Methylthioadenosine Phosphorylase at 1.7 Å Resolution Provides Insights Into Substrate Binding and Catalysis Todd C Appleby1, Mark D Erion2 and Steven E Ealick3*

View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by Elsevier - Publisher Connector

Research Article 629

The structure of human 5′-deoxy-5′-methylthioadenosine phosphorylase at 1.7 Å resolution provides insights into substrate binding and catalysis Todd C Appleby1, Mark D Erion2 and Steven E Ealick3*

Background: 5′-Deoxy-5′-methylthioadenosine phosphorylase (MTAP) Addresses: 1Section of Biochemistry, Molecular and catalyzes the reversible phosphorolysis of 5′-deoxy-5′-methylthioadenosine Cell Biology, Cornell University, Ithaca, NY 14853, USA and 2Metabasis Therapeutics, Inc., 9390 Town (MTA) to adenine and 5-methylthio-D-ribose-1-phosphate. MTA is a by-product Center Drive, San Diego, California 92121, USA of polyamine biosynthesis, which is essential for cell growth and proliferation. and 3Department of Chemistry and Chemical This salvage reaction is the principle source of free adenine in human cells. Biology, Cornell University, Ithaca, NY 14853, USA. Because of its importance in coupling the purine salvage pathway to polyamine *Corresponding author. biosynthesis MTAP is a potential chemotherapeutic target. Email: [email protected]

Results: We have determined the crystal structure of MTAP at 1.7 Å resolution Key words: methylthioadenosine phosphorylase, using multiwavelength anomalous diffraction phasing techniques. MTAP is a multiwavelength anomalous diffraction, purine trimer comprised of three identical subunits. Each subunit consists of a single nucleoside phosphorylase, salvage pathway α β β / domain containing a central eight-stranded mixed sheet, a smaller five- Received: 18 January 1999 stranded mixed β sheet and six α helices. The native structure revealed the Revisions requested: 17 February 1999 presence of an adenine molecule in the purine-binding site. The structure of Revisions received: 4 March 1999 MTAP with methylthioadenosine and sulfate ion soaked into the active site was Accepted: 12 March 1999 also determined using diffraction data to 1.7 Å resolution. Published: 26 May 1999

Conclusions: The overall quaternary structure and subunit topology of MTAP are Structure June 1999, 7:629–641 similar to mammalian purine nucleoside phosphorylase (PNP). The structures of http://biomednet.com/elecref/0969212600700629 the MTAP–ligand complexes provide a map of the active site and suggest © Elsevier Science Ltd ISSN 0969-2126 possible roles for specific residues in substrate binding and catalysis. Residues accounting for the differences in substrate specificity between MTAP and PNP are also identified. Detailed information about the structure and chemical nature of the MTAP active site will aid in the rational design of inhibitors of this potential chemotherapeutic target. The MTAP structure represents the first structure of a mammalian PNP that is specific for 6-aminopurines.

Introduction Trypanosoma brucei brucei [8], bovine liver [9] and human Human 5′-deoxy-5′-methylthioadenosine phosphorylase liver [10]. Biochemical evidence suggests that mam- (MTAP; EC 2.4.2.28) functions in the purine salvage malian MTAP is a trimer made up of three identical sub- pathway where it catalyzes the reversible phosphorolysis units of 32 kDa [7,9], although in at least one case the of 5′-deoxy-5′-methylthioadenosine (MTA), a 6-amino- enzyme was reported as a dimer [10]. Each subunit in the purine nucleoside formed during polyamine biosynthesis human enzyme contains 283 amino acid residues. Protein [1]. The products of the reaction are free adenine and sequence analysis reveals that human MTAP shares 5-methylthio-D-ribose-1-phosphate (MTR-1-P). Adenine 25–28% identity with members of the mammalian purine is subsequently recycled back into nucleotides through nucleoside phosphorylase (PNP) family of trimeric the action of phosphoribosyltransferases [2], while MTR- enzymes. PNP also functions in the purine salvage 1-P is utilized in methionine biosynthesis [3]. MTAP pathway where it catalyzes the phosphate-dependent activity is responsible for essentially all of the free cleavage of nucleosides [11]. In mammals, PNP is adenine generated in human cells [4], indicating that this responsible for the phosphorolysis of the 6-oxopurine enzyme performs an important function in the purine nucleosides, guanosine and inosine, whereas MTAP is salvage pathway. highly specific for 6-aminopurine nucleoside substrates [8,10]. In addition, human MTAP readily accepts MTAP activity was first characterized in rat ventral 5′-deoxyadenosine and 6-aminopurine nucleosides con- prostate in 1969 [5]. The enzyme has since been purified taining a halogen, haloalkyl or alkylthio group at the from a number of organisms and tissues including 5′ position of the sugar as substrates, but displays little or the thermophilic bacterium Caldariella acidophila [6], no activity with adenosine or adenosine analogs contain- mouse Ascites sarcoma 180 cells [2], human placenta [7], ing a 5′-hydroxyl group [8,10,12]. 630 Structure 1999, Vol 7 No 6

The crystal structures of PNP from human erythrocyte glycol, 200 mM Tris-HCl pH 7.4 and 2 mM DTT and [13] and bovine spleen [14] have been previously obtain a maximum size of 0.4–0.5 mm in length and reported, and extensive studies aimed at identifying 0.2 mm in diameter across the hexagonal face. The crys- residues involved in substrate binding and catalysis have tals belong to space group P321 with unit-cell constants of been performed on the human enzyme [15–17]. Kinetic a=b=120.8 Å and c = 44.5 Å. The Matthews number 3 studies of PNP mutants and structural studies of PNP (VM) [19] is 3.0 Å /Da assuming one monomer of molecu- and PNP–ligand complexes identified active-site lar weight 31.2 kDa in the asymmetric unit, which corre- residues that interact with the nucleoside and phosphate. sponds to a solvent content of ~59%. The 25% ethylene These studies suggested that Asn243 and Glu201 are glycol in the mother liquor also acts as a cryoprotectant, responsible for the high specificity of PNP for 6-oxo- allowing the MTAP crystals to freeze straight out of the purines while His257 is believed to hydrogen bond to the drop. These crystals diffract beyond 1.7 Å resolution when ribose moiety of the nucleoside in the sugar-binding using synchrotron radiation. region. The phosphate-binding pocket of PNP contains an arginine residue in addition to several polar Although SeMet–MTAP was fully active, crystals failed sidechains. The PNP studies also led to a proposed to grow under the native crystallization conditions. Explo- enzyme mechanism: phosphorolysis is thought to ration of an expanded search grid around the native con- proceed through an SN1-type mechanism in which the ditions revealed that a substantial drop in reservoir pH (to glycosidic bond is cleaved first, forming an oxocarbenium pH 5.9) and increased ionic strength were required for ion intermediate that is stabilized and subsequently growth of SeMet–MTAP crystals. Diffraction quality attacked by one of the nucleophilic phosphate oxygens crystals of SeMet–MTAP are long, thin hexagonal rods [16]. Structural studies of human MTAP together with that obtain a size of 0.5–0.6 mm in length and 0.1 mm in structural comparisons to the functionally related mam- diameter across the hexagonal face. These crystals also malian PNP will help to establish a molecular basis for belong to space group P321 with unit-cell constants of the observed differences in substrate specificity between a=b=121.0 Å and c = 44.3 Å. the two enzymes and may offer insight into a possible catalytic mechanism for MTAP. Structure determination With one monomer containing 286 residues in the asym- In this report, we describe the 1.7 Å resolution crystal metric unit and nine selenomethionine residues per structures of human MTAP bound to adenine alone and of monomer, the favorable ratio of roughly 32 residues per MTAP in complex with MTA and sulfate. Se atom suggested that the structure determination of MTAP using multiwavelength anomalous diffraction Results and discussion (MAD) phasing techniques was feasible [20]. A single, Protein production and crystallization frozen SeMet–MTAP crystal was used to collect a three- Expression of the polyhistidine-tag–human-MTAP cDNA wavelength MAD data set on beamline F2 at the Cornell construct in Escherichia coli BL21(DE3) cells [18] resulted in High Energy Synchrotron Source (CHESS). Although the desired fusion protein (6His–MTAP) at levels approach- technical considerations limited the data to 2.8 Å resolu- ing 20% of the total soluble protein fraction from the host tion, direct methods succeeded in finding eight out of the cells (data not shown). After thrombin cleavage of the 6His- nine selenium atom positions [21]. The missing Se atom tag, three residues from the tag remain fused to the natural was later determined to belong to the N-terminal N-terminal end of MTAP (Gly-Ser-His-MTAP). The selenomethionine residue that is believed to be disor- resulting protein, however, is fully active indicating that the dered along with ten other N-terminal residues that reside additional residues do not disturb subunit–subunit interac- in a solvent channel of the crystal. The resulting electron- tions or catalytic activity. Therefore, the proteolized recom- density map was of sufficient quality to trace most of the binant product is considered equivalent to native human polypeptide chain and model a majority of the backbone MTAP. The final yield from the protein purification proce- and sidechain atoms. dure is 24 mg of purified material per liter of host cell culture. Production of the selenomethionyl derivative of A second MAD data set collected to significantly higher MTAP (SeMet–MTAP) gave similar results to the native resolution on beamline X12C at the National Synchrotron enzyme, although the yield was about twofold lower. The Light Source (NSLS) was used together with the previ- reduced yield is primarily accounted for by approximately ously determined eight Se sites to calculate higher resolu- twofold lower expression of the 6His–MTAP construct in tion experimental phases. This 2.2 Å resolution map the methionine auxotroph. when solvent-flattened was of extremely high quality, and allowed some previously missing residues to be The optimized crystallization conditions for native MTAP modeled. Figure 1 shows a representative region of the yield hexagonal rod-shaped crystals that grow in 12% (w/v) electron-density map calculated from the 2.2 Å resolution polyethylene glycol (PEG) 6000, 25% (v/v) ethylene experimental MAD phases. Research Article Human 5′-deoxy-5′-methylthioadenosine phosphorylase Appleby, Erion and Ealick 631

Figure 1

Example of the experimental electron density. Stereoview of the solvent-flattened F186 F186 W189 W189 experimental map calculated using MAD phases to 2.2 Å resolution showing a representative section of the model. The map is contoured at the 1σ level and superimposed on the refined models. (The figure was produced using the program CHAIN [40].)

T188 T188

R187 R187

Structure

Overall structure electron density in this region eight native N-terminal The final model of MTAP consists of residues 9–224 and residues and three residual polyhistidine-tag residues are 230–281. The overall fold of MTAP is very similar to also missing from the final model. mammalian PNP [13,14]. Figure 2 illustrates the structure of the MTAP monomer, which consists of a single α/β The crystal structure reveals that human MTAP is a trimer domain. The central portion of the molecule is made up of containing three identical subunits related by C3 symmetry. a large eight-stranded mixed β sheet with topology β2, β3, The enzyme is positioned in the cell such that the molecular β4, β1, β5, β11, β6, β8. In addition, strands β5 and β11 threefold axis of the trimer is along a crystallographic three- extend into a smaller five-stranded mixed β sheet with fold axis. The trimer measures ~35 Å along the axis and topology β7, β9, β10, β5, β11. The β sheets are flanked by ~75 Å across. There are few contacts between trimers in the six α helices (Figure 2b). Strands β5 and β11, helix α5 and crystal, and consequently the majority of the trimer surface four separate loop structures all contribute residues to the is exposed to solvent channels which make up 59% of the active site of the enzyme. cell volume. The subunit contacts within a trimer are more extensive and primarily involve residues from strand β7 and Major topological differences between PNP and MTAP the loop connecting β7 to helix α2 of one subunit, and helix occur at the N and C termini of the polypeptide chain. The α3 and the loop connecting strand β9 to α3 of a neighboring first ordered residues at the N terminus of MTAP belong subunit. At the trimer interface, direct contacts between to strand β1. As the residues Met1–Thr8 in MTAP appear identical residues related by the threefold axis include a to be disordered in the crystal structure, it is unlikely that hydrophobic interaction of three Trp189 residues and a these N-terminal residues participate in forming an addi- hydrogen-bond network involving three Thr118 residues. tional α helix or β strand. In PNP, however, there is an 11-residue N-terminal α helix which is connected by a loop The active site to the first β strand. MTAP also contains an additional The active site of each monomer is located near the inter- 12-residue C-terminal α helix (α6) which is not present in face between subunits. Figure 3 illustrates the MTAP PNP. The location of this helix positions Leu279 of one trimer and the location of the three active sites. In MTAP, subunit near the hydrophobic 5′-methylthio-binding the active site of a given subunit is composed entirely of pocket of a neighboring subunit in the trimer. residues from that subunit with the exception of His137 and Leu279, which belong to a neighboring subunit. This The only major break in the electron density occurs in the trimeric arrangement of subunits is very similar to that loop between β11 and α5 and involves residues Lys225, seen in mammalian PNP [13,14]. Glu226, His227, Glu228 and Glu229. This flexible loop is believed to extend out away from the trimer into a large Although no substrates or inhibitors of MTAP were solvent channel of the crystal, and the missing density in added during the purification or crystallization of the this region probably results from disorder. The N terminus enzyme, the MAD-phased electron-density maps also extends into a solvent channel, and due to the poor revealed strong planar density in the purine-binding site. 632 Structure 1999, Vol 7 No 6

Figure 2

(a) (b) 258 α5 233

β11 220

264 β5 97 α 3 α6 α5 275 195 β10 193

C 180 α3 188 β10 β9 β7 β 6 172 β9 168 α4 β 8 α1 β11 116 β7 112

α2 β β5 158 146 2 C 45 β 32 59 16 3 β4 β1 198 α2 74 106 162 α6 β2 β3 β4 β1 α1 β5 β11 α4 β6 β8 N 109 164 50 83 207 27 53 11 88 211

N (c)

180 180 100 100 220 220

240 240

160 160 20 20 120 120 60 200 60 200 140 140 40 40 80 80 260 260 N N C Structure C

Structure of the MTAP monomer. (a) Ribbon representation of the (b) Topology diagram of the β-sheet structure in the MTAP monomer: secondary structure elements in the human MTAP monomer: β strands are represented by arrows and α helices by cylinders. β strands, labeled β1–β11, are represented by flat green arrows and Secondary structure elements are labeled accordingly, as in (a). The α helices, labeled α1–α6, are colored magenta. Regions of random first and last residues in the β strands and α helices are also identified. coil are shown in blue and the two turns of 310 helix are colored cyan. (The figure was produced using Microsoft PowerPoint [Microsoft The short yellow loop between β11 and α5 represents residues Corporation, Redmond, WA].) (c) Stereoview Cα trace of an MTAP 225–229, which are missing in the final model. The refined positions of subunit. Every twentieth residue is labeled. The broken line represents MTA and sulfate are represented by ball-and-stick models with atoms the disordered loop containing residues 225–229. MTA and sulfate color-coded: red, oxygen atoms; blue, nitrogen atoms; gray, carbon are modeled in red. (The figure was produced using the program atoms; and yellow, sulfur atoms. The N and C termini are labeled. MOLSCRIPT [41].)

This density persisted throughout refinement against the presence of the purine and allows us to identify key native diffraction data, and an adenine molecule was residues that account for base specificity. eventually included in this region of the model. Several hydrogen bonds are possible between the adenine base Diffraction data were also collected on a native MTAP and sidechain atoms, which offers further support for the crystal soaked in MTA and ammonium sulfate. MTA is Research Article Human 5′-deoxy-5′-methylthioadenosine phosphorylase Appleby, Erion and Ealick 633

Figure 3

Stereoview of the MTAP trimer. The trimer is viewed down the molecular/crystallographic threefold axis. Each subunit is shown in a different color, with MTA and sulfate modeled in red in each of the three active sites. Broken lines indicate residues 225–229, which are missing in the final model. (The figure was produced using the program MOLSCRIPT [41].)

Structure

the natural nucleoside substrate for the enzyme and torsion angle of 145° [16], which suggests that MTAP and should bind in the active site. Sulfate was chosen as a PNP bind nucleosides in a similar manner. Figure 4b phosphate substitute on the basis that the two molecules shows a schematic representation of the ligands bound in share similar chemical characteristics and that sulfate is an the active site. inhibitor of the catalytic reaction. The putative phosphate-binding site will be referred to as the sulfate/phos- The base-binding site phate-binding site hereafter, in consideration of the fact The electron density reveals that Phe177, Ser178, Asp220, that important structural information about this region of Asp222 and a water molecule facilitate the binding of the the active site is obtained based on the structure of the purine base. Extensive hydrogen bonding between the sulfate-containing crystal. enzyme and ligand in this region of the active site provides specificity for 6-aminopurine nucleosides. If Asp220 Although difference density is strong for the sulfate ion in MTAP is protonated (see later section), then the dis- and nucleoside base, the density for the nucleoside sugar tance between Asp220 OD1 and N7 of adenine (2.97 Å) moiety is not as clearly defined. This could be due to suggests a hydrogen bond. It appears that Asp220 can also increased flexibility of the ribose unit, but is more likely form hydrogen bonds with N6 of adenine and the back- to be a result of low occupancy indicating that MTA may bone nitrogen of Asp222. A hydrogen bond between not have fully displaced the adenine during the crystal- Asp220 OD2 and Asp222 N may contribute to the ability soaking experiment. Although the difference density for of Asp222 to adopt an unfavorable φ/ψ (53°/–121°) confor- the methylthioribose group was not as strong as the mation. Asp222 can form a weak hydrogen bond with N6 density for the sulfate ion and base, a reliable model for of adenine while also participating in a hydrogen-bond the methylthioriobosyl group could be easily built. network with a water molecule, Ser178 OG and Ser178 N. This set of interactions orients the water molecule cor- The active site of MTAP contains three distinct regions rectly so that it can donate a hydrogen bond to the unpro- corresponding to the base-, methylthioribose- and tonated N1 of adenine. The plane of the phenyl group of sulfate/phosphate-binding sites. Figure 4a shows the ori- Phe177 is nearly perpendicular to the plane of the purine entation of the ligands with respect to important active- ring system, generating a herringbone-type base-stacking site residues. The geometric arrangement of the interaction between the enzyme and the base. sulfate/phosphate- and nucleoside-binding sites places the sulfate ion on the α face of the ribose of MTA. Active-site The methylthioribose-binding site residues position the sulfate ion such that O4 is 4.06 Å Met196, Val233, Val236 and Leu237 in addition to His237 away from C1′ of the nucleoside. MTA is modeled with a and Leu279 from a neighboring subunit are present in the glycosidic torsion angle (O4′–C1′–N9–C4) of 126° and a methylthioribose-binding pocket of MTAP. The enzyme C4′-endo sugar pucker for the methylthioribose moiety. appears to form only one hydrogen bond with the sugar This is an unusual conformation compared to purine moiety of MTA. The amide nitrogen of Met196 is a rea- nucleosides in solution, which generally have a glycosidic sonable hydrogen-bonding distance (3.06 Å) from the torsion angle near 0° (syn) or 180° (anti) [22]. Crystallo- 2′-hydroxyl of the ribose ring. Val233, Val236 and Leu237 graphic analysis of PNP in complex with various nucleo- together with Leu279 from a neighboring subunit form a sides and nucleoside analogs reveals an average glycosidic hydrophobic pocket near the 5′-methylthio group of 634 Structure 1999, Vol 7 No 6

Figure 4

The MTAP active site. (a) Stereoview of the (a) active site showing key residues. MTA and sulfate are colored magenta and are labeled accordingly. Active-site residues from one monomer are in yellow, and residues from the neighboring monomer are colored cyan. The blue sphere represents the active-site water molecule. (The figure was made using the program RIBBONS [42].) (b) Schematic representation of the MTAP active site. Residues of MTAP are shown boxed and labeled appropriately. Shaded residues identify structurally equivalent residues in PNP. An asterisk is used to denote an active-site residue contributed from a neighboring monomer. Dotted lines indicate proposed protein–protein (b) or protein–ligand hydrogen bonds. (The figure N243 V245 was produced using CHEMDRAW N D220 OD2 D222 [CambridgeSoft, Cambridge, MA].) OD1 OG1

T242 N WAT5 T219 OG1 N N V260 MTA OG L261 L237 N S178 E201 N V236 N O F177 F200 L279* C S O N M196 M219 H257 V233 O OG1 T93 F159* H137* O OG1 T197 S220 O S OG1 O O R60 R84 S33 T18 N NH1

N NE2 A94 H61 A116 H86 Structure

MTA. The face of the imidazole ring of His137 from a sidechain may undergo a conformational change upon neighboring subunit also packs against the 5′ region of the sulfate/phosphate binding. ribose moiety. Comparisons to human PNP The sulfate/phosphate-binding site The high degree of structural homology between MTAP The sulfate/phosphate-binding pocket contains Thr18, and PNP is not readily apparent from primary sequence Arg60, His61, Thr93, Ala94 and Thr197. These residues data. Sequence alignments alone reveal that MTAP shares contribute a complement of sidechain and backbone approximately 28% identity with bovine PNP and about atoms that facilitate the binding of a negatively charged 27% identity with human PNP. However, the similar ion. The positively charged guanidinium group of Arg60 functions suggest, and the highly conserved topology provides a favorable charge–charge interaction with the demonstrates, that they are homologous proteins. A struc- sulfate ion. In addition, Thr18 OG1, Thr93 OG1, Thr197 tural alignment of MTAP to human and bovine PNP OG1 and His61 NE2 form hydrogen bonds with the (Figure 5) reveals a similar geometric arrangement of the oxygen atoms of sulfate. The amide nitrogens of Thr18 nucleoside- and phosphate-binding sites, as well as a and Ala94 also donate hydrogen bonds to one of the similar overall active-site architecture. An alignment of the sulfate oxygens. The conformation of Thr18 is different in protein sequences based on regions of structural homology the two MTAP structures, suggesting that the Thr18 (Figure 5) highlights several key residue changes that may Research Article Human 5′-deoxy-5′-methylthioadenosine phosphorylase Appleby, Erion and Ealick 635

Figure 5

Structural alignment and structure-based sequence comparison of MTAP and PNP. (a) (a) Stereoview of the structural alignment of human MTAP (red), human PNP (cyan), and bovine PNP (yellow). Only the Cα trace for a single subunit of each enzyme is shown. MTA and sulfate are modeled in black stick form to indicate the active-site location. (b) The structure-based sequence alignment of the three enzymes. Residues with similar properties are shaded in gray and identical residues are denoted by black arrowheads. Secondary structure elements are represented by arrows (β strands) and cylinders (α helices) for both MTAP (above the alignment in black) and human PNP (below the alignment in gray). Active-site residues are highlighted in magenta (phosphate-binding site), green (sugar-binding site) and cyan (base-binding site). Numbers indicate every tenth residue based on the MTAP sequence. (The figure was made using the program ALSCRIPT [43].) (b) β1 β2

10 20 30 40 HuMTAP MASGTTTTAVKI GI I GGT GLDD PEI LEGRTEKYVDTPFGKPS HuPNP MENGYTYEDYKNTAEWLLSHTKHRPQVAI I CGS GLGGLTDKLTQAQI FDYSEI PNFPRST BvPNP MANGYTYEDYQDTAKWLLSHTEQRPQVAVI CGS GLGGLVNKLTQAQTFDYSEI PNFPEST

α1 β1 α2 β2

β3 β4 α1 β5

50 60 70 80 90 100 DALI LGKI KNVDCVLLA RHGRQHTI MPSKVNYQANI WALKEEGCTHVI VTTACGSL REEI VPGHAGRL VFGFL NGRACVMMQGR F H M Y EGYPLWKVTFPVRVFHLLGVDTLVVTNA AGGL NPKF VPGHAGRL VFGI L NGRACVMMQGR F H M Y EGYPFWKVTFPVRVFRLLGVETLVVTNA AGGL NPNF

β3 β4 α3 β5

β6 β7 α2 β8

110 120 130 140 150 160 QPGDI VI I DQF I DRT T MRPQSF YDGSHSCARGVCH I PMAEPFCPKTREVLI ETAKKLG LRCHSKG EVGDI MLI RDHI NL PGFSGQNPL RGPNDERFGDRF PAMSDAYDRTMRQRALSTWKQMGEQRELQEG EVGDI MLI RDHI NL PGFSGENPL RGPNEERFGVRF PAMSDAYDRDMRQKAHSTWKQMGEQRELQEG

β6 β7 α4

β9 α3 β10 α4 β11

170 180 190 200 210 220 230 TMVTI EGPRFSSRAESFMFRTWGADVI NMT TVPEVVLAKEAGI CYASI AMATD Y D CWK EHEEAVS TYVMVAGPSFETVAECRVLQKLGADAVGMS TVPEVI VARHCGLRVFGFSLI TN KVI MDYESLEKAN TYVMLGGPNFETVAECRL L RNL GADAVGMS TVPEVI VARHCGLRVFGFSLI TN KVI MDTESQGKAN

β8 α5 β9 α6 β10

α5 α6

240 250 260 270 280 V DRVLKTLKE NANKAKSLLLTTI PQI GSTEWSETLHNLKNMAQFSVL LPRH H EE VLAAGKQAAQKLEQFVSI LMASI PLPDKAS H EEVLEAGKQ AAQKLEQFVSLLMASI PVSGHTG

α7 Structure

account for the differences in substrate specificity change is reasonable considering the fact that PNP is spe- between MTAP and PNP. cific for 6-oxopurines and therefore requires a good hydrogen-bond donor for both N7 and O6 of the base, a task Base specificity well suited to the amino group of Asn243. Interestingly, it In MTAP, Asp220 appears to have an important role in has been shown that mutating Asn243 to an alanine in nucleobase specificity by acting as both a hydrogen-bond human PNP results in a 470-fold decrease in catalytic effi- donor to N7 and a hydrogen-bond acceptor for N6 of ciency, which is primarily due to a 1000-fold decrease in adenine. In PNP, Asp220 is replaced by Asn243. This kcat [15]. This observation suggests that Asn243 is also an 636 Structure 1999, Vol 7 No 6

important catalytic residue. The ability of a nucleoside Asn115 occupies this region in PNP, but the structure phosphorylase to achieve such a significant change in base reveals that the sidechain points away from the phos- specificity by replacing a catalytically important residue phate-binding pocket. The arginine residue in the without losing functionality and still maintaining the same sulfate/phosphate-binding region of MTAP (Arg60) is overall active-site architecture has been demonstrated (see structurally conserved in PNP (Arg84), suggesting that later section) [17]. mammalian PNP and MTAP might utilize this residue in a similar fashion for substrate binding and catalysis. Asp222 in MTAP can form a weak hydrogen bond with Although His61 in MTAP and His86 in PNP appear to N6 of adenine and can participate along with Ser178 in the have similar roles in sulfate/phosphate binding, there is a positioning of a water molecule to act as a hydrogen-bond difference in the backbone structure of the two enzymes donor to N1, which is not protonated in 6-aminopurines. in this region of the active site. The structural alignment PNP utilizes Glu201 in place of Ser178 and the active site places the Cα atoms for the two histidine residues nearly water molecule of MTAP, thus providing a good hydro- 7 Å apart. The sidechain conformations for the respective gen-bond acceptor for N1, which is protonated in 6-oxo- histidines, however, result in a much closer alignment of purines. MTAP and PNP both utilize an active-site the imidazole rings, which provide the interaction with the phenylalanine residue (Phe177 and Phe200, respectively), sulfate/phosphate ion. which can provide a herringbone stacking interaction with the purine ring of either 6-aminopurine or 6-oxopurine Implications for catalysis nucleoside substrates. Given the similar function and overall conserved active- site architecture of MTAP and PNP, it is likely that these Sugar specificity enzymes share a common catalytic mechanism. The PNP

A comparison of residues in the ribose-binding site reveals reaction is believed to proceed through an SN1-type mech- another significant difference between MTAP and PNP anism in which the glycosidic bond is cleaved first forming that may account for the altered substrate specificity. The an oxocarbenium intermediate [16]. The oxocarbenium affinity of MTAP for 5′-deoxynucleosides and nucleosides ion is then stabilized and subsequently attacked by a nega- with a halogen, haloalkyl or alkylthio group at the 5′ positively charged phosphate oxygen. Crystallographic studies tion appears to result from the general hydrophobicity and of human PNP bound to 6-oxopurine nucleosides reveal complete lack of hydrogen-bond donors or acceptors in this that the carboxamido group of Asn243 is in a good position region of the active site. In PNP, His257 occupies the same to stabilize the transition-state structure by donating a position as Val233 in MTAP and appears to form a hydro- hydrogen bond to N7 of the purine. The biochemical evi- gen bond with the 5′-hydroxyl group of guanosine and dence obtained from the Asn243→Ala mutant discussed inosine. The smaller alkyl sidechain of Val233 in MTAP above also supports this important role for Asn243 in catal- may provide a more suitable environment for nucleosides ysis. A large decrease in catalytic efficiency for a with larger hydrophobic groups at the 5′ position. Glu201→Ala mutant of PNP revealed that this residue also has a role in catalysis, although the primary factor was

His137 of MTAP and Phe159 of PNP are both con- a large increase in Km for the nucleoside substrate [16]. tributed by a neighboring subunit and occupy nearly iden- The PNP structure shows that in the phosphate-binding tical positions in the ribose-binding site of their respective site, His86 hydrogen bonds to phosphate and to the car- enzymes. The face of the ring system in both His137 and boxyl group of Glu89. The orientation of Glu89, His86 and Phe159 packs against the 5′ region of the nucleoside sub- phosphate is similar to the catalytic triad seen in the serine strate, and probably forms van der Waals contacts with the proteases [23]. This arrangement of functional groups may ribose moiety. In addition to contributing to the active site act to increase the nucleophilicity of the phosphate. of a neighboring subunit, His137 in MTAP also hydrogen bonds to a water molecule, which in turn is hydrogen Structural analysis suggests that MTAP may utilize a bonded to His65 of the neighboring subunit. This hydro- similar reaction mechanism. MTAP, like PNP, binds gen-bonding scheme may represent an important β-nucleosides in a high-energy C4′-endo conformation, subunit–subunit interaction in the MTAP trimer. which is believed to put strain on the substrate and may help weaken the glycosidic bond. The sulfate/phosphate- Sulfate/phosphate binding binding site of MTAP positions the sulfate ion on the Comparisons of the residues involved in sulfate/phosphate α face of the MTA ribose moiety. The closest sulfate binding in MTAP and PNP reveal the highest conserva- oxygen in the MTAP structure, O4, is 4.06 Å away from tion of functional groups compared to the other active-site C1′ of the nucleoside, and is in a good position to initially regions. The two serine residues involved in sulfate/phos- stabilize the partial positive charge on O4′ of the proposed phate binding in PNP, Ser33 and Ser220, are replaced by oxocarbenium ion intermediate. It is conceivable that a Thr18 and Thr197 in MTAP. MTAP also contains conformational change in MTAP during the reaction an additional sulfate/phosphate-binding residue, Thr93. brings the phosphate oxygen into closer proximity to the Research Article Human 5′-deoxy-5′-methylthioadenosine phosphorylase Appleby, Erion and Ealick 637

electrophilic C1′ allowing for the subsequent nucleophilic nucleosides, but retains the ability to efficiently catalyze attack. It has been shown that a buried aspartic acid cleavage of the glycosidic bond. residue can exhibit a significant increase in pKa allowing it to be protonated under normal physiological conditions Replacing the catalytically important Glu201 in PNP with (~pH 7) [24]. If Asp220 in MTAP is protonated, this the Asp222–Ser178–water network of interactions in residue is in an excellent position to protonate N7 of the MTAP accommodates the unprotonated N1 of the purine in order to accommodate the flow of negative 6-aminopurine while conserving catalytic efficiency. It is charge to the base that occurs during glycosidic bond believed that Glu201 in PNP offers specificity for 6-oxo- cleavage (Figure 6). The proposed hydrogen bond purines and helps to hold the nucleoside in a catalytically between Asp220 OD2 and the backbone nitrogen of competent orientation, which is reflected by the large

Asp222 may have an important role in anchoring the change in Km and smaller change in kcat of the Asp220 sidechain so that Asp220 OD1 is in a good position Glu201→Ala PNP mutant [15,16]. Therefore, the changes to protonate N7 of the purine during catalysis. Thus, it in this region of the active site of MTAP to accommodate appears that in addition to contributing to the specificity the 6-aminopurine are not expected to cause significant for 6-aminopurines, Asp220 and Asp222 in MTAP may alterations in the catalytic mechanism. The functionally also help to conserve the catalytic mechanism proposed conserved His86 in PNP and His61 in MTAP appear to for mammalian PNP. have similar roles in phosphate binding. Unlike the Glu89-His86-phosphate triad that is seen in the PNP The ability of a protonated aspartic acid residue to assist in structure, His61 in MTAP hydrogen bonds to a backbone the enzyme-catalyzed phosphorolysis of nucleosides was carbonyl oxygen rather than an acidic group. Furthermore, predicted following a mutational study on human PNP in the Glu89→Ala mutation in PNP did not have a large which Asn243 was mutated to an aspartic acid residue effect on catalytic activity. These results suggest that a [17]. The Asn243→Asp mutant displayed a lower affinity catalytic triad is not required for efficient glycosidic bond for inosine and guanosine in addition to a decrease in kcat cleavage in MTAP or PNP. relative to wild-type PNP. In contrast, adenosine was a good substrate for the mutant and revealed a 5000-fold Biological implications ′ ′ increase in kcat when compared to the rate of phosphoro- Human 5 -deoxy-5 -methylthioadenosine phosphorylase lysis of adenosine by wild-type PNP. The catalytic effi- (MTAP) functions in the purine salvage pathway where ciency of the mutant utilizing adenosine as a substrate was it catalyzes the reversible phosphorolysis of 5′-deoxy-5′- only 80-fold lower than the catalytic efficiency of wild- methythioadenosine (MTA). We report here the crystal type PNP with its natural substrate, inosine. This single structure of human MTAP. Structural comparisons mutation resulted in an enzyme that exhibits a change in reveal that the enzyme is a member of the mammalian substrate specificity from 6-oxopurine to 6-aminopurine purine nucleoside phosphorylase (PNP) family of trimeric enzymes, which contain a unique α/β fold. Struc- tural studies of MTAP bound to adenine and in complex Figure 6 with MTA and sulfate have identified important active- site residues. A detailed view of the MTAP active site confirms the functional similarities of MTAP and PNP, Asp220 while also providing insight into the structural changes required for the differences in substrate specificity

O O between the two enzymes. Analysis of the MTAP structure also suggests that the previously proposed catalytic NH2 H mechanism for PNP can be extended to MTAP as well. δ – N N The absence of MTAP activity in several malignant H3C N human cell lines has been reported [25]. The loss of S δ + N O enzyme function in these cells is now believed to be a result of homozygous deletions on chromosome 9 of the – O O closely linked MTAP and p16/MTS1 tumor suppressor HO OH P genes [26]. As the absence of the tumor suppressor is – HO O probably responsible for the onset of malignancy, the Structure lack of MTAP activity appears to be a consequence of cancer development and not a primary cause. Neverthe- Proposed transition-state structure for the enzyme-catalyzed less, the absence of MTAP activity does have an impor- phosphorolysis of MTA. (The figure was produced using CHEMDRAW [CambridgeSoft, Cambridge, MA].) tant impact on purine metabolism in these cells. It has been shown that MTAP-deficient (MTAP–) carcinoma 638 Structure 1999, Vol 7 No 6

cells display an increased rate of de novo purine synthe- free poly-histidine-tag. The purified MTAP was buffer-exchanged and sis, when compared to the same cells transfected with subsequently concentrated to 23 mg/ml in 10 mM Tris-HCl pH 7.4 and 2 mM DTT. The purified material was stored at –80°C. The protein con- and expressing MTAP cDNA [27]. These results centration was determined by the Bradford method [29] using protein suggest that in the absence of MTAP activity, cells assay dye reagent (BioRad) with bovine serum albumin as a standard. cannot recycle adenine back into nucleotides and must Purity was verified by running samples on a 12% polyacrylamide gel fol- therefore increase their rate of de novo AMP synthesis. lowed by Coomassie staining (data not shown). Spectrophotometric MTAP assays following the method of Toorchen and Miller [10] were An important consequence of this finding is that the performed to verify enzyme activity. MTAP– cells should be more susceptible to the growth inhibitory effects of chemotherapeutic drugs that specifi- Expression and purification of SeMet–MTAP cally target the purine biosynthetic pathway. In fact, Production of SeMet–MTAP follows essentially the same protocol as MTAP– cancer cell lines do display increased sensitivity above with a few alterations. The expression host used for production of SeMet–MTAP is the E. coli B834(DE3) methionine auxotroph towards methotrexate and azaserine in the presence of [Novagen Inc.]. The 1 L induction cultures contain M9 media supple- MTA, whereas cancer cell lines with MTAP activity mented with 40 µg/ml L-amino acids (not including methionine), 1X + (MTAP cells) were not as severely affected [27,28]. In BME vitamin solution (GibcoBRL), 0.4% (w/v) glucose, 2 mM MgSO4, 25 µg/ml FeSO .7H O, 30 µg/ml kanamycin, 0.1 mM CaCl and view of these observations, it may be possible to enhance 4 2 2 µ L + 40 g/ml -selenomethionine. These induction cultures were inoculated the treatment of MTAP tumors by the co-administra- with cells from a 1 ml saturated starter-culture containing the above tion of strong MTAP inhibitors together with traditional media with L-methionine in place of L-selenomethionine in order to chemotherapeutic compounds that specifically target the promote growth. The starter cells were then pelleted and washed in the de novo purine biosynthetic machinery. The structure of induction culture media immediately before inoculation in order to remove as much residual L-methionine as possible. SeMet–MTAP is MTAP provides a detailed view of the active site and also purified and stored in the presence of 5 mM DTT rather than 2 mM offers insight into the roles that specific residues play in DTT in order to protect against oxidation of the selenium atoms. substrate binding and catalysis. This information will assist in the identification and design of strong inhibitors Crystallization of native MTAP of MTAP should the enzyme prove to be a useful Initial crystallization conditions for native MTAP were determined using the sparse matrix screens, crystal screen and crystal screen 2 chemotherapeutic target. (Hampton Research). Several conditions yielded crystals of varying quality. The most promising crystals grew from condition 37 of the Materials and methods crystal screen 2 kit (10% PEG 8000, 0.1 M Hepes pH 7.5, 8% ethyl- Unless otherwise noted, all chemicals and reagents were purchased ene glycol). Subsequent optimization of the initial conditions resulted in from the Sigma Chemical Company, St Louis, MO. diffraction-quality crystals that grow at room temperature using the hanging-drop vapor diffusion technique. Drops (4 µl) contained a 1:1 Expression and purification of native MTAP mixture of protein and reservoir solutions. The protein solution is MTAP cDNA isolated from a human placenta cDNA gene library was described above. The reservoir solution contained 12% (w/v) PEG amplified using the polymerase chain reaction (PCR) and cloned into 6000, 25% (v/v) ethylene glycol, 200 mM Tris-HCl pH 7.4 and 2 mM the expression vector pKK223 (K Takabayashi and MDE, unpublished DTT. Drops were equilibrated against 1.0 ml of reservoir solution. results). The MTAP cDNA was then subcloned into the pET-28a expression vector [Novagen Inc.] using traditional molecular cloning tech- Crystallization of SeMet–MTAP niques. This new expression construct, which codes for an N-terminal Diffraction-quality crystals of SeMet–MTAP were grown at room tem- thrombin-cleavable polyhistidine-tag, was subsequently transformed into perature using the hanging-drop vapor diffusion method described E. coli BL21(DE3) cells [Novagen Inc.]. Induction cultures containing above. The protein solution contained 25 mg/ml SeMet–MTAP, 10 mM 1 L of Luria broth (LB) media and 30 µg/ml of kanamycin were inocu- Tris-HCl pH 7.4, 5 mM DTT and 400 mM NaCl. The reservoir solution lated with 1 ml saturated starter-cultures and incubated at 37°C. When contained 12% (w/v) PEG 6000, 25% (v/v) ethylene glycol, 80 mM the induction cultures reached an OD600 of ~0.6, the cells were MES pH 5.9 and 5 mM DTT. induced with 1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) for 4 h at 37°C. Cells were harvested by centrifugation and resuspended in MTA and sulfate crystal soak 40 ml ice cold binding buffer (5 mM imidazole, 500 mM NaCl, 20 mM Two week old native MTAP crystals were transferred from their original Tris-HCl pH 7.9) [Novagen Inc.]. All subsequent protein purification hanging drops containing 12% (w/v) PEG 6000, 25% ethylene glycol, steps were carried out at 4°C or on ice. Cells were lysed using a French 200 mM Tris-HCl pH 7.4 and 2 mM DTT, prior to equilibration, into an press followed by sonication. Following a high-speed centrifugation step artificial mother liquor containing the above solution supplemented with to remove insoluble material, the clarified cell extract was filtered and 10 mM NH4SO4 and 4 mM MTA. The soaks were carried out overnight applied to a 2.5 ml HisBind immobilized nickel column [Novagen Inc.], (16 h) before crystals were looped and frozen for data collection. which was previously equilibrated with binding buffer. The column was washed with ten volumes of binding buffer followed by six volumes of MAD data (2.8 Å resolution) wash buffer (60 mM imidazole, 500 mM NaCl, 20 mM Tris-HCl pH 7.9) A three-wavelength MAD data set was collected on a single frozen [Novagen Inc.]. 6His–MTAP was eluted with six volumes of 300 mM imi- (–180°C) SeMet–MTAP crystal on the F2 beamline at CHESS dazole, 500 mM NaCl and 20 mM Tris-HCl pH 7.9. The protein solution (Cornell University, Ithaca, NY). An X-ray absorption spectrum (in the was then buffer-exchanged into 20 mM Tris-HCl pH 8.4, 150 mM NaCl, vicinity of the Se K-absorption edge) for SeMet–MTAP was deter- 2.5 mM CaCl2 and 2 mM dithiothreitol (DTT) using an ultra-filtration mined by recording fluorescence measurements on a solid-state device. To cleave the polyhistidine-tag, the protein solution was treated detector during an energy scan of the crystal. Diffraction data were with 0.5 U of biotinylated thrombin [Novagen Inc.] per mg of 6His–MTAP then collected to 2.8 Å resolution at wavelengths corresponding to at 4°C for ~16 h. The reaction mixture was then passed over a strepta- the inflection point of the spectrum (0.97938 Å), the peak of the vidin agarose column [Novagen Inc.] to remove the biotinylated pro- spectrum (0.97915 Å), and a reference wavelength above the peak tease. The protein solution was buffer-exchanged back into binding (0.96768 Å). The data were acquired on a 1K charged-couple device buffer and passed over a second nickel column in order to remove the (CCD) detector [30] using 1° oscillations with 30 s exposures. The Research Article Human 5′-deoxy-5′-methylthioadenosine phosphorylase Appleby, Erion and Ealick 639

crystal-to-detector distance was 70 mm and the detector area was MAD data (2.2 Å resolution) 51 mm by 51 mm. In order to minimize systematic errors, Bijvoet pairs A higher resolution three-wavelength MAD data set was later col- at the different wavelengths were acquired close in time by collecting lected on a single frozen (–180°C) SeMet–MTAP crystal on the X12C the data in 5° wedges simulating inverse beam geometry (5° starting beamline at the NSLS (Brookhaven National Laboratory, Upton, NY). at φ followed by 5° starting at φ + 180°) and alternating wavelengths As before, an absorption spectrum was determined and data were throughout the experiment until a total of 80° (40° starting at φ =0° collected at wavelengths corresponding to the inflection point and 40° starting at φ = 180°) were collected for each wavelength. (0.97913 Å), the peak (0.97874 Å), and a reference wavelength above the peak (0.95000 Å). The data were acquired to 2.2 Å resolu- The data were processed with the program DENZO and tion on the single-module Brandeis CCD detector using 1° oscillations SCALEPACK [31]. Unmerged data output from SCALEPACK (with and 60 s exposures. The crystal-to-detector distance was 101.8 mm the ‘NO MERGE ORIGINAL INDEX’ flag set) for all three wavelengths and the detector area was 100 mm by 100 mm. This time, however, were used as input for the MADSYS series of programs [20]. the data set at each wavelength was collected in its entirety simulating ANOSCL was used to perform a local scaling routine intended to min- inverse beam geometry (30° starting at φ = 0° followed by 30° starting imize the systematic error between Bijvoet pairs, thus allowing for at φ = 180°), before moving on to the next wavelength. accurate determination of the anomalous signal. A second local scaling routine, WVLSCL, was then run to minimize the systematic The NSLS MAD data were also processed using DENZO and errors between data from different wavelengths in order to enhance SCALEPACK. This data together with the previously determined eight determination of the dispersive signal. MADLSQ was subsequently Se-atom positions was directly input into MLPHARE [36] from the used to calculate structure-factor amplitudes for the normal scattering CCP4 suite of programs to refine heavy-atom parameters and calculate from all atoms in the structure (|FT|), structure-factor amplitudes for the experimental phases to 2.2 Å resolution. DM was run using the higher normal scattering from the selenium atoms alone (|FA|), and the phase resolution data and phases in reflection omit mode with the ‘NCYCLE ∆φ φ φ differences between the structure factors FT and FA ( = T– A) for AUTO’ flag set. The program terminated after four cycles of solvent-flat- the unmerged reflection data. Finally, MERGIT was run to merge the tening and histogram matching. The figure of merit was 0.59 prior to results for multiple observations of the same reflection. The SHELXS density modification and was 0.87 after density modification. Data col- software package [21] was used to calculate E values based on the lection statistics are summarized in Table 1. FA values and direct methods were used to find eight out of the nine selenium atom positions. The eight Se positions were used together Model building and refinement with data from all three wavelengths to refine heavy-atom parameters The interactive computer graphics program O [37] was used to build a and calculate experimental phases using the MADSYS programs Cα trace for an MTAP monomer into the original 2.8 Å resolution MAD ASLSQ and MADABCD, respectively. Density modification was per- map. Chain tracing proceeded smoothly as electron density was con- formed using the initial phases and structure-factor amplitudes from tinuous for the majority of the molecule. Mainchain and sidechain atoms the reference wavelength in the program DM [32] of the CCP4 suite were subsequently built into the remaining density. Electron density for [33]. DM was run in reflection omit mode with the ‘NCYCLE AUTO’ residues 129–134 and 261–265 was discontinuous and difficult to flag set. The program terminated after two cycles of solvent-flattening interpret. The initial 2.8 Å resolution model did not include residues and histogram matching. The figure of merit was 0.65 prior to density 1–9, 225–229 and the C-terminal residues 272–283, as density for modification and was 0.84 after density modification. these regions of the protein was very weak or missing.

Data were also collected on the F1 beamline at CHESS using a native At this point it was possible to refit the 2.8 Å resolution model MTAP crystal and a native MTAP crystal soaked in sulfate and MTA. Both according to the new 2.2 Å resolution experimental map. Well data sets were collected to 1.7 Å resolution on a Quantum-4 CCD detec- resolved bumps in the higher resolution map, corresponding to the tor (ADSC) using 1° oscillations until greater than 90% overall complete- carbonyl oxygens, allowed for adjustment of mainchain atom posi- ness was observed. The Quantum-4 data sets were both indexed with tions. Several sidechain atoms were also adjusted. Significant MOSFLM [34] and scaled using SCALA [35] from the CCP4 suite of improvements to the electron density in the regions of residues programs. Data collection statistics are summarized in Table 1. 129–134 and 261–265 were also evident and the positions of these

Table 1

Summary of diffraction data.

Data type/source Se MAD (CHESS F2) Se MAD (NSLS X12C) Mono (CHESS F1)

Inflection Peak Reference Inflection Peak Reference Native Complex

Wavelength (Å) 0.97938 0.97915 0.96768 0.97913 0.97874 0.95000 0.91900 0.91900 Resolution (Å) 2.80 2.80 2.80 2.20 2.20 2.20 1.70 1.70 No. observations 41,898 41,818 42,258 69,120 67,457 69,285 276,082 393,633 No. unique 9087 9059 9083 18,938 18,924 19,026 39,479 42,292 Completeness (%) 97.7 (91.8) 97.7 (91.4) 98.2 (93.7) 99.0 (97.4) 99.0 (97.5) 99.0 (96.8) 93.1 (83.7) 96.8 (93.2)

Rsym*(%) 4.9 (7.0) 5.6 (7.8) 5.3 (8.0) 5.9 (7.8) 7.3 (9.3) 6.4 (8.1) 4.5 (7.8) 5.3 (10.6) Average I/σI 25.8 (17.2) 22.4 (15.5) 23.6 (15.8) 18.9 (9.8) 16.7 (9.1) 17.7 (9.2) 12.4 (8.1) 9.4 (6.5) f′ values –8.727 –8.513 –3.596 –7.814 –6.708 –2.722 – – f′′ values 3.242 4.273 3.859 3.244 4.544 3.260 – –

Values for the outer resolution shell are given in parentheses: Cornell CHESS native data, 1.79–1.70 Å; CHESS MTAP–MTA–sulfate complex ΣΣ Σ High Energy Synchrotron Source (CHESS) MAD data, 2.90–2.80 Å; data, 1.79–1.70 Å. *Rsym = i |Ii–|/ , where is the mean National Synchrotron Light Source (NSLS) MAD data, 2.28–2.20 Å; intensity of the N reflections with intensities Ii and common indices h,k,l. 640 Structure 1999, Vol 7 No 6

Table 2 nucleoside, however, was weak and difficult to interpret. A sulfate molecule was modeled into the phosphate-binding site density, and the Refinement statistics and model quality. sidechain of Thr18 was adjusted so that the oxygen atom fit into the new density. Subsequent rounds of SA followed by model adjustment began Native MTA–sulfate with data from 8.0–2.3 Å resolution and was eventually extended to 1.7 Å complex resolution. Water molecules were added beginning at 2.0 Å resolution. The nucleoside was left out of the model until later stages of refinement. Resolution range (Å) 20.0–1.7 8.0–1.7 Refinement statistics for this model are given in Table 2. Total number of nonhydrogen atoms 2251 2242 Number of protein atoms 2068 2068 Structure-based sequence alignment Number of water atoms 173 149 A structural alignment of MTAP to human and bovine PNP was per- Number of ligand atoms 10 25 formed by submitting the three-dimensional coordinates of each model Number of reflections in refinement 34,211 41,325 to the DALI server (http://www2.ebi.ac.uk/dali/). The results of the Number of reflections in test set 3806 4174 structural alignments were verified by graphically viewing the superim- R factor (%) 18.1 20.2 posed molecules in the program O. Further improvements to the struc- Rfree (%) 20.2 22.6 ture alignments were made using the LSQ program in O. The MTAP Rms deviation from ideal geometry protein sequence was then manually aligned to the bovine and human bonds (Å) 0.007 0.008 PNP protein sequences based on the observed structural homology angles (o) 1.258 1.249 between the MTAP and PNP structures. Ramachandran plot most favored region (%) 88.9 88.5 Accession numbers additional allowed region (%) 10.3 10.7 The atomic coordinates for the MTAP–adenine and MTAP–MTA–sulfate generously allowed region (%) 0.9 0.9 models have been deposited in the Protein Data Bank with accession disallowed region (%) 0.0 0.0 numbers 1cb0 and 1cg6, respectively. Average B factors (Å2) mainchain 14.9 19.2 sidechain 18.6 22.3 Acknowledgements water 24.0 26.7 The authors would like to acknowledge Ken Takabayashi for providing the adenine 9.8 15.3 initial MTAP clone and Robert Sweet for providing access to beamline methylthioribose – 33.6 X12-C at the NSLS, which is supported by the United States Department of Energy Offices of Health and Environmental Research and of Basic Energy SO4 – 31.8 Sciences and by the National Science Foundation. This work was supported by the Biomedical Research Resource Program of the National Insti- tutes of Health, through MacCHESS grant RR-01646, the WM Keck Foundation and the Lucille P Markey Charitable Trust. residues were modified accordingly. Previously missing mainchain atoms and most of the sidechain atoms for residues in the C-terminal helix and C-terminal tail up to residue 279 were also easily modeled References using the new electron-density map. Strong planar density was also 1. Pegg, A.E. & Williams-Ashman, H.G. (1969). On the role of found in the region that was believed to be the purine-binding site. S-adenosyl-L-methionine in the biosynthesis of spermidine by rat Initially, this density was not modeled to see if it would persist prostate. J. Biol. Chem. 244, 682-693. during refinement. 2. Savarese, T.M., Crabtree, G.W. & Parks, R.E., Jr. (1981). 5′-Methylthioadenosine phosphorylase-1. Substrate activity of The improved initial model was refined against the 1.7 Å resolution data 5′-deoxyadenosine with the enzyme from sarcoma 180 cells. from the native MTAP crystal using X-PLOR [38]. Several rounds of Biochem. Pharmacol. 30, 189-199. simulated annealing (SA) [39] followed by manual re-fitting of the 3. Backlund, P.S., Jr. & Smith, R.A. (1981). Methionine synthesis from 5′-methylthioadenosine in rat liver. J. Biol. Chem. 256, 1533-1535. model in the program O were performed. Refinement began using bulk- 4. Kamatani, N. & Carson, D.A. (1981). Dependence of adenine solvent-corrected data to 2.3 Å resolution. After the data were production upon polyamine synthesis in cultured human lymphoblasts. extended to 2.0 Å resolution, water molecules were included in the Biochim. Biophys. Acta 675, 344-350. model during subsequent rounds of refinement based on 3σ peaks in 5. Pegg, A.E. & Williams-Ashman, H.G. (1969). Phosphate-stimulated ′ Fo–Fc maps that contained a reasonable hydrogen bond to at least one breakdown of 5 -methylthioadenosine by rat ventral prostate. other protein, ligand or solvent atom. The data were eventually Biochem. J. 115, 241-247. extended to 1.7 Å resolution. At this point, the density in the proposed 6. Carteni-Farina, M., et al., & Zappia, V. (1979). 5′-Methylthioadenosine purine-binding site was still very strong and an adenine molecule was phosphorylase from Caldariella acidophila. Purification and properties. modeled into this region. Density for residues 280 and 281 was Eur. J. Biochem. 101, 317-324. 7. Ragione, F.D., Carteni-Farina, M., Gragnaniello, V., Schettino, M.I. & revealed during refinement, but density for the N-terminal residues 1–8 Zappia, V. (1986). Purification and characterization of 5′-deoxy-5′- (plus Gly-Ser-His from the 6His-tag), surface loop residues 225–229 methylthioadenosine phosphorylase from human placenta. J. Biol. and C-terminal residues 282 and 283 never appeared. The final model Chem. 261, 12324-12329. includes 173 water molecules, one adenine molecule and 2068 protein 8. Ghoda, L.Y., et al., & Bacchi, C.J. (1988). Substrate specificities of atoms from 268 of the 283 natural residues. Refinement statistics for 5′-deoxy-5′-methylthioadenosine phosphorylase from Trypanosoma this model are given in Table 2. brucei and mammalian cells. Mol. Biochem. Parasitol. 27, 109-118. 9. Ragione, F.D., Oliva, A., Gragnaniello, V., Russo, G.L., Palumbo, R. & The refined native MTAP model was then used as an initial model for Zappia, V. (1990). Physiochemical and immunological studies on ′ ′ X-PLOR refinement of the MTA–sulfate crystal soak data. Data from mammalian 5 -deoxy-5 -methylthioadenosine phosphorylase. J. Biol. Chem. 265, 6241-6246. 8.0–3.0 Å resolution were used in rigid-body refinement for an overall 10. Toorchen, D. & Miller, R.L. (1991). Purification and characterization of adjustment of the model to fit the new data. An Fo–Fc difference electron- 5′-deoxy-5′-methylthioadenosine (MTA) phosphorylase from human density map was then calculated. Strong difference density was clearly liver. Biochem. Pharmacol. 41, 2023-2030. visible for a sulfate ion. Density near Thr18 suggested that the sidechain 11. Parks, R.E., Jr. & Agarwal, R.P. (1972). Purine nucleoside was now in a different conformation. The presence of the nucleoside phosphorylase. In The Enzymes. 3rd Edition. (Boyer, P.D., ed.), base was also evident. Density for the methylthioribose portion of the 7, pp. 483-514, Academic Press, Inc., New York. Research Article Human 5′-deoxy-5′-methylthioadenosine phosphorylase Appleby, Erion and Ealick 641

12. Sufrin, J.R., et al., & Porter, C.W. (1991). Targeting 5′-deoxy-5′- 37. Jones, T.A., Zou, J.Y., Cowan, S.W. & Kjeldgaard, M. (1991). Improved (methylthio)adenosine phosphorylase by 5′-haloalkyl analogues of methods for building protein models in electron density maps and the 5′-deoxy-5′-(methylthio)adenosine. J. Med. Chem. 34, 2600-2606. location of errors in these models. Acta Crystallogr. A 47, 110-119. 13. Ealick, S.E., et al., & Bugg, C.E. (1990). Three-dimensional structure 38. Brünger, A.T. (1988). X-PLOR Manual, Version 3.1. Yale University, of human erythrocyte purine nucleoside phosphorylase at 3.2 Å New Haven, CT. resolution. J. Biol. Chem. 265, 1812-1820. 39. Brünger, A.T., Krukowski, A. & Erickson, J. (1990). Slow-cooling 14. Koellner, G., Luic, M., Shugar, D., Saenger, W. & Bzowska, A. (1997). protocols for crystallographic refinement by simulated annealing. Acta Crystal structure of calf spleen purine nucleoside phosphorylase in a Crystallogr. A 46, 585-593. complex with hypoxanthine at 2.15 Å resolution. J. Mol. Biol. 40. Sack, J. (1988). CHAIN — a crystallographic modeling program. 265, 202-216. J. Mol. Graph. 6, 244-245. 15. Erion, M.D., et al., & Ealick S.E. (1997). Purine nucleoside 41. Kraulis, P.J. (1991). MOLSCRIPT: a program to produce both detailed phosphorylase. 1. Structure-function studies. Biochemistry. and schematic plots of protein structures. J. Appl. Crystallogr. 36, 11725-11734. 24, 946-950. 16. Erion, M.D., Stoeckler, J.D., Guida, W.C., Walter, R.L. & Ealick S.E. 42. Carson, M. (1987). Ribbon models of macromolecules. J. Mol. Graph. (1997). Purine nucleoside phosphorylase. 2. Catalytic mechanism. 5, 103-106. Biochemistry 36, 11735-11748. 43. Barton, G.J. (1993). ALSCRIPT: a tool to format multiple sequence 17. Stoeckler, J.D., et al., & Erion, M.D. (1997). Purine nucleoside alignments. Protein Eng. 6, 37-40. phosphorylase. 3. Reversal of purine base specificty by site-directed mutagenesis. Biochemistry 36, 11749-11756. 18. Studier, F.W., Rosenberg, A.H., Dunn, J.J. & Dubendorff, J.W. (1990). Use of T7 RNA polymerase to direct expression of cloned genes. Methods Enzymol. 185, 60-89. 19. Matthews, B.W. (1968). Solvent content of protein crystals. J. Mol. Biol. 33, 491-497. 20. Hendrickson, W.A. (1991). Determination of macromolecular structures from anomalous diffraction from synchrotron radiation. Science 254, 51-58. 21. Sheldrick, G.M. (1990). Phase annealing in SHELX-90; direct methods for larger structures. Acta Crystallogr. A 46, 467-473. 22. Saenger, W. (1984). Principles of Nucleic Acid Structure. (Cantor, C.R., ed.), pp. 51-104, Springer-Verlag New York, Inc., New York. 23. Blow, D.M., Birktoft, J.J. & Hartley, B.S. (1969). Role of a buried acid group in the mechanism of action of chymotrypsin. Nature 221, 337-340. 24. Langsetmo, K., Fuchs, J.A. & Woodward, C. (1991). The conserved, buried aspartic acid in oxidized Escherichia coli thioredoxin has a pKa of 7.5. Its titration produces a related shift in global stability. Biochemistry 30, 7603-7609. 25. Ragione, F.D., et al., & Zappia, V. (1995). 5′-Deoxy-5′- methylthioadenosine phosphorylase and p16INK4 deficiency in multiple tumor cell lines. Oncogene 10, 827-833. 26. Nobori, T., et al., & Carson, D.A. (1996). Genomic cloning of methylthioadenosine phosphorylase: a purine metabolic enzyme deficient in multiple different cancers. Proc. Natl Acad. Sci. USA 93, 6203-6208. 27. Chen, Z., Olopade, O.I. & Savarese, T.M. (1997). Expression of methylthioadenosine phosphorylase cDNA in p16–, MTAP– malignant cells: restoration of methylthioadenosine phosphorylase-dependent salvage pathways and alterations of sensitivity to inhibitors of purine de novo synthesis. Mol. Pharmacol. 52, 903-911. 28. Kamatani, N., Nelson-Rees, W.A. & Carson, D.A. (1981). Selective killing of human malignant cell lines deficient in methylthioadenosine phosphorylase, a purine metabolic enzyme. Proc. Natl Acad. Sci. USA 78, 1219-1223. 29. Bradford, M. (1976). Quantitation of microgram quantities of protein utilizing the principle of protein-dye binding. Anal. Biochem. 72, 248-254. 30. Tate, M.W., Eikenberry, E.F., Barna, S.L., Wall, M.E., Lowrance, J.L. & Gruner, S.M. (1995). A large-format high-resolution area X-ray detector based on a fiber-optically bonded charge-coupled device (CCD). J. Appl. Crystallogr. 28, 196-205. 31. Otwinowski, Z. & Minor,W. (1997). Processing of X-ray diffraction data collected in oscillation mode. Methods Enzymol. 276, 307-326. 32. Cowtan, K. (1994). Joint CCP4 and ESF-EACBM Newsletter on Protein Crystallography. 31, 34-38. 33. Collaborative Computational Project, Number 4. (1994). The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D 50, 760-763. 34. Leslie, A.G.W., Brick, P. & Wonacott, A.T. (1986). An improved package for the measurement of oscillation photographs. In Daresbury Laboratory Information Quarterly for Protein Crystallography. 18, pp. 33-39, SERC Daresbury Laboratory, Warrington, UK. 35. Evans, P.R. (1993). Data reduction. In Proceedings of the CCP4 Study Weekend, Data Collection & Processing. 114-122. 36. Otwinowski, Z. (1991). Maximum-likelihood refinement of heavy atom parameters. In Isomorphous Replacement and Anomalous Scattering. (Wolf, W., Evans, P.R. & Leslie, A.G.W., eds), pp. 80-88, SERC Proceedings, Daresbury Laboratory, Warrington, UK.