In Complex with the Nonsteroidal Anti-Inflammatory Drugs Flufenamic Acid and Indomethacin

[CANCER RESEARCH 64, 1802–1810, March 1, 2004]

Crystal Structures of Prostaglandin D2 11-Ketoreductase (AKR1C3) in Complex with the Nonsteroidal Anti-Inflammatory Drugs Flufenamic Acid and Indomethacin

Andrew L. Lovering,1 Jon P. Ride,1 Christopher M. Bunce,1 Julian C. Desmond,2 Stephen M. Cummings,1 and Scott A. White1 The Schools of 1Biosciences and 2Medicine, The University of Birmingham, Birmingham, United Kingdom

ABSTRACT this purpose depends on the precise determination of the mechanisms whereby they exert their antitumor effects. It is becoming increasingly well established that nonsteroidal anti- Because inflammatory diseases of the gastrointestinal tract predis- inflammatory drugs (NSAID) protect against tumors of the gastrointesti- pose to the development of cancer and because COX-2 expression is nal tract and that they may also protect against a variety of other tumors. elevated during the progression of colon carcinoma, it has become These activities have been widely attributed to the inhibition of cylooxy- genases (COX) and, in particular, COX-2. However, several observations widely accepted that the anti-inflammatory and antineoplastic actions have indicated that other targets may be involved. Besides targeting COX, of NSAID are interrelated and mediated by COX-2 inhibition (1, 2, certain NSAID also inhibit enzymes belonging to the aldo-keto reductase 15, 16). However, several lines of evidence indicate that the thera- (AKR) family, including AKR1C3. We have demonstrated previously that peutic actions of NSAID may also include targets other than COX. overexpression of AKR1C3 acts to suppress cell differentiation and pro- For example, the doses of aspirin required to treat chronic inflamma- mote proliferation in myeloid cells. However, this enzyme has a broad tory diseases are greater than those required to inhibit COX. Also, tissue distribution and therefore represents a novel candidate for the aspirin derivatives that are not efficient COX inhibitors remain anti- target of the COX-independent antineoplastic actions of NSAID. Here we inflammatory (17–19). Similarly, the NSAID doses required to dem- report on the X-ray crystal structures of AKR1C3 complexed with the onstrate antineoplastic activities in vitro have almost always been NSAID indomethacin (1.8 Å resolution) or flufenamic acid (1.7 Å resolu- greater than those required for mere inhibition of either COX-1 or tion). One molecule of indomethacin is bound in the active site, whereas COX-2 (reviewed in Ref. 20). In addition, it has been demonstrated flufenamic acid binds to both the active site and the ␤-hairpin loop, at the opposite end of the central ␤-barrel. Two other crystal structures (1.20 that a spectrum of COX-2-selective and nonselective NSAID dis- and 2.1 Å resolution) show acetate bound in the active site occupying the played invariant antiproliferative and proapoptotic actions against proposed oxyanion hole. The data underline AKR1C3 as a COX-inde- transformed embryo fibroblasts, irrespective of whether the cells were Ϫ/Ϫ Ϫ/Ϫ Ϫ/Ϫ pendent target for NSAID and will provide a structural basis for the derived from wild-type, COX-1 , COX-2 , or COX-1 and Ϫ Ϫ future development of new cancer therapies with reduced COX-depend- COX-2 / mice (4). These observations combine to indicate strongly ent side effects. that a second non-COX target exists in the antineoplastic actions of NSAID. Besides inhibiting COX-1 and COX-2, NSAID also inhibit mem- INTRODUCTION bers of the NAD(P)H-dependent aldo-keto reductase (AKR) family, including the human enzyme AKR1C3, also known variously as Prostaglandins (PG) regulate diverse biological functions during 3␣-hydroxysteroid dehydrogenase type 2 (21), 17␤-hydroxysteroid both homeostasis and inflammation. A key step in the production of dehydrogenase type 5 (22), and PGD 11-ketoreductase (23). PG is the oxidation of arachidonate by cyclooxygenase (COX). Sep- 2 AKR1C3 has a broad tissue distribution, including tissues where arate genes encode for COX-1 and COX-2. COX-1 is constitutively NSAID have been shown to protect against cancer. In terms of K and expressed in diverse tissues and mediates homeostatic PG synthesis. m k /K , PGD represents a strong candidate for a physiologically In contrast, COX-2 expression is induced during inflammatory re- cat m 2 relevant in vivo substrate (23). AKR1C3 converts PGD to PGF ␣ sponses, creating elevated synthesis of PG that in turn drive aspects of 2 2 (23), which in the adipocyte model has been shown to block differ- the inflammatory response. Nonsteroidal anti-inflammatory drugs entiation by indirect antagonism of peroxisome proliferator-activated (NSAID) are drugs used to control inflammatory diseases and do so receptor (PPAR) ␥ (24). When not metabolized to PGF ␣, PGD is by inhibition of COX and, in particular, COX-2 activity. It has 2 2 nonenzymatically converted to PGJ2 and thence stepwise to 15- become generally accepted that NSAID also protect against progres- deoxy-⌬12,14-PGJ , a natural activating ligand for PPAR␥ (25). sion of gastrointestinal tumors, and there is increasing evidence that 2 Thus, the relative PGD2 11- ketoreductase activity of AKR1C3 is they may also protect against a variety of other cancers including likely to be a key determinant of PPAR␥ activity within cells. prostate carcinoma and, most recently, leukemia (1–7). NSAID have We and others have shown that the human acute myeloid leukemia also been shown to be antiproliferative against a broad spectrum of in cell lines HL-60 and KG1 express AKRIC3 (26, 27) and that treat- vivo and in vitro models of human malignancies, resulting in increased ment of HL-60 cells with AKR1C3 inhibitors [including indometha- apoptosis and/or cell differentiation (8–14). Together, these findings cin (IMN)] results in increased sensitivity to the antiproliferative and have led to the concept of cancer chemoprevention in individuals at prodifferentiative actions of all-trans-retinoic acid and 1␣25-dihy- risk. However, developing the full potential of NSAID-like drugs for droxyvitamin D3 (27, 28). Similarly, exposure of HL-60 cells to an

excess of PGD2 mimics the action of AKR1C3 inhibitors in promot- Received 9/9/03; revised 11/28/03; accepted 1/6/04. ing all-trans-retinoic acid-induced differentiation. Reciprocally, over- Grant support: A grant from the Leukemia Research Fund. The costs of publication of this article were defrayed in part by the payment of page expression of AKR1C3 in HL-60 cells increases resistance to all- ␣ charges. This article must therefore be hereby marked advertisement in accordance with trans-retinoic acid and 1 25-dihydroxyvitamin D3 (29). Importantly, 18 U.S.C. Section 1734 solely to indicate this fact. we recently observed that the capacity of PGD and the AKRIC3- Notes: A. L. Lovering was a recipient of a Ph.D. scholarship from the School of 2 Biosciences, University of Birmingham. Present address for A. L. Lovering: The Depart- inhibiting NSAID IMN to each promote the differentiation of HL-60 ment of Biochemistry, University of British Columbia, Vancouver, Canada. acute myeloid leukemia cells was negated by the PPAR␥ antagonist Requests for reprints: Scott A. White, The School of Biosciences, The University of Birmingham, Edgbaston, Birmingham, United Kingdom B15 2TT. Phone: 44-121-414- GW9662. Furthermore, a large body of recent work has determined 7534; Fax: 44-121-414-5925; E-mail: [email protected]. that PPAR␥ is a key regulator of proliferation differentiation and 1802

Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2004 American Association for Cancer Research. CRYSTAL STRUCTURES OF AKR1C3 apoptosis in diverse cells and that both synthetic and natural PPAR␥ Table 1 Comparison of the kinetic parameters for recombinant, His6-tagged AKR1C3 ligands exert antineoplastic activity in diverse in vitro and animal (this study) with those for recombinant, untagged AKR1C3 (Reference no.) models of neoplasia (for recent examples, see Refs. 30–34). We have His6-tagged AKR1C3 Recombinant AKR1C3 therefore proposed that AKR1C3 provides a plausible target for Km kcat Km kcat Reference Ϫ1 Ϫ1 non-COX-dependent antineoplastic activities activities of NSAID (␮M) (min ) (␮M) (min ) no. (29). 5␣DHTa reduction 7.2 Ϯ 0.3 0.12 19.8 Ϯ 7.2 0.26 47 The possibility that the chemoprotective actions of NSAID against Androsterone reduction 15.9 Ϯ 0.4 0.51 8.96 Ϯ 1.2 0.37 47 3␣-Androstanediol 32.0 Ϯ 2.0 0.47 27.2 Ϯ 4.2 0.15 47 cancer are mediated via a non-COX mechanism may provide new oxidation clinical in roads to the management of these diseases. At present, the 19.1 0.5 48 use of NSAID is limited by their COX-associated toxicities. It may a 5␣DHT, 5␣-dihydrotestosterone. therefore be beneficial to derive cancer drugs that are more directed against AKR1C3 and that limit these toxicities. Thus, to explore the mechanism of NSAID inhibition of AKR1C3 and enable the future Buster reagent (Novagen) containing Benzonase DNase (Novagen), using 10 development of non-COX-selective inhibitors with potential antitu- ml lysis reagent/400 ml original culture. After incubation at room temperature mor and other clinically beneficial effects, we have initiated a struc- with shaking (150 rpm) for 25–30 min, the cell debris was pelleted by ture-function study of recombinant AKR1C3 and report here on the centrifugation (48,000 ϫ g, 30 min, 4°C). The cleared lysate was mixed with high-resolution crystal structures of ternary complexes containing nickel-nitrilotriacetic acid His-bind resin (Ni-NTA; Novagen) using 0.5 ml ϩ resin/400 ml original culture, followed by gentle shaking at 4°C for 30 min. AKR1C3, NADP , and one of acetate, flufenamic acid (FLF), or IMN The mixture was loaded into an empty column, and unbound protein was (Fig. 1). removed by washing with 8 ml of 50 mM sodium phosphate buffer (pH 8.0) containing 300 mM NaCl. After an additional wash with 20 mM imidazole in MATERIALS AND METHODS the same buffer, the recombinant protein was eluted in 1.5 column volumes of 100 mM imidazole. After the addition of DTT to a concentration of 2 mM, the Materials. Enzyme substrates, cofactors, and chemicals were obtained protein was further purified by fast protein liquid chromatography gel filtration from Sigma-Aldrich (Dorset, United Kingdom) unless otherwise stated. on a Superdex 200 HR10/30 column using 10 mM potassium phosphate, 1 mM Ϫ Oligonucleotide primers were made by Alta Bioscience (Birmingham, EDTA, and 1 mM DTT (pH 7.0) at a flow rate of 0.4 ml⅐min 1. Protein was United Kingdom). Restriction endonucleases were obtained from Promega monitored by absorbance at 280 nm, and purity of the fractions (1 ml) was ϩ (Southampton, United Kingdom). assessed by SDS-PAGE. NADP was added to a final concentration of 2 mM, Production of AKR1C3 Expression Vector. The AKR1C3 open reading and the protein was concentrated by centrifugation (3,000 ϫ g,4°C) on a frame was amplified from a cDNA clone (KIAA0119; Kazusa DNA Research Vivaspin concentrator (Mr 30,000 cutoff; Vivascience) to varying concentra- Institute, Chiba, Japan) by PCR using primers based on the 5Ј end (5Ј- tions between 15 and 45 mg⅐mlϪ1. Approximately 20 mg of highly pure, CAGCATATGGATTCCAAACAGCAG-3Ј) and the 3Ј end (5Ј-CAGCTC- His-tagged recombinant protein suitable for crystallography could be purified GAGATATTCATCTGAATATGGTAT-3Ј) that introduced an NdeI site that from 1 liter of cell culture. overlapped the start codon and a downstream XhoI cloning site (excluding the Measurement of Steady-State Kinetic Parameters. Enzyme activity was stop codon). The resulting PCR fragment was cloned into pGEMT-Easy vector measured as either reduction of 5␣-dihydrotestosterone or androsterone or (Promega). After digestion with NdeI and XhoI, the fragment was gel purified oxidation of 3␣-androstanediol. Reduction reactions were monitored in 1-ml -and ligated into pET21b(؉) (Novagen), yielding pET21b-AKR1C3. The di- volumes containing 5–100 ␮M 5␣-dihydrotestosterone or 2–50 ␮M androste rection and nucleotide sequence of the cloned AKR1C3 gene in the pET21b rone in 3% (v/v) acetonitrile, 150 ␮M NADPH, and 10 mM potassium phos- vector were confirmed by DNA sequencing. The plasmid pET21b-AKR1C3 phate buffer (pH 7.0) at 30°C. Initial velocities were measured by observing was designed to introduce a COOH-terminal His6 tag to aid purification. the rate of change of absorbance of pyridine nucleotide at 340 nm (⑀ 6270 Ϫ1⅐ Ϫ1 Expression and Purification of Recombinant AKR1C3. Overnight cul- M cm ) in 1 ml, with a 1-cm light path. Calculation of kcat and Km values tures of Escherichia coli BL21(DE3) cells (Novagen) expressing the pET21b- used the Leonora program (35), yielding estimates of the kinetic constants and Ϫ AKR1C3 construct in LB medium containing 100 ␮g⅐ml 1 ampicillin were their associated SEs (Table 1). For inhibition studies, IMN and FLF were seeded (0.1%) into 400-ml batches of fresh LB medium containing ampicillin added at several different concentrations to give final concentrations varying and then incubated at 37°C for 16 h, with shaking at 220 rpm. Expression between 0.5 and 50 ␮M, and steroid substrate concentrations were then varied of the protein occurred without the need for induction with isopropyl-␤-D- as indicated above. The type of inhibition and the inhibition constants were thiogalactopyranoside. Bacterial cultures were pelleted by centrifugation calculated using the Leonora program. All reactions were initiated by the (3,800 ϫ g, 15 min), and the cells were disrupted by resuspension in Bug- addition of enzyme. A fluorescence assay was used to determine the kinetic constants for the oxidation of 3␣-androstanediol. The fluorescence emission of NADPH at 450 nm (5-nm slit width) with excitation at 340 nm (15-nm slit width) was monitored at 30°C. Each cuvette contained a 3-ml reaction mixture with 10 mM ϩ potassium phosphate buffer (pH 7.0), 200 ␮M NADP , and 3␣-androstanediol (1–40 ␮M) with 3% DMSO as solvent. A standard curve was constructed by monitoring fluorescence changes with incremental additions of NADPH. FLF and IMN were used in inhibition studies, and data were analyzed as described above. Crystallization and Data Collection. Crystals were grown by the hang- ing-drop vapor-diffusion method in 6-␮l drops. Form I crystals were obtained from a 1:1 mixture of purified protein [16 mg⅐mlϪ1, in a buffer containing 10 mM potassium phosphate buffer (pH 7.0), 1 mM DTT, 1 mM ϩ EDTA, and 2 mM NADP ] and a reservoir solution containing 25% (w/v) Fig. 1. Chemical structures of the following nonsteroidal anti-inflammatory drug inhibitors: A, flufenamic acid; and B, indomethacin. The numbering scheme has been polyethylene glycol 4000 (Fluka), 100 mM sodium citrate (pH 6.0), 2.5% adopted from previous Protein Data Bank entries. As drawn, the chlorobenzoyl group of (v/v) 2-methyl-2,4-pentanediol, and 800 mM ammonium acetate. Crystals indomethacin is defined as being in the cis conformation with respect to the indole ring. were soaked in artificial mother liquor supplemented with 15% (v/v) The trans conformation would have the chlorobenzoyl group pointing up to the right 2-methyl-2,4-pentanediol shortly before flash cooling to 100 K in a cryo- (adopted from Ref. 51). Due to steric clashing between positions 5 and 11, the two ring ϫ ϫ systems cannot be in the same plane when indomethacin is in the cis conformation (see stream. Form I crystals were approximately 0.6 0.3 0.2 mm, taking 6 main text). days to grow to maximum size, and diffracted beyond 1.1 Å resolution. A 1803

Table 2 X-ray data collection and refinement statistics ϩ ϩ ϩ ϩ Acetate, NADP Flufenamic acid, NADP Indomethacin, NADP Acetate, NADP Ligands Space group P212121 P212121 P212121 P3221 Crystal form I I I II Cell parameters (Å, °) 55.81 ϫ 63.07 ϫ 96.20 55.80 ϫ 63.00 ϫ 96.29 55.68 ϫ 63.12 ϫ 96.11 58.06 ϫ 58.06 ϫ 206.32 ␣, ␤, ␥ ϭ 90 ␣, ␤, ␥ ϭ 90 ␣, ␤, ␥ ϭ 90 ␣, ␤, ␥ ϭ 120 Data collection statisticsa Resolution (Å)38–1.2 (1.26–1.20) 48–1.8 (1.90–1.80) 20–1.7 (1.79–1.70) 51–2.1 (2.21–2.10) Completeness (%) 98.5 (90.1) 99.7 (98.1) 98.0 (88.3) 94.7 (74.6) Multiplicity 5.0 (2.6) 4.4 (2.8) 3.4 (2.7) 10.6 (1.9) I/␴I 16.4 (4.1) 13.6 (3.8) 18.7 (4.7) 22.0 (2.6) No. of observations 522,433 (36,225) 141,396 (12,414) 127,566 (12,797) 243,584 (4,813) No. of unique reflections 105,032 (13,800) 32,084 (4,459) 37,296 (4,785) 23,069 (2,550) b Rsym 5.9 (24.2) 7.8 (25.6) 4.9 (20.0) 8.7 (22.0) Refinement statistics Wilson B/average Bc 9.3/13.5 17.7/19.2 16.54/23.7 23.14/22.68 No. of non-H atoms/waters 3075/434 2868/236 2964/363 2889/296 rmsdd bonds/angles 0.006/1.3 0.007/1.2 0.009/1.6 0.009/1.2 Ramachandrane 260/20/0/1 263/16/1/1 262/18/0/1 259/22/0/1 f R/Rfree 14.6/15.6 16.8/20.5 17.2/19.8 15.8/19.6 a Values in parentheses are for the outer shell. b ϭ͚ ʈ ͉ Ϫ ͉ ʈ ͚ ͉ ͉ Rsym hkl Fo Fc / hkl Fo . c Wilson B factor, as reported by Truncate (36), and average B factor for all atoms following calculation of “total” B values using TLSANL (42). d Root mean square deviation from ideal values. e Values represent the number of nonglycine, nonproline amino acids in the core, allowed, generously allowed, and forbidden regions of the Ramachandran phi-psi plot (56), according to the program Procheck (45). f 5% of the data were kept aside for cross-validation and not used for refinement purposes. The same Rfree set was kept for each of the form I data sets. complete data set was collected to 1.20 Å on beam line ID14-1 (␭ ϭ 0.93 0.64 and a packing coefficient of 0.53. The structure was initially refined Å) at the European Synchrotron Radiation Facility using an Area Detector using the rigid body, simulated annealing, and geometry minimization Systems Corporation (ADSC) Quantum 4 CCD detector. Form II crystals protocols against a maximum likelihood target with the program CNS were grown in similar conditions, except that only 400 mM ammonium before modeling in amino acid differences. This refined structure was then acetate was used. Crystals grew to a maximum size of 0.4 ϫ 0.3 ϫ 0.2 mm used to solve acetate complex I (crystal form I, native) using CNS molec- ϳ over 4 days and diffracted to 2.0 Å resolution. A complete data set was ular replacement. Crystals soaked with FLF or IMN were isomorphous to collected to 2.1 Å resolution on beam line ID29 (␭ ϭ 0.93 Å)atthe the native crystals (Table 2). 2mFo-DFc SIGMAA-weighted (40) electron European Synchrotron Radiation Facility using an ADSC Q210 detector. density maps were calculated after partial refinement of the native structure For the FLF and IMN inhibitor soaks, crystals were placed into a cryobuffer against structure factor amplitudes for the soaked crystals to check for containing 25% (w/v) polyethylene glycol 4000, 100 mM sodium citrate incorporation of the inhibitor. The structures were initially refined using (pH 6.0), 10% (w/v) ethylene glycol, 10% (w/v) DMSO, 800 mM NaCl, and CNS against a maximum likelihood target and later refined using Refmac5 5mM of either FLF or IMN, for a period of 20 min before flash cooling in (41) incorporating maximum likelihood and TLS refinement (42). Molec- the cryostream. Data were collected on beam line ID14-2 (␭ ϭ 0.93 Å)at ular models were inspected, and manual adjustments were made using the the European Synchrotron Radiation Facility using an ADSC detector. All diffraction data were integrated and scaled using either Mosflm/Scala (36) graphics package TURBO-FRODO (43) using 2mFo-DFc and mFo-DFc or Denzo/Scalepack (Ref. 37; Table 2). SIGMAA-weighted electron density maps as a guide. Correct modeling of Structure Determination. Acetate complex II (crystal form II) was ligands was checked using CNS simulated annealing omit maps (44). solved by the molecular replacement method as implemented in CNS (38) Finally, all models were validated using the program Procheck (45). The using chain A of the AKR1C2 structure [Ref. 39; Protein Data Bank (PDB) final statistics are given in Table 2. Structure superimpositions were entry 1IHI; Table 3; Fig. 5] as a search model. A clear solution (top peak calculated using the “magic fit” and “improve fit” features of the program in both rotation and translation functions) was seen with a monitor value of Swiss PDB Viewer (46).

Table 3 Structural comparison with closest homologs and selected AKR members AKR Sequence identity Resolution of AKR members classificationa rmsd (Å)b (%) PDB IDc structure (Å) Reference no. Human 3␣ HSDd type II 1C3 100 Acetate complex I Reference structure 1S1P 1.20 This work Flufenamic acid complex 0.22 (315)/0.63 (2524) 1S2C 1.8 This work Indomethacin complex 0.25 (315)/0.75 (2524) 1S2A 1.7 This work Acetate complex II 0.24 (315)/0.64 (2524) 1S1R 2.1 This work Human 3␣ HSD type III 1C2 87 ϩ Ternary complex: NADP & ursodeoxycholate 0.66 (303) 1IHI:A 3.0 39 ϩ Quaternary complex: NADP , testosterone, & acetate 0.63 (300) 1J96:A 1.25 52 Rat 3␣ HSD 1C9 69 Polypeptide only 1.02 (280) 1RAL 3.0 57 ϩ Binary complex: NADP 0.74 (279) 1LW1:B 2.7 58 ϩ Ternary complex: NADP , testosterone 0.88 (301) 1AFS:A 2.5 59 Porcine aldose reductase 1B6 49 60 Ternary complex: NADPH & sorbinil 1.29 (296) 1AH0 2.3 Human aldose reductase 1B1 48 ϩ Binary complex: NADP 0.92 (291) 1ADS 1.65 61 a Classified according to AKR nomenclature (62). b Root mean square deviation from the reference structure (AKR1C3 acetate complex I) calculated using C␣ atoms (and all atoms for the four structures reported here). The number of atom pairs used is shown in parentheses. c When more than one molecule per asymmetric unit exists, the molecule giving the lowest rmsd value is used, and the chain identifier is given. PDB, Protein Data Bank. d HSD, hydroxysteroid dehydrogenase. 1804

Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2004 American Association for Cancer Research. CRYSTAL STRUCTURES OF AKR1C3 ␣ ␤ RESULTS As expected from sequence homology, AKR1C3 folds as an 8 8

superbarrel, capped at the NH2-terminal end with an antiparallel Kinetic Characterization of the Recombinant Protein. Because hairpin loop (Figs. 2, 4D, and 5). At the COOH-terminal end of the our study used His-tagged and not wild-type protein, we first wished barrel, the loops connecting each ␤-strand to the following ␣-helix to determine that the tag did not alter the biochemical properties of contribute amino acid side-chains into the active site (Fig. 2). Two of AKR1C3. We therefore compared the properties of our tagged en- these loops are particularly extended: amino acids 116–143 (loop A); zyme with those reported previously for nontagged AKR1C3. These and amino acids 217–238 (loop B). Two additional helices, H1 and experiments used oxidation of 3␣-androstanediol and reduction of H2, precede helix ␣7 and follow helix ␣8, respectively. Following androsterone. These substrates were chosen in preference to PGD for 2 helix H2, the COOH-terminal tail, also known as loop C, forms an the pragmatic reasons that they and their products are more chemi- extended, meandering coil, occupying the space near to the active site cally stable and that the kinetics of the enzyme have been more between loops A and B and forming interactions with both. extensively analyzed in the past with these substrates. These tests The Active Site. The active site (Fig. 2) is located at the COOH- confirmed that the ability of His -tagged recombinant AKR1C3 pro- 6 terminal end of the central ␤-barrel. The NADPϩ is in an extended tein to reduce 5␣-dihydrotestosterone or androsterone and oxidize conformation, so that atom C2 of the nicotinamide ring is positioned 3␣-androstanediol yielded kinetic constants similar to those published above the center of the barrel with the A side (as defined by Ref. 49) previously (47, 48) for untagged recombinant protein (Table 1). facing the barrel at an angle of ϳ45° to the barrel axis and stacked Inhibition by the NSAID FLF and IMN. When tested in the ␤ oxidative direction with varying concentrations of 3␣-androstanediol, against the side-chain of Tyr216 from strand 7 (Figs. 2 and 4A). The both FLF and IMN displayed potent competitive inhibition against the B face of the nicotinamide ring is exposed to the active site, with the C4-pro-R position available for hydride transfer. The NADPϩ diphos- steroid substrate with K values of 0.14 Ϯ 0.01 and 0.27 Ϯ 0.01 ␮M, i ␤ respectively. However, when tested in the reductive direction against phate moiety straddles a gap formed between the ends of -strands 7 varying concentrations of androsterone, the inhibitors were less po- and 8 and forms H-bonds to Ser217N, Ser217OG, Leu219N, Ser221N, Ј tent, and the data for both showed a better fit to a mixed type of Gln222NE2, and Lys270N. The 2 -phosphate group is salt bridged to Lys270 and Arg276 and H-bonded to Ser271OG and Tyr272N, pre- inhibition. The Ki values corresponding to the competitive components were calculated to be 3.1 Ϯ 0.5 and 2.1 Ϯ 0.4 ␮M for FLF and sumably to favor NADP(H) over NAD(H), thereby avoiding a costly IMN, respectively, with uncompetitive components of 4.4 Ϯ 0.3 and transhydrogenation cycle between the NAD(H) and NADP(H) pools. 4.6 Ϯ 0.5 ␮M. The adenine ring is sandwiched between the hydrophobic part of the The Three-Dimensional Structure of AKR1C3. Two crystal Arg276 side-chain (atoms CG and CD) on one side and the aliphatic forms were obtained from purified protein, differing only in the side-chains of Leu219, Leu236, and Ala253 on the other, while concentration of ammonium acetate required for crystal growth. De- forming several H-bonds around the ring edge. The substrate-binding spite the similarity in growth conditions, crystal form I diffracts to site is located on the other side of the nicotinamide ring from the significantly higher resolution (Table 2), and consequently, most of central ␤-barrel and consists mainly of hydrophobic, aromatic amino our analysis and studies are concentrated on this form. The crystal acid side-chains. In both acetate complexes, the acetate molecule is in form I structure (Fig. 2) consists of amino acids 6 to 320, NADPϩ,1 an identical position, close to and in an approximate stacking arrange- molecule each of acetate and 2-methyl-2,4-pentanediol, and 434 water ment with the nicotinamide ring (Fig. 4A). The methyl group of the molecules and represents an enzyme:NADPϩ:acetate ternary complex acetate molecule points into a hydrophobic pocket comprising Tyr24, (Figs. 3A and 4A), hereafter referred to as acetate complex I. The Tyr55, Leu54, Trp227, and Phe306. The carboxylate carbon is 3.17 Å refined crystal form II structure (acetate complex II) consists of amino away from the nicotinamide C4 position on the pro-R side. Oxygen acids 6 to 321, NADPϩ, one molecule each of acetate and 2-methyl- OXT of the acetate is H-bonded to His117NE2 (2.93 Å) and Tyr55OH 2,4-pentanediol, and 286 water molecules. Overall, the electron den- (2.47 Å), which is itself H-bonded to Lys84NZ (2.89 Å). Lys84NZ is sity maps for both models were extremely clear and well defined. The H-bonded to Asp50OD2 (2.66 Å) and Ser51O (2.87 Å). The four

COOH-terminal His6 tag was disordered in both models and com- amino acids Asp50, Tyr55, Lys84, and His117 are strongly conserved pletely absent in the electron density maps. There was also some weak across the AKR family and have been proposed to form a catalytic electron density in the hydrophobic, substrate-entry channel, indicat- tetrad that catalyzes the oxidation of alcohol or reduction of ketone ing disordered solvent molecules that could not be modeled in acetate functional groups via a “push-pull mechanism” (50). Briefly, in the complexes I and II. reduction of carbonyls, it is proposed that the hydride ion equivalent

Fig. 2. Wall-eyed stereo view of AKR1C3 shown in a ribbon representation. The ␣-helices are red, the ␤-strands are yellow. The secondary structure element nomenclature has been adopted from Ref. 61. NADPϩ is shown in ball-and-stick form with oxygens colored red, nitrogens colored blue, carbons colored gray, and phosphorous colored orange. Superimposed in the active site is acetate (red), flufenamic acid (FLF) 1 (green), and indomethacin (blue). FLF2 (purple) is shown at the proposed FLF2-binding site (see Fig. 4D). The diagram was prepared using Molscript (63).

1805

Fig. 3. SigmaA-weighted electron density maps for (A) acetate and the NADPϩ nicotinamide head group from acetate complex I at 1.20 Å resolution, (B) flufenamic acid (FLF) 1 and (C) FLF2 from the FLF ternary complex structure at 1.8 Å resolution, and (D) indomethacin and solvent peak 1 from the indomethacin ternary complex at 1.7 Å resolution. In each case, the final refined 2Fo-Fc map [colored red, contoured at either 1 ␴ (A, B, and D)or0.8␴ (C), and using all reflection data between ϱ and the given resolution] is superimposed with the bias-free simulated annealing omit map (colored in blue) calculated using the established protocol in CNS (38) with an initial annealing temperature of 2000 K (44). The atoms are colored as follows: oxygen, red; nitrogen, blue; and carbon, gray. The diagram was prepared using Bobscript (64).

on the pro-R C4 position of the NADPH nicotinamide attacks the displaced by 1.2 Å toward the inhibitor and an ϳ100° rotation about substrate carbonyl to form a transitory tetrahedral oxyanion, which its CA–CB bond to allow the aromatic side-chain to interact better abstracts the proton from Tyr55OH to form tyrosinate-55 anion and with the trifluoromethyl-benzene ring. The indole ring of Trp227 is the product alcohol. The tyrosinate anion is stabilized by the adjacent, also tilted by ϳ15° to enlarge the active site on binding of the positively charged Lys84NZ (50). inhibitor. The dihedral angle between the two rings of FLF1, defined The acetate complexes I and II are structurally very similar over the as C7-C6-C1Ј-C7Ј is Ϫ113° (Figs. 1A and 4B). The majority of entire protein fold (Table 3), despite different crystal packing arrange- interactions between FLF1 and its nearest neighbors are van der Waal ments. When comparing the rotamer angles of acetate complexes I contacts, but there is a bifurcated H-bond from FLF1N to FLF1O1 and II, only 38 of a total of 321 amino acids have a difference in chi1 (2.85 Å, predicted D–H:A angle ϭ 123°) and from FLF1N to the angle of more than 20°. amide oxygen of the nicotinamide ring (2.91 Å, predicted D–H:A NSAID Complexes. The ternary complexes with either FLF or angle ϭ 134°). In addition, there is an H-bond between FLF1F1 and IMN bound to AKR1C3:NADPϩ were obtained from soaking exper- Tyr216OH (2.97 Å). iments using form I crystals after replacing the ammonium acetate in At the second FLF-binding site, there is clear electron density to the crystal stabilization solution (artificial mother liquor) with sodium position the two aromatic rings and also the carboxylic acid group of chloride. The FLF ternary complex consists of amino acids 6 to 320, FLF2. The electron density for the trifluoromethyl group is very weak NADPϩ, 2 molecules of FLF, and 237 water molecules, whereas the and poorly defined, suggesting that FLF2 is bound with a lower IMN ternary complex consists of amino acids 6 to 320, NADPϩ, occupancy and possibly in two alternative conformations, with the IMN, DMSO, and 363 water molecules. Strong and clearly defined trifluoromethyl group either cis or trans to the carboxylic acid with electron density in the active site region allowed the unambiguous dihedral angles (see above) of 48° or Ϫ133°, respectively. Fig. 4D fitting of a molecule of either FLF (FLF1) or IMN (Fig. 3), interacting shows FLF2 in its hydrophobic binding pocket: ring 1 (C1-C6) with amino acid side-chains from ␤-strands 4, 5, 6, and 7, as well as interacts with the side-chains of Val8, Val18, Leu261, Gly264, and loops B and C (Fig. 4, B and C). Of the 17 amino acid side-chains with Val266; ring 2 (C1Ј--C6Ј) packs against Gln6, atoms CB to CD of at least one atom within 4.0 Å of FLF1 and/or IMN (marked in Fig. Arg258, Gln262, and Phe284. Arg301 from a crystallographic sym- 5), 8 are aromatic, and 9 are invariant among AKR1C1–AKR1C4. metry-related molecule is positioned close to FLF2 ring 1, so that the During the latter stages of refinement, it became apparent that a plane of the Arg301 guanidinium group is parallel to and approxi- second molecule of FLF (FLF2) was bound next to the ␤-hairpin loop mately 3.5 Å away from the plane of the FLF2 carboxylate group. at the NH2 terminus (Figs. 3C and 4D). With IMN, the chlorobenzoyl group is in the cis conformation, with In the active site, FLF1 binds next to the nicotinamide ring, with the respect to the indole ring, as defined earlier (see Fig. 1B). The planes carboxylate group occupying a similar position to the acetate group in of the two ring systems are almost perpendicular to each other. With acetate complexes I and II, and with a similar H-bonding pattern (Fig. the indole ring positioned in the plane of Fig. 1, the chlorobenzene 4, A and B). There is little perturbation of the active site structure on ring points toward the viewer, with a C*– N– C9– C10 dihedral angle binding FLF1. Phe311 (loop C) is affected the most, with its CA atom of 44° [cf. –59.16° in the IMN:COX-2 complex, PDB entry 4cox 1806

Fig. 4. AϪC, wall-eyed stereo views of AKR1C3 active site. Only the nicotinamide, ribose, and diphosphate moieties of NADPϩ are shown. A, acetate complex I. Atoms are colored as follows: carbon, gray; oxygen, red; nitrogen, blue; and phosphorous, pink. Carbon atoms in the ligand are colored green. Water 36 has been shown as a red sphere. Atom OXT of the acetate is H- bonded to Tyr55OH and His117NE2. Also shown, but not labeled for clarity, is Tyr216, which is behind the nicotinamide ring of NADPϩ. B, flufenamic acid (FLF) 1 from the FLF ternary complex. Atoms are colored as described in A. Water 150 is shown as a red sphere. C, indomethacin from the indomethacin ternary complex. Solvent peak 1 is shown as a red sphere. D, the FLF2-binding pocket with two alternative conformations of FLF2 shown in green (see also Fig. 2). The polypeptide backbone atoms have been indicated and are colored according to secondary structure type: ␣-helix, red; ␤-strand, yellow; and coil, gray. The diagram was prepared using Molscript (63).

(51)]. There are few H-bonds between IMN and the enzyme:NADPϩ However, there are important differences between AKR1C3 and the complex; the remaining contacts are van der Waal interactions (Fig. other AKR1C family members in the active site and in loop A, which 4C). The indole ring, which has forced the side-chain of Phe306 to must be responsible for the differences in activities (47). Although rotate 120° about the CA-CB bond, is approximately perpendicular to AKR1C3 is structurally most similar to AKR1C2, loop A of AKR1C3 ϩ the NADP nicotinamide ring, with the shortest distance (ϳ3.3 Å) is shifted toward loop C (Asn134CA to Asn134CA distance of 6.2 Å between atom C4 of the nicotinamide and IMNC5. The bridging in the superimposed structures of PDB:1J96_A and acetate complex I) carbonyl is at an angle of ϳ40° to the plane of the nicotinamide ring, and is structurally more similar to loop A of aldose reductase (PDB with the C and O atoms 4.0 and 4.2 Å away from the nicotinamide C4 entries 1AH0 and 1AH4). atom, respectively. The carbonyl oxygen is too far away from Acetate in the Active Site. The electron density in the active site Tyr55OH to H-bond directly. Instead, there is an unidentifiable sol- of acetate complex I clearly shows a molecule of acetate, used as an vent peak (Figs. 3D and 4C) occupying the same position as the additive during the crystallization process. The presence of acetate in acetate OXT (Fig. 4A), and it is 2.0, 2.6, and 2.8 Å from the IMN the protein drop significantly improved the growth and quality of carbonyl oxygen, Tyr55OH, and His117NE2, respectively. The car- crystals. Analysis of all available AKR structures in the PDB databank boxylate group points toward and interacts with the oxygen atoms ϩ is particularly revealing; eight structures, representing nine polypep- NO1 and NO2 from the nicotinamide half of the NADP diphosphate tide chains, have a carboxylic acid-containing compound bound in the moiety, forming three H-bonds, and an additional H-bond is formed active site. In all of them, the carboxylate group is bound next to the between IMNO2 and Gln222N. conserved tyrosine and histidine residues in the active site. There are three additional chains with bound acetate in the active site. In one DISCUSSION example, PDB entry 1J96, an acetate molecule has bound in the active ␣ ␤ site in preference to testosterone, an inhibitor, which is also present AKR1C3 Structure. AKR proteins have a well-established 8 8 architecture, where differences in the active site loops determine and bound in the active site entrance channel (52). When superim- specificity of substrate and inhibitor binding. The four structures posing all of the protein structures with an active site-bound carbox- presented here, the first of AKR1C3, are very similar to each other ylate, it can be seen that the carboxylate adopts one of two possible (Table 3) and homologous to other members of the AKR family, orientations. Seven of the structures (from a total of 12) have carboxy- particularly the closely related AKR1C family members (Fig. 5). late groups superimposing with the acetates in the two acetate com- 1807

Fig. 5. A Clustal-W sequence alignment (65) of AKR1C1–AKR1C4. The sequence is shown in single-letter amino acid code for AKR1C3. Only nonidentical amino acids are indicated in the other three sequences. There are no gaps or insertions. The secondary structure elements are indicated with shading and labeled. AKR1C3 shares 88%, 87%, and 84% identities with AKR1C1, AKR1C2, and AKR1C4, respectively. An updated multi- sequence alignment of the entire AKR family is kept at the AKR homepage (http://www.med. upenn.edu/akr/; Ref. 62).

plexes presented here. However, in all of the structures, one of the peak, which is only 2.0 Å from IMNO1 and 2.6 and 2.8 Å from carboxylate oxygens is H-bonded to both atoms OH of the active site Tyr55OH and His117NE2, respectively. To test whether solvent peak tyrosine and NE2 of the histidine. In our acetate complex I structure, 1 was actually covalently linked to the carbonyl oxygen, we modeled one of the acetate oxygens, labeled OXT, is H-bonded to Tyr55OH in a hydroperoxide derivative of IMN and subjected this to several with a short distance of 2.47 Å. An oxyanion hole, analogous to that cycles of test refinement using the program Refmac5 (41). The re- in serine proteases, has been proposed for the AKR family, but instead sultant Fo-Fc difference electron density clearly showed a negative of main-chain amides stabilizing the build up of charge on the oxygen, peak between IMNO1 and solvent peak 1. Also, there were no peaks the hole is proposed to consist of the conserved active site tyrosine in an anomalous (Fϩ Ϫ FϪ) map corresponding to this position. In the (Tyr55), histidine (His117), and nicotinamide ring. Analysis of the absence of any further data, we cannot identify the origin of this ligand-bound AKR structures deposited in the PDB reveals five solvent peak. The FLF carboxylate group has a lower pKa (3.9 in free structures with short H-bonds (Յ2.60 Å) to atom OH of the active site solution) than IMN (4.5 in free solution) and therefore is more likely tyrosine and three structures with short H-bonds between the carbox- to be charged, even in the solvent-protected active site. ylate oxygen and atom NE2 of the active site histidine. In addition, In this study, we observed a 10–20-fold increase in the values of Ki there is one structure of an inhibitor-bound complex (PDB entry for the inhibitors when inhibiting reductive compared with oxidative 1AFS; Table 3) where the carbonyl oxygen (atom O3) of testosterone reactions. Both drugs were potent competitive inhibitors against 3␣- is Ͻ2.6 Å from the active site tyrosine OH group in both of the androstanediol in the oxidative direction, in agreement with the drugs non-crystallographic symmetry (NCS)-related structures (labeled binding to the active site of the enzyme:NADPϩ complex as observed chains A and B). It should be noted that the accuracy of bond lengths in the crystal structures. However, the drugs were less potent inhibi- and interatomic distances is very dependent on the resolution of the tors of the reduction of androsterone, suggesting that the affinity of the X-ray structure. Most of the AKR X-ray structures deposited in the drugs for the enzyme:NADPH complex may be lower than that for the Ն ϩ PDB have a resolution lower than 2.0 Å (dmin 2.0 Å). However, at enzyme:NADP complex. The inhibition was clearly of a mixed type, a resolution of 1.20 Å in the acetate complex I, the overall estimated suggestive of a second inhibitor-binding site on the enzyme. Interest- coordinate error is estimated at Ϯ0.037, Ϯ0.034, or Ϯ0.028 Å, based ingly, equilibrium dialysis studies on the binding of NSAID to the on R value, free R value, or maximum likelihood, respectively. related AKR1C9 from rat liver (53) revealed the presence of two NSAID Inhibition. Analysis of the two NSAID ternary crystal complexes, a high-affinity ternary complex corresponding to enzyme: ϩ ϭ ␮ structures shows that FLF and IMN bind at the active site. The NAD :IMN (Kd 1–2 M for IMN) and a low-affinity binary ϭ ␮ carboxylate of FLF1 occupies the acetate-binding site, as expected, complex corresponding to enzyme:IMN (Kd 22 M). It seems likely with oxygen atom O1 occupying the proposed oxyanion hole (Figs. 4, that a similar effect is being seen in these studies, and this is supported A and B). The H-bond between Tyr55OH and atom FLF1O1 is also by the presence of a second FLF molecule observed next to the short (2.6 Å). The other end of FLF1 H-bonds via the atom F1 to ␤-hairpin loop at the bottom of the TIM barrel in the crystal structure Tyr216OH, which ring stacks with the NADPϩ nicotinamide ring and (Figs. 2 and 4D). The higher average B-factor for FLF2, compared thus forms a “molecular clamp” around the nicotinamide. IMN also with those for FLF1 and the protein, indicates that this binding site has contains a carboxylate, but surprisingly, this does not bind in the lower affinity for FLF. However, no second molecule of IMN could acetate-binding pocket (Fig. 4C). Instead, it H-bonds to atoms NO1 be observed in the corresponding hydrophobic pocket in the and NO2 of the NADPϩ diphosphate moiety, indicating that the AKR1C3:IMN complex structure. carboxylate group exists predominantly in the neutral, protonated The kinetic data also suggest that it is possible that the conforma- form. The IMN carbonyl group (atoms C9 and O1) is pointing in tion of IMN and FLF1 in the active site of AKR1C3 is dependent on toward the oxyanion hole. The hole itself is occupied by a solvent the oxidation state of the nucleotide. We have attempted to model 1808

IMN in alternative conformations with the carboxylate group super- (NSAIDs) on cyclooxygenase-null embryo fibroblasts. J. Exp. Med., 190: 451–459, imposed on the acetate of acetate complex I (data not shown). In these 1999. 5. Wechter, W. J., Kantoci, D., Murray, E. D., Quiggle, D. D., Leipold, D. D., Gibson, models, steric hindrance exists between IMN and the aromatic side- K. M., and McCracken, J. D. R-Flurbiprofen chemoprevention and treatment of chains that make up the active site, although these clashes could intestinal adenomas in the apc(min)ϩ mouse model: implications for prophylaxis and treatment of colon cancer. Cancer Res., 57: 4316–4324, 1997. potentially be avoided by displacement of the side-chains. Due to the 6. Wechter, W. J., Leipold, D. D., Murray, E. D., Quiggle, D., McCracken, J. D., plasticity of the active site side-chains [observed in this study and Barrios, R. S., and Greenberg, N. M. E-7869 (R-flurbiprofen) inhibits progression of proposed by Penning et al. (47)], we cannot rule out other binding prostate cancer in the tramp mouse. Cancer Res., 60: 2203–2208, 2000. 7. Kasum, C. M., Blair, C. K., Folsom, A. R., and Ross, J. A. Non-steroidal anti- modes of IMN. Additional studies are required. inflammatory drug use and risk of adult leukemia. Cancer Epidemiol. Biomark. Prev., A Comparison of IMN Inhibition of AKR1C3 and COX. 12: 534–537, 2003. NSAID inhibitors are better known for their ability to inhibit COX. It 8. Yamazaki, R., Kusunoki, N., Matsuzaki, T., Hashimoto, S., and Kawai, S. Nonste- roidal anti-inflammatory drugs induce apoptosis in association with activation of is interesting to compare the IMN ternary complex with the structures peroxisome proliferator-activated receptor ␥ in rheumatoid synovial cells. J. Pharma- of IMN-bound COX-2, determined to 2.9 Å resolution (51). As with col. Exp. Ther., 302: 18–25, 2002. AKR1C3, IMN is bound to COX-2 at the bottom of a hydrophobic 9. Liu, J. J., Wang, J. Y., Hertervig, E., Nilsson, A., and Duan, R. D. Sulindac induces apoptosis, inhibits proliferation and activates caspase-3 in HEP G2 cells. Anticancer channel and blocks substrate access to the active site. The carboxylate Res., 22: 263–266, 2002. group salt bridges to a conserved Arg120 side-chain. The bound IMN 10. Rodriguez-Burford, C., Barnes, M. N., Oelschlager, D. K., Myers, R. B., Talley, L. I., is also in a cis conformation, as defined earlier, but the chlorobenzoyl Partridge, E. E., and Grizzle, W. E. Effects of nonsteroidal anti-inflammatory agents ϳ (NSAIDs) on ovarian carcinoma cell lines: preclinical evaluation of NSAIDs as group is rotated about the N-C9 bond by 100° with respect to chemopreventive agents. Clin. Cancer Res., 8: 202–209, 2002. IMN-bound AKR1C3, so that with the indole ring in the plane of the 11. Mohammed, S. I., Bennett, P. F., Craig, B. A., Glickman, N. W., Mutsaers, A. J., paper (Fig. 1B), it is pointing away from the viewer. Structures also Snyder, P. W., Widmer, W. R., DeGortari, A. E., Bonney, P. L., and Knapp, D. W. Effects of the cyclooxygenase inhibitor, piroxicam, on tumor response, apoptosis, and exist for iodo-IMN-bound COX-1 (54), but due to the limited reso- angiogenesis in a canine model of human invasive urinary bladder cancer. Cancer lution of that study (4.5 Å), the authors could not unambiguously Res., 62: 356–358, 2002. position IMN into electron density, and so two models, cis and trans, 12. Thurnher, D., Bakroeva, M., Schutz, G., Pelzmann, M., Formanek, M., Knerer, B., and Kornfehl, J. Non-steroidal anti-inflammatory drugs induce apoptosis in head and were fitted. A detailed comparison of the IMN-bound complexes of neck cancer cell lines. Acta Oto-Laryngol., 121: 957–962, 2001. AKR1C3 and COX-2 should enable the development of NSAID that 13. Stack, E., and DuBois, R. N. Role of cyclooxygenase inhibitors for the prevention of colorectal cancer. Gastroenterol. Clin. N. Am., 30: 1001–1010, 2001. are more AKR1C3 or COX-2 selective. 14. Shureiqi, I., Xu, X. C., Chen, D. N., Lotan, R., Morris, J. S., Fischer, S. M., and Targetting AKR1C3 in Cancer. There is current widespread in- Lippman, S. M. Nonsteroidal anti-inflammatory drugs induce apoptosis in esophageal terest in exploiting PPAR␥ not only in hematological malignancies cancer cells by restoring 15-lipoxygenase-1 expression. Cancer Res., 61: 4879–4884, 2001. but also in solid tumors and diabetes. As a result, there have been 15. Dannenberg, A. J., Altori, N. K., Boyle, J. O., Dang, C., Howe, L. R., Weksler, B. B., great efforts to develop pharmacologically active synthetic PPAR␥ and Subbaramaiah, K. Cyclo-oxygenase 2: a pharmacological target for the preven- ligands. Regrettably, the early-generation drugs have proven severely tion of cancer. Lancet Oncol., 2: 544–551, 2001. 16. Turini, M. E., and DuBois, R. N. Cyclooxygenase-2: a therapeutic target. Annu. Rev. hepatotoxic to humans (55). Thus, AKR1C3 provides not only a Med., 53: 35–57, 2002. strong candidate for the COX-independent target of NSAIDs but also 17. Chiabrando, C., Castelli, M. G., Cozzi, E., Fanelli, R., Campoleoni, A., Balotta, C., a means of therapeutically targeting PPAR␥ using established drugs Latini, R., and Garattini, S. Antiinflammatory action of salicylates: aspirin is not a prodrug for salicylate against rat carrageenin pleurisy. Eur. Pharmacol., 159: 257– of known toxicology. Furthermore, the structural data reported here 264, 1989. should help enable the development of novel NSAID-like drugs with 18. April, P., Abeles, M., Baraf, H., Cohen, S., Curran, N., Doucette, M., Ekholm, B., improved properties and better selectivity against COX. It remains to Goldlust, B., Knee, C. M., Lee, E., et al. Does the acetyl group of aspirin contribute to the antiinflammatory efficacy of salicylic acid in the treatment of rheumatoid be seen whether drugs can be developed with sufficient selectivity arthritis? Semin. Arthritis Rheum., 19: 20–28, 1990. against other members of the AKR1C family. Indeed, many AKR 19. Preston, S. J., Arnold, M. H., Beller, E. M., Brooks, P. M., and Buchanan, W. W. enzymes are inhibited by NSAID. But, the sequence differences Comparative analgesic and anti-inflammatory properties of sodium salicylate and acetylsalicylic acid (aspirin) in rheumatoid arthritis. Br. J. Clin. Pharmacol., 27: among the AKR1C family are mostly in the three loops (loops A, B 607–611, 1989. and C; Fig. 5) that define the active site entrance channel and ligand- 20. Tegeder, I., Pfeilschifter, J., and Geisslinger, G. Cyclooxygenase-independent actions of cyclooxygenase inhibitors. FASEB J., 15: 2057–2072, 2001. binding pockets, and there are also observed differences in the activity 21. Khanna, M., Qin, K. N., Wang, R. W., and Cheng, K. C. Substrate-specificity, gene and substrate profiles of this family (23, 47), suggestive of structure structure, and tissue-specific distribution of multiple human 3-␣-hydroxysteroid de- differences that could be exploited in a drug design. hydrogenases. J. Biol. Chem., 270: 20162–20168, 1995. 22. Peltoketo, H., Luu-The, V., Simard, J., and Adamski, J. 17␤-Hydroxysteroid dehydrogenase (HSD)/17-ketosteroid reductase (KSR) family; nomenclature and main characteristics of the 17HSD/KSR enzymes. J. Mol. Endocrinol., 23: 1–11, 1999. ACKNOWLEDGMENTS 23. Matsuura, K., Shiraishi, H., Hara, A., Sato, K., Deyashiki, Y., Ninomiya, M., and Sakai, S. Identification of a principal mRNA species for human 3␣-hydroxysteroid We thank Audrey Boniface for technical assistance, Klaus Fu¨tterer for dehydrogenase isoform (AKR1C3) that exhibits high prostaglandin D-2 11-ketoreductase activity. J. Biochem. (Tokyo), 124: 940–946, 1998. fruitful discussions, and the European Synchrotron Radiation Facility for travel 24. Reginato, M. J., Krakow, S. L., Bailey, S. T., and Lazar, M. A. Prostaglandins and access to synchrotron facilities and help during data collection. We are promote and block adipogenesis through opposing effects on peroxisome proliferator- grateful to the Medical Research Council for part funding of the BIP compu- activated receptor ␥. J. Biol. Chem., 273: 1855–1858, 1998. tational suite and to Miklos Cserzo and Tony Pemberton for system adminis- 25. Straus, D. S., and Glass, C. K. Cyclopentenone prostaglandins: new insights on biological activities and cellular targets. Med. Res. Rev., 21: 185–210, 2001. tration. 26. Nagase, T., Miyajima, N., Tanaka, A., Sazuka, T., Seki, N., Sato, S., Tabata, S., Ishikawa, K., Kawarabayasi, Y., Kotani, H., et al. Prediction of the coding sequences of unidentified human genes. III. The coding sequences of 40 new genes (kiaa0081- REFERENCES kiaa0120) deduced by analysis of cDNA clones from human cell line KG-1. DNA Res., 2: 37–43, 1995. 1. Xu, X. C. Cox-2 inhibitors in cancer treatment and prevention, a recent development. 27. Mills, K. I., Gilkes, A. F., Sweeney, M., Choudhry, M. A., Woodgate, L. J., Bunce, Anti-Cancer Drugs, 13: 127–137, 2002. C. M., Brown, G., and Burnett, A. K. Identification of a retinoic acid responsive 2. Gupta, R. A., and DuBois, R. N. Colorectal cancer prevention and treatment by aldoketoreductase expressed in HL60 leukaemic cells. FEBS Lett., 440: 158–162, inhibition of cyclooxygenase-2. Nat. Rev. Cancer, 1: 11–21, 2001. 1998. 3. Grosch, S., Tegeder, I., Niederberger, E., Brautigam, L., and Geisslinger, G. Cox-2 28. Bunce, C. M., Mountford, J. C., French, P. J., Mole, D. J., Durham, J., Michell, R. H., independent induction of cell cycle arrest and apoptosis in colon cancer cells by the and Brown, G. Potentiation of myeloid differentiation by anti-inflammatory agents, selective cox-2 inhibitor celecoxib. FASEB J., 15: U300–U321, 2001. by steroids and by retinoic acid involves a single intracellular target, probably an 4. Zhang, X. P., Morham, S. G., Langenbach, R., and Young, D. A. Malignant trans- enzyme of the aldoketoreductase family. Biochim. Biophys. Acta, 1311: 189–198, formation and antineoplastic actions of nonsteroidal antiinflammatory drugs 1996. 1809

29. Desmond, J. C., Mountford, J. C., Drayson, M. T., Walker, E. A., Hewison, M., Ride, plasticity and tissue distribution reveals roles in the inactivation and formation of J. P., Luong, Q. T., Hayden, R. E., Vanin, E. F., and Bunce, C. M. The aldo-keto male and female sex hormones. Biochem. J., 351: 67–77, 2000. reductase AKR1C3 is a novel suppressor of cell differentiation that provides a 48. Penning, T. M., Burczynski, M. E., Jez, J. M., Lin, H. K., Ma, H. C., Moore, M., plausible target for the non-cyclooxygenase-dependent antineoplastic actions of non- Ratnam, K., and Palackal, N. Structure-function aspects and inhibitor design of type steroidal anti-inflammatory drugs. Cancer Res., 63: 505–512, 2003. 517␤-hydroxysteroid dehydrogenase (AKR1C3). Mol. Cell. Endocrinol., 171: 137– 30. Jackson, L., Wahli, W., Michalik, L., Watson, S. A., Morris, T., Anderton, K., Bell, 149, 2001. D. R., Smith, J. A., Hawkey, C. J., and Bennett, A. J. Potential role for peroxisome 49. Rose, I. A., Hanson, K. R., Wildinson, K. D., and Wimmer, M. J. A suggestion for proliferator activated receptor (PPAR) in preventing colon cancer. Gut, 52: 1317– naming faces of ring compounds. Proc. Natl. Acad. Sci. USA, 77: 2439–2441, 1980. 1322, 2003. 50. Schlegel, B. P., Jez, J. M., and Penning, T. M. Mutagenesis of 3␣-hydroxysteroid 31. Farrow, B., O’Connor, K. L., Hashimoto, K., Iwamura, T., and Evers, B. M. Selective dehydrogenase reveals a “push-pull” mechanism for proton transfer in aldo-keto activation of PPAR␥ inhibits pancreatic cancer invasion and decreases expression of reductases. Biochemistry, 37: 3538–3548, 1998. tissue plasminogen activator. Surgery, 134: 206–212, 2003. 51. Kurumbail, R. G., Stevens, A. M., Gierse, J. K., McDonald, J. J., Stegeman, R. A., 32. Ying, H., Suzuki, H., Zhao, L., Willingham, M. C., Meltzer, P., and Cheng, S. Y. Pak, J. Y., Gildehaus, D., Miyashiro, J. M., Penning, T. D., Seibert, K., Isakson, P. C., Mutant thyroid hormone receptor ␤ represses the expression and transcriptional and Stallings, W. C. Structural basis for selective inhibition of cyclooxygenase-2 by activity of peroxisome proliferator-activated receptor ␥ during thyroid carcinogenesis. anti-inflammatory agents. Nature (Lond.), 384: 644–648, 1996. Cancer Res., 63: 5274–5280, 2003. 52. Nahoum, V., Gangloff, A., Legrand, P., Zhu, D. W., Cantin, L., Zhorov, B. S., 33. Lapillonne, H., Konopleva, M., Tsao, T., Gold, D., McQueen, T., Sutherland, R. L., Luu-The, V., Labrie, F., Breton, R., and Lin, S. X. Structure of the human 3␣- Madden, T., and Andreeff, M. Activation of peroxisome proliferator-activated recep- hydroxysteroid dehydrogenase type 3 in complex with testosterone and NADP at tor ␥ by a novel synthetic triterpenoid 2-cyano-3,12-dioxooleana-1,9-dien-28-oic acid 1.25-Å resolution. J. Biol. Chem., 276: 42091–42098, 2001. induces growth arrest and apoptosis in breast cancer cells. Cancer Res., 63: 5926– 53. Askonas, L. J., Ricigliano, J. W., and Penning, T. M. The kinetic mechanism 5939, 2003. catalyzed by homogeneous rat-liver 3-␣-hydroxysteroid dehydrogenase: evidence for 34. Chen, G. G., Xu, H., Lee, J. F., Subramaniam, M., Leung, K. L., Wang, S. H., Chan, binary and ternary dead-end complexes containing nonsteroidal antiinflammatory U. P., and Spelsberg, T. C. 15-Hydroxy-eicosatetraenoic acid arrests growth of drugs. Biochem. J., 278: 835–841, 1991. colorectal cancer cells via a peroxisome proliferator-activated receptor ␥-dependent 54. Loll, P. J., Picot, D., Ekabo, O., and Garavito, R. M. Synthesis and use of iodinated pathway. Int. J. Cancer, 107: 837–843, 2003. nonsteroidal antiinflammatory drug analogs as crystallographic probes of the pros- 35. Cornish-Bowden, A. Analysis of Enzyme Kinetic Data. New York: Oxford Univer- taglandin H-2 synthase cyclooxygenase active site. Biochemistry, 35: 7330–7340, sity Press, 1995. 1996. 36. Bailey, S. The ccp4 suite: programs for protein crystallography. Acta Crystallogr. 55. Isley, W. L. Hepatotoxicity of thiazolidinediones. Expert Opin. Drug Saf., 2: 581– Sect. D Biol. Crystallogr., 50: 760–763, 1994. 586, 2003. 37. Otwinowski, Z., and Minor, W. Processing of x-ray diffraction data collected in 56. Ramachandran, G. N., and Sasisekharan, V. Conformation of polypeptides and oscillation mode. Methods Enzymol., 276: 307–326, 1997. proteins. Adv. Protein Chem., 23: 283–438, 1968. 38. Brunger, A. T., Adams, P. D., Clore, G. M., DeLano, W. L., Gros, P., Grosse- 57. Hoog, S. S., Pawlowski, J. E., Alzari, P. M., Penning, T. M., and Lewis, M. Kunstleve, R. W., Jiang, J. S., Kuszewski, J., Nilges, M., Pannu, N. S., Read, R. J., Three-dimensional structure of rat-liver 3-␣-hydroxysteroid dihydrodiol dehydrogen- Rice, L. M., Simonson, T., and Warren, G. L. Crystallography and NMR system: a ase, a member of the aldo-keto reductase superfamily. Proc. Natl. Acad. Sci. USA, new software suite for macromolecular structure determination. Acta Crystallogr. 91: 2517–2521, 1994. Sect. D Biol. Crystallogr., 54: 905–921, 1998. 58. Bennett, M. J., Schlegel, B. P., Jez, J. M., Penning, T. M., and Lewis, M. Structure 39. Jin, Y., Stayrook, S. E., Albert, R. H., Palackal, N. T., Penning, T. M., and Lewis, M. of 3␣-hydroxysteroid/dihydrodiol dehydrogenase complexed with NADP. Biochem- Crystal structure of human type III 3 ␣-hydroxysteroid dehydrogenase/bile acid istry, 35: 10702–10711, 1996. binding protein complexed with NADPϩ and ursodeoxycholate. Biochemistry, 40: 59. Bennett, M. J., Albert, R. H., Jez, J. M., Ma, H. C., Penning, T. M., and Lewis, M. 10161–10168, 2001. Steroid recognition and regulation of hormone action: crystal structure of testosterone 40. Read, R. J. Improved Fourier coefficients for maps using phases from partial struc- and NADPϩ bound to 3␣-hydroxysteroid dihydrodiol dehydrogenase. Structure, 5: tures with errors. Acta Crystallogr. Sect. A, 42: 140–149, 1986. 799–812, 1997. 41. Murshudov, G. N., Vagin, A. A., and Dodson, E. J. Refinement of macromolecular 60. Urzhumtsev, A., TeteFavier, F., Mitschler, A., Barbanton, J., Barth, P., Urzhumtseva, structures by the maximum-likelihood method. Acta Crystallogr. Sect. D Biol. Crys- L., Biellmann, J. F., Podjarny, A. D., and Moras, D. A “specificity” pocket inferred tallogr., 53: 240–255, 1997. from the crystal structures of the complexes of aldose reductase with the pharma- 42. Winn, M. D., Isupov, M. N., and Murshudov, G. N. Use of TLS parameters to model ceutically important inhibitors tolrestat and sorbinil. Structure, 5: 601–612, 1997. anisotropic displacements in macromolecular refinement. Acta Crystallogr. Sect. D 61. Wilson, D. K., Bohren, K. M., Gabbay, K. H., and Quiocho, F. A. An unlikely sugar Biol. Crystallogr., 57: 122–133, 2001. substrate site in the 1.65 angstrom structure of the human aldose reductase holoen- 43. Roussel, A. and Cambillau, C. Silicon Graphics Geometry Partners Directory, pp. zyme implicated in diabetic complications. Science (Wash. DC), 257: 81–84, 1992. 77–79. Mountain View, CA: Silicon Graphics, 1989. 62. Hyndman, D., Bauman, D. R., Heredia, V. V., and Penning, T. M. The aldo-keto 44. Hodel, A., Kim, S. H., and Brunger, A. T. Model bias in macromolecular crystal reductase superfamily homepage. Chem. Biol. Interact., 143–144: 621–631, 2003. structures. Acta Crystallogr. Sect. A, 48: 851–858, 1992. 63. Kraulis, P. J. Molscript: a program to produce both detailed and schematic plots of 45. Laskowski, R. A., Macarthur, M. W., Moss, D. S., and Thornton, J. M. Procheck: a protein structures. J. Appl. Crystallogr., 24: 946–950, 1991. program to check the stereochemical quality of protein structures. J. Appl. Crystal- 64. Esnouf, R. M. Further additions to Molscript Version 1.4, including reading and logr., 26: 283–291, 1993. contouring of electron-density maps. Acta Crystallogr. Sect. D Biol. Crystallogr., 55: 46. Guex, N., and Peitsch, M. C. Swiss-model and the Swiss-pdbviewer: an environment 938–940, 1999. for comparative protein modeling. Electrophoresis, 18: 2714–2723, 1997. 65. Thompson, J. D., Higgins, D. G., and Gibson, T. J. Clustal W: improving the 47. Penning, T. M., Burczynski, M. E., Jez, J. M., Hung, C. F., Lin, H. K., Ma, H. C., sensitivity of progressive multiple sequence alignment through sequence weighting, Moore, M., Palackal, N., and Ratnam, K. Human 3 ␣-hydroxysteroid dehydrogenase position-specific gap penalties and weight matrix choice. Nucleic Acids Res., 22: isoforms (AKR1C1-AKR1C4) of the aldo-keto reductase superfamily: functional 4673–4680, 1994.

1810

Downloaded from cancerres.aacrjournals.org on September 30, 2021. © 2004 American Association for Cancer Research. Crystal Structures of Prostaglandin D2 11-Ketoreductase (AKR1C3) in Complex with the Nonsteroidal Anti-Inflammatory Drugs Flufenamic Acid and Indomethacin

Andrew L. Lovering, Jon P. Ride, Christopher M. Bunce, et al.

Cancer Res 2004;64:1802-1810.

Updated version Access the most recent version of this article at: http://cancerres.aacrjournals.org/content/64/5/1802

Cited articles This article cites 61 articles, 19 of which you can access for free at: http://cancerres.aacrjournals.org/content/64/5/1802.full#ref-list-1

Citing articles This article has been cited by 11 HighWire-hosted articles. Access the articles at: http://cancerres.aacrjournals.org/content/64/5/1802.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/64/5/1802. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.