Three-dimensional Structure of a Hyperthermophilic 5′-Deoxy-5′-methylthioadenosine Phosphorylase from Sulfolobus solfataricus*

The structure of 5′-deoxy-5′-methylthioadenosine phosphorylase from Sulfolobus solfataricus (SsMTAP) has been determined alone, as ternary complexes with sulfate plus substrates 5′-deoxy-5′-methylthioadenosine, adenosine, or guanosine, or with the noncleavable substrate analog Formycin B and as binary complexes with phosphate or sulfate alone. The structure of unliganded SsMTAP was refined at 2.5-Å resolution and the structures of the complexes were refined at resolutions ranging from 1.6 to 2.0 Å. SsMTAP is unusual both for its broad substrate specificity and for its extreme thermal stability. The hexameric structure of SsMTAP is similar to that of purine-nucleoside phosphorylase (PNP) from Escherichia coli, however, only SsMTAP accepts 5′-deoxy-5′-methylthioadenosine as a substrate. The active site of SsMTAP is similar to that of E. coli PNP with 13 of 18 nearest residues being identical. The main differences are at Thr89, which corresponds to serine in E. coliPNP, and Glu163, which corresponds to proline in E. coli PNP. In addition, a water molecule is found near the purine N-7 position in the guanosine complex of SsMTAP. Thr89 is near the 5′-position of the nucleoside and may account for the ability of SsMTAP to accept either hydrophobic or hydrophilic substituents in that position. Unlike E. coli PNP, the structures of SsMTAP reveal a substrate-induced conformational change involving Glu163. This residue is located at the interface between subunits and swings in toward the active site upon nucleoside binding. The high-resolution structures of SsMTAP suggest that the transition state is stabilized in different ways for 6-amino versus6-oxo substrates. SsMTAP has optimal activity at 120 °C and retains full activity after 2 h at 100 °C. Examination of the three-dimensional structure of SsMTAP suggests that unlike most thermophilic enzymes, disulfide linkages play a key in role in its thermal stability.

SsMTAP has been classified as a hyperthermophilic enzyme since its optimum temperature (120°C) is higher than the boiling point of water (1). The enzyme is also characterized by extreme thermostability, demonstrating full activity after 2 h at 100°C and showing an apparent melting temperature of 132°C. In addition, the enzyme remains stable in the presence of protein denaturants, detergents, and organic solvents, even at elevated temperatures. While human MTAP requires reducing agents in order to maintain activity (5), SsMTAP does not. In fact, the presence of several disulfide bonds is believed to play a role in the extreme thermostability and durability of the enzyme.
Previous work suggested that SsMTAP is a hexamer (160 kDa) consisting of six identical subunits of 27 kDa each. Early biochemical studies on the hexameric SsMTAP also revealed that this enzyme shares the broader substrate binding characteristics of hexameric PNPs (1). In fact, although SsMTAP is classified as an MTA phosphorylase, it shares negligible sequence homology with human MTAP (14% identity) and reveals a significant sequence identity to Escherichia coli PNP (32%). Like E. coli PNP, SsMTAP will utilize inosine, guanosine, and adenosine as substrates. Unique to SsMTAP, however, is the ability to cleave MTA as well.
MTA phosphorylases are members of the purine-nucleoside phosphorylase (PNP) family of enzymes, which function in the salvage pathways of cells (6). This family of proteins can be roughly divided into two classes, the hexameric PNPs and trimeric PNPs, based on the quaternary structure of the physiologically relevant form of the enzyme. Interestingly, the monomers of trimeric and hexameric PNPs are structurally similar even though the oligomeric assemblies are unrelated (7). Results of early biochemical studies and sequence homology suggest that SsMTAP belongs to the hexameric class of enzymes (1). In addition, this enzyme is the first hexameric nucleoside phosphorylase shown to utilize MTA as a substrate. In contrast, human MTAP is a member of the trimeric class of enzymes. Recently, the complete genome sequence revealed that S. solfataricus contains a second enzyme that is homologous to trimeric human MTAP (GenBank TM accession number AE006641). However, no obvious PNP is encoded in the S. solfataricus genome.
Many of the enzymes in the PNP family have been well characterized, and crystal structures have been obtained for both trimeric and hexameric PNPs. Atomic resolution structures of human (8), bovine (9 -11), and Cellulomonas sp. (12) PNPs have been determined, and a human MTAP structure has recently been reported (13). Each of these structures reveals a similar trimeric arrangement of subunits. The crystal structure of E. coli PNP has also been described and represents an example from the hexameric class of enzymes (7,14). The structure of uridine phosphorylase from E. coli (15) represents a second example of the hexameric class. The structures of thymidine phosphorylase from E. coli (16,17) and pyrimidinenucleoside phosphorylase from Bacillus stearothermophilus (18) show that these enzymes belong to a separate structural family.
Here, we report the crystal structures of unliganded SsMTAP and binary and ternary complexes of SsMTAP bound to substrates and substrate analogs. Our studies on MTAP from S. solfataricus are aimed toward understanding the structural basis for its broad substrate specificity and toward understanding the factors that contribute to its extreme thermostability.

EXPERIMENTAL PROCEDURES
Protein Production and Crystallization-Recombinant SsMTAP utilized for these structural studies was expressed and purified according to the methods of Cacciapuoti et al. (19). The recombinant enzyme is similar to the wild-type enzyme regarding molecular weight, hexameric structure, presence of intersubunit disulfide bonds, substrate specificity, and specific activity. However, the recombinant SsMTAP demonstrates both lower thermophilicity and thermostability (19).
The protein was concentrated to ϳ7-10 mg/ml using ultrafiltration. All SsMTAP crystals were grown at either room temperature or 18°C using the hanging drop vapor diffusion technique. The drops contained 1.2 l of protein solution and 1.0 l of reservoir solution suspended over 850 l of reservoir solution. For the native SsMTAP crystals, the reservoir solution contained 28 -30% dioxane, 12% 2-methyl-2,4-pentanediol, 0.12 M MgCl 2 , 0.04 M NaCl, and 0.1 M Tris-HCl, pH 7.4. Microseeding with previously grown crystals improved the size and diffraction quality of the crystals. Native crystals of SsMTAP are platelike with typical dimensions of 0.25 mm ϫ 0.25 mm ϫ 0.1 mm. The 12% 2-methyl-2,4-pentanediol in the mother liquor acts as a suitable cryoprotectant when freezing the crystals.
Preparation of Protein-Ligand Complexes-Crystals containing SsMTAP complexed with guanosine, adenosine, MTA, or Formycin B (FMB) plus sulfate ion were generated by co-crystallization. The conditions were identical to those given above for the native SsMTAP crystallization except that the reservoir solution contained 28 -30% dioxane, 12% 2-methyl-2,4-pentanediol, 0.12 M MgCl 2 , 0.03 M MgSO 4 , and 0.1 M Tris-HCl, pH 7.4. In order to obtain the structure of the enzyme bound to phosphate, magnesium sulfate was replaced by ammonium chloride and potassium phosphate was added to the reservoir solution. The final reservoir solution contained 28 -30% dioxane, 12% 2-methyl-2,4-pentanediol, 0.12 M NH 4 Cl, 0.02 M KH 2 PO 4 , and 0.1 M Tris-HCl, pH 7.4. We also attempted to prepare SsMTAP crystals containing FMB and phosphate using these conditions but only phosphate was observed in the active site (see below).
X-ray Data Collection and Structure Determination-Data for unliganded SsMTAP were collected using a single frozen native SsMTAP crystal at the National Synchrotron Light Source (Brookhaven National Laboratory, Upton, NY). The crystals diffracted to 2.5-Å resolution and a complete data set was collected using the oscillation method. Data were subsequently processed using DENZO and SCALEPACK (20). Crystals of the native SsMTAP belong to the orthorhombic space group C222 1 . Data for SsMTAP crystals containing sulfate and guanosine, adenosine, MTA, or FMB were collected at the Cornell High Energy Synchrotron Source (CHESS) (Cornell University, Ithaca, NY). Data for the SsMTAP crystals grown in the presence of phosphate with and without FMB were also collected at CHESS. All data for SsMTAP complexes were collected using frozen crystals. The CHESS data were processed with MOSFLM (21) and scaled using SCALA (22) from the CCP4 suite of programs (23).
The native SsMTAP crystals and SsMTAP crystals containing MTA/ sulfate, FMB/sulfate, FMB/phosphate, and sulfate alone all crystallized in the orthorhombic space group C222 1 . The SsMTAP crystals containing guanosine/sulfate, adenosine/sulfate, and phosphate alone all crystallized in the monoclinic space group P2 1 . The unit cell parameters for the two space groups are related with a O ϳ a M , b O ϳ 2b M , and c O ϳ c M sin␤ M . There are three subunits per asymmetric unit in the orthorhombic form and the hexamer is formed using a crystallographic 2-fold axis. There is an entire hexamer in the asymmetric unit of the monoclinic form. The Matthews number (V M ) (24) is 2.5 Å 3 /Da for both crystal forms, which corresponds to a solvent content of about 52%. Both forms grow under similar conditions and have similar crystal morphologies. The actual crystal form can only be determined by examining the x-ray intensity data. Data collection statistics for all data sets are given in Table I.
The structure of orthorhombic SsMTAP was determined using molecular replacement with E. coli PNP (PDB entry code: 1ECP) (7) as the search model (sequence identity is 32%). Data between 10-and 4-Å resolution were included in the cross-rotation and translation functions, which were carried out using the CNS software package (25). The rotated and translated search model was carried through a round of rigid-body refinement. Electron density maps were calculated using the observed amplitudes measured from the native SsMTAP crystal and phases calculated from the refined search model. The interactive computer graphics program O (26) was then used to replace residues in the E. coli search model with corresponding residues in the SsMTAP sequence. Several rounds of torsion angle simulated annealing using CNS followed by manual refitting of the model in O were performed. Water molecules were included in the model during subsequent rounds of refinement. This refined model was then used as a search model in molecular replacement to determine the structure of the monoclinic form of SsMTAP, which contains an entire hexamer in the asymmetric unit. Refinements of all structures were performed with iterative cycles of torsion angle simulated annealing using CNS and manual refitting of the models in O.
The structures of the complexes were determined by refining the SsMTAP structure in the appropriate space group (P2 1 or C222 1 ). After an initial round of simulated annealing refinement, difference Fourier maps were used to reveal the active site contents (Fig. 1). The ligand atoms were included in the model followed by iterative cycles of model building and refinement. Individual isotropic B-factors and water molecules were included during the later stages of refinement. Crystallographic data and refinement statistics for all structures are reported in Table II.

RESULTS AND DISCUSSION
Overall Structure of SsMTAP-The crystal structure reveals that SsMTAP is a hexamer containing six identical subunits of ϳ27 kDa each, confirming the results from previous biochemical studies (1). The enzyme is disc shaped and is ϳ40 Å thick with a diameter of about 106 Å across the hexameric face. The molecule displays D 3 symmetry and can be described as a trimer of dimers with three symmetric intersubunit disulfide bonds linking the dimers to one another. Fig. 2 shows the hexameric arrangement of subunits and the location of the disulfide bonds. The observation of only three disulfide bonds contrasts with results from earlier work, which predicted the SCHEME 1. presence of six intersubunit disulfide bonds and suggested that the hexamer was arranged as a dimer of trimers (1). The quaternary structure of SsMTAP is very similar to that of the E. coli PNP hexamer (7,14).
Each SsMTAP monomer contains one active site, which is located near a dimer interface. The two active sites in the dimeric unit are related by a molecular 2-fold axis and are separated by about 20 Å. The dimers are formed mainly by contacts between residues from helix ␣2, helix ␣3, and three loops. These dimers form hexamers through contacts involving residues on strand ␤6, helix ␣3, helix ␣4, strand ␤7, and three loops. The hexamer forming contacts include disulfide bonds between pairs of 2-fold related Cys 125 residues. Cys 125 is located on the loop between helix ␣3 and helix ␣4 of each subunit near the center of the hexamer. Because the disulfide linkage spans a molecular 2-fold axis, only three disulfide bonds form per hexamer.
The final models of SsMTAP and its complexes consist of all residues except for 1 or 2 residues at the N terminus of each chain and 3 or 4 residues in the loop between the last strand, ␤9, and the C-terminal ␣ helix. These residues showed weak electron density and are believed to be flexible or disordered. The overall fold of SsMTAP (Fig. 3) is very similar to E. coli PNP (7,14), consisting of a single ␣/␤ domain. The central portion of the molecule is made up of a large eight-stranded mixed ␤ sheet with topology ␤2, ␤3, ␤4, ␤1, ␤5, ␤9, ␤6, and ␤7. Strand ␤8 is a 5-residue strand that forms hydrogen bonds with the ends of strands ␤7 and ␤5 resulting in a smaller 5-stranded ␤ sheet with topology ␤6Ј, ␤7Ј, ␤8, ␤5Ј, and ␤9Ј, where the prime designates the C-terminal end of that strand. The two  fused ␤ sheets form a distorted ␤ barrel near the active site. The core ␤ sheet structure is flanked by seven ␣ helices.
Although no sulfate or phosphate was added during crystallization of native SsMTAP, some residual electron density was observed in the predicted phosphate-binding site. Modeling sulfate or phosphate in this region resulted in a poor fit to the electron density and refinement resulted in extremely high B-factors (Ͼ70 Å 2 ) for either of these ions when it was included in the model. Based on these results neither sulfate nor phosphate is present in the final model. It is possible that the  electron density is due to phosphate that is present at very low occupancy or that the phosphate-binding site is occupied by several disordered water molecules.
Structures of the Binary and Ternary Complexes-Co-crystallization of SsMTAP with the nucleoside substrate guanosine in the presence of sulfate resulted in a ternary complex of SsMTAP-Guo-SO 4 . Sulfate was chosen as a phosphate substitute on the basis that the two anions share similar chemical characteristics and that sulfate is an inhibitor of the catalytic reaction. There is one hexamer in the asymmetric unit and each of the six active sites contains one molecule of guanosine bound in the nucleoside-binding site and one sulfate ion bound in the phosphate-binding site. The refined guanosine shows a glycosidic torsion angle (O4Ј-C1Ј-N9-C4) of 109°and a C4Ј-endo sugar pucker. This is an unusual conformation compared with nucleosides in solution, which generally have a glycosidic torsion angle near 0°(syn) or 180°(anti) and a C2Ј-endo or C3Ј-exo sugar pucker (27). However, the nucleoside conformation is similar to that found in the mammalian PNP-nucleoside complexes (11).
Co-crystallization of SsMTAP with the nucleoside substrate adenosine in the presence of sulfate resulted in a ternary complex of SsMTAP-Ado-SO 4 . There is one hexamer in the asymmetric unit and difference electron density maps revealed some variation in the adenosine occupancy among the six active sites. As a result, adenosine was modeled in only four of the six active sites. One sulfate ion was bound in each of the six phosphate-binding sites. Adenosine is observed with a glycosidic torsion angle of 116°and a C4Ј-endo sugar pucker.
Co-crystallization of SsMTAP with the nucleoside substrate MTA in the presence of sulfate resulted in a ternary complex of SsMTAP-MTA-SO 4 . There is one-half of a hexamer in the asymmetric unit. Each of the three active sites contains one molecule of MTA bound in the nucleoside-binding site and one sulfate ion bound in the phosphate-binding site. Both the initial F o Ϫ F c and the final 2F o Ϫ F c difference maps reveal weak density for the 5Ј-methylthio group while density for the purine base and the remainder of the ribose moiety is well defined. It is possible that the inability of the 5Ј-methylthio group to form a hydrogen bond with the side chain of His 5 may allow this portion of the molecule to become flexible. Consequently, the weak density for the methylthio group likely results from high thermal motion or disorder. MTA is modeled with a glycosidic torsion angle of 124°and a C4Ј-endo sugar pucker.
Co-crystallization of SsMTAP with either sulfate or phosphate resulted in the binary complexes SsMTAP-SO 4 and SsMTAP-PO 4 , respectively. For each complex there is one hexamer in the asymmetric unit and each of the six active sites contains one sulfate/phosphate ion bound in the phosphatebinding site. Closer inspection of the electron density maps for the SsMTAP-PO 4 complex revealed the presence of one well ordered molecule of Tris bound to each phosphate ion and extending into the ribose-binding site. Interestingly, although Tris buffer is used at a concentration of 0.1 M for all co-crystallization experiments, none of the sulfate containing structures revealed the presence of a bound Tris molecule in the active site.
Co-crystallization of SsMTAP with the noncleavable inosine analog FMB and phosphate revealed only phosphate bound in the active site. In this complex, there is one-half of a hexamer in the asymmetric unit and each of the three active sites contains one molecule of phosphate ion. Further inspection of the electron density map revealed that a molecule of Tris was bound instead of FMB. In contrast, co-crystallization with FMB in the presence of sulfate revealed both sulfate and FMB in the active site. The SsMTAP-FMB-SO 4 ternary complex has one-half of a hexamer in the asymmetric unit and each of the three active sites contain one molecule of FMB bound in the nucleoside-binding site and one molecule of sulfate ion bound in the phosphate-binding site.
Structure of the Active Site-The active site of SsMTAP was characterized using the binary and ternary complexes described above. The SsMTAP active site is located near the surface of the molecule in a groove formed at the dimer interface (Fig. 2). In contrast to the buried active site of the trimeric PNPs, the SsMTAP active site is relatively open. This exposed active site is similar to that seen in the hexameric E. coli PNP structure (7, 14) (Fig. 4a). The active site of SsMTAP consists of about 18 residues from eight separate polypeptide regions. These include Gly 21 , Arg 25 , Ile 64 , Arg 86 , Thr 89 , Thr 90, Gly 91 , Phe 160 , Glu 163 , Val 179 , Glu 180 , Met 181 , Glu 182 , Ser 204 , and Asp 205 from one subunit and Val 4 , His 5 , and Arg 43 from an adjacent subunit. The SsMTAP monomer can be divided into two structural domains by cutting at the end of strand ␤5. In general, the residues in the N-terminal structural domain are involved primarily in phosphate binding, while the residues in the C-terminal structural domain are involved primarily in nucleoside binding. Of the 18 active site residues 13 are identical in E. coli PNP (Fig. 4b). The active site of SsMTAP can be divided into three main regions: the phosphate-binding site, the purine-binding site, and the ribose base-binding site. This geometric arrangement of bound substrates is very similar to those seen in structures of both trimeric and hexameric PNPs (7,8,13,14,28). The binding sites are discussed in detail below.
The Phosphate-binding Site-The phosphate-binding site was characterized using ternary complexes of SsMTAP with nucleosides and sulfate and binary complexes with sulfate or phosphate alone. Residues in the phosphate binding site include Gly 21 , Arg 25 , Thr 89 , and Arg 86 from one monomer and Arg 43 from an adjacent monomer. In addition, the 2Ј-and 3Ј-hydroxyl groups of the bound nucleoside substrate contribute hydrogen bonds. The arrangement of interactions when sulfate is bound in the active site is the same for all of the SsMTAP ternary complexes as well as for the SsMTAP-SO 4 binary complex, and is illustrated in Fig. 5a. The phosphatebinding sites of E. coli PNP (7,14) and E. coli (15) uridine phosphorylase use the same functional groups and the same binding geometry as SsMTAP, except that Thr 89 is replaced by serine in these enzymes. In contrast, the phosphate-binding sites of trimeric enzymes in the PNP family are similar to each other (8, 13) but distinctly different from those of the hexameric enzymes.
The phosphate-binding site from the SsMTAP-PO 4 binary complex is very similar to that seen in the sulfate containing structures (Fig. 5b). One key difference involves Arg 25 . This residue seems to be slightly flexible in the sulfate complexes, showing some variation among the crystallographically independent monomers, and in some active sites did not contact the sulfate ion directly. In the phosphate complex, however, Arg 25 is well ordered and donates two hydrogen bonds to two different phosphate oxygen atoms. Also in complexes containing phosphate (but not in the ones containing sulfate) the carbonyl oxygen of Ile 64 is hydrogen bonded to a water molecule which in turn forms a hydrogen bond with the phosphate.
In complexes containing phosphate a well ordered molecule of Tris was also observed. The Tris-binding site is formed by amino acids that normally participate in ribose binding (Fig.  5b). In addition to donating a hydrogen bond to the phosphate from the amino group, the Tris hydroxyl groups also hydrogen bond to the carboxylate of Glu 182 and the side chain of His 5 from the neighboring subunit. An additional hydrogen bond is formed between a Tris hydroxyl group and the backbone amide group of Met 181 . Since these residues are involved in binding the ribose moiety of nucleosides, FMB is excluded from the phosphate containing structures by interactions between SsMTAP and a Tris molecule.
The Purine-binding Site-The purine-binding site was characterized using the ternary complexes guanosine, adenosine, and MTA. Fig. 6a illustrates the active site structure when guanosine is bound. The purine base binding site is composed of residues Phe 160 , Glu 163 , Val 179 , Met 181 , Ser 204 , and Asp 205 . A number of van der Waal contacts are also formed with strands ␤5 and ␤8 as they pass by the purine-binding site. Glu 163 accepts hydrogen bonds from both N-1 and the C-2 amino group of the purine base. In mammalian PNP, Glu 201 forms similar hydrogen bond with the purine base (11). In the SsMTAP-Guo-SO 4 complex the side chain conformation of Asp 205 is such that its carboxylate group is oriented away from the purine base. Ser 204 hydrogen bonds to a water molecule that in turn donates a hydrogen bond to N-7 of the nucleoside. Ser 204 and Asp 205 are highly conserved in hexameric PNPs. In E. coli PNP, the equivalent aspartic acid is believed to be protonated and to stabilize the transition state through a hydrogen bond to N-7 (7). A similar role is proposed for Asp 200 in human MTAP (13) and for Asn 243 in trimeric PNPs (11,29). The plane of the aromatic ring of Phe 160 is nearly perpendicular to the plane of the purine base generating a herringbone-type stacking interaction between the enzyme and the base. Met 181 contacts the purine base, Phe 160 and the hydrophobic side of the ribosyl group. A similar interaction, involving an aromatic residue and methionine, is found in all known nucleoside phosphorylase structures (7,8,(11)(12)(13)(14)28). Strand ␤5 and strand ␤8, which includes Val 179 , are located on the opposite side of the purine base. Fig. 6b shows the nucleoside-binding region of SsMTAP bound to adenosine. Most of the structural features of the active site are similar to the SsMTAP-Guo-SO 4 complex. One key difference involves Asp 205 . In SsMTAP-Ado-SO 4 , the carboxylate group of Asp 205 is near N-7 and the 6-amino group of the purine. This conformation also allows Ser 204 to donate a hydrogen bond to the carboxylate group of Asp 205 , which in turn is positioned to protonate N-7. As observed for the guanosine complex, Glu 163 is positioned near the N1-C2 edge of the purine base. In the case of adenosine, N-1 is unprotonated so Glu 163 must be protonated to form a hydrogen bond (see discussion below). The structure of the SsMTAP active site containing MTA is shown in Fig. 6c. The active site contacts are very similar to those observed in the adenosine complex with potential hydrogen bonds between the MTA and Asp 205 and between MTA and Glu 163 .
The Ribose-binding Site-Potential hydrogen bonds in the ribose-binding site are formed between Glu 182 and the 2Ј-and 3Ј-hydroxyl groups and between His 5 of a neighboring monomer and the 5Ј-hydroxyl group (Fig. 6). Additional hydrogen bonds are formed between the 2Ј-and 3Ј-hydroxyl groups and the sulfate oxygen atoms. Met 181 provides another key interaction, packing against the hydrophobic face of the ribose, Phe 160 and the purine base. His 5 , Met 181 , and Glu 182 are highly conserved among the hexameric members of the PNP family (7,15). A methionine residue equivalent to Met 181 has been identified in all known nucleoside phosphorylase structures (7,8,(11)(12)(13)(14)28). Interestingly, mutation of the equivalent methionine in human PNP had little affect on its activity (29), despite its high degree of conservation. The active site positions the phosphate for nucleophilic attack at the C1Ј carbon atom of the nucleoside. In the SsMTAP substrate complexes the distance between C1Ј and the closest sulfate oxygen atom is 3.5-4.0.
Since MTA lacks the 5Ј-hydroxyl group present in guanosine and adenosine, it no longer requires a hydrogen bond donor or acceptor in this vicinity of the active site. Instead, binding would be enhanced by hydrophobic interactions between the 5Ј-methylthio group and neighboring protein atoms. Even though the active sites of E. coli PNP and SsMTAP are similar (13 out of 18 residues are identical) only SsMTAP accepts MTA as a substrate. Residues that are located in the vicinity of the 5Ј-methylthio group include Ile 64 , Thr 89 , Phe 160 , Met 181 , and Glu 182 from one monomer and Val 4 , His 5 , and Arg 43 from an adjacent monomer. Of these, only Val 4 , Ile 64 , and Thr 89 are different than in E. coli PNP, in which case these residues are Pro 4 , Met 64 , and Ser 90 , respectively. Although the hydrophobicities of these residues are similar in the two enzymes, the structures of the side chains are different and this may contribute to the formation of a binding pocket for the 5Ј-methylthio group. One residue that might contribute to a hydrophobic environment near the 5Ј-position in SsMTAP is Thr 89 . In the E. coli PNP-FMB complex, a water molecule bridges the equiva- lent Ser 90 hydroxyl group and the 5Ј-hydroxyl group of FMB (14). In SsMTAP Thr 89 could serve a similar role for substrates containing a 5Ј-hydroxyl group and then rotate about the C␣-C␤ bond to provide a more hydrophobic environment for the 5Ј-methylthio group of MTA. It is interesting to note that human MTAP, which is specific for MTA, contains two active site threonines (13) that are replaced by serines in the structurally homologous human PNP (8), which requires a 5Ј-hydroxyl group. In addition, human MTAP utilizes a histidine to valine substitution to further increase the hydrophobicity of the 5Ј-methylthio group environment.
A more subtle difference between the active site structures of the SsMTAP-MTA-SO 4 complex and the SsMTAP-Guo-SO 4 or SsMTAP-Ado-SO 4 complexes involves the side chain conformation of Glu 182 in the ribose-binding site and the glycosidic torsion angle of MTA. For MTA, an increase in the glycosidic torsion angle of the MTA (124°) compared with adenosine (116°) and guanosine (109°) causes the sugar to rotate away from His 5 of the neighboring subunit. This movement results in the loss of a hydrogen bond between the carboxylate oxygen of the side chain and the 3Ј-hydroxyl group of MTA and a conformational change in the Glu 182 side chain (Fig. 6c). It is possible that the loss of the hydrogen bond between Glu 182 and the 3Ј-hydroxyl of MTA and the change in the glycosidic torsion angle, together with other less obvious structural changes, allows SsMTAP to adjust to the hydrophobic 5Ј-substituent.
Substrate-induced Conformational Change-In the nucleoside-free native SsMTAP structure and SsMTAP structures containing only sulfate or phosphate, a portion of the loop connecting strand ␤7 to helix ␣5 (residues 162-166) is in a relatively extended conformation. With the loop in this conformation, the side chain of Glu 163 is pointing away from the active site. Upon nucleoside binding a conformational change takes place and a portion of the ␤7-␣5 loop transforms into a 3 10 helix. As a consequence of this structural rearrangement, the carboxylate group of Glu 163 moves into a position where it can interact with the N1-C2 edge of the purine base. Fig. 7 illustrates the conformational change. This 3 10 helix occurs in the structures of all three SsMTAP-substrate complexes. In E. coli PNP, Glu 163 is replaced by proline (7,14). In addition to lacking the side chain carboxylate, proline provides rigidity and prevents a similar conformational change from occurring.
Unusual Binding Geometry in the Formycin B-Sulfate Complex- Fig. 8a depicts the structure of SsMTAP containing FMB. In SsMTAP, the noncleavable nucleoside analog binds in a dramatically different conformation than the nucleoside substrates. FMB is modeled with a glycosidic torsion angle of Ϫ63°, but because the ribose group remains more or less fixed, the purine base is rotated by ϳ180°compared with the guanosine, adenosine, and MTA complexes. As a consequence of this unusual binding geometry, the carboxylate group of Asp 205 is oriented down toward the nucleoside-binding site, but instead of being near the N-7 and N-6 positions of the purine, it accepts a hydrogen bond from N-1. The 6-oxo group is linked indirectly to Asp 205 by a water molecule. In the FMB complex, the ribose moiety adopts a C2Ј-exo conformation instead of the C4Ј-endo conformation seen in the SsMTAP-substrate complexes. In this binding geometry the 3Ј-hydroxyl group is no longer able to donate a hydrogen bond to Glu 182 . In addition, Glu 163 in the SsMTAP-FMB-sulfate complex is in the open (unliganded) conformation. Because of its unusual geometry, the N1-C2 edge of the FMB base is pointed away from the region of the active site where Glu 163 is located and the hydrogen bond to N-1 cannot form. The structure of E. coli PNP (PDB entry code: 1A69) (14) complexed with FMB and sulfate is also available for comparison (Fig. 8b). In this case, the structure reveals that FMB binds to E. coli PNP with a glycosidic torsion angle (101°) similar to the values seen in the SsMTAP-substrate complexes. The different modes for FMB binding displayed by the two enzymes is intriguing considering the overall similarity of the active site compositions and architectures. One difference is the pH of crystallization. However, SsMTAP, which displays the unusual binding geometry, was crystallized near physiological conditions at pH 7.4, whereas E. coli PNP was crystallized at pH 5.2 to 5.4 (14).
Implications for the Catalytic Mechanism of SsMTAP-Enzymes in the PNP family catalyze the phosphorolytic cleavage of the glycosidic bonds with varying substrate specificities and efficiencies. In addition to differences in quaternary structure, it has been shown that in general, the mammalian trimeric PNPs have a higher substrate specificity and efficiency than the hexameric enzymes (6,30,31). The crystal structures of these enzymes together with biochemical and mutational studies (29,32,33) have allowed detailed mapping of the active sites for several enzymes in the PNP family and specific roles in substrate binding and catalysis have been assigned. While all of these enzymes utilize inorganic phosphate to cleave the glycosidic bond of nucleosides, key structural differences exist which account for the variation in substrate specificities. Human and bovine PNP utilize Asn 243 and Glu 201 (8,11,28) to specifically bind 6-oxopurine nucleosides. These enzymes also prefer nucleosides with a 5Ј-hydroxyl group, and utilize His 257 to form a hydrogen bond to the ribose moiety at this position. Human MTAP on the other hand uses two aspartic acid residues (Asp 220 and Asp 222 ) and a structural water molecule in the purine-binding site to specifically bind 6-aminopurine nucleosides (13). This enzyme also creates a hydrophobic pocket for the 5Ј-methylthio group primarily by replacing His 257 of the mammalian PNPs with valine. The hydrophobic pocket also account for its higher affinity for 5Ј-deoxynucleosides and nucleosides with a 5Ј-alkyl or 5Ј-haloalkyl group (34,35). The structure of hexameric E. coli PNP reveals a more exposed active site, which is believed to contribute to the much broader substrate specificity displayed by this class of enzymes. The purine-binding site utilizes Asp 204 , a serine and water molecules in adapting to the hydrogen bonding requirements of both 6-oxo and 6-aminopurine nucleosides (7,14). While E. coli PNP will accommodate a variety of purine nucleosides, the enzyme does not utilize MTA as a substrate. Like E. coli PNP, SsMTAP accepts both 6-oxo and 6-aminopurine nucleoside substrates. However, only SsMTAP accepts MTA. Therefore, the active site of SsMTAP must accommodate two different hydrogen bonding schemes at N-1 and N-6, and both hydrophilic and hydrophobic substituents at the 5Ј-position while maintaining catalytic efficiency. Comparison of the active site of SsMTAP with those of human and bovine PNPs, human MTAP, and E. coli PNP suggests both similarities and differences.
During phosphorolytic cleavage of the glycosidic bond, electron density accumulates on the purine base. As the transition state forms, the negative charge is stabilized through hydrogen bond donors. Studies on human PNP suggested that most of the negative charge accumulates at N-7 and that the side chain of a highly conserved Asn 243 donates a hydrogen bond to offset the partial charge (32). Nucleosides that are alkylated at the 7-position are excellent substrates because of the electron withdrawing properties of the 7-substituent, even though the alkyl group interferes with the hydrogen bond to N-7. In mammalian PNPs Glu 201 accepts a hydrogen bond from N-1 but may also provide some transition state stabilization by positioning a water molecule as a hydrogen bond donor to O-6 (32). This interaction would be further enhanced if Glu 201 were protonated. Tebbe et al. (12) have argued a key role for glutamic acid in the mechanism of Cellulomonas sp. PNP. In E. coli PNP the hydrogen bond to N-7 is provided by Asp 204 (7) and in human MTAP by Asp 220 (13), requiring that the aspartic acids be mostly in the protonated state. The expected pK a of a buried aspartic acid or glutamic acid suggests that the carboxylate can exist in the protonated form to a significant extent (36). For example, the pK a of Glu 35 in lysozyme has been measured to be about 6.2 by 1 H NMR (37).
In SsMTAP, the residue nearest N-7 is Asp 205 . The high resolution structures of the adenosine and MTA complexes show that for 6-aminopurine nucleoside substrates, Asp 205 is in position to hydrogen bond with N-7, analogous to Asp 204 in E. coli PNP. In both SsMTAP and E. coli PNP, the aspartic acid is held in position by hydrogen bonds to a conserved serine and the 6-amino group of the purine. In contrast, Asp 205 in the guanosine complex is oriented away from the purine-binding site and is not in hydrogen bond contact with N-7. Instead, a well ordered water molecule is observed within hydrogen bond- FIG. 8. Active site drawing of the SsMTAP FMB-sulfate complex. a, the binding geometry for FMB. b, the binding geometry for the E. coli PNP-FMB complex for comparison. The coordinates were taken from PDB entry code 1A69 (14). Hydrogen bonds are shown as dashed lines with the corresponding donor-acceptor distance labeled. Residues belongs to the neighboring subunit are designated with an asterisk (*).
ing distance of N-7 and bridging to Ser 204 and a second water molecule is observed bridging O-6 and Asp 205 . These water molecules, which are also hydrogen bonded to each other could stabilize the transition state by donating hydrogen bonds to N-7 and O-6. One striking difference between SsMTAP and E. coli PNP is Glu 163 . In unliganded SsMTAP Glu 163 is pointed away from the active site, however, in all three SsMTAPsubstrate complexes, the carboxylate group is within hydrogen bonding distance of N-1. Since N-1 is protonated for 6-oxopurine nucleosides but not in 6-aminopurine nucleosides, Glu 163 must serve as acceptor for the former and as a donor for the latter. As a donor, Glu 163 must be protonated and the resulting hydrogen bond provides yet another mechanism for stabilizing the negatively charged transition state. Similar hydrogen bonds are formed with 6-oxopurines in the case of Glu 201 in the mammalian PNPs (11,28) and like the trimeric PNPs, a water molecule is located near the Glu 163 carboxylate in SsMTAP, however, this water molecule is too far from O-6 to form an additional hydrogen bond.
Although the protonation states of Glu 163 and Asp 205 cannot be determined from the crystallographic studies alone it is likely that SsMTAP utilizes different catalytic strategies for different substrates. In the case of 6-aminopurine nucleosides, transition stabilization might be provided by hydrogen bonds from Asp 205 , Glu 163 , or both. In the case of 6-oxopurine nucleosides, water molecules provide the hydrogen bonds, while Glu 163 only serves to orient the substrate. Finally, it is possible that the position of Asp 205 in the SsMTAP-guanosine complex is an artifact although the crystals were prepared at pH 7.4 and the position of Asp 205 was consistent for all six crystallographically independent monomers. Further kinetic of SsMTAP mutants should further clarify the roles of specific amino acids.
Thermostability of SsMTAP-In addition to the broad substrate specificity displayed by SsMTAP, the extreme thermophilicity of the enzyme (optimal enzymatic activity at 120°C) has generated much interest as well. We compared the threedimensional structure of SsMTAP with that of E. coli PNP (optimal enzymatic activity at 45°C) to identify structural features that might result in thermostability. The comparison is complicated by the low sequence identity (ϳ32%), making it difficult to determine which of the many residue changes contribute most significantly to the increased stability of SsMTAP. Previous work in the area of understanding the structural mechanisms of protein stability has identified some common features of thermophilic proteins (38,39). In general, thermostable proteins have an increased number of polar and ionic interactions compared with their mesophilic counterparts. Also, reduced internal cavities and more tightly packed hydrophobic cores are also observed in the thermophilic proteins. It is interesting that disulfide bonds do not appear to be a common feature of naturally occurring thermophilic proteins.
We examined the factors listed above to assess the structural basis for the extreme thermal stability of SsMTAP. Structural analysis using the program WHATIF (40) reveals that the E. coli PNP monomer contains 30 ion pairs (separated by a distance of less than 7 Å) while the SsMTAP monomer contains only 21. Even more striking is the fact that E. coli PNP contains twice as many strong ion pairs (separated by a distance of less than 4 Å) than SsMTAP. The number of optimal hydrogen bonds potentially formed in the two proteins was also determined using the program WHATIF. The results show that the number of hydrogen bonds that are present in E. coli PNP (137) and SsMTAP (132) are similar. The structures of the two enzymes possess nearly identical folds and the number of hydrophobic residues is similar in both enzymes. Therefore, the thermal stability of SsMTAP is probably not related to the compactness of the core structure. Finally, thermostability may be gained from the shortening of loops connecting secondary structural elements. However, E. coli PNP contains 238 residues per subunit while SsMTAP is approximately the same size with 236 residues per subunit. Consequently, the only obvious structural feature of SsMTAP that might account for its increased thermostability is the presence of the three intersubunit disulfide bonds.
Although the presence of disulfide bonds has not been implicated as a general feature of naturally occurring extremely thermophilic proteins, it appears that these might be an important structural mechanism for the thermostability of SsMTAP since SsMTAP lacks any of the common stabilizing features. This is consistent with previous studies of thermostabilty of SsMTAP in the presence of reducing agents (1) in which a significant loss of activity was observed at 100°C in the presence 0.1 M dithiothreitol. The structure of SsMTAP shows that the disulfide bridges are strategically placed to stabilize the hexamer. In SsMTAP there are two types of 2-fold axes (Fig. 2). The first type of 2-fold axis relates pairs of subunits that when joined generate complete active sites. The contacts between these subunits are extensive. The second type of 2-fold axis joins the closely packed dimeric units. This interface is far away from the active site and has few intersubunit contacts. The three disulfide bonds in SsMTAP link subunits that are related by the second type of 2-fold symmetry axis but not the first and consequently provide overall stability to the SsMTAP hexamer.
Recently there have been reports of single or multiple disulfide bonds in intracellular proteins from hyperthermophilic microrganisms (41)(42)(43)(44)(45). In all cases, these disulfide bonds are not present in the homologous mesophilic proteins. The occurrence of three intersubunit disulfide bonds in SsMTAP suggests the possibility that the formation of these covalent links could be a more general strategy for stabilizing hyperthermophilic enzymes.