Inhibition of Angiogenesis in Vivo by Plasminogen Activator Inhibitor-1*

The process of angiogenesis is important in both normal and pathologic physiology. However, the mechanisms whereby factors such as basic fibroblast growth factor promote the formation of new blood vessels are not known. In the present study, we demonstrate that exogenously added plasminogen activator inhibitor-1 (PAI-1) at therapeutic concentrations is a potent inhibitor of basic fibroblast growth factor-induced angiogenesis in the chicken chorioallantoic membrane. By using specific PAI-1 mutants with either their vitronectin binding or proteinase inhibitor activities ablated, we show that the inhibition of angiogenesis appears to occur via two distinct but apparently overlapping pathways. The first is dependent on PAI-1 inhibition of proteinase activity, most likely chicken plasmin, while the second is independent of PAI-1's anti-proteinase activity and instead appears to act through PAI-1 binding to vitronectin. Together, these data suggest that PAI-1 may be an important factor regulating angiogenesis in vivo.

ogen activator both activate plasminogen to the broad spectrum enzyme plasmin, which is thought to directly degrade basement membrane proteins, and to activate latent metalloproteinases (4,5). Plasmin activity has been implicated in tissue remodeling (6), and migrating cells both in vitro and in vivo express elevated levels of uPA, together with its receptor (7)(8)(9). Thus, PAs at the cell surface are thought to promote cell migration through breakdown of the extracellular matrix. However, recent studies suggest that the role of uPA as well as other proteinases may not be to cause generalized matrix destruction but instead may be to expose cryptic cell attachment sites necessary for cell migration (10 -12).
PAI-1 interacts with a number of nonproteinase ligands, including the cell adhesion protein vitronectin (13,14). Vitronectin is a multifunctional glycoprotein present in plasma, platelets, and the extravascular matrix. Like PAI-1, vitronectin is deposited at sites of disease or injury (15,16), where it binds multiple ligands, including collagens, fibrin, uPA receptor, integrins, and PAI-1 (17)(18)(19)(20)(21). Plasma vitronectin is in a closed conformation that does not interact with integrins. However, upon binding to a surface at a site of injury, vitronectin undergoes a conformational change, which exposes the integrin binding site (22). Similarly, active PAI-1 converts the closed plasma form of vitronectin to the open integrin binding conformation (22). Paradoxically, however, PAI-1 binding to vitronectin blocks the adjacent RGD integrin binding site on vitronectin (10,11,14). Thus, the association of PAI-1 with vitronectin simultaneously induces the cell binding conformation but blocks the cell binding site. However, this block is likely to be transient, since reaction of PAI-1 with a proteinase induces a rapid conformational change in PAI-1 that causes a large reduction in its affinity for vitronectin. Likewise, complex formation between PAI-1 and a proteinase induces an equally dramatic increase in PAI-1 affinity for some members of the low density lipoprotein receptor family, such as the low density lipoprotein receptor-related protein, gp330/megalin, and the very low density lipoprotein receptor (23)(24)(25). This leads to rapid repartitioning of PAI-1 from vitronectin in the matrix to the endocytic receptors and subsequent cellular degradation of the PAI-1-proteinase complex (23). Thus, PAI-1 and vitronectin can be thought of as forming a molecular switch (10,19,23,26). When PAI-1 is bound to vitronectin, the integrin adhesion site is cryptic, and cell migration is blocked. However, association with a proteinase induces the conformational change associated with the switch in PAI-1 affinity from vitronectin to low density lipoprotein receptor-related protein, resulting in clearance of the PAI-1 and allowing cellular integrins access to the RGD adhesion site. Thus, PAI-1 bound to vitronectin within the extracellular matrix may serve to regulate cell migration, both by regulating specific proteolysis and by controlling access of cellular integrins to specific adhesion sites within the matrix. Accordingly, the present study examines the potential of exogenously added PAI-1 to regulate vascular cell migration in vivo by testing the hypothesis that PAI-1 can inhibit angiogenesis in vivo in both a vitronectin-dependent and a proteinasedependent manner.

EXPERIMENTAL PROCEDURES
Proteins and Reagents-To maintain PAI-1 functional activity during the extended incubations required for these in vivo assays, all recombinant PAI-1 proteins, unless otherwise noted, were constructed on the constitutively active PAI-1 background (14-1b) (27). This form of PAI-1 has been shown to be essentially identical to wild-type PAI-1 with respect to inhibitory activity, as well as binding to vitronectin, heparin, and low density lipoprotein receptor-related protein but has a half-life nearly 75-fold longer than wild-type PAI-1 at 37°C (145 versus 2 h) (23,27). The 14-1b PAI-1 and latent wild type PAI-1 were from Molecular Innovations. All other forms were constructed using the Transformer site-directed mutagenesis kit (CLONTECH) according to the manufacturer's instructions and isolated as described (28), except that preparations having significant detectable endotoxin levels following phenyl-Sepharose chromatography were further purified by polymyxin B chromatography (Detoxi-Gel; Pierce). Final endotoxin levels for all proteins used were below the FDA limit for parenteral drugs of 5 endotoxin units/kg (29). The different forms of PAI-1 used were as follows: 14-1b PAI-1 (PAI-1); latent PAI-1 (PAI-1 L ); PAI-1 with mutation of Gln 123 to Lys (PAI-1 K ), which has been previously shown to have a specific defect in vitronectin binding but retains full inhibitory activity (30); PAI-1 with mutation of Arg 346 to Ala (PAI-1 A ), which binds vitronectin with wild-type affinity but does not inhibit PAs (10); and finally, a double mutant, Thr 333 to Arg, Ala 335 to Arg (PAI-1 R ), which binds vitronectin normally but has no inhibitory activity toward any proteinase. Cleaved forms of PAI-1 were made by incubating PAI-1 with pancreatic elastase at a molar ratio of 100:1 (PAI-1/elastase) for 30 min followed by inactivation of the elastase with phenylmethylsulfonyl fluoride (1 mM) (31). This results in cleavage of the reactive center loop at a single site and induces the conformational change in PAI-1 associated with reactive center loop insertion and loss of vitronectin affinity (13,31). Chicken vitronectin was purified from chicken plasma as described (32), and antiserum was raised in rabbits by standard techniques.
Binding Assays-Competitive inhibition binding assays were performed as described (23) using a form of 14-1b PAI-1 containing a hexapeptide tag at the amino terminus (PAI-1 HMK ). This form of PAI-1 can be labeled with 32 P without loss of inhibitory activity or binding affinity for vitronectin and is similar to a form of PAI-1 previously described (23) that was constructed on the wild-type background.
In Vivo Angiogenesis Assay-The chicken CAM angiogenesis assay was performed essentially as described (33). Briefly, filter disks saturated with FGF-2 were applied to the chicken CAM. Next, 25 l of PAI-1 variants or antibodies diluted in phosphate-buffered saline to the concentrations indicated were then applied at times 0, 24, and 48 h, and the CAM tissues were harvested at 72 h. Angiogenesis was evaluated as described (33) by a person unaware of the codes for the different experimental groups. The data are calculated as angiogenic index, which is determined by subtracting the mean branching observed in the absence of FGF-2 treatment from the branch points in samples treated with FGF-2 and presented as the percentage of the FGF-2 only-treated control.
Proteinase Assays-SDS-PAGE zymography was used for analysis of proteinase levels in CAM tissue. Briefly, tissues were harvested after 72 h of treatment, rapidly frozen on dry ice, and stored at Ϫ70°C. Immediately prior to analysis, frozen tissues were weighed and then extracted in nonreducing SDS-PAGE sample buffer, after which equivalent amounts of tissue extract were subjected to SDS-PAGE on a 10% gel (Novex). Chicken kidney extracts containing uPA were prepared as described (34) and were included as positive controls. Following electrophoresis, gels were soaked in 2.5% Triton X-100 twice for 45 min each to remove the SDS. The gel was then applied to a 1% agarose gel containing 1% casein with or without 30 g/ml of chicken plasminogen (Molecular Innovations) and incubated in a humid chamber at 37°C. Samples were also analyzed on casein indicator gels containing 100 g/ml of anti-human uPA (Chemicon).
The second order rate constant for the inhibition of chicken plasmin (Molecular Innovations) by the constitutively active human PAI-1 was determined as described (34). The relative inhibitory activity of the PAI-1 K mutant versus chicken plasmin was compared with active PAI-1 by titration as described (35), except that the substrate Val-Leu-Lys-p-nitroanilide (Sigma) was used. The relative inhibition of chicken uPA by the various forms of PAI-1 was examined by incubation of PAI-1 with kidney extracts followed by SDS-PAGE zymography, and none of the forms of PAI-1 tested effectively blocked chicken uPA (data not shown). Immunoblot analysis of purified chicken plasminogen, plasmin, and CAM extracts was performed essentially as described (34). PAI-1 and plasmin complexes were formed by reacting equimolar amounts of PAI-1 and plasmin for 30 min at room temperature prior to the addition of SDS-PAGE sample buffer.
To determine whether PAI-1 was inhibiting plasmin within the CAM tissue an ELISA specific for PAI-1-plasmin complexes was developed. Briefly, a monoclonal antibody, MA33B8 (36) (Molecular Innovations), was coated onto microtiter plates at 2 g/ml in TBS for 1 h at 37°C. This antibody has previously been shown to bind reactive center loop inserted forms of PAI-1, such as PAI-1 in complex with a proteinase, with high affinity (37). Following coating, the plates were blocked for 1 h using 2% bovine serum albumin, 5% nonfat dried milk in TBS (blocking buffer), after which samples of CAM tissue extracts in blocking buffer were added to the plate and incubated at 37°C for 1 h followed by washing with TBS, 0.01% Tween 20. The plate was then incubated with a 1:10,000 dilution of rabbit anti-chicken plasminogen antiserum in blocking buffer for 1 h at 37°C, followed by washing with TBS, 0.01% Tween 20. Next, goat anti-rabbit IgG conjugated to alkaline phosphatase was added at a 1:5000 dilution in blocking buffer and incubated for 1 h at 37°C, followed by washing. The plasmin-PAI-1 complexes were then detected using the substrate p-nitrophenyl phosphate, disodium (Sigma) at a concentration of 4 mg/ml in TBS. The values obtained for the plasmin-PAI-1 complexes present in the CAM tissues were compared with a standard curve of purified chicken plasmin-PAI-1 complexes. To ensure the specificity of the ELISA, preliminary experiments were also performed. These demonstrated that the constitutively active form of PAI-1 (14-1b) used in our studies did not bind to MA33B8 coated onto microtiter plates prior to complex formation with chicken plasmin when detected with rabbit anti-human PAI-1 antiserum (data not shown). Likewise, the ELISA was shown to be specific for PAI-1-plasmin complexes when the rabbit anti-chicken plasminogen antiserum was used, since purified PAI-1⅐uPA complexes were not detected in this assay (data not shown).
Immunofluorescence-For immunofluorescence analysis of frozen samples, CAM tissues were embedded in OCT and snap-frozen in a 2-methylbutane dry ice slurry, and stored at Ϫ80°C until processing. Sections were cut at 5 m and fixed for 10 min in methanol/acetone 1:1 and then rinsed in phosphate-buffered saline. Samples were then stained with rabbit anti-chicken vitronectin antibodies and analyzed by immunofluorescence using a rhodamine-labeled goat anti-rabbit IgG (Pierce) for detection.

PAI-1 Blocks Angiogenesis in the Chick CAM-Previously,
we demonstrated that vitronectin enhanced the migration of smooth muscle cells in vitro and that PAI-1 was an efficient inhibitor of this enhancement (10). This inhibition was independent of PAI-1's anti-proteinase activity and instead was dependent on PAI-1 binding to matrix vitronectin, where it prevented the association of promigratory integrins with vitronectin (10,11). Similarly, vitronectin can also stimulate the migration of endothelial cells in vitro, and as with smooth muscle cells this stimulation is blocked by PAI-1. 2 Together, these data suggest that vitronectin stimulates vascular cell migration and that PAI-1 blocks this enhancement by preventing the association of ␣ v ␤ 3 or other promigratory integrins with vitronectin in the matrix.
Since PAI-1 and vitronectin appear to act together in vitro to regulate vascular cell migration, we tested the ability of PAI-1 to inhibit angiogenesis in vivo. For these studies, we examined the effect of PAI-1 on FGF-2-induced angiogenesis in the chicken CAM assay. Fig. 1 shows the results obtained following treatment for 3 days with or without FGF-2 or FGF-2 plus PAI-1. A and B show CAM tissue incubated with either vehicle alone or with FGF-2, respectively. As expected, FGF-2 induced a strong angiogenic response in the CAM tissue, as is evident from the increased number of branching vessels. C and D show the effect of 1 M latent-PAI-1 (PAI-1 L ; C) or active-PAI-1 (D) on FGF-2-induced angiogenesis. These data indicate that only the active form of PAI-1 is an effective inhibitor of FGF-2mediated angiogenesis in the CAM assay. Quantitation of these results demonstrated that active PAI-1 completely blocks angiogenesis, whereas PAI-1 L has no significant effect ( Fig. 2A). Moreover, the inhibition of angiogenesis by active-PAI-1 is dose-dependent with an IC 50 value of 0.112 Ϯ 0.005 M (Fig.  2B). Remarkably, this value is nearly identical to the reported K d of 0.100 -0.125 M for PAI-1 binding to native vitronectin in solution (13,38), suggesting that similar to vascular cell migration in vitro, PAI-1 inhibition of CAM angiogenesis might be at least in part vitronectin dependent. However, since the active conformation of PAI-1 both binds vitronectin with high affinity and inhibits proteinases, whereas PAI-1 L has neither property (13), these results do not reveal whether the inhibition of angiogenesis is dependent on either proteinase inhibition, vitronectin binding, or both.
Proteinase Inhibition by PAI-1 Inhibits FGF-2-induced Angiogenesis in Vivo-To dissect the contribution of PAI-1 proteinase-inhibitory activity away from its vitronectin binding properties, on the inhibition of angiogenesis, a specific PAI-1 mutant was constructed that binds vitronectin poorly but retains wild-type inhibitory activity (30). This mutant, PAI-1 K , has a greater than 250-fold lower affinity for vitronectin than does wild-type PAI-1 and should inhibit proteinase activity in the CAM tissue without significantly blocking vascular cell adhesion to vitronectin (13). Fig. 3 demonstrates that this mutant, like fully functional PAI-1, significantly inhibits vessel formation in the CAM in a dose-dependent manner with an IC 50 of 0.204 Ϯ 0.017 M. This value is ϳ2-fold higher than that of fully functional PAI-1, which is 0.112 M. This suggests that PAI-1's anti-proteinase activity is critical for its inhibition of angiogenesis and that its binding to vitronectin is likely to be less important for this inhibition. However, it is also possible that these two activities overlap, such that high levels of antiproteinase activity may mask the importance of vitronectin binding. This would be the case, for example, if proteolytic remodeling of the matrix were required to expose cryptic vitronectin adhesion sites, which then were able to promote vascular cell migration.
Human PAI-1 Inhibits Plasmin Activity in the CAM-To directly examine proteinase activity during FGF-2-induced angiogenesis, CAM extracts were analyzed by SDS-PAGE zymography. These results demonstrate that there is a marked increase in uPA activity in CAM tissue following 72 h of treatment with FGF-2 (Fig. 4A, lane 4), and PAI-1 does not appear to significantly reduce this activity (Fig. 4A, lane 5). The lack of inhibition of chicken uPA by active human PAI-1 is consistent with previous findings (39) and suggests that human PAI-1's anti-proteinase activity in this system is probably directed toward another enzyme. Since the most obvious function of uPA is to activate plasminogen to plasmin, and since PAI-1 has been reported to be a very efficient inhibitor of plasmin (40), we examined the ability of human PAI-1 to directly inhibit chicken plasmin. These data indicated that human PAI-1 is a very efficient inhibitor of chicken plasmin, having a secondorder rate constant of ϳ7.6 ϫ 10 5 M Ϫ1 s Ϫ1 (data not shown). This is essentially identical to the rate of 6.6 ϫ 10 5 M Ϫ1 s Ϫ1 reported for bovine PAI-1 inhibition of human plasmin (40). Human PAI-1 also formed SDS-stable high molecular weight complexes with chicken plasmin as expected for inhibition of a serine proteinase by a serpin (Fig. 4B, lane 4) (41). Similar to uPA, the levels of plasminogen present in the CAM also increased upon FGF-2 treatment; however, unlike uPA, the addition of PAI-1 to the CAM reduced the amount of plasminogen present in 72-h extracts to levels similar to unstimulated CAM tissue (Fig. 4B, lanes 5-7). This reduction most likely is a result of PAI-1 inhibition of angiogenesis, leading to reduced blood volume in the tissue, as well as the expected rapid clearance and degradation of PAI-1-plasmin complexes (23). The observation that PAI-1 reduces plasminogen but not uPA in FGF-2treated tissue also suggests that the inhibition of angiogenesis is not due to PAI-1 blocking FGF-2 signaling directly, since uPA, which should be produced by cells locally in response to FGF-2 (42, 43), is not significantly reduced. In contrast, plasminogen, which is produced in the liver and secreted into the blood, would only be expected to increase in the CAM when new vessels are established.
To see if human PAI-1 actually does inhibit chicken plasmin in the CAM tissue, a sensitive, PAI-1-plasmin complex-specific ELISA was developed that utilizes a conformation-specific anti-PAI-1 monoclonal antibody that does not bind to active PAI-1 but does bind PAI-1 in complex with a proteinase (37). The captured PAI-1-plasmin complex is then detected with a specific anti-chicken plasminogen antibody. Extracts from CAM tissue after 72 h of treatment with FGF-2 or with FGF-2 and 1 M active PAI-1 were examined in this assay. The results of this analysis are shown in Table I and indicate that PAI-1 plasmin complexes are detected in all PAI-1-treated CAM samples tested but not in CAM tissues that were not treated with human PAI-1. The levels ranged from 6.2 to 145 pg of complex per mg of CAM tissue. Thus, these results demonstrate that exogenously added human PAI-1 can inhibit chicken plasmin in vivo. Taken together, these data suggest that PAI-1 can inhibit angiogenesis in the CAM in a proteinase-dependent manner but that the target enzyme is unlikely to be chicken uPA, since it has been shown to be resistant to inhibition by human PAI-1 (39). Instead, based on the rate constant of inhibition and the presence of PAI-1⅐plasmin complexes in vivo, human PAI-1 most likely acts in the CAM by inhibiting the activity of chicken plasmin, thereby blocking plasmin-mediated proangiogenic events, such as basement membrane and fibrin matrix remodeling, and activation of matrix metalloproteinases (5,44,45). Consistent with this interpretation, several studies have demonstrated that specific plasmin inhibitors can block angiogenesis in vitro. (46 -48).

PAI-1 Binding to Vitronectin Inhibits FGF-2-induced Angiogenesis in Vivo-
The results above suggest that the proteinaseinhibitory properties of PAI-1 are important for its inhibition of angiogenesis in vivo. However, since PAI-1 inhibition of vascular cell migration in vitro was shown to be dependent on vitronectin binding (10,11), and since the PAI-1 K mutant was not as effective an angiogenesis inhibitor as fully functional PAI-1, we also examined whether or not PAI-1 binding to vitronectin in the absence of proteinase inhibition plays a significant role in controlling FGF-2-induced angiogenesis. For this analysis, it was first necessary to determine whether human PAI-1 binds chicken vitronectin with high affinity. Since the amino acid residues critical for the binding of human PAI-1 to human vitronectin were recently identified (14), an alignment of the somatomedin B regions of all vitronectin sequences FIG. 5. Human PAI-1 binds chicken vitronectin with high affinity and inhibits angiogenesis. A, solid phase competitive inhibition binding assays with increasing concentrations of PAI-1 or PAI-1 mutants competing for the binding of 32 P-labeled PAI-1 to chicken vitronectin. E, active PAI-1; Ⅺ, PAI-1 A ; ‚, PAI-1 K . The curves for PAI-1 and PAI-1 A were generated with a four-parameter fit as described (65)

TABLE II Sequence comparison of the PAI-1 binding site in vitronectin from different species
Sequence alignment is shown of the segment of the somatomedin B domain of vitronectin using all available vitronectin sequences in GenBank™. Amino acid residues found to be important for PAI-1 binding to vitronectin are in boldface type and underlined (14), and the RGD integrin binding site is in boldface italic type. found in the GenBank TM data base to date was performed (Table II). This analysis revealed that all of the residues that have been identified as being essential for PAI-1 binding to human vitronectin are conserved across all species thus far sequenced. This suggests that the structure of this region of vitronectin has been well conserved and that human PAI-1 may bind chicken vitronectin with high affinity. This was tested directly in a competitive inhibition binding assay, which demonstrated that human PAI-1 bound to chicken vitronectin with high affinity (Fig. 5A). Furthermore, just as with human vitronectin (13), a PAI-1 mutant (PAI-1 A ) that does not inhibit PA activity but has previously been shown to bind human vitronectin, bound chicken vitronectin normally, whereas the PAI-1 K mutant did not (Fig. 5A). The IC 50 values calculated from these data were 17 Ϯ 2.8 nM for fully active PAI-1 and 21 Ϯ 2.5 nM for PAI-1 A . These values are essentially identical to results obtained in similar experiments with human vitronectin, which yielded an IC 50 of 17 nM (data not shown). Together, these results indicate that human PAI-1 binding to chicken vitronectin is indistinguishable from its binding to human vitronectin.
We next examined whether PAI-1 binding to vitronectin in the absence of PA inhibition plays a significant role in its inhibition of angiogenesis. In the first experiments, we examined the ability of the PAI-1 A mutant to block angiogenesis. This mutant has its P 1 residue changed from Arg to Ala, which converts it from an inhibitor of PAs to an inhibitor of elastase 3 but does not affect its binding to vitronectin (Fig. 5A). These data are shown in Fig. 5B and indicate that the PAI-1 A mutant is able to block angiogenesis in the CAM assay but that it is ϳ50-fold less effective than the PAI-1 K mutant demonstrating ϳ50% inhibition of angiogenesis at 10 M. These results suggest that PAI-1 binding to vitronectin is less important for its inhibition of angiogenesis than is PAI-1 inhibition of proteinase activity. However, since the association of PAI-1 A with a proteinase, either as an inhibitor or a substrate, would result in its cleavage and therefore loss of vitronectin affinity (13,49), and since proteinase activity in general is thought to be increased in areas of active angiogenesis, then it is also possible that PAI-1 A is being cleaved and inactivated in the CAM. Such proteolytic inactivation of PAI-1 by both neutrophil proteinases and matrix metalloproteinases has been previously reported (50,51). Therefore, to eliminate this possibility, a new PAI-1 mutant was designed to have no inhibitory activity toward any proteinase and to retain vitronectin binding even after cleavage by a proteinase. This mutant was based on previous studies demonstrating that proteinase inhibition and loss of vitronectin binding requires a rapid conformational change in PAI-1 that involves insertion of its reactive center loop into the interior of PAI-1 ␤-sheet A (41,49). The mutant, PAI-1 R , was constructed on the wild-type PAI-1 background and is disabled by the introduction of two Arg residues at positions 333 and 335 (residues P 14 and P 12 of the reactive center loop). These residues are in the hinge region of the reactive center loop, and upon reaction with a proteinase, they normally fold into ␤-sheet A with their side chains oriented toward the interior of PAI-1. Mutations in this region have been found to greatly reduce the rate of conformational rearrangement associated with proteinase inhibition, converting the inhibitor to a sub-3 S. Stefansson and D. A. Lawrence, unpublished data.
FIG. 6. PAI-1 R retains high affinity for vitronectin after proteinase cleavage. Solid phase competitive inhibition binding assays with increasing concentrations of PAI-1 or PAI-1 R , either native or cleaved, competing for the binding of 32 P-labeled PAI-1 to chicken vitronectin. E, PAI-1; ‚, PAI-1 R ; q, cleaved PAI-1; OE, cleaved PAI-1 R . The curves for PAI-1, PAI-1 R , and cleaved PAI-1 R were generated with a four-parameter fit as described (65), and the error bars represent the S.E.

FIG. 7. Inhibition of angiogenesis by noninhibitory PAI-1 R or vitronectin antibodies.
A, quantitation of the CAM assay with fully active PAI-1 (E) and PAI-1 R (Ⅺ). The inset shows an expanded view of the origin. The curves for active PAI-1 and PAI-1 R were generated with a four-parameter fit as described (65). B, quantitation of angiogenesis in the CAM assay with or without treatment with either affinity-purified anti-vitronectin antibodies or normal rabbit IgG at 15 g/day. p values of Ͻ0.05 relative to the FGF-2-only control (None) are shown. In both A and B, the data are presented as the percentage of the FGF-2-only control (100%), and n Ն 6 for each condition. C and D show a vessel in a section of CAM tissue harvested after 3 days of treatment with FGF-2 and immediately frozen prior to analysis. C is the light image, and D is the same section stained with antichicken vitronectin antibodies and detected with rhodamine labeled goat antirabbit IgG. The original magnification was ϫ 400. strate (31,49). Therefore, these two substitutions would be expected to greatly retard the insertion of the reactive center loop into ␤-sheet A upon interaction with a proteinase and also to prevent the mutant from adopting the latent conformation. Since loop insertion also results in loss of vitronectin affinity (13), then the PAI-1 R mutant should not only fail to inhibit all proteinases; it should also retain significant vitronectin affinity after reaction with a proteinase. This is seen in Fig. 6, where the native forms of both active-PAI-1 and PAI-1 R are shown to inhibit the binding of 32 P-PAI-1 to vitronectin. However, following cleavage in their reactive center loops and incubation for 24 h, only the PAI-1 R mutant retains the ability to compete for vitronectin binding (compare the filled triangles with the filled circles in Fig. 6). This indicates that, as predicted, the conformational rearrangement of the PAI-1 R mutant is very slow, and thus it should retain high affinity for vitronectin even in an environment that is high in proteinase activity. Fig. 7A demonstrates that PAI-1 R is a very effective inhibitor of FGFmediated angiogenesis. Since the PAI-1 R is devoid of all proteinase-inhibitory activity, this indicates that angiogenesis can be substantially blocked by PAI-1 in a proteinase-independent manner. This inhibition is also dose-dependent, and PAI-1 R appears to be significantly more effective than PAI-1 A , inhibiting angiogenesis by ϳ41% at 0.1 M (Fig. 7A), compared with only 26% for PAI-1 A at a 10-fold higher concentration (see 1 M; Fig. 6B). This difference is probably due to the ability of PAI-1 R to retain vitronectin affinity even in the presence of proteinase activity.
Detailed analysis of the IC 50 values for the inhibition of angiogenesis by PAI-1 R , and comparison with fully active PAI-1 appear to yield identical results (compare 0.115 Ϯ 0.010 M for PAI-1 R and 0.115 Ϯ 0.033 M for PAI-1 in this experiment), suggesting that the noninhibitory PAI-1 R mutant is as effective an inhibitor of angiogenesis as fully functional PAI-1. However, examination of the two dose-response curves clearly indicates that despite very similar IC 50 values, the two fits are not the same, with the PAI-1 R mutant having a very different slope and range compared with active PAI-1. This can be seen in Fig.  7A, where the fit for active PAI-1 yields a range of 100% and clearly goes to zero, while the range obtained with PAI-1 R is only 85% and instead of going to zero approaches an asymptotic value of ϳ85% inhibition of angiogenesis. This suggests that, as in vitro (52), vitronectin in the matrix may enhance vascular cell migration and possibly vessel formation. In this case, it may be that the inhibition of angiogenesis by PAI-1 R results from the capacity of this noninhibitory mutant to block vitronectin's enhancement of vascular cell migration and/or vessel formation in vivo. If this were the mechanism, then the PAI-1 R mutant should reduce angiogenesis but be unable to block it completely, even in the presence of excess PAI-1 R . This appears to be exactly the result obtained (Fig. 7A) and suggests that, in vivo, vitronectin enhances angiogenesis, possibly by increasing the rate of vascular cell migration. To test whether simply blocking access to vitronectin could inhibit angiogenesis, CAMs were treated with FGF-2 in the presence or absence of an affinity-purified polyclonal antibody to chicken vitronectin. This is shown in Fig. 7B and demonstrates that antibodies to vitronectin reduce angiogenesis by 57% in the CAM. This confirms that blocking vitronectin is sufficient to reduce angiogenesis and supports the conclusion that the PAI-1 R mutant inhibits angiogenesis by preventing cellular access to vitronectin in the matrix. Finally, this proposition is further supported by immunofluorescence analysis of FGF-2-treated CAM tissue, which confirms that vitronectin is present in the vascular wall in a diffuse pattern that suggests its association with the vascular basement membrane (Fig. 7, C and D).
The inhibition of angiogenesis by PAI-1 R and the antivitronectin antibody complements previous studies demonstrating that integrin antagonists such as synthetic RGD containing peptides and function-blocking antibodies to the vitronectin receptor, integrin ␣ v ␤ 3 , can inhibit FGF-2-mediated angiogenesis (53,54). While these earlier studies indicated that ␣ v ␤ 3 was required for angiogenesis, they did not indicate whether or not there was a specific matrix component that was necessary for this process, since ␣ v ␤ 3 binding is not strictly limited to vitronectin. The observation that a nonproteinase inhibitory PAI-1 mutant as well as antibodies to vitronectin can significantly reduce angiogenesis suggests that the interaction of ␣ v ␤ 3 with vitronectin is important for a normal angiogenic response to FGF-2. Furthermore, since neither agent can completely inhibit angiogenesis even at concentrations where vitronectin is likely to be saturated, it would appear that other matrix proteins can also facilitate vessel formation but that angiogenesis is likely to be optimal when vitronectin is present in the provisional angiogenic matrix. This is consistent with the in vitro cell migration results, where the inclusion of vitronectin in the matrix enhances the migration of both smooth muscle cells and endothelial cells and where PAI-1 was able to block this enhancement (10,26). 2 Of course, it is not possible to unambiguously derive the molecular mechanism by which PAI-1 inhibits angiogenesis. However, taken together, the data presented here suggest that both PAI-1 and vitronectin play a role in angiogenesis and that both PAI-1's proteinase inhibitory activity and its binding to vitronectin may be significant components of PAI-1's inhibition of angiogenesis.
In conclusion, the present study demonstrates that exogenously added PAI-1 at supraphysiologic concentrations is a potent inhibitor of angiogenesis in the CAM. We have also shown that this inhibition appears to act via two separable but probably overlapping pathways, one of which involves inhibition of proteinase activity, most likely chicken plasmin, and another that is independent of PAI-1 proteinase inhibitory activity but requires high affinity binding of PAI-1 to vitronectin. That PAI-1 can regulate angiogenesis is consistent with the reported regulation and expression of PAI-1 in vitro, since angiogenic factors such as FGF and vascular endothelial cell growth factor have both been shown to induce PAI-1 expression in endothelial cells (55,56). PAI-1 in turn has been shown to regulate the expression of FGF in injured endothelial cells, suggesting a possible feedback loop for these two proteins (57). PAI-1 has also been shown to be specifically induced in cells surrounding angiogenic vessels, increasing only in cells directly in contact with the sprouting endothelium (58). However, previous in vivo studies examining the role of PAI-1 in tumor growth and angiogenesis have been more controversial, since different studies have yielded contradictory results. In one study, tumor cells transfected with the PAI-1 cDNA showed a slight reduction in tumor-associated angiogenesis (59), whereas two other studies using PAI-1-deficient mice suggested alternatively that either tumor angiogenesis requires PAI-1 (60) or that PAI-1 levels have no effect on tumor growth (61). These conflicting results might be due to the ability of PAI-1 to influence angiogenesis in diverse ways. For example, an absence of PAI-1 could lead to excessive proteolysis resulting in loss of cell adhesion sites and inhibition of angiogenesis as was suggested by Bajou et al. (60). Alternatively, our results indicate that inhibition of proteolysis can also inhibit angiogenesis. This could result from an inhibition of fibrinolysis in the provisional angiogenic matrix (45,62), which could impact endothelial cell migration. In this model, the effect would be similar to the reduced wound healing response associated with plasminogen deficiency, which is thought to be due to an in-ability of cells to migrate into the fibrin matrix in the absence of plasmin (63,64). Finally, our results suggest that PAI-1's association with vitronectin can also regulate angiogenesis. It may be that in some circumstances vitronectin and PAI-1 act together in vivo to either promote or inhibit angiogenesis. In this model, vitronectin present in the provisional matrix might enhance angiogenesis by promoting vascular cell migration, and PAI-1, by controlling access to the integrin adhesion site on vitronectin, might regulate this enhancement.