- Email: [email protected]
Crystal Structure of Human Thymidine Phosphorylase in Complex with a Small Molecule Inhibitor
Structure, Vol. 12, 75–84, January, 2004, 2004 Elsevier Science Ltd. All rights reserved.
DOI 10.1016/j.str.2003.11.018
Crystal Structure of Human Thymidine Phosphorylase in Complex with a Small Molecule Inhibitor Richard A. Norman, Simon T. Barry, Michael Bate, Jason Breed, Jeremy G. Colls, Richard J. Ernill, Richard W.A. Luke, Claire A. Minshull, Mark S.B. McAlister, Eileen J. McCall, Helen H.J. McMiken, Dougie S. Paterson, David Timms, Julie A. Tucker, and Richard A. Pauptit* AstraZeneca Alderley Park Macclesfield, Cheshire SK10 4TG United Kingdom
Summary Human thymidine phosphorylase (HTP), also known as platelet-derived endothelial cell growth factor (PDECGF), is overexpressed in certain solid tumors where it is linked to poor prognosis. HTP expression is utilized for certain chemotherapeutic strategies and is also thought to play a role in tumor angiogenesis. We determined the structure of HTP bound to the small molecule inhibitor 5-chloro-6-[1-(2-iminopyrrolidinyl) methyl] uracil hydrochloride (TPI). The inhibitor appears to mimic the substrate transition state, which may help explain the potency of this inhibitor and the catalytic mechanism of pyrimidine nucleotide phosphorylases (PYNPs). Further, we have confirmed the validity of the HTP structure as a template for structure-based drug design by predicting binding affinities for TPI and other known HTP inhibitors using in silico docking techniques. This work provides the first structural insight into the binding mode of any inhibitor to this important drug target and forms the basis for designing novel inhibitors for use in anticancer therapy. Introduction HTP (EC 2.4.2.4) has a primary catabolic role in pyrimidine nucleoside metabolism, catalyzing the reversible phosphorolysis of thymidine, deoxyuridine, and their analogs (except deoxycytidine) to their respective bases and 2-deoxyribose-1-phosphate (2-dR-1-P) (Krenitsky et al., 1981). HTP is a member of the PYNP family which in higher species consists of two PYNPs, TP and uridine phosphorylase (UP), and in lower organisms consists of only one PYNP with dual (TP and UP) activity (Krenitsky et al., 1965; Pugmire and Ealick, 2002). Kinetic studies using E. coli TP (EcTP) and rabbit muscle TP have shown that the enzyme follows a sequential mechanism whereby phosphate is the first substrate to bind and 2-dR-1-P is the last product to leave (Krenitsky, 1968; Schwartz, 1971). These results are supported by evidence that the enzyme is stabilized by phosphate and 2-dR-1-P, but not by thymidine or thymine (Krenitsky and Tuttle, 1982). Structural studies on B. stearothermophilus PYNP (BsPYNP), bound to pseudouridine and phosphate, have *Correspondence: [email protected]
led Pugmire and Ealick to propose a catalytic mechanism for this family of enzymes in which the pyrimidine nucleoside is bound in a high-energy conformation causing the weakening of the glycosidic bond. The glycosidic bond is weakened further by flow of electrons from O-5⬘ of the ribose moiety to the pyrimidine ring, while the phosphate ion is positioned to attack the C-1⬘ position of the ribose ring to yield 2-dR-1-P and the free pyrimidine base (Pugmire and Ealick, 1998, 2002) (Figure 1). HTP is able to transfer the deoxyribosyl moiety from one pyrimidine base to another thus playing an additional vital role in base/nucleotide salvage (Zimmerman and Seidenberg, 1964a, 1964b). HTP is now understood to be identical to PD-ECGF, which is an angiogenic activity identified in platelet lysates (Miyazono et al., 1987), and to gliostatin which has important physiological roles in neuronal and glial development (Asai et al., 1992). The activity of HTP correlates with hypoxia and the induction of angiogenesis in a number of malignant and hyperproliferative disease states. HTP overexpression and elevated enzyme activity have been found in the plasma of cancer patients (Pauly et al., 1977), and HTP expression has been linked with the invasive/metastatic capacity of gastric (Maeda et al., 1996; Takebayashi et al., 1996b), bladder (Mizutani et al., 1997; O’Brien et al., 1995, 1996), and colorectal (Takebayashi et al., 1996a) tumors. HTP activity is also elevated in ovarian (Reynolds et al., 1994), oesophageal, and pancreatic cancers (Takebayashi et al., 1996a), and has been shown to be associated with a more aggressive phenotype in nonsmall cell lung cancer node-negative disease (Fox et al., 1996). HTP overexpression has been implicated in other angioproliferative disorders, including rheumatoid arthritis (Takeuchi et al., 1994), psoriasis (Creamer et al., 1997), and hyperproliferation of the vasa vasorum in artherosclerosis (Ignatescu et al., 1999). The in vivo angiogenic activity associated with HTP is thought to be mediated by dephosphorylated 2-deoxyribose and although the mechanism is unclear it does require the enzyme activity of HTP (Miyadera et al., 1995; Moghaddam et al., 1995; Sumizawa et al., 1993). The activity of HTP inactivates certain chemotherapeutic agents such as TFT (trifluorothymidine) (Fukushima et al., 2000) and can also be used to catalyze the conversion of prodrugs such as tegafur (1-(tetrahydro2-furanyl)-5-fluorouracil) and capecitabine (N4-pentyloxycarbonyl-5⬘-deoxy-5-fluorocytidine) to anticancer nucleotides which interfere with DNA replication (Patterson et al., 1995; Miwa et al., 1998). Whereas the latter type of therapy would be incompatible with HTP inhibition, inhibitors of HTP may be used to provide alternative agents to control angiogenesis and in combination with TFT therapy. These alternative approaches would prove especially attractive in cases where primary fluoropyrimidine therapy has failed. Recently, a number of HTP inhibitors have been reported, the most potent of which is TPI (5-chloro-6-[1(2-iminopyrrolidinyl) methyl] uracil hydrochloride) with Ki ⫽ 20 nM. TPI causes a substantial reduction in tumor
Structure 76
Figure 1. Reaction Catalyzed by HTP
growth when given continuously to mice carrying HTP overexpressing tumors by increasing the proportion of apoptotic cells and possibly also inhibiting angiogenesis (Matsushita et al., 1999). The combination of TFT and TPI in a 1:0.5 molar ratio (TAS-102) is currently tested in phase I trials and can be administered orally (Thomas et al., 2002). Less potent HTP inhibitors have been designed using models of HTP based on the structure of EcTP or BsPYNP (Cole et al., 2003; Price et al., 2003). Both bacterial enzymes (ⵑ90 kDa) possess around 40% sequence identity with HTP and, like the larger mammalian TPs (ⵑ110 kDa), are homodimers (Desgranges et al., 1981; Kraut and Yamada, 1971; Pugmire and Ealick, 1998; Schwartz, 1978). Each subunit of the bacterial enzymes consists of a small ␣-helical domain (␣ domain) and a large mixed ␣-helical and  sheet domain (␣/ domain) separated by a large cleft containing the substrate binding site (Pugmire and Ealick, 1998; Walter et al., 1990). Movement of the two domains allows for an open (empty) or closed (substrate-bound) conformation of the enzyme (Pugmire and Ealick, 1998; Pugmire et al., 1998). Here, we report the crystal structure of HTP in complex with TPI in an active (closed) conformation. Since this is the first structure of any PYNP family member bound to an inhibitor, it will provide the basis for further structure-based drug design studies with the aim of increasing current inhibitor potency and specificity, or developing new anticancer drug candidates. Results and Discussion HTP Proteolysis and Structure Determination Crystals of recombinant HTP which diffract to 3.5 A˚ have been reported (Spraggon et al., 1993). In our hands, however, crystallization trials with HTP (residues 12– 482) have for many years failed to produce diffractiongrade crystals. Sequence alignments (data not shown) of HTP with EcTP and BsPYNP show a glycine-rich N-terminal extension of 32 residues in HTP which is assumed to lack defined secondary structure. Since disordered N- or C termini of proteins could conceivably lead to crystallization difficulties, limited proteolysis was performed with a range of proteases in an attempt to define a minimal domain more amenable to successful crystallization. This yielded a product from which residues 409 and 410 had been cleaved, as shown by Edman sequencing and electrospray mass spectrometry, and
which crystallized. Whereas we had initially assumed we had removed a portion of the C-terminal domain, remote from the active site, analysis of the crystals by SDS-PAGE showed the crystallized product included two fragments: 12–408 and 411–482. Functional assays carried out on the proteolyzed HTP show that it possesses the same degree of activity as full-length HTP toward its substrate (data not shown). Crystals of HTP belong to the space group C2 and the asymmetric unit contains a single HTP protomer. The three-dimensional structure of HTP was determined by the molecular replacement method, using the crystal structure of BsPYNP as the search model (see Experimental Procedures). The structure was refined to an R factor of 18.9% and an Rfree of 25.8% (Table 1). The final Fo-Fc electron density for the bound TPI is shown in Figure 2. Recombinant HTP spans residues 12–482; however, the proteolytic treatment removed residues 409 and 410 causing the remainder of the loop (residues 407–408 and 411–414) to become disordered. No electron density was observed for residues 12–32 (N terminus), 238–239, and 481–482 (C terminus). The structure of HTP contains one molecule of TPI in the active site. Structure of HTP and Comparison to Other PYNPs The HTP protomer (Figure 3A) has a similar fold to that reported for EcTP and BsPYNP (Pugmire and Ealick, 1998; Walter et al., 1990) and comprises an ␣ domain and a mixed ␣/ domain connected by three polypeptide loops (Figure 3B). The ␣ domain consists of six ␣ helices (␣1–␣4 and ␣8–␣9), and the ␣/ domain consists of a central mixed  sheet (1–5, 13) surrounded by ␣ helices (␣5–␣7 and ␣10–␣16) and two small antiparallel  sheets (consisting of strands 6, 7, 9, and 11, and 8, 10, and 12 respectively) flanked by two ␣ helices (␣17 and ␣18). The three loops act as a hinge allowing the two domains to move between the open (inactive) and closed (active) conformations, bringing together the active site residues (Pugmire and Ealick, 1998; Pugmire et al., 1998). The main structural differences between HTP, EcTP, and BsPYNP are in the ␣/ domain where HTP has an additional helix (␣15), an extra turn in ␣16, and a more extended C-terminal region. The small antiparallel  sheet (8, 10, and 12) is present in HTP and BsPYNP but not in EcTP. An HTP dimer is generated from the single protomer that constitutes the asymmetric unit by rotation about the crystallographic 2-fold axis (Figure 3C). The core of
Structure of Human Thymidine Phosphorylase 77
Table 1. Data Collection, Refinement Statistics, and Model Quality TP-TPI Complex Resolution range (A˚) Spacegroup Z Cell dimensions a b c  Wavelength (A˚) Observed reflections Unique reflections Highest resolution shell (A˚) Redundancy I/(I) Completeness (%) Rmerge (%) Rwork (%) Rfreeb (%) Rms bond lengths (A˚) Rms bond angles (⬚) Number of nonhydrogen atoms Protein Inhibitor Water Mean temperature factors (A˚2) Protein Inhibitor Ramachandran plotc (%) Residues in most favored regions Residues in additional allowed regions Residues in disallowed regions
91.3–2.1 C2 4 70.0 A˚ 66.1 A˚ 96.9 A˚ 106.5⬚ 0.9780 32,998 15,827 2.21-2.11 2.1 (1.3)a 15.5 (1.8) 63.7 (11.2) 3.0 (19.3) 18.8 (34.8) 25.8 (57.7) 0.015 1.5 3,214 16 75 23.9 19.9 94.5 5.5 0
a
Values in parentheses correspond to the high resolution shell. Rfree was calculated against 5.4% of the complete dataset excluded from refinement. c Statistics obtained from PROCHECK (Laskowski et al., 1993). b
the dimer interface is analogous to that of the BsPYNP structure and consists of a coiled coil formed by helices ␣3 from each monomer. Also, helix ␣1 of one monomer packs against the loop connecting helices ␣8 and ␣9 of the second monomer. In the BsPYNP structure, residues 364–371, which correspond to the clipped loop of HTP, extend across the active site cleft and appear to act as a
latch, holding the ␣ and ␣/ domains together. Residues A366, K367, and K368 (in BsPYNP) from the tip of this loop make contacts with the opposing molecule in the dimer. It has been proposed that dimer interactions around this loop have a functional role in closing the active site cleft (Pugmire and Ealick, 1998). Residues S409 and R410 (corresponding to A366 and K367 in BsPYNP) are not present in the HTP that we crystallized and consequently this loop is disordered in our structure. However, HTP still adopts a closed conformation (see below) and the species crystallized has been shown to be fully active. Thus, for HTP, this region is not essential for the formation of an active enzyme. Structural studies on BsPYNP (Pugmire and Ealick, 1998) suggest that binding of phosphate results in the formation of a key hydrogen bond between residues H116 and G205 (corresponding to H150 and A239 of HTP), causing an 8⬚ rotation of the ␣/ domain with respect to the ␣ domain compared to the structure in the absence of phosphate. Subsequent binding of the pyrimidine base would be responsible for an additional 20⬚ rotation to form the fully closed active site (Pugmire and Ealick, 1998, 2002). Phosphate is absent from the catalytic site in our structure, A239 is disordered, and therefore no hydrogen bond is observed between H150 and A239. Our HTP structure appears to be in the active (closed) conformation. Alignment of the ␣ domain of HTP with the corresponding ␣ domains of EcTP and BsPYNP gave rms deviations of 0.82 and 0.67 A˚, respectively, for 100 equivalent C␣ atoms. Alignment of the HTP ␣/ domain with the corresponding ␣/ domains of EcTP and BsPYNP gave rms deviations of 1.94 and 1.46 A˚, respectively, over 282 C␣ atoms. However, an alignment of the entire HTP protomer with the EcTP and BsPYNP structures gave rms deviations of 3.55 and 1.71 A˚, respectively, over 383 C␣ atoms. While the individual domains superpose well, the overall structural similarity between HTP and EcTP is less than that for HTP and BsPYNP. By analogy to BsPYNP, HTP is in the active (closed) conformation (Figure 4). Thus, the presence of the inhibitor TPI alone, without phosphate, is sufficient to cause closure of the ␣ and ␣/ domains to form the active conformation. The structure of BsPYNP contains a metal binding site
Figure 2. Electron Density Corresponding to TPI Stereo view of the final 2.1 A˚ Fo-Fc omit map to within 2 A˚ of the inhibitor molecule contoured at 3.0 (cyan). Nitrogen atoms are blue; oxygen atoms, red; chlorine atom, lime green; carbon atoms and bonds, light purple. All molecular graphics were produced using Bobscript (Esnouf, 1999) and Raster3D (Merrit and Murphy, 2003) except Figure 7, which was produced using PyMOL (DeLano, 2002).
Structure 78
Structure of Human Thymidine Phosphorylase 79
Figure 4. Fold Comparison of HTP Comparison of the fold of HTP with EcTP and BsPYNP. The three molecules are shown in the same orientation, and colored red to blue (N terminus to C terminus). Bound pseudouridine (light purple) and phosphate (red) in the BsPYNP structure and TPI (light purple) in the HTP structure are shown as cpk models. Arrows indicate domain movement giving the active conformation of the enzyme.
adjacent to the phosphate binding site. This unidentified penta-coordinate metal interacts with the backbone carbonyl oxygen atoms of residues L243, A246, and G88, and the side chains of E255 and T90 (BsPYNP). In the HTP structure, there is no metal ion present, but the N atom of K124 (which corresponds to T90 of BsPYNP) adopts a position similar to the BsPYNP metal ion and interacts with the equivalent residues, possibly fulfilling the same structural role. Inhibitor Binding, Specificity and Insights into the Catalytic Mechanism of HTP The HTP active site may be described as having two components, each interacting with one part of the TPI molecule. The first component of the active site is formed by residues from the ␣ domain and H116 from the ␣/ domain, which interact with the 5-chlorouracil moiety. The second component of the active site, which interacts with the 2-iminopyrrolidinyl moiety, is formed by S117 from the ␣/ domain and a network of water molecules which occupy the phosphate binding site (Figure 5A). The 5-chlorouracil of TPI makes strong hydrogenbonding interactions with R202, S217, K221, and H116 (Figures 5A and 5B). These residues are conserved in the BsPYNP structure (R168, S183, K187, and H82 respectively) and, with the exception of H82 (H116 in HTP), form a similar hydrogen-bonding network with the uracil moiety of the substrate analog pseudouridine (Figure 5C). The number of hydrogen bonds between BsPYNP and the uracil ring is lower than between HTP and TPI, and the equivalent distances are longer, consistent with the tighter binding of TPI to HTP. Another contribution to the tight interactions between HTP and the 5-chlorouracil of TPI is the chlorine atom, which is held in a hydrophobic pocket formed by V208 and I214 of the ␣
domain and L148 and V241 of the ␣/ domain, and forces the 2-oxo, 3-amido, and 4-oxo groups of the 5-chlorouracil moiety closer to their interacting partners relative to uracil. The C␥2 of V241 lies 4.4 A˚ from the chlorine, indicative of good van der Waals interactions between the two atoms. By analogy, the methyl group of thymine would occupy the same position as the chlorine and would make similar interactions with V241. Uracil, however, lacks a substituent at this position making the distance between the 5 position and the C␥2 of V241 approximately 6.0 A˚, creating a hydrophobic cavity. Interestingly, of the four residues which form the hydrophobic pocket in HTP, only V241 is not conserved in BsPYNP: it corresponds to F207. In the bacterial structure, the distance between the 5 position of uracil and the C⑀2 of F207 is 3.9 A˚, consistent with good van der Waals interactions between these atoms. Binding of thymine would result in steric clashes unless the phenyl ring of F207 were rotated by 90⬚ around the C-C␥ bond to accommodate the thymine methyl group at the 5 position. Such rotation of the phenyl ring around the C-C␥ bond may allow BsPYNP to accept both uracil and thymine as substrates, without the need for extensive structural rearrangement of the active site. This suggests the F207 side chain in BsPYNP may be less discriminating than V241 in HTP. Whereas the basis for specificity has been ascribed to the high affinity of HTP for the 2⬘-deoxyribose moiety of the thymidine substrate (Pugmire and Ealick, 2002), there may also be a contribution from V241. Unfortunately, the structure of HTP complexed with TPI offers no direct insights as to the basis of specific interactions between HTP and the thymidine 2⬘-deoxyribose moiety. The 2-iminopyrrolidinyl moiety of TPI makes two hydrogen bonds, one to the carbonyl oxygen of S117 and one to a water molecule (Figures 5A and 5B). The posi-
Figure 3. Structure of HTP (A) Stereo view of the HTP monomer showing the main-chain trace of the ␣ domain (cyan), the hinge regions (green), and the mixed ␣/ domain (brown). TPI (light purple) is shown as a cpk model in the active site situated between the ␣ and ␣/ domains. (B) Topology diagram of HTP color coded in the same manner as in (A). (C) Stereo view of the HTP dimer generated by applying 2-fold crystallographic symmetry to the monomer. HTP monomers are colored cyan and brown and TPI (light purple) is shown as a cpk model. The position of the proteolytically cleaved loop is indicated by an arrow.
Structure of Human Thymidine Phosphorylase 81
tively charged character of the 2-iminopyrrolidine may mimic the oxocarbenium ion-like transition state proposed for thymidine phosphorolysis (Figure 1). This suggests an orientation for the ribose moiety of thymidine which is rotated by approximately 180⬚ with respect to previous models (Pugmire and Ealick, 2002). Water molecules W1 and W2 in HTP (Figures 5A and 5B) mimic the phosphate oxygen atoms in the BsPYNP structure, suggesting that these oxygen positions could be relevant and the orientation of the phosphate during catalysis could be similar to that observed in the BsPYNP crystal structure (Figure 5A). The role of H116, which is absolutely conserved in the PYNP family, has been the subject of some debate. It is believed to be involved in either stabilizing the transition state during catalysis or in donating a proton to the N-1 of the pyrimidine ring following glycosidic cleavage (Figure 1). Given that the pKa of TPI has been measured at 6.05 (our unpublished data), it is likely that at the pH of crystallization (pH 6.5) the 1-amido group of 5-chlorouracil is anionic and makes a hydrogen bond with the protonated N⑀2 of H116, and N␦2 of H116 serves as the hydrogen-bond acceptor for N of K222. The side chains of both D114 and E225 are positioned within hydrogen-bonding distance of each other and of K222 in an unusual arrangement (Figure 5A), suggesting this triad forms a protonshuttling mechanism whose role is to deliver a proton to H116. Catalysis would therefore proceed as previously described followed by protonation of the pyrimidine ring at the 1-amido position by H116 via the proton shuttle. Thus, the structure provides evidence for the role of H116 as the proton donor during thymidine phosphorolysis. In addition, the resemblance of TPI to the zwitterionic transition state during thymidine phosphorolysis may help explain its high potency. Implications for Drug Design In order to establish whether the HTP structure could provide the basis for in silico-based drug design, calculations were carried out in an attempt to predict binding affinities for TPI and other known HTP inhibitors. Docking TPI into the HTP active site using GLIDE (Eldridge et al., 1997) yielded a bound position for the inhibitor in substantial agreement with experiment (rms deviation ⫽ 0.42 A˚ for the nonhydrogen atoms) (Figure 6A). The ionic status of TPI on binding to the enzyme was investigated using a binding energy calculation based on PoissonBoltzmann electrostatics (Grant et al., 2001). Calculations for the TPI binding mode found by experiment indicate that the inhibitor binds as the zwitterion (Table 2). Three further ligands (Figure 6B) exhibiting a diverse
Figure 6. Structures of HTP Inhibitors Used in Docking Studies (A) View of the HTP active site color coded and labeled in the same manner as in Figure 5A showing the superposed docked TPI molecule (yellow). (B) TPI. (C) Compound II, 6-amino-5-bromouracil (Langen et al., 1967). (D) Compound III, 5-chloro-6-[(2-nitroimidazol1-yl) methyl] uracil (Cole et al., 2003). (E) Compound IV, N-(2,4-dioxo1,2,3,4-tertahydrothieno[3,2-d] pyrimidin-7-yl) guanidine (Price et al., 2003).
range of chemical features and affinities were also docked using GLIDE and their binding energies estimated using the same electrostatic protocol as for TPI. For these ligands, experimental values of pKa were unavailable and these were estimated using ACD software (Advanced Chemistry Development, http://www.acdlabs. com), thus allowing corrected estimates of the binding constants (Kcorr) to be made. In the case of ligands (II) and
Figure 5. Active Site (A) Stereo view of the HTP active site highlighting important interactions. No phosphate is present in HTP, but the position of the phosphate in the BsPYNP crystal structure is indicated. Nitrogen atoms are blue; oxygen atoms, red; chlorine atom, lime green; phosphate atom, lavender; carbon atoms and bonds, light purple (inhibitor) and white (protein). Residues which form direct interactions with the inhibitor have their labels colored light purple and residues which form the hydrophobic pocket around the chlorine atom have their labels colored green. Other residues and selected water molecules are labeled in black. (B) Schematic representation of the HTP active site highlighting important interactions. The ring positions of the 5-chlorouracil moiety are numbered and hydrogen-bonding distances (A˚) are labeled in blue. (C) Schematic representation of the BsPYNP active site highlighting important interactions. The ring positions of the uracil moiety are numbered and hydrogen-bonding distances (A˚) are labeled in blue.
Structure 82
Table 2. Docking and Scoring Results for Some HTP Ligands Compound TPI cation TPI zwitterion II neutral II anion III neutral III anion IV cation IV zwitterion
pKaa 6.1 (6.9) (7.1) (7.3) (9.5)
Kcalc
Kcorrb
Inactived 345 nM 74 mM 0.375 M Inactived 6.0 M Inactived 1.85 M
Inactived 360 nM 111 mM 0.6 M Inactived 10.8 M Inactived 235 M
IC50c
of phosphate. The HTP structure provides a framework including many replaceable water molecules, which offers a design opportunity for further developing existing inhibitors or to design novel inhibitors (Figure 7).
5–20 nM 1.6 M 21.7 M 76 M
a
Values in parentheses were calculated using ACD software (Advanced Chemistry Development, http://www.acdlabs.com). Kcorr is the corrected value of Kcalc to account for the fraction of the binding form of the compound that would be present in the experimental assay solution. c Experimental IC50 data were taken from (Cole et al., 2003) (TPI, compound II, and compound III) and from (Price et al., 2003) (TPI and compound IV). d As interpreted from positive ⌬G. b
(III), the anions would appear to be the active entities. For ligand (IV), as for TPI, it would appear the zwitterion is bound and the poorer affinity relates to the higher pKa estimated as a consequence of the annelated thiophene ring. The common pyrimidine-dione substructure docks in a similar manner for all the ligands and the estimated affinities are correctly ranked in comparison to the experimental IC50s. The estimated TPI binding affinity is poorer than expected and we have examined the effect of introducing phosphate into the binding site. This gives rise to an enhanced calculated affinity for TPI (0.5 nM) and IV (200 nM) but is deleterious to the binding of the anions II (14 mM) and III (4 mM), probably due to the electrostatic repulsion between like charges. We know from the structure that phosphate binding, although mandatory for the enzyme mechanism, is not a necessary requirement for inhibitor binding. These studies indicate that the HTP structure can be used to predict docking modes and relative binding affinities for candidate inhibitors of HTP. In addition, the probable ionic status of the bound form of the ligand can be suggested. Structure-based design studies to date have been impeded by the lack of an HTP structure and the limitations of homology models based on bacterial structures. An active conformation of the enzyme is needed for these studies and the structure of HTP bound to an inhibitor shows that this can be achieved in the absence
Experimental Procedures Sample Preparation and Crystallization HTP was expressed in E. coli from plasmid pMOAL-10T (Moghaddam and Bicknell, 1992). The plasmid encodes a product consisting of glutathione S-transferase (GST) fused to HTP residues 12–482 (based on Swissprot accession number P19971) via a thrombincleavable linker. Cells were induced overnight at 15⬚C with 0.1 mM IPTG to maximize soluble accumulation, harvested, and lysed in buffer A (20 mM Tris/HCl, pH 7.4, 100 mM NaCl, 5 mM DTT) by highpressure homogenization. Cell debris were removed by centrifugation and the lysate supernatant loaded onto a glutathione-sepharose column equilibrated with buffer A. The column was washed with at least 10 vol of buffer A prior to eluting the product with buffer A containing 20 mM reduced glutathione. The product was incubated for 4 hr at 37⬚C with thrombin (10 U/mg fusion protein) and separated from the cleaved GST by size-exclusion chromatography on a Superdex 75 column equilibrated with buffer B (20 mM Tris/HCl, pH 7.5, 150 mM NaCl, 1 mM DTT). HTP was subjected to limited proteolysis by trypsin (1:100 w/w; 40 min at 37⬚C) followed by addition of 1 mM phenylmethyl sulphonyl fluoride. Proteolyzed HTP was loaded onto a Superdex 75 column equilibrated in buffer C (20 mM Tris/HCl, pH 7.4, 50 mM NaCl) and concentrated to 8 mg ml⫺1. Prior to crystallization, 5 mM TPI (Taiho Pharmaceutical Co. Ltd., Japan) was added to the protein and incubated at 4⬚C for 30 min. The HTPTPI complex was crystallized at 15⬚C using the hanging-drop vapor diffusion method by mixing equal volumes of complex solution and precipitant (8% PEG 1000 [w/v], 50 mM malate/imidazole, pH 6.5, 100 mM MgCl2). Crystals grew in some wells only and reached their full size after approximately 3 weeks. Crystals were harvested into cryoprotectant solution (9% PEG 1000, 50 mM malate/imidazole, pH 6.5, 100 mM MgCl2, 20% [v/v] glycerol) and flash cooled in a nitrogen stream at 100 K. Structure Determination and Refinement HTP diffraction data were collected from a single crystal at 100 K on an ADSC CCD detector on beamline 14-2 at the SRS (Daresbury, UK) and processed with MOSFLM and SCALA (CCP4, 1994; Leslie, 1992). Diffraction data extends to 2.1 A˚, but the data are essentially complete only to 2.7 A˚. Statistics for the data collection are given in Table 1. The structure was solved by the molecular replacement method using a modification (nonconserved residues truncated to alanine or glycine; ligand and water molecules excluded) of the crystal structure of BsPYNP (PDB code 1brw, chain B) as a search model. Rotation and translation searches were carried out using the program AMORE (CCP4, 1994). The statistically most promising solution was eventually proven correct, though it was with considerable difficulty that sufficient clarity and connectivity could be obtained in the first derived electron density maps to allow rebuilding
Figure 7. Inhibitor Design Orthogonal views of the HTP active site with bound TPI (light purple). Van der Waals radii for bound water molecules are shown (red), highlighting the available space exploitable in drug design.
Structure of Human Thymidine Phosphorylase 83
and refinement to be initiated. The connectivity was more apparent when in the first instance the resolution was limited to 2.6 A˚, where the completeness of the data was 90%. This gave readily interpretable density for the secondary structure elements. Automated model building using MAID (Levitt, 2001) proved a useful tool to reposition the secondary structure elements slightly from their positions in the trial model. Continued rounds of simulated annealing, energy minimization and temperature factor refinement using CNX (Accelrys, www.accelrys.com) extending the resolution to the limit of 2.1 A˚ eventually resulted in the connectivity between the secondary structure elements becoming very clear in the electron density, allowing the building of a complete model. Final refinement using CNX and Refmac5 (CCP4, 1994) lead to an R factor of 18.9% and an Rfree of 25.8%. Statistics for the refinement and quality of the model are given in Table 1.
References
Docking and Scoring Studies HTP was prepared for docking and scoring studies as follows: WHATCHECK (Hooft et al., 1996) analysis suggested that two histidines (H116 and H441) exist as the N⑀2(H) tautomer and one is protonated (H319) with the remainder as the N␦1(H) tautomer. For residues modeled with dual conformations, the “A” locations for R329 and M273 were chosen. Of the two juxtaposed carboxylates, D114 and E225, the latter was protonated. CHARMM (Mackerell et al., 1998) was used to add the hydrogen atoms and their positions were optimized while keeping the heavy atoms fixed. Crystallographic water molecules were then removed to allow free docking of the ligands. For the phosphate studies, dihydrogen phosphate was added in the position found in the BsPYNP structure. Ligand dockings were performed using GLIDE 2.5 (Eldridge et al., 1997) using the default settings, in extended sampling mode. Ligands were treated as being flexible and the van der Waals radii of the ligand nonpolar atoms were scaled by a factor of 0.8. Binding affinities were calculated using the following equation: ⌬Gcalc ⫽ PB ⫹ ␥⌬ASA ⫹ ␣NR ⫹ RBE, where PB is the change in coulombic interaction energy and electrostatic desolvation energy in kcal/mol on complexation as determined by the Poisson-Boltzmann treatment (Grant et al., 2001). Atomic charges for both the protein and the ligand were generated using the MMFF force field (Halgren and Halgren, 1996). ⌬ASA is the change in accessible area on complexation and is converted to a nonpolar (hydrophobic) contribution by means of the ␥ coefficient (0.025 kcal/mol/A2). NR is the number of rotatable bonds and is converted to a measure of the loss of conformational entropy on binding by the coefficient ␣ (0.33 kcal/mol/bond). RBE represents the loss in rigid body entropy on complex formation and is assigned a fixed value of 4.71 kcal/mol. The values for Kcalc were generated using ⌬Gcalc ⫽ ⫺RTln(Kcalc).
Creamer, D., Jaggar, R., Allen, M., Bicknell, R., and Barker, J. (1997). Overexpression of the angiogenic factor platelet-derived endothelial cell growth factor/thymidine phosphorylase in psoriatic epidermis. Br. J. Dermatol. 137, 851–855.
Estimates of compound pKa’s were made using ACD software (Advanced Chemistry Development, http://www.acdlabs.com); these pKa’s were used to correct the calculated binding affinities in accordance with the fraction of active species present at the assay pH of 7.4 using the Henderson-Hasselbach equation (Albert and Serjeant, 1962).
Acknowledgments We would like to thank Roy Bicknell (Angiogenesis Group, Molecular Oncology Laboratory, Weatherall Institute of Molecular Medicine, Oxford, UK) for the gift of plasmid pMOAL-10T and Delphine Dorison-Duval for determining the pKa value of TPI. We would also like to thank Jon Read and Antonio Pineda-Lucena and in particular Philip Edwards (Manchester University) for helpful discussions. We gratefully acknowledge the staff, in particular James Nicholson, and use of facilities at the SRS, Daresbury, UK.
Received: August 27, 2003 Revised: September 24, 2003 Accepted: October 1, 2003 Published: January 13, 2004
Albert, A., and Serjeant, E.P. (1962). Ionization Constants of Acids and Bases. A Laboratory Manual. (London: Methuen and Co. Ltd.). Asai, K., Nakanishi, K., Isobe, I., Eksioglu, Y.Z., Hirano, A., Hama, K., Miyamoto, T., and Kato, T. (1992). Neurotrophic action of gliostatin on cortical neurons. Identity of gliostatin and platelet-derived endothelial cell growth factor. J. Biol. Chem. 267, 20311–20316. CCP4 (Collaborative Computational Project 4) (1994). The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D 50, 760–763. Cole, C., Reigan, P., Gbaj, A., Edwards, P.N., Douglas, K.T., Stratford, I.J., Freeman, S., and Jaffar, M. (2003). Potential tumor-selective nitroimidazolylmethyluracil prodrug derivatives: inhibitors of the angiogenic enzyme thymidine phosphorylase. J. Med. Chem. 46, 207–209.
DeLano, W.L. (2002). The PyMOL Molecular Graphics System (San Carlos, CA: DeLano Scientific). Desgranges, C., Razaka, G., Rabaud, M., and Bricaud, H. (1981). Catabolism of thymidine in human blood platelets: purification and properties of thymidine phosphorylase. Biochim. Biophys. Acta 654, 211–218. Eldridge, M.D., Murray, C.W., Auton, T.R., Paolini, G.V., Mee, R.P., and Murray, C.W. (1997). Empirical scoring functions.1. The development of a fast empirical scoring function to estimate the binding affinity of ligands in receptor complexes. J. Comput. Aided Mol. Des. 11, 425–445. Esnouf, R.M. (1999). Further additions to MolScript version 1.4, including reading and contouring of electron-density maps. Acta Crystallogr. D 55, 938–940. Fox, S.B., Westwood, M., Moghaddam, A., Comley, M., Turley, H., Whitehouse, R.M., Bicknell, R., Gatter, K.C., and Harris, A.L. (1996). The angiogenic factor platelet-derived endothelial cell growth factor/thymidine phosphorylase is up-regulated in breast cancer epithelium and endothelium. Br. J. Cancer 73, 275–280. Fukushima, M., Suzuki, N., Emura, T., Yano, S., Kazuno, H., Tada, Y., Yamada, Y., and Asao, T. (2000). Structure and activity of specific inhibitors of thymidine phosphorylase to potentiate the function of antitumor 2⬘-deoxyribonucleosides. Biochem. Pharmacol. 59, 1227– 1236. Grant, J.A., Pickup, B.T., and Nicholls, A. (2001). A smooth permittivity function for Poisson-Boltzmann solvation methods. J. Comput. Chem. 22, 608–640. Halgren, T.A., and Halgren, T.A. (1996). Merck molecular force field.1. Basis, form, scope, parameterization, and performance of mmff94. J. Comput. Chem. 17, 490–519. Hooft, R.W.W., Sander, C., Vriend, G., and Hooft, R.W.W. (1996). Positioning hydrogen atoms by optimizing hydrogen-bond networks in protein structures. Proteins 26, 363–376. Ignatescu, M.C., Gharehbaghi-Schnell, E., Hassan, A., Rezaie-Majd, S., Korschineck, I., Schleef, R.R., Glogar, H.D., and Lang, I.M. (1999). Expression of the angiogenic protein, platelet-derived endothelial cell growth factor, in coronary atherosclerotic plaques: in vivo correlation of lesional microvessel density and constrictive vascular remodeling. Arterioscler. Thromb. Vasc. Biol. 19, 2340–2347. Kraut, A., and Yamada, E.W. (1971). Cytoplasmic uridine phosphoryaseft liver. Characterization and kinetics. J. Biol. Chem. 246, 2021– 2030. Krenitsky, T.A. (1968). Pentosyl transfer mechanisms of the mammalian nucleoside phosphorylases. J. Biol. Chem. 243, 2871–2875. Krenitsky, T.A., and Tuttle, J.V. (1982). Correlation of substratestabilization patterns with proposed mechanisms for three nucleoside phosphorylases. Biochim. Biophys. Acta 703, 247–249. Krenitsky, T.A., Mellors, J.W., and Barclay, R.K. (1965). Pyrimidine nucleosidases: their classification and relationship to uric acid ribonucleoside phosphorylase. J. Biol. Chem. 240, 1281–1286.
Structure 84
Krenitsky, T.A., Koszalka, G.W., and Tuttle, J.V. (1981). Purine nucleoside synthesis, an efficient method employing nucleoside phosphorylases. Biochemistry 20, 3615–3621. Langen, P., Etzold, G., Barwolff, D., and Preussel, B. (1967). Inhibition of thymidine phosphorylase by 6-aminothymine and derivatives of 6-aminouracil. Biochem. Pharmacol. 16, 1833–1837.
Price, M.L., Guida, W.C., Jackson, T.E., Nydick, J.A., Gladstone, P.L., Juarez, J.C., Donate, F., and Ternansky, R.J. (2003). Design of novel N-(2,4-dioxo-1,2,3,4-tetrahydro-thieno[3,2-d]pyrimidin-7-yl)guanidines as thymidine phosphorylase inhibitors, and flexible docking to a homology model. Bioorg. Med. Chem. Lett. 13, 107–110.
Laskowski, R.A., Moss, D.S., and Thornton, J.M. (1993). Main-chain bond lengths and bond angles in protein structures. J. Mol. Biol. 231, 1049–1067.
Pugmire, M.J., and Ealick, S.E. (1998). The crystal structure of pyrimidine nucleoside phosphorylase in a closed conformation. Structure 6, 1467–1479.
Leslie, A.G.W. (1992). Recent changes to the MOSFLM package for processing film and image plate data. Joint CCP4 ⫹ ESF-EAMCB Newsletter on Protein Crystallography No. 26.
Pugmire, M.J., and Ealick, S.E. (2002). Structural analyses reveal two distinct families of nucleoside phosphorylases. Biochem. J. 361, 1–25.
Levitt, D.G. (2001). A new software routine that automates the fitting of protein X-ray crystallographic electron-density maps. Acta Crystallogr. D 57, 7–9.
Pugmire, M.J., Cook, W.J., Jasanoff, A., Walter, M.R., and Ealick, S.E. (1998). Structural and theoretical studies suggest domain movement produces an active conformation of thymidine phosphorylase. J. Mol. Biol. 281, 285–299.
Mackerell, A.D., Bashford, D., Bellott, M., Dunbrack, R.L., Evanseck, J.D., Field, M.J., Fischer, S., Gao, J., Guo, H., Ha, S., et al. (1998). All-atom empirical potential for molecular modeling and dynamics studies of proteins. J. Phys. Chem. B 102, 3586–3616. Maeda, K., Chung, Y.S., Ogawa, Y., Takatsuka, S., Kang, S.M., Ogawa, M., Sawada, T., Onoda, N., Kato, Y., and Sowa, M. (1996). Thymidine phosphorylase/platelet-derived endothelial cell growth factor expression associated with hepatic metastasis in gastric carcinoma. Br. J. Cancer 73, 884–888. Matsushita, S., Nitanda, T., Furukawa, T., Sumizawa, T., Tani, A., Nishimoto, K., Akiba, S., Miyadera, K., Fukushima, M., Yamada, Y., et al. (1999). The effect of a thymidine phosphorylase inhibitor on angiogenesis and apoptosis in tumors. Cancer Res. 59, 1911–1916. Merrit, E.A., and Murphy, M.E.P. (2003). Raster 3D version 2.0—a program for photorealistic molecular graphics. Acta Crystallogr. D 50, 869–873. Miwa, M., Ura, M., Nishida, M., Sawada, N., Ishikawa, T., Mori, K., Shimma, N., Umeda, I., and Ishitsuka, H. (1998). Design of a novel oral fluoropyrimidine carbamate, capecitabine, which generates 5-fluorouracil selectively in tumours by enzymes concentrated in human liver and cancer tissue. Eur. J. Cancer 34, 1274–1281. Miyadera, K., Sumizawa, T., Haraguchi, M., Yoshida, H., Konstanty, W., Yamada, Y., and Akiyama, S. (1995). Role of thymidine phosphorylase activity in the angiogenic effect of platelet derived endothelial cell growth factor/thymidine phosphorylase. Cancer Res. 55, 1687– 1690. Miyazono, K., Okabe, T., Urabe, A., Takaku, F., and Heldin, C.H. (1987). Purification and properties of an endothelial cell growth factor from human platelets. J. Biol. Chem. 262, 4098–4103. Mizutani, Y., Okada, Y., and Yoshida, O. (1997). Expression of platelet-derived endothelial cell growth factor in bladder carcinoma. Cancer 79, 1190–1194. Moghaddam, A., and Bicknell, R. (1992). Expression of plateletderived endothelial cell growth factor in Escherichia coli and confirmation of its thymidine phosphorylase activity. Biochemistry 31, 12141–12146. Moghaddam, A., Zhang, H.T., Fan, T.P., Hu, D.E., Lees, V.C., Turley, H., Fox, S.B., Gatter, K.C., Harris, A.L., and Bicknell, R. (1995). Thymidine phosphorylase is angiogenic and promotes tumor growth. Proc. Natl. Acad. Sci. USA 92, 998–1002. O’Brien, T., Cranston, D., Fuggle, S., Bicknell, R., and Harris, A.L. (1995). Different angiogenic pathways characterize superficial and invasive bladder cancer. Cancer Res. 55, 510–513. O’Brien, T.S., Fox, S.B., Dickinson, A.J., Turley, H., Westwood, M., Moghaddam, A., Gatter, K.C., Bicknell, R., and Harris, A.L. (1996). Expression of the angiogenic factor thymidine phosphorylase/platelet-derived endothelial cell growth factor in primary bladder cancers. Cancer Res. 56, 4799–4804. Patterson, A.V., Zhang, H., Moghaddam, A., Bicknell, R., Talbot, D.C., Stratford, I.J., and Harris, A.L. (1995). Increased sensitivity to the prodrug 5⬘-deoxy-5-fluorouridine and modulation of 5-fluoro2⬘-deoxyuridine sensitivity in MCF-7 cells transfected with thymidine phosphorylase. Br. J. Cancer 72, 669–675. Pauly, J.L., Schuller, M.G., Zelcer, A.A., Kirss, T.A., Gore, S.S., and Germain, M.J. (1977). Identification and comparative analysis of thymidine phosphorylase in the plasma of healthy subjects and cancer patients. J. Natl. Cancer Inst. 58, 1587–1590.
Reynolds, K., Farzaneh, F., Collins, W.P., Campbell, S., Bourne, T.H., Lawton, F., Moghaddam, A., Harris, A.L., and Bicknell, R. (1994). Association of ovarian malignancy with expression of plateletderived endothelial cell growth factor. J. Natl. Cancer Inst. 86, 1234– 1238. Schwartz, M. (1971). Thymidine phosphorylase from Escherichia coli. Properties and kinetics. Eur. J. Biochem. 21, 191–198. Schwartz, M. (1978). Thymidine phosphorylase from Escherichia coli. Methods Enzymol. 51, 442–445. Spraggon, M., Stuart, D., Ponting, C., Finnis, C., Sleep, D., and Jones, Y. (1993). Crystallization and x-ray diffraction study of recombinant platelet-derived endothelial cell growth factor. J. Mol. Biol. 234, 879–880. Sumizawa, T., Furukawa, T., Haraguchi, M., Yoshimura, A., Takeyasu, A., Ishizawa, M., Yamada, Y., and Akiyama, S. (1993). Thymidine phosphorylase activity associated with platelet-derived endothelial cell growth factor. J. Biochem. (Tokyo) 114, 9–14. Takebayashi, Y., Akiyama, S., Akiba, S., Yamada, K., Miyadera, K., Sumizawa, T., Yamada, Y., Murata, F., and Aikou, T. (1996a). Clinicopathologic and prognostic significance of an angiogenic factor, thymidine phosphorylase, in human colorectal carcinoma. J. Natl. Cancer Inst. 88, 1110–1117. Takebayashi, Y., Miyadera, K., Akiyama, S., Hokita, S., Yamada, K., Akiba, S., Yamada, Y., Sumizawa, T., and Aikou, T. (1996b). Expression of thymidine phosphorylase in human gastric carcinoma. Jpn. J. Cancer Res. 87, 288–295. Takeuchi, M., Otsuka, T., Matsui, N., Asai, K., Hirano, T., Moriyama, A., Isobe, I., Eksioglu, Y.Z., Matsukawa, K., and Kato, T. (1994). Aberrant production of gliostatin/platelet-derived endothelial cell growth factor in rheumatoid synovium. Arthritis Rheum. 37, 662–672. Thomas, M.B., Hoff, P.M., Carter, S., Bland, G., Lassere, Y., Wolff, R., Xiong, H., Hayakawa, T., and Abbruzzese, J. (2002). A dosefinding, safety and pharmacokinetics study of TAS-102, an antitumor/antiangiogenic agent given orally on a once-daily schedule for five-days every three weeks in patients with solid tumors. Proc. Am. Assoc. Cancer Res. 43, 554. Walter, M.R., Cook, W.J., Cole, L.B., Short, S.A., Koszalka, G.W., Krenitsky, T.A., and Ealick, S.E. (1990). Three-dimensional structure of thymidine phosphorylase from Escherichia coli at 2.8A˚ resolution. J. Biol. Chem. 265, 14016–14022. Zimmerman, M., and Seidenberg, J. (1964a). Deoxyribosyl transfer. I. Thymidine phosphorylase and nucleoside deoxyribosyltransferase in normal and malignant tissues. J. Biol. Chem. 239, 2618–2621. Zimmerman, M., and Seidenberg, J. (1964b). Deoxyribosyl transfer. II. Nucleoside: pyrimidine deoxyribosyltransferase activity of three partially purified thymidine phosphorylases. J. Biol. Chem. 239, 2622–2627. Accession Numbers The atomic coordinates and structure factors have been deposited in the Protein Data Bank (accession code 1UOU).
DOI 10.1016/j.str.2003.11.018
Crystal Structure of Human Thymidine Phosphorylase in Complex with a Small Molecule Inhibitor Richard A. Norman, Simon T. Barry, Michael Bate, Jason Breed, Jeremy G. Colls, Richard J. Ernill, Richard W.A. Luke, Claire A. Minshull, Mark S.B. McAlister, Eileen J. McCall, Helen H.J. McMiken, Dougie S. Paterson, David Timms, Julie A. Tucker, and Richard A. Pauptit* AstraZeneca Alderley Park Macclesfield, Cheshire SK10 4TG United Kingdom
Summary Human thymidine phosphorylase (HTP), also known as platelet-derived endothelial cell growth factor (PDECGF), is overexpressed in certain solid tumors where it is linked to poor prognosis. HTP expression is utilized for certain chemotherapeutic strategies and is also thought to play a role in tumor angiogenesis. We determined the structure of HTP bound to the small molecule inhibitor 5-chloro-6-[1-(2-iminopyrrolidinyl) methyl] uracil hydrochloride (TPI). The inhibitor appears to mimic the substrate transition state, which may help explain the potency of this inhibitor and the catalytic mechanism of pyrimidine nucleotide phosphorylases (PYNPs). Further, we have confirmed the validity of the HTP structure as a template for structure-based drug design by predicting binding affinities for TPI and other known HTP inhibitors using in silico docking techniques. This work provides the first structural insight into the binding mode of any inhibitor to this important drug target and forms the basis for designing novel inhibitors for use in anticancer therapy. Introduction HTP (EC 2.4.2.4) has a primary catabolic role in pyrimidine nucleoside metabolism, catalyzing the reversible phosphorolysis of thymidine, deoxyuridine, and their analogs (except deoxycytidine) to their respective bases and 2-deoxyribose-1-phosphate (2-dR-1-P) (Krenitsky et al., 1981). HTP is a member of the PYNP family which in higher species consists of two PYNPs, TP and uridine phosphorylase (UP), and in lower organisms consists of only one PYNP with dual (TP and UP) activity (Krenitsky et al., 1965; Pugmire and Ealick, 2002). Kinetic studies using E. coli TP (EcTP) and rabbit muscle TP have shown that the enzyme follows a sequential mechanism whereby phosphate is the first substrate to bind and 2-dR-1-P is the last product to leave (Krenitsky, 1968; Schwartz, 1971). These results are supported by evidence that the enzyme is stabilized by phosphate and 2-dR-1-P, but not by thymidine or thymine (Krenitsky and Tuttle, 1982). Structural studies on B. stearothermophilus PYNP (BsPYNP), bound to pseudouridine and phosphate, have *Correspondence: [email protected]
led Pugmire and Ealick to propose a catalytic mechanism for this family of enzymes in which the pyrimidine nucleoside is bound in a high-energy conformation causing the weakening of the glycosidic bond. The glycosidic bond is weakened further by flow of electrons from O-5⬘ of the ribose moiety to the pyrimidine ring, while the phosphate ion is positioned to attack the C-1⬘ position of the ribose ring to yield 2-dR-1-P and the free pyrimidine base (Pugmire and Ealick, 1998, 2002) (Figure 1). HTP is able to transfer the deoxyribosyl moiety from one pyrimidine base to another thus playing an additional vital role in base/nucleotide salvage (Zimmerman and Seidenberg, 1964a, 1964b). HTP is now understood to be identical to PD-ECGF, which is an angiogenic activity identified in platelet lysates (Miyazono et al., 1987), and to gliostatin which has important physiological roles in neuronal and glial development (Asai et al., 1992). The activity of HTP correlates with hypoxia and the induction of angiogenesis in a number of malignant and hyperproliferative disease states. HTP overexpression and elevated enzyme activity have been found in the plasma of cancer patients (Pauly et al., 1977), and HTP expression has been linked with the invasive/metastatic capacity of gastric (Maeda et al., 1996; Takebayashi et al., 1996b), bladder (Mizutani et al., 1997; O’Brien et al., 1995, 1996), and colorectal (Takebayashi et al., 1996a) tumors. HTP activity is also elevated in ovarian (Reynolds et al., 1994), oesophageal, and pancreatic cancers (Takebayashi et al., 1996a), and has been shown to be associated with a more aggressive phenotype in nonsmall cell lung cancer node-negative disease (Fox et al., 1996). HTP overexpression has been implicated in other angioproliferative disorders, including rheumatoid arthritis (Takeuchi et al., 1994), psoriasis (Creamer et al., 1997), and hyperproliferation of the vasa vasorum in artherosclerosis (Ignatescu et al., 1999). The in vivo angiogenic activity associated with HTP is thought to be mediated by dephosphorylated 2-deoxyribose and although the mechanism is unclear it does require the enzyme activity of HTP (Miyadera et al., 1995; Moghaddam et al., 1995; Sumizawa et al., 1993). The activity of HTP inactivates certain chemotherapeutic agents such as TFT (trifluorothymidine) (Fukushima et al., 2000) and can also be used to catalyze the conversion of prodrugs such as tegafur (1-(tetrahydro2-furanyl)-5-fluorouracil) and capecitabine (N4-pentyloxycarbonyl-5⬘-deoxy-5-fluorocytidine) to anticancer nucleotides which interfere with DNA replication (Patterson et al., 1995; Miwa et al., 1998). Whereas the latter type of therapy would be incompatible with HTP inhibition, inhibitors of HTP may be used to provide alternative agents to control angiogenesis and in combination with TFT therapy. These alternative approaches would prove especially attractive in cases where primary fluoropyrimidine therapy has failed. Recently, a number of HTP inhibitors have been reported, the most potent of which is TPI (5-chloro-6-[1(2-iminopyrrolidinyl) methyl] uracil hydrochloride) with Ki ⫽ 20 nM. TPI causes a substantial reduction in tumor
Structure 76
Figure 1. Reaction Catalyzed by HTP
growth when given continuously to mice carrying HTP overexpressing tumors by increasing the proportion of apoptotic cells and possibly also inhibiting angiogenesis (Matsushita et al., 1999). The combination of TFT and TPI in a 1:0.5 molar ratio (TAS-102) is currently tested in phase I trials and can be administered orally (Thomas et al., 2002). Less potent HTP inhibitors have been designed using models of HTP based on the structure of EcTP or BsPYNP (Cole et al., 2003; Price et al., 2003). Both bacterial enzymes (ⵑ90 kDa) possess around 40% sequence identity with HTP and, like the larger mammalian TPs (ⵑ110 kDa), are homodimers (Desgranges et al., 1981; Kraut and Yamada, 1971; Pugmire and Ealick, 1998; Schwartz, 1978). Each subunit of the bacterial enzymes consists of a small ␣-helical domain (␣ domain) and a large mixed ␣-helical and  sheet domain (␣/ domain) separated by a large cleft containing the substrate binding site (Pugmire and Ealick, 1998; Walter et al., 1990). Movement of the two domains allows for an open (empty) or closed (substrate-bound) conformation of the enzyme (Pugmire and Ealick, 1998; Pugmire et al., 1998). Here, we report the crystal structure of HTP in complex with TPI in an active (closed) conformation. Since this is the first structure of any PYNP family member bound to an inhibitor, it will provide the basis for further structure-based drug design studies with the aim of increasing current inhibitor potency and specificity, or developing new anticancer drug candidates. Results and Discussion HTP Proteolysis and Structure Determination Crystals of recombinant HTP which diffract to 3.5 A˚ have been reported (Spraggon et al., 1993). In our hands, however, crystallization trials with HTP (residues 12– 482) have for many years failed to produce diffractiongrade crystals. Sequence alignments (data not shown) of HTP with EcTP and BsPYNP show a glycine-rich N-terminal extension of 32 residues in HTP which is assumed to lack defined secondary structure. Since disordered N- or C termini of proteins could conceivably lead to crystallization difficulties, limited proteolysis was performed with a range of proteases in an attempt to define a minimal domain more amenable to successful crystallization. This yielded a product from which residues 409 and 410 had been cleaved, as shown by Edman sequencing and electrospray mass spectrometry, and
which crystallized. Whereas we had initially assumed we had removed a portion of the C-terminal domain, remote from the active site, analysis of the crystals by SDS-PAGE showed the crystallized product included two fragments: 12–408 and 411–482. Functional assays carried out on the proteolyzed HTP show that it possesses the same degree of activity as full-length HTP toward its substrate (data not shown). Crystals of HTP belong to the space group C2 and the asymmetric unit contains a single HTP protomer. The three-dimensional structure of HTP was determined by the molecular replacement method, using the crystal structure of BsPYNP as the search model (see Experimental Procedures). The structure was refined to an R factor of 18.9% and an Rfree of 25.8% (Table 1). The final Fo-Fc electron density for the bound TPI is shown in Figure 2. Recombinant HTP spans residues 12–482; however, the proteolytic treatment removed residues 409 and 410 causing the remainder of the loop (residues 407–408 and 411–414) to become disordered. No electron density was observed for residues 12–32 (N terminus), 238–239, and 481–482 (C terminus). The structure of HTP contains one molecule of TPI in the active site. Structure of HTP and Comparison to Other PYNPs The HTP protomer (Figure 3A) has a similar fold to that reported for EcTP and BsPYNP (Pugmire and Ealick, 1998; Walter et al., 1990) and comprises an ␣ domain and a mixed ␣/ domain connected by three polypeptide loops (Figure 3B). The ␣ domain consists of six ␣ helices (␣1–␣4 and ␣8–␣9), and the ␣/ domain consists of a central mixed  sheet (1–5, 13) surrounded by ␣ helices (␣5–␣7 and ␣10–␣16) and two small antiparallel  sheets (consisting of strands 6, 7, 9, and 11, and 8, 10, and 12 respectively) flanked by two ␣ helices (␣17 and ␣18). The three loops act as a hinge allowing the two domains to move between the open (inactive) and closed (active) conformations, bringing together the active site residues (Pugmire and Ealick, 1998; Pugmire et al., 1998). The main structural differences between HTP, EcTP, and BsPYNP are in the ␣/ domain where HTP has an additional helix (␣15), an extra turn in ␣16, and a more extended C-terminal region. The small antiparallel  sheet (8, 10, and 12) is present in HTP and BsPYNP but not in EcTP. An HTP dimer is generated from the single protomer that constitutes the asymmetric unit by rotation about the crystallographic 2-fold axis (Figure 3C). The core of
Structure of Human Thymidine Phosphorylase 77
Table 1. Data Collection, Refinement Statistics, and Model Quality TP-TPI Complex Resolution range (A˚) Spacegroup Z Cell dimensions a b c  Wavelength (A˚) Observed reflections Unique reflections Highest resolution shell (A˚) Redundancy I/(I) Completeness (%) Rmerge (%) Rwork (%) Rfreeb (%) Rms bond lengths (A˚) Rms bond angles (⬚) Number of nonhydrogen atoms Protein Inhibitor Water Mean temperature factors (A˚2) Protein Inhibitor Ramachandran plotc (%) Residues in most favored regions Residues in additional allowed regions Residues in disallowed regions
91.3–2.1 C2 4 70.0 A˚ 66.1 A˚ 96.9 A˚ 106.5⬚ 0.9780 32,998 15,827 2.21-2.11 2.1 (1.3)a 15.5 (1.8) 63.7 (11.2) 3.0 (19.3) 18.8 (34.8) 25.8 (57.7) 0.015 1.5 3,214 16 75 23.9 19.9 94.5 5.5 0
a
Values in parentheses correspond to the high resolution shell. Rfree was calculated against 5.4% of the complete dataset excluded from refinement. c Statistics obtained from PROCHECK (Laskowski et al., 1993). b
the dimer interface is analogous to that of the BsPYNP structure and consists of a coiled coil formed by helices ␣3 from each monomer. Also, helix ␣1 of one monomer packs against the loop connecting helices ␣8 and ␣9 of the second monomer. In the BsPYNP structure, residues 364–371, which correspond to the clipped loop of HTP, extend across the active site cleft and appear to act as a
latch, holding the ␣ and ␣/ domains together. Residues A366, K367, and K368 (in BsPYNP) from the tip of this loop make contacts with the opposing molecule in the dimer. It has been proposed that dimer interactions around this loop have a functional role in closing the active site cleft (Pugmire and Ealick, 1998). Residues S409 and R410 (corresponding to A366 and K367 in BsPYNP) are not present in the HTP that we crystallized and consequently this loop is disordered in our structure. However, HTP still adopts a closed conformation (see below) and the species crystallized has been shown to be fully active. Thus, for HTP, this region is not essential for the formation of an active enzyme. Structural studies on BsPYNP (Pugmire and Ealick, 1998) suggest that binding of phosphate results in the formation of a key hydrogen bond between residues H116 and G205 (corresponding to H150 and A239 of HTP), causing an 8⬚ rotation of the ␣/ domain with respect to the ␣ domain compared to the structure in the absence of phosphate. Subsequent binding of the pyrimidine base would be responsible for an additional 20⬚ rotation to form the fully closed active site (Pugmire and Ealick, 1998, 2002). Phosphate is absent from the catalytic site in our structure, A239 is disordered, and therefore no hydrogen bond is observed between H150 and A239. Our HTP structure appears to be in the active (closed) conformation. Alignment of the ␣ domain of HTP with the corresponding ␣ domains of EcTP and BsPYNP gave rms deviations of 0.82 and 0.67 A˚, respectively, for 100 equivalent C␣ atoms. Alignment of the HTP ␣/ domain with the corresponding ␣/ domains of EcTP and BsPYNP gave rms deviations of 1.94 and 1.46 A˚, respectively, over 282 C␣ atoms. However, an alignment of the entire HTP protomer with the EcTP and BsPYNP structures gave rms deviations of 3.55 and 1.71 A˚, respectively, over 383 C␣ atoms. While the individual domains superpose well, the overall structural similarity between HTP and EcTP is less than that for HTP and BsPYNP. By analogy to BsPYNP, HTP is in the active (closed) conformation (Figure 4). Thus, the presence of the inhibitor TPI alone, without phosphate, is sufficient to cause closure of the ␣ and ␣/ domains to form the active conformation. The structure of BsPYNP contains a metal binding site
Figure 2. Electron Density Corresponding to TPI Stereo view of the final 2.1 A˚ Fo-Fc omit map to within 2 A˚ of the inhibitor molecule contoured at 3.0 (cyan). Nitrogen atoms are blue; oxygen atoms, red; chlorine atom, lime green; carbon atoms and bonds, light purple. All molecular graphics were produced using Bobscript (Esnouf, 1999) and Raster3D (Merrit and Murphy, 2003) except Figure 7, which was produced using PyMOL (DeLano, 2002).
Structure 78
Structure of Human Thymidine Phosphorylase 79
Figure 4. Fold Comparison of HTP Comparison of the fold of HTP with EcTP and BsPYNP. The three molecules are shown in the same orientation, and colored red to blue (N terminus to C terminus). Bound pseudouridine (light purple) and phosphate (red) in the BsPYNP structure and TPI (light purple) in the HTP structure are shown as cpk models. Arrows indicate domain movement giving the active conformation of the enzyme.
adjacent to the phosphate binding site. This unidentified penta-coordinate metal interacts with the backbone carbonyl oxygen atoms of residues L243, A246, and G88, and the side chains of E255 and T90 (BsPYNP). In the HTP structure, there is no metal ion present, but the N atom of K124 (which corresponds to T90 of BsPYNP) adopts a position similar to the BsPYNP metal ion and interacts with the equivalent residues, possibly fulfilling the same structural role. Inhibitor Binding, Specificity and Insights into the Catalytic Mechanism of HTP The HTP active site may be described as having two components, each interacting with one part of the TPI molecule. The first component of the active site is formed by residues from the ␣ domain and H116 from the ␣/ domain, which interact with the 5-chlorouracil moiety. The second component of the active site, which interacts with the 2-iminopyrrolidinyl moiety, is formed by S117 from the ␣/ domain and a network of water molecules which occupy the phosphate binding site (Figure 5A). The 5-chlorouracil of TPI makes strong hydrogenbonding interactions with R202, S217, K221, and H116 (Figures 5A and 5B). These residues are conserved in the BsPYNP structure (R168, S183, K187, and H82 respectively) and, with the exception of H82 (H116 in HTP), form a similar hydrogen-bonding network with the uracil moiety of the substrate analog pseudouridine (Figure 5C). The number of hydrogen bonds between BsPYNP and the uracil ring is lower than between HTP and TPI, and the equivalent distances are longer, consistent with the tighter binding of TPI to HTP. Another contribution to the tight interactions between HTP and the 5-chlorouracil of TPI is the chlorine atom, which is held in a hydrophobic pocket formed by V208 and I214 of the ␣
domain and L148 and V241 of the ␣/ domain, and forces the 2-oxo, 3-amido, and 4-oxo groups of the 5-chlorouracil moiety closer to their interacting partners relative to uracil. The C␥2 of V241 lies 4.4 A˚ from the chlorine, indicative of good van der Waals interactions between the two atoms. By analogy, the methyl group of thymine would occupy the same position as the chlorine and would make similar interactions with V241. Uracil, however, lacks a substituent at this position making the distance between the 5 position and the C␥2 of V241 approximately 6.0 A˚, creating a hydrophobic cavity. Interestingly, of the four residues which form the hydrophobic pocket in HTP, only V241 is not conserved in BsPYNP: it corresponds to F207. In the bacterial structure, the distance between the 5 position of uracil and the C⑀2 of F207 is 3.9 A˚, consistent with good van der Waals interactions between these atoms. Binding of thymine would result in steric clashes unless the phenyl ring of F207 were rotated by 90⬚ around the C-C␥ bond to accommodate the thymine methyl group at the 5 position. Such rotation of the phenyl ring around the C-C␥ bond may allow BsPYNP to accept both uracil and thymine as substrates, without the need for extensive structural rearrangement of the active site. This suggests the F207 side chain in BsPYNP may be less discriminating than V241 in HTP. Whereas the basis for specificity has been ascribed to the high affinity of HTP for the 2⬘-deoxyribose moiety of the thymidine substrate (Pugmire and Ealick, 2002), there may also be a contribution from V241. Unfortunately, the structure of HTP complexed with TPI offers no direct insights as to the basis of specific interactions between HTP and the thymidine 2⬘-deoxyribose moiety. The 2-iminopyrrolidinyl moiety of TPI makes two hydrogen bonds, one to the carbonyl oxygen of S117 and one to a water molecule (Figures 5A and 5B). The posi-
Figure 3. Structure of HTP (A) Stereo view of the HTP monomer showing the main-chain trace of the ␣ domain (cyan), the hinge regions (green), and the mixed ␣/ domain (brown). TPI (light purple) is shown as a cpk model in the active site situated between the ␣ and ␣/ domains. (B) Topology diagram of HTP color coded in the same manner as in (A). (C) Stereo view of the HTP dimer generated by applying 2-fold crystallographic symmetry to the monomer. HTP monomers are colored cyan and brown and TPI (light purple) is shown as a cpk model. The position of the proteolytically cleaved loop is indicated by an arrow.
Structure of Human Thymidine Phosphorylase 81
tively charged character of the 2-iminopyrrolidine may mimic the oxocarbenium ion-like transition state proposed for thymidine phosphorolysis (Figure 1). This suggests an orientation for the ribose moiety of thymidine which is rotated by approximately 180⬚ with respect to previous models (Pugmire and Ealick, 2002). Water molecules W1 and W2 in HTP (Figures 5A and 5B) mimic the phosphate oxygen atoms in the BsPYNP structure, suggesting that these oxygen positions could be relevant and the orientation of the phosphate during catalysis could be similar to that observed in the BsPYNP crystal structure (Figure 5A). The role of H116, which is absolutely conserved in the PYNP family, has been the subject of some debate. It is believed to be involved in either stabilizing the transition state during catalysis or in donating a proton to the N-1 of the pyrimidine ring following glycosidic cleavage (Figure 1). Given that the pKa of TPI has been measured at 6.05 (our unpublished data), it is likely that at the pH of crystallization (pH 6.5) the 1-amido group of 5-chlorouracil is anionic and makes a hydrogen bond with the protonated N⑀2 of H116, and N␦2 of H116 serves as the hydrogen-bond acceptor for N of K222. The side chains of both D114 and E225 are positioned within hydrogen-bonding distance of each other and of K222 in an unusual arrangement (Figure 5A), suggesting this triad forms a protonshuttling mechanism whose role is to deliver a proton to H116. Catalysis would therefore proceed as previously described followed by protonation of the pyrimidine ring at the 1-amido position by H116 via the proton shuttle. Thus, the structure provides evidence for the role of H116 as the proton donor during thymidine phosphorolysis. In addition, the resemblance of TPI to the zwitterionic transition state during thymidine phosphorolysis may help explain its high potency. Implications for Drug Design In order to establish whether the HTP structure could provide the basis for in silico-based drug design, calculations were carried out in an attempt to predict binding affinities for TPI and other known HTP inhibitors. Docking TPI into the HTP active site using GLIDE (Eldridge et al., 1997) yielded a bound position for the inhibitor in substantial agreement with experiment (rms deviation ⫽ 0.42 A˚ for the nonhydrogen atoms) (Figure 6A). The ionic status of TPI on binding to the enzyme was investigated using a binding energy calculation based on PoissonBoltzmann electrostatics (Grant et al., 2001). Calculations for the TPI binding mode found by experiment indicate that the inhibitor binds as the zwitterion (Table 2). Three further ligands (Figure 6B) exhibiting a diverse
Figure 6. Structures of HTP Inhibitors Used in Docking Studies (A) View of the HTP active site color coded and labeled in the same manner as in Figure 5A showing the superposed docked TPI molecule (yellow). (B) TPI. (C) Compound II, 6-amino-5-bromouracil (Langen et al., 1967). (D) Compound III, 5-chloro-6-[(2-nitroimidazol1-yl) methyl] uracil (Cole et al., 2003). (E) Compound IV, N-(2,4-dioxo1,2,3,4-tertahydrothieno[3,2-d] pyrimidin-7-yl) guanidine (Price et al., 2003).
range of chemical features and affinities were also docked using GLIDE and their binding energies estimated using the same electrostatic protocol as for TPI. For these ligands, experimental values of pKa were unavailable and these were estimated using ACD software (Advanced Chemistry Development, http://www.acdlabs. com), thus allowing corrected estimates of the binding constants (Kcorr) to be made. In the case of ligands (II) and
Figure 5. Active Site (A) Stereo view of the HTP active site highlighting important interactions. No phosphate is present in HTP, but the position of the phosphate in the BsPYNP crystal structure is indicated. Nitrogen atoms are blue; oxygen atoms, red; chlorine atom, lime green; phosphate atom, lavender; carbon atoms and bonds, light purple (inhibitor) and white (protein). Residues which form direct interactions with the inhibitor have their labels colored light purple and residues which form the hydrophobic pocket around the chlorine atom have their labels colored green. Other residues and selected water molecules are labeled in black. (B) Schematic representation of the HTP active site highlighting important interactions. The ring positions of the 5-chlorouracil moiety are numbered and hydrogen-bonding distances (A˚) are labeled in blue. (C) Schematic representation of the BsPYNP active site highlighting important interactions. The ring positions of the uracil moiety are numbered and hydrogen-bonding distances (A˚) are labeled in blue.
Structure 82
Table 2. Docking and Scoring Results for Some HTP Ligands Compound TPI cation TPI zwitterion II neutral II anion III neutral III anion IV cation IV zwitterion
pKaa 6.1 (6.9) (7.1) (7.3) (9.5)
Kcalc
Kcorrb
Inactived 345 nM 74 mM 0.375 M Inactived 6.0 M Inactived 1.85 M
Inactived 360 nM 111 mM 0.6 M Inactived 10.8 M Inactived 235 M
IC50c
of phosphate. The HTP structure provides a framework including many replaceable water molecules, which offers a design opportunity for further developing existing inhibitors or to design novel inhibitors (Figure 7).
5–20 nM 1.6 M 21.7 M 76 M
a
Values in parentheses were calculated using ACD software (Advanced Chemistry Development, http://www.acdlabs.com). Kcorr is the corrected value of Kcalc to account for the fraction of the binding form of the compound that would be present in the experimental assay solution. c Experimental IC50 data were taken from (Cole et al., 2003) (TPI, compound II, and compound III) and from (Price et al., 2003) (TPI and compound IV). d As interpreted from positive ⌬G. b
(III), the anions would appear to be the active entities. For ligand (IV), as for TPI, it would appear the zwitterion is bound and the poorer affinity relates to the higher pKa estimated as a consequence of the annelated thiophene ring. The common pyrimidine-dione substructure docks in a similar manner for all the ligands and the estimated affinities are correctly ranked in comparison to the experimental IC50s. The estimated TPI binding affinity is poorer than expected and we have examined the effect of introducing phosphate into the binding site. This gives rise to an enhanced calculated affinity for TPI (0.5 nM) and IV (200 nM) but is deleterious to the binding of the anions II (14 mM) and III (4 mM), probably due to the electrostatic repulsion between like charges. We know from the structure that phosphate binding, although mandatory for the enzyme mechanism, is not a necessary requirement for inhibitor binding. These studies indicate that the HTP structure can be used to predict docking modes and relative binding affinities for candidate inhibitors of HTP. In addition, the probable ionic status of the bound form of the ligand can be suggested. Structure-based design studies to date have been impeded by the lack of an HTP structure and the limitations of homology models based on bacterial structures. An active conformation of the enzyme is needed for these studies and the structure of HTP bound to an inhibitor shows that this can be achieved in the absence
Experimental Procedures Sample Preparation and Crystallization HTP was expressed in E. coli from plasmid pMOAL-10T (Moghaddam and Bicknell, 1992). The plasmid encodes a product consisting of glutathione S-transferase (GST) fused to HTP residues 12–482 (based on Swissprot accession number P19971) via a thrombincleavable linker. Cells were induced overnight at 15⬚C with 0.1 mM IPTG to maximize soluble accumulation, harvested, and lysed in buffer A (20 mM Tris/HCl, pH 7.4, 100 mM NaCl, 5 mM DTT) by highpressure homogenization. Cell debris were removed by centrifugation and the lysate supernatant loaded onto a glutathione-sepharose column equilibrated with buffer A. The column was washed with at least 10 vol of buffer A prior to eluting the product with buffer A containing 20 mM reduced glutathione. The product was incubated for 4 hr at 37⬚C with thrombin (10 U/mg fusion protein) and separated from the cleaved GST by size-exclusion chromatography on a Superdex 75 column equilibrated with buffer B (20 mM Tris/HCl, pH 7.5, 150 mM NaCl, 1 mM DTT). HTP was subjected to limited proteolysis by trypsin (1:100 w/w; 40 min at 37⬚C) followed by addition of 1 mM phenylmethyl sulphonyl fluoride. Proteolyzed HTP was loaded onto a Superdex 75 column equilibrated in buffer C (20 mM Tris/HCl, pH 7.4, 50 mM NaCl) and concentrated to 8 mg ml⫺1. Prior to crystallization, 5 mM TPI (Taiho Pharmaceutical Co. Ltd., Japan) was added to the protein and incubated at 4⬚C for 30 min. The HTPTPI complex was crystallized at 15⬚C using the hanging-drop vapor diffusion method by mixing equal volumes of complex solution and precipitant (8% PEG 1000 [w/v], 50 mM malate/imidazole, pH 6.5, 100 mM MgCl2). Crystals grew in some wells only and reached their full size after approximately 3 weeks. Crystals were harvested into cryoprotectant solution (9% PEG 1000, 50 mM malate/imidazole, pH 6.5, 100 mM MgCl2, 20% [v/v] glycerol) and flash cooled in a nitrogen stream at 100 K. Structure Determination and Refinement HTP diffraction data were collected from a single crystal at 100 K on an ADSC CCD detector on beamline 14-2 at the SRS (Daresbury, UK) and processed with MOSFLM and SCALA (CCP4, 1994; Leslie, 1992). Diffraction data extends to 2.1 A˚, but the data are essentially complete only to 2.7 A˚. Statistics for the data collection are given in Table 1. The structure was solved by the molecular replacement method using a modification (nonconserved residues truncated to alanine or glycine; ligand and water molecules excluded) of the crystal structure of BsPYNP (PDB code 1brw, chain B) as a search model. Rotation and translation searches were carried out using the program AMORE (CCP4, 1994). The statistically most promising solution was eventually proven correct, though it was with considerable difficulty that sufficient clarity and connectivity could be obtained in the first derived electron density maps to allow rebuilding
Figure 7. Inhibitor Design Orthogonal views of the HTP active site with bound TPI (light purple). Van der Waals radii for bound water molecules are shown (red), highlighting the available space exploitable in drug design.
Structure of Human Thymidine Phosphorylase 83
and refinement to be initiated. The connectivity was more apparent when in the first instance the resolution was limited to 2.6 A˚, where the completeness of the data was 90%. This gave readily interpretable density for the secondary structure elements. Automated model building using MAID (Levitt, 2001) proved a useful tool to reposition the secondary structure elements slightly from their positions in the trial model. Continued rounds of simulated annealing, energy minimization and temperature factor refinement using CNX (Accelrys, www.accelrys.com) extending the resolution to the limit of 2.1 A˚ eventually resulted in the connectivity between the secondary structure elements becoming very clear in the electron density, allowing the building of a complete model. Final refinement using CNX and Refmac5 (CCP4, 1994) lead to an R factor of 18.9% and an Rfree of 25.8%. Statistics for the refinement and quality of the model are given in Table 1.
References
Docking and Scoring Studies HTP was prepared for docking and scoring studies as follows: WHATCHECK (Hooft et al., 1996) analysis suggested that two histidines (H116 and H441) exist as the N⑀2(H) tautomer and one is protonated (H319) with the remainder as the N␦1(H) tautomer. For residues modeled with dual conformations, the “A” locations for R329 and M273 were chosen. Of the two juxtaposed carboxylates, D114 and E225, the latter was protonated. CHARMM (Mackerell et al., 1998) was used to add the hydrogen atoms and their positions were optimized while keeping the heavy atoms fixed. Crystallographic water molecules were then removed to allow free docking of the ligands. For the phosphate studies, dihydrogen phosphate was added in the position found in the BsPYNP structure. Ligand dockings were performed using GLIDE 2.5 (Eldridge et al., 1997) using the default settings, in extended sampling mode. Ligands were treated as being flexible and the van der Waals radii of the ligand nonpolar atoms were scaled by a factor of 0.8. Binding affinities were calculated using the following equation: ⌬Gcalc ⫽ PB ⫹ ␥⌬ASA ⫹ ␣NR ⫹ RBE, where PB is the change in coulombic interaction energy and electrostatic desolvation energy in kcal/mol on complexation as determined by the Poisson-Boltzmann treatment (Grant et al., 2001). Atomic charges for both the protein and the ligand were generated using the MMFF force field (Halgren and Halgren, 1996). ⌬ASA is the change in accessible area on complexation and is converted to a nonpolar (hydrophobic) contribution by means of the ␥ coefficient (0.025 kcal/mol/A2). NR is the number of rotatable bonds and is converted to a measure of the loss of conformational entropy on binding by the coefficient ␣ (0.33 kcal/mol/bond). RBE represents the loss in rigid body entropy on complex formation and is assigned a fixed value of 4.71 kcal/mol. The values for Kcalc were generated using ⌬Gcalc ⫽ ⫺RTln(Kcalc).
Creamer, D., Jaggar, R., Allen, M., Bicknell, R., and Barker, J. (1997). Overexpression of the angiogenic factor platelet-derived endothelial cell growth factor/thymidine phosphorylase in psoriatic epidermis. Br. J. Dermatol. 137, 851–855.
Estimates of compound pKa’s were made using ACD software (Advanced Chemistry Development, http://www.acdlabs.com); these pKa’s were used to correct the calculated binding affinities in accordance with the fraction of active species present at the assay pH of 7.4 using the Henderson-Hasselbach equation (Albert and Serjeant, 1962).
Acknowledgments We would like to thank Roy Bicknell (Angiogenesis Group, Molecular Oncology Laboratory, Weatherall Institute of Molecular Medicine, Oxford, UK) for the gift of plasmid pMOAL-10T and Delphine Dorison-Duval for determining the pKa value of TPI. We would also like to thank Jon Read and Antonio Pineda-Lucena and in particular Philip Edwards (Manchester University) for helpful discussions. We gratefully acknowledge the staff, in particular James Nicholson, and use of facilities at the SRS, Daresbury, UK.
Received: August 27, 2003 Revised: September 24, 2003 Accepted: October 1, 2003 Published: January 13, 2004
Albert, A., and Serjeant, E.P. (1962). Ionization Constants of Acids and Bases. A Laboratory Manual. (London: Methuen and Co. Ltd.). Asai, K., Nakanishi, K., Isobe, I., Eksioglu, Y.Z., Hirano, A., Hama, K., Miyamoto, T., and Kato, T. (1992). Neurotrophic action of gliostatin on cortical neurons. Identity of gliostatin and platelet-derived endothelial cell growth factor. J. Biol. Chem. 267, 20311–20316. CCP4 (Collaborative Computational Project 4) (1994). The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D 50, 760–763. Cole, C., Reigan, P., Gbaj, A., Edwards, P.N., Douglas, K.T., Stratford, I.J., Freeman, S., and Jaffar, M. (2003). Potential tumor-selective nitroimidazolylmethyluracil prodrug derivatives: inhibitors of the angiogenic enzyme thymidine phosphorylase. J. Med. Chem. 46, 207–209.
DeLano, W.L. (2002). The PyMOL Molecular Graphics System (San Carlos, CA: DeLano Scientific). Desgranges, C., Razaka, G., Rabaud, M., and Bricaud, H. (1981). Catabolism of thymidine in human blood platelets: purification and properties of thymidine phosphorylase. Biochim. Biophys. Acta 654, 211–218. Eldridge, M.D., Murray, C.W., Auton, T.R., Paolini, G.V., Mee, R.P., and Murray, C.W. (1997). Empirical scoring functions.1. The development of a fast empirical scoring function to estimate the binding affinity of ligands in receptor complexes. J. Comput. Aided Mol. Des. 11, 425–445. Esnouf, R.M. (1999). Further additions to MolScript version 1.4, including reading and contouring of electron-density maps. Acta Crystallogr. D 55, 938–940. Fox, S.B., Westwood, M., Moghaddam, A., Comley, M., Turley, H., Whitehouse, R.M., Bicknell, R., Gatter, K.C., and Harris, A.L. (1996). The angiogenic factor platelet-derived endothelial cell growth factor/thymidine phosphorylase is up-regulated in breast cancer epithelium and endothelium. Br. J. Cancer 73, 275–280. Fukushima, M., Suzuki, N., Emura, T., Yano, S., Kazuno, H., Tada, Y., Yamada, Y., and Asao, T. (2000). Structure and activity of specific inhibitors of thymidine phosphorylase to potentiate the function of antitumor 2⬘-deoxyribonucleosides. Biochem. Pharmacol. 59, 1227– 1236. Grant, J.A., Pickup, B.T., and Nicholls, A. (2001). A smooth permittivity function for Poisson-Boltzmann solvation methods. J. Comput. Chem. 22, 608–640. Halgren, T.A., and Halgren, T.A. (1996). Merck molecular force field.1. Basis, form, scope, parameterization, and performance of mmff94. J. Comput. Chem. 17, 490–519. Hooft, R.W.W., Sander, C., Vriend, G., and Hooft, R.W.W. (1996). Positioning hydrogen atoms by optimizing hydrogen-bond networks in protein structures. Proteins 26, 363–376. Ignatescu, M.C., Gharehbaghi-Schnell, E., Hassan, A., Rezaie-Majd, S., Korschineck, I., Schleef, R.R., Glogar, H.D., and Lang, I.M. (1999). Expression of the angiogenic protein, platelet-derived endothelial cell growth factor, in coronary atherosclerotic plaques: in vivo correlation of lesional microvessel density and constrictive vascular remodeling. Arterioscler. Thromb. Vasc. Biol. 19, 2340–2347. Kraut, A., and Yamada, E.W. (1971). Cytoplasmic uridine phosphoryaseft liver. Characterization and kinetics. J. Biol. Chem. 246, 2021– 2030. Krenitsky, T.A. (1968). Pentosyl transfer mechanisms of the mammalian nucleoside phosphorylases. J. Biol. Chem. 243, 2871–2875. Krenitsky, T.A., and Tuttle, J.V. (1982). Correlation of substratestabilization patterns with proposed mechanisms for three nucleoside phosphorylases. Biochim. Biophys. Acta 703, 247–249. Krenitsky, T.A., Mellors, J.W., and Barclay, R.K. (1965). Pyrimidine nucleosidases: their classification and relationship to uric acid ribonucleoside phosphorylase. J. Biol. Chem. 240, 1281–1286.
Structure 84
Krenitsky, T.A., Koszalka, G.W., and Tuttle, J.V. (1981). Purine nucleoside synthesis, an efficient method employing nucleoside phosphorylases. Biochemistry 20, 3615–3621. Langen, P., Etzold, G., Barwolff, D., and Preussel, B. (1967). Inhibition of thymidine phosphorylase by 6-aminothymine and derivatives of 6-aminouracil. Biochem. Pharmacol. 16, 1833–1837.
Price, M.L., Guida, W.C., Jackson, T.E., Nydick, J.A., Gladstone, P.L., Juarez, J.C., Donate, F., and Ternansky, R.J. (2003). Design of novel N-(2,4-dioxo-1,2,3,4-tetrahydro-thieno[3,2-d]pyrimidin-7-yl)guanidines as thymidine phosphorylase inhibitors, and flexible docking to a homology model. Bioorg. Med. Chem. Lett. 13, 107–110.
Laskowski, R.A., Moss, D.S., and Thornton, J.M. (1993). Main-chain bond lengths and bond angles in protein structures. J. Mol. Biol. 231, 1049–1067.
Pugmire, M.J., and Ealick, S.E. (1998). The crystal structure of pyrimidine nucleoside phosphorylase in a closed conformation. Structure 6, 1467–1479.
Leslie, A.G.W. (1992). Recent changes to the MOSFLM package for processing film and image plate data. Joint CCP4 ⫹ ESF-EAMCB Newsletter on Protein Crystallography No. 26.
Pugmire, M.J., and Ealick, S.E. (2002). Structural analyses reveal two distinct families of nucleoside phosphorylases. Biochem. J. 361, 1–25.
Levitt, D.G. (2001). A new software routine that automates the fitting of protein X-ray crystallographic electron-density maps. Acta Crystallogr. D 57, 7–9.
Pugmire, M.J., Cook, W.J., Jasanoff, A., Walter, M.R., and Ealick, S.E. (1998). Structural and theoretical studies suggest domain movement produces an active conformation of thymidine phosphorylase. J. Mol. Biol. 281, 285–299.
Mackerell, A.D., Bashford, D., Bellott, M., Dunbrack, R.L., Evanseck, J.D., Field, M.J., Fischer, S., Gao, J., Guo, H., Ha, S., et al. (1998). All-atom empirical potential for molecular modeling and dynamics studies of proteins. J. Phys. Chem. B 102, 3586–3616. Maeda, K., Chung, Y.S., Ogawa, Y., Takatsuka, S., Kang, S.M., Ogawa, M., Sawada, T., Onoda, N., Kato, Y., and Sowa, M. (1996). Thymidine phosphorylase/platelet-derived endothelial cell growth factor expression associated with hepatic metastasis in gastric carcinoma. Br. J. Cancer 73, 884–888. Matsushita, S., Nitanda, T., Furukawa, T., Sumizawa, T., Tani, A., Nishimoto, K., Akiba, S., Miyadera, K., Fukushima, M., Yamada, Y., et al. (1999). The effect of a thymidine phosphorylase inhibitor on angiogenesis and apoptosis in tumors. Cancer Res. 59, 1911–1916. Merrit, E.A., and Murphy, M.E.P. (2003). Raster 3D version 2.0—a program for photorealistic molecular graphics. Acta Crystallogr. D 50, 869–873. Miwa, M., Ura, M., Nishida, M., Sawada, N., Ishikawa, T., Mori, K., Shimma, N., Umeda, I., and Ishitsuka, H. (1998). Design of a novel oral fluoropyrimidine carbamate, capecitabine, which generates 5-fluorouracil selectively in tumours by enzymes concentrated in human liver and cancer tissue. Eur. J. Cancer 34, 1274–1281. Miyadera, K., Sumizawa, T., Haraguchi, M., Yoshida, H., Konstanty, W., Yamada, Y., and Akiyama, S. (1995). Role of thymidine phosphorylase activity in the angiogenic effect of platelet derived endothelial cell growth factor/thymidine phosphorylase. Cancer Res. 55, 1687– 1690. Miyazono, K., Okabe, T., Urabe, A., Takaku, F., and Heldin, C.H. (1987). Purification and properties of an endothelial cell growth factor from human platelets. J. Biol. Chem. 262, 4098–4103. Mizutani, Y., Okada, Y., and Yoshida, O. (1997). Expression of platelet-derived endothelial cell growth factor in bladder carcinoma. Cancer 79, 1190–1194. Moghaddam, A., and Bicknell, R. (1992). Expression of plateletderived endothelial cell growth factor in Escherichia coli and confirmation of its thymidine phosphorylase activity. Biochemistry 31, 12141–12146. Moghaddam, A., Zhang, H.T., Fan, T.P., Hu, D.E., Lees, V.C., Turley, H., Fox, S.B., Gatter, K.C., Harris, A.L., and Bicknell, R. (1995). Thymidine phosphorylase is angiogenic and promotes tumor growth. Proc. Natl. Acad. Sci. USA 92, 998–1002. O’Brien, T., Cranston, D., Fuggle, S., Bicknell, R., and Harris, A.L. (1995). Different angiogenic pathways characterize superficial and invasive bladder cancer. Cancer Res. 55, 510–513. O’Brien, T.S., Fox, S.B., Dickinson, A.J., Turley, H., Westwood, M., Moghaddam, A., Gatter, K.C., Bicknell, R., and Harris, A.L. (1996). Expression of the angiogenic factor thymidine phosphorylase/platelet-derived endothelial cell growth factor in primary bladder cancers. Cancer Res. 56, 4799–4804. Patterson, A.V., Zhang, H., Moghaddam, A., Bicknell, R., Talbot, D.C., Stratford, I.J., and Harris, A.L. (1995). Increased sensitivity to the prodrug 5⬘-deoxy-5-fluorouridine and modulation of 5-fluoro2⬘-deoxyuridine sensitivity in MCF-7 cells transfected with thymidine phosphorylase. Br. J. Cancer 72, 669–675. Pauly, J.L., Schuller, M.G., Zelcer, A.A., Kirss, T.A., Gore, S.S., and Germain, M.J. (1977). Identification and comparative analysis of thymidine phosphorylase in the plasma of healthy subjects and cancer patients. J. Natl. Cancer Inst. 58, 1587–1590.
Reynolds, K., Farzaneh, F., Collins, W.P., Campbell, S., Bourne, T.H., Lawton, F., Moghaddam, A., Harris, A.L., and Bicknell, R. (1994). Association of ovarian malignancy with expression of plateletderived endothelial cell growth factor. J. Natl. Cancer Inst. 86, 1234– 1238. Schwartz, M. (1971). Thymidine phosphorylase from Escherichia coli. Properties and kinetics. Eur. J. Biochem. 21, 191–198. Schwartz, M. (1978). Thymidine phosphorylase from Escherichia coli. Methods Enzymol. 51, 442–445. Spraggon, M., Stuart, D., Ponting, C., Finnis, C., Sleep, D., and Jones, Y. (1993). Crystallization and x-ray diffraction study of recombinant platelet-derived endothelial cell growth factor. J. Mol. Biol. 234, 879–880. Sumizawa, T., Furukawa, T., Haraguchi, M., Yoshimura, A., Takeyasu, A., Ishizawa, M., Yamada, Y., and Akiyama, S. (1993). Thymidine phosphorylase activity associated with platelet-derived endothelial cell growth factor. J. Biochem. (Tokyo) 114, 9–14. Takebayashi, Y., Akiyama, S., Akiba, S., Yamada, K., Miyadera, K., Sumizawa, T., Yamada, Y., Murata, F., and Aikou, T. (1996a). Clinicopathologic and prognostic significance of an angiogenic factor, thymidine phosphorylase, in human colorectal carcinoma. J. Natl. Cancer Inst. 88, 1110–1117. Takebayashi, Y., Miyadera, K., Akiyama, S., Hokita, S., Yamada, K., Akiba, S., Yamada, Y., Sumizawa, T., and Aikou, T. (1996b). Expression of thymidine phosphorylase in human gastric carcinoma. Jpn. J. Cancer Res. 87, 288–295. Takeuchi, M., Otsuka, T., Matsui, N., Asai, K., Hirano, T., Moriyama, A., Isobe, I., Eksioglu, Y.Z., Matsukawa, K., and Kato, T. (1994). Aberrant production of gliostatin/platelet-derived endothelial cell growth factor in rheumatoid synovium. Arthritis Rheum. 37, 662–672. Thomas, M.B., Hoff, P.M., Carter, S., Bland, G., Lassere, Y., Wolff, R., Xiong, H., Hayakawa, T., and Abbruzzese, J. (2002). A dosefinding, safety and pharmacokinetics study of TAS-102, an antitumor/antiangiogenic agent given orally on a once-daily schedule for five-days every three weeks in patients with solid tumors. Proc. Am. Assoc. Cancer Res. 43, 554. Walter, M.R., Cook, W.J., Cole, L.B., Short, S.A., Koszalka, G.W., Krenitsky, T.A., and Ealick, S.E. (1990). Three-dimensional structure of thymidine phosphorylase from Escherichia coli at 2.8A˚ resolution. J. Biol. Chem. 265, 14016–14022. Zimmerman, M., and Seidenberg, J. (1964a). Deoxyribosyl transfer. I. Thymidine phosphorylase and nucleoside deoxyribosyltransferase in normal and malignant tissues. J. Biol. Chem. 239, 2618–2621. Zimmerman, M., and Seidenberg, J. (1964b). Deoxyribosyl transfer. II. Nucleoside: pyrimidine deoxyribosyltransferase activity of three partially purified thymidine phosphorylases. J. Biol. Chem. 239, 2622–2627. Accession Numbers The atomic coordinates and structure factors have been deposited in the Protein Data Bank (accession code 1UOU).