Thrombin Functions through Its RGD Sequence in a Non-canonical Conformation*

Previous studies have suggested that thrombin interacts with integrins in endothelial cells through its RGD (Arg-187, Gly-188, Asp-189) sequence. All existing crystal structures of thrombin show that most of this sequence is buried under the 220-loop and therefore interaction via RGD implies either partial unfolding of the enzyme or its proteolytic digestion. Here, we demonstrate that surface-absorbed thrombin promotes attachment and migration of endothelial cells through interaction with αvβ3 and α5β1 integrins. Using site-directed mutants of thrombin we prove that this effect is mediated by the RGD sequence and does not require catalytic activity. The effect is abrogated when residues of the RGD sequence are mutated to Ala and is not observed with proteases like trypsin and tissue-type plasminogen activator, unless the RGD sequence is introduced at position 187–189. The potent inhibitor hirudin does not abrogate the effect, suggesting that thrombin functions through its RGD sequence in a non-canonical conformation. A 1.9-Å resolution crystal structure of free thrombin grown in the presence of high salt (400 mm KCl) shows two molecules in the asymmetric unit, one of which assumes an unprecedented conformation with the autolysis loop shifted 20 Å away from its canonical position, the 220-loop entirely disordered, and the RGD sequence exposed to the solvent.

Thrombin is a multifunctional serine protease that plays key roles in blood coagulation and vascular development (1). The procoagulant role unfolds upon cleavage of fibrinogen but also the activation of other clotting factors like fXIII, fVIII, fV, and fXI. The prothrombotic role depends on the cleavage of protease-activated receptors, which are also responsible for the signaling functions of the enzyme during embryonic development (2,3). Finally, the anticoagulant role of the enzyme depends on the cleavage of protein C with the assistance of the cofactor thrombomodulin (4). In addition to these well characterized roles that require catalytic activity, thrombin promotes chemotaxis even when its active site is blocked (5). Thrombin also promotes endothelial cell adhesion upon interaction with glypican (6) by exploiting its RGD (Arg-187, Gly-188, Asp-189) sequence (7,8), a motif involved in integrin binding (9) that is not present in many other proteases. During evolution, the RGD sequence of thrombin was retained in almost all species, from hagfish to human (10). Conservation of this sequence, however, may be due to other factors. Arg-187 makes an important ion-pair interaction with Asp-222 in the 220-loop to stabilize the Na ϩ binding environment that guarantees optimal catalytic activity (11,12). Asp-189, on the other hand, defines the trypsin-like primary specificity of the enzyme (13).
The possible functional role of the RGD sequence in thrombin is called into question by the crystal structure where Arg-187 is exposed to the solvent, but Gly-188 and Asp-189 are almost completely buried under the 220-loop (14). As a possible solution to this conundrum, Bar-Shavit et al. (7,8) have suggested that thrombin functions through its RGD sequence only after denaturation or proteolytic cleavage, either by thrombin itself (7) or plasmin (8). However, recent studies have shown that surface-absorbed thrombin, or its active site inhibited form, promote attachment, migration, and survival of endothelial cells via the ␣ ␤ 3 integrin (15,16). Other studies have provided evidence that thrombin in solution can interact with ␣ ␤ 3 integrin in a way that is inhibited by RGD mimetics (17,18). Hence, the enzyme may expose the RGD sequence independent of denaturation or proteolysis, as reported here.

MATERIALS AND METHODS
Site-directed Mutagenesis-Site-directed mutants of human thrombin, tissue-type plasminogen activator (tPA), 1 and rat trypsin were expressed, purified, and tested for activity as described (19 -22). The thrombin mutant G188A (chymotrypsinogen numbering) was prepared in this study, and its k cat /K m values toward physiologic and synthetic substrates were found to be compromised up to 100-fold relative to wild type. All other thrombin mutants were characterized previously (11,13). The trypsin mutant trypsin-RGD was made by replacing the 184a-188 sequence FLEGGK with the thrombin 184a-188 sequence YKP-DEGKRG. The tPA mutant tPA-RGD was made by replacing the 184a-188 sequence DTRSGGPQANLH with the thrombin 186a-188 sequence YKPDEGKRG. The thrombin sequence was inserted in trypsin and tPA between Gly-184 and Asp-189 that are conserved and occupy spatially equivalent positions in all three proteases. The activity of trypsin-RGD toward chromogenic substrates was reduced 10-fold relative to wild type. The activity of tPA-RGD was significantly more compromised, reaching 10,000-fold toward the natural substrate plasminogen.
Cell CA). Forty-eight-well plates were coated overnight with 1 g/well wildtype or mutant thrombins, tPA, or trypsins. Coating was confirmed and quantified by protein determination (Bradford assay) after solubilization in NaOH. In control experiments with mutants defective for attachment, the amount of protein was increased up to 30 g/well. Bovine serum albumin (BSA) was used as a reference. The wells were then blocked for 60 min with 3% BSA in phosphate-buffered saline at 37°C. HUVECs were detached with 0.526 mM EDTA/phosphate-buffered saline, suspended in serum-free medium containing 0.3% BSA (SFM-0.3% BSA) and incubated in the presence or absence of the GRGDSP peptide (Bachem, King of Prussia, PA) or mouse monoclonal antibody against human ␣ ␤ 3 integrin (clone LM609; Chemicon International Inc., Temecula, CA). The GRGDSP peptide is used for the affinity purification of fibronectin receptor (␣ 5 ␤ 1 integrin), as it contains the RGD integrin recognition site of the fibronectin cell binding domain. It is also a potent inhibitor of cell attachment to fibronectin through interaction with ␣ 5 ␤ 1 integrin. Cell suspensions (10 5 /well) were then added to the wells, and the plates were incubated at 37°C for 60 min. The nonadherent cells were aspirated and the number of the adhered cells was measured in triplicate wells by means of the endogenous enzyme hexosaminidase. In representative samples, endothelial cells attached on surface-absorbed proteins were fixed, stained, and photographed using an inverted microscope (Olympus IX70) equipped with photometric CoolSNAP HQ camera in ϫ10 objective lens magnification.
Cell Migration-Cell migration was studied using the modified Boyden chamber assay (Chemicon International Inc. kit: QCM TM 24-well colorimetric cell migration assay) where the upper and lower chambers are separated by 8-m pore polycarbonate filters. In haptotactic cell motility assay, the undersurface of the membrane filter was precoated with 1 g/well wild-type or mutant thrombins, tPA, or trypsin. In control experiments with mutants defective for migration, the amount of protein was increased up to 30 g/well. BSA was used as a reference. To modulate the migration toward immobilized thrombin, lower chambers were filled with SFM-0.3% BSA containing the GRGDSP peptide or mouse monoclonal antibody against human ␣ v ␤ 3 integrin (clone LM609). Endothelial cells were added to the upper compartment of the chamber at a density of 10 3 /l in SFM-0.3% BSA and were incubated for 6 h at 37°C allowing migration in the lower chamber. Cells on the filters that did not migrate (cells on upper surface) were removed by wiping with cotton swabs. Migrated cells were assessed according to the manufacturer's protocol. All experiments were carried out in triplicate.
Crystallization Studies-The R77aA mutant of thrombin, devoid of its site of autoproteolytic digestion (11), was crystallized to resolve the structure of thrombin bound to K ϩ and free of any inhibitors. The protein was concentrated to 5.6 mg/ml in 50 mM choline chloride, 20 mM Tris, pH 7.4. Crystallization was achieved at 25°C by vapor diffusion against 20% polyethylene glycol 2000-monomethyl ether, 0.1 M Bis-Tris, pH 6.6, and 400 mM KCl. Equal volumes of the protein sample and reservoir solution were mixed (2 l each) to prepare the sitting drops. Diffraction quality crystals (ϳ0.3 ϫ 0.06 ϫ 0.06 mm) grew after 7 days. Crystals were cryoprotected in paratone oil prior to flash-freezing. Data were collected from a single crystal at the Advanced Photon Source (beamline 14-BM-C, Argonne National Laboratory) and processed using the HKL2000 package. The crystals were orthorhombic, in space group P2 1 2 1 2 1 and contained two molecules per asymmetric unit (Table I). The structure was solved by molecular replacement using the coordinates of the thrombin-PPACK complex (11) and the program MOLREP from the CCP4 package (23). Crystallographic refinement was carried out by simulated annealing and conjugated-gradient minimization using CNS (23), and model building was performed with the program XtalView (24). Coordinates have been deposited in the Protein Data Bank (accession code 2A0Q).

RESULTS AND DISCUSSION
Recent studies have documented that surface-absorbed thrombin promotes endothelial cell attachment and migration via interaction with ␣ v ␤ 3 integrin (15). The effect appears to be mediated by the RGD sequence, but no direct demonstration could be provided in that or any previous studies (6 -8, 16). Fig. 1 documents the ability of immobilized wild-type thrombin to mediate the attachment of HUVECs. Inactivation of thrombin with the S195A mutation, which replaces the active site Ser with Ala, did not abrogate this property. This is in agreement with previous studies of thrombin inactivated at the active site with diisopropylphosphofluoridate (15). The ability to promote endothelial cell attachment was also retained by mutants significantly compro-mised in their cleavage of fibrinogen or PARs (e.g. E80A, Y225A, D221A, D222A) (19), as shown in Fig. 2. In contrast, mutants of the RGD sequence (R187A, G188A, and D189A) failed to support cell attachment (Figs. 1 and 2) even when the amount of protein used in the assay was increased 30-fold. The effect is highly selective because mutants of residues Lys-186d (K186dA) and Ala-190 (A190S) flanking the RGD sequence behaved like wild type. In a Boyden chamber assay, HUVECs readily migrated through a microporous membrane toward surface absorbed with wild-type thrombin or any mutant not carrying substitutions of the RGD sequence, but RGD mutants were ineffective in promoting cell migration (Fig. 2).
A striking demonstration of the involvement of the RGD sequence comes from experiments carried out with trypsin and tPA. These proteases are highly homologous to thrombin but do not possess the RGD sequence. Accordingly, they were unable

FIG. 1. Attachment of endothelial cells to wild-type (wt) and mutant thrombins.
HUVECs were incubated on wells coated with 1 g/well of BSA or wild-type or mutant thrombins S195A, R187A, G188A, and D189A. Attached cells were photographed with an inverted microscope after 1 h incubation at 37°C. to induce attachment and migration of HUVECs (Fig. 2). However, when residues 187 and 188 were replaced with Arg-Gly as found in thrombin (residue 189 is Asp in both trypsin and tPA), the mutants acquired the ability to induce attachment and migration of HUVECs as wild-type thrombin (Fig. 2). Need of the RGD sequence embedded in a proper protein scaffold is also demonstrated by the lack of activity of the thrombin peptide KRGDAC (Fig. 2), as opposed to the activity shown by the longer 23-amino acid thrombin peptide TP508 encompassing the RGD sequence (16).
Endothelial cell attachment and migration toward immobilized wild-type or S195A mutant thrombin was found to depend on ␣ v ␤ 3 and ␣ 5 ␤ 1 integrins. When endothelial cells were pretreated with 10 g/ml anti-␣ v ␤ 3 antibody (LM609) or 100 M GRGDSP peptide, which blocks the fibronectin receptor ␣ 5 ␤ 1 , attachment and migration of cells were significantly inhibited (Fig. 2). The effect was observed for both wild-type thrombin and the S195A mutant.
To explore further the nature of thrombin conformation under the conditions employed in our cell attachment and migration assays, thrombin was challenged with the potent and highly selective inhibitor hirudin. Hirudin binds to thrombin in the femtomolar range (25) and is an exceptionally good probe of the conformational state of the enzyme (26) because it covers 20% of its solvent exposed surface area (27). When thrombin was preincubated with hirudin at a 1:20 ratio and the enzymeinhibitor complex was surface-absorbed, endothelial cells could not attach and migrate (Fig. 2). On the other hand, when thrombin was first surface-absorbed and hirudin was added to the media with the cells, the inhibitor was unable to abrogate attachment or migration. These findings suggest that when thrombin is anchored to the support, it assumes a non-canonical conformation that exposes the RGD sequence and prevents hirudin binding. The inability to bind hirudin presages a drastic conformational rearrangement of the enzyme.
Evidence that thrombin can indeed expose the RGD sequence through a drastic conformational transition comes from a serendipitous observation garnered from the crystal structure of the enzyme solved under experimental conditions never before explored. In an effort to understand the molecular basis of thrombin preference for Na ϩ versus K ϩ (28), the enzyme was crystallized free of any inhibitors in the presence of 400 mM KCl. The high concentration of K ϩ was made necessary by the significantly weaker affinity of this cation compared with Na ϩ (29). The two molecules in the asymmetric unit are related by a rotation of 173° (Fig. 3). Molecule-1 has K ϩ bound to the Na ϩ site and assumes a conformation similar to the Na ϩ -bound F form (11). The second molecule in the asymmetric unit, molecule-2, has no K ϩ bound to the Na ϩ site and assumes a conformation that has no counterpart in over 180 structures of thrombin currently deposited in the Protein Data Bank.
Packing of the two molecules in the asymmetric unit is quite different from that of the Na ϩ -bound F form of thrombin and includes a second K ϩ bound at the interface. The intermolecular contacts are more extensive in the presence of KCl, with molecule-1 almost crushing molecule-2 in the contact region and squeezing the autolysis loop away from its canonical position (Fig. 3, blue). The entire segment from Asn-143 to Lys-149e comprising the autolysis loop (residues 144 -149e) of molecule-1 is disordered and its electron density could not be traced, as typically found in thrombin structures (11). On the other hand, this region could be traced for molecule-2 up to Thr-147 and from Gln-151 on (Fig. 3, blue). Remarkably, the proximal segment of the autolysis loop of molecule-2 overlaps that of the Na ϩ -bound F structure of thrombin up to the C␣ atom of Gly-140 and then repositions itself toward exosite I by making a sharp turn that dislodges the sequence around Glu-146 more than 20 Å away from its canonical position. The two backbone traces overlap again at the level of the backbone oxygen atom of Ser-153. One consequence of this sharp rearrangement of the autolysis loop of molecule-2 is the breaking of a key salt bridge between the side chains of Glu-146 and Arg-221a. At the other end of the interface, the entire 186-loop of molecule-2 is pulled and Pro-186 moves upward toward the 220-loop causing a total disorder in the region from Asp-221 to Gly-223 that cannot be traced in the electron density. These destabilizing interactions in the 186-and 220-loops are not seen in molecule-1, because the K ϩ bound in the cation site

FIG. 2. Requirement of the RGD sequence of thrombin for attachment (black bars) and migration (gray bars) of endothelial cells.
In cell attachment experiments, HUVECs were plated on wells coated with 1 g/well wild-type or mutant thrombins, trypsins, or tPAs and allowed to attach for 1 h at 37°C. Trypsin-RGD and tPA-RGD refer to trypsin and tPA mutants containing the RGD sequence at position 187-189 as in thrombin. The effect of hirudin was studied by preincubating thrombin with the inhibitor (1:20 molar ratio) and then coating the wells with the complex (Hirudin-wt) or alternatively by adding hirudin (5 g/ml) with the cells to the wells coated with wild-type thrombin (ϩHirudin). The effect of integrin antagonists was studied by addition of LM609 (10 g/ml) or GRGDSP (100 M) with the cells to wells coated with wild-type (ϩLM609, ϩGRGDSP) or S195A mutant (S195AϩLM609, S915AϩGRGDSP) thrombin. The same reagents at the same concentrations were used to coat the undersurface of membrane filters in migration experiments, and cells were added to the upper compartment of the Boyden chamber. Values obtained in the presence of BSA were used as controls in both attachment and migration experiments to calculate the fold increase shown in the plot. In migration assays, cells on the lower surface were also counted manually under the microscope in six predetermined high magnification microscopic fields. The number per field ranged from 89 Ϯ 14 for wild-type thrombin to 14 Ϯ 7 for mutants of the RGD sequence, with 4 Ϯ 3 measured in the control with BSA.
stabilizes the architecture of the region. The massive disruption of the cation binding site in molecule-2 is completed by a collapse of the ␤-strand from Gly-216 to Cys-220. The unprecedented structural changes in molecule-2 culminate in the exposure of the RGD sequence to the solvent (Fig. 4).
The drastic changes in molecule-2 reported here are not triggered by the R77aA mutation that resides far from the intermolecular interface (Fig. 3) generated by the crystal packing. The changes are likely the result of the crystallization conditions, namely the absence of any inhibitor and the presence of high (400 mM) KCl. Thrombin can therefore assume a non-canonical conformation where the 220-loop is disordered and the RGD sequence is exposed to the solvent. This conformation would have little affinity toward hirudin, consistent with the results in Fig. 2, because of the massive disruption of the 220-loop hosting critical residues for hirudin recognition (26). The collapse of the primary specificity pocket would also compromise catalytic activity, which is, however, not required for RGD function (Figs. 1 and 2).
We believe that exposure of the RGD sequence requires attachment to solid supports where thrombin is subject to constraints and forces not seen in solution. These conditions may be mimicked by surface-coating or crystallographic packing and may reproduce the crowded environment of the extracellular matrix where thrombin migrates upon vascular injury (6 -8, 15, 16). However, recent studies have suggested that native thrombin in solution can interact with ␣ v ␤ 3 integrin in a way that can be inhibited specifically by RGD mimetics (17,18), which raises the possibility that exposure of the RGD sequence may also occur independent of matrix attachment. The conformation of molecule-2 reported here likely captures thrombin in the act of completely unraveling the RGD sequence and the 220-loop region, which may be revealed in future studies of the free form of the enzyme. FIG. 3. Spatial arrangement of the two molecules in the asymmetric unit in crystals of thrombin grown in the presence of 400 mM KCl. The two molecules come into contact through an extended interface that involves the autolysis loop of molecule-2 and the 186-loop of both molecules. Highlighted are the regions from Gly-140 to Ser-153 (green), containing the autolysis loop 144 -149e, and from Gly-218 to Tyr-225 (blue) containing the 220-loop. There are two K ϩ (red balls, arrows) in the structure, one bound to molecule-1 in the cation binding site and the second at the interface. Note how the 218 -225 region is clearly visible in molecule-1 but only partially so (residues 221-223 are missing) in molecule-2. Also note the drastic, 20 Å dislodgement of the autolysis loop in molecule-2 that makes a sharp turn toward exosite I and away from its canonical position close to the 220-loop (11,14). The site of mutation, R77aA, is indicated by * in both molecules and is remote from the packing interface. Molecule-1 assumes the standard conformation of thrombin in this region, with Asp-222 in the 220-loop covering Gly-188 and Asp-189 of the RGD sequence, but leaving Arg-187 exposed to the solvent. The drastic changes observed in molecule-2 result in complete disorder of the 221-223 sequence, not traceable in the electron density map, and exposure of the RGD motif.