Structure-based Engineering of Species Selectivity in the Interaction between Urokinase and Its Receptor

The high affinity interaction between the urokinase-type plasminogen activator (uPA) and its glycolipid-anchored receptor (uPAR) is decisive for cell surface-associated plasminogen activation. Because plasmin activity controls fibrinolysis in a variety of pathological conditions, including cancer and wound healing, several intervention studies have focused on targeting the uPA·uPAR interaction in vivo. Evaluations of such studies in xenotransplanted tumor models are, however, complicated by the pronounced species selectivity in this interaction. We now report the molecular basis underlying this difference by solving the crystal structure for the murine uPA·uPAR complex and demonstrate by extensive surface plasmon resonance studies that the kinetic rate constants for this interaction can be swapped completely between these orthologs by exchanging only two residues. This study not only discloses the structural basis required for a successful rational design of the species selectivity in the uPA·uPAR interaction, which is highly relevant for functional studies in mouse models, but it also suggests the possible development of general inhibitors that will target the uPA·uPAR interaction across species barriers.

The urokinase-type plasminogen activator receptor (uPAR) 3 is a glycolipid-anchored membrane protein (1) that recognizes the serine protease urokinase-type plasminogen activator (uPA) and its zymogen (pro-uPA) with very high affinity and specificity (2). This interaction is exclusively mediated by the N-terminal growth factor-like domain (GFD  ) of uPA, and it is indispensable for the focalized uPA-mediated plasminogen activation that occurs at cell surfaces both in vitro (3,4) and in vivo (5)(6)(7). Several independent studies have correlated uPAR expression in vivo to e.g. neutrophil infiltration (8 -10) and to various pathological conditions such as cancer invasion and metastasis (11)(12)(13), hepatic fibrin deposition (14,15), and kidney barrier function (16). Accordingly, uPAR has been proposed as a promising molecular target for intervention and/or cytotoxin-based cancer therapies (14,17,18). With the goal of future monitoring efficacy of such treatment modalities, we have developed a high affinity peptide antagonist of the human uPA⅐uPAR interaction (19), which has proven well suited for non-invasive in vivo imaging of uPAR by positron emission tomography (20).
The extracellular domains of uPAR comprise three homologous Ly-6/uPAR (LU)-type modules, all of which adopt the three-fingered folding topology that is defined by the structures of the single domain snake venom ␣-neurotoxins (21)(22)(23). Crystal structures recently solved for human uPAR in complex with the abovementioned peptide antagonist (24), the N-terminal fragment (ATF) of uPA (25), or the somatomedin B (SMB) domain of vitronectin (26) all confirm this topology, and more importantly, they consistently reveal the presence of a large hydrophobic uPA-binding cavity in uPAR that requires all three LU domains for its assembly. In total, Ͼ2000 Å 2 of solvent-accessible surface is buried at the interface of the ATF⅐uPAR complex (25,27), and the functional epitope for this high affinity interaction (K D ϳ 0.2 nM) has been identified for the human uPAR by systematic alanine-scanning mutagenesis (28,29).
Biochemical analyses of the binding properties for human and murine components reveal, however, that the uPA⅐uPAR interaction exhibits species selectivity, where affinities for cross-species interactions are significantly reduced (30,31). This species barrier is important and has to be considered when testing the pharmacological effects of various uPAR antagonists in mouse cancer models. Likewise, this difference also has to be taken into account when testing the functional roles of uPAR in proteolysis and signaling in vivo during, e.g. cancer invasion and metastasis using xenotransplanted tumor models (32). To obtain a detailed molecular characterization of the structures responsible for this species selectivity, we have now solved the crystal structure of the mouse ATF⅐uPAR complex at 3.1 Å and compared this to the orthologous human complex. This enabled the present structure-driven design of a genetically engineered mouse uPA that preferentially binds human uPAR by swapping only two positions in the GFD. The reciprocal swapping is also demonstrated for human uPA. This knowledge provides an important structural platform guiding further studies on the functional aspects of uPAR and its ligands in normal physiology and pathophysiology using, e.g. transgenic mice, where this species barrier has been eliminated. Our structural data may furthermore assist future rational development of antagonists of the uPA⅐uPAR interaction that are not compromised by the species barrier.
Expression and Purification of Soluble Recombinant Mouse uPAR and Mouse ATF-Mouse pro-uPA (muPA, residues 1-413), mouse ATF (mATF, residues 1-143), as well as a secreted form of mouse uPAR (muPAR, residues 1-275) were all expressed by Drosophila melanogaster S2 cells after stable transfection with pMT-expression vectors entailing cDNAs encoding the signal and protein sequences for the respective proteins. Expression was induced by 0.5 mM Cu 2 SO 4 for 7 days at 25°C, and in the case of pro-uPA the media also contained 10 g/ml aprotinin to prevent proteolytic cleavage at the activation site (29). muPAR and a selected number of single site mutants thereof were purified by immunoaffinity chromatography using an immobilized mouse monoclonal anti-uPAR antibody (KOR-1) raised against purified huPAR in a uPAR-deficient transgenic mouse (35). Studies by mass spectrometry using a protocol developed for huPAR (36) revealed that our purified muPAR preparation was heterogeneously glycosylated with roughly 25% carrying bi-antennary glycans on three sites, 50% on four sites, and 25% on five sites. Subsequent peptide mass mapping by MALDI-tandem mass spectrometry (MS) revealed that the recombinant muPAR carried N-linked glycans on Asn 52 , Asn 160 , Asn 170 , Asn 198 , and Asn 259 , whereas the potential sites Asn 9 and Asn 231 remained unmodified (data not shown).
Crystallization and Data Collection for the mATF⅐muPAR Complex-The mATF⅐muPAR complex was initially formed at 65 M before separation by size-exclusion chromatography (Superdex75 TM in 20 mM Tris-HCl, 50 mM NaCl, pH 8.0). Subsequent deglycosylation was required for optimal crystal growth, and the purified mATF⅐muPAR complex was therefore further concentrated to 10 mg/ml and incubated overnight at room temperature with 10 units peptide N-glycosidase F/mg complex. The crystals were generated by the microdialysis method (Hampton Research, Aliso Viejo, CA) using a precipitant solution of 5-7% (w/v) polyethylene glycol 3350, 50 mM Bis-Tris at pH 5.5. Crystals were harvested using a solution of 20% (w/v) polyethylene glycol 3350, 20% (v/v) glycerol, 50 mM Bis-Tris at pH 5.5. This solution was also used as a cryoprotectant for x-ray diffraction data collection at Ϫ160°C.
Structure Determination and Refinement-The x-ray diffraction data (supplemental Table S1) for crystals of mATF⅐muPAR complexes were collected using synchrotron radiation at NSLS beam lines x12b and x29 of the National Synchrotron Light Sources at Brookhaven National Laboratory. The crystals diffracted to 3.1 Å with the beam line x29. The structure of the hATF⅐huPAR complex (1FD6) (25) were positioned into the mATF⅐muPAR crystals by molecular replacement (37). The orientations and positions of each of the five modules of the complex (DI, DII, and DIII in muPAR and GFD and kringle in mATF) were refined by a rigid-body refinement protocol at 6 Å and then 4 Å. This structure was subjected to several iterative rounds of refinement by using programs CNS (38) and REFMAC (39), and by using a manual model adjustment with programs COOT (40) and O (41). There are two mATF⅐muPAR complexes in the crystallographic asymmetric unit. No non-crystallographic symmetry restrains were used between these two complexes during refinement. The structure was refined to an R value of 0.239 and an R free value of 0.339 (supplemental Table S1). The final structures were analyzed by PROCHECK (42), PyMOL (43), and MOLSOFT ICM (44). The coordinates of the reported structure have been deposited in the Protein Data Bank (3LAQ).
Binding Kinetics Assessed by Surface Plasmon Resonance-Interaction studies were carried out in real-time using a Biacore 3000 TM instrument (Biacore, Uppsala, Sweden). In all experiments, 10 mM HEPES, 150 mM NaCl, 3 mM EDTA, and 0.005% (v/v) surfactant P-20 at pH 7.4 was used as running buffer. Purified, recombinant pro-uPA was immobilized on a 100 nm carboxymethyl-dextran matrix (CM5 sensor chip) at two different levels of 0.25 and 1 g/ml, providing surface densities between 8 and 32 fmol/mm 2 . Covalent coupling was performed with N-hydroxy-succinimide/N-ethyl-NЈ- [3-(diethylamino)propyl]-carbodiimide in 10 mM sodium acetate, pH 5.0, at a flow rate of 5 l/min for 6 min. To obtain as optimal preparations as possible for the evaluation of the kinetics for the uPA⅐uPAR interaction, purified receptor preparations were subjected to size-exclusion chromatography (Superdex200 TM ) in Biacore running buffer prior to analysis to remove any traces of aggregated material. Serial 2-fold dilutions of this monomeric uPAR (0.4 -400 nM) were subsequently passed over three flow cells with immobilized huPA, muPA, and one mock coupled flow cell. Association was recorded for 300 s followed by a dissocia- tion phase of 775 s at a flow rate of 50 l/min at 20°C. The resulting sensorgrams were processed using double referencing (29). After each run, the sensor chip was regenerated by two consecutive injections of 0.1 M acetic acid in 0.5 M NaCl. The kinetic rate constants, k on and k off , were derived from these real-time interaction analyses by fitting the association and dissociation phases to a simple bimolecular interaction model using the BIAevaluation 4.1 software (Biacore), as described in detail previously (29). For weak interactions, K D and R max were also calculated from the equilibrium binding isotherm by nonlinear curve fitting assuming saturation of a single binding site, where R eq is the binding level at equilibrium, and R max is the binding capacity of the chip.

RESULTS AND DISCUSSION
Species Selectivity of the uPA⅐uPAR Interaction-To enable the first quantitative determination of the magnitude of the species barrier in the uPA⅐uPAR interaction by surface plasmon resonance, the relevant human and murine pro-uPAs and uPARs were expressed as secreted recombinant proteins by Drosophila S2 cells and purified by affinity chromatography. Studies on the interaction between purified muPAR and immobilized muPA by surface plasmon resonance (Fig. 1) revealed a high affinity interaction (K D ϳ 0.17 nM) that is comparable to that measured in parallel for the orthologous human components (K D ϳ 0.24 nM). This similarity also extends to the individual kinetic rate constants. Analyses of the corresponding mixed human and mouse components clearly reveal the lower affinity of the cross-species interactions. muPAR thus recognizes huPA with a 400-fold higher K D compared with huPAR, and this effect is entirely caused by an increase in k off . Along the same lines, huPAR recognizes muPA with a 79-fold lower affinity compared with muPAR, but in this case both kinetic rate constants are affected (Fig. 1E). Although this kinetic study reveals a relatively stringent species selectivity in the uPA⅐uPAR interaction, it does, nevertheless, also emphasize that the frequently used argument of a complete and infinite crossspecies barrier between man and mouse is ambiguous (31,32,45). To obtain a more detailed structural insight into the possible mechanisms responsible for this species difference, we subsequently solved the three-dimensional structure of mATF⅐muPAR by x-ray crystallography.
Overall Crystal Structure of the Murine ATF⅐uPAR Complex-The crystal structure of the mATF⅐muPAR complex was determined to 3.1 Å and refined to an R value of 0.238 and an R free value of 0.338 (supplemental Table S1). Most residues (91.1%) have allowed dihedral angle geometries, as estimated by PRO-CHECK (42). The crystals contain two ATF⅐uPAR complexes in the crystallographic asymmetric unit. The model derived for mATF includes residues 8 -132 as residues 1-7 were disordered, resembling previous findings for hATF in complex with huPAR (25)(26)(27). Electron densities corresponding to residues 82-92 and 227-231 of muPAR were also absent, and these regions were consequently omitted from the current model.
The structure we determine for muPAR consists of three homologous LU domains (denoted DI, DII, and DIII), forming a large hydrophobic ligand-binding cavity similar to the one observed in human uPAR (Fig. 2). The tight complex with mATF is primarily established by the burial of the ␤-hairpin from the GFD module of mATF in this central cavity of the receptor (Fig. 2B). The two mATF⅐muPAR complexes present in the asymmetrical unit cell are quite similar, with a root mean standard deviation (r.m.s.d.) of 1.3 Å for 1527 peptide mainchain atoms. No stereochemical restrains were placed between these two molecules during model refinement, suggesting that the current structure is not perturbed by crystal packing effects. The r.m.s.d. for the two mATF in the unit cell was 0.75 Å for 496 main-chain atoms. Furthermore, low r.m.s.d. values were also obtained for the individual modules in ATF, i.e. GFD (0.77 Å) FIGURE 2. Overall structure of the murine ATF⅐uPAR complex. The structure solved for the murine ATF⅐uPAR complex by x-ray crystallography is shown in A as a ribbon representation with the N-linked carbohydrates, which are defined by the electron density maps, as green sticks. The individual LUdomains in muPAR are shown in yellow (DI: residues 1-81), blue (DII: residues 94 -178), and red (DIII: residues 179 -274). The mATF (residues 8 -132), which comprises a growth factor domain (GFD) and kringle domain, is shown in cyan. The engagement of the ␤-hairpin of GFD in the tight interaction with uPAR is illustrated in B by a combined surface representation (muPAR) and ribbon diagram (mATF). Oxygen, nitrogen, carbon, and sulfur atoms in the surface representation are shown as red, blue, white, and yellow, whereas carbohydrates are shown in green. The position of the glycosyl-phosphatidylinositol anchor (GPI) is indicated in both panels. Nomenclatures for domain and secondary structure in uPAR follow previously established guidelines (24). and kringle (0.72 Å), which signify a low flexibility of the linker region between these modules in the receptor-bound state as opposed to the flexible nature observed for hATF in solution by NMR (46).
Consistent with the relatively high sequence identity between murine and human proteins (71% for uPA and 62% for uPAR, Fig. 3), the overall structure of the mATF⅐muPAR complex is quite similar to its human counterpart (25-27) displaying an r.m.s.d. of 2.5 Å for 1456 main-chain atoms (Fig. 4A). Although this translates into an almost identical overall topology of the mouse and human ATF⅐uPAR complexes, notable differences do nevertheless exist.
First, a small but clearly distinct 12°shift in the orientation of the central three-stranded ␤-sheet in uPAR DIII was observed between these species (Fig. 4A, inset). This rearrangement of DIII is in perfect agreement with previous notions of significant interdomain flexibilities in uPAR (24,25). The repositioning of this ␤-sheet is accompanied by a slightly different organization of loop 3 connecting ␣IIIE and ␤IIIF. Relative to its human ortholog, this loop contains one additional residue (Leu 258 ) in muPAR (Fig. 3), and the adjacent Asn 259 is modified by N-linked carbohydrate (Fig. 4A). In huPAR, Asn 259 forms an interdomain hydrogen bond to His 47 in DI, and no glycosylation is found in this region (25,27).
Second, a 3-residue deletion in loop 2 connecting ␤-strands ␤IIC and ␤IID in muPAR DII (Fig. 3) leads to a shorter and structurally well ordered loop (residue 130 -140), as illustrated in Fig. 4B. The corresponding loop in huPAR (residues 129 -142) is longer and partly disordered. Importantly, this loop harbors one of the key residues for uPA binding (Asp 140 in huPAR and Asp 138 in muPAR), and it undergoes a major repositioning of Ͼ10 Å depending on the ligand, which is loaded into the hydrophobic binding cavity (24,25,29). Intriguingly, the 3-residues truncation of this loop in DII is conserved among all published uPAR sequences from non-primate mammals (supplemental Fig. S1).
Structural Origin of the Species Selectivity by Elements Residing in GFD-A detailed analysis of the ligand-binding interfaces in the structures solved of human and murine uPA⅐uPAR complexes reveals several shared as well as unique properties. As mentioned previously, our crystal structure of the mATF⅐muPAR complex clearly shows that this interaction is primarily governed by the burial of the ␤-hairpin of GFD into the central cavity of muPAR (Fig. 2), which aligns well with the architecture and function disclosed for the human ortholog (25,48). In particular, the conserved Lys-Tyr-Phe triad along with the proximal Ile at the tip of the ⍀-loop in the ␤-hairpin are intimately engaged in receptor binding in both species. We denote this conserved interface "site 1" (Fig. 5, A, B, and E). In fact, these four residues account for no less than 42% of the total contact area in muPA (supplemental Table S2) and 54% in huPA (25). Accordingly, Ͼ80% of the exposed surface of Phe 26 in muPA is thus buried upon receptor binding, which contributes Ͼ100 Å 2 to the total contact surface of 683 Å 2 (supplemental Table S2). Further evidence for the existence of such a conserved hot spot for uPAR binding in the ⍀-loop of uPA (site 1) is provided by site-directed mutagenesis, which reveals that Phe 25 alone contributes no less than 3.8 kcal/mol to the human uPA⅐uPAR interaction (Table 1). A logic interpretation of our data leads to the hypothesis that a functional binding pocket, which accommodates these hot spot residues of site 1 in uPA, is conserved among receptors from different species. Targeting this allegedly conserved binding site in uPAR with small molecule mimetics could thus provide inhibitors that would enable pharmacological intervention across the species barrier.
With a view to the species selectivity in receptor binding it is noteworthy that the ␤-hairpin of GFD harbors only four positions that differ between man and mouse, i.e. Asn 22 /Tyr 23 , Asn 27 /Arg 28 , His 29 / Arg 30 , and Trp 30 /Arg 31 (color coded in Fig. 5E). Because GFD as an isolated domain maintains all structural information required for highaffinity uPAR binding (48), one or more of these four substitutions most likely accounts for this selectivity. Based on structural considerations, the Asn 22 /Tyr 23 and Trp 30 / Arg 31 substitutions, which we define as "site 2," appear particularly attractive as candidates for the species selectivity in uPAR recognition. First, these residues provide, along with residues from site 1, the largest contribution to the contact area with uPAR in both species (supplemental Table S2). Second, despite this contribution, they nevertheless display quite different chemical properties and three-dimensional arrangements in the orthologous uPAR complexes (Fig.  5, C and D), as opposed to those of site 1. Finally, a search in the MER-OPS data base (49) reveals that this Tyr-Arg dyad is evolutionally conserved among non-primate mammals (Fig. 5E), whereas the corresponding Asn-Trp dyad is restricted to uPA from primates only.
Quite the opposite applies for the remaining two positions in the

. Superimposition of human and mouse uPAR structures.
A shows the superimposed structures we have solved for human and mouse uPAR in complex with ATF after alignment on uPAR DI. The structures are shown as ribbon diagrams with DI, DII, and DIII in muPAR colored yellow, blue, and red, respectively, whereas huPAR is shown in gray. For clarity ATF is omitted from these representations. The inset shows the superimposition of the 3-stranded ␤-sheets of DIII after a 90°vertical rotation, which clearly reveals the 12°tilt of this ␤-sheet between species (highlighted by curved arrows). B shows a superimposition of the "ligand-loading loop" in uPAR DII. The connecting ␤-strands (␤IIC and ␤IID) are colored blue in the mATF⅐muPAR complex (3LAQ), green in the hATF⅐huPAR complex (2I9B), and red in the peptide⅐huPAR complex (1YWH). The repositioning of Asp 140 in huPAR upon uPA binding is indicated by the hatched arrow. The well defined structure of this short loop in muPAR (blue) positions the equivalent Asp 138 at an identical position relative to the bound ATF⅐mATF is shown as a combined surface and ribbon representation with its key residues for receptor binding shown as sticks. Nomenclatures for domain and secondary structure in uPAR follow previously established guidelines (24).
␤-hairpin of GFD that differ between muPA and huPA (i.e. Asn 27 /Arg 28 and His 29 /Arg 30 ). These residues are not conserved at all between non-primate mammals (Fig. 5E), and their side chains remain partially exposed and exhibit less striking interactions with uPAR in their respective receptor complexes (supplemental Tables S2 and S3). Accordingly, they are not likely to provide a significant contribution to the thermodynamics of receptor binding.
Swapping Species Selectivity by Structure-driven Design of the ␤-Hairpin in GFD-Because the binding between uPA and uPAR is governed entirely by the GFD module of uPA (48), and the majority of the side chains buried in the uPA⅐uPAR inter-face are indeed confined to the ␤-hairpin of GFD (25), we focused our subsequent mutagenesis efforts on this region to identify the major functional determinants of the species selectivity. Based on the previous structural examination of the mATF⅐muPAR interface, we first designed a quadruple muPA mutant having humanized all four positions in the ␤-hairpin of GFD, i.e. Tyr 23 3 Asn, Arg 28 3 Asn, Arg 30 3 His, and Arg 31 3 Trp. The resulting mouse pro-uPA Y23N/R28N/R30H/R31W mutant had an affinity toward huPAR that was indistinguishable from that of huPA (Table 2). Interestingly, this improvement was achieved through combined effects on both k on and k off , and this muPA mutant now exhibits kinetic rate constants for its interaction with huPAR that are identical to those recorded for the cognate ligand, huPA ( Table 2). The accompanying attenuation in the affinity for muPAR of muPA Y23N/R28N/R30H/R31W also replicates the kinetics observed for the mixed huPA⅐ muPAR interaction (Table 2), thus again demonstrating a perfect swapping of the species selectivity by humanizing only the ␤-hairpin of GFD. These kinetic data are essentially cooperated by IC 50 values determined for a recombinant mGFD carrying the same four mutations (50). To further define the minimal structural elements controlling this species barrier, we prepared a few additional muPA mutants guided by the structural considerations mentioned above. Ultimately, we were able to successfully swap the kinetic rate constants for the interactions between muPA and human or murine uPAR by introducing only two mutations in the GFD, i.e. muPA Y23N/R31W (Table 2), thus demonstrating a clear structure-activity relationship for site 2. Concordantly, it was also possible to graft the reciprocal species selectivity by introducing the equivalent mutations in huPA N22Y/W30R ( Table 2). To explore this relationship further, we created the single site mutants muPA R31W and huPA W30R , but none  A and B, whereas the non-conserved site 2 residues in the ␤-hairpin of GFD are compared in C and D. Selected residues in GFD are shown as sticks, whereas those in uPAR are identified by numbers in italics. A sequence comparison of the receptor-binding ␤-hairpin from various mammalian uPA is shown in E. Non-conserved residues between man and mouse are highlighted by color coding (blue/green), the conserved hot spot is indicated (site 1, gray) as are the major determinants for the species selectivity (site 2, black).

TABLE 1 Kinetics for the human uPA⅐uPAR interaction: impact of mutations in the ␤-hairpin of GFD
Kinetic rate constants (k on and k off ) for the interactions between human uPAR and human pro-uPA carrying single-site mutations in the ␤-hairpin of GFD were determined by surface plasmon resonance. These rate constants were derived from analyzing serial 2-fold dilutions of human uPAR covering the concentration range of 1.5-400 nM and highly purified pro-uPA mutants immobilized at low levels on a CM5 sensor. Measurements were performed at 20°C with flow rates of 50 l/min. The equilibrium binding constants (K D ) were calculated from the mean values of the corresponding rate constants (k off /k on ). a Differences in the Gibbs free energy (⌬⌬G) of uPAR binding between uPA mutant (⌬G mut ) and wild-type uPA (⌬G wt ) were calculated by the following equation: ⌬⌬G ϭ ⌬G mut Ϫ ⌬G wt ϭ RT ln(K D mut/K D wt), where R ϭ 1.99 cal/mol K and T ϭ 293 K. b Indicates that the rate constants for these two mutants are too fast to be measured reliably, and the equilibrium binding constants are determined by fitting data at equilibrium.

Immobilized ligand
of these succeeded in grafting any uPAR-binding selectivity to either species, and they even showed impaired binding to both huPAR and muPAR ( Table 2). As opposed to this, a huPA N22Y mutant actually attains a partial swapping toward muPA-like properties, because it gains an 18-fold decrease in the K D for muPAR accompanied by a 79-fold increase for huPAR (Table 2). Contribution to Species Selectivity by Elements Residing in uPAR-A structural comparison of residues lining the ligandbinding cavity in the mouse and human ATF⅐uPAR complexes revealed an almost complete conservation of the 20 residues forming the functional epitope for uPA binding in huPAR (29). Among these key residues for uPA binding, we replaced four conserved positions in muPAR individually to alanine, i.e. Arg 53 , Tyr 57 , Leu 66 , and Asp 138 . In compliance with a conserved topology of the ligand-binding cavity in uPAR, this translates into a 13-to 50-fold increase in the K D values for muPA (supplemental Table S4), where Asp 138 has the largest contact area in the complex (51.5 Å 2 , supplemental Table S2) and also experiences the largest impact upon alanine replacement. Only one of the functional key residues lining the uPA-binding cavity is altered in muPAR, i.e. Leu 55 in huPAR is replaced by Met 55 in muPAR (Fig. 5, A  and B). Because this position is close to the critical Asn 22 / Tyr 23 of site 2 in the bound GFDs (supplemental Fig. S2), it could obviously be a contributing factor to the species selectivity. Replacing Met 55 in muPAR with the human ortholog by site-directed mutagenesis does, however, not improve the affinity of the mixed huPA⅐muPAR M55L complex (supple-mental Table S4), which indicates that this difference is not relevant for the species barrier.
By scrutinizing the remaining positions in the receptor interfaces, which did not show an impact by the previous alanine scanning mutagenesis of huPAR (29), a few additional candidates were uncovered for this species barrier. In particular, the presence of a negatively charged residue in muPAR (Glu 31 ) proximal to the variable Arg 31 /Trp 30 position in site 2 of the bound GFDs constitutes a promising candidate (Fig. 5, C and D, and supplemental Fig. S2). To explore the functional significance of this difference, we exchanged Glu 31 in muPAR to leucine, the residue found at the equivalent position in huPAR. Although Glu 31 only has a minimal contact surface area with muPA (12 Å 2 ), it does form a hydrogen bond to the essential Arg 31 at site 2 of the bound mGFD (supplemental Table S3 and Fig. S2). Accordingly, its substitution to leucine in muPAR increases the affinity for huPA by Ͼ10-fold, i.e. the K D decreases from 68 nM for muPAR wt to only 5.1 nM for muPAR E31L (Table 3). This gain of function is fully accounted for by an increased stability of the corresponding huPA⅐muPAR E31L complex, because k off is the only rate constant improved by this mutation. A similar gain of function was observed in other muPAR mutants containing the Glu 31 3 Leu in combination with other irrelevant substitutions (data not shown), which substantiates the impact of this position on the species barrier between man and mouse.
Importantly, this Glu 31 3 Leu mutation in muPAR also rescues completely the loss of high affinity binding of muPA R31W

TABLE 2 Swapping kinetics of the uPA⅐uPAR interaction between human and mouse proteins: impact of mutations in GFD
Kinetic rate constants (k on and k off ) for the different species combinations of uPA and uPAR were determined by surface plasmon resonance using a Biacore 3000 as outlined in the legend to Table 1  ( Table 3). Because muPAR E31L maintains an unaltered affinity for muPA wt , these data imply that Glu 31 can only be tolerated at the hydrophobic ligand-binding interface if its carboxylgroup can be hydrogen-bonded to Arg 31 in uPA. This notion is further substantiated by the observation that the partial swapping of the species selectivity by huPA N22Y actually can be fully accomplished by either the additional elimination of Glu 31 in muPAR E31L or by the introduction of Arg 30 as in a huPA N22Y/W30R double mutant (Table 3). These data raise the interesting possibility of a molecular co-evolution of the binding interface between uPA and uPAR. SMB Binding Is Not Compromised by Species Differences-Another well characterized uPAR-ligand interaction is the binding of the provisional matrix protein vitronectin. This interplay facilitates adhesion and cell migration in vitro (51) in a process that is regulated by the endogenous saturation of uPAR by uPA (52,53). The binding of vitronectin to uPAR is governed by the SMB domain, and the epitope for this interaction has recently been defined both functionally (54) and structurally (26). The topology of the interface between huPAR and hSMB is shown in Fig. 6A with key residues contributing to the affinity of this interaction shown as sticks. Comparison of the SMB interface with muPAR and huPAR in complex with ATF reveals an almost complete conservation of all residues except for a 180°rotation of the indol side chain of Trp 32 (Fig. 6B). Combining this structural information with the fact that the SMB domains of vitronectin are conserved during evolution predicts that a similar cross-species barrier is not likely to exist for this interaction. This proposition is verified experimentally by surface plasmon resonance studies using purified components. A K D of 0.80 M is thus determined for the cross-species interaction of hSMB with mATF⅐muPAR complexes (Fig. 6C), which is comparable to the K D of 0.47 M measured previously for the orthologous human components (54). Importantly, this study demonstrates that the molecular basis for the uPAR⅐vitronectin interaction is not compromised per se in xenotransplanted tumor models, but an indirect effect may nevertheless be anticipated in such studies due to different levels of uPAR occupancy by uPA in tumor cells and the tumor-associated stroma as alluded to previously.
Conclusion-Through structural and functional studies, we have gained a detailed molecular understanding of the interaction between uPAR and its bona fide protease ligand uPA, enabling the present rational design of the species selectivity in this interaction. Importantly, this study suggests that uPAR can possibly be targeted across different species if specific low molecular weight inhibitors primarily engaging the binding pocket for site 1 residues can be developed. The pronounced hydrophobicity of the entire binding cavity of uPAR may nonetheless still create some impediments for the development of small specific inhibitors with optimal pharmacological profiles.
Future tumor studies in animal models will undoubtedly also benefit significantly from the structure-function relationships uncovered here for the uPA⅐uPAR interaction. As reviewed previously (17), the vast majority of preclinical intervention studies aimed at pharmacological inhibition of FIGURE 6. Cross-species binding of human SMB to ATF⅐uPAR complexes. The architecture of the hSMB-binding interface with human uPA⅐uPAR as revealed by the x-ray structure (PDB code 3BT1) is illustrated in A with hSMB in a ribbon diagram and huPAR in a surface representation. Functionally important residues in both molecules (54) are shown as sticks. Conservation of the SMB binding interface on uPAR is illustrated in B after superimposition of uPAR DI using the following color coding: gray (hATF⅐huPAR⅐hSMB, 3BT1), magenta (hATF⅐huPAR, 2FD6), and yellow (mATF⅐muPAR, 3LAQ). Although the hot spot residue Arg 91 is conserved between species (Fig. 3) it is nevertheless only defined by electron densities in the crystals of the tri-molecular complex ATF⅐uPAR⅐SMB, where it is stabilized by its ionic interaction with Asp 22 in SMB. C, quantification of the equilibrium binding of hSMB to preformed mATF⅐muPAR complexes by surface plasmon resonance. Covalently immobilized muPAR (231 resonance units (RU) ϳ 7 fmol/mm 2 ) on the sensor chip was initially saturated by injection of 200 nM mATF (yielding a B max of 50 RU ϳ 3.1 fmol/mm 2 ), and the interaction of this mATF⅐muPAR complex with hSMB was subsequently probed by additional injections of a 2-fold dilution of hSMB (10 nM to 10 M). The recorded sensorgrams are shown in the left panel, whereas the derived equilibrium binding isotherms for hSMB are shown in the right panels along with the non-linear curve fitting assuming a simple 1:1 binding. This analysis yields a K D of 0.8 M and a B max of 15 RU ϳ 2.4 fmol/mm 2 for the hSMB interaction with the immobilized mATF⅐muPAR. uPAR in xenotransplanted tumors do not take into account the different targeting efficiency that is encountered by cancer cells and the host-derived tumor-associated stromal cells. This is of paramount importance as most human cancers indeed exhibit a pronounced expression of uPAR by such "tumor-educated stromal cells" (11,55). A general uPAR antagonist engaging the binding pocket normally occupied by site 1 residues would thus circumvent the need for such precautions.
A direct application of this rational design of the species selectivity is the recent generation of a knock-in mouse strain carrying a Plau gene, where site 2 residues have been humanized. 4 These mice are well suited for dissecting the biological significance of the uPA⅐uPAR interaction in vivo, because the only functions that are expected to be compromised in these mice are those related to the surface localization of uPA.