Did evolution create a flexible ligand-binding cavity in the urokinase receptor through deletion of a plesiotypic disulfide bond?

The urokinase receptor (uPAR) is a founding member of a small protein family with multiple Ly6/uPAR (LU) domains. The motif defining these LU domains contains five plesiotypic disulfide bonds stabilizing its prototypical three-fingered fold having three protruding loops. Notwithstanding the detailed knowledge on structure-function relationships in uPAR, one puzzling enigma remains unexplored. Why does the first LU domain in uPAR (DI) lack one of its consensus disulfide bonds, when the absence of this particular disulfide bond impairs the correct folding of other single LU domain-containing proteins? Here, using a variety of contemporary biophysical methods, we found that reintroducing the two missing half-cystines in uPAR DI caused the spontaneous formation of the corresponding consensus 7–8 LU domain disulfide bond. Importantly, constraints due to this cross-link impaired (i) the binding of uPAR to its primary ligand urokinase and (ii) the flexible interdomain assembly of the three LU domains in uPAR. We conclude that the evolutionary deletion of this particular disulfide bond in uPAR DI may have enabled the assembly of a high-affinity urokinase-binding cavity involving all three LU domains in uPAR. Of note, an analogous neofunctionalization occurred in snake venom α-neurotoxins upon loss of another pair of the plesiotypic LU domain half-cystines. In summary, elimination of the 7–8 consensus disulfide bond in the first LU domain of uPAR did have significant functional and structural consequences.

The urokinase-type plasminogen activator receptor (uPAR) 2 is an extracellular membrane protein composed of three homo-logous Ly6/uPAR-type (LU) domains and a C-terminal glycosylphosphatidylinositol (GPI) membrane anchor (1). It serves to focalize plasminogen activation on cell surfaces via its highaffinity binding to the urokinase-type plasminogen activator (uPA) (2). In so doing, it facilitates extravascular fibrin surveillance reducing the adverse effects of chronic inflammation caused by unremitting fibrin deposition (3). Besides promoting pericellular proteolysis, the uPA⅐uPAR interaction also stimulates cell adhesion and migration via direct and indirect interactions with vitronectin and integrins (4 -9). Elegant transgenic mouse models show that the interaction between uPA and uPAR promotes hepatic fibrin clearance (3) and improves neuronal recovery after either cerebral ischemia (10,11) or spinal cord injury (12). Notwithstanding these beneficial effects, the uPA⅐uPAR interaction may also elicit detrimental pathological effects, particularly in relationship to chronic inflammation. In genetic mouse models, the interplay between uPA and uPAR augments the pathogenesis of collagen-induced arthritis (13,14). In line with these causal correlations, high plasma levels of shed uPA/uPAR predict poor prognosis in several pathologic conditions with inflammatory lesions, e.g. bacterial infections (15,16), kidney disease (17,18), and invasive and metastatic solid cancers (19). The latter association spurred a considerable interest in developing uPAR-specific targeting strategies intended for use in cancer therapy (20 -24). These initiatives are now being supplemented by the development of uPAR-targeting probes for noninvasive imaging of uPAR expression using either (i) positron emission tomography to guide patient staging (25)(26)(27) or (ii) near-IR fluorescence to guide precision cancer surgery by improving margin resection (28 -30).
Crystal structures of uPAR solved in complex with its natural protein ligands (Fig. 1, C and D) (31)(32)(33)(34), small molecule antagonists (21,35,36), or antibodies (31,37) reveal that all three LU domains in uPAR combine to form a compact globular structure. This assembly creates (i) a large and hydrophobic uPAbinding cavity comprising elements from all three LU domains and (ii) a smaller peripheral binding site for the somatomedin B (SMB) domain of vitronectin at the interface between the first (DI) and second (DII) LU domain in uPAR. Although these binding sites are nonoverlapping they do, nevertheless, interact cooperatively. Prior uPA-binding thus increases uPAR's affinity for vitronectin (7) and in so doing it leads to increased cell adhesion and migration (5,8,9). We showed previously that the dynamic assembly of the LU domains in uPAR enables this allosteric regulation of ligand binding (5,38). Combining our biophysical and functional data led us to conclude that uPA occupancy drives uPAR into a more closed and compact conformation and this increases the affinity for SMB (5). Locking uPAR permanently in this compact conformation, by introducing a nonnative disulfide bond between DI and DIII, by-passed the cooperativity of uPA binding and generated a constitutive high-affinity binding site for SMB (6,39). Our data, furthermore, suggested that uPAR DI plays a dominating role in this flexibility (38). Given that this particular domain differs from the otherwise invariant LU-consensus motif, inasmuch as it lacks one of the 5 plesiotypic disulfide bonds (40), we speculate that the loss of this structural constraint could have been instrumental for evolving high-affinity uPA binding and gaining cooperativity in vitronectin binding. This proposition is not unprecedented as snake venom ␣-neurotoxins (representing secreted single LU-domain proteins) developed potent neurotoxicity toward synapsid targets and underwent neofunctionalization after deletion of a plesiotypic disulfide bond (41). In this study, we therefore introduced the lacking consensus disulfide bond in human uPAR DI and studied the consequences thereof on the structural flexibility and ligand binding properties of intact uPAR.

Loss of a consensus disulfide bond in uPAR DI
Sequence alignments of the three homologous LU domains in human uPAR clearly show that the 5-disulfide bond signature, considered a plesiotypic trait of ancient three-fingered neurotoxins (41), is maintained in both uPAR DII and uPAR DIII (Fig. 1A). This primordial disulfide pattern was first identified in the basal-type ␣-neurotoxins (e.g. denmotoxin) that are present in venoms of nonfront-fanged snakes (Colubridae) that feed preferentially on nonmammalian prey (41,42). In accordance with this feeding behavior, their ␣-neurotoxins have only weak affinity for mammalian nicotinic acetylcholine receptors (41,42). In the more advanced elapid snakes, the ␣-neurotoxins gained high affinity for mammalian acetylcholine receptors via deletion of the 2-3 LU consensus disulfide bond in loop 1 (41,42), as illustrated in Fig. 1A. Intriguingly, this sequence alignment and our experimental disulfide assignments (31,32,35,40) reveal that uPAR DI unexpectedly lacks another of the invariant disulfide bonds that defines the ancient LU protein domain fold. In this case, the missing disulfide (denoted 7-8 in Fig. 1) is located at the base of loop 3 connecting ␤-strands E and F (Fig. 1, E and F) and it is absent from DIs of all known mammalian orthologues of uPAR (Fig. S1). The deletion of this particular disulfide bond in uPAR during evolution is remarkable given that single LU domain-containing proteins, such as GPIHBP1, CD59, and -bungarotoxin require this disulfide bond for their correct protein folding and function (43)(44)(45).

Reintroducing the missing disulfide bond in uPAR DI
To assign the most probable site(s) for reintroducing the primordial 7-8 consensus disulfide bond in uPAR DI, we exam-ined primary sequence alignments of individual LU domains. This clearly pinpointed Thr 51 and Val 70 as the most promising candidate pair (Fig. 1A, Figs. S1 and S8). This notion was further substantiated by the pairwise C ␤ -C ␤ atom distances in seven different crystal structures available for uPAR in complex with various ligands (Table S5). The C ␤ -C ␤ distances of these Thr 51 -Val 70 pairs were 6.4 Ϯ 0.5 Å, which is slightly longer than the distances for the bona fide 7-8 consensus disulfide bonds present in DII (4.0 Ϯ 0.2 Å) and DIII (3.9 Ϯ 0.4 Å). Nonetheless, evaluations focused only on minimizing structural perturbations highlighted yet another possible candidate pair, as the C ␤ -C ␤ atoms for Lys 50 and Val 70 were only 5.0 Ϯ 0.4 Å apart.
Based on these considerations, we chose to express both uPAR T51C-V70C and uPAR K50C-V70C (residues 1-283) in Drosophila S2-cells and purify the secreted proteins. To confirm the oxidation status of the introduced cysteine residues (i.e. validating that they are indeed engaged in disulfide bond formation), we subjected uPAR to limited proteolysis with chymotrypsin under nondenaturing conditions. We optimized the conditions to hydrolyze predominately the Tyr 87 -Ser 88 peptide bond in the linker region between DI and DII and to a lesser extent the Tyr 57 -Arg 58 peptide bond located in loop 3 of DI. Mass spectrometry confirmed that all cysteine residues in these protein preparations were engaged in disulfide bonding (Table  1, Fig. S2).

Introducing the 7-8 disulfide bond desensitizes DI to enzymatic deglycosylation and limited proteolysis in intact uPAR
We suspected that the inherent flexibility of DI as well as its assembly with DII and DIII were likely to be perturbed by the constraints introduced by the additional disulfide bond. Several independent lines of experimental evidence support this assumption.
Introducing the 7-8 disulfide bond rendered the glycan attached to DI (Asn 52 ) resistant toward PNGase F-mediated release, when incubating intact uPAR with the deglycosidase under nondenaturing conditions. As shown in Fig. 2A, PNGase F readily removed the glycan attached to DI from full-length uPAR wt , but was unable to do so in uPAR T51C-V70C . Likewise, uPAR K50C-V70C was a very poor substrate for PNGase F. Notably, a similar resilience toward PNGase F was acquired by uPAR wt when it was driven into its closed conformation either by uPA binding (46) or by insertion of a remote interdomain disulfide bond (uPAR H47C-N259C , Fig. 1G) (6). The accessibility of the glycan on Asn 52 to PNGase F was thus markedly compromised by the structural constraints originating from the introduction of the 7-8 disulfide bond between ␤-strands IE and IF (Fig. 1F) and this recapitulated the properties observed for uPAR in complex with uPA. All glycans attached to DII or DIII proved resistant to PNGase F in all tested uPAR variants.
Probing the various uPAR mutants by limited proteolysis with chymotrypsin revealed that the cleavage efficiency of the exposed linker region between DI and DII (Tyr 87 -Ser 88 ) was largely unaffected by the constraints from the additional disulfide bonds or by occupancy with GFD (Fig. 2B). In contrast, the sensi-

Consensus disulfide bonds in LU domains
tivity toward a secondary cleavage site at Tyr 57 -Arg 58 within loop 3 proved markedly different between the various uPAR mutants. Although uPAR T51C-V70C appeared slightly more prone to this cleavage compared with uPAR wt , uPAR H47C-N259C was refractory. We suspect that this resistance reflects that DI remains fully integrated with DIIDIII in uPAR H47C-N259C despite having a cleaved linker region between DI and DII (6). Maintaining the compact globular three-domain assembly of intact uPAR would thus shield Tyr 57 -Arg 58 from proteolysis. Aligned with that proposition, we observed that GFD occupancy also prevents the cleavage at this position (Fig. 2B). An unexpected fuzziness in the electrophoretic mobility of the 1-57 fragment from uPAR K50C-V70C complicated the kinetic evaluation of the second chymotrypsin cleavage event in this mutant. The origin of this aberrant mobility remains unclear. Mass spectrometry of uPAR K50C-V70C digested for the 24 h showed only the expected mass of DI with one internal pep-

Consensus disulfide bonds in LU domains
tide cleavage (ϩ18 Da). Furthermore, this mass collapsed completely into the expected mass for the 1-57 fragment upon reduction (data not shown).
One plausible mechanism explaining these differences in the sensitivity of Tyr 57 -Arg 58 to chymotrypsin cleavage is that they report on the half-life of the assembled three LU domains after the initial Tyr 87 -Ser 88 cleavage. In one extreme case, DI would remain covalently tethered to DIIDIII thus sterically shielding Tyr 57 -Arg 58 (e.g. in uPAR H47C-N259C ). In another case, little or no DI would remain attached to DIIDIII thus allowing significant cleavage at this position (e.g. in uPAR T51C-V70C ). To test this possibility, we performed size exclusion chromatography of samples freshly treated with chymotrypsin (Fig. 3). The elution profiles from the size exclusion column revealed: (i) that uPAR DI released by limited proteolysis from uPAR wt remains partly associated to DIIDIII (Fig. 3A); (ii) that prior occupancy with GFD greatly enhances this association (Fig. 3C); and (iii) that introduction of the 7-8 consensus disulfide bond completely eliminates the noncovalent association of DI and DIIDIII (Fig. 3, B and D). Combined, these data provide further evidence supporting the notion that the constraints from the 7-8 disulfide bond lower the propensity for the globular assembly of DI-DII-DIII, which to some extent relies on a prominent contribution from ␤IE to the DI-DII interface (Fig. 1, B and E).

Flexibility in the globular three-domain assembly of uPAR by small angle X-ray scattering
To gain further insights into possible differences in the interdomain flexibility between the various disulfide-constrained uPAR conformers, we performed small angle X-ray scattering (SAXS) analyses. We analyzed both uPAR and uPAR⅐ATF complexes to compare the domain flexibility before and after ligand-induced compaction of the receptor. To maximize sample monodispersity, we fractionated uPAR and uPAR⅐ATF complexes by size exclusion chromatography before collecting batch scattering data by SAXS. From the normalized scattering data, we first derived the intra-particle distance distribution function, p(r) providing model-independent information on the shape parameters ( Fig. 4). Comparison of the different unoccupied uPAR mutants revealed that only uPAR H47C-N259C exhibits a symmetrical, bell-shaped p(r) function indicative of a compact and globular structure with a radius of gyration (R g ) of 22.4 Ϯ 0.1 Å and a maximal particle dimension (D max ) of 70 Ϯ 5 Å (Table S2). In contrast, reintroducing the missing 7-8 consensus disulfide in uPAR DI did not lead to large scale changes in the overall shape parameters, the p(r) functions for uPAR K50C-V70C and uPAR T51C-V70C are almost superimposable onto that of uPAR wt (Fig. 4B). Upon ATF binding, all uPAR disulfide conformers compacted into similar sized particles with a D max of 90 Å (Fig. 4D, Tables S1-S4). A small shift to higher distances in the p(r) function of the uPAR wt ⅐ATF complex is observed relative to ATF complexes with the uPAR disulfide mutants, but whether this reflects a significant structural difference is unclear. Nonetheless, transforming the scattering data into dimensionless Kratky plots provides a clear ranking of the unoccupied uPAR disulfide conformers into three groups according to their degree of flexibility. This transformation of the SAXS data are particularly useful to obtain a semi-quantitative analysis of the propensity of a given protein to adopt a globular fold (represented by a bell-shaped curve Table 1 Verification of disulfide bond status by MS uPAR DI (residues 1-87) was excised from intact uPAR 1-283 by limited chymotrypsin digestion and the molecular masses were determined by LC-ESI-MS and maximum entropy (MaxEnt1) deconvolution of the charge state distributions of the proteins (settings, Gaussian FWHM: 1.0 and resolution: 0.25 Da/channel). The calculated masses are from the primary sequences including a paucimannosidic glycan (Man 3 GlcNAc 2 Fuc; 1,038.5 Da) tethered to Asn 52 in S2-cells produced uPAR (42

Consensus disulfide bonds in LU domains
with peak at 1.104 for qR g ϭ ͌3) or be intrinsically disordered (represented by a hyperbolic curve with a plateau around qR g values of 1.5-2.0). From the Kratky plots presented in Fig. 5A, uPAR wt exhibits the greatest flexibility and uPAR H47C-N259C is stable and globular. Both uPAR variants with an intact 7-8 disulfide bond show an intermediate flexibility (Fig. 5A). In the presence of ATF, this difference in flexibility largely disappears, as illustrated by the Kratky plots in Fig. 5B. SAXS-driven ensemble modeling of the scattering data by the ensemble optimization method (EOM) yielded a similar conclusion regarding the flexibility of unoccupied uPAR (Fig. 5, C and D). In this analysis, a homogeneous ensemble of relatively compact structures provides an excellent fit to the scattering data for uPAR H47C-N259C with low flexibility metrics (R flex of 47%, relative to the threshold of randomness: 85%, determined from the random pool) calculated from the probability distributions (47). In contrast, a heterogeneous ensemble comprising both extended and compact conformations with a relatively high R flex of 81% is required to fit the scattering data for uPAR wt . Intermediate ensemble compositions with R flex values of 74 and . Asterisks identify peak fractions and the yellow arrows show DI associated to DIIDIII; white arrows show detached DI. A, uPAR wt (note, 20 -30% of DI co-elutes with DIIDIII). B, uPAR T51C-V70C (no co-elution). C, uPAR wt in the presence of a 4-fold molar excess of GFD (note, Ͼ90% of DI co-elutes with DIIDIII and the peak eluted earlier indicative of the formation of a trimolecular DI⅐GFD⅐DIIDIII complex). D, uPAR K50C-V70C (no co-elution). E, uPAR H47C-N259C (note, 100% DI co-elutes with DIIDIII due to the covalent tether between DI and DIII). F, uPAR C6S-C12S (note, 5-10% of DI co-elutes with DIIDIII).

Consensus disulfide bonds in LU domains
73% provide good fits to uPAR T51C-V70C and uPAR K50C-V70C , respectively. These properties are also evident from the variations in shape observed in the refined ab initio surface envelopes reconstructed from the experimental SAXS data (Fig. S4). Extended models are generated for the highly flexible variants (uPAR wt , uPAR T51C-V70C , and uPAR K50C-V70C ), in contrast to the compact models generated for the less flexible uPAR H47C-N259C and all uPAR⅐ATF complexes.

Dynamics of uPAR DI
Previously we showed with hydrogen-deuterium exchange MS (HDX-MS) that uPAR DI undergoes a pronounced change in flexibility during the compaction of intact uPAR that occurs on uPA binding (38). In particular, peptides spanning the third loop of DI (i.e. ␤IE and ␤IF) experience significant reductions in their deuterium uptake when uPAR is driven into its compact state by uPA binding (38). Due to the covalent tethering of ␤IE and ␤IF by the 7-8 consensus disulfide bond in LU domains, we suspected that it could stabilize the ␤-sheet between strands E and F in uPAR DI and this could in part be reconciled with the reduced flexibility observed in the Kratky plot of the SAXS data (Fig. 5A). To probe this proposition further, we performed a continuous deuterium labeling of the different uPAR disulfide variants in the presence or absence of saturating levels of GFD and determined the deuterium uptake values by MS after pepsin digestion with special emphasis on peptide (57)(58)(59)(60)(61)(62)(63)(64)(65)(66). The deuterium uptake plots for this peptide reveals a considerable variability in the flexibility of ␤IE and ␤IF in the different disulfide variants (Fig. 6A). Interestingly, the ranking of the deuterium uptake recapitulates to some extent the flexibility assigned

Consensus disulfide bonds in LU domains
by Kratky plots of the scattering data, i.e. uPAR T51C-V70C Ն uPAR wt Ͼ uPAR K50C-V70C Ͼ uPAR H47C-N259C . Yet one notable difference was apparent. The deuterium uptake for residues 57-66 differed significantly between the two uPAR variants with introduced 7-8 disulfide bonds, uPAR T51C-V70C exhibiting the far greater uptake thus resembling uPAR wt (Fig.  6A). All isotope envelopes are unimodal at the shortest exchange time (10 s) in the presence and absence of GFD (Fig. S6), demonstrating that the amount of misfolded protein is negligible (48). This in turn signifies that the faster exchange kinetics in uPAR T51C-V70C as compared with uPAR K50C-V70C reflects increased dynamics rather than irreversible protein misfolding (Fig. S6).
These differences in exchange rates of the 57-66 fragment among the tested uPAR disulfide conformers is almost erased in the corresponding uPAR⅐GFD complexes (Fig. 6B). The compaction of uPAR by ligand binding thus reduces the macromolecular interdomain flexibility as well as the intradomain flexibility in DI, as monitored by SAXS and HDX-MS, respectively. Deuterium uptake plots for other regions in uPAR are shown in Fig. S5, but none of those show as prominent effects as peptide(57-66) (reporting on ␤IE and ␤IF).

Binding kinetics of uPA to the uPAR disulfide variants
To determine the kinetic rate constants for the uPA⅐uPAR interactions by surface plasmon resonance, we developed a cap-

Consensus disulfide bonds in LU domains
turing system, which enabled a homogenous presentation of uPAR via an antibody-mediated noncovalent tether onto the biosensor chip. To accomplish this, we chose ATN-615 as capturing mAb for uPAR because its functional epitope is located distant to the uPA-binding cavity (Fig. S7) and it forms a very stable complex with uPAR (k off being 6 ϫ 10 Ϫ5 s Ϫ1 , Fig. 7A). As shown in Fig. 7 and Table 2, this experimental setup provided kinetic data of high-end quality and it revealed a tight interaction between uPAR wt and ATF 1-143 (K D is 20 pM with a k on of 1 ϫ 10 7 M Ϫ1 s Ϫ1 and a k off of 2 ϫ 10 Ϫ4 s Ϫ1 ). As shown in Table  2, all the tested uPAR mutants had comparable association rate constants (k on ), but they differed significantly in dissociation rate constants (k off ). Notably, reintroduction of the 7-8 consensus disulfide bond at the position defined by sequence predictions led to a Ͼ40-fold decrease in the stability of the corresponding uPAR T51C-V70C ⅐ATF complex as reflected by the greater k off value (Fig. 7, B versus E). Importantly, mutating the positions chosen for the 7-8 disulfide bonds individually to alanine did not recapitulate this gross impairment in stability (i.e. T51A and V70A). In contrast, reintroducing the disulfide bond at the position judged to cause minimal structural perturbation led to a moderate 3.5-fold increase in the k off for the corresponding uPAR K50C-V70C ⅐ ATF complex (Fig. 7D).

SMB binding to the various disulfide variants of uPAR
To assess the low-affinity binding between the SMB domain of vitronectin and uPAR by surface plasmon resonance (SPR) in solution, we used a slightly different format. In this setup, 2-fold serial dilutions of SMB reached equilibrium binding with uPAR or uPAR⅐ATF complexes in solution before they were captured on the sensor surface by the anti-uPAR mAb R24 (Fig. 8A). The virtue of this system is that it uses relatively high concentrations of uPAR and ATF (100 and 150 nM), which makes certain that ATF saturates uPAR despite some of the analyzed disulfide mutants having impaired uPA binding, e.g. uPAR T51C-V70C . In accordance with previous studies (5-7), uPAR wt bound SMB with a K D of 3 M, whereas no binding of SMB to uPAR R91D could be detected (Fig. 7B).
As shown in Table 2 and Fig. 7, the introduction of the 7-8 consensus disulfide bond in uPAR DI led to a minor decline in the affinity of SMB for intact uPAR. With one exception, all uPAR mutants exhibited an increased affinity for SMB upon saturation with ATF. As reported previously (6), the affinity of uPAR H47C-N259C for SMB did not change upon ATF binding as it already presents the higher affinity without bound ATF (Fig.  7C). The present SPR platform is thus capable of detecting the allosteric effect of ATF saturation on the SMB binding to uPAR in solution, albeit the magnitude of this effect is slightly less pronounced compared with the previous solid-phase based detection systems (6,7). Nonetheless, this analysis clearly showed that the allosteric impact of ATF on SMB binding remain intact in both uPAR T51C-V70C and uPAR K50C-V70C showing that reintroduction of the 7-8 consensus disulfide bond in uPAR DI does not uncouple the beneficial effect of ATF on SMB binding. In fact, binding of ATF managed to compensate for the lower inherent affinity of uPAR T51C-V70C and uPAR K50C-V70C for SMB and restored an affinity comparable with that of uPAR wt ⅐ATF (Fig. 7, B versus D and E).

Impact of deleting the 2-3 LU disulfide bond in uPAR DI
Finally, we explored the impact of deleting the 2-3 disulfide bond in uPAR DI, the event that presumably led to neofunctionalization of this LU domain in snake venom ␣-neurotoxins (41). In all our functional tests, the uPAR C6S-C12S mutant behaved essentially as uPAR wt despite having only three of the five plesiotypic LU domain disulfides in DI. We found that (i) DI associate with DIIDIII after chymotrypsin cleavage of the linker region in uPAR C6S-C12S (Fig. 3F); (ii) the affinity of uPAR C6S-C12S for uPA and SMB is comparable with those of uPAR wt (Figs. 7F and 8F); and (iii) ATF-binding stimulates SMB binding to the same extent as uPAR wt (Fig. 8F). These data indicate that deletion of the 2-3 disulfide bond in

Consensus disulfide bonds in LU domains
uPAR DI has no deleterious effect on the known ligand interactions with uPAR.

Discussion
Loss and gain of disulfide-bonded cysteine residues (halfcystines) occurs very infrequently during evolution. One possible reason for this low frequency is that missense mutations of consensus half-cystines rarely survive selection and become permanently integrated in the genomes due to the deleterious effects of the reactive free thiol group in the partner half-cystine (49). Interestingly, one study found that in 99% of the cases, where missense mutations of half-cystines were maintained during evolution, both pairs of half-cystines were in fact replaced in concert (50). Circumstantial evidence from human genetics on LU domains align well with this proposition. One example illustrating this relationship is the pathological outcome of natural missense mutations in GPIHBP1. This protein has a single LU domain and it plays an essential role in intravascular triglyceride hydrolysis by shuttling the lipoprotein lipase to the capillary lumen (51). Individuals with a dysfunctional GPIHBP1 develop severe chylomicronemia. Notably, the majority of the missense mutations in human GPIHBP1 causing disease involves the replacement of single half-cystines (43) or introduction of an unpaired single cysteine (52). A similar pattern emerges for disease-causing mutations in the secreted single LU-domain protein SLURP1, where the dysfunctional protein is associated with a human skin disorder called mal de Meleda (53).
Notwithstanding the need for a concerted replacement of both half-cystines to eliminate a given disulfide bond, such rare events have in fact occurred in the evolution and diver-

Kinetics of uPAR⅐ATF interactions and K D of SMB binding
Analyses with single cycle protocols provided association (k on ) and dissociation (k off ) rate constants for the interactions between ATF in solution and different uPAR mutants captured on mAb ATN-615. This setup uses 3-4 rounds of single cycle injections each including five serial 2-fold dilutions of ATF  , which in combination covers the concentration range of 0.03-2 nM. Fitting with non-linear regression to a simple bimolecular interaction model yielded the kinetic rate constants and the stoichiometry was calculated as the molar ratio between captured ligand and calculated R max for the analyte. Fig. 7 shows the corresponding sensorgrams. Equilibrium binding of SMB to uPAR and uPAR⅐ATF complexes were measured by SPR after capture to the mAb R24 as described in Fig. 8

Consensus disulfide bonds in LU domains
sification of LU domain proteins. The evolution of snake venom ␣-neurotoxins clearly emphasizes this relationship, where the deletion of the 2-3 disulfide bond allegedly coincided with neofunctionalization and development of high potency toward synapsid neuronal acetylcholine receptors (41). One possible mechanism underlying this association is that the deletion of this particular disulfide bond would relax the LU domain scaffold, which allowed subsequent exploitation of new binding interfaces.
In the present study, we found that elimination of the 2-3 disulfide bond in the first LU domain of uPAR did not have notable functional consequences. Another report showed that elimination of this disulfide bond in the single LU domain protein CD59 also had no impact on its complement regulatory function (44). Based on these studies, it would thus appear that the plesiotypic 2-3 disulfide bond is not essential for the overall structural integrity of LU domain proteins, which would explain why its deletion could cause functional diversification in ␣-neurotoxins.
Paradoxically, the one plesiotypic LU domain disulfide bond that actually is lost in uPAR is the 7-8 disulfide in DI. In contrast to the 2-3 disulfide, this disulfide bond appears to be indispensable for the correct folding and function of single LU domain proteins in general (45,51,53,54). Furthermore, the deletion of this particular disulfide bond occurs in very few proteins and exclusively in the first LU domain of proteins with 2 or more LU domains e.g. uPAR/PLAUR (31,35,40), C4.4A/ LYPD3 (55,56), and Haldisin/LYPD5 (57). Bearing in mind that flexibility between the individual LU domains in uPAR plays an important role for ligand binding (5, 6, 38) it becomes highly pertinent to ask the question: did deletion of this plesiotypic 7-8 disulfide in uPAR DI facilitate the evolution of a highaffinity ligand-binding cavity for uPA by enabling a dynamic assembly of its three LU domains?
To address this question, we first needed to identify the putative positions of the two deleted half-cystines. Our comprehensive alignments of more than 50 annotated orthologous uPAR sequences from the class of Mammalia revealed an invariant distance (5 amino acids) between the 6th and 7th cysteines in the second and third LU domains of all orthologues (not shown, but Fig. 1A, and Figs. S1 and S8 provide representative alignments). Moreover, the 5th and 7th positions after the 6th consensus cysteine were invariantly a lysine and a glycosylated asparagine in uPAR DI (equivalent to Lys 50 and Asn 52 in human uPAR DI). In contrast, the 6th position, the assumed position of the missing 7th consensus cysteine, varied between species (Fig.  S1). The 8th and 9th consensus cysteines were always neighboring residues in those uPAR domains, where all 5 plesiotypic disulfide bonds were preserved. Importantly, uPAR-like orthologues annotated from the class of Reptilia including lizards, snakes, turtles, and crocodilians (Fig. S8) faithfully replicate these properties. It is therefore beyond any reasonable doubt that the correct position, from an evolutionary point view, for the introduction of the missing 7-8 disulfide bond in human uPAR DI is indeed Thr 51 and Val 70 .
The corresponding uPAR T51C-V70C mutant expressed well in S2-cells and all cysteines were engaged in the expected disulfide bonding in the purified protein. Nonetheless, our biochemical and biophysical analysis on the purified uPAR T51C-V70C revealed that the disulfide constrained DI behaves very different to that of uPAR wt . Among other features, we observed with size exclusion chromatography that the introduction of the 7-8 disulfide abrogated the interdomain interaction between DI and DIIDIII in uPAR T51C-V70C (Fig. 3). Importantly, we show that the penalty for introducing the missing disulfide in uPAR DI is a Ͼ40-fold reduction in its affinity for uPA, which would provide uPAR T51C-V70C with a K D that is at least 40-fold above the plasma concentration of pro-uPA in humans (58). It would therefore appear that the evolutionary deletion of the plesiotypic 7-8 disulfide bond in uPAR DI was essential for creating a high-affinity binding site for uPA via a flexible assembly of all three LU domains in uPAR. This proposition is well aligned with studies on the co-evolution of uPA and uPAR (32,59). Based on crystal structures of human and murine uPAR⅐ATF complexes and extensive mutagenesis (3,31,32,60), the functional hotspot residues in the ␤-hairpin of GFD for uPAR binding is well characterized in these species. The molecular basis for the species selectivity in the uPAR⅐uPA interaction between primates and nonprimate mammals is represented by the concerted replacement of the Asn 22 3 Tyr 23 and Trp 30 3 Arg 31 dyad in human and mouse uPA (32) (Fig. S9). Moreover, the majority of the important residues in the uPAR-binding motif of GFD from mammals are also conserved within the class of Reptilia (Fig. S9A) and this class is also the earliest class where a bona fide uPAR orthologue with three LU domains can be traced (59). In this phylogenetic class, uPAR DI has already lost its plesiotypic 7-8 disulfide (Fig. S8). In the class of Sarcopterygii, a uPAR-like protein with three LU domains has recently been identified in the African lungfish Protopterus (59). This protein is unique in the sense that it retains the full signature of an ancient LU domain with all five plesiotypic disulfide bonds present in all three domains (Fig. S8). Examination of the ␤-hairpin of GFD in the corresponding uPA orthologue reveals that the binding motif known from Mammalia and Reptilia has not yet evolved (Fig. S9A). It is therefore possible that this interaction is not operational in species belonging to lobe-finned fishes, but further functional studies on purified proteins is required before a definitive conclusion can be made.
Evolution of the low-affinity interaction between uPAR and the SMB domain of vitronectin follows a slightly different path. Although the hotspot residues in SMB for the vitronectin interaction with human uPAR is highly conserved all the way to Sarcopterygii (Fig. S10A), the hotspot residue in uPAR (i.e. Arg 91 in the linker region between DI and DII) is only conserved within Mammalia (Fig. S10B). It is therefore likely that the co-evolution, which shaped the uPARbinding site in SMB, originally was driven by the interaction between vitronectin and the cognate uPA-inhibitor plasminogen activator inhibitor type 1, in which the hotspot residue for SMB-binding (i.e. Arg 101 ) is conserved in lobe-finned fishes, bony fishes, and cartilaginous fishes (61). Later in evolution, the class of Mammalia presumably developed the uPAR⅐vitronectin interaction through convergent evolution.

Generation of uPAR 1-87 by limited proteolysis with chymotrypsin
The N-terminal LU domain of human uPAR (DI, residues 1-87) was excised by limited proteolysis with chymotrypsin using conditions that preferentially led to hydrolysis of the Tyr 87 -Ser 88 peptide bond in uPAR. In brief, 45 g of intact uPAR was incubated with 9 ng of chymotrypsin (E:S of 1:5000) in 54 l of PBS, pH 7.4, for 120 min at 25°C and the released uPAR DI 1-87 was isolated by size exclusion chromatography with a Superdex TM 75 HR10/300 column TM (GE Healthcare) operated at 0.5 ml/min in PBS (40). Time course experiments were conducted at higher chymotrypsin ratios (E:S of 1:750) for up to 24 h, which led to an additional cleavage at Tyr 57 -Arg 58 in uPAR wt .

Deglycosylation of intact uPAR under native conditions
Purified uPAR variants (10 g in 10 l of PBS) were incubated for 1 h at 25°C with 1 unit of PNGase F. To discriminate between effects on DI and DIIDIII, the PNGase F-treated samples were further incubated for 1 h with 2 ng of chymotrypsin (1:5000 (w/w)) to cleave the Tyr 87 -Ser 88 peptide bond in the linker region. Heating at 95°C in SDS-PAGE sample buffer containing 20 mM DTT stopped the reaction. After cooling and addition of 50 mM iodoacetamide, samples were analyzed by SDS-PAGE and visualized by Coomassie G250 staining.

Surface plasmon resonance
We determined the binding kinetics for the uPA⅐uPAR interactions with SPR measurements on a Biacore T200 TM system (GE Healthcare). To accomplish this, we immobilized a polyclonal rabbit anti-mouse immunoglobulin antibody (GE Healthcare) as the first capture layer (30 g/ml in 10 mM sodium acetate, pH 5.0) on a CM5 sensor chip using N-hydroxysuccinimide and N-ethyl-N- (3-(diethylamino)propyl)carbodiimide. This yielded a surface density of 1100 resonance units (RU), which corresponds to 7.3 fmol of mAb/mm 2 (assuming one RU ϳ 1 pg/mm 2 ). Injection of 1 M ethanolamine inactivated excess NHS-esters. To prepare the second capture layer, we injected 50 nM of a high-affinity mouse monoclonal anti-uPAR mAb (ATN-615) for 350 s at 20 l/min in the active flow cell only, which led to ϳ0.8 fmol of mAb/mm 2 (120 RU). The last capture step involved a 300-s injection of 50 nM uPAR at 20 l/min in both flow cells, which led to a capture level of 1.6 -2.0 uPAR molecules bound per ATN-615 in the active flow cell. Kinetic rate constants for the various uPA⅐uPAR interactions were determined with single cycle protocols by which five serial 2-fold dilutions of the interaction partner (ATF) were injected for 200 s without intervening regeneration and followed by a longer dissociation phase (1,000 -3,000 s dependent on the dissociation rate constant k off ). These real-time interactions were measured at 50 l/min in 10 mM HEPES, 150 mM NaCl, 3 mM EDTA, and 0.05% (v/v) surfactant P-20 at pH 7.4 at 20°C. Three consecutive injections of 10 mM glycine/HCl, pH 1.7, at the end of each single cycle regenerated the chip.
Fitting of the double blank referenced data by nonlinear regression to a bimolecular interaction model, assuming pseudo-first order reaction conditions, yielded the association (k on ) and dissociation (k off ) rate constants, the K D (k off /k on ), as well as the binding capacity (R max ). We used the evaluation software supplied with the instrument for global fitting (BiacoreT200 Evaluation TM 3.0).
Our experimental protocol for measuring equilibrium binding between SMB and uPAR or uPAR⅐ATF complexes relies entirely on steady state conditions in solution between 100 nM uPAR Ϯ 150 nM ATF and serial 2-fold dilutions of SMB (0.3-10 M). Binding of SMB was monitored by the mass increase of uPAR or uPAR⅐ATF complexes during capture by the anti-uPAR mAb R24 as a function of the added SMB concentration (37). We used amine chemistry to immobilize R24 directly on a CM5 chip at a density of 651 RU (4.3 fmol of mAb/mm 2 ). We designed this atypical binding protocol to minimize the effects of the vastly different inherent stabilities of the bimolecular uPAR wt ⅐ATF and uPAR T51C-V70C ⅐ATF complexes (see Table 2). Using concentrations of uPAR and ATF that were at least 100fold above the highest measured K D we made sure that uPAR was saturated with ATF during the steady state binding to SMB in solution.

Hydrogen-deuterium exchange
Purified uPAR was diluted in PBS to a final concentration of 30 M in the absence or presence of 2-fold molar excess of GFD  , and subsequent incubation for 15 min at 25°C secured complete complex formation. To initiate isotopic exchange, samples were diluted 10-fold in PBS exchange D 2 O buffer (10 mM Na 2 HPO 4 , 150 mM NaCl in D 2 O, pH read 7.2), which resulted in 90% D 2 O (v/v) in the final labeling solutions. Aliquots of 60 pmol of uPAR were withdrawn after 10, 100, and 1000 s of incubation at 25°C. We used acidification and lowered the temperature to efficiently prevent further exchange by adding 1 volume of ice-cold quench buffer (0.1 M Na 2 HPO 4 , 0.8 M tris-(2-carboxyethyl)phosphine, 2 M urea in H 2 O, pH 2.5) to the withdrawn sample and kept it on ice for 3 min to allow reduction of disulfide bonds. Subsequently, the quenched samples were snap frozen in liquid nitrogen and stored at Ϫ80°C until analysis. All samples were labeled in triplicates except

Consensus disulfide bonds in LU domains
uPAR T51C-V70C , which was measured in duplicates. To prepare an undeuterated control, the samples were prepared as described above except we replaced the PBS exchange D 2 O buffer with the corresponding protiated solvent. We prepared a fulldeuterium control by incubating the sample in the exchange buffer for 72 h at 37°C before quenching.

Determining deuterium uptake by UPLC-ESI-MS
Quenched and reduced uPAR samples (60 pmol) were thawed and immediately injected into a cooled (0°C) nanoAC-QUITY UPLC reversed-phased chromatographic system equipped with HDX technology (Waters, Milford, MA), using an ice-cold syringe to minimize back-exchange. The sample entered a 100-l injection loop prior to the on-line digestion of the protein, which was carried out by a 2.0 ϫ 20-mm column (IDEX Upchurch Analytical Guard Column, Oak Harbor, WA) packed with agarose-immobilized pepsin (Thermo Scientific Pierce, Rockford, IL), located in a compartment with a temperature of 20°C. The on-line digestion occurred at a flow rate of 300 l/min in 0.23% (v/v) formic acid in H 2 O and the generated peptic peptides were trapped on a short guard column and desalted for 3 min (ACQUITY UPLC BEH C18 VanGuard Pre-Column, 1.7 m, 2.1 ϫ 5 mm, Waters). Subsequently, the peptides were separated at a flow rate of 40 l/min on a 1.0 ϫ 50-mm analytical column (ACQUITY UPLC BEH C18, 1.7 m, 1.0 ϫ 50 mm, Waters) with a 12-min linear gradient from 5-50% (v/v) acetonitrile containing 0.23% (v/v) formic acid. In some experiments, we used a longer analytical column (100 mm).
Eluted peptides were introduced into a quadrupole TOF mass spectrometer (Synapt G2, Waters Corp.) by electrospray ionization. Acquired MS spectra was lock-mass corrected against leucine enkephalin and calibrated against sodium iodide.
Extensive cleaning reduced carryover between runs to less than 5%. In brief, we washed the injection loop with 200 l of 50% (v/v) MeOH in H 2 O containing 0.23% (v/v) formic acid, followed by 200 l of 0.23% (v/v) formic acid followed by a blank gradient.
The peptic peptides from uPAR were identified by collisioninduced dissociation with data independent (MS E ) acquisition mode. Protein Lynx Global Server 3.0 (PLGS) software (Waters) searched and identified the peptic peptides from uPAR. Data processing with DynamX 3.0 (Waters) determined the deuterium content in each peptide.

Small-angle X-ray scattering
Synchrotron radiation X-ray scattering data were collected on the EMBL P12 beamline of the storage ring PETRA III (DESY, Hamburg) and EMBL X33 beamline of the storage ring DORIS (DESY, Hamburg) (Tables S1-S4), using PILATUS 2M and 1M pixel detectors (DECTRIS, Switzerland), respectively. Batch experiments of uPAR and uPAR⅐ATF complexes (purified by size exclusion chromatography) were measured in 20 mM PBS, 5% glycerol, pH 7.4, Ϯ 50 mM NaSO 4 , while flowing through a temperature controlled capillary (P12: 1.2-mm inner diameter, X33: 1.7-mm inner diameter) at 10°C. Twenty frames of 0.05-s exposure time (P12) or four frames of 30-s exposure time (X33) were collected. The sample-to-detector distance was 2.7 m (P12) and 3.1 m (X33), covering a range of momentum transfer 0.002 Å Ϫ1 Յ s Ն 0.5 Å Ϫ1 and 0.008 Å Ϫ1 Յ s Ն 0.6 Å Ϫ1 (s ϭ 4sin/, where 2 is the scattering angle, and ϭ 1.24 Å is the X-ray wavelength), respectively. Based on comparison of successive frames, no detectable radiation damage was observed. Data from the detector were normalized to the transmitted beam intensity, averaged, placed on absolute scale relative to water, and the scattering of buffer solutions subtracted. All data manipulations were performed using PRIMUSqt and the ATSAS software package (64). The forward scattering I(0) and radius of gyration, R g were determined from Guinier analysis (65) assuming that at very small angles (s Յ 1.3/R g ) the intensity is represented as I(s) ϭ I(0)exp(Ϫ(sR g )2/ 3)). These parameters were also estimated from the full scattering curves using the indirect Fourier transform method implemented in the program GNOM (66), along with the distance distribution function p(r) and the maximum particle dimensions D max . Molecular masses of solutes were estimated from I(0) by computation of partial specific volume and the contrast between the glycosylated protein sequence and the chemical components of the solution using the MULCH server (http:// smb-research.smb.usyd.edu.au/NCVWeb/) 3 (67). Computation of theoretical scattering intensities was performed using the program CRYSOL (68).

Ab initio shape determination
Low resolution shapes were reconstructed from SAXS data using the programs DAMMIF (69), which represents the macromolecule as a densely packed interconnected configuration of beads or chain-like ensemble of dummy residues, respectively, that best fits the experimental data I exp (s) by minimizing the discrepancy, where N is the number of experimental points, c is a scaling factor, and I calc (s j ) and (s j ) are the calculated intensity and the experimental error at the momentum transfer s j , respectively. Multiple modeling runs were conducted to verify the stability of the solution, and to establish the most typical 3D reconstructions according to a spatial discrepancy measure using DAMAVER (70).

Ensemble modeling
Crystal structures of uPAR and uPAR⅐ATF complex (PDB IDs 2FD6, 3U74, and 3U73) were used as template for flexible modeling of the uPAR domain structures in solution. Glycosylation was introduced into the models using the GLYCOSYLA-TION routine of ATSAS (64). The program EOM (47) was used to generate a pool of 10,000 uPAR conformations with flexible linkers defined between the three LU domains. A generic algorithm was then employed to select subsets of conformers from the pool that best fit the experimental scattering data for each