Dual-reporter high-throughput screen for small-molecule in vivo inhibitors of plasminogen activator inhibitor type-1 yields a clinical lead candidate

Plasminogen activator inhibitor type-1 (PAI-1) is a serine protease inhibitor (serpin) implicated in numerous pathological processes, including coronary heart disease, arterial and venous thrombosis, and chronic fibrotic diseases. These associations have made PAI-1 an attractive pharmaceutical target. However, the complexity of the serpin inhibitory mechanism, the inherent metastability of serpins, and the high-affinity association of PAI-1 with vitronectin in vivo have made it difficult to identify pharmacologically effective small-molecule inhibitors. Moreover, the majority of current small-molecule PAI-1 inhibitors are poor pharmaceutical candidates. To this end and to find leads that can be efficiently applied to in vivo settings, we developed a dual-reporter high-throughput screen (HTS) that reduced the rate of nonspecific and promiscuous hits and identified leads that inhibit human PAI-1 in the high-protein environments present in vivo. Using this system, we screened >152,000 pure compounds and 27,000 natural product extracts (NPEs), reducing the apparent hit rate by almost 10-fold compared with previous screening approaches. Furthermore, screening in a high-protein environment permitted the identification of compounds that retained activity in both ex vivo plasma and in vivo. Following lead identification, subsequent medicinal chemistry and structure–activity relationship (SAR) studies identified a lead clinical candidate, MDI-2268, having excellent pharmacokinetics, potent activity against vitronectin-bound PAI-1 in vivo, and efficacy in a murine model of venous thrombosis. This rigorous HTS approach eliminates promiscuous candidate leads, significantly accelerates the process of identifying PAI-1 inhibitors that can be rapidly deployed in vivo, and has enabled identification of a potent lead compound.

Plasminogen activator inhibitor type-1 (PAI-1) 3 is a member of the largest and most widely distributed family of protease inhibitors, the serine protease inhibitor (serpin) family (1). It is the physiologic inhibitor of tissue and urokinase plasminogen activator (tPA and uPA). In normal physiology, PAI-1 regulates processes such as fibrinolysis and wound healing. However, elevated PAI-1 levels have been associated with a myriad of both acute and chronic disorders, including thrombosis; fibrosis of the lung, liver, and kidney; and atherosclerosis (2)(3)(4)(5). PAI-1 has also been implicated in the development of metabolic dysregulation and diabetes (6 -10) and aging (11,12). Increased expression of PAI-1 is associated with a poor prognosis in many cancers, including breast, colorectal, and neurological cancers (13). However, the mechanism(s) by which PAI-1 may contribute to metabolic disorders, aging, or cancer remains unclear. Together, these findings suggest that therapeutic inactivators of PAI-1 may have significant benefit in the treatment of a wide variety of diseases and could provide important tools with which to decipher the biological underpinnings of PAI-1's role in disease.
Although a number of small-molecule inhibitors of PAI-1 have been reported to be active in vitro, the majority of these ligands display reduced efficacy in vivo. For instance, tiplaxtinin (PAI-039) and the natural product, annonacinone, inhibit PAI-1 function in vitro, but do not target vitronectin-bound PAI-1, the predominant form of PAI-1 in vivo (14,15). Another PAI-1specific synthetic inhibitor, PAI-749, showed no effect on thrombus formation in human blood ex vivo despite its potency by in vitro and ex vivo preclinical evaluation (16). Additionally, although bis-ANS inhibits PAI-1 with low micromolar affinity in vitro, it lacks the specificity necessary for in vivo applications, due to nonspecific binding to hydrophobic protein surfaces (17,18). Additional inhibitors that have been characterized in vitro are peptides (19) that will likely encounter stability issues in vivo. To identify new inhibitors, many high-throughput screens (HTSs) have been performed (20,21). We previously described a pilot screen in which we identified a novel class of polyphenolbased PAI-1 inactivators (22). Despite their nanomolar affinity for PAI-1 in vitro, high concentrations of these ligands were required for inactivation in vivo, introducing an increased potential for off-target interactions. Furthermore, attempts to increase in vivo efficacy of the polyphenols by chemical modification have not yet yielded scaffolds that retain significant activity in vivo. 4 A recent series of acylanthranilic acid derivatives have shown efficacy in animal models of disease (23)(24)(25); however, their activities against vitronectin-bound PAI-1 have yet to be reported. Together, this limited collection of PAI-1 inhibitors indicates the need for further screens of diverse small-molecule libraries to identify novel inactivators of PAI-1 for in vivo applications and for the potential development of novel therapeutic candidates. However, screening platforms to date have been only marginally successful due to high incidences of fluorometric and colorimetric interference and nonspecific protein binding by lead compounds in follow-up studies (26).
To specifically address these two issues in screening for inactivators of serpins and PAI-1 in particular, we developed a dual-spectroscopic reporter screening system designed to markedly reduce the rate of compounds that falsely test positive through spectroscopic interference by mimicking the reporter substrates and to enrich for hits that retain activity in a high protein background by eliminating compounds with promiscuous binding. A compound was classified as a PAI-1 inactivator only upon the generation of two distinct spectroscopic signals that correlate with reduced PAI-1 activity. Over 152,000 purified compounds and 27,000 natural product extracts were screened with this HTS. This approach yielded a 5-13-fold reduction in the rate of apparent positive compounds compared with our previous screens, accelerating the confirmation process and promoting efficient follow-up studies. Furthermore, one PAI-1 inactivator identified by the screen was directly active in ex vivo plasma and in vivo without further chemical synthetic manipulation, indicating that screening in high-carrier protein buffers may streamline the process to studies of in vivo efficacy. Here, we report the development of the dual-reporter system, the execution of this system in an HTS against PAI-1, and the evaluation of two novel PAI-1 inactivators in vitro and in vivo.

A colorimetric/fluorescent dual-reporter pair for measuring PAI-1 activity
An aminomethylcoumarin (AMC) fluorescent substrate for uPA was chosen as the first reporter of PAI-1 activity for the screening platform due to its retention of signal in a high pro-tein background (Table S1). However, due to the challenge of identifying two nonoverlapping fluorescent reporters whose signals are retained in serum albumin, we sought to adapt a colorimetric substrate to pair with the AMC substrate. Colorimetric reporters are able to quench the intrinsic fluorescence of white plastic microplates, and this observation has been employed previously for HTS (27,28). This phenomenon is observed with yellow chromophores, including the common reporter used in protease substrates, p-nitroaniline (pNA). To confirm this effect, cleavage of 50 M L-pyroglutamyl-glycyl-Larginine-p-nitroaniline hydrochloride (pEGR-pNA) by uPA was first tested in HEPES-buffered saline (HBS). The pNA substrate alone or in a mixture with 5 nM uPA was incubated for 30 min in a white 96-well microplate, after which the fluorescence spectrum of the microplate was recorded (excitation 430 nm, emission 450 -550 nm). Upon pEGR-pNA cleavage by uPA, pNA quenched the white-plate fluorescence (WPF) by 60.2% ( Fig. 1A and Table 1). Because we were interested in pairing this pNA substrate with AMC to develop a dual-reporter system, we tested whether cleavage of 10 M Z-glycyl-glycyl-arginine-7amido-4-methylcoumarin hydrochloride (ZGGR-AMC) substrate by uPA could also be monitored in white microplates by recording its fluorescence spectra (excitation 385 nm, emission 420 -520 nm) (Fig. 1B). The quench in WPF by pNA was only modestly decreased (54.5%) in the presence of 15 mg/ml BSA ( Table 1). As expected, AMC fluorescence was quenched nearly 7-fold in HBS relative to HBS with 15 mg/ml BSA (compare the fluorescence change of 200.6-fold in HBS alone with 28.6-fold

Dual-reporter high-throughput screen for inhibitors of PAI-1
in HBS with BSA; Table 1) but still retained a high signal above background. Together, these data indicate that cleavage of pNA and AMC substrates is easily detectable via fluorescence in the BSA-containing buffer and that they might be paired as a dualreporter system to monitor uPA activity and therefore PAI-1 inhibition.
To use pEGR-pNA and ZGGR-AMC in a single mixture, it was first necessary to investigate potential spectroscopic interference. Excitation of the white plate at 430 nm falls outside of the AMC excitation spectrum. Likewise, excitation of AMC at 380 nm falls outside of the known white plate excitation spectrum. Therefore, we predicted minimal interference between the reporters. To confirm this, the effect of the AMC fluorescence on WPF was recorded (excitation 430 nm, emission 470 nm), and as expected, AMC had no effect on WPF at the wavelength required to observe quenching by pNA ( Fig. 2A, filled bars). In addition, the WPF showed no signal in the range at which AMC fluorescence is recorded (excitation 385 nm, emission 460 nm) (Fig. 2B, open bars). Importantly, in the absence of uPA, no quenching of WPF by the pNA substrate was observed, and no AMC signal was detected. Likewise, inhibition of uPA with 10 nM PAI-1 prior to substrate addition yielded only background signal levels (Fig. 2, A and B).
Because the goal of the dual-reporter system is to utilize a mixture of two reporter substrates, it was also necessary to confirm that both reporters retained their signals when mixed. To examine this, turnover of the pEGR-pNA and ZGGR-AMC by 5 nM uPA was allowed to reach completion in separate reactions, after which point the two cleaved reporters were mixed and fluorescence was recorded. The cleaved pNA substrate alone or in a mixture with AMC quenched WPF by 58.1 Ϯ 1.0% and 59.0 Ϯ 0.9%, respectively, indicating that the addition of AMC had no effect on the WPF readout (Fig. 2C). However, the -fold increase in fluorescence by the AMC reporter alone was modestly reduced from 56.2-to 35.6-fold in the pNA/AMC mixture (Fig. 2D). This result might be expected, as the pNA absorbance spectrum (29) overlaps with the emission of the AMC (30). Despite this overlap, a large signal over background is retained in the pNA/AMC mixture. Taken together, these data indicate that WPF and AMC can be appropriately paired for measuring residual protease activity, and therefore PAI-1 activity, in a single mixture in white microplates.

Validation of the dual-reporter system
Next, the assay was optimized for a 384-well microplate format for screening large-compound libraries (Figs. S2 and S3) (details provided in supporting Results). Prior to executing the high-throughput screen, we sought to spectroscopically confirm the use of the dual-reporter system as well as to validate the addition of BSA to the screening platform ( Fig S4). First, tiplaxtinin (PAI-039) (Fig. 3A) was used to confirm that the dualreporter assay detects PAI-1 inactivators in a dose-dependent manner. Tiplaxtinin is a well-characterized PAI-1 inactivator with an IC 50 of ϳ10 M in buffered conditions (14). Tiplaxtinin was incubated for 15 min with PAI-1, followed by the addition of uPA and the pNA/AMC substrate mixture in a low-volume 384-well format. As expected, upon increasing the concentration of tiplaxtinin in HBS, WPF decreased due to quenching by free pNA as PAI-1 was dose-dependently inactivated (Fig. 3C). Simultaneously, AMC fluorescence increased, and the dual-reporter readout resulted in nearly identical IC 50 values of 28 -29 M, which are comparable with previously reported values (14).
As described above, by combining the WPF and AMC reporters, we predicted that compounds that show a change in only one of the reporter signals are likely not active against PAI-1 but instead are compounds that mimic the spectroscopic properties of one of the reporters. To test this, we used curcumin (Fig. 3B), a yellow chromophore with a similar UV absorption spectrum as pNA. In an earlier screen, curcumin was shown to have no effect on PAI-1 activity (data not shown), making it an ideal candidate to test reporter mimicry for assay validation. PAI-1 was incubated with 30 M curcumin prior to uPA addition, after which the assay was carried out as described in low-volume 384-well format. Upon reading the WPF, curcumin appears to inhibit PAI-1 activity, quenching the WPF to the same extent as pNA in the positive control (Fig. 3D). However, no change in AMC fluorescence was observed (Fig. 3E), indi- , only pNA quenches WPF (54%), whereas AMC had no effect on the fluorescence of white microplates. B, similarly, pNA showed no fluorescence at 385/460 nm, whereas a 24.6-fold change in AMC fluorescence was observed. Importantly, preincubation of uPA with PAI-1 returns each of these signals to background levels. C and D, cleavage of uPA and AMC substrates by uPA was allowed to reach completion, after which the reporters were mixed with either buffer or other reporter to determine the extent of spectroscopic interference. C, AMC had no effect on the ability for pNA to quench WPF. D, in the presence of pNA, the change in AMC fluorescence (35.6-fold) is modestly reduced relative to AMC alone (56.2-fold). This may be expected, as the pNA absorbance spectrum slightly overlaps with the emission spectrum of AMC. All assays were performed in triplicate, and data are shown with means Ϯ S.D. (error bars).

Dual-reporter high-throughput screen for inhibitors of PAI-1
cating that curcumin did not inhibit PAI-1 activity. It is possible that curcumin, or other chromophores present in the screening libraries, could quench the AMC fluorescence, as was observed with pNA ( Fig. 2D), and some potential PAI-1 inactivators could be missed using this stringent criterion for defining a hit. However, the incidence of falsely high hit rates when only one reporter is used presents a significant challenge for efficient follow-up and confirmation, and we anticipate the benefits of reducing high rates of false positives will outweigh the small potential for false negative results. Thus, for our HTS, an initial readout like that described with curcumin would not be considered an active compound and would not be followed up for further study.
Finally, to validate the use of BSA in the HTS platform, we analyzed the MicroSource Spectrum 2000 (MS2000) library used in a screen described previously (22). In follow-up studies of this prior screen, we determined that many of the hits from the MS2000 library were inactive in plasma due to nonspecific protein binding (data not shown). Indeed, the synthetic derivative CDE-066 based upon the polyphenol scaffolds identified in the pilot screen showed a 3,200-fold decrease in potency in plasma relative to buffered conditions (22). Thus, we predicted that screening the MS2000 side-by-side in HBS and HBS containing 15 mg/ml BSA would result in distinct hit sets. Namely, in HBS, we expected that the same hits would be identified as described previously (22), but that these compounds would be inactive when screened in more stringent conditions in 15 mg/ml BSA. A subset of compounds from the two MS2000 screens is shown in Fig. 3F. Indeed, when screening in HBS, we found nearly identical hits as our pilot screen, which included the polyphenols tannic acid, epigallocatechin-digallate, and hexachlorophene (Fig. 3F). Moreover, as predicted, each of these hits was inactive in the MS2000 screen containing 15 mg/ml BSA (Fig. 3G). Together, these data indicate that the use of BSA in the dual-reporter screening platform might enrich for compounds whose activity is retained in high-protein environments. We predict that these hits should display a more efficient transition into plasma and in vivo settings. Finally, revisiting the MS2000 library further validates the use of the dual-reporter screen, as we observed changes in both the WPF and AMC signals in expected hits in buffered conditions. ). Curcumin, a yellow chromophore that does not influence the activity of PAI-1, was used to validate that the WPF/AMC dual-reporter assay is insensitive to colorful, noninhibitory compounds. D, the pNA readout reveals quenching of WPF, suggesting that curcumin may have inactivated PAI-1. E, the AMC readout, however, resulted in no change in reporter fluorescence, indicating that curcumin indeed has no effect on PAI-1 activity. This type of result, where only one reporter signal is affected, was considered a nonactive compound in screening follow-up. A previously screened pilot library (22) was rescreened in HBS without (F) and with (G) 15 mg/ml BSA using the dual-reporter assay. AMC signals are shown in closed circles, whereas WPF signals are shown in open circles. True hits from the initial pilot screen (dashed ovals) were similarly identified here in HBS, validating the use of the pNA/AMC mixture. As expected, none of these hits were active in BSA, indicating that the dual-reporter system in BSA may enrich for PAI-1 ligands that retain function in excess protein. All assays except for screens (F and G) were performed in n ϭ 3-6 replicates as shown.

Dual-reporter high-throughput screen for inhibitors of PAI-1 Primary high-throughput screening with a dual-reporter system reveals reduction in hit rates
The final HTS protocol is described in the legend to Fig. 4. Briefly, 200 nl of a compound was added per well, yielding a compound concentration of ϳ32 M in 3.2% DMSO in the presence of 15 nM PAI-1 in HBS-BSA. Following a 15-min incubation, 3 l of 15 nM uPA in HBS-BSA was added per well, for final concentrations of 10 nM PAI-1 and 5 nM uPA. A 2:1 PAI-1/uPA ratio was chosen to enrich for the most active compounds, as greater than half of the PAI-1 must be inactivated before a signal is generated. After an additional 15-min incubation, 3 l of the pNA/AMC substrate mixture was added in HBS-BSA to yield final concentrations of 200 M pEGR-pNA and 100 M ZGGR-AMC. The mixture of uPA and substrates in the absence of PAI-1 served as a positive control, whereas the negative control consisted of the mixture of uPA and substrates along with PAI-1. Following a 90-min incubation to allow for substrate turnover, quenching of WPF by pNA (excitation 430 nm, emission 470 nm) and AMC fluorescence (excitation 380 nm, emission 470 nm) were recorded.
Using this HTS protocol, 152,899 purified compounds were screened from 15 different collections in the University of Michigan Center for Chemical Genomics (CCG). Compounds were considered a hit if observed changes in both the WPF and AMC signals were greater than three S.D. values (Ͼ3 S.D.) from the negative control. Compounds that displayed a change in only one of the reporter signals were not considered for further evaluation. An example of these hit criteria are shown with a subset of 100 compounds in Fig. 5A. Hits are circled where both the WPF and AMC signals display a change. In contrast, compounds that were likely mimicking the spectroscopic properties of the protease reporters, wherein only one of the signals showed a change, are boxed. The average Z-factor values (ZЈ), which serve as a statistical gauge for the quality of the HTS assay for WPF and AMC were 0.72 and 0.68, respectively, indicating . The high-throughput WPF/AMC dual-reporter system. PAI-1 in 15 mg/ml BSA was incubated for 15 min in the presence of a small molecule, followed by the addition of uPA. After 15 min, a mixture of pEGR-pNA and ZGGR-AMC substrates was added, and the final mixture was incubated for 90 min to allow for substrate turnover. Dual fluorescence was read (pNA quenching of WPF using excitation/emission of 430/470 nm; AMC fluorescence using excitation/emission of 380/470 nm). Compounds that inactivate PAI-1 such that residual uPA activity results in a change in signal greater than Ͼ3 S.D. away from the mean of the negative control in both the WPF and AMC signals are considered a hit for PAI-1 inactivation. Figure 5. Primary screening with the dual-reporter system reveals a dramatic reduction in levels of promiscuous hits and/or spectroscopic interference. A, compounds were considered hits if a change in both the WPF and AMC signals was observed (dashed ovals), whereas compounds that displayed a change in only one signal were considered nonactive (solid boxes). A subset of 100 compounds from the primary screen is shown to exemplify these hit criteria, where AMC signal is shown in filled circles and WPF signal is shown in open circles. B, the AMC signal alone of a 2,000-compound subset of the primary screen displays a Ͼ20% hit rate. C, the same 2,000 compounds are shown with the added WPF hit criteria, where the single set of gray circles represents compounds that show anti-PAI-1 activity via both AMC and WPF. A comparison of hit rates for single-and dual-reporter HTS systems for the entire screen is shown in Table 2.

Dual-reporter high-throughput screen for inhibitors of PAI-1
that the WPF/AMC dual-reporter system is a highly reliable screen, because a Z value between 0.5 and 1.0 is considered to denote an excellent assay (31). Because the WPF and AMC fluorescence reads were recorded individually, the data from each signal could be compared separately to explore the hit rates using pNA or AMC versus the dual-reporter system. As expected, we observed an unacceptably high hit rate for each reporter alone. Analysis of the AMC reporter alone revealed that 20.3% of compounds displayed a change in signal Ͼ3 S.D. from the negative control (Table 2). A representative subset of 2,000 compounds from this analysis is shown in Fig. 5B, demonstrating this excessively high hit rate for a primary screen using only an AMC reporter. For the WPF reporter only, the hit rate was lower than with AMC but still unacceptably high at 8.7% (Table 2). Upon applying both signals for the hit criteria, the overall hit rate was significantly reduced to 1.5% (2,363 compounds) ( Table 2). This compared with a hit rate of 7.2-18.8%, depending upon which a single reporter was used. The dramatic reduction in hit rate is demonstrated in Fig. 5C, where the same subset of 2,000 compounds are shown but with the added criterion that a change in WPF signal greater than 3 S.D. from the negative control was also required in addition to the change in AMC signal. Together, these HTS results demonstrate the usefulness and efficiency of applying a dual-reporter system compared with a single-reporter assay for ruling out compounds that spectroscopically mimic a reporter signal.
Based upon the reduction in hit rates observed with the purified compound collections (Table 2), we predicted that the dual-reporter system may be useful in screening natural product extracts (NPEs) as natural product collections often display significant spectroscopic interference with both fluorescence and colorimetric HTS assays (32). To the best of our knowledge, few NPE screens have been executed against PAI-1, perhaps reflecting this spectroscopic challenge, and leaving this extremely valuable resource for identifying diverse PAI-1 inactivators untapped. By applying the dual-reporter system to screen NPEs, we hypothesized that the incidence of spectroscopic interference might be greatly reduced and, thus, the efficiency of follow-up studies enhanced. Toward this goal, 27,739 NPEs were screened in the CCG using the HTS protocol described (Fig. 4). Similar to analysis of the purified compound primary screen, single-reporter data could be compared with the dual-reporter readouts. As expected, we observed a high rate of 9.4% upon analyzing the WPF data alone (Table 2). However, with the AMC reporter alone, we encountered the surprisingly high hit rate of 67.5%, indicating that the majority of NPEs are fluorescent to some degree in the same wavelength range as AMC. Analysis of the NPE primary screen using both the WPF and AMC reporters resulted in an overall hit rate of 7.5% (1,884 NPEs), suggesting that most of the AMC reporter screen hits were due to NPE fluorescence at 460 nm. Although 7.5% may seem like a significantly high hit rate, the NPE libraries in the CCG consist of multiple extraction conditions for each strain. Thus, the true hit rate is likely significantly less than 7.5%. Confirmation of the 1,884 NPEs was carried out as described under "Experimental procedures." The confirmation round of NPE screening yielded 117 extracts, and a follow-up study of a subset of the most potent of these extracts is under way.
Confirmation and counterscreen testing of the remaining 2,363 purified compounds generated in the primary screen was performed as described under "Experimental procedures." Briefly, because all confirmation and counterscreen data were performed in triplicate, we chose a compound for doseresponse only if it displayed a signal (Ͼ3 S.D.) in two of the three wells of the confirmation screen and also showed no activity (Ͻ3 S.D.) in two of the wells in the counterscreen. Based on these criteria, 395 compounds were chosen for dose-response testing in the CCG (Table S2). Thirty-nine compounds passed dose-response testing and were further scrutinized with regard to known toxicities, promiscuity in CCG screens, pharmacologic structural characteristics, and follow-up manual doseresponse testing of commercially available compounds. Only two of the 39 compounds, CCG-6920 and CCG-7844 (Fig. S1, A  and B), satisfied all of the aforementioned criteria and were further investigated as lead compounds. We propose that this final low confirmation rate consisting of two compounds is due to the highly stringent conditions under which PAI-1 was screened, including the dual-reporter system containing a high BSA concentration and with a 2:1 PAI-1/uPA ratio. Further, a pre-read was performed during confirmation testing (see "Experimental procedures"), which serves to exclude compounds that spectroscopically mimic both pNA and AMC. These compounds would not have been excluded in the primary screen. However, to simplify our confirmation analysis, the pre-read data were also used to exclude compounds that alone displayed greater than a 3 S.D. change in signal from the negative controls with both reporters. Thus, highly colored and/or fluorescent compounds that inactivated PAI-1 were excluded in follow-up analysis, and several hits could have been lost at this stage.

Characterization of novel PAI-1 inactivators in plasma and in vivo
Follow-up testing of the two hits from dose-response testing revealed that each compound was active in 15 mg/ml BSA (data not shown). However, CCG-6920 was inactive against murine PAI-1 (data not shown) and was therefore not evaluated further because the in vivo characterizations and efficacy studies are conducted in mice. Curiously, commercially available CCG-7844 was markedly less active compared with the DMSO stock used in the HTS. This is a common issue when transitioning from screening centers to purchasing compounds, as stocks in screening centers are often stored for long durations in DMSO, and, as a result, unstable compounds may break down. Inspection of the CCG-7844 structure revealed a possible breakdown product that we designated CCG-7844 BP (Fig. S1C). This compound was purchased and tested for direct inactivation of

Dual-reporter high-throughput screen for inhibitors of PAI-1
PAI-1 in 15 mg/ml BSA (Fig. 6A). Using the dual-reporter system, we found that CCG-7844 BP inhibited PAI-1 with equivalent IC 50 values for each reporter (58.7-62.5 M; Fig. 5A). Furthermore, MS analysis identified a chemical species in the commercial preparation of CCG-7844 with an m/z identical to that of commercially available CCG-7844 BP , supporting the hypothesis that the breakdown product of the chemical library stock was indeed the active inhibitor picked up in the screen.
To examine the specificity of CCG-7844 BP for PAI-1 relative to other closely related serpins, single AMC reporter assays were used. These assays were constructed identically to the PAI-1 assay, but with different serpins, enzymes, and their specific substrates. These experiments compared the inhibitory activity of CCG-7844 BP for PAI-1 with its ability to inhibit antithrombin, ␣ 1 -antitrypsin, and ␣ 2 -antiplasmin. These three serpins have a high degree of structural similarity to PAI-1 and are present in plasma at much higher concentrations than PAI-1 and therefore could present potential off-target interactions for CCG-7844 BP . These data demonstrate that whereas CCG-7844 BP is an efficient inhibitor of PAI-1, it shows no inhibitory activity against antithrombin, ␣ 1 -antitrypsin, or ␣ 2 -antiplasmin at concentrations up to 1 mM, indicating that CCG-7844 BP specifically targets PAI-1 (Fig. 6B).
To determine whether this screen also could enrich for leads with a high likelihood of efficiently translating into in vivo settings, we examined whether CCG-7844 BP was also able to inhibit PAI-1 activity in human plasma (Fig. 6C). This is a critical test for determining the potential in vivo efficacy of any PAI-1 inhibitor, because in plasma, nonspecific protein binding of promiscuous molecules is a potential problem. Also, PAI-1 in plasma is predominantly found in association with vitronectin, and many previously characterized PAI-1 inactivators, including tiplaxtinin, display potent and specific inactivation of PAI-1 in buffer, but fail to inhibit PAI-1 bound to vitronectin or in plasma (14,22,33). Therefore, their potency in vivo is compromised. For these studies, glycosylated human PAI-1 was added to either buffer containing BSA, buffer containing BSA supplemented with 10 nM vitronectin, or PAI-1-depleted human plasma followed by the addition of varying concentrations of CCG-7844 BP (1 M to 1 mM), and the remaining PAI-1 activity . E, to test the efficacy of CCG-7844 BP in vivo, blood was withdrawn from the femoral artery of transgenic mice overexpressing murine PAI-1. The mice were then treated with intra-arterial injections of either 0.75, 1.5, or 3.0 mg/kg CCG-7844 BP via the femoral artery. Blood samples were then drawn 30 min following the injection, and the amount of active PAI-1 was determined and compared with pre-injection baseline levels. F, a time course of the effect of the 1.5 mg/kg of CCG-7844 BP was performed as in E, but with blood drawn at 0, 5, 15, 30, and 60 min, and levels of active PAI-1 murine plasma were determined. In A-D, each point represents duplicate samples analyzed in three independent experiments. Error bars representing ϮS.D. are shown where the errors are larger than the symbol. Data in E and F were analyzed for significance via one-way ANOVA and individually tested against vehicle via Fisher's least significant difference test (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). Data are shown as mean Ϯ S.D. n ϭ 3-6 as indicated.

Dual-reporter high-throughput screen for inhibitors of PAI-1
was determined. We found that the inhibitory activity of CCG-7844 BP against PAI-1 was unaffected either in ex vivo plasma or by the addition of vitronectin (Fig. 6C, open symbols) (IC 50 ϭ 44 Ϯ 2.5 M in plasma compared with 50 Ϯ 3.4 or 55 Ϯ 7.5 M in BSA-containing buffer or BSA-containing buffer with vitronectin, respectively). In contrast, as reported previously (14), the activity of tiplaxtinin is markedly attenuated in the presence of vitronectin such that no IC 50 can be determined (Fig. 6C, compare IC 50 ϭ 1.2 Ϯ 0.3 mM in HBS-BSA (filled circles) with HBS-BSA containing vitronectin (filled triangles)). These data demonstrate that CCG-7844 BP retains its potency in plasma and is not adversely affected by PAI-1 binding to vitronectin or by the high-protein environment of plasma. In addition to this spectroscopic approach, PAI-1 inactivation by CCG-7844 BP in human plasma was independently confirmed by an additional method based on our immunoassay described previously (22) with an IC 50 of 46 Ϯ 4.4 M. To the best of our knowledge, CCG-7844 BP is the first small-molecule PAI-1 inactivator that efficiently inhibits vitronectin-bound PAI-1 and with uncompromised potency in plasma compared with a buffer system.
Based on the ex vivo data in human plasma that suggests that CCG-7844 BP is likely to retain activity in vivo, we sought to evaluate the activity of CCG-7844 BP in transgenic mice overexpressing murine PAI-1 as described previously (22). For this analysis, we first determined that CCG-7844 BP inhibited murine PAI-1 as efficiently as human PAI-1, with IC 50 values of 43 Ϯ 7.5 M in plasma, 58 Ϯ 9.5 M in HBS-BSA, and 50 Ϯ 6.0 M in HBS-BSA containing 10 nM vitronectin (Fig. 6D). Next, transgenic mice overexpressing murine PAI-1 were anesthetized, the femoral artery was cannulated, and blood was drawn at time 0, after which CCG-7844 BP was injected via the same line at 0.75, 1.5, or 3.0 mg/kg. Blood was drawn again 30 min later, and the amount of active PAI-1 was determined and compared with baseline. A dose-dependent decrease in active PAI-1 levels was observed, with an ϳ50% reduction in active PAI-1 with both 1.5 and 3.0 mg/kg of CCG-7844 BP (Fig. 6E). To reach this same effective level of inactivation, our previously characterized polyphenol PAI-1 inactivator, CDE-066, required a 10-fold higher dose (30 mg/kg) (22). These data demonstrate the utility of our HTS design for accelerating the identification of PAI-1 inhibitors with a high likelihood of retaining activity in vivo.

Medicinal chemistry and structure-activity relationship (SAR) studies to improve pharmacokinetic properties
The data above establish the efficiency of our HTS; however, the apparent very low protein binding selected for by our HTS could present the potential problem that a small molecule, such as CCG-7844 BP , with very low protein binding could be rapidly cleared. To test this, a time course of the effect of the 1.5 mg/kg CCG-7844 BP dose was performed and showed that the compound's effect was relatively short-lived in vivo and that by 60 min post-administration, PAI-1 activity was not significantly different from baseline (Fig. 6F). Therefore, a systematic SAR study, which will be described in a separate report, was undertaken to probe the structural requirements of CCG-7844 BP for PAI-1 inactivation and to develop compounds with similar potency toward PAI-1 but with prolonged activity in vivo. One compound identified in this study, MDI-2268 (Fig. S1D), was found to have activity similar to that of CCG-7844 BP in vitro (Fig. 7A) but in vivo showed sustained PAI-1-inhibitory activity at 90 min following intraperitoneal (IP) injection (Fig. 7B). Based on these results, a pharmacokinetic study of the half-life and bioavailability of MDI-2268 was performed in rats, and these data indicate that MDI-2268 has excellent pharmacokinetics. Following intravenous administration, the half-life of MDI-2268 is 30 min, and after oral administration, the half-life is 3.4 h (Fig. 7C). The bioavailability of MDI-2268 was determined from the area under the curve of these data, and MDI-2268 was found to be 57% bioavailable, suggesting that it could be dosed orally. To test this possibility, PAI-1-overexpressing mice were dosed by oral gavage with increasing amounts of MDI-2268 from 0.3 to 10 mg/kg. The percentage of active PAI-1 in plasma was determined 90 min later and compared with vehicle-treated control mice. These data demonstrate that MDI-2268 is orally active against PAI-1 (Fig. 7D).
Together, these results suggest that MDI-2268 has the potential to be an effective inhibitor of PAI-1 during pathologic events where PAI-1 activity has been shown to play a role. Therefore, to test whether MDI-2268 is an effective inhibitor of PAI-1 in vivo under pathologic conditions, we examined the efficacy of MDI-2268 for the treatment of deep vein thrombosis (VT) in a murine model of VT (Fig. 8). Current treatments for VT involve the use of anticoagulation, which has a significant risk of bleeding. In contrast, stimulation of the fibrinolytic system could be a safer option for thrombosis treatment because the coagulation system will be unaffected, and PAI-1 is considered a potential therapeutic target in VT (24, 34 -36). For these studies, VT was induced via the electrolytic inferior vena cava model (EIM) as described (37). After induction, mice received either 3 mg/kg MDI-2268, or low-molecular weight heparin (LMWH), or vehicle alone three times/day by IP injection. Thrombi were harvested 2 days following VT induction, and thrombus weight was recorded. Bleeding risk was also assessed in a separate cohort of mice 90 min after IP injection of 3 mg/kg of either MDI-2268 or LMWH. These data indicated that MDI-2268 treatment was as efficacious as LMWH and was associated with a 62% decrease in thrombus weight compared with controls (Fig. 7A). Importantly, unlike LMWH, MDI-2268 was also safe and did not change the bleeding time. In contrast, LMWH demonstrated significantly prolonged bleeding times compared with the other groups (Fig. 7B). Thus, MDI-2268 is both as safe and as effective as LMWH at treating VT with the advantage that it has a lower risk of bleeding. MDI-2268 may represent a new and attractive avenue for the treatment of thrombotic disorders such as VT and other disorders in which excess PAI-1 is implicated.

Discussion
This study was initiated to address two problems noted in previous HTSs for small-molecule inhibitors of PAI-1: high apparent hit rates of compounds with spectroscopic properties similar to the screen reporter and hits that lose efficacy when translated into in vivo settings either due to their poor activity against vitronectin-bound PAI-1 or due to their high degree of promiscuous binding. Small compounds and especially NPEs

Dual-reporter high-throughput screen for inhibitors of PAI-1
are notorious for having intrinsic properties that interfere with HTS reporters, either high colorimetric and/or fluorimetric background interference or significant quenching of reporter signals (32,38). To avoid blue-shifted signals commonly inherent in small compounds, red-shifted reporters have been recommended for HTS (38). This was a strategy used by Grant et al. (39) in their dual-reporter system, which combined a rhodamine-conjugated substrate and an AMC-conjugated substrate. To increase the stringency of our HTS against nonselective serum protein binding, we also included BSA in all reaction buffers. Unfortunately, BSA was found to interfere with the fluorescence signal of all red-shifted dyes that we evaluated, including rhodamine-110, BOPIDY, and resorufin.
Because the use of a red-shifted dye was excluded by the addition of BSA to our HTS, we devised a novel system using two reporters with opposite signal changes. With the popular AMC reporter, inactivation of PAI-1 leads to an enhanced fluorescence signal. In contrast, with the WPF reporter, inactivation of PAI-1 leads to a decreased fluorescence signal. For a small compound to pass the primary HTS as a false positive in this dual-reporter system, it must act to both quench a fluores-cence signal and act as a fluorophore. The low likelihood of this event significantly increases the stringency of our HTS.
Two additional sources of stringency were also utilized. First, we maintained a 2-fold molar excess of PAI-1 over uPA in the screen as in an earlier study (22). This required that more than half of the PAI-1 in each reaction had to be inactivated before a positive uPA-based signal could be detected, decreasing the sensitivity of our HTS and increasing its selectivity toward compounds with more potent anti-PAI-1 activity. This also increased the signal/noise ratio by ensuring that uPA was completely inactivated by PAI-1 in the absence of compound. Second, the inclusion of 15 mg/ml BSA selected against compounds that exhibit high nonspecific protein binding. This resulted in the absence of any polyphenolic compounds from the primary HTS-positive compound list even though polyphenols represent some of the highest-affinity PAI-1-inactivating compounds discovered to date (22,40). This is likely because polyphenols are known to have affinities for plasma albumin in the 5-100 M range (41,42). Whereas the polyphenols inactivate PAI-1 with apparent affinities of 3-100 nM (22), the abundance of albumin in serum at 33-55 mg/ml or ϳ0.7 M over-

Dual-reporter high-throughput screen for inhibitors of PAI-1
comes the ability of polyphenols to target plasma PAI-1, which normally circulates at 0.1-0.4 nM (43). Thus, the addition of BSA to the HTS dramatically improved the stringency of the screen for specific compounds that were not prone to promiscuous binding, like the polyphenols. After triage, only two compounds (CCG-6920 and CCG-7844) were shown to dose-dependently inactivate PAI-1 in both halves of the dual-reporter system in the presence of 15 mg/ml BSA. Whereas the selection against promiscuously binding molecules eliminated compounds that were poor pharmaceutical candidates, this in turn created a different problem, in that our final lead molecule, CCG-7844 BP , likely had such low nonspecific protein binding that its activity in vivo was very short-lived. However, engineering in nonspecific protein binding turned out not to be a significant problem, and our SAR studies yielded MDI-2268, which retains the in vivo efficacy of CCG-7844 BP but also has excellent pharmacokinetic properties.
The success of an HTS also depends on the quality and integrity of the compound library used. The water content of storage mixtures, the temperature, the storage duration, the light expo-sure, the gas under which compounds are placed, and the number of freeze-thaw cycles to which the compounds are subjected are factors that can affect the rate of compound decay. Decay is typically measured through biophysical means, such as NMR and MS (44,45). Less well-known is the rate of compound decay into active breakdown products, a process that is difficult to study because activity can only be defined in the context of a screen target. However, the identification of CCG-7844 BP is a reminder that compound decay can also generate new activities.
Finally, the studies presented here demonstrate some of the unique problems associated with screening for small-molecule inhibitors that retain activity in vivo against metastable proteins like serpins. Considerations for such an HTS are different from those for typical screens targeting an enzyme active site or receptor-ligand interaction. By reducing the rate of spectroscopic interference, as well as the rate of hits by promiscuous protein-binding molecules, we have been able to identify a lead candidate molecule, MDI-2268, with significant clinical potential. The metastable nature of serpins and the major conformational transformation required for their mechanism of action make them uniquely susceptible to the binding of relatively nonspecific ligands (26). By removing promiscuous hits from our HTS, we have been able to identify a class of molecules with high potential for translation into in vivo settings. MDI-2268 shows both excellent pharmacokinetic properties and efficacy in vivo against pathologic thrombosis, demonstrating the potential of a dual-reporter high protein screen for the rapid identification of pharmaceutical candidates targeting PAI-1 and potentially serpins more generally.

Materials and reagents
See the supporting information.

Fluorescent protease reporters in non-and high-protein buffers
Plasmin or uPA was combined with fluorogenic substrates at the final concentrations listed in Table S1 in black, nonbinding 96-well microplates in triplicate. Turnover of the substrate was monitored in 10 mM HEPES, 150 mM NaCl, pH 7.4, containing 0.005% Tween 20 (HBS) in the absence and presence of 10 -15 mg/ml (1-1.5% w/v) BSA. After incubation at room temperature for 15-30 min, the fluorescence was read on a SpectraMax M5 microplate reader at the excitation and emission wavelengths listed in Table S1 using automatic cutoffs determined by the SpectraMax software. The -fold change in fluorescence in the absence and presence of BSA was determined relative to the fluorescent reporter alone.

Dual-reporter high-throughput screen for inhibitors of PAI-1
(50 M final) or 40 M ZGGR-AMC (10 M final). The final volume after the addition of all reagents was 120 l. Following a 30-min incubation, the WPF emission spectrum was read using an excitation wavelength of 430 nm (emission cutoff ϭ 450 nm). The AMC fluorescence emission spectrum was read using an excitation wavelength of 385 nm (emission cutoff ϭ 420 nm). Fixed wavelength WPF reads were performed using excitation and emission wavelengths of 430 and 470 nm, respectively (emission cutoff ϭ 455 nm), whereas fixed wavelength AMC fluorescence reads were performed using excitation and emission wavelengths of 385 and 460 nm, respectively (emission cutoff ϭ 455 nm). The -fold change in WPF signal (⌬F WPF ) was determined by Equation 1, where F WPF,0 is the fluorescence intensity in the absence of uPA and F WPF,U is the fluorescence intensity in the presence of the enzyme, uPA. The percentage of WPF quenched was determined by the expression, (1 Ϫ ⌬F WPF ) ϫ 100%. The -fold change in AMC fluorescence signal (⌬F AMC ) was determined by Equation 2, where F AMC,0 is the fluorescence intensity in the absence of uPA and F AMC,U is the fluorescence intensity in the presence of uPA.
Spectroscopic interference between pNA and AMC was tested as follows. Aliquots of pEGR-pNA or ZGGR-AMC were first incubated with 5 nM uPA until completely cleaved (90 min, room temperature). Each reporter was then diluted 2-fold with either HBS or the other reporter for final concentrations of 50 M cleaved pEGR-pNA and 10 M cleaved ZGGR-AMC. WPF and AMC fluorescence signals were then recorded using the above fixed wavelengths.

Conversion to 384-well microplate format and assay optimization
See the supporting information.

Evaluation of tiplaxtinin (PAI-039) and curcumin by a dualreporter system
PAI-1 at 10 nM was incubated with 3.2-56 M tiplaxtinin in HBS containing 0.1% DMSO for 15 min in 384-well low-volume white microplates followed by incubation with 5 nM uPA for 15 min. A mixture of 200 M pEGR-pNA and 100 M ZGGR-AMC was added, and the reaction was incubated for an additional 30 min, after which WPF and AMC fluorescence were read at the described wavelengths. The final assay volume was 12 l. The data were expressed as normalized fluorescence intensity (F normalized ), as determined by Equation 3, where F is the raw fluorescence intensity, whereas F max and F min are the maximum and minimum fluorescence intensities, respectively. For AMC fluorescence, F max and F min were provided by control reactions containing uPA and a mixture of uPA and PAI-1, respectively, whereas for WPF, the controls were reversed. The data were fit by a nonlinear regression analysis (log [inhibitor] versus response) with variable slope using GraphPad Prism version 7. A single concentration of 30 M curcumin was reacted with 10 nM PAI-1, 5 nM uPA, and a mixture of pEGR-pNA and ZGGR-AMC as above, and the raw WPF and AMC fluorescence intensities were analyzed.

Primary high-throughput screening
High-throughput screening was carried out using the protocol outlined in Fig. 4  For primary screening, all reagents except for compounds were added using a Thermo Scientific Multidrop Combi. Each sample well contained 6 l of 15 nM PAI-1. One compound (0.2 l, ϳ2 mM) was stamped per well using a Beckman Biomek FX with a pin-tool attachment liquid-handling system, resulting in a final compound concentration of ϳ33 M (3.3% DMSO). Concentrations of the NPEs are unknown, as they are added as a solventextracted mixture. Following the addition of uPA and then substrate mixture as described (Fig. 4), fluorescence was recorded using a BMG Labtech Pherastar plate reader. WPF was read using excitation and emission wavelengths of 430 and 470 nm, respectively, and AMC fluorescence was read using 380 and 470 nm, respectively. All data were analyzed using Tripos Benchware Dataminer. Compounds that displayed a change in WPF and AMC fluorescence signal that was Ͼ3 S.D. from the negative control were chosen for confirmation.

Counterscreen, confirmation, and dose-response testing
Confirmation testing for both purified compounds and NPEs was performed as described for the primary assay except that compounds were stamped in triplicate. In addition, a counterscreen read of the confirmation plates was performed after the addition of PAI-1, compound, and uPA, but prior to the addition of the substrate mixture. Only compounds that showed a difference of Ͻ3 S.D. in WPF and AMC signals relative to the negative control prior to the addition of substrates, but a Ͼ3 S.D. change in WPF and AMC signals in at least two wells after the addition of the substrates, were selected for dose-response testing. For dose-response testing, 0.03-0.6 l of compound were stamped in duplicate using a TTP Labtech Mosquito X1 liquid-handling system, resulting in approximate final compound concentrations of 12-250 M. Development was carried Dual-reporter high-throughput screen for inhibitors of PAI-1 out as described for the primary screen. NPEs were not included in dose-response testing because the concentrations of active compounds in these mixtures are unknown.

Inactivation of PAI-1 in plasma
Dose-dependent inactivation of glycosylated human PAI-1 in human plasma by CCG-7844 BP was performed as described previously (22). Inactivation of murine PAI-1 by CCG-7844 BP in murine plasma was similarly tested. The control assay buffer was modified as follows: 40 mM HEPES, pH 7.4, 100 mM NaCl, 0.005% Tween 20, 10% DMSO, and 15 mg/ml BSA with or without 10 nM human vitronectin.

Inactivation of PAI-1 in vivo
All animal procedures were approved by the University of Michigan Institutional Animal Care and Use Committee and performed in accordance with all applicable national and/or institutional guidelines for the care and use of animals. Transgenic C57Bl6J mice heterozygous for murine PAI-1 overexpression (46) were weighed and then anesthetized with 2% isoflurane throughout the experiment. The femoral artery was exposed and cannulated using a polyethylene catheter. Blood was drawn from the femoral artery at time 0 before CCG-7844 BP administration. Then CCG-7844 BP at 0.75, 1.5, or 3.0 mg/kg in Ringer buffer with 0.1% DMSO was injected intraarterially via the femoral catheter, and blood was drawn again 30 min later. A time course of the effect of the 1.5 mg/kg CCG-7844 BP was performed with the same method but with blood drawn at 0, 5, 15, 30, and 60 min. Levels of active PAI-1 murine plasma were determined as described previously (22). For analysis of MDI-2268, the compound was dissolved in 0.1% DMSO in lactated Ringer buffer for IP injections. For gavage administration, the compound was suspended in 0.5% methyl cellulose in water. In both cases, 90 min after MDI-2268 administration, the mice were anesthetized with 2% isoflurane and sacrificed by vena cava puncture and terminal bleed. The percentage of active PAI-1 in the plasma was then determined as described (22).

Synthesis and pharmacokinetic analysis of MDI-2268
The synthesis of MDI-2268 is described in the supporting Experimental procedures. The pharmacokinetic analysis of MDI-2268 was performed by the University of Michigan pharmacokinetic and mass spectrometry core. Briefly, MDI-2268 at 3 mg/ml in 15% DMSO, 15% PEG 400, and 70% PBS was injected via the tail vein (intravenously) at 15 mg/kg or given orally at 30 mg/kg to three rats in each group. At the time points shown, blood samples were collected and centrifuged immediately at 15,000 ϫ g for 10 min. The plasma was collected and frozen at Ϫ80°C for later analysis. The analytical curves were constructed with 11 non-zero standards by plotting the peak area ratio of MDI-2268 to the internal standard versus the concentration. The concentration range evaluated was from 2.5 to 5,000 ng/ml. A blank sample (matrix sample processed without internal standard) was used to exclude contamination or interference.

Murine model of venous thrombosis
The EIM is a murine model of venous thrombosis that has been described previously (37). Briefly, thrombosis is induced under flow conditions by a small current run through a copper wire applied to the anterior vein wall. This releases free radicals within the vessel, which in turn activate the endothelial cells, initiating coagulation. The EIM produces a thrombus that is highly consistent in size and develops in the presence of blood flow (37). After induction of the thrombus, mice received 3 mg/kg MDI-2268, 3 mg/kg LMWH, or vehicle by IP injection 3 times/day. Thrombi were harvested 2 days after induction, and their weights were recorded. To evaluate the safety of MDI-2268, bleeding times were evaluated in a separate cohort of mice 90 min after IP injection of either LMWH or MDI-2268 at 3 mg/kg. Ninety minutes later, mice were anesthetized with chloral hydrate (450 mg/kg; Fisher), and 5 mm of the tail tip was cut with a razor blade and immediately placed in 12 ml of PBS at 37°C and timed until blood flow stopped.