Re-engineering of Human Urokinase Provides a System for Structure-based Drug Design at High Resolution and Reveals a Novel Structural Subsite*

Inhibition of urokinase has been shown to slow tumor growth and metastasis. To utilize structure-based drug design, human urokinase was re-engineered to provide a more optimal crystal form. The redesigned protein consists of residues Ile16-Lys243 (in the chymotrypsin numbering system; for the urokinase numbering system it is Ile159-Lys404) and two point mutations, C122A and N145Q (C279A and N302Q). The protein yields crystals that diffract to ultra-high resolution at a synchrotron source. The native structure has been refined to 1.5 Å resolution. This new crystal form contains an accessible active site that facilitates compound soaking, which was used to determine the co-crystal structures of urokinase in complex with the small molecule inhibitors amiloride, 4-iodo-benzo(b)thiophene-2-carboxamidine and phenylguanidine at 2.0–2.2 Å resolution. All three inhibitors bind at the primary binding pocket of urokinase. The structures of amiloride and 4-iodo-benzo(b)thiophene-2-carboxamidine also reveal that each of their halogen atoms are bound at a novel structural subsite adjacent to the primary binding pocket. This site consists of residues Gly218, Ser146, and Cys191–Cys220 and the side chain of Lys143. This pocket could be utilized in future drug design efforts. Crystal structures of these three inhibitors in complex with urokinase reveal strategies for the design of more potent nonpeptidic urokinase inhibitors.

Cancer cell invasion, the spread and growth of tumor metastases, is a primary cause of mortality and morbidity of malignancy (2), and this invasion requires the degradation of basement membranes and other extracellular protein structures. Urokinase has been shown to be strongly associated with tumor cells (3) and to play a role in basement membrane degradation via a cascade mechanism involving activation of plasminogen and the metalloproteases (4 -6). Furthermore, inhibitors of urokinase have been reported to slow tumor metastasis as well as growth of the primary tumor (7)(8)(9)(10)(11)(12)(13)(14)(15). These inhibitors include the small molecules 4-iodo benzo(b)thiophene-2-carboxamidine (B428), 1 4-benzodioxolanyletheyl benzo(b)thiophene-2-carboxamidine (B623) (12)(13)(14), and amiloride (8,15). These compounds are competitive inhibitors of uroki-nase and have been proposed to bind at the primary binding pocket common to all trypsin-like serine proteases (15). However, none of these compounds posses all of the characteristics of a good therapeutic agent for the treatment of cancer.
Structure-based drug design has become an important tool for improving the potency and pharmacological characteristics of compounds toward providing therapeutic agents. This method has contributed to the development of potent and specific inhibitors for many targets such as HIV protease, cyclooxygenase-2, influenza neuraminidase, and the metalloproteinases (16 -22). To most efficiently apply crystallographydriven structure-based drug design, it is preferable that the crystals have certain properties. One property is that active site of the target is open in the crystal lattice. This molecular packing permits the diffusion and binding of compounds into the active site and eliminates the need to optimize crystal growth in the presence of each inhibitor. Another important property is that the crystals reproducibly diffract to high resolution (2.5-2.0 Å). It is preferable that this data quality is achievable on a conventional rotating anode source, thereby eliminating the need for travel to synchrotron facilities. The higher resolution data facilitate unambiguous map interpretation and minimize the average atomic positional error (23). Hence, an appropriate crystal form can greatly facilitate the process of structure-based drug design. A crystal system exists for urokinase, although it does not fully encompass the preferred properties outlined above.
Human low molecular weight (LMW) urokinase has been crystallized in complex with the peptidic inhibitor Glu-Gly-Argchloromethyl ketone (1). This structure reveals the geometry of the urokinase active site as well as the orientation of a peptide inhibitor in the substrate-binding groove. However, the LMW urokinase crystals diffract to lower resolution (2.5 Å resolution, synchrotron radiation; 3.0 Å resolution, rotating anode source) and utilize co-crystallization to achieve the target-ligand complex. In addition, the active site is in close contact with another molecule because of a noncrystallographic 2-fold axis near the active site. This interaction could limit minor ligand induced conformational shifts and perhaps distort the active site conformation. Furthermore, the noncrystallographic and crystallographic packing effectively blocks the active site such that it would be difficult to diffuse small molecules into the active site in this crystal form (if they were not blocked by the irreversible covalent inhibitor). Hence, although this system may be used for modeling of small molecule urokinase inhibitors, it may not provide an ideal system for structure-based drug design. Therefore, to design an anti-cancer therapeutic, a new crystal form of human urokinase was sought to facilitate the application of structure-based drug design. The strategy utilized protein engineering and information from the reported LMW urokinase structure to design an altered protein sequence to yield a new crystal form.
The new form of urokinase, micro-urokinase, crystallizes under conditions very similar to the low molecular weight form (1), although crystal packing and data quality are very different. This new crystal form contains a monomer in the asymmetric unit and diffracts to ultra-high resolution (d min ϭ 1.03 Å). In addition, this crystal form has an open active site permitting direct diffusion of compounds into the apo-crystals and is therefore ideal for providing precise structure determinations for urokinase ligand complexes by the soaking technique.
The re-engineered crystal system and soaking technique were utilized to determine the co-crystal structure of urokinase in complex with a series of small molecule inhibitors at 2.0 or 2.2 Å resolution. Two of these inhibitors, amiloride (24), and B428 (25,26), have been shown to reduce tumor size and metastasis (8,(12)(13)(14)(15), whereas the effect of the third, phenylguanidine (27) has not been reported to date. These complex structures were completed to determine the binding orientation of each compound to urokinase. This information in turn may be utilized to design molecules of increased potency toward discovery of an anti-cancer therapeutic compound.

EXPERIMENTAL PROCEDURES
Recombinant Micro-urokinase-Micro-urokinase was engineered by polymerase chain reaction manipulations using a human urokinase cDNA as a template (28). The C279A and N302Q mutations were made by the method of polymerase chain reaction based site-directed mutagenesis. Urokinase native leader sequence was fused directly to Ile 159 by polymerase chain reaction. This product was ligated to a baculovirus transfer vector pJVP10z (29). The final expression vector sequence was confirmed by DNA sequencing.
The pJVP10z-micro-urokinase vector was transfected into Sf9 cells by the calcium phosphate precipitation method using the BaculoGold kit from PharMingen (San Diego, CA). Single recombinant virus expressing micro-urokinase was plaque purified by standard methods, and a large stock of the virus was prepared. Large scale expression of micro-urokinase was performed in suspension in High-Five cells, (Invitrogen, San Diego, CA) growing in Excel 405 serum free medium (JRH Biosciences, Lenexa, KS) at 27°C. Urokinase activity in the supernatant was measured by amidolysis of the chromogenic urokinase substrate H-D-pyroglutamyl-Gly-L-Arg-p-nitroanilide (S2444; Helena Laboratories, Beaumont, TX). The culture supernatant was harvested as the starting material for purification. Protease inhibitors, iodoacetamide (10 mM), benzamidine (5 mM), and EDTA (1 mM) were added to the pooled culture medium. The medium was diluted 5-fold with 5 mM HEPES, pH 7.5, and filtered through 1.2 and 0.2-m membranes. The micro-urokinase protein was captured onto Sartorius membrane adsorber S100 (Sartorius, Edgewood, NY) by passing the medium through the membrane at a flow rate of 50 ϳ100 ml/min. After extensive washing with 10 mM HEPES, pH 7.5, containing 10 mM iodoacetamide, 5 mM benzamidine, and 1 mM EDTA, micro-urokinase was eluted from S100 membrane with a NaCl gradient (20 -500 mM, 200 ml) in 10 mM HEPES buffer, pH 7.5, 10 mM iodoacetamide, 5 mM benzamidine, 1 mM EDTA. The eluate was diluted 10-fold with the above 10 mM HEPES buffer containing inhibitors, and loaded onto a S20 column (Bio-Rad). Microurokinase was eluted with a 20ϫ column volume NaCl gradient (20 -500 mM). No inhibitors were used in the elution buffers. The eluate was then diluted 5-fold with 10 mM HEPES buffer, pH 7.5, and loaded onto a heparin-agarose (Sigma) column. Micro-urokinase was eluted with a NaCl gradient from 10 -250 mM. The heparin column eluate of microurokinase was applied to a benzamidine-agarose (Sigma) column equilibrated with 10 mM HEPES buffer, pH 7.5, 200 mM NaCl. The column was washed with the equilibration buffer, and the urokinase was eluted with 50 mM NaOAc, pH 4.5, 500 mM NaCl. The micro-urokinase eluate was concentrated to 4 ml by ultrafiltration and applied to a Sephadex G-75 column equilibrated with 20 mM NaOAc, pH 4.5, 100 mM NaCl. The single peak containing micro-urokinase was collected and lyophilized as the final product.
Amidolytic Kinetics of Urokinase and Micro-urokinase-The effects of synthetic inhibitors on the steady state amidolytic activity of LMW urokinase or micro-urokinase toward the chromogenic substrate, S2444 (Helena Laboratories), was characterized by the formation of p-nitroanaline (30). Briefly, 0 -50 M concentration of inhibitors were tested against 25 IU/ml (0.14 ng/ml) LMW urokinase or micro-urokinase and 0.4 -4.0 mM concentrations of S2444 in 200 l volumes in phosphate-buffered saline and 0.01% bovine serum albumin, pH 7.4. Incubations were performed at 37°C with absorbance at 405 nm recorded every 11 s for 20 min. Data were plotted as 1/S versus 1/v for Lineweaver-Burk analysis and the calculation of inhibition constants. K i values were obtained from replots of the resultant slopes versus I (26,31).
Protein Crystallography-Crystals were obtained by the hanging drop vapor diffusion method. A typical well solution of 0.15 M Li 2 SO 4 , 20% polyethylene glycol MW 4000 in succinate buffer, pH 4.8 -6.0, was used. On the coverslip, 2 l of well solution is mixed with 2 l of protein solution, and the slip is sealed over the well. Crystallization occurred at 18 -24°C within 24 h. The protein solution was composed of 6 mg/ml (0.21 mM) micro-urokinase in 10 mM citrate, pH 4.0, 3 mM ⑀-amino caproic acid p-carbethoxyphenyl ester chloride with 1% Me 2 SO cosolvent. The resultant micro-urokinase crystals are composed of enzyme with an empty active site. The compound ⑀-amino caproic acid p-carbethoxyphenyl ester chloride is reported to inhibit urokinase with an apparent K i of 0.3 M at neutral pH and was co-crystallized with urokinase in an attempt to obtain a complex structure (32). Repeated tests with this compound resulted in a structure with an active site occupied only by ordered solvent molecules even at 1.5 Å resolution. Hence, we have hypothesized that this inhibitor is degraded during the crystallization experiment albeit critical for obtaining urokinase crystals. Studies are underway to try to understand the mechanism of this phenomenon.
The micro-urokinase crystals belong to the space group P2 1 2 1 2 1 with unit cell dimensions of a ϭ 55.16 Å, b ϭ 53.00 Å, c ϭ 82.30 Å and ␣ ϭ ␤ ϭ ␥ ϭ 90°and diffract beyond 1.5 Å on a Rigaku RTP 300 RC rotating anode source equipped with an RAXISII detector. In addition, a 1.03 Å resolution native data set was collected on a CCD detector at beam line F1 of the Cornell High Energy Synchrotron Source in Ithaca, NY. All data were collected at 100 -160 K and processed by the program package DENZO (33). Before crystals were frozen, they were passed through a solution of 0.15 M Li 2 SO 4 , 20% polyethylene glycol MW 4000, succinate buffer, pH 4.8 -6.0, and 20% glycerol for cryogenic protection. Data were collected at low temperature to preserve the diffraction of the crystal throughout data acquisition. The crystal structure was determined by the molecular replacement method using the program AMORE (34). The LMW urokinase structure was used as the search probe (1) (Protein Data Bank entry 1LMW) against the RAXISII data.
The structure was refined to 1.5 Å resolution using the synchrotron data and the program package XPLOR (35) by a combination of rigid body, simulated annealing maximum likelihood refinement, and maximum likelihood positional refinement. Electron density maps to 1.5 Å resolution were inspected on a Silicon Graphics INDIGO2 workstation using the program package QUANTA 97 (Molecular Simulations, Inc). At 1.5 Å resolution constrained individual temperature factor refinement was also included in the refinement cycle. Electron density maps to 1.5 Å resolution were examined, and water molecules and bound ions were identified as positive peaks in the F o Ϫ F c map at least 4 above noise. Refinement continued with automatic water addition using the XWAT feature of SHELXL (36). Final refinement steps included cycles of model building where disorder and additional solvent molecules were added. The final R-factor is 19.2% with a R free of 21.8%.
To obtain the amiloride, B428, or phenylguanidine micro-urokinase complex structures, crystals of urokinase were placed in 50 l of crystallization mother liquor to which 0.5 l of a 1 mg/10 l compound solution was added. The solid compound was obtained from the Abbott chemical repository and was initially dissolved in Me 2 SO. Crystals were allowed to incubate for 12-15 h at 24°C and prepared for data collection in a manner identical to that of the native crystals. Data were collected on a Rigaku RTP 300 RC rotating anode source equipped with an RAXISII detector at 160 K by the method of flash freezing. Data were processed using the HKL program suite (33). Initial electron density maps were calculated using the program package XPLOR (35) and the 1.5 Å native model. All electron density maps were inspected on a Silicon Graphics INDIGO2 workstation using QUANTA 97, and the orientation of all compounds were clearly visualized in the initial 2F o Ϫ F c map. The complexes were refined to 2.0 Å resolution using the program package XPLOR. Refinement consisted of alternating steps of positional and B-factor refinement. Ordered solvent molecules were identified as positive peaks in the F o Ϫ F c map that were 4 above noise. Table I summarizes statistics for all micro-urokinase models. All data are between 89 and 90% complete with a merging R sym between 7 and 11% and an I/ between 12 and 15. The native model is refined to a R factor of 19.2% and R free of 21.8% at 1.5 Å resolution. The overall B-factor for the protein is 12 Å 2 , and the overall B-factor for the 337 ordered solvent molecules is 26 Å 2 . The current native model also contains three ordered sulfate ions, and two alternate side chain conformations located at the active site. All backbone atoms are well defined in the final 2F o Ϫ F c map with atomic B-factors at or below 30 Å 2 . The B428 model is refined to 2.0 Å resolution with a R factor of 20.9% and a R free of 27.7%, while the amiloride model is refined to 2.2 Å resolution with a R factor of 21.5% and a R free of 29.1%. The phenylguanidine model is refined to 2.0 Å resolution with a R factor of 18.9% and a R free of 22.1%. Data for the complex structures were of quality comparable with that of native structures collected under the same conditions on a rotating anode source.

RESULTS
Redesign of LMW Urokinase-To redesign the LMW urokinase sequence for the purpose of improving the crystal characteristics, the LMW urokinase coordinate file (Protein Data Bank entry 1LMW) was examined for sequences of excessively high B-factor, suggesting areas of disorder. The hypothesis is that areas of high disorder in the structure may contribute to the overall disorder of the crystals and/or may interfere with optimal crystal packing. The LMW urokinase structure consists of residues 136 -158 of the A-chain and 159 -411 of the B-chain connected by a disulfide bridge between Cys 148 and Cys 279 (urokinase numbering). 2 The B-chain corresponds to the serine protease domain, whereas the 21 residue A-chain lacks the kringle and epidermal growth factor domains present in full-length urokinase. The A-chain is reported to be an area of high disorder (1), and examination of the protein data bank coordinate file (Protein Data Bank entry 1LMW) reveals that residues 148 -155 of the A-chain have an average B-factor of 64 Å 2 ranging from 26 Å 2 for the disulfide-linked sulfur of residue Cys 148 to 110 Å 2 for Pro 155 . The very high B-factors for the LMW urokinase A-chain confirm this observation. Consequently, the A-chain was removed as a first step in the redesign. Furthermore, to remove the resultant free thiol on the B-chain, Cys 148 was mutated to an alanine.
Further examination of the LMW urokinase coordinate file indicates a second area of disorder consisting of residues 405-411 of the C terminus where the average B-factor is 147 Å 2 . Residues 407-411 represent a five residue extension in urokinase relative to other trypsin-like serine proteases. However, because residues 405-406 also have high atomic B-factors, the entire 405-411 segment was removed. The final potential site for disorder is the glycosylation site at residue 302. This glycosylation site was removed by an N302Q mutation to facilitate expression of the glycosylation-free protein in baculovirus. Hence, the re-engineered urokinase (micro-urokinase) consists of residues Ile 159 -Lys 404 (Ile 16 -Lys 243 chymotrypsin numbering system) with the two point mutations C279A (C122A) and N302Q (N145Q).
Micro-urokinase Crystal Packing-Micro-urokinase crystallizes with a monomer in the asymmetric unit (P2 1 2 1 2 1 ), whereas the LMW urokinase crystal form has a dimer in the asymmetric unit (R3) with intimate contacts at the substratebinding site. Specifically, in LMW urokinase, residues 94 -101 from each molecule (chymotrypsin numbering system as aligned by Ref. 1) 2 form a series of intermolecular main chain hydrogen bonds resulting in an extended four stranded ␤-sheet (1). From the LMW urokinase structure, it was seen that this loop decreases the size of the S 4 pocket relative to that at the substrate-binding site of other serine proteases such as thrombin, Factor Xa and tissue plasminogen activator (1,(37)(38)(39). Hence, this loop provides a critical structural feature of the substrate-binding groove. However, because of the close crystal contact at this site in the LMW urokinase crystals, the possibility existed that the structure of the substrate-binding site may be distorted or conformationally restricted. The new crystal form of micro-urokinase lacks the close crystal contact present in LMW urokinase, and an overlay of the two structures indicates that the conformation of this loop is essentially identical in the two crystal forms. Consequently, it is unlikely that packing in either crystal system affects the conformation of this loop and the resultant shape of the S 4 pocket, although the more open micro-urokinase packing may allow for inhibitorinduced conformational shifts.
Examination of crystal packing at the A-chain-binding cleft gives insight into why micro-urokinase yields different lattice packing and better diffracting crystals (a sample of the final 2F o Ϫ F c electron density map at 1.5 Å resolution is shown in Fig. 1A). In LMW urokinase, the A-chain binds in a cleft composed of residues 25-29, 116 -122, and 201-208. In the crystal structure of micro-urokinase, there is no A-chain, and the Achain-binding cleft is partially occupied by a symmetry related molecule. Specifically, a hydrophobic loop extending from 144 to 150 in the symmetry related molecule is directly bound at the A-chain site such that Tyr 149 -OH of the loop is involved in two hydrogen bonds at the A-chain cleft (Ser 202 -N and Ser 135 -O). In LMW urokinase, the A-chain blocks this set of interactions. Thus, in micro-urokinase, removal of the A-chain exposes a new "binding site" for the 144 -150 loop of another microurokinase molecule permitting a new lattice to form. This interaction at the A-chain cleft probably contributes to the improved crystal quality by being both a site of nucleation as well as by facilitating very close contact between adjacent molecules. 2 The urokinase numbering system is used for discussion of the sequence re-engineering work, whereas the chymotrypsin numbering system as aligned by Ref. 1 is used for discussion of the serine protease domain structure for micro-urokinase.
Value of the R factor where 10% of the data were randomly removed from the refinement.
Micro-urokinase and LMW urokinase are nearly identical in structure (overall rms deviation for main chain atoms, 0.8 Å) with one significant structural change near a site of re-engineering. As discussed above, removal of the A-chain results in an empty cavity. One loop (201-210) forming this site undergoes a conformational shift relative to LMW urokinase with rms deviation (main chain) ranging from 1.1 to 1.8 Å with the largest shift being for Arg 206 . However, although this loop is involved in a crystal packing interaction, the conformation of the 144 -150 of the symmetry related molecule is the same for both micro-urokinase and LMW urokinase. Other sites of variation include the flexible loop at residues 37-37D (rms deviation main chain, 1.7-3.5 Å), residues 17-19 (rms deviation main chain, 1.1-2.1 Å) and residues 185B-186 (rms deviation main chain, 1.7 Å). All areas were of high b-factor in the LMW urokinase structure (b-factor Ͼ 60 -90 Å 2 ) but of significantly lower b-factor in the micro-urokinase structure (b-factor Ͻ 20 Å 2 ) with the exception of residues 17-19, which were of low b-factors in both structures. The 17-19 segment was clearly defined in the final 2F o Ϫ F c electron density maps of microurokinase and is not near any re-engineered sites. Residues 185B-186 were remodeled in the higher-resolution structure. In the lower resolution LMW urokinase structure, Trp 186 was exposed to solvent and Gln 185B was buried. The higher resolution data clearly placed Trp 186 in the protein core with Gln 185B exposed to solvent.
Active Site of Native Micro-urokinase-Like the overall molecular fold, the active sites of LMW urokinase and microurokinase are nearly identical (rms deviation, Ͻ0.8 Å). The higher resolution data did not depict any large side chain movements relative to LMW urokinase but did show an alternate side chain conformation for two residues (Fig. 1, B and C) in addition to a bound sulfate ion (see Fig. 3C). The sulfate ion is bound near the oxyanion hole (40), where O1 is accepting hydrogen bonds from Gly 193 -NH (2.8 Å) and Ser 195 -OH (2.8 Å), whereas O 2 is accepting a hydrogen bond from His 57 -N⑀2 (2.8 Å). Hence, the higher resolution data revealed more structural details at the active site.
In Fig. 1B, native 1.5 Å 2F o Ϫ F c (contoured at 1 ) and F o Ϫ F c (contoured at 3 ) electron density maps depict that the side chain of His 99 is in multiple conformations. These maps were calculated before the alternate conformation had been included in the model. As presented in Fig. 1B, one His 99 conformation is identical to that observed with LMW urokinase. In this conformation, His 99 -N␦1 accepts a hydrogen bond from Tyr 94 -OH (2.9 Å). In the alternate conformation (modeled into the green positive peak; Fig. 1B), the His 99 imidazole is rotated approximately 90°about the C␤-C␥ bond resulting in a different hydrogen bonding pattern. Here, His 99 -N␦1 can donate a hydrogen bond to Asp 102 -O␦1 (3.2 Å). The His 99 side chain forms part of both the S 4 and S 2 pockets. Hence, a change in the conformation of His 99 results in a change in the overall shape of S 2 and S 4 , suggesting that the side chain movement would effect a drug design strategy directed toward the substratebinding groove.
The side chain of Cys 42 is also observed in two side chain conformations and is near the active site (Fig. 1C). In what is likely the major conformation, the Cys 42 -Cys 58 disulfide bridge is intact. However, in the alternate conformation, the disulfide is broken and the Cys 42 thiol group lies in a small hydrophobic pocket formed by the side chains of Phe 59 , Ile 29 , and Val 41 . This side chain shift is unexpected as the Cys 42 -Cys 58 disulfide bridge is present all trypsin-like serine protease structures, and its proximity to the catalytic triad suggests that it may structurally stabilize the active site. Hence, one might expect the catalytic activity to be affected when this disulfide bridge is broken. On the other hand, one must note that this observation occurs in the solid state and that further solution work would be necessary to determine its physiological significance.  Examination of crystal packing at the active site reveals that the micro-urokinase molecules pack forming a solvent channel that leads to the active site groove. Therefore, small molecule inhibitors may diffuse into the crystal and bind at the active site. This is important from a structure-based drug design perspective because it facilitates soaking as a method of forming protein-compound complex crystals. The soaking method was used to obtain crystal structures with the three known urokinase inhibitors, B428, amiloride, and phenylguanidine. These structures were obtained at high resolution and provide a starting point for structure-based drug design of a nonpeptidic urokinase inhibitor.
B428 -B428 has been reported to inhibit human urokinase with an IC 50 value of 0.320 M (Refs. 25 and 26 and Table II). B428 inhibition was tested versus LMW urokinase and microurokinase, and Fig. 2 presents the Lineweaver-Burke analysis for the effect of B428 on the activity of micro-urokinase. The results show that B428 competitively inhibits micro-urokinase as observed for the native enzyme (25,26). As listed in Table II, B428 inhibits LMW urokinase with a K i of 0.490 M while inhibiting micro-urokinase with a K i of 0.512 M. Hence, K i values for the native and re-engineered forms of the protein are essentially identical and are consistent with reported IC 50 values (25,26).
The B428-micro-urokinase co-crystal structure was completed to 2.0 Å resolution. In the complex structure, the 2F o Ϫ F c and F o Ϫ F c maps indicate that His 99 is in two conformations as observed in the native structure although Cys 42 is observed only in the conformation in which the Cys 42 -Cys 58 disulfide bridge is intact. It is unclear why only one conformation is observed for the Cys 42 -Cys 58 disulfide. In the native structure, the alternate conformation became visible at high resolution. Hence, one possibility is that second conformation is not visible in the lower resolution electron density map. Another explanation is that inhibitor binding may induce a shift to a single conformation or that the inhibitor may only bind to the protein form where the disulfide is intact. Further experiments at high resolution will be necessary to fully understand this phenomenon. Fig. 3A shows the 2F o Ϫ F c (contoured at 1 ) and F o Ϫ F c (contoured at 3 ) electron density maps calculated in the absence of inhibitor and before any refinement cycles. All atoms of the inhibitor are clearly defined in both maps, and the compound is found to bind at the S 1 pocket as might be predicted from its net positive charge.
Interactions between B428 and the S 1 pocket are consistent with observations for trypsin and other trypsin-like enzymes (41)(42)(43)(44)(45). Nearly all atoms of B428 are in van der Waals' or hydrogen bonding contact with the S 1 site (Fig. 3, B and C).  (Fig. 3B). Hence, both hydrophobic and hydrophilic interactions occur at S 1 .
In addition to interactions at S 1 , the 4-iodo group is pointing out of the S 1 pocket away from the substrate-binding groove and is making van der Waals' interactions with the side chain of Cys 220 and the main chain atoms of Gly 218 . These residues form part of a subpocket composed of the disulfide bridge at Cys 191 -Cys 220 , residues Gly 218 and Ser 146 , and the side chain of Lys 143 . This pocket has been termed the S 1 ␤ pocket because of its proximity to the primary S 1 site (Fig. 3C). It is reported that the 4-iodo group of B428 confers a 10-fold increase in binding potency relative to the 4-hydro compound (25,26). This observation is consistent with the B428-urokinase crystal structure where the 4-iodo group partially accesses the S 1 ␤ pocket. Fur-  Table II. FIG . 3. A, initial 2F o Ϫ F c (purple) and F o Ϫ F c (green) maps contoured at 1 and 3 , respectively, for the binding site of B428 before refinement. B, molecular surface as calculated by the program package QUANTA (Molecular Simulations Inc.) depicting interactions between B428 and micro-urokinase. The inhibitor and inhibitor surface are shown in orange, whereas the protein and the protein surface are shown in cyan. C, view of B428 bound at the S 1 site of urokinase. The S 2 site between His 57 and His 99 is also shown as well as the S 4 site. An ordered sulfate ion is also shown bound near the oxyanion hole. thermore, B623 inhibits urokinase with an IC50 of 0.07 M (25,26). Based upon the crystal structure of B428-micro-urokinase, it is possible that this larger 4-substituent is occupying more of the S 1 ␤ pocket 3 and consequently binds more tightly to urokinase. Hence, access to this novel pocket has been shown to confer an increase in binding potency and may serve as a site for further substitution in structure-based drug design.
Examination of the crystal structure of B428-urokinase shows that the 5 and 6 positions of the benzo(b)thiophene-2carboxamidine are also open for substitution, whereas the 3 and 7 positions are buried within the S 1 pocket and therefore less likely to accommodate a substituent. Of these, the 5 position does not directly point toward any pockets of the urokinase molecule because it points toward Gln 192 and out toward bulk solvent. Hence, substitution at this position is less likely to confer a large increase in binding potency. On the other hand, the 6 position points toward the urokinase catalytic site although the position appears partially blocked by the side chain of the active site Ser 195 . The distance from Ser 195 -OH to the 6 position carbon is 3.2 Å; therefore incorporation of a substitution at this position may require a shifting of the benzothiophene scaffold away from Ser 195 . Additionally, substitutions at the 6 position would not orient toward the substrate-binding groove accessed by Glu-Gly-Arg-chloromethyl ketone. Substitutions at the 6 position would have to bend back toward the substrate-binding site or access other subsites. Nevertheless, the 4 and 6 positions appear to be the best substitution sites toward increasing the binding potency of B428, and both sets of substitutions will likely occupy sites apart from the substratebinding groove.
Amiloride-Amiloride has been reported to inhibit human urokinase with a K i (24) or IC 50 of 7 M (25, 26). As observed with B428, amiloride also competitively inhibits LMW uroki- nase and micro-urokinase with similar values (K i ϭ 7.2 M for LMW urokinase, and K i ϭ 6.9 M for micro-urokinase). Amiloride is a weaker urokinase inhibitor than B428 (Table II) but may have more favorable pharmacological properties because the compound is an orally active commercial drug (46). To compare the binding modes of amiloride and B428 and to establish strategies for development of a more potent amiloridebased urokinase inhibitor, the co-crystal structure of amiloride micro-urokinase was completed at 2.2 Å resolution.
Examination of the 2F o Ϫ F c (contoured at 1 ) and F o Ϫ F c (contoured at 3 ) electron density maps at the active site shows that all atoms of the inhibitor are clearly defined in both maps (Fig. 4A). In addition, the maps show His 99 in two conformations and the Cys 42 -Cys 58 disulfide bridge intact as observed in the B428 complex. The data also indicate that amiloride binds at the S 1 pocket as observed with B428 (Fig. 4C).
The crystal structure of amiloride-micro-urokinase indicates that amiloride is making more hydrogen bonding interactions at the S 1 site than B428 while maintaining some of the van der Waals' interactions within the pocket. The size of the amiloride pyrazine scaffold is smaller than the B428 benzothiophene such that even though the pyrazine ring is in contact with the rim of the S 1 pocket as observed for B428, the extent of the packing interactions is smaller. In place of the thiophene ring, the 3-amino and 2-acylguanidine groups of amiloride are making hydrogen bonding interactions. Specifically, the 3-amino group is packed underneath the side chain of Ser 195 as shown in Fig. 4B where it is donating a hydrogen bond to Ser 195 -O␥ (3.1 Å). The carbonyl of the acyl guanindine group is accepting a hydrogen bond (2.9 Å) from a buried solvent molecule bound directly above Tyr 228 . The guanidine-NH is donating a hydrogen bond to Gly 218 -O (3.1 Å). As observed with B428, the amide-like nitrogens are donating hydrogen bonds to Gly 218 -O (2.7 Å) and Asp 189 -O␦1(3.0 Å) or to Asp 189 -O␦2 (3.0 Å), and Ser 190 -O␥ (2.7 Å). The hydrogen bonding geometry of the guanidinium group is also very similar to that observed for ArgP 1 in the Glu-Gly-Arg-chloromethyl ketone-LMW urokinase structure (1). Hence, although the core scaffolds of both B428 and amiloride are bound at the S 1 pocket, the nature of the interactions within the pocket are different.
The crystal structure of amiloride-micro-urokinase reveals strategies for structure-based drug design of a more potent small molecule inhibitor. One potential site of substitution is the 6 position. The 6-chloro group of amiloride is accessing the S 1 ␤ pocket as observed for the 4-iodo group of B428. Specifically the 6-chloro group is in hydrophobic contact with the side chain of Cys 220 and the main chain atoms of Gly 218 (Fig. 4C). Thus, although the chemical structures of B428 and amiloride are very different, interactions at the S 1 ␤ pocket are nearly identical. Because of this similarity, one might substitute the 6-chloro position of amiloride with larger groups such as iodine (present in B428) or a benzodioxol arylethenyl (present in B623), which were both shown to enhance the activity in the benzo(b)thiophene-2-carboxamidine series. The 3 position of amiloride within the S 1 pocket is another site for substitution. However, substitutions at this site are expected to point toward Gln 192 and then out toward bulk solvent as observed for the 5 position of B428. Thus, use of a rigid linker may be necessary to redirect substitutions toward the protein including the substrate-binding groove. In summary, substitutions of the amiloride scaffold should occur at the 5 and 6 positions to provide direct access to the S 1 ␤ pocket or indirect access to other sites on the protein.
Phenylguanidine-Phenylguanidine inhibits urokinase with a K i of 20.6 M (27) and is therefore a weaker inhibitor of urokinase than either amiloride or B428 (Table II). This inhib-itor also competitively inhibits micro-urokinase with a K i consistent with the LMW form (K i ϭ 20.6 M LMW for urokinase, and K i ϭ 17.4 M for micro-urokinase). To compare the binding mode of this inhibitor to amiloride and B428 and to determine potential sites of substitution, the co-crystal structure of phenylguanidine-micro-urokinase was completed at 2.0 Å resolution.
The phenylguanidine-micro-urokinase active site structure is very similar to that in the presence of B428 and amiloride. His 99 is observed in multiple conformations while the Cys 42 -Cys 58 disulfide bridge is intact. Additionally, the 2F o Ϫ F c (contoured at 1 ) and F o Ϫ F c (contoured at 3 ) electron density maps (Fig. 5A) obtained using the urokinase model in the absence of inhibitor and before any refinement cycles shows that all atoms of the inhibitor are clearly defined in both maps. The inhibitor was found to bind at the S 1 pocket (Fig. 5B).
Even though both amiloride and phenylguanidine have scaffolds of the same size, the phenyl ring of phenylguanidine binds very differently from the pyrazine ring of amiloride (Fig. 5, B  and C). Specifically, the phenylguanidine ring packs underneath Ser 195 and is interacting with the main chain atoms of Val 213 -Trp 215 as well as the side chain of Val 213 . The ring also interacts with the main chain atoms of Ser 190 -Cys 191 as well as the side chain of Ser 190 . The differential ring packing is most likely due to amiloride possessing one additional linker atom between the guanidine and aromatic groups relative to phenylguanidine (Table II) because the guanidine groups are oriented very similarly. Specifically, the guanidine-NH is donating a hydrogen bond to Gly 218 -O (3.0 Å), whereas the amidine-like nitrogens are donating hydrogen bonds to Gly 218 -O (2.9 Å) and Asp 189 -O␦1 (2.9 Å) or to Asp 189 -O␦2 (3.0 Å) and Ser 190 -O␥ (3.3 Å). Thus, it is likely that the core scaffold of amiloride (pyrazine ring) orients differently than the phenyl group of phenylguanidine because the binding is being driven by the hydrogen bonding geometry of the guanidine groups rather than the van der Waals'/hydrogen bonding interactions of the core groups even though interactions of the core groups most certainly contribute to the compound binding.
The phenyl guanidine urokinase structure also shows that Gln 192 has changed conformation and is in hydrophobic contact with the inhibitor (Fig. 5B) such that it is blocking the entrance to the S 1 ␤ pocket. In the native and the B428 or amiloride complex structures, the S 1 ␤ pocket is open where Gln 192 is accepting a hydrogen bond from Lys 143 (3.3 Å) and donating a hydrogen bond to Tyr 151 (3.1 Å). Thus, a conformational shift of this side chain requires breaking two hydrogen bonds. This is not the case for other serine proteases such as thrombin where there is no hydrogen bonding partner for Glu 192 in either position. Here, there is less of an energy barrier to a conformational shift of Glu 192 , and the side chain may be found in both conformations (49,50). For urokinase, it appears that the binding of certain inhibitors such as phenyl guanidine does break the two Gln 192 hydrogen bonds and conformationally shift Gln 192 to maximize hydrophobic desolvation of the compound. Hence, Gln 192 may be induced to shift conformation and because Gln 192 may act as a switch to the entrance to S 1 ␤ from S 1 , noting the orientation of this side chain is important in a drug design strategy.
The crystal structure of phenylguanidine-urokinase suggests a structure-based drug design strategy different from that with B428 or amiloride. Both B428 and amiloride are capable of directly accessing the S 1 ␤ pocket, whereas the binding orientation of phenylguanidine is such that a similar interaction cannot be achieved by direct substitution of the phenyl ring ( Fig. 5C) even with movement of Gln192 to the S 1 ␤ open position. Specifically, as shown in Fig. 5 (B and C), the 2 and 3 positions could point toward the S 1 ␤ pocket but are too far away to support direct interaction with S 1 ␤. In fact, substitution of the phenyl ring with halogens at both the 2 and 3 positions did not result in any increase in inhibitory potency (27). On the other hand, substitution at position 4 with a chloro-or trifluromethyl-group resulted in an increase in inhibition to K i values of 6.8 and 6.5 M, respectively (27). This 4 substitution is expected to orient toward the side chain of Ser 195 and may obtain binding energy from a favorable van der Waals' packing interaction with Ser 195 and the S 1 pocket. The 5 and 6 positions are within the S 1 pocket and therefore less open for substitution. Because interactions with the S 1 ␤ pocket are expected to confer an increase in binding potency and because phenylguanidine may not directly access this site, modification of the scaffold may be a promising drug design strategy for this series.
Further examination of an overlay of the crystal structures of phenyl guanidine and amiloride micro-urokinase (Fig. 5C) shows that the binding of the two scaffolds is complementary. The lack of overlap between the two groups suggests that the phenyl and pyrazine rings could be fused to form a 1-naphth-ylguanidine system. The naphthyl ring would be expected to occupy the sites of both core scaffolds and could therefore maintain the positive characteristics of both the phenylguanidine and amiloride series. This would include utilization of the 4-chloro or 4-trifluromethyl substitutions in the phenylguanidine series as well as access to the S 1 ␤ pocket exploited by amiloride and B428. Hence, a merging of the amiloride and phenylguanidine scaffolds would be predicted to benefit from the additivity of both sites and create a more potent and easily optimized urokinase inhibitor. DISCUSSION Urokinase inhibitors have been shown to affect tumor metastasis and growth in vivo making urokinase an attractive anti-cancer target. However, these existing compounds lack all of the properties necessary for a therapeutic agent and require optimization. Crystallography driven structure-based drug design based on a series of ligand-protein crystal structures can be utilized to optimize urokinase inhibition. The properties of the protein crystals can affect the efficiency of structure-based drug design because a larger number of more accurate struc- FIG. 5. A, initial 2F o Ϫ F c (purple) and F o Ϫ F c (green) maps contoured at 1 and 3 , respectively, for the binding site of phenyl guanidine before refinement. B, molecular surface micro-urokinase as calculated by the program package QUANTA (Molecular Simulations Inc.) depicting interactions between B428 and micro-urokinase. The inhibitor and inhibitor surface are shown in orange, whereas the protein and protein surface are shown in cyan. C, overlay of the crystal structures of amiloride (purple) and phenyl guanidine (black) micro-urokinase, showing that the two scaffolds occupy different areas of the S 1 pocket. tures provides a better description of the relationship between binding interactions and binding energy. Fortunately, advances in molecular biology can be used to engineer the protein to obtain crystal systems that facilitate faster and more exact structure determinations and enhance the drug design cycle (47). Such a method has been used to design a crystal system for human urokinase for optimization of a urokinase inhibitor.
The sequence of LMW urokinase was redesigned to produce a new crystal form that would permit a more ideal system for structure-based drug design. Specifically, LMW urokinase was re-engineered to minimize the areas of disorder that may likely cause suboptimal crystal packing. This recombinant protein, micro-urokinase, produces crystals with close packing interactions at the A-chain cleft, which would be blocked in LMW urokinase. This close molecular packing results in crystals that diffract to high resolution on a rotating anode source (1.6 -2.0 Å). However, even though the micro-urokinase molecules are closely packed, the active site is both unoccupied and open to solvent channels in the crystal. This property readily allows compounds to be diffused into the crystal and has facilitated the determination of crystal structures in the presence of three reported urokinase inhibitors toward design of an anti-cancer agent.
The micro-urokinase crystal system and soaking method was used to determine the co-crystal structures of micro-urokinase complexed with the inhibitors B428 (25,26), amiloride (24), and phenylguanidine (27). Each of the co-crystal structures gives insight into favorable compound-protein interactions that contribute to the binding of these inhibitors to urokinase. The primary binding force is likely the hydrogen bonds between each inhibitor's amidine or guanidine group and Asp 189 . This salt bridge interaction is common to many guanidine or amidine complexes with trypsin or trypsin-like serine proteases such as thrombin, factor Xa, or tissue plasminogen activator (41)(42)(43)(44)(45) and is observed for Arg-P 1 in the Glu-Gly-Arg-chloromethyl ketone LMW urokinase structure (1). In addition to the hydrogen bonding interactions, van der Waals' packing between the core scaffold and the S 1 pocket may also contribute to the overall binding energy. Hydrophobic packing at the S 1 pocket is the primary binding interaction between substrates/ inhibitors in the chymotrypsin family of proteases where the S 1 pocket contains no charged groups (48 -51). Additionally, a series of thrombin inhibitors that lack a positively charged group to interact with Asp 189 have been described (52,53). Hence, both hydrophilic and hydrophobic interactions at the S 1 pocket contribute to the binding of B428, amiloride, and phenylguanidine, and these interactions are present in other crystal structures.
Examination of the urokinase structures reveals a new additional binding site adjacent to the S 1 pocket. The site, termed the S 1 ␤ subpocket, is composed of the disulfide bridge at Cys 191 -Cys 220 , residues Ser 146 and Gly 218 , and the side chain of Lys 214 . The S 1 ␤ subpocket is also present in the LMW urokinase structure (Protein Data Bank entry 1LMW) and is away from any re-engineered sites. The crystal structure of phenyl guanidine urokinase reveals that Gln 192 may act as a switch for the closing and opening of S 1 ␤. In the native and B428 or amiloride complex structures, the S 1 ␤ pocket is open, and Gln 192 is involved in two hydrogen bonds (Lys 143 and Tyr 151 ). However, in the presence of other inhibitors such as phenyl guanidine or Glu-Gly-Arg-chloromethyl ketone (1), the hydrogen bonds are broken, and the conformation of Gln 192 shifted such that its side chain is in van der Waals' contact with the inhibitor. In this conformation, the entrance to S 1 ␤ is blocked, and the shift is most likely induced to maximize interactions with the inhibitor. Hence, although the S 1 ␤ pocket may be blocked by the induced movement of Gln 192 , its proximity to S 1 makes it an attractive subsite for structure-based drug design.
The halogen atoms of B428 and amiloride are interacting with the entrance to the S 1 ␤ subsite (Gly 218 -Cys 220 ). Interactions at this site have been shown to confer a significant increase in inhibitory potency for the benzo(b)thiophene-2-carboxamidine series where the 4-iodo group (IC 50 ϭ 0.32 M) or 4-benzodioxolanyletheyl (IC 50 ϭ 0.07 M) inhibit more strongly than the 4-hydro compound (IC 50 ϭ 3.7 M) (25,26). The increase in potency observed for both substitutions is most likely due to packing interactions at the S 1 ␤ pocket. Phenylguanidine lacks a halogen atom to access the S 1 ␤ pocket, and examination of the structure reveals that the pocket can not be easily accessed by a direct substitution of the phenylguanidine ring. However, an overlay of the phenylguanidine crystal structure with that of amiloride reveals that the two scaffolds could be merged to form a 1-guanadyl naphthalene. This compound could, in turn, access the S 1 ␤ pocket. Hence, urokinase cocrystal structures with B428, amiloride, and phenylguanidine indicate that all three scaffolds may provide either direct or indirect access to the S 1 ␤ pocket. Furthermore, this newly described subsite has great potential for the future design of more potent urokinase inhibitors for the treatment of cancer.