% of Triton X-100 (v/v), 0.01 M enzyme was added. All reactions were carried out in 5 M FPP and 50 M [ 14 C]IPP along with 100 mM Hepes-KOH buffer (pH 7.5), 50 mM KCl, and 0.5 mM MgCl 2 . Portions of the reaction mixture were periodically withdrawn within 10% substrate depletion and mixed with 10 mM EDTA for reaction termination. The radiolabeled products were then extracted with 1-butanol, and the radioactivities associated with aqueous and organic phases were separately quantitated by using a Beckman LS6500 scintillation counter. Under the saturated concentration of FPP and IPP (Ͼ5-fold K m value), the rate of IPP condensation was determined as k cat (s Ϫ1 ).

RESULTS AND DISCUSSION
Overall Structure-The refined model of UPPs in complex with sulfate, Mg 2ϩ , and Triton contains amino acid residues 13-71 and 86 -240 in one subunit (monomer A) and 17-241 in another (monomer B). Fig. 1, A and B show ribbon diagrams of the dimeric protein structure. There are two sulfate ions in the active site of each monomer and two Triton X-100 molecules bound to the active site tunnel of monomer B. Two magnesium ions are bound to equivalent positions at the dimer interface. Not included in this figures are a fifth sulfate ion and an unidentified compound with a five-member ring, both observed at crystal contact regions, plus 688 water molecules. The two UPPs subunits have very similar structures as those observed in the apoenzyme crystal (11), with monomers A and B assuming the closed and open conformations, respectively. The entire dimers of UPPs can be superimposed with a root mean square deviation of 1.59 Å between bound and apo forms for all 3,383 equivalent non-hydrogen atoms. Excluding the terminal residues that have large deviations, the root mean square deviation is 1.18 Å for 3,300 atoms and 0.594 Å if only 1,656 backbone atoms are compared.
The most significant difference in the protein model of apo-UPPs and the current UPPs model is the presence of the additional loop 72-82 in monomer B, as shown in Fig. 1C. In the final difference Fourier map, although some of the side chains still lack clear densities, the polypeptide backbone is well defined. The loop was not in contact with neighboring molecules, and it became visible probably because the bound Triton molecules stabilized this particular conformation of UPPs in the crystal. The residues of 79 -82 extended the helix ␣3 by more than one turn at the N terminus, which is capped with Pro-78. The succeeding residue, Ser-83, was moved by more than 2 Å away from the active site. Indeed, the entire ␣3 helix was displaced by ϳ1-2 Å before the kink at Glu-96 and ϳ1 Å for residues 97-103. It became more straightened than in the original model. The adjacent helix ␣4 was also shifted by about 1 Å. Thus, the structure of monomer B in the present UPPs crystal has an even more open conformation than that of apo-UPPs. In contrast, monomer A remained largely unchanged as in the apo-UPPs crystal. Only slight shifts were observed in the helices ␣3 and ␣4, with maximal displacements of 1 Å, although the kink of helix ␣3 seems to occur at Glu-96 also rather than at Ala-92, as in apo-UPPs. Apparently, monomer A assumes a closed conformation, as in the apoenzyme. The Bvalues of residues 88 -96 are between 40 and 60 Å 2 , whereas those of 86 -87 are over 60 Å 2 , indicating some disorder in this region. Other regions with B-values higher than 40 Å 2 are located in residues 114 -115 of monomer A and 74 -81 of monomer B, corresponding to the loops of ␤C-␣4 and ␤B-␣3, respectively. Movement of these two helices, which constitute the forewing in the butterfly-shaped UPPs molecule (11), accounts for the conversion between open and closed conformations and, thus, modulates binding and release of the substrate and product from the tunnel-shaped active site.
Adjacent to the flexible loop, residues 69 -71 at the C terminus of strand ␤B show maximal displacements (Fig. 1C). Specifically, the entire residues of Ala-69, Phe-70, and Ser-71 in monomer B were moved by 3.3, 5.8, and 7.5 Å, respectively. The phenyl group of Phe-70 was moved by 14.0 Å from the original position in apo-UPPs, where it was only Ͻ2 Å away from a bound sulfate ion in the current structure. Such structural changes presumably were a result of the bound sulfate ion and Triton molecules in the active site. It now interacts directly with the tert-octyl phenyl head group of Triton (see below). The neighboring side chains of Leu-85, Leu-88, Phe-89, Trp-91, and Ser-95 in the ␣3 helix were also rotated to accommodate the Triton molecule. Similar rearrangements for residues 69 -71 were also observed in monomer A, but the side chains in helix ␣3 remained in their original positions, perhaps because of the absence of the Triton molecule. The N-terminal part of helix ␣3 that encompasses residues 79 -85 and the connecting loop of 72-78 to strand ␤B were not observed in monomer A. Presumably, these residues were flexible, as in the apoenzyme.
The Sulfate Ions-Both FPP and IPP substrates, as well as the reaction intermediates and the final product UPP, have a highly negatively charged pyrophosphate moiety in the molecule. In the previously solved crystal structure of farnesyl pyrophosphate synthase, the allylic and homoallylic substrates are bound via Mg 2ϩ ions, which are coordinated with the active site Asp residues in the conserved DDXXD motifs (9). However, a similar DDXXD motif was not found in cis-prenyltransferases. In contrast, the crystal structure of UPPs from M. luteus (10) showed a sulfate ion bound to a highly conserved structural P-loop that contains the positively charged Arg-32 (equivalent to Arg-30 in the E. coli enzyme). This structural P-loop is supposed to interact with the pyrophosphate of the allylic substrate FPP. A second binding site for pyrophosphate of the homoallylic substrate IPP was also proposed (10,16). It involves two positively charged arginine residues, 197 and 203 (Arg-194 and Arg-200 in E. coli). In both cases, Mg 2ϩ ions were also supposed to participate in substrate binding by bridging the pyrophosphates of FPP and IPP with the acidic side chains of Asp-29 and Glu-216* (Asp26 and Glu213* in E. coli). 2 In our refined UPPs structure there are five sulfate ions, and both monomers have two sulfates bound in the active site, as FIG. 1. Overall structure of UPPs. A and B, two orthogonal views of the enzyme with bound sulfate, magnesium ions, and Triton X-100 molecules. The left monomer, A, is colored in red and green for ␣-helices and ␤-strands, respectively, whereas the right monomer, B, is in magenta and cyan, respectively. A hydrophobic active site tunnel is surrounded by two ␣ helices (␣2 and ␣3) and two ␤ stands (␤B and ␤D) and is flanked with a loop connecting ␤B and ␣3. However, the loop is visible only for monomer B, where two Triton molecules (T1 and T2, shown in green and yellow) are bound. There are two sulfate ions (S1 and S2) bound to each monomer adjacent to the tunnel. Two magnesium ions are bound to His-199 (H199) and Glu-213 (E213) from different monomers at the dimer interface. C, stereo view of the loop connecting ␤B and ␣3. This loop, found previously as a disordered structure, is more ordered with the Triton molecules and ions bound in the active site. The loop encompasses a connection between the ␤B strand and the ␣3 helix, shown in cyan and purple, and the residues of 79 -82 extended the helix by more than one turn. The current model is superimposed on that of apo-UPPs. Conformational changes are shown with the strand and helix colored in green and red, respectively, for the apoenzyme. Side chains having significantly different dispositions in the two models are indicated with bonds colored pink and blue for the apo-UPPs and the current model, respectively. The sulfate ion, which is bound to Arg-102 (R102) and the two water molecules intercalating in the ␣3 helix at the kink, are also shown. This figure as well as Figs. 2, 3B, and 5A were prepared using MolScript (28) and Raster3D (29).
shown in Figs. 1 and 2. The first sulfate ion (S1) forms five well defined hydrogen bonds with the backbone nitrogen atoms of Gly-27, Gly-29, and Arg-30 and the side-chain NE and NH2 atoms of Arg-30. The side chain of the less conserved Arg-39 is 2.8 Å from the S1 sulfate in monomer A, but the distance is 4.0 Å in monomer B. Nevertheless, this Arg-39 as well as the Arg-77 in the flexible loop may both participate in substrate binding and charge neutralization in case of a more negative pyrophosphate. The second sulfate ion (S2) forms four hydrogen bonds with the side-chain nitrogen atoms NH2 of Arg-194, NE and NH2 of Arg-200, and the oxygen OG1 of Ser-202, which is hydrogen bonded to NH1 of Arg-194. One of the S2 oxygen atoms in monomer A is 2.7 Å from the backbone N of the C-terminal Arg-242*, indicating a fifth hydrogen bond. The side chain of Arg-241* is 3.4 Å from the S2 ion. Again, these two arginines and the nearby Arg-239* may interact with the pyrophosphate group of the substrate, although none of them is strictly conserved.
A fifth sulfate ion was observed on the outer surface of monomer B (see Fig. 1C). It is bound directly to the side chain of Arg-102, forming two well defined hydrogen bonds with the NH1 and NH2 atoms. Two additional water-mediated hydrogen bonds were also observed between this sulfate and the backbone oxygen atom of Ser-95. It may also interact with the nearby side chains of Lys-98 and a symmetry-related Lys-33, but no direct bonds were seen. One of the two mediating water For comparison, the model of apo-UPPs is superimposed and shown with pink bonds. Panels A and B were drawn using monomers B and A, respectively. Two Mg 2ϩ ions are bound to equivalent sites at the dimer interface, and one is shown in panel C. C, the octahedral architecture of each cation is constituted by the side chains of His-199 (H199) and Glu213* (E213*) from different subunits and four water molecules. The water molecules are further hydrogen bonded to protein atoms, forming a network that maintains an ordered structure in the nearby region. Bonds colored in pink and blue are for monomers A and B, respectively, whereas the superimposed apo-UPPs model is shown using thin sticks. molecules was also hydrogen-bonded to the backbone nitrogen atom of Ser-99 and a third water molecule, which was bound to the backbone oxygen and nitrogen atoms of Asp-94 and Lys-98, respectively. Insertion of these two water molecules between the normally hydrogen-bonded backbone atoms actually produced the kink at Glu-96 in the ␣3 helix. Similar intercalation of waters in the ␣3 helix was also observed for monomer A, but the densities were weaker.
Compared with the original model of apoenzyme, quite a few rearrangements in the UPPs structure occurred upon sulfate binding. In monomer A, the side chain of Arg-30 remains in original position, but the guanidium group was flipped over to bind the S1 sulfate ion. Originally, it was hydrogen bonded to Asp-26. The side chain of Asp-26 was rotated 120°for the 1 angle upon binding sulfate, and the carboxyl group became bonded to the side chain of Arg-194 (Fig. 2B). The side chain of Arg-200 formed a salt bridge with Glu-213* in apo-UPPs. In the present structure, the guanidium group was moved 4 Å upon swinging the side chain by 120°for the 1 angle to interact with the S2 ion. The side chain of Arg-39 also underwent slight reorientation. In monomer B of the apo-UPPs, the side chain of Arg-30 had a different conformation, which did not interact with Asp-26. Upon sulfate binding, it moved 5 Å toward the S1 ion by rotating almost 180°about the 3 angle and made identical interactions with the sulfate as in monomer A ( Fig.  2A). The side chain of Asp26 in monomer B was also rotated 120°and made hydrogen bonds with Arg-194. The side chain of Arg-200, originally bound to Glu-213* and Glu-240*, was moved 5 Å toward the S2 ion by 120°rotation about the 3 angle. Arg-39 remained almost unchanged, as in monomer A. Consistent with the model proposed for the UPPs from M. luteus, the S1 and S2 sulfate ions observed in our crystal structure may represent the locations of the pyrophosphate groups of the allylic and homoallylic substrates FPP and IPP, respectively, in E. coli UPPs. However, no magnesium ion was observed to make direct interactions with the bound sulfate ions. As shown below, the magnesium ions are bound in other places, and they function in a different way than those in FPPs or other enzymes with pyrophosphate substrates.
The Magnesium Binding Sites-The role of the metal ion has often been argued in metal-requiring enzymes. In prenyltransferases, the common mechanism may be that Mg 2ϩ chelated by Asp residues coordinates with the pyrophosphate moiety of substrate FPP and facilitates the nucleophilic attack by making the pyrophosphate a better leaving group. The trans-type prenyltransferases all have two DDXXD motifs responsible for allylic substrate (FPP) and homoallylic substrate (IPP) binding (24). The Asp residues in the motif play essential role in Mg 2ϩ binding, and the substitution of these Asp residues with Ala led to the remarkable decrease of substrate affinity and turnover number (25,26). However, none of these motifs is found in the cis-type enzymes. In the UPPs from E. coli, a possible candidate for binding Mg 2ϩ in the active site is Asp-26 (or Asp-29 in the M. luteus enzyme). As shown above, this residue did have significant conformation change of the side chain upon binding sulfate, but no Mg 2ϩ ion seemed to be involved. Nevertheless, it remains unanswered whether Mg 2ϩ will participate in binding if the anions are actually pyrophosphates.
In the refined structure of UPPs crystallized in the presence of 5 mM MgCl 2 , there are two Mg 2ϩ ions bound to the enzyme (Fig. 1). The two Mg 2ϩ binding sites are equivalent. They are located in the dimer interface and related by the molecular dyad axis. As shown in Fig. 2C, each ion is octahedrally coordinated with six ligands; one of them is the ND1 atom in the side chain of His199, another is the OE1 atom of Glu213*, and the other four are water molecules. The water molecules are directly hydrogen bonded with the backbone O of Gly197, the side chain OE1 of Glu198 and the backbone N of His-199 in one subunit, as well as the side chain OE2 of Glu213* and the backbone O of Ala235* in another subunit. The hydrogen bond network is further extended with the involvement of ordered water molecules in this region. The Mg 2ϩ binding site is 11 Å from the S2 site for sulfate ion in both monomers, and there is no direct interaction between the bound cation and anion.
In the previous structure of apo-UPPs, the C termini of both subunits were disordered, wherein no densities were observed beyond residue 240 of monomer A and residue 238 of monomer B. In the current structure, the C termini are also disordered. However, some notable rearrangements were observed. In both subunits, the imidazole ring of His-199 was rotated 90°into a proper orientation for coordinating the Mg 2ϩ ion (Fig. 2C). Without the cation, the Mg 2ϩ binding site was occupied by the positively charged side chain of Arg-239*. It was moved 11 Å to the other side of peptide backbone in the presence of Mg 2ϩ ion and redirected toward the S2 sulfate ion in the active site. The side chain of Glu-240* was salt bridged with that of Arg-200 in the apo-UPPs, but in the current model the backbone atoms were displaced by 4.5 Å, and the side chain also moved 10 Å away from the active site, facing the solvent. Similar conformations of Arg-239 and Glu-240 are seen in both monomers, whereas the additional Arg-241 of monomer B is 3.4 Å from the S2 ion in monomer A. As shown above, Arg-200 is directly bonded to the S2 sulfate ion, while Arg-239* and Arg-241* may also be involved. Consequently, the binding of Mg 2ϩ is likely to generate a more ordered structure of the C terminus for interactions with the pyrophosphate substrate.
Effects of Mutants on the UPPs Activity-As shown in Table  II, our results from a previous mutagenesis study showed that substitution of Asp-26 in E. coli UPPs by alanine decreased the k cat to only one-thousandth (10 Ϫ3 ) of that for the wild-type enzyme without significant change of the K m values for FPP and IPP (14). Therefore, Asp-26 is important for catalysis but not for substrate binding. The IPP condensation mechanism of the enzymes for polyprenyl pyrophosphate synthesis has been well established (5,6). A carbocation is first generated by eliminating the pyrophosphate in the allylic substrate, with the assistance of charge neutralization by Mg 2ϩ or protonation of the leaving group. A proton on the second carbon of the ho- moallylic substrate is then subtracted prior to its nucleophilic attack on the allylic carbocation. The carboxyl group of Asp-26, located between the two bound sulfate ions in the active site, is a good candidate to accept proton from the substrate IPP during catalysis.
Although it is uncertain whether Mg 2ϩ ions participate in the catalysis by UPPs, the imidazole group of His-43 seems to serve as a proton donor to the pyrophosphate. It is located between the S1 sulfate ion and the Triton molecule, which represent the pyrophosphate and the hydrocarbon moieties of the substrate, respectively. As shown in Table II, the mutation of His-43 to Ala decreased the k cat to only one thousandth (10 Ϫ3 ) of the original value. A moderate increase of the K m values was also observed. As a proton donor, the imidazole group should be protonated, and, because of the positive charge, it is likely to also participate in substrate binding.
Regarding the role of Mg 2ϩ in substrate binding, our fluorescence studies showed that the FPP substrate still binds to UPPs and quenches its intrinsic fluorescence even in the absence of Mg 2ϩ (17). However, the IPP binding absolutely requires Mg 2ϩ . As discussed above, binding of Mg 2ϩ resulted in significant structural changes in the C-terminal regions of the enzyme, which allowed side chain rotations of Arg-200 and Arg-239* into a proper orientation for binding the S2 sulfate ion. Previous studies showed that Glu-213 is involved in IPP binding, because replacement of Glu-213 with Ala resulted in a significant 70-fold increase of IPP K m value and a 100-fold reduction of k cat (14,16) (Table II). In the current structure, His-199 and Glu-213* constitute two ligands for binding Mg 2ϩ , and we further examined the importance of His-199 by mutating it to Ala. As listed in Table II, the mutant H199A shows a four times larger K m value for IPP. However, its role seems not as important as that of Glu-213. The His-199/Glu-213 double mutant also displays 70-fold larger IPP K m and 1,000-fold smaller k cat values. Therefore, Glu-213 is essential in binding Mg 2ϩ , but His-199 is optional. These results are consistent with the previous observation that both carboxyl oxygen atoms in the side chain of Glu-213 contribute to Mg 2ϩ binding. On the other hand, in all three mutants the K m values for FPP remained nearly unchanged, providing further evidence for the distinction between S1 and S2 site in binding FPP and IPP.
Structures of the Bound Triton Molecules-In the refined structure of UPPs studied here, no FPP molecule was observed in the active site, although the crystallization condition included a significant level of the substrate. Instead, two molecules of Triton X-100 were clearly seen in the electron density maps. These are shown in Fig. 3. The first Triton molecule (T1) has 24 non-hydrogen atoms with an extended conformation for its PEG moiety. The second molecule (T2) has 30 non-hydrogen atoms with the PEG moiety folded back on itself, resulting in a circular shape of the molecule. As shown in Fig. 1, both Tritons are bound to the active site tunnel of monomer B, which has an open conformation. The T1 molecule occupies the lower or inner part of the tunnel, whereas the T2 molecule binds to the upper or outer part, and it appears to block the opening of the tunnel. In the tunnel of monomer A there were elongated densities probably corresponding to fragments of other Triton molecules, but the densities were not clear enough for model building, although the first PEG fragment was observed in this monomer A of the apo-UPPs crystal (11).
Interaction between the Triton molecules and the active site residues of UPPs are mostly hydrophobic, as expected. The "head" moiety of Triton consisting of a phenyl ring and a tert-octyl group is entirely hydrophobic. In the T1 molecule, it interacts with the side chains of the Ala-47, Val-50, Val-54, Ala-92, Leu-100, Leu-107, Leu-139, and Ile-141 of UPPs. Details of the hydrophobic interactions are listed in Table III. Near the opening of the tunnel, the head group seems to be penned up by a salt bridge between Arg-51 and Glu-96 from the helices ␣2 and ␣3, respectively, which constitute part of the tunnel wall. The amphipathic PEG tail of T1, for which the present model contains three ethylene glycol units, is halfexposed to the solvent. The distal part extends away from the tunnel beyond the opposing side chains of Ser-55 and His-103 near its bottom. The head group of the second Triton molecule, T2, also makes direct hydrophobic interactions with that of T1 where the minimal distance between them is 4 Å. It also interacts with the side chains of Ala-69, Phe-70, Phe-89, Ala-92, Leu-93, Ile-109, Phe-116, Leu-120, Ile-124, Ile-141, and Ala-143 in the hydrophobic cleft of the active site (Table III). Above  (right, T2). The models are superimposed on the final 2F o Ϫ F c map contoured at the 1.0 level. Densities in the initial Fourier maps were comparable with the clear densities shown here, especially for the tert-octyl phenyl head groups. During refinement, the PEG tails were extended, yielding final numbers of 24 and 30 atoms in the T1 and T2 molecules, respectively. This figure was produced using BobScript (30) and Raster3D. B, the monomer B is viewed with the active site tunnel in the front. This subunit is in complex with two sulfates (S1 and S2), a magnesium ion (Mg), and two molecules of Triton X-100 (T1 and T2). C, a schematic diagram for the cis-and trans-prenyl moieties of product UPP. The all-trans C 15 portion of the product, which is located in the bottom of the active site cleft, can be represented by the linear Triton molecule, T1. The other circular Triton T2 resembles the cis-prenyl portion of the product. The Triton molecules together may mimic the entire product in the tunnel. this hydrophobic cluster, two adjacent side chains of Trp-75 and Leu-85 tend to form a lid to cover the active site cleft. The PEG tail makes a U-turn at the second ethylene glycol unit, which is opposed by Met-25 and Trp-221. The third unit is in contact with Gly-46 and Val-50, and it is close to the side chain of His-43, with a distance of 4.5 Å. The 4 th and 5 th ethylene glycol units fold back on the phenyl head group, and they also interact with the tert-octyl group of the T1 molecule as well as the side chains of Phe-89 and Ala-92. The remaining 4 -5 units of the PEG tail protruded out of the active site and were not observed because of disorder.
The hydrophobic nature of the Triton molecules and their propensity to replace the natural substrate FPP in binding to the active site cleft suggest that the Triton structure should somehow resemble the reaction intermediates or product of UPPs, as shown in Fig. 3, B and C. Because of the different stereochemistry (cis) UPPs catalyzes from that (all-trans) of substrate FPP, the product UPP likely adopts a folded structure. The length of the observed PEG tail of the T1 molecule is comparable with that of a straightened all-trans farnesyl group of the substrate, whereas the remainder of T1 together with the entire T2 molecule has about the same size of the cis-polyprenyl moiety of the product. Thus, the linear part the T1 and the other parts of Triton molecules mimics the trans-and cisprenyl parts of the product, respectively. In addition, the shortest distance between the Triton molecules and the sulfate ions S1 and S2 are 7.6 and 12.4 Å, respectively. Consequently, it is more likely for the S1 site to serve for binding to the pyrophosphate moiety of the allylic substrate FPP.
High Concentration of Triton Inhibits the Enzyme Activity-Our previous studies have shown that Triton X-100 at a low concentration of 0.1% increased the steady-state k cat of E. coli UPPs reaction by 190-fold (18). The rationale for this activity stimulation by Triton can be attributed to its ability to provide the preferred interaction with the product, which is highly hydrophobic, and thus facilitate product release from the active site, which is the rate-limiting step of steady-state catalysis. In the present structure, two Triton molecules were found to occupy the UPPs active site and prohibit binding of the natural substrate FPP. Therefore, we determined the UPPs activity in the presence of high concentration of Triton X-100 as used in the protein crystallization experiments. As shown in Fig. 4, the enzyme activity is increased with the addition of a low concentration of Triton, which is similar to the previous results. However, when the concentration is higher than 1%, the UPPs  ions induce conformational changes in the C-terminal region to facilitate binding of IPP. The two Triton molecules in the active site cleft may represent the trans-and cis-prenyl moieties of the product. In a plausible mechanism of UPPs, the deprotonated IPP acts as a nucleophile to attack the carbocation formed by FPP with the release of its pyrophosphate group, similar to other prenyltransferase reactions (27). As suggested by our structural analysis, Asp-26 and His-43 play key roles in the enzyme reaction in which Asp-26 acts as general base to deprotonate IPP and His-43 provides a proton to FPP for the dissection of its pyrophosphate group. Both of their mutations to Ala resulted in a significant decrease of k cat (Table II). The loop of ␤B-␣3 containing Glu-73, Trp-75, and Arg-77 may also be involved in the catalysis. Our previous site-directed mutagenesis studies have shown that the mutation of these three residues to Ala led to lower substrate affinity and catalytic activity (11). An overview of the active site structure with the proposed reaction mechanism for UPPs catalysis is shown in Fig. 5.
The subunit conformations with the primary substrate analogue of sulfate (monomer A) and the additional product analogue of Triton (monomer B) are different, particularly in the ␣3 helix. The interconversion between two conformations of open and closed forms is implicated in substrate binding and product release. The conformational change during substrate binding and catalysis of UPPs has been probed previously using steady-state and stopped-flow fluorometers (17). It was shown that FPP binding quenches the fluorescence of Trp-91 in the ␣3 helix, which moves toward the active site during substrate binding and thereby results in a closed conformation to provide better interaction of UPPs with the substrate. After the reaction, the crowding prenyl chain of the product shifts the UPPs structure to an open conformer for product release.
FIG. 5. Active site structure and catalytic mechanism of UPPs. A, active site amino acid residues with sulfate and Mg2ϩ ions. Between the two sulfates (S1 and S2) is located a catalytic Asp-26 (D26), which may act as a general base to activate IPP. His-43 (H43) around the sulfate may act as general acid to protonate FPP. The Mg 2ϩ is coordinated with His-199 (H199) and Glu-213 (E213). One sulfate is surrounded by Arg-30 (R30), Arg-39 (R39), and His-43 (H43), and another is bound with Arg-194 (R194), Arg-200 (R200), and Arg-241* (R241*). On the flexible loop, Glu-73 (E73) and Arg-77 (R77) may also be involved in catalysis, whereas Trp-75 (W75) may protect the reaction intermediates in the active site cleft. B, surface of the active site. The surface is color coded from red to blue according to charge potential from Ϫ15 to 15 k B T, where k B is the Boltzmann constant (ϭ 1.381 ϫ 10 Ϫ23 J⅐k Ϫ1 ) and T is Kelvin temperature. The oxygen and sulfur atoms in the sulfate ions are shown in green and yellow, whereas the two Triton molecules are in cyan and magenta. This figure was generated using GRASP (31). C, the proposed reaction mechanism of UPPs. Based on the crystal structure, the IPP condensation reaction is likely to begin with deprotonation of IPP by Asp-26, followed by attack of the deprotonated IPP on the carbocation formed from FPP by His-43-assisted dissociation of the pyrophosphate group. See "Results and Discussion" for further descriptions. D, model of UPPs turnover. The synthesis of UPP is initiated with binding of the 15-carbon allylic substrate FPP to an empty active site of UPPs. It proceeds with binding of a 5-carbon homoallylic substrate IPP and subsequent condensation with the allylic substrate from which a pyrophosphate ion is released. The condensation is repeated seven more times until it yields the final 55-carbon product UPP. At this point some conformational changes occur in the enzyme, which may involve the ␣3 helix, and allow release of the UPP molecule.
As a summary, the UPPs turnover is shown in Fig. 5D. The binding of FPP followed by an incoming IPP initiates the condensation reaction. The condensation occurs with the release of pyrophosphate from FPP, leading to addition of five carbon atoms to the growing hydrocarbon chain. A similar reaction is repeated by incorporating another IPP molecule, and it proceeds until the FPP chain elongation yields the C 55 final product. The all-trans C 15 portion of the product reaches the bottom of active site cleft, which was represented by the linear Triton molecule (T1). The other curved Triton (T2) resembles the cis-prenyl portion of the product.
The UPPs structure studied here shows two conformations; monomer A is in the "closed" form because it does not contain a well defined Triton or PEG molecule bound in the active site, and monomer B is in the "full-open" form because two productlike Triton molecules are bound. However, in the structure of apo-UPPs, monomer B has an "open" conformation although the active site is empty. Interestingly, another UPPs crystal contains a bound Triton in the active site of monomer A, which also has a closed conformation. 3 Thus, the closed form of monomer A, and the open and full-open forms of monomer B probably correspond to "start," "idle," and "stop" status, respectively. Although the loop of 72-82 can be seen in the present UPPs crystal, the precise functions of the strictly conserved residues in this region remain undetermined. To visualize other conformations and fully elucidate the catalytic process, further crystallographic work on UPPs complexes with various substrate analogues is in progress.