Characterization of the Functional Epitope on the Urokinase Receptor

The high affinity interaction between the serine protease urokinase-type plasminogen activator (uPA) and its glycolipid-anchored receptor (uPAR) represents one of the key regulatory steps in cell surface-associated plasminogen activation. On the basis on our crystal structure solved for uPAR in complex with a peptide antagonist, we recently proposed a model for the corresponding complex with the growth factor-like domain of uPA (Llinas et al. (2005) EMBO J. 24, 1655-1663). In the present study, we provide experimental evidence that consolidates and further develops this model using data from a comprehensive alanine scanning mutagenesis of uPAR combined with low resolution distance constraints defined within the complex using chemical cross-linkers as molecular rulers. The kinetic rate constants for the interaction between pro-uPA and 244 purified uPAR mutants with single-site replacements were determined by surface plasmon resonance. This complete alanine scanning of uPAR highlighted the involvement of 20 surface-exposed side chains in this interaction. Mutations causing ΔΔG ≥ 1 kcal/mol for the uPA interaction are all located within or at the rim of the central cavity uniquely formed by the assembly of all three domains in uPAR, whereas none are found outside this crevice. Identification of specific cross-linking sites in uPAR and pro-uPA enabled us to build a model of the uPAR·uPA complex in which the kringle domain of uPA was positioned by the constraints established by the range of these cross-linkers. The nature of this interaction is predominantly hydrophobic and highly asymmetric, thus emphasizing the importance of the shape and size of the central cavity when designing low molecular mass antagonists of the uPAR/uPA interaction.

The urokinase-type plasminogen activator receptor (uPAR) 2 is a glycosylphosphatidylinositol-anchored membrane glycoprotein (1) that has a primary role in focalizing plasminogen activation at the cell surface through its specific high affinity interaction with the urokinase-type plasminogen activator (uPA). Besides facilitating the generation of plasmin activity in the vicinity of uPAR-expressing cells, which is directly or indirectly involved in remodeling of the extracellular matrix (2,3), the uPAR/pro-uPA interaction also assists in regulating other aspects of cell adhesion and migration. Among these molecular processes is the direct interaction with matrix-deposited vitronectin (4); the modulation of integrin function, in particular ␣ M ␤ 2 , ␣ 5 ␤ 1 , and ␣ 3 ␤ 1 (5)(6)(7)(8); and the activation of the chemotactic FPRL1 receptor (9). As an increased expression level of uPAR is often found in the invasive areas of various human cancers and correlates with poor prognosis (10), the uPAR/uPA interaction and uPA catalytic activity are considered relevant molecular targets for drug development (11)(12)(13). Intervention strategies developed for targeting the uPAR/uPA interaction with a view to cancer therapy include recombinant fusion proteins containing the receptor-binding module of uPA (14,15), anti-uPAR monoclonal antibodies (16), synthetic peptide antagonists (17)(18)(19), down-regulation of uPAR expression by small interfering RNAs (20,21), and a modified anthrax toxin that is specifically activated by the enzymatic activity of receptor-bound uPA (22).
Although the uPAR/uPA interaction thus plays an active role in orchestrating these processes at the biochemical level, an accurate experimental description of the three-dimensional structure of this complex has yet to be solved by x-ray crystallography. During the last decade, a number of studies have reported on the assignment of the uPA-binding site on uPAR using a plethora of different techniques, including chemical * This work was supported by grants from the John and Birthe Meyer Foun-protection analysis (23), single site-directed mutagenesis (24,25), hydrogen-deuterium exchange (26,27), chemical crosslinking (28), overlapping synthetic peptides (29), and entire domain swapping analysis (30,31). However, it is impossible to reconcile these findings in one unifying molecular model of the uPAR⅐uPA complex, demonstrating the need to revisit the molecular basis of this high affinity interaction. Some of the inconsistencies between these previous studies undoubtedly arise because of the inherent coupling between maintenance of the three-domain structure of uPAR and its high affinity uPA binding (32). With the appearance of our experimentally determined three-dimensional structure of uPAR in complex with a synthetic peptide antagonist solved by x-ray crystallography (33), data from single-site mutagenesis of uPAR can now be interpreted with much more confidence. In the present study, we have thus assigned the functional epitope on uPAR for uPA binding as delineated by a comprehensive alanine scanning mutagenesis of intact soluble uPAR combined with kinetic measurements on the uPAR/pro-uPA interaction by surface plasmon resonance. These data have been supplemented by low resolution distance constraints established by various chemical cross-linkers acting as "molecular rulers" after identification of their primary target sites. Combining these experimental observations with the proposal of a convergent binding site displayed on both the growth factor-like domain of uPA and a synthetic peptide antagonist (33) provided the information necessary to enable the first computer modeling of the uPAR⅐ATF complex.
Expression and Purification of Soluble Human Recombinant uPAR Mutants-Soluble forms of human uPAR were expressed in and secreted by Drosophila melanogaster Schneider 2 (S2) cells, which were stably transfected with pMTC/uPAR (residues 1-283) 3 (34). These proteins are secreted to the conditioned medium because of a deletion of the carboxyl-terminal signal sequence that is required for glycolipid anchoring (1,26). Single-site alanine replacements were introduced into pMTC/ uPAR by site-directed mutagenesis using a previously designed three-gene cassette approach (34), and the corresponding 244 soluble uPAR mutants were expressed in S2 cells. All constructs were verified by DNA sequencing using an ABI PRISM 310 genetic analyzer (Applied Biosystems, Foster City, CA). The majority of the secreted uPAR mutants were purified from the conditioned medium by immunoaffinity chromatography using an immobilized anti-uPAR monoclonal antibody (R2), followed by reverse-phase HPLC using a Vydac C4 column (0.46 ϫ 25 cm) and a linear gradient (40 min) from 0 to 70% (v/v) acetonitrile in 0.1% (v/v) trifluoroacetic acid. Immunoaffinity purification of uPAR mutants changed at positions 274 -277 (covering the epitope for antibody R2) was performing using anti-uPAR monoclonal antibody R3, the functional epitope of which resides on uPAR domain I. As judged by SDS-PAGE of ϳ5 g of reduced and alkylated sample, the purity of these uPAR preparations were generally Ͼ95% (supplemental Fig. S1). The identities of individual purified uPAR mutants were also verified at the protein level by peptide mass mapping and tandem mass spectrometric sequencing after in-gel digestion of reduced and carbamidomethylated protein using an endoproteinase (modified porcine trypsin, chymotrypsin, Lys-C, or Asp-N) before mass assignment by an Autoflex II TM MALDI-TOF/TOF mass spectrometer (Bruker Daltonik GmbH, Bremen, Germany) (35). If the mutated residue was located in a glycopeptide, the corresponding in-gel digest was also subjected to deglycosylation by N-glycosidase F. Protein concentrations of the purified uPAR mutants were determined spectrophotometrically using an extinction coefficient (E 280 nM 1% ) of 9.2 (36), except for those mutants involving Trp 32 or Trp 129 , for which an estimated E 280 nM 1% of 6.8 was used. Expression and Purification of Murine uPAR and Murine ATF-Soluble mouse recombinant wild-type uPAR and the uPAR R43K mutant (residues 1-275) 3 were expressed in S2 cells transfected with pMTC/smuPAR and purified along the same lines as described for human uPAR, except for the immunoaffinity chromatography, which was based on an anti-uPAR monoclonal antibody raised in uPAR-deficient mice (antibody KOR-1). The ATF (residues 1-143) 3 of mouse uPA was also expressed in S2 cells transfected with pMTC/mATF and subsequently purified from the medium by ligand affinity chromatography using immobilized purified mutant uPAR, followed by reverse-phase HPLC under same conditions as described for human uPAR.
Expression and Purification of Human Recombinant Pro-uPA, Pro-uPA Mutants, and the Carboxyl-terminally His 6tagged Growth Factor-like Domain (GFD)-Human pro-uPA (residues 1-411) 3 with its active-site Ser 356 replaced with alanine was produced by Drosophila S2 cells stably transfected with the pMTB/uPA S356A vector. Expression of pro-uPA S356A was induced by 0.5 mM Cu 2 SO 4 as described previously (34). The S2 medium also contained 10 g/ml aprotinin to prevent proteolytic cleavage at the activation site of pro-uPA. Pro-uPA was isolated by immunoaffinity chromatography using an immobilized anti-uPA monoclonal antibody (clone 6). The purity of this pro-uPA was Ͼ95%, and the preparation showed only negligible conversion to two-chain uPA as judged by SDS-PAGE (supplemental Fig. S2). The identity of the purified pro-uPA S356A was verified by electrospray ionization mass spectrometry, which revealed two components with a mass of either 47,511.6 or 47,365.8 Da, corresponding to the disulfide-bonded polypeptide chain of pro-uPA S356A with one N-linked glycosylation at Asn 302 (1039.0 Da) and, in ϳ60% of the molecules, one additional O-linked fucose at Thr 18 (146.1 Da). Following the same protocol, 12 different pro-uPA mutants with single-site replacement of lysine with alanine within the amino-terminal region were produced (supplemental Fig. S2). The protein concentration of a stock solution of pro-uPA S356A was determined accurately by amino acid composition analysis (37), and the various lysine mutants were quantified using an E 280 nM 1% of 18.5. The GFD (residues 1-48) of human uPA was expressed as a carboxyl-terminally His 6 -tagged protein in Pichia pastoris strain X-33 using the yeast expression vector pPICZ␣A/GFD-His according to the protocols of Invitrogen. The secreted GFD-His 6 was purified from the medium by adsorption onto an Ni 2ϩ chelate column (HiTrap TM chelating HP), followed by reverse-phase HPLC using a Vydac C4 column (0.46 ϫ 25 cm) and a linear gradient (40 min) from 0 to 35% (v/v) acetonitrile in 0.1% (v/v) trifluoroacetic acid. The identity of the purified GFD was verified by MALDI mass spectrometry (average mass of 6189.6 Da; no fucosylation present at Thr 18 ), and the protein concentration was determined using an estimated E 280 nM 1% of 11.8.
Assessing uPAR Binding to Immobilized Pro-uPA by Surface Plasmon Resonance-All real-time interaction studies were carried out on a Biacore 3000 TM (Biacore International AB, Uppsala, Sweden) in running buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, and 0.005% (v/v) surfactant P20, pH 7.4). Human pro-uPA S356A (0.2-0.5 g/ml) was immobilized covalently at pH 5.0 on a carboxymethylated dextran matrix (CM5 sensor chip) using NHS/EDC as described previously (24), which yielded coupling levels in three different flow cells covering the range of 200 -1000 resonance units, corresponding to 0.2-1 ng/mm 2 pro-uPA. For comparative evaluation of the kinetics of the uPAR/pro-uPA interaction, serial 2-fold dilutions of the various uPAR mutants (0.8 -100 nM in running buffer) were analyzed in parallel for uPA binding at a flow rate of 20 l/min at 20°C. Association was recorded for 300 s, followed by a dissociation phase of 775 s, and the derived sensorgrams for such an interaction analysis using the wild-type receptor are shown in supplemental Fig. S3. After each run, the sensor chip was regenerated by two consecutive injections of 10 l of 0.1 M acetic acid and 0.5 M NaCl. Data processing was accomplished by double referencing, during which each sensorgram was corrected by subtraction of the signal from both a mock-coupled flow cell and an appropriate buffer run. The kinetic rate constants k on and k off were derived by nonlinear fitting of the association and dissociation phases to a simple bimolecular interaction model (uPA ϩ uPAR 7 uPAR⅐uPA) using BIAevaluation Version 4.1 software (Biacore International AB). In brief, k off values were initially derived by fitting the collected data to dR/dt ϭ Ϫk off R, and the corresponding k on was subsequently fitted to dR/dt ϭ k on [uPAR](R max Ϫ R) Ϫ k off R, assuming pseudo first-order reaction kinetics essentially as described (38). The curve fitting to the recorded data for the wild-type uPAR (uPAR wt )/uPA interaction is shown in supplemental Fig. S3 along with the corresponding residual plot. The change in free energy (⌬⌬G) caused by each mutation was cal-culated from ⌬⌬G ϭ ⌬G mut Ϫ ⌬G wt ϭ RT ϫ ln(K d(mut) /K d(wt) ), where K d values were derived from the ratio of the kinetic constants k off and k on , R ϭ 1.99 cal/mol K, and T ϭ 293 K.
Because of the limitations enforced by the logistics of this experiment, it was impossible to analyze freshly prepared protein samples for each of the large number of mutants. The entire library of uPAR alanine mutants was therefore kept frozen at Ϫ80°C, and individual mutants were thawed and diluted in running buffer immediately before analysis. At appropriate time intervals, uPAR wt was analyzed in parallel and served as a reference for calculation of ⌬⌬G. A number of uPAR mutants, including some of those that exhibited severely impaired kinetics for the pro-uPA interaction, were subjected to size exclusion chromatography (Superdex 200 TM in 10 mM phosphate and 0.1 M NaCl, pH 7.4) to separate uPAR monomer from any aggregated protein before reanalysis of the monomer fraction by surface plasmon resonance without prior freezing. Generally, the k off values did not changed significantly in this second analysis, but the k on values occasionally increased by up to 2-fold after size exclusion chromatography.
Covalent Conjugation of uPAR⅐Ligand Complexes by Bifunctional Chemical Cross-linkers-Covalent chemical conjugation via ⑀-amines of optimally positioned lysine residues in preformed uPAR⅐pro-uPA or uPAR⅐ATF complexes was introduced by various homobifunctional NHS esters with variable spacer arm lengths (DSS, ϳ11.4 Å; DSG, ϳ7.7 Å; and DST, ϳ6.4 Å). Concentrated stock solutions of these chemical crosslinkers were prepared in dry Me 2 SO and were diluted in Me 2 SO to the working solutions (5-20 mM) immediately prior to use. Noncovalent complexes between 4 nM 125 I-labeled ATF and either 20 nM uPAR wt or single-site lysine mutant uPAR were allowed to preform during a 60-min incubation at 4°C in crosslinking buffer (0.1 M Tris, pH 8.1, containing 0.2% (w/v) CHAPS). Covalent cross-linking of the preformed complexes was accomplished by a 10 -15 min incubation with the respective cross-linker (0.1-1 mM), after which the reaction was quenched by addition of 100 mM CH 3 COONH 4 . In other experiments, 4 nM 125 I-labeled uPAR was subjected to chemical cross-linking to either pro-uPA S356A or one of the 12 different single-site lysine mutants of pro-uPA. Cross-linking efficiency was visualized by SDS-PAGE of the reduced and alkylated samples, followed by autoradiography.
Zero-length Cross-linking of uPAR⅐GFD Complexes Using Carbodiimide-Monomeric uPAR (10 M; prepared by size exclusion chromatography as describe above) was mixed with 90 M GFD for 1 h at 4°C to allow the formation of noncovalent uPAR⅐GFD complexes. Zero-length cross-linking between optimally positioned amino and carboxylate groups was achieved by additional incubation overnight at 20°C with 5 mM EDC added from a freshly prepared stock solution in H 2 O. Cross-linking efficiency was assessed by SDS-PAGE of reduced and carbamidomethylated samples. The relevant Coomassie Brilliant Blue-stained bands were excised from the dried gel and subjected to in-gel trypsin digestion in parallel using either 16 Oor 18 O-enriched H 2 O as solvent. This solvent-induced "isotope coding" of tryptic peptides, which causes a specific mass shift of 8 Da for cross-linked peptides with two carboxyl termini (39), facilitated the identification of these particular peptides by subsequent mass spectrometric analyses by nanoflow reversephase liquid chromatography connected to a quadrupole timeof-flight mass spectrometer (Q-Tof Ultima, Micromass, Manchester, UK). The further identification of such crosslinked peptides was enabled by the high mass accuracy and precision obtained using a 7-Tesla linear quadrupole ion trap-Fourier transform ion cyclotron resonance mass spectrometer (LTQ FT TM , Thermo Electron Corp., Bremen) with a modified nanoelectrospray ion source (Proxeon Biosystems, Odense, Denmark).
Molecular Modeling of uPAR Complexes-Molecular modeling of uPAR⅐GFD complexes based on a convergence in the receptor-binding motif between the ␤-hairpin in the GFD of uPA and the ␣-helix of a linear peptide antagonist was performed essentially as described (33). In brief, a high temperature molecular dynamics procedure was applied in which ambiguous constraints were introduced between the side chains of (i) Tyr 24 in GFD and Val 125 , Leu 168 , His 251 , Leu 252 , Ala 255 , and His 166 in uPAR; (ii) Phe 25 in GFD and Leu 55 , Leu 123 , Val 125 , Leu 150 , Leu 168 , and His 251 in uPAR; (iii) Ile 28 in GFD and Arg 53 , Leu 55 , Leu 66 , and Leu 150 in uPAR; and (iv) Trp 30 in GFD and Val 29 , Leu 31 , Leu 55 , Tyr 57 , and Leu 66 in uPAR. 600 structures of complexes were generated, and the 300 with the best van der Waals interaction energies were selected to perform a theoretical alanine scanning for residues 21-30 in GFD. For all these 300 structures, the calculated ⌬⌬G values were compared with those derived from experimental mutagenesis studies as published previously (33). To refine these structures, an additional ambiguous constraint between Asp 140 and any atom of GFD was introduced. 30 structures were refined, leading to four final structures with a reasonable agreement between the experimental and calculated ⌬⌬G values. For these four structures, the average ⌬⌬G values are 0. To further validate this model, individual ⌬⌬G values were calculated for selected single-site mutants in uPAR and compared with those determined experimentally. Generally, these values are mutually concordant and are, as expected, dominated by positions that have relatively small impacts on the uPA interaction (⌬⌬G Ͻ 1 kcal/mol). This pairwise comparison revealed, for example, only six positions in uPAR domain I (of 62 tested) where the deviations between the calculated and determined ⌬⌬G values actually exceeded the estimated error of 0.5 kcal/mol. Furthermore, these six positions represent charged residues for which the effect of the solvent is improperly weighted by the use of an implicit solvent.
Binding free energies were calculated using the Poisson-Boltzmann equation (40). All calculations were performed using CHARMM Version c28a3 (41) with the force field par_all22_ prot_lipid.inp and top_all22_prot_lipid.inp (42). Finally, based on these structures, a model of the uPAR⅐ATF complex was built in which GFD was superimposed onto that of the corresponding uPAR⅐GFD model. Distance constraint obtained by chemical cross-linking between Lys 98 from the kringle domain of uPA and Lys 43 from uPAR was introduced in MODELLER Version 6.0 to position the kringle domain in the complex. The coordinates for the NMR-derived solution structure for human ATF (residues 6 -135) and the crystal structure for human uPAR (residues 1-283) were used for these modeling studies (Protein Data Bank accession codes 1URK (43) and 1YWH (33), respectively).

RESULTS
Several laboratories have previously shown that a secreted form of human uPAR can be produced by various eukaryotic expression systems, e.g. Chinese hamster ovary cells (24,44) and D. melanogaster S2 cells (18,25,34), after deletion of the carboxyl-terminal signal sequence, which is instrumental for the in vivo tethering of uPAR to the cell surface by a glycosylphosphatidylinositol membrane anchor (1). This soluble recombinant uPAR (residues 1-283) 3 represents a convenient surrogate for the glycosylphosphatidylinositol-anchored uPAR, as in a range of biochemical assays, it retains many of the functional properties inherent to the native protein, including its high affinity for uPA (24,32). In this study, we exploited Drosophila S2 cells as a host for the stable expression of a large number of uPAR mutants with defined single-site mutations covering almost 90% of the fully processed uPAR sequence (residues 1-283). To avoid compromising residues in uPAR that are of structural importance, we focused our mutagenesis efforts on residues that are not included in the consensus sequence defining the Ly-6/uPAR/␣-neurotoxin protein domain family (26,33,45), i.e. all cysteines and the conserved Asn residue that immediately follows the last cysteine of each domain were omitted. The resultant 244 uPAR mutants were expressed well by the S2 cells, and the purities and yields of the mutant proteins were comparable with those of uPAR wt (supplemental Fig. S1).
Impact of Single-site Mutagenesis in uPAR on Pro-uPA Binding Kinetics-To address the functional importance of specific amino acid side chains in human uPAR, we produced a comprehensive uPAR mutant library in which defined side chains of the soluble receptor were individually deleted beyond the C ␤ atom by alanine substitution. Native alanine residues in uPAR were changed to serine, whereas N-linked glycosylation sites were replaced with glutamine. The impact of these specific side chain deletions on the uPA interaction were assessed by measuring the kinetic rate constants for the interaction between immobilized pro-uPA S356A and the respective purified uPAR single-site mutants by surface plasmon resonance. A complete inventory of these experimentally determined rate constants is provided in supplemental Table SI. During the course of this study, we observed that repeat determinations of the dissociation rate constant for the interaction between uPAR and pro-uPA S356A were very reproducible over time and insensitive to cycles of freeze-thawing of the purified protein preparation. In contrast, the association rate constant could vary by as much as 2-fold under the same conditions. Because of the logistics of the present experimental setup comparing a very large number of purified uPAR mutants, it was, however, impossible to ensure exact identical handling of all samples. In the following data evaluation, we therefore initially focused on k off values, which seemed reasonable, as this rate constant furthermore yielded the largest contribution to the global change in the Gibbs free energy of binding (⌬⌬G). The impact of specific side chain deletion primarily on k off rather than on k on is concordant with reports for other protein/protein interactions probed by alanine scanning mutagenesis (46,47). The quality of the data evaluation for these interactions is illustrated in supplemental Fig. S3, showing the actual fits to the data of sensorgrams obtained for the interaction between uPAR wt and immobilized pro-uPA S356A . The dissociation time recorded for this particular interaction was considered adequate to identify uPAR mutants with impaired k off values caused by specific alanine replacements.
The k off values determined for the 244 different uPAR⅐pro-uPA complexes are illustrated by the histograms shown in Fig. 1 (uPAR domain I), Fig. 2 (uPAR domain II), and Fig. 3 (uPAR domain III). Each histogram shows threshold levels for the effects of the mutations, arbitrarily set to 2.5, 5, and 10 times the dissociation rate of uPAR wt . As evident from Fig. 1, nine residues in uPAR domain I displayed a Ն2.5-fold increase in the dissociation rate constant upon alanine substitution (i.e. Arg 25 , Thr 27 , Leu 40 , Lys 50 , Thr 51 , Arg 53 , Leu 55 , Tyr 57 , and Leu 66 ), and two residues exhibited a Ն5-fold increase in k off (Leu 55 and Leu 66 ). According to the crystal structure solved for uPAR in complex with a uPA antagonist peptide (33), all nine residues exhibit a surface accessibility of Ͼ15 Å 2 in the unoccupied receptor after removal of the antagonist. The involvement of some of these residues in uPA binding (Thr 51 , Arg 53 , Leu 55 , Tyr 57 , and Leu 66 ) has been identified previously by alanine scanning mutagenesis and chemical protection analysis (23,24).
A similar scanning of uPAR domain II identified 21 different residues with a Ն2.5-fold increase in k off upon alanine replacement ( Fig. 2 and supplemental Table SI). However, 10 of these residues (Ser 97 , Ser 100 , Leu 113 , Asp 124 , Leu 144 , Arg 145 , Gly 146 , Gly 148 , Phe 165 , and His 166 ) do not pass the minimum threshold of 15 Å 2 , which we have set for an acceptable surface accessibility. Of the remaining 11 residues in uPAR domain II that, according to these criteria, are important for uPA binding (Asp 102 , Ser 104 , Glu 106 , Val 125 , Thr 127 , Asp 140 , Asp 141 , His 143 , Leu 150 , Pro 151 , and Leu 168 ), seven displayed a Ն5-fold increase in k off (shown in boldface), and two of these displayed as much as 23-fold (Asp 102 ) and 36-fold (Leu 150 ) increases in k off . This analysis clearly emphasizes the important role of uPAR domain II in uPA binding.
In contrast to uPAR domains I and II, very few residues in uPAR domain III were found to contribute to the stability of the uPAR⅐pro-uPA complexes (i.e. Met 219 , Gly 227 , and Phe 256 ). In addition, these effects were at the lower end of the arbitrary threshold (ϳ2.5-fold increase in k off ) (Fig. 3) and were most likely due to local perturbations of uPAR domain III because none of these residues occupies a surface-exposed position in the structure (Ͻ15 Å 2 of surface accessibility). A compilation of these data highlighting residues in uPAR that we have identified as being important for uPA binding by single-site alanine scanning mutagenesis and surface plasmon resonance is shown in Table 1. This functional epitope for uPA binding is also highlighted in the surface representation of the uPAR structure shown in  Positioning the Kringle Domain in the uPAR⅐Pro-uPA Complex by Chemical Cross-linking-As chemical cross-linking has developed into a valuable tool for low resolution structural modeling (48,49), we used this technique to establish distance constraints within the uPAR⅐pro-uPA complex.
Covalent conjugation of human uPAR to radiolabeled pro-uPA by the amine-specific NHS-based cross-linker DSS has been used as a sensitive tool for the detection of uPAR in various biological fluids and detergent lysates from neoplastic cell lines (1,28,50,51). However, this cross-linking reaction is not observed when using murine proteins, despite an equivalently high affinity uPAR/uPA interaction (18, 50),  suggesting the absence of optimally positioned lysine side chains in the noncovalent complex.
The observation that mutation of any of the 10 lysine residues in human uPAR did not have a pronounced impact on the affinity for pro-uPA (supplemental Table SI) enabled us to identify the lysine residues involved in the cross-linking reaction. In these experiments, complexes were formed using saturating amounts of the unlabeled mutant component (in this case, uPAR). The noncovalent uPAR⅐ATF complexes were conjugated by a short exposure to various NHS-based cross-linkers with different spacer arm lengths (DSS, ϳ11.4 Å; DSG, ϳ7.7 Å; and DST, ϳ6.4 Å). The short exposure to the active NHS ester ensured that only optimally positioned lysines were crosslinked. The autoradiograph in Fig. 5 shows that all mutants except uPAR K43A formed covalent complexes with 125 I-labeled ATF under these conditions with DSG as cross-linker. Similar results were also obtained with DSS and DST (data not shown), illustrating that the ⑀-amino group of Lys 43 in uPAR domain I can be positioned within 6.4 -11.4 Å of its target amine in ATF. The difference in susceptibility among the various uPAR lysine mutants for chemical conjugation to ATF cannot be ascribed to different levels of noncovalent complexes, as the mutants have comparable K d values for pro-uPA (supplemental Table SI), and preincubation using concentrations well above the K d ensured that equal levels of noncovalent complexes were formed.
Identification of Lys 43 as the specific cross-linking site in human uPAR provides a rational structural basis for the observed cross-species barrier in the cross-linking potential of amine-reactive esters because a non-reactive arginine residue occupies the equivalent position in murine uPAR. To confirm this, we introduced a lysine at this position in murine uPAR and created a gain-of-function mutant, as the resultant murine uPAR R43K now efficiently formed a covalent complex with murine ATF using DSS (Fig. 6, eighth lane versus seventh lane). The affinity of murine uPAR wt and uPAR R43K for murine pro-uPA was comparable as judged by surface plasmon resonance studies (data not shown). Albeit very faint, the band in the autoradiograph in Fig. 6 (fourth lane) clearly shows that murine uPAR R43K also gained some cross-linking potential for human ATF. It is possible that the chemical conjugation efficacy between murine uPAR R43K and human ATF even resembles the one obtained in the pure human system, as saturation is by far reached under these conditions because of a Ͼ300-fold reduction in affinity for the noncovalent complex formation in the mixed-species experiment (18). This suggests that the corresponding target site in ATF probably is conserved between these two species and that the species specificity in cross-linking is entirely governed by the lysine-to-arginine substitution at position 43 in the receptor.
To identify the corresponding target site(s) in pro-uPA for these amine-reactive cross-linkers, we expressed and purified 12 pro-uPA single-site mutants, targeting all lysine positions in the non-catalytic modular part of uPA corresponding to ATF (supplemental Fig. S2). All these pro-uPA mutants displayed unaltered binding kinetics for immobilized uPAR wt as assessed by surface plasmon resonance, with the exception of pro-uPA K23A , which exhibited a moderately reduced affinity (52) (data not shown). Cross-linking experiments were performed as described above using excess unlabeled mutants (in this case, pro-uPA). As shown in Fig. 7, pro-uPA K98A clearly encountered the more severe impact on its cross-linking competence as evidenced by both reduced complex formation and accumulation of unreacted 125 I-labeled uPAR. As previously inferred from the reverse cross-linking experiments described above, the target site in pro-uPA (Lys 98 ) is indeed conserved between man and mouse.
Cross-linking uPAR⅐GFD Complexes by EDC-To provide additional and independent intermolecular distance constraints within the uPAR⅐GFD complex, we also performed zero-length conjugation of preformed complexes between uPAR wt and GFD-His using EDC. The traditional stabilization with NHS was omitted in this case to avoid stable "dead-end" modifications of carboxylate groups in either of the proteins. As evident from the experiment shown in Fig. 8 (lanes 1), monomeric uPAR wt underwent extensive oligomerization during EDC-induced cross-linking. Occupancy of uPAR by GFD-His completely prevented this receptor oligomerization, which was replaced with a specific conjugated uPAR⅐GFD complex (Fig. 8, lanes 2). These complexes contained one molecule of each protein as revealed by MALDI mass spectrometry (data not shown). Limited chymotrypsin treatment of the cross-linked complexes, leading to cleavage between uPAR domains I and II after Tyr 87 (28,44), clearly revealed that uPAR domain I was by far the major partner for the intermolecular cross-linking (Fig. 8, lanes 3). Specific cross-linking sites were subsequently identified by mass spectrometry of the paired samples after parallel in-gel hydrolysis by trypsin using solvents enriched in 16 O-and 18 O-labeled H 2 O. As evident from the masses deter-

constants for selected uPAR mutants with an impaired interaction with immobilized pro-uPA S356A as determined by surface plasmon resonance
The interactions between immobilized pro-uPA (200 -1000 resonance units) and the respective purified uPAR mutants were measured at 20°C for serial 2-fold dilutions ranging from 1 to 100 nM uPAR and analyzed in three flow cells with different levels of immobilized ligand. Shown are the means for the rate constants k on and k off derived from these data by nonlinear least-squares curve fitting using BIA evaluation Version 4.1 software, but only residues experiencing a Ն2.5-fold increase in k off and having a surface accessibility of Ͼ15 Å 2 are included. The K d was calculated from the means of the corresponding rate constants (K d ϭ k off /k on ). The change in the Gibbs free energy was calculated as ⌬⌬G ϭ RT ϫ ln(K d(mut) /K d(wt) ). For comparison, the accessibility derived from the crystal structure of uPAR is shown (33). mined for tryptic uPAR peptides 31-43 and 117-139 (containing several glutamic and aspartic acids), the present EDC cross-linking procedure did not create unproductive dead-end modifications of the carboxylates (Table 2). So far, we have assigned a single intermolecular cross-link between uPAR and GFD. This was introduced between the ␣-amino group of Ser 1 in GFD and the carbox- The structure of human uPAR determined by x-ray crystallography in complex with a peptide antagonist is shown as a surface representation after removal of the peptide (33). All residues subjected to alanine scanning are shown in wheat, whereas the residues defining the consensus sequence for the Ly-6/uPAR modules are shown in white. To highlight the functional epitope on uPAR for uPA binding, the residues selected as being important for this interaction (Table 1) are shown in red (⌬⌬G Ͼ 1 kcal/mol) and blue (⌬⌬G ϭ 0.5-1 kcal/mol). The carbohydrate moieties defined by the crystal packing in the x-ray structure are shown in green. A shows the front of uPAR with the distinct ligand-binding cavity; B shows this cavity at a higher magnification; and C shows the back of uPAR. The glycolipid attachment site is indicated (glycosylphosphatidylinositol (GPI)). These images were generated with PyMOL (DeLano Scientific) using Protein Data Bank code 1YWH for uPAR molecule E in the unit cell for the octamer (33). ) and 20 nM uPAR (residues 1-283) were exposed to 400 M DSG (range of 7.7 Å) for 10 min at room temperature before the reaction was quenched. After reduction and alkylation, these samples were subjected to SDS-PAGE, followed by autoradiography. Mapping the uPAR/uPA Interface ylate of Asp 11 in uPAR domain I, as we identified this particular conjugate between uPAR-(8 -13) and GFD-(1-23) ( Table 2) in both our cross-linked samples containing either intact uPAR or uPAR domain I (Fig. 8). However, additional unidentified intermolecular cross-links must be present in the uPAR⅐GFD complexes, as we also found GFD-(1-23) as the free amino-terminal tryptic fragment in our purified conjugates (Table 2).

DISCUSSION
To delineate the functional importance of specific amino acids in human uPAR on its high affinity interaction with the cognate protease ligand uPA, we have, in this study, produced and characterized a comprehensive uPAR mutant library. As predicted from homology considerations (26,45) and later corroborated by the three-dimensional x-ray structure (33), the extracellular uPAR is organized as a modular receptor composed of three homologous domains belonging to the Ly-6/ uPAR/␣-neurotoxin protein domain fold. Biochemical studies using limited proteolysis to obtain specific cleavages in either of the two interdomain linker peptides in uPAR revealed that maintenance of the three-domain structure of uPAR constitutes an indispensable condition for expression of high affinity uPA binding (32,53). In keeping with this observation, the present alanine scanning mutagenesis clearly showed that the high affinity binding of pro-uPA critically depends on structural elements residing in both uPAR domain I (nine residues) and in uPAR domain II (21 residues), thus providing a structural basis for the composite nature of the uPA-binding site in uPAR. Although the alanine scanning mutagenesis did not reveal any important role for uPAR domain III in uPA binding, this domain nevertheless appears to play an auxiliary role in this high affinity interaction. This supposition is based on the observation that neither uPAR domain I (32) nor uPAR domains I and II (53) can form very stable complexes with GFD as opposed to either intact uPAR or a mixture containing uPAR domain I and uPAR domains II and III. This phenomenon most likely reflects a stabilizing role of uPAR domain III in the assembly of a functional ligand-binding cavity in the three-domain uPAR.
A significant body of biochemical data exists in the literature to highlight the role of uPAR domain I in uPA binding (23,24,28,32,44). This relationship is further substantiated by this study, which identified new residues involved in uPA binding (Arg 25 , Thr 27 , and Leu 40 ) and which validated the importance of previously assigned residues (Thr 51 , Arg 53 , Leu 55 , Tyr 57 , and Leu 66 ). Interestingly, the side chains of these residues line a significant fraction of the "right" wall of the central ligand-binding cavity of uPAR (Fig. 4) as defined by the crystal structure of uPAR in complex with a peptide antagonist (33).
The present alanine-scanning mutagenesis also unexpectedly uncovered that uPA binding is particularly sensitive to mutagenesis in uPAR domain II (Fig. 2), where 11 surface-exposed residues were considered to be important. The involvement of this domain in uPA binding was previously inferred primarily from indirect evidence. Proteolytic cleavage of the linker region between uPAR domains I and II thus led to a Ͼ1500-fold reduction in affinity for uPA equivalent to ⌬⌬G Ն 4 kcal/mol (32). In a mutagenesis study, Bdeir et al. (25) focused on uPAR domain II, but only interrogated five residues in this domain and reported that the double mutation K139A/H143A caused a 5-fold reduction in k off . This finding is concordant with our data that uPAR H143A displayed a similar decrease in k off . Bdeir et al. also reported that the triple mutation R137A/ R142A/R145A had a severe impact on uPA binding, which was predominantly enforced by a reduction in k on . With a view to our present data, this effect is most likely caused by structural perturbation because, of the corresponding single mutants, only R145A had a significant effect on uPA binding ( Fig. 2 and supplemental Table SI) and because Arg 145 has a surface accessibility of only 0.5 Å 2 in our x-ray structure (33).  The latter was subjected to limited proteolysis by 12 nM chymotrypsin to generate uPAR domain I (DI ) and uPAR domains II and III (DIIϩIII) (lanes 3). After reduction and carbamidomethylation, these samples were analyzed by SDS-PAGE, and the cross-linking pattern was visualized by Coomassie Brilliant Blue (BB) staining (A) or by Western blotting using either a monoclonal anti-His antibody (B) or anti-uPAR monoclonal antibody S2 (which recognizes uPAR domain I) (C ) or S1 (which recognizes uPAR domains II and III) (D). Arrows indicate the uPAR domains and their corresponding GFD conjugates, which are also highlighted in lanes 3 by asterisks.
The importance of uPAR domain II in coordinating the assembly of a functional ligand-binding cavity in the intact uPAR is also highlighted by the observation that several buried residues in this domain have a severe impact on uPA binding when mutated individually (Fig. 2). One ␤-strand in particular exhibits a very low tolerance to mutagenesis, i.e. ␤IID 3 in uPAR domain II (residues 142-149). We have previously proposed that this bent ␤-strand plays a key role in the internal dynamics of this modular receptor by acting as a hinge centered on the RGC sequence and that this hinge enables movement between domains I and II; the angle between these domains varies from 107.8°to 112.2°, as observed in our crystal structure (33). Impairment of this interdomain flexibility through structural perturbations imposed by mutagenesis in ␤IID is therefore likely to affect the architecture of the central cavity in uPAR, with resultant effects on uPA binding to the composite functional epitope comprising residues from both uPAR domains I and II.
Examination of the ⌬⌬G values presented in Table 1 reveals that no single side chain has the predominant contribution to the free energy of uPA binding, i.e. typical "hot spots" (54) are absent from the receptor interface with uPA. This property may be related in part to the architecture of the uPA-binding site, where the engagement of the deep central cavity is governed primarily by a flexible surface comprising hydrophobic and aliphatic residues. Alanine scanning mutagenesis of uPAR will therefore not significantly affect the hydrophobic nature of this ligand interface and will be accompanied by compensatory effects that may attenuate the impact observed on the ⌬⌬G values. Nevertheless, the present alanine scanning mutagenesis clearly demonstrated the important role of several hydrophobic amino acid residues from uPAR domains I and II, which line the central cavity (Fig. 4). As illustrated in Fig. 4B, two tracks of residues involved in uPA binding seem to irradiate from Leu 150 at the floor of the cavity to either Asp 102 or Asp 140 at the rim of this gorge. Notably, no residue causing a shift in ⌬⌬G Ͼ 1 kcal/ mol is found exposed on the opposite side of uPAR, and the only residue in this region causing an intermediate effect (T51A) is localized close to the entrance of the binding cavity (Fig. 4C).
On the basis of these biochemical data, we have now built a model for human uPAR in complex with the ATF of uPA in which the kringle domain is positioned by the low resolution distance constraint established by our cross-linking experiments (Fig. 9). This model merges the information derived from the present experiments with our previous proposal of a convergent receptor-binding site displayed by the ␤-hairpin in the GFD of uPA and a linear peptide antagonist developed by combinatorial chemistry (18,33). The long ␤-hairpin of GFD engages the deep central cavity in uPAR, and the resultant interface buries 1629 Å 2 of accessible surface area of uPA and 1318 Å 2 of uPAR in the complex. Remarkably, in this model, the complementary receptor/ligand interface of the uPAR⅐GFD complex is highly asymmetrical. In uPAR, 27 residues participate in the formation of this interface, but none of the individual contributions actually exceed 6% of the total buried accessible surface area. As opposed to this, only four residues in GFD provide ϳ40% of the buried surface (Lys 23 , Tyr 24 , Phe 25 , and Trp 30 ). This distinction in geometry may have a bearing on the determined ⌬⌬G values after mutagenesis, where no real hot spots are found among the many residues forming the large hydrophobic cavity in uPAR.
The intimate engagement of the ␤-hairpin of GFD in the interaction with the central cavity of uPAR provides a plausible binding mechanism that amalgamates the majority of the biochemical data reported in the literature for this interaction into one unifying model. First, the location of the GFD-binding site at a composite interface involving several residues in both uPAR domains I and II lining the central cavity explains why the maintenance of the three-domain structure of uPAR is required to yield high affinity uPA binding (32,53). This architecture also provides an explanation for the observation that the isolated GFD module can induce the specific assembly of a stable trimolecular complex comprising uPAR domain I, GFD, and uPAR domains II and III, whereas all the corresponding bimolecular complexes are very short-lived. A similar property has been reported for ATF (55). Second, the model buries the amino acid side chains in both uPAR and uPA that become protected against chemical modification upon complex formation at the interface of the complex, i.e. Tyr 57 in uPAR and Lys 23 and Tyr 24 in uPA (23,26). Third, residues highlighted as being important for this interaction by single-site mutagenesis are also confined to the ␤-hairpin of GFD (52) or, in the case of a Accurate peptide masses were determined by nanoflow liquid chromatography/electrospray ionization Fourier transform ion cyclotron resonance mass spectrometry. b Mass differences observed between peptides derived from in-gel trypsin cleavage in 18 O-and 16 O-enriched solvents were determined by nanoflow liquid chromatography/ electrospray ionization quadrupole time-of-flight mass spectrometry. A ⌬m of 8 Da indicates a peptide conjugate with two carboxyl termini, and a ⌬m of 4 Da indicates a peptide with a single carboxyl terminus; when only 2 Da is observed, it stems from the fact that this particular peptide is a poor pseudo-substrate for trypsin because of the presence of a glutamic acid in the P2 position. c Masses for cross-linked peptides were calculated using GPMAW Version 7.0. The actual cross-linking site between uPAR-(8 -13) and GFD-(1-23), i.e. Asp 11 and Ser 1 , was unambiguously defined by the respective sequences, as only one pair of residues satisfied the reactivity of the employed cross-linking chemistry. d Numbers indicate the samples from the gel in Fig. 8 that contained the following specific peptides: 1) uPAR⅐GFD-His; 2) uPAR domain I⅐GFD-His; 3) uPAR; and 4) uPAR domain I. e The tryptic peptides representing GFD-(1-23) became deamidated during the in-gel procedures before addition of trypsin irrespective of its cross-linking status. Accordingly, the ⌬m caused by the differential 18  uPAR, are located within or at the rim of the central ligandbinding cavity (Fig. 4) (24). Two exceptions to this are Asp 102 and Asp 140 , which are located somewhat distantly from the docked GFD module at the rim of the central cavity in uPAR (Fig. 9). One likely explanation for this apparent inconsistency relates to the mobility of these residues because both exhibit B-factors above the average for uPAR, and Asp 102 in particular is located in a flexible region of loop 1 in uPAR domain II that is relatively less well defined in the structure (33). The kringle domain of uPA is positioned on the outside of uPAR domain I in this model, causing ATF to virtually embrace this domain in a "loose-fitting sandwich" (Fig. 9). This conformation places the ⑀-amino groups of uPAR Lys 43 and uPA Lys 98 within the deduced cross-linking distance. However, it should be emphasized that the entire range of spacer arm lengths for the homobifunctional cross-linkers (6.4 -11.4 Å) used in this study leads to a specific productive conjugation of uPAR to pro-uPA, which indicates certain flexibility in the position of the kringle domain of uPA in relation to uPAR. Such a flexible and transient nature of the interplay between the kringle domain and uPAR is compatible with the observation that GFD and the kringle domain behave as completely independent modules in the NMR-derived structure of ATF (43) and with our previous observation that GFD and ATF exhibit comparable kinetic rate constants for the interaction with uPAR (56). The proximity of the flexible amino terminus of GFD to Asp 11 in uPAR domain I established by zero-length cross-linking is also compatible with the proposed model of the uPAR⅐ATF complex. Finally, our model of the uPAR⅐ATF complex provides insight into the molecular basis for the competitive inhi-bition exerted by anti-uPAR monoclonal antibody R3 on the uPAR/uPA interaction. Although the functional epitope for antibody R3 (i.e. Glu 33 , Leu 61 , and Lys 62 ) (24,26) does not coincide with that determined here for uPA, it is clear from the model that the corresponding structural epitopes must overlap to such an extent that steric hindrance would preclude concomitant binding of ATF and antibody R3 (Fig. 9).
During the revision of this manuscript, the crystal structure of human uPAR in complex with ATF and a Fab fragment of an anti-uPAR monoclonal antibody was published (57). This structure nicely confirms the overall positioning of uPAR and ATF in our model, where the ␤-hairpin of GFD engages the central ligand-binding cavity of uPAR and where the kringle domain is located close to uPAR domain I. Notably, the distance in the crystal structure between Lys 43 in uPAR and Lys 98 in the kringle domain of ATF (10.8 Å) is indeed within the range defined by the chemical cross-linkers employed in this study. However, one important distinction between the crystal structure of uPAR in complex with ATF (57) compared with the corresponding complex with a peptide antagonist (33) relates to the relative orientation between uPAR domains I and II. The angle between these two receptor domains is thus reduced by ϳ20°in the uPAR⅐ATF complex because of a conformational shift centered on the previously mentioned "hinge" region between the corresponding interdomain ␤-strands, i.e. ␤IE and ␤IID (33,57). This rearrangement causes Asp 140 in uPAR domain II to move closer to GFD, where it is engaged in a hydrogen bond with Ser 21 (57), thus providing a possible structural basis for the change in ⌬⌬G for uPA binding, which we observed with the D140A mutation. . Model of a uPAR⅐ATF complex. A model of the uPAR⅐ATF complex based on experimental data from alanine scanning and chemical cross-linking is shown in A. uPAR is illustrated as a surface representation in wheat, and ATF is shown as a ribbon diagram with green ␤-sheets and red ␣-helixes. Residues in uPAR identified as important for uPA binding by the present alanine scanning mutagenesis are shown in red (⌬⌬G Ͼ 1 kcal/mol) and blue (⌬⌬G ϭ 0.5-1 kcal/mol). The position of the cross-linking site in the kringle domain of uPA is indicated, whereas the corresponding target site in uPAR (Lys 43 ) is not visible from this view. The functional epitope for monoclonal anti-uPAR antibody R3 (Glu 33 , Leu 61 , and Lys 62 ), which inhibits uPA binding, is highlighted by a yellow surface representation (24,26). The proposed engagement of the ␤-hairpin of the GFD module of uPA in binding to the central cavity of uPAR is shown at a higher magnification in B. This illustration was generated using PyMOL.
Combined with the recently published crystal structure of the uPAR⅐ATF complex, the present functional study thus provides some of the structure-function relationships that are required to understand the molecular mechanisms controlling the interactions between uPAR and its biological ligands such as pro-uPA, vitronectin, and certain integrins. This information may also prove essential for "decoding" the well established species barrier of both the uPAR/uPA interaction and the selective inhibition of this interaction by peptide antagonists. The pronounced species barrier between man and mouse is a considerable problem that has to be tackled when testing the pharmacological effects of uPAR antagonists in various mouse cancer models (11).