Binding of the anticancer drug ZD1694 to E. coli thymidylate synthase: assessing specificity and affinity

Research Article

1317

Binding of the anticancer drug ZD1694 to E. coli thymidylate synthase: assessing specificity and affinity Earl E Rutenber and Robert M Stroud* Background: Thymidylate synthase (TS) catalyzes the reductive methylation of deoxyuridine monophosphate (dUMP) by 5,10-methylenetetrahydrofolate (CH2H4folate) to form deoxythymidine monophosphate (dTMP) and dihydrofolate (H2folate). The essential role of TS in the cell life cycle makes it an attractive target for the development of substrate and cofactor-based inhibitors that may find efficacy as anticancer and antiproliferative drugs. Antifolates that compete specifically with the binding of CH2H4 folate include the cofactor analog CB3717 (10-propargyl-5,8dideazafolate). However, the development of potent cofactor analog inhibitors of TS, such as CB3717, as drugs has been slowed by their toxicity, which often becomes apparent as hepatic and renal toxicity mediated by the specific chemistry of the compound. Attempts to abolish toxicity in human patients while preserving potency against the target enzyme, have led to the development of ZD1694, which has already shown significant activity against colorectal tumours.

Address: Department of Biochemistry and Biophysics, University of California at San Francisco, San Francisco, California 94143–0448, USA. *Corresponding author. E-mail: [email protected] Key words: anti-cancer drug, folate analog, protein-ligand interaction, structure-based drug design, ZD1694 Received: 5 August 1996 Revisions requested: 2 September 1996 Revisions received: 16 September 1996 Accepted: 17 September 1996

Structure 15 November 1996, 4:1317–1324 Results: The three dimensional crystallographic structure of ZD1694 in complex with dUMP and Escherichia coli TS has been determined to a resolution of 2.2 Å. © Current Biology Ltd ISSN 0969-2126 This was used to evaluate the specific structural determinants of ZD1694 potency and to correlate structure/activity relationships between it and the closely related ligand, CB3717. ZD1694 binds to TS in the same manner as CB3717 and H2folate, but a methyl group on its quinazoline ring, its thiophene ring and the methyl group at N10 are compensated for by plastic accommodation of the enzyme active site coupled with specific rearrangement in the solvent structure. A specific hydrogen bond between the protein and the inhibitor CB3717 is absent in the case of ZD1694 whose monoglutamate tail is reoriented and more well ordered.

Conclusions: The binding mode of ZD1694 to thymidylate synthase has been determined at atomic resolution. ZD1694 forms a ternary complex with dUMP and participates in the multi-step TS reaction through the covalent bond formation between dUMP and Cys146 thereby competing with CH2H4folate at the active site. Analysis of this inhibitor ternary complex structure and comparison with that of CB3717 reveals that the enzyme accommodates the differences between the two inhibitors with small shifts in the positions of key active site residues and by repositioning an active site water molecule, thereby preserving a general binding mode of these inhibitors.

Introduction Thymidylate synthase (TS; EC 2.1.1.45) is recognized and exploited as the target of various anticancer drugs because it catalyzes the reductive methylation of deoxyuridine monophosphate (dUMP) in the final step in the sole de novo pathway for synthesis of deoxythymidine monophosphate (dTMP) and is therefore an essential enzyme of DNA synthesis. This unique role makes TS an ideal candidate for the design of potent inhibitors which will serve as lead compounds in the development of anticancer drugs. Drugs that compete directly or indirectly with dUMP binding to TS include the mechanism-based derivatives of pyrimidines, such as the well known prodrug 5-fluorouracil (5FU) [1,2] which ultimately forms a covalent mechanism-based inhibitory complex with TS.

Antifolates that compete specifically with the binding of the fully reduced form of folate, CH2H4folate, include the analog 10-propargyl-5,8-dideazafolate, CB3717, discovered by Jones and colleagues [3–5]. CB3717, which inhibits human TS in vitro with Ki ~10 nM and displays antitumor activity in vivo, demonstrates excellent clinical efficacy; however, CB3717 has a poor therapeutic index due to its low solubility at urinary and hepatic pH. Clinical trials were halted after a small proportion of patients exhibited renal and hepatic toxicity [3–7]. Numerous analogs of the quinazloine-based CB3717 have been made and screened to characterize those with improved solubility. A 2-desamino-2-methyl compound, ZD1694 (sold under the trade name Tomudex) was found

1318

Structure 1996, Vol 4 No 11

to have high solubility. Though it has weaker binding to free TS (Ki = 670 nM), it has improved interaction with the enzyme, folylpolyglutamate synthase (FPGS), that attaches up to five a- and g-linked glutamates to folates inside the cell. FPGS-mediated polyglutamylation after cellular uptake serves to sequester the analog within the target cell and also improves its potency against free TS by factors up to >100 [8]. ZD1694 therefore is a much better in vivo inhibitor of TS than is CB3717. ZD1694 is a potent inhibitor of L1210 (ascitic murine line) cell growth and leads to regression and elimination of ascitic tumors in whole mice [8]. The antitumor activity of ZD1694 is prevented by co-administration of thymidine indicating that TS is the target of its action in vivo [8]. The antitumor activity is diminished by co-administration of folinic acid, but this effect is decreased when folinic acid is given as a four hour delayed ‘rescue’, indicating that ZD1694 competes with folinic acid for cellular uptake [8]. ZD1694, approved for human therapy in several European countries, has recently completed phase III clinical trials and has shown significant activity against colorectal tumors [9]. ZD1694 differs chemically from its predecessor CB3717 at three specific sites (Fig. 1). Firstly, the CB3717 quinazoline 2-amino group, a hydrogen bond donor, is changed to a methyl group in ZD1694. This change abolishes intermolecular crosslinking at high pH which presumably prevents crystallization in the kidney and diminishes the serious side effect of renal failure. In the complex of Escherichia coli thymidylate synthase (ECTS) with dUMP and CB3717, this amino group forms hydrogen bonds to the protein.

Secondly, the p-aminobenzoate (PABA) ring of CB3717 is changed to a thiophene ring in ZD1694. The most potent cytotoxic compounds are analogs in which the PABA ring is replaced by a thiophene ring [10]. The thiophene ring presumably mediates the improved interaction with FPGS, a major factor in producing high levels of in vivo potency. The thiophene and PABA rings are isosteric although the thiophene ring sulfur confers greater electronegativity to the ring system. And thirdly, the CB3717 N10-propargyl substituent is truncated to a methyl group in ZD1694. The propargyl group confers increased affinity by ~tenfold relative to CH3 and is essential for potency of CB3717 [6]. Its role in binding, as seen in the ECTS-dUMP-CB3717 structure [11], is hydrophobic since the propargyl group extends into a hydrophobic site and packs against the aromatic ring of Phe176 and the backbone atoms of Gly173. We have now determined the structure of ECTS in ternary complex with dUMP and ZD1694 to investigate how the enzyme differentially accommodates these fundamental and chemical modifications of the inhibitor, and to define further sites where subsequent modifications may be focused to optimize selectivity or affinity. The crystal structure of the ECTS-dUMP-ZD1694 complex was determined and refined to 2.2 Å resolution.

Results and discussion Overall features of the protein and binding

Two molecules of ZD1694 are bound, one in each of two symmetrically arranged active sites (Figs 2,3). The enzyme is in the ‘closed’ conformation that is triggered by binding of both substrate and folate (or a folate analog) to TS

Figure 1 Chemical structures of (a) CB3717 and (b) ZD1694 showing key differences between the two.

(a) propargyl

O 4

HN

CH 2 N

5

3

2 1

H 2N

CO2H

CH 2

4a

6

8a

7

10

CONH

C

H

CH 2CH 2CO2H PABA

glutamate

8

N quinazoline

(b)

O

CH 3

4

HN

3

4a

2 1

H 3C

CH 2 N

5

10

CONH

7 8

N 2-methyl quinazoline

C

H

CH 2CH 2CO2H

6

8a

CO2H

S

thiophene

glutamate

Research Article Structure of thymidylate synthase with ZD1694 Rutenber and Stroud

1319

Figure 2 Cross-eyed stereo view of the protein a carbon trace for a single monomer of the TS structure showing the locations of residues. ZD1694 and substrate dUMP are also shown as all atom representations in the active site. (Figure generated using the program CHAIN [27].)

191

233 64

39

76 87 256

47

1

264 214

[11–14] or by binding of antifolates alone [15,16]. The ‘closed’ conformation differs from the ‘open’ conformation, observed for unliganded TS [17] by the concerted inward movement of many secondary structural elements. The resulting conformational change extends throughout the entire protein and occurs by segmental accommodation [11]. This conformational change serves to sequester reactants from bulk solvent and to position the substrate and cofactor for catalysis. The structure of the protein components of the ZD1694 and CB3717 complexes are very Figure 3

Stylized ribbon representation of ZD1694 bound with substrate dUMP to the TS dimer showing the relative locations of substrate and inhibitor binding. One inhibitor molecule (ZD1694) and one substrate molecule (dUMP) bind in each of two, twofold-related active sites. In each active site, dUMP binds closest to the dimer interface. (Figure generated using the program MidasPlus [25,26].)

120

191

233 104 131 149 164 202

76 87 256

28 16

64 39

47

1 264

214

120 104 131 149 164 202

28 16

similar. Least squares superimposition guided by Ca atoms of the ZD1694 complex and the CB3717 complex resulted in a root mean square (rms) deviation of 0.19 Å on positions of protein Ca atoms and a rms deviation of 0.75 Å on positions of all protein atoms. Inhibitor binding at the active site

The bound conformation and orientation of ZD1694 is essentially the same as observed in TS ternary complexes with folates [12], or CB3717 [11]. Key interactions between protein sidechains and CB3717, unless otherwise noted, are preserved in the ZD1694 structure. The refined omit difference electron density map calculated after omitting the catalytic active site residue Cys146, the substrate dUMP and the inhibitor ZD1694 clearly shows resolved electron density for every atom excluded (Fig. 4). The substrate dUMP binds at the bottom of the large active site with the pyrimidine ring in an anti orientation and the ribose ring in a C2′ endo conformation. A covalent bond, seen clearly in the difference electron density map, is formed between the Sg of Cys146 and the C6 position of the pyrimidine ring. ZD1694 binds adjacent to the substrate, securing the pyrimidine ring of dUMP in its productive binding to the catalytic nucleophile, Sg of Cys146 (Fig. 5). In the normal course of the catalytic reaction, this covalent bond activates the C5 position of the pyrimidine ring for bond formation with the iminium ion methylene from the CH2H4folate cofactor. In ZD1694, as there is no reactive methylene to donate, the receptive substrate intermediate is generated once the Sg–C6 bond is formed and stabilized by the binding of ZD1694. This arrangement and orientation of substrate and cofactor analog is also similar in detail to that of the product and oxidized cofactor bound to an active site mutant of ECTS, in the product complex structure [12]. This shows that the binding of dUMP and ZD1694 to TS causes the normal

1320

Structure 1996, Vol 4 No 11

Figure 4

Cross-eyed stereo view of refined omit difference electron density map calculated with terms [(Fo–Fc) acalc] after omitting

Cys146, dUMP (DUM 565) and ZD1694 (D16 566) from one active site and refining the structure. The final refined model for the

reaction to proceed as far as the covalent bond formation between dUMP C6 and the Cys146 Sg, whereupon a covalent inhibitory complex is formed. Superimposition and comparison of the ECTS-dUMPZD1694 and the ECTS-dUMP-CB3717 structures and the ECTS-dUMP-CB3717 structure reveals a set of 28 residues in the first monomer (residues 1–264), that differ in position by more than four times the amount expected based on the average B factor of the compared atoms [18]. Most of these residues are located on the solvent-accessible surface of the protein and do not interact with the bound ligands. Specific residues at the enzyme inhibitor interface have statistically significant positional shifts. The residues from both structures within 5.0 Å of the Figure 5

Overview of the cofactor analog ZD1694 bound in the active site of ECTS. The protein backbone is shown as stylized ribbon in cyan. ZD1694, dUMP and residues lining the active site are shown colored by atom type. (Figure generated using the program MidasPlus [25,26].)

omitted residue and the ligands are superimposed. The electron density, contoured at 2.5 s, is strong and continuous for the covalent bond between Sg of Cys146 and C6 of dUMP. Under various refinement schemes in which the van der Waals interaction energies between substrate dUMP and protein were omitted, the unrestrained distance between these two atoms refines to that of a reasonable covalent bond (~1.85 Å). Figure generated using the program CHAIN [27].

bound inhibitor and the surrounding active site are shown in Figure 6. In the ECTS-dUMP-ZD1694 structure the sidechains of Leu172 and Phe176 pack against the thiophene ring and have a rms positional shift of 1.9 Å and 2.2 Å respectively, relative to the ECTS-dUMP-CB3717 structure. The rearrangement of these sidechains accommodates the increased steric bulk of the sulfur atom and the small changes in the position of the thiophene ring relative to the PABA ring in the ECTS-dUMP-CB3717 structure. The sidechain of Glu82 lies farther from the inhibitor making its closest approach, of 6.8 Å, between Glu82 Oε2 and the carbonyl oxygen of ZD1694 adjacent to the glutamic acid tail. Nonetheless, the Glu82 sidechain has a rms shift of 1.9 Å relative to the ECTS-dUMPCB3717 structure. This rearrangement may be due to shifts in the position of the glutamic acid tail and bound water molecules nearby. The 2-desamino-2-methyl quinazoline ring of the ZD1694 packs against the uridine ring of the substrate dUMP in the mode observed for the quinazoline ring of CB3717. In ZD1694, a methyl group occupies the 2- position of this ring system. This methyl group is bound in a pocket formed by side chains of residues Asp169, Tyr209, the carbonyl oxygen of Ala263 and a well ordered water molecule, Wat580 . In the ECTS-dUMP-ZD1694 structure the interatomic distances between the methyl carbon and the closest atoms of each of these residues ranges from 3.3 to 3.8 Å (Table 1). In the ECTS-dUMP-CB3717 structure, the 2-amino group of CB3717 makes a strong 2.8 Å hydrogen bond to the carbonyl of the penultimate residue, Ala263, and much weaker 3.6 Å interactions with the Od of Asp169 and OH group of Tyr209. The 2-amino group is also coordinated by a well ordered water molecule with a hydrogen bond of 3.1 Å. The loss of strong electrostatic interactions at the 2-position and the entropic cost of packing the hydrophobic methyl group in a polar pocket may explain the tenfold loss of affinity seen with ZD1694 compared with CB3717, and the structure described here suggests ways in which higher affinity could be evoked

Research Article Structure of thymidylate synthase with ZD1694 Rutenber and Stroud

Figure 6

1321

surface potential [16]. Key interactions between the monoglutamate tail and the protein are mediated through a network of ordered water molecules (Fig. 7). The two oxygens of the a-carboxylate form hydrogen bonds to three water molecules which in turn are hydrogen bonded to the main-chain protein atoms, His51 N and Ile258 O, and to the side chain atom Lys48 Nz. One Oε atom of the ZD1694 g-carboxylate group forms a hydrogen bond to Og of Ser54 and both oxygen atoms form a bifurcated hydrogen bond with Nd of His51. In the ECTS-dUMP-ZD1694 structure the slightly different orientation of substituents on the five-membered thiophene ring relative to the six-membered PABA ring causes a shift of 0.3 Å in the Ca atom of the glutamic acid tail. Comparison to human thymidylate synthase binding

Overlap, guided by superposition of protein Ca’s, of ZD1694 and CB3717. The local environment of the active site and sidechains from residues within 5 Å of the bound inhibitor are shown, with sidechains and ligands from the ZD1694 structure colored by atom and those from the CB3717 structure colored green. (Figure generated using program MidasPlus [25,26]).

in subsequent modifications. A similar rearrangement in the water structure at this site is seen in an ECTS mutant where asparagine replaces Asp169 (C Sage and RS, unpublished results). The volume occupied by the extended propargyl substituent of CB3717 is vacant in the ZD1694 structure, in which the backbone Ca of Gly173 is 0.9 Å closer to the N10-methyl group of ZD1694. The propargyl group of CB3717 is responsible for the highest specificity of the 10-substituted quinazoline-based cofactor analogs. In the ECTS-dUMP-CB3717 structure, the propargyl group packs against the conserved Phe176 and the linear axis of the three carbon group points toward the backbone of Gly173. The triple bonded terminal carbon is 3.3Å from Nd of Asn 177, 3.3 Å from N3 of dUMP and 3.5 Å from O4 of dUMP. In the ECTS-dUMP-ZD1694 structure, the N10 methyl group is a r the negatively charged monoglutamate tail is a shallow groove carrying a positive electrostatic

Thymidylate synthase is a highly conserved enzyme with a high number of conserved residues comprising the active site [17]. Visual inspection of the active site relative to the sequence alignment of ECTS and human thymidylate synthase (HTS) reveals that sequence variation does occur in proximity to the bound inhibitor, but the nature of the substitutions are unlikely to affect the binding mode of ZD1694 in the HTS active site. In the ECTS-dUMP-ZD1694 structure there are 20 residues that have at least one atom within 5 Å of the bound inhibitor. Of these residues, 14 are conserved in the sequence alignment between ECTS and HTS, and two are amino acid substitutions to a smaller side chain. The remaining four amino acid substitutions are to a larger side chain and could affect the overall binding mode. These side chain substitutions are His51, Thr78, Ala260 and Val262 in ECTS for Phe, Lys, Met and Met in HTS, respectively. However, the side chains of His51, Thr78 and Ala260 all lie closest to the glutamate tail of ZD1694 and comprise the shallow groove in which the tail is bound, a location least likely to affect inhibitor binding. His51 and Thr78 mediate interactions with the glutamate tail through hydrogen bonds to well ordered solvent molecules. In contrast, the side chain of Val262 packs against the quinazoline ring of ZD1694 and although large perturbations in the protein structure of this region could affect inhibitor binding, the site of this side chain could be occupied by the bulkier side chain of methionine.

Table 1 Interatomic distances for analogous atoms in ECTS−dUMP−ZD1694 and ECTS−dUMP−CB3717 structures. CB3717 NA2 NA2 NA2 NA2 N10

Asp169 Tyr209 Ala263 Wat580 Gly173

OD2 OH O O Ca

D (Å)

ZD1694

3.7 3.7 2.8 3.1 6.0

CM2 CM2 CM2 CM2 N10

D (Å) Asp169 Tyr209 Ala263 Wat580 Gly173

OD2 OH O O Ca

3.8 3.7 3.3 3.4 5.4

1322

Structure 1996, Vol 4 No 11

Figure 7

dUMP 565 OR3

Ala 263 N

CB

CR3

CR2 C6

CR4

CR1

CA

C5

CR5

C OP1

O

N1

OR1

OR5

C4

C2

O4 P

O2 N3

OP2 OP3

N

2.83

Phe 176

C

Trp 83 O

2.94

WAT 811

Leu 143

O

Ile 79

CD1

WAT 580

NE1

O

3.27

2.98

3.01

CP1

CZ2

C8 CA

C16

WAT 643

C7

C15

CB

C11

3.08

C

CD2 CE3

N1 C8A

N10

CH2

CM2 C2

C6 C9

C14

OG

CB

CE2

ZD1694

O O

CG

3.12

Ser 54 N

CA

CZ3

C4A

O

Schematic diagram of the ligand binding determinants as displayed by the program LIGPLOT [28]. The hydrogen bond network of ordered water molecules around the glutamate tail of ZD1694 is illustrated. The bonds of ZD1694 are colored purple and those of the protein are colored black. Hydrogen bonds are black dashes with the bond length noted. Protein residues and ligand atoms involved in hydrophobic contacts with the ligand are noted with green eyelashes.

C5

N3 C4

3.11

S13

C

OE2

Tyr 209

2.72

N

OD2

OA4

OD1

CB CD NE2

His 51

3.21

CE1

WAT 770

OE1

2.92 ND1

Trp 80

CA O CG

CD2

2.87

CB

CT

2.84

O2 2.68

O1

CA CA C

CG

O CA

CB

O

2.97

C

O

Asp 169

Leu 172 N

N

N

O

WAT 774

C

CG

Gly 173

2.86 2.87 NZ

O

3.29

WAT 653

CE

3.02

3.21 O

CD

WAT 771 CG

O

WAT 769

CB

2.72 O

N

CA O CA

C

N C

CG1 CB

CD1

Lys 48

CG2

Ile 258

Biological implications Thymidylate synthase (TS) catalyzes the committed step in the sole de novo pathway for biosynthesis of thymidine and is therefore important for the production of DNA, but is not required for the synthesis of mRNA. Thus inhibitors that target TS, act preferentially on cells that are rapidly proliferating with an elevated demand for DNA synthesis, such as cancer cells. In a well studied, multi-step mechanism [19,20] TS catalyzes the reductive methylation of deoxyuridine monophosphate (dUMP) by transfer and subsequent reduction of a methylene group from the cofactor methylenetetrahydrofolate (5,10CH2THF) to yield deoxythymidine monophosphate (dTMP) and dihydrofolate (H2folate). Recently, ZD1694 (trade name Tomudex) has emerged from phase III clinical trials as a promising and effective alternative to the only widely approved anticancer therapy that targets thymidylate synthase, the pro-drug 5fluorouracil (5FU) [9]. ZD1694 is less toxic than its predecessor drug candidate, CB3717, which was removed from clinical trials, despite its effectiveness against TS, after small numbers of patients developed renal and hepatic toxicities. ZD1694 is as effective as the approved combination regime of 5FU and leucovorin (LV) in limiting disease progress and mediating survival [9]. Additionally, patients who received ZD1694 had shorter hospital

stays, higher response rates, reduced toxicity and more frequent palliative benefits when compared with patients who received 5-FU plus LV in the treatment of advanced colorectal cancers [9]. The crystal structure of ZD1694 bound by E. coli TS has now been determined at a resolution of 2.2 Å and used to evaluate the structural determinants of the ligand– protein interaction. This structure also provides the opportunity for comparison of the binding mode of ZD1694 with that of both the closely related cofactor analog, CB3717 and the cofactor product, H2folate. ZD1694 acts as an inhibitor of the multi-step TS reaction by catalyzing the first part of the reaction, covalent bond formation between substrate dUMP and the catalytic thiol Cys146, thereby competing with cofactor 5,10CH2THF at the active site. The enzyme is therefore ‘trapped’ at a reversible step midway along the reaction pathway, because ZD1694 lacks the iminium ion methylene substituent required for continuation of the normal reaction. The chemical structure of ZD1694 differs from that of CB3717 in three distinct ways, each of which is compensated for by plastic accommodation of the protein and rearrangement of the solvent structure, thereby preserving the general binding mode of these inhibitors. The modification of a key amino group in CB3717 to a methyl group at the same position in ZD1694 abrogates

Research Article Structure of thymidylate synthase with ZD1694 Rutenber and Stroud

a specific hydrogen bond between the inhibitor and an enzyme-ligated water molecule at the active site. The loss of this energetically favorable interaction may account for the 67-fold difference in binding affinities for the free enzyme. The p-aminobenzoate (PABA) moiety of CB3717 is changed to a thiophene ring in ZD1694 thereby improving the metabolically important interaction between ZD1694 and the enzyme which catalyzes polyglutamylation, folylpolyglutamate synthase (FPGS). The same hydrophobic sidechains from active site residues stack against both the PABA and the thiophene ring systems in much the same manner. The shifts in these residue positions probably do not represent large differences in binding affinities of CB3717 and ZD1694. The propargyl substituent at N10 in CB3717 is truncated to a methyl group in ZD1694. The binding of this smaller functional group leads to a slight shift of the protein backbone toward the inhibitor creating a small cavity unoccupied by either inhibitor atoms or ordered solvent molecules. This energetically unfavorable cavity may account for a small portion of the differences in binding affinities between CB3717 and ZD1694.

Materials and methods Protein and co-crystallization The thyA gene in pUC19 was transformed into E. coli X2913. After induction to 5 to 10 % of the cellular protein, the enzyme was purified as described [21], stored as an ammonium sulfate precipitate at –80° C and dialyzed against 20 mM KPO4, pH 8.0, 3.8 mM dithiothriotol (DTT), and 0.05 mM disodium EDTA for crystallization of the inhibited enzyme. Ternary complex crystals were grown by hanging drop co-crystallization of TS with the substrate dUMP and ZD1694. A 5.0 ml drop of protein solution (20 mM KPO4 pH 8.0, 1.5 mg ml–1 TS, 1.9 mM ZD1694, 5.8 mM dUMP, 3.8 mM MgCl2, 3.8 mM DTT, 0.05 mM EDTA and 1.25 M (NH4)2SO4) was equilibrated at 23° C against an excess of precipitant solution (2.5 M (NH4)2SO4 and 20 mM KPO4 pH 8.0). Crystals with hexagonal rod morphology grew in 2–3 days, reaching a size of 650 mm × 200 mm × 200 mm. Diffraction intensities were collected using a Siemens X-100 position sensitive area detector system, CuKa (1.54 Å) radiation, frames of width ∆Ω = 0.2° degree and data collection

1323

times of 300 seconds per frame. A single crystal yielded a complete dataset with a total of 249 261 observations to 2.2 Å. Intensities from 600 frames recorded using two crystal orientations were integrated, scaled and reduced using the Siemens software in space group P63 with unit cell dimensions a = b = 126.8 Å, c = 67.56 Å, a = b = 90°, g = 120° to give 30 531 unique reflections with an overall Rsym of 10.05 % calculated on intensities. Intensities from 600 frames recorded using two crystal orientations were intergrated, scaled and reduced using the Siemens software in space group P63 with unit cell dimensions a = b = 126.8 Å, c = 67.56 Å, a = b = 90°, g = 120° to give 30 531 unique reflections with an overall Rsym of 10.05 % calculated on intensities (Table 2).

Structure solution and refinement The structure was solved by isomorphous molecular replacement using the reported ECTS-dUMP-CB3717 ternary complex structure [11], with the ligands and water molecules omitted, as a model. Rigid body least squares minimization of the model position using X-PLOR [22] gave an R factor of 29.2 % using all data between 7.0 Å and 2.2 Å. A difference electron density map ((Fo–Fc)ac) calculated using the rigid-body refined model displayed clear, detailed density for the ligands, ZD1694 and dUMP, and for well ordered solvent molecules. The ligands were modeled into the difference electron density using FRODO [23]. The complete model was subjected to alternating cycles of positional and B factor refinement, including two rounds of simulated annealing [24] and hand rebuilding. The final R factor is 17.1 % for all data between 7.0 Å and 2.2 Å for a model with a total of 4649 atoms, including 245 water molecules. A summary of the data processing and refinement statistics is presented in Table 2. ECTS-dUMP-ZD1694 was superimposed on ECTS-dUMP-CB3717 by automated selection of a core of Ca atoms in the dimer based on difference distance matrices [17] and minimizing the rms deviation in these Ca positions in the two structures, using the programs NEWDOME and GEM [12]. The ECTS-dUMP-ZD1694 structure was compared to the ECTS-dUMP-CB3717 structure by first deriving an estimate of uncertainty in position expressed as the expectation in standard deviation, sxyz(B), for each atom in both structures as a function of the averaged B factor of the compared atoms. The calculated sxyz (B), resolution and other determinants of the analysis are based on comparison of independently determined structures of proteins seen where there are multiple copies located at different positions in the same crystal [18].

Accession numbers Coordinates have been deposited with the Protein Data Bank.

Table 2 Crystallographic data and refinement statistics for the ECTS−dUMP−ZD1694 complex Resolution (Å)

Crystallographic data Rsym (I) (%)* Av. I/sig(I) No. of unique reflections, possible No. of reflections, collected No. of observations,total Completeness of data (%) Redundancy Rcryst of refined structure (%)†

∞–3.81

3.81–3.02

3.02–2.64

2.64–2.40

2.40–2.20

5.55 21.3 2139 6218 66 227 99.9 10.7 12.9 (7.0–4.0 Å)

11.0 7.49 2109 6106 59 336 99.7 9.7 14.1

21.7 3.32 2096 6094 50 753 99.8 8.3 18.7

32.8 2.06 2089 6066 41677 100.0 6.9 21.1

35.1 1.56 2109 6047 31268 99.0 5.2 26.4

*Rsym = (ΣhklΣΝi=l[Iavg–Ii]2 / [ΣhklΣΝi=l(Ii)2])½. †Rcryst = Σ|(|Fo|–|Fc|)| / Σ|Fo|. Rms deviations from ideality in final model were: 0.014 Å (Bond lengths); 3.53° (Bond angles); 22.5° (Dihedral angles) and 3.91°

(Improper angles). Improper angles define chiral centres and planar groups of atoms.

1324

Structure 1996, Vol 4 No 11

Acknowledgements Work was supported by grant CA 63081 from NIH to RMS.

References 1. Hartman, K.R. & Heidelberger, C. (1958). Studies on fluorinated pyrimidines: XIII. Inhibition of thymidylate synthase. J. Biol. Chem. 236, 3006–3016. 2. Heidelberger, C. (1965). Fluorinated pyrimidines. Prog. Nucleic Acid Res. Mol. Biol. 4, 1–50. 3. Jones, T.R., et al., & Harrap, K.R. (1981). A potent antitumor quinazoline inhibitor of thymidylate synthase: synthesis, biological properties and therapeutic results in mice. Eur. J. Cancer 17, 11–19. 4. Jackson, R.C., Jackman, A.L. & Calvert, A.H. (1983). Biochemical effects of a quinazoline inhibitor of thymidylate synthase, CB3717, on human lymphoblastoid cells. Biochem. Pharmacol. 32, 3783–3790. 5. Jackman, A.L., et al., & Calvert, A.H. (1990). Activity of the thymidylate synthase inhibitor 2-desamino-N10-propargyl-5,8-dideazafolic acid and related compounds in murine (L1210) and human (W1L2) systems in vitro and in L1210 in vivo. Cancer Res. 50, 5212–5218. 6. Alison, D.L., et al., & Calvert, A.H. (1985). The clinical pharmacokinetics of the novel antifolate N10-propargyl-5,8-dideazafolic acid (CB3717). Cancer Chemother. Pharmacol. 14, 265–271. 7. Newell, D.R., et al., & O’Connor, P. (1986). Pharmacokinetics of the thymidylate synthase inhibitor N10-propargyl-5,8-dideazafolic acid (CB3717) in the mouse. Cancer Treat. Rep. 70, 971–979. 8. Jackman, A.L., et al., & Stephens, T.C. (1991). ICI D1694, a quinazoline antifolate thymidylate synthase inhibitor that is a potent inhibitor of L1210 tumor cell growth in vitro and in vivo: a new agent for clinical study. Cancer Res. 51, 5579–5586. 9. Cunningham, D., et al., & Perez-Manga, G. (1995). ‘Tomudex’ (ZD1694): results of a randomised trial in advanced colorectal cancer demonstrate efficacy and reduced mucositis and leucopenia. The ‘Tomudex’ Colorectal Cancer Study Group. Eur. J. Cancer A 31, 1945–1954. 10. Marsham, P.R., et al., & Calvert, A.H. (1991). Quinazoline antifolate thymidylate synthase inhibitors: heterocyclic benzoyl ring modifications J. Med. Chem. 34, 1594–1605. 11. Montfort, W.R., et al., & Stroud, R.M. (1990). Structure, multiple site binding, and segmental accommodation in thymidylate synthase on binding dUMP and an anti-folate. Biochemistry 29, 6964–6977. [Published erratum appears in Biochemistry (1990) 29, 10864]. 12. Fauman, E.B., Rutenber, E.E., Maley, G.F., Maley, F. & Stroud, R.M. (1994). Water-mediated substrate/product discrimination: the product complex of thymidylate synthase at 1.83 Å Biochemistry 33, 1502–1511. 13. Matthews, D.A., et al., & Freer, S. (1990). Stereochemical mechanism of action for thymidylate synthase based on the X-ray structure of the covalent inhibitory ternary complex with 5-fluoro-2′-deoxyuridylate and 5,10-methylenetetrahydrofolate. J. Mol. Biol. 214, 937–948. 14. Matthews, D.A., Appelt, K., Oatley, S.J. & Xuong, N.H. (1990). Crystal structure of Escherichia coli thymidylate synthase containing bound 5-fluoro-2′-deoxyuridylate and 10-propargyl-5,8-dideazafolate. J. Mol. Biol. 214, 923–936. 15. Kamb, A., Finer-Moore, J.S. & Stroud, R.M. (1992). Cofactor triggers the conformational change in thymidylate synthase: implications for an ordered binding mechanism. Biochemistry 31, 12876–12884. 16. Kamb, A., Finer-Moore, J., Calvert, A.H. & Stroud, R.M. (1992). Structural basis for recognition of polyglutamyl folates by thymidylate synthase. Biochemistry 31, 9883–9890. 17. Perry, K.M., et al., & Stroud, R.M. (1990). Plastic adaptation toward mutations in proteins: structural comparison of thymidylate synthases. Proteins 8, 315–333. 18. Stroud, R.M. & Fauman, E.B. (1995). Significance of structural changes in proteins: expected errors in refined protein structures. Protein Sci. 4, 2392–2404. 19. Stroud, R.M. & Finer-Moore, J.S. (1993). Stereochemistry of a multistep/bipartite methyl transfer reaction: thymidylate synthase. FASEB J. 7, 671–677. 20. Carreras, C.W. & Santi, D.V. (1995). The catalytic mechanism and structure of thymidylate synthase. Annu. Rev. Biochem. 64, 721–762. 21. Maley, G.F. & Maley, F. (1988). Properties of a defined mutant of Escherichia coli thymidylate synthase. J. Biol. Chem. 263, 7620–7627. 22. Brünger, A.T. (1992). X-PLOR Version 3.1. A system for X-ray crystallography and NMR, Yale University Press, New Haven and London.

23. Jones, T.A. (1985). Diffraction methods for biological macromolecules. Interactive computer graphics: FRODO. Methods. Enzymol. 115, 157–171. 24. Brünger, A.T. (1988). Crystallographic refinement by simulated annealing. Application to a 2.8 Å resolution structure of aspartate aminotransferase. J. Mol. Biol. 203, 803–816. 25. Huang, C.C., Pettersen, E.F., Klein, T.E., Ferrin, T.E. & Langridge, R. (1991). Conic: a fast renderer for space-filling molecules with shadows. J. Mol. Graphics 9, 230–236. 26. Ferrin, T.E., Huang, C.C., Jarvis, L.E. & Langridge, R. (1988). The MIDAS Display System. J. Mol. Graphics 6, 13–37. 27. Sack, J. (1988). CHAIN - A crystallographic modeling program. J. Mol. Graphics 6, 244–245. 28. Wallace, A.C., Laskowski, R.A. & Thornton, J.M. (1995). LIGPLOT: a program to generate schematic diagrams of protein-ligand interactions. Protein Eng. 8, 127–134.