Mechanism of Inactivation of Plasminogen Activator Inhibitor-1 by a Small Molecule Inhibitor*

The inactivation of plasminogen activator inhibitor-1 (PAI-1) by the small molecule PAI-1 inhibitor PAI-039 (tiplaxtinin) has been investigated using enzymatic analysis, direct binding studies, site-directed mutagenesis, and molecular modeling studies. Previously PAI-039 has been shown to exhibit in vivo activity in various animal models, but the mechanism of inhibition is unknown. PAI-039 bound specifically to the active conformation of PAI-1 and exhibited reversible inactivation of PAI-1 in vitro. SDS-PAGE indicated that PAI-039 inactivated PAI-1 predominantly through induction of PAI-1 substrate behavior. Preincubation of PAI-1 with vitronectin, but not bovine serum albumin, blocked PAI-039 activity while analysis of the reciprocal experiment demonstrated that preincubation of PAI-1 with PAI-039 blocked the binding of PAI-1 to vitronectin. Together, these data suggest that the site of interaction of the drug on PAI-1 is inaccessible when PAI-1 is bound to vitronectin and may overlap with the PAI-1 vitronectin binding domain. This was confirmed by site-directed mutagenesis and molecular modeling studies, which suggest that the binding epitope for PAI-039 is localized adjacent to the previously identified interaction site for vitronectin. Thus, these studies provide a detailed characterization of the mechanism of inhibition of PAI-1 by PAI-039 against free, but not vitronectin-bound PAI-1, suggesting for the first time a novel pool of PAI-1 exists that is vulnerable to inhibition by inactivators that bind at the vitronectin binding site.

Plasminogen activator inhibitor-1 (PAI-1) 3 is a member of the serine protease inhibitor (serpin) gene family and is the principal inhibitor of both urokinase type plasminogen activa-tor (uPA) and tissue type PA (tPA) (1)(2)(3). In healthy individuals, PAI-1 expression is low and generally confined to vascular smooth muscle cells, adipocytes, and the platelet precursor, megakaryocytes (4 -7). However, its expression in many other cell types, including vascular endothelial cells, hepatocytes, and the stromal cells surrounding tumors, can be strongly up-regulated by a number of stimuli, including cytokines, growth factors, and endotoxin (6 -11). PAI-1 is also stored in platelet ␣-granules where it can be released upon platelet activation (12). Thus, its local concentration at sites of vascular injury or inflammation can be very high relative to normal plasma levels (13,14).
Structurally, PAI-1 represents an example of the general serpin property of conformational plasticity. PAI-1 exists in multiple conformational states, including active, latent, proteasecomplexed, and cleaved forms (15). Active PAI-1 decays to the latent form with a half-life of 1-2 h at 37°C, and exposure to denaturants can partially return the latent form to the active state (16). The biological significance of the latent form remains unknown. Like latent PAI-1, the cleaved form is also inactive, but unlike latent PAI-1 this form cannot be reactivated. The labile nature of PAI-1 appears to be critical for its function as a protease inhibitor, and like other serpins, PAI-1 acts as a "suicide inhibitor" that reacts only once with a target protease. This is because during serpin inhibition the protease cleaves the serpin reactive center loop (RCL) to the point of an acyl⅐enzyme intermediate, and this cleavage is coupled with a major conformational rearrangement of the serpin involving RCL insertion into ␤-sheet A of the serpin forming an SDS-stable acyl⅐enzyme complex, in which the PAI-1 structure resembles the cleaved form (17)(18)(19).
In addition to uPA and tPA, PAI-1 also interacts with a number of non-protease ligands (1). These include heparin, the cell adhesion protein vitronectin, and members of the endocytic low density lipoprotein receptor family, such as the low density lipoprotein receptor-related protein, GP330/Megalin, and the very low density lipoprotein receptor (20 -26). These non-protease interactions are important for both PAI-1 localization and for its function (27)(28)(29)(30)(31)(32).
High plasma PAI-1 levels are associated with both acute diseases such as sepsis and myocardial infarction (8,33), and with chronic disorders, including cancer, atherosclerosis, and type 2 diabetes mellitus (34 -37). The association of PAI-1 with these syndromes has led to the suggestion that PAI-1 may contribute to the pathology of disease. However, the mechanistic role that PAI-1 plays in disease development is not clear and is likely to be complex, because PAI-1 can act through at least two distinctly different pathways, either by modulating fibrinolysis through the regulation of plasminogen activators, or by influencing tissue remodeling through the direct regulation of cell migration (31, 38 -41). The possibility that PAI-1 plays a direct role in a variety of diseases makes it an attractive target for small molecule drug development; however, the structural complexity of PAI-1 has made the identification and development of small molecule PAI-1-inactivating agents challenging (42)(43)(44)(45). Recently, a novel, orally active small molecule PAI-1 inhibitor, PAI-039, also called tiplaxtinin, has been described and examined in several animal models of disease where PAI-1 is thought to play a role (46 -52). These studies indicate that inhibition of PAI-1 activity is efficacious; however, the mechanism of action of this clinically relevant small molecule PAI-1 inactivating drug was not shown. Using PAI-039 as a molecular tool to investigate the inactivation of PAI-1, we have identified a mechanism suggesting the existence of a novel pool of PAI-1 vulnerable to inhibition by compounds that interact at the vitronectin binding domain.

EXPERIMENTAL PROCEDURES
PAI-039 was synthesized and characterized for chemical identity and purity at Wyeth Research (Collegeville, PA). Stock solutions were prepared in 100% Me 2 SO prior to dilution in assay buffer. Human native vitronectin, recombinant human wild-type PAI-1 (either non-glycosylated, expressed, and isolated from Escherichia coli or glycosylated, expressed, and isolated from insect cells), and rabbit polyclonal antibodies directed against human PAI-1 were from Molecular Innovations (Southfield, MI). High molecular weight human uPA was from Laboratories Serono S.A. (Aubonne, Switzerland). The chromogenic uPA substrate, pGlu-Gly-Arg-p-nitroanilide, was from Sigma, and the chromogenic tPA substrate, Pefachrome tPA, was from Centerchem Inc. (Norwalk, CT). Polystyrene 96-well Microtest Falcon plates were from BD Biosciences. All supplies and reagents for SDS-PAGE were from Bio-Rad. All supplies and buffers for binding assays with Biacore 3000 sensor chips were from Biacore AB (Uppsala, Sweden). All other chemical reagents used in the study were from Fisher Scientific (Pittsburgh, PA).
Direct PAI-1 Activity Assay with Chromogenic tPA/uPA Substrates-All direct chromogenic assays with purified human PAI-1 and site-directed mutants were performed as described previously (53,54). For determination of the effect of vitronectin on drug activity, 40 nM PAI-1 was incubated in 200 l of assay buffer, either alone (control), or with 200 nM vitronectin, or 4 M bovine serum albumin for 30 min at 23°C, after which 10 l of each sample was added to microtiter plate wells containing 80 l of PAI-039 in assay buffer for final concentrations before substrate addition of either 80, 60, 40, 30, 20, 10, 5, or 0 M PAI-039. The assay plate was briefly mixed on the plate reader, and then the mixture was incubated at 23°C for 15 min. Next 10 l of 50 nM uPA was added to each well, the plate was mixed briefly again, and the mixture was incubated for 5 min at 37°C before adding 100 l of substrate. The plate was then incubated at 37°C for 5 min before reading ⌬A 405 in kinetic mode for 10 min at 37°C. The data were plotted against the drug concentration and fit to an IC 50 equation (background corrected) using the GraFit Software program (Erithacus Software Ltd., Horley, UK).
Surface Plasmon Resonance Analysis-Direct binding of PAI-039 to PAI-1 and PAI-1 to native human vitronectin was monitored using a Biacore 3000 optical biosensor (Biacore AB, Uppsala, Sweden). Purified recombinant human latent and active PAI-1 were immobilized at the level of ϳ5000 response units (RU), passing working solutions of 20 g/ml in 10 mM sodium acetate, pH 5.0, through flow cells on the first chip. The second chip was immobilized with native vitronectin at the level of 600 RU under the same conditions. The reference flow cell surfaces were immobilized with ovalbumin at the level of 5000 RU on the first chip and 600 RU on the second chip, and both served as controls for nonspecific drug binding. Remaining binding sites on the chip surfaces were blocked by 1 M ethanolamine, pH 8.5. All binding reactions were performed in standard HBS-P (0.01 M HEPES, pH 7.4, 0.15 M NaCl, 0.005% v/v surfactant P20) buffer, pH 7.4 (Biacore, Uppsala, Sweden), containing 10% of Me 2 SO as running buffer. Binding of PAI-039 to PAI-1 was monitored at 25°C at a flow rate of 30 l/min for 2 min, followed by 2 min of dissociation with running buffer. Chip surfaces were regenerated with a 1-min pulse of 2 M NaCl, prepared in running buffer, followed by 2-min washing with running buffer to remove the high salt solution. All injections were performed using the Wizard Customized Application program in an automated mode. Control runs included the binding of several monoclonal antibodies raised against the human PAI-1⅐tPA complex. Direct binding of PAI-039 to immobilized PAI-1 was measured using 100ϫ dilutions of working solutions in running buffer over a range of concentrations (0.1 to 50 M). The binding of active and PAI-039-reacted human PAI-1 to vitronectin was studied at 10 nM PAI-1 and 0.08 -50 M PAI-039. All binding experiments were done at least twice and were corrected for background and bulk refractive index by subtraction of the reference flow cell and running buffer responses. All collected data were analyzed with BIAevaluation 3.1 Software (Biacore, AB, Sweden). For binding of PAI-039 to immobilized PAI-1, the apparent affinity was calculated from an equilibrium analysis of the maximum response (RU) at each drug concentration using single binding site model. The inhibition of PAI-1 binding to vitronectin by PAI-039 was analyzed by reacting a constant amount of PAI-1 with increasing concentrations of PAI-039 prior to SPR analysis, and determining the concentration of PAI-1 competent to bind vitronectin from the slope of the association phase of the binding curves. This analysis accurately measures the concentration of PAI-1 able to bind vitronectin, because our previous experiments performed with increasing concentrations of purified active PAI-1 in the absence of PAI-039 have shown that there is a linear relationship between the slope of the initial association phase of PAI-1 binding to vitronectin and the concentration of active PAI-1 (not shown).
Reversibility Assay-PAI-1 (9 nM) was incubated in an assay plate, with 80 M PAI-039 or vehicle in assay buffer (10 mM HEPES, pH 7.4, 150 mM NaCl, 0.05% Tween 20) at 23°C for 15 min prior to dilution. The concentrated PAI-1-PAI-039 or PAI-1-vehicle solutions were then serially diluted, 2-fold in parallel, in assay buffer. UPA (10 l of 100 nM) was then added to each well after dilution of PAI-1-PAI-039 (final volume 100 l), and the assay plate was incubated at 37°C for 5 min before adding substrate. After substrate addition the assay plate was incubated at 37°C for 5 min before reading ⌬A 405 in kinetic mode for 10 min at 37°C. The percent restoration of PAI-1 activity at each dilution was calculated from the ratio of the activity in the PAI-039-treated samples to the activity in control samples Molecular Docking-The previously determined structure for the active form of PAI-1 (PDB code: 1B3K) was used as an initial template. This is a stable mutant of PAI-1 that has four mutated residues N150H, K154T, Q319L, and M354I (55). All hydrogens and the four mutated residues were first minimized using the AMBER force field (56). Next PAI-039 was docked to PAI-1 using the induced fit docking protocol developed by Schrodinger. Induced fit docking is based on the docking program Glide with the refinement module in Prime (Schrodinger, Inc.), which was reported to accurately predict ligand binding modes and concomitant structural changes in a receptor. Compounds were initially docked using Standard Precision Glide at reduced van der Waals scaling against three types of proteins: (i) R1, original active form of PAI-1; (ii) R2, the active form of PAI-1 with Arg-76 mutated to Ala; or (iii) R3, the active form of PAI-1 with Arg-118 mutated to Ala. After initial docking, prime refinement and scoring were performed on the initial constructed protein-ligand complex. The residues within 5 Å of ligand poses were treated as being flexible. Both protonation states of carboxylic acids were considered for PAI-039.
Site-directed Mutagenesis-To identify the PAI-1 binding site for PAI-039, a variety of site-directed mutants was constructed using the QuikChange site-directed mutagenesis kit from Stratagene (La Jolla, CA) as described before (53). Based upon computer modeling, the following amino acid residues were chosen for substitution by alanine: Arg-76, Tyr-79, Lys-80, Thr-93, Thr-94, Asp-95, Phe-114, Arg-118, Thr-120, Lys-122, and His-143. All mutations were confirmed by sequencing throughout the entire PAI-1 coding region of mutant plasmids isolated from three clones of each individual mutant. The functional characterization of these mutants was based on the IC 50 of PAI-039 determined by direct chromogenic assay and by SPR analysis of vitronectin binding.

PAI-039 Binds to PAI-1 and Inhibits Its Anti-proteolytic
Activity against Both tPA and uPA-To examine the capacity of PAI-039 to block PAI-1 inhibitory activity, single step chromogenic assays were performed. For this analysis, purified recombinant PAI-1, either glycosylated or non-glycosylated, was preincubated with increasing concentrations of PAI-039 followed by the addition of uPA or tPA, and the remaining PAI-1 inhibitory activity was determined. These data are shown in Fig. 1, which demonstrates a dose-dependent inhibition of PAI-1 activity by PAI-039 with calculated IC 50 values for PAI-1 inactivation by PAI-039 from 9 to 12 M in this assay system ( Table  1). The inhibition profile was similar regardless of the PAI-1 target enzyme, or the glycosylation state of PAI-1, indicating that the effect is specific for PAI-1 and independent of the protease.
Direct binding studies of the interaction between PAI-1 and PAI-039 were also performed using SPR analysis. For these studies PAI-1 in either the active or latent conformation was coupled to a sensor chip, and the binding of PAI-039 was determined in a Biacore 3000. These data indicate that PAI-039 binds to PAI-1 in the active conformation ( Fig. 2A); however, no specific binding is observed with PAI-1 in the latent conformation (inset, Fig. 2A). Fig. 2A shows the SPR binding curves for PAI-039 to PAI-1, and Fig. 2B shows a plot of the maximal binding of PAI-039 at equilibrium versus the PAI-039 concentration. Fitting the data in panel B with a single binding site model yields an apparent K d for the binding of PAI-039 to active PAI-1 of ϳ15 M.
PAI-039 Binding to PAI-1 Is Reversible-The SPR data in Fig.  2A also demonstrate that the binding of PAI-039 to PAI-1 is  reversible, because the drug rapidly dissociates from PAI-1 during the wash phase. Therefore, to determine if the inactivation of PAI-1 by PAI-039 is also reversible, experiments were performed where PAI-1 was incubated with PAI-039 in the absence of plasminogen activators at a concentration that fully inactivates PAI-1 when plasminogen activators are present. After this preincubation step, the samples were diluted to the PAI-039 concentrations indicated in Fig. 3 and reacted with uPA. The restoration of PAI-1 anti-protease activity was then determined in a chromogenic assay as in Fig. 1. These studies demonstrate that the dilution of PAI-039 in the reaction mixtures restored PAI-1 inhibitory activity to levels consistent with the final concentration of drug in each sample (Fig. 3). This indicates that, like the binding of PAI-039 to PAI-1, its ability to block PAI-1 activity is also reversible in the absence of plasminogen activators. PAI-039 Blocks Covalent Complex Formation and Converts PAI-1 to a Substrate for PAs-In the absence of drug, PAI-1 forms stable covalent complexes with uPA or tPA of 93-105 kDa, respectively (Fig. 4, A and B). After pretreatment with increasing concentrations of PAI-039, the intensity of these higher molecular mass bands with both PAs decreased, indicating a reduction in complex formation due to drug-induced   (1:1) four times. UPA was then added to each sample and allowed to react with the PAI-1 for 5 min at 37°C before addition of chromogenic substrate. p-Nitroanilide release was monitored spectrophotometrically, and the ⌬A 405 for each sample was used to determine restoration of PAI-1 activity as a percentage of control activity without drug. This was then plotted versus the final drug concentration. The data were calculated from four independent experiments and represent the mean Ϯ S.E. PAI-1 inactivation (Fig. 4, A and B). Associated with the loss of high molecular weight complexes is a concomitant increase in the intensity of the band representing the cleaved, substrate form of PAI-1. This suggests that the predominant mechanism of inactivation by PAI-039 is to block productive inhibition of PAs by inducing turnover of the complex and cleavage of PAI-1 by the PAs as a substrate. At the highest PAI-039 concentration tested (80 M) there was a reduction in the cleaved PAI-1 band compared with the 40 M dose. This suggests that at high PAI-039 concentrations, well above the IC 50 and K d values, PAI-039 blocks PAI-1 activity by an alternate mechanism. The nature of this secondary mechanism is not clear; however, it is unlikely that this was due to the induction of a conformational change in PAI-1 to the latent form by the drug at these concentrations, because the data in Fig. 3 above indicated that even at these high concentrations of PAI-039 the inactivation of PAI-1 was reversible, whereas the conversion of PAI-1 to the latent conformation was not reversible under these conditions. Taken together, these data suggest the mechanistic basis for PAI-1 inactivation by PAI-039 is induction of substrate behavior. Thus, PAI-1 inactivation by the drug would remain reversible until the drugbound PAI-1 is reacted with a plasminogen activator where upon PAI-1 is cleaved as a substrate and the inactivation becomes irreversible. Consistent with this interpretation, no PAI-1 activity was restored in reversibility experiments similar to those shown in Fig. 3 when plasminogen activator was added prior to dilution of the drug-treated samples (data not shown).
PAI-1 Binding to PAI-039 or Vitronectin Is Mutually Exclusive-In blood, PAI-1 circulates bound to vitronectin, and this high affinity binding also facilitates the localization of PAI-1 within the tissue matrix. To determine if PAI-039 binding to PAI-1 could influence PAI-1 association with vitronectin, SPR experiments were performed with immobilized vitronectin. PAI-1 was first preincubated with increasing concentrations of PAI-039 followed by SPR analysis of PAI-1 binding to vitronectin. Fig. 5 shows the results of these studies; panel A demonstrates that the binding of PAI-1 to vitronectin is reduced as the PAI-039 concentration increases, and panel B shows the inhibition binding curves calculated from the slopes of the association phase of both glycosylated and non-glycosylated PAI-1 binding to vitronectin. The IC 50 values calculated  50 fit for glycosylated (OE) and non-glycosylated (F) PAI-1. The initial association phase data for each concentration of PAI-039 were linearly fit to determine the slope of each binding curve, and these data were converted to the concentration of residual active PAI-1 using the standard curve for active PAI-1 binding to vitronectin. (Control experiments indicated that there is a linear relationship between the slope of the initial association phase and the concentration of active PAI-1.) The amount of active PAI-1 in the presence of PAI-039 was expressed as percent active PAI-1 remaining, and these data were plotted to determine the IC 50 . B, data from two independent experiments are plotted against PAI-039 concentrations and fitted with IC 50 equation from GraFit. C, PAI-1 (40 nM) was incubated with 5-fold molar excess of human vitronectin (F), 100-fold molar excess of bovine serum albumin (▫), or assay buffer (E) for 30 min at room temperature. Each of these was diluted 10-fold into assay buffer containing PAI-039 and incubated for 15 min at room temperature. Residual PAI-1 activity was determined in a direct chromogenic assay using uPA as described under "Experimental Procedures." The data were collected in five independent experiments (mean Ϯ S.E.).
for the inhibition of PAI-1 binding to vitronectin by PAI-039 (ϳ13-17 M) were very similar to both the IC 50 s for inhibition reported above and to the K d for PAI-039 binding to PAI-1. This indicates that PAI-039 not only inhibits PAI-1 anti-protease activity but also blocks its association with vitronectin.
We also examined if the prior association of PAI-1 with vitronectin could influence the activity of PAI-039 on PAI-1.
For these studies PAI-1 was preincubated with either vitronectin or bovine serum albumin, and the inactivation of PAI-1 by PAI-039 was examined by the chromogenic assay for PAI-1 activity. These results indicated that a 5ϫ molar excess of vitronectin completely prevented the drug-induced inactivation of PAI-1, whereas preincubation of PAI-1 with a 100ϫ molar excess of bovine serum albumin had no effect on the ability of PAI-039 to inhibit PAI-1 activity (Fig. 5C). This implies that PAI-1 bound to vitronectin is protected from inactivation by PAI-039, and together with the data in A and B, these results suggest that the binding of PAI-039 or vitronectin to PAI-1 is mutually exclusive. We can speculate from these results that PAI-039 may bind near the PAI-1-vitronectin binding site and that, when bound, PAI-039 sterically blocks vitronectin binding to PAI-1. Likewise, if the order of addition is reversed and PAI-1 is first permitted to associate with vitronectin, then PAI-1 is protected from PAI-039 inactivation, possibly because the PAI-039 binding site is blocked by vitronectin.
Modeling of PAI-039 Binding to PAI-1 and Mapping of the Binding Site by Mutagenesis-Molecular modeling studies were next performed to search for potential binding sites on PAI-1 for PAI-039. The potential sites were searched based on the shape of the protein with the aid of "LigandFit" module of Cerius 2 (Accelrys, San Diego, CA). The method employs a cavity detection algorithm (57) for detecting invaginations in the protein as candidate binding site regions. With the additional information on the physicochemical properties, charge, hydrophobicity, and the ligand interactions, a hypothetical conformational binding site for PAI-039 in PAI-1 was identified in a cleft on the PAI-1 structure in the region of helices D and E, ␤-strand 2A, and the amino acid residues connecting helix E with the ␤-sheet strand 1A (Fig. 6). Several potential interactions were suggested in this region, including potential interactions between the carboxyl group of PAI-039 and Arg-76 and Arg-118, and between the trifluoromethyl group and the Asp-95 side chain and backbone. In addition, stacking interactions were observed between PAI-039 and Tyr-79. We also note that this putative binding cleft is located immediately adjacent to the previously identified vitronectin binding site on PAI-1 shown in yellow in Fig. 6A (20,25). This is consistent with the results shown in Fig. 5.
To examine the predictive value of the model, we performed extensive site-directed mutagenesis in this region, and the efficacy of PAI-039 toward each of these mutants was examined in Escherichia coli cell extracts. However, before performing analysis of the mutants we first examined the activity of PAI-039 against PAI-1 in E. coli cell extracts and compared this to its activity against purified PAI-1 (Fig. 7). These data demonstrated that the IC 50 of PAI-039 in cell extracts from E. coli expressing wild-type PAI-1 was identical to the IC 50 of PAI-039 against purified PAI-1. This indicated that we could examine the activity of PAI-039 against a large number of PAI-1 mutants without the need to individually purify each mutant. The results of these experiments are shown in Table 2, and a representative titration with one of the site-directed mutants (D95A) is shown in Fig. 7. These data demonstrate that several mutations in and around the proposed binding site increased the IC 50 of PAI-039 by Ͼ20%, suggesting that these mutants did not interact as strongly with PAI-039 as did wild-type PAI-1. These residues were R76A and Y79A in helix D and T93A, T94A, and D95A in ␤-strand 2A. The three most severely affected residues were R76A, T94A, and D95A, which all showed increases in IC 50 values of ϳ50%. Interestingly, two mutations actually reduced the IC 50 of PAI-039. These were T120A and K122A in the surface loop connecting helix E to ␤-strand 1A. Neither of these residues was predicted to interact directly with PAI-039 in our modeling studies, but they are located immediately adjacent to the cleft and may alter the conformation of the binding cleft. Alternatively, the larger side chains on the wild-type residues may partially shield access to the cleft by PAI-039 in wild-type PAI-1. Thus, comparison of the mutagenesis data with the modeling analysis in Fig. 6 suggests that the model likely represents a close approximation of the binding interaction; however, the precise orientation of the PAI-039 may be slightly different than shown, because the mutation of the Arg-118 has little effect on PAI-039 activity. Furthermore, the relatively modest effect of each mutation suggests that the drug binds to PAI-1 largely through hydrophobic interactions, because mutations that disrupt hydrogen bonds or ionic interactions would be expected to show significantly greater increases in IC 50 values. This result is also consistent with the hydrophobic nature of PAI-039.

DISCUSSION
PAI-1 inhibits the serine proteases tPA and uPA, and thereby regulates fibrinolysis and tissue remodeling, affecting diverse pathologies, including thrombosis and atherosclerosis. PAI-1 has therefore been termed a common denominator in cardiovascular and metabolic disease, not only because it occupies a central mechanistic position regulating the initiation of proteolysis, but also due to its significant synthesis by platelets, smooth muscle cells, and adipocytes. Thus, PAI-1 inhibitors could have potential therapeutic benefit in many diseases where plasma and tissue levels of PAI-1 are elevated, including deep vein thrombosis, cancer, and type 2 diabetes. Monoclonal antibody inhibitors of PAI-1 have been characterized (58), but their demonstrated efficacy in preclinical models of disease is limited, and use in humans is often excluded because of poor pharmacological and immunological properties. Therefore, the development of effective small molecule PAI-1 inactivating agents with oral bioavailability and oral efficacy will greatly improve the potential for targeting PAI-1 for therapeutic intervention. Previously we described the characterization of a benzofuran derivative, WAY-140312, which was the first small molecule PAI-1 inhibitor to exhibit oral efficacy (45). A different chemical scaffold was used to develop a second generation of molecules, of which PAI-039 (tiplaxtinin) is representative.
PAI-1 initially forms a reversible Michaelis complex with its target protease, followed by irreversible inactivation induced by Purified PAI-1 (PP) and PAI-1 in bacterial cell extracts (CE) was first titrated with uPA, then the amount of PAI-1 determined to inhibit 80% of 5 nM uPA (final concentration) was preincubated with PAI-039 for 15 min at room temperature. Residual PAI-1 activity was then determined in a direct chromogenic assay using uPA as described under "Experimental Procedures." The data were collected in three independent experiments (mean Ϯ S.E.).

TABLE 2 IC 50 values for the inhibition of PAI-1 activity by PAI-039 presented as the percent change in activity relative to wild-type PAI-1 (WT)
The IC 50 values were calculated for the reaction of non-glycosylated PAI-1 with uPA in a direct chromogenic assay.

Region
Mutation a Mutants that showed differences in reactivity with PAI-039 Ͼ20% compared to wild-type PAI-1.
insertion of the PAI-1 RCL into ␤-sheet A of PAI-1. This complicated mechanism of protease inhibition is subject to interference at several levels, and monoclonal antibodies have been shown to inactivate PAI-1 by three different mechanisms: 1) by preventing formation of the Michaelis complex; 2) by accelerating the transition to the latent conformation; or 3) by inducing turnover of the PAI-1⅐protease complex as a substrate (59 -61). Although fewer studies have characterized small molecule inactivators, several important investigations have been published. The binding site for the anthranilic acid derivative AR-H029953XX was modeled in a manner similar to the approaches we have used but employed fewer mutants. Because the acidic function was required for activity, Bjorquist et al. (42) correctly anticipated that binding would occur in a hydrophobic cleft near residues Phe-114 and Val-121. Mutations made specifically in this region identified Arg-76, Arg-115, and Arg-118 as necessary for the inhibitory activity of the molecule. Using compounds with intrinsic fluorescence, a similar hydrophobic region in the flexible joint of PAI-1 was identified by Egelund et al. (43) as a binding site for a variety of compounds, including AR-H029953XX. This group also observed that the activity of these neutralizers was reduced in the presence of vitronectin. Further characterization of the interaction of these compounds with PAI-1 was complicated by their weak inhibitory potency and physical chemical properties, and depending on the assay conditions, inactivation was either through conversion to the substrate form or by polymerization of PAI-1.
In our studies PAI-039 was used at concentrations below the critical micelle concentration (not shown), allowing analysis of mechanism without the complication of secondary physical chemical effects on PAI-1. The binding of PAI-039 was within the same hydrophobic region of PAI-1 and like AR-H029953XX involved the Arg-76 residue, but our modeling and mutagenesis analysis suggests that PAI-039 bridges this hydrophobic cleft to the Asp-95 residue on the edge of ␤-sheet A. The inactivation of PAI-1 at PAI-039 concentrations near the IC 50 therefore appears to occur not by prevention of the interaction between PAI-1 and the protease, but by inhibiting the formation of a stable covalent complex, and through the conversion of PAI-1 to a substrate for PAs. This could be due to the ability of PAI-039 to hinder the conformational change in ␤-sheet A that is necessary to accommodate the insertion of the PAI-1 RCL during the protease inhibition reaction (17,18). The localization of the PAI-039 binding site in the cleft at the edge of ␤-sheet A would be consistent with this mechanism as would the reversibility of PAI-039, because the conformational change in ␤-sheet A and RCL insertion are not thought to occur until PAI-1 interacts with a protease. This model is also consistent with the mutual exclusion of PAI-039 and vitronectin binding, because the vitronectin binding site is localized immediately adjacent to this binding cleft on the edge of ␤-sheet A.
Our experiments have also extensively characterized other novel aspects of the interactions of PAI-039 with PAI-1. For example, we show that inhibition occurs equally with non-glycosylated and glycosylated PAI-1. We also found that PAI-039 bound reversibly, and its activity was affected by the presence of vitronectin. The effect of vitronectin was specific, because 100fold molar excess of bovine serum albumin had no effect on the interaction with PAI-1. In their analysis, Egelund et al. (43) hypothesized that the vitronectin-neutralizing effect seen with the compounds that they used could hamper in vivo activity, because most PAI-1 circulates bound to vitronectin. Consistent with this logical hypothesis we found that preincubation of PAI-1 in plasma from PAI-1 null mice protected the PAI-1 from inactivation by PAI-039, whereas PAI-1 preincubated in plasma from double vitronectin/PAI-1 null mice was not protected (not shown). Nevertheless, PAI-039 has been shown to exhibit potent efficacy in vivo, with a hemostatic profile of normal coagulation, but with accelerated fibrinolysis, which is similar to that observed following vascular injury in PAI-1 knockout mice (46,47,50,51). PAI-039 has also been shown to block the inhibition of cell adhesion and migration by PAI-1 in cell culture systems and to inhibit angiogenesis in vivo in a PAI-1dependent manner (52). Although there is always a possibility that PAI-039 is working through some as yet undiscovered mechanism, it is clear that it is a potent in vitro inhibitor of free PAI-1 that exhibits excellent properties for a drug candidate. Therefore, an alternative explanation of the observed in vivo activity could be based upon the dynamic interaction between PAI-1 and vitronectin and the fact that normal turnover creates a novel pool of PAI-1 susceptible to attack by PAI-039. This is similar to what has been shown in vivo by infusion of a dominant negative PAI-1 capable of binding vitronectin but without the ability to inhibit its target proteases (62)(63)(64). Treatment with this mutant PAI-1 showed a significant in vivo effect in a model of severe renal fibrosis by competing vitronectin-bound active PAI-1 with vitronectin-bound inactive PAI-1, indicating that wild-type PAI-1 activity can be blocked in vivo even in the presence of normal vitronectin levels. These data suggest that the PAI-1 vitronectin interaction is dynamic in vivo and thus that PAI-1 is susceptible to inactivation in vivo. These data are further supported by studies in human adipose tissue, which is considered the largest tissue source of PAI-1. Human adipocytes actively synthesize PAI-1, but not vitronectin, implying that PAI-1 is secreted from these cells in the free conformation (65). Thus, we speculate that, in normal individuals where tissue synthesis of PAI-1 is low, and plasma PAI-1 circulates in complex with vitronectin at sub-nanomolar levels, the drug would be expected to have a minimal effect on PAI-1 activity; however, in situations where PAI-1 synthesis is elevated in tissues, such as adipose, as for example in type 2 diabetes (66), the drug would be expected to produce a significant inhibitory effect. Although this hypothesis of an adipose tissue-specific micro-environment for susceptible PAI-1 is novel, the concept is similar to that proposed by Rosenberg and Aird for the vascular-bed-specific regulation of hemostasis (67). Our studies suggest that a similar concept may apply to the interaction of PAI-1 and vitronectin in the local tissue milieu. Finally, it is worth noting that the interaction of PAI-1 with PAI-039 is also dynamic, and PAI-039 binds active PAI-1 reversibly. This suggests that if the binding of PAI-039 to PAI-1 remains reversible in the presence of a protease, while the inactivation becomes irreversible due to the cleavage of the PAI-1 RCL by the protease, then this unique mechanism of reversible binding and irreversible inactivation may permit PAI-039 to act catalytically against multiple PAI-1 molecules.