Structural Basis for the Binding Specificity of Human Recepteur d'Origine Nantais (RON) Receptor Tyrosine Kinase to Macrophage-stimulating Protein*

Background: RON and MET receptors bind their ligands MSP and HGF selectively and activate different signaling pathways. Results: Crystallographic and analytical ultracentrifugation studies provide important information about RON-MSP interaction. Conclusion: RON-MSP and MET-HGF exhibit 2:2 complex stoichiometry, but differences within the respective interfaces explain the strict ligand-receptor specificity. Significance: Signaling pathways must be exquisitely regulated with no cross-reactivity between related systems. Recepteur d'origine nantais (RON) receptor tyrosine kinase and its ligand, serum macrophage-stimulating protein (MSP), play important roles in inflammation, cell growth, migration, and epithelial to mesenchymal transition during tumor development. The binding of mature MSPαβ (disulfide-linked α- and β-chains) to RON ectodomain modulates receptor dimerization, followed by autophosphorylation of tyrosines in the cytoplasmic receptor kinase domains. Receptor recognition is mediated by binding of MSP β-chain (MSPβ) to the RON Sema. Here we report the structure of RON Sema-PSI-IPT1 (SPI1) domains in complex with MSPβ at 3.0 Å resolution. The MSPβ serine protease-like β-barrel uses the degenerate serine protease active site to recognize blades 2, 3, and 4 of the β-propeller fold of RON Sema. Despite the sequence homology between RON and MET receptor tyrosine kinase and between MSP and hepatocyte growth factor, it is well established that there is no cross-reactivity between the two receptor-ligand systems. Comparison of the structure of RON SPI1 in complex with MSPβ and that of MET receptor tyrosine kinase Sema-PSI in complex with hepatocyte growth factor β-chain reveals the receptor-ligand selectivity determinants. Analytical ultracentrifugation studies of the SPI1-MSPβ interaction confirm the formation of a 1:1 complex. SPI1 and MSPαβ also associate primarily as a 1:1 complex with a binding affinity similar to that of SPI1-MSPβ. In addition, the SPI1-MSPαβ ultracentrifuge studies reveal a low abundance 2:2 complex with ∼10-fold lower binding affinity compared with the 1:1 species. These results support the hypothesis that the α-chain of MSPαβ mediates RON dimerization.

, implicating it in the maintenance and restructuring of the extracellular matrix during cellular growth and migration processes. Consequently, RON is an important target for cancer therapies using anti-RON monoclonal antibodies as well as small molecule kinase inhibitors (22)(23)(24).
The RON polypeptide comprises an extracellular ligand binding domain and a cytoplasmic tyrosine kinase domain, connected by a short membrane-spanning region. The RON ectodomain is subdivided into six domains: the N-terminal semaphorin domain (Sema), a small cysteine-rich plexin-semaphorin-integrin domain (PSI), and four immunoglobulinplexin-transcription factor domains (IPT [1][2][3][4] ). Precursor RON is a single-chain glycosylated protein that undergoes proteolytic maturation at a consensus furin cleavage site (Arg 309 -Gly 310 in the Sema) prior to translocation onto the cell surface (1). The mature receptor consists of a 40-kDa RON ␣-chain containing the N-terminal half of Sema and a 145-kDa RON ␤-chain containing the rest of the protein. Currently, the MSPmediated activation of the RON receptor is presumed to be similar to the proposed activation mechanisms of MET by its ligand, HGF, a protein homologous to MSP (25). In other words, the binding of mature MSP (comprising a disulfidelinked ␣and ␤-chain heterodimer, hereafter termed MSP␣␤) to the RON ectodomain initiates the formation of a signalingcompetent RON dimer on the cell surface, juxtaposing the cytoplasmic kinase domains sufficiently close to induce autophosphorylation of conserved tyrosine residues, which leads to downstream signal transduction (4,26,27). The ras/mitogenactivated protein kinase (MAPK), phosphatidylinositol 3-kinase (PI3K)/Akt, focal adhesion kinase, and ␤-catenin molecules are activated by the MSP-mediated RON signal transduction pathways (28,29).
The RON-specific ligand is the MST1 gene product, MSP. The 80-kDa serum growth factor is composed of six domains: the N-terminal hairpin domain (N domain), four Kringle domains (K 1 -K 4 ), and a chymotrypsin-like serine protease (SP) domain that is devoid of a catalytic triad (replaced by Gln 522 , Gln 568 , and Tyr 661 ) (30). MSP shares ϳ50% sequence identity with HGF (Fig. 1B), and both ligands belong to the plasminogen-like growth factor family (31). Circulating MSP is synthesized in liver cells as a single-chain precursor (pro-MSP) that does not bind to RON (32,33). Under cellular stress, pro-MSP undergoes proteolytic maturation to become a disulfide-linked MSP␣␤, which binds and activates RON (33). Several serine proteases (kallikreins, matriptase, hepsin, and human airway trypsin-like protease) recognize the cleavage site (Arg 483 -Val 484 ) between the K 4 and SP domains (33). The 50-kDa ␣-chain (MSP␣) contains the N and K 1 -K 4 domains, whereas the 30-kDa ␤-chain (MSP␤) comprises the SP domain (4,30,31). Both ␣and ␤-chains of MSP are essential for its biological activity; however, the receptor-specific binding to RON Sema is mediated by the MSP ␤-chain alone (33)(34)(35). Mutagenesis studies identified the interacting residue pair, Arg 682 /Glu 648 , and the neighboring Arg 683 in MSP␤ as critical for RON receptor binding and activation (32)(33)(34)36). The interaction between MSP␣ and RON is weak and not always detectable (33,34). By contrast, HGF␣ binds to the MET with an affinity higher than that of HGF␤ (37). In fact, splice variants of HGF, NK 1 and NK 2 , function as MET agonist and antagonist, respectively (38). To gain insights into the structural determinants of RON-MSP specificity, the crystal structure of the RON SPI 1 -MSP␤ complex was determined at 3.0 Å, and the binding interaction was characterized using analytical ultracentrifugation (AUC).
The human MST1 gene was amplified from a cDNA clone (ID 5190966, Open Biosystems, Inc.). The MSP proteins, containing a C-terminal His 6 tag, were purified from the D. melanogaster S2 conditioned medium as previously described (40). MSP (Gln 19 -Gly 711 ), MSP␣ (Gln 19 -Lys 464 ), and MSP␤ (Phe 465 -Gly 711 , which includes 19 linker residues to the ␣-chain to facilitate five physiologically relevant disulfide bonds (40,41)) were stored in 20 mM MES, pH 6, 0.1 M NaCl, 0.02% (v/v) sodium azide at Ϫ80°C. MSP␤ used in crystallization studies also contained a C672S mutation, introduced to prevent an incorrect disulfide bond formation between Cys 672 and Cys 588 and to maintain the Cys 468 -Cys 588 linkage (41 Crystallization, Data Collection, and Structure Determination-Crystals of RON SPI 1 in complex with MSP␤ were obtained at room temperature using the vapor diffusion method. MSP␤ and RON SPI 1 at an ϳ1:1 molar ratio were mixed to yield ϳ60 M concentration. The drops comprised equal volumes of SPI 1 -MSP␤ and mother liquor containing 0.1 M Tris-HCl, pH 8.5, 20% (w/v) PEG 4000, 8% (v/v) isopropyl alcohol, and 4% (v/v) polypropylene glycol 400 (derived from condition 41 of Hampton Crystal Screen I). For data collection, thin plate-shaped RON SPI 1 -MSP␤ crystals were transferred to mother liquor supplemented with 30% (v/v) glycerol and flashcooled in liquid nitrogen. Diffraction data were collected at the General Medicine and Cancer Institutes Collaborative Access Team microbeamline at the Advanced Photon Source (Argonne National Laboratory, Argonne, IL), which was equipped with a MARmosiac CCD detector. Diffraction data were processed with XDS to a resolution of 2.8 Å (42). The structure was determined by molecular replacement (in the space group P2 1 2 1 2 1 ) using the program PHASER (43) with the free RON Sema (PDB code 4FWW) and free MSP␤ (PDB code 2ASU) structures as the search models (39,41). Problems with the progress of the refinement due to pseudomerohedral twinning were tracked with the programs SFCHECK and XTRIAGE in PHENIX (44,45). Structure refinement, including a pseudomerohedral twinning rule, was conducted at 3.0 Å resolution using the program REFMAC5 (46). Model building and structure modification was performed using the interactive computer graphics program COOT (47). Molecular interfaces were calculated using PISA and ProFace (48,49). Topological complementarity of interacting surfaces was calculated using the shape correlation statistic program SC (50), as implemented in CCP4. Figures were prepared with PyMOL (DeLano Scientific), MOLSCRIPT, and RASTER3D (51,52).
Analytical Ultracentrifugation-Sedimentation velocity (SV) and sedimentation equilibrium (SE) experiments were performed at 20°C using a ProteomeLab Beckman XL-A ultracentrifuge with an absorbance optical system and a 4-hole An60-Ti rotor (Beckman Coulter). For SV, 400 l of protein, dialyzed in PBS, pH 7.4, and 420 l of PBS were loaded into the sample and reference sectors of the dual-sector charcoal-filled epon centerpieces. The cells were centrifuged at 50,000 rpm, and the absorbance data for 0.125-30 M proteins were collected at 230, 250, or 280 nm to obtain linear signals of Ͻ1.25 absorbance units. The absorbance signal was monitored in a continuous mode with a step size of 0.003 cm and a single reading per step. Sedimentation coefficients were calculated from SV profiles using the program SEDFIT (53). The continuous c(s) distributions were calculated assuming a direct sedimentation boundary model using the Lamm equation with maximum entropy regularization at a confidence level of 1 S.D.
For SE, the sample sector of dual-sector centerpieces was filled with 140 -180 l of protein (0.5-14 M), and the reference sector was filled with 150 -190 l of PBS. Each SE experiment was conducted at three or four speeds (8,000 -24,000 rpm) at 20°C, increasing from the lowest to the highest speed. SE experiments of SPI 1 -MSP␣␤ association were determined at 4°C because analysis of the SV experiments, followed by SDS-PAGE, indicated occasional limited degradation of the protein mixture at 20°C. Equilibrium was considered as reached when the RMSD value of successive scans taken at 3-h periods was below the noise level as determined by SEDFIT. Absorbance was scanned at a wavelength interval of 0.001 cm with 20 replicates/step. The SE curves were analyzed using the non-linear regression analysis program SEDPHAT to obtain the K D , based on the Boltzmann distributions of ideal species in the centrifu-gal field (54). The SDS-PAGE and Western blotting assays under non-denaturing and denaturing conditions were used to evaluate protein integrity at the end of SV and SE experiments. The density and viscosity of buffers at 20°C and 4°C were calculated using SEDNTERP (55). The partial specific volumes of glycosylated proteins were calculated as published (56). The structure-based hydrodynamic properties of proteins were calculated using the bead shell-modeling program HYDROPRO (57). The c(s) distributions and SE profiles were prepared with the program GUSSI (C. A. Brautigam, University of Texas Southwestern Medical Center).

RESULTS AND DISCUSSION
The RON-MSP interaction was investigated using biophysical and structural approaches to shed light on the receptorligand specificity in this and related systems. The characterization of moderately to strongly binding complexes provided explanations for the MSP specificity to RON relative to other receptors, which support and expand our understanding of RON signaling from previous investigations (32-36, 40, 41, 58). The crystal structure revealed the detailed receptor-ligand interactions within a 1:1 complex. The AUC analysis examined the stoichiometry of the interaction between RON SPI 1 and MSP␤ at lower concentration than that used in the crystallization, and also examined the interaction with the full-length mature MSP␣␤. The AUC showed the same 1:1 stoichiometry for the RON SPI 1 -MSP␤ complex as in the crystals. In contrast, in the presence of MSP␣␤, the majority of the complexes exhibited the 1:1 stoichiometry but also revealed a minor higher stoichiometry species, relevant to the physiological function of RON in signal transduction.
Structure Determination-Data processed at 2.8 Å resolution in space group P222 showed systematic intensity absences along all principle axes, consistent with space group P2 1 2 1 2 1 . However, the intensity statistics indicated a possible twinning by pseudomerohedry (͉E 2 Ϫ 1͉ ϭ 0.621, L test ϭ 0.377). Molecular replacement in space group P2 1 2 1 2 1 identified a single RON Sema and a single MSP␤ with Z-scores for the rotation and translation functions of RFZ ϭ 10.1 and TFZ ϭ 19.2 for Sema, and RFZ ϭ 6.8 and TFZ ϭ 28.2 for MSP␤. Thus, refinement commenced using the data processed in the P2 1 2 1 2 1 space group. Once it became clear that the refinement was not progressing, the data were reprocessed in space group P1, yielding unit cell dimensions of a ϭ 63.9 Å, b ϭ 107.1 Å, c ϭ 147.5 Å, ␣ ϭ 90.1°, ␤ ϭ 90.1°, ␥ ϭ 90.1°(i.e. all crystal cell angles were close to 90°). Next, the 2.8 Å resolution data were processed in space group P2 1 , using each of the principle orthorhombic cell axes as the potential unique monoclinic b axis. The resulting three R merge values were 0.195, 0.202, and 0.212 for the choice of the orthorhombic unit cell a, b, and c, respectively. The high R merge values may be attributed to the decrease in diffraction intensity below 3.0 Å resolution (͗I/I͘ Ͻ 1.5). All three data sets yielded molecular replacement solutions with two complexes in the asymmetric unit, which exhibited non-crystallographic 2-fold screw symmetry along the corresponding non-unique crystal axes. Refinement was carried out at 3.0 Å resolution using the data sets that yielded the two better R merge values. The correct unit cell choice was determined to be a ϭ 106. 6 0°based on the packing of molecules in the asymmetric unit. The correct crystal cell parameters exhibited identical RON SPI 1 -MSP␤ interfaces of the two complexes in the asymmetric unit. In contrast, one of the complexes in the incorrect choice of unique unit cell axis had small but systematically longer distances between interacting receptor and ligand residues. Presumably, this distortion was introduced by the incorrect assignment of two unit cell angles, the one assigned exactly 90°and the second slightly different from 90°. It should be emphasized that the structure of individual molecules remained the same in both the correct and incorrect cell units; only the packing of the molecules in the crystals was subtly different. Table 1 provides the data processing and refinement statistics.
Finally, through the entire study period, extensive attempts to improve the crystals were unsuccessful. Nevertheless, the quality of the structure reported here is sufficient to shed light on the biologically important questions.
Structure of RON SPI 1 -The biologically active MSP is generated by proteolytic cleavage at Arg 483 -Val 484 in the linker region between the ␣and ␤-chains, yielding the disulfidelinked MSP␣␤. The single-chain MSP␤ construct used in the crystallization included the uncleaved 19-amino acid linker region (Cys 468 -Arg 483 ) between the ␣and ␤-chains, ensuring that all of the cysteine residues in MSP␤ form disulfide bonds. Previous SPR studies showed that this single-chain MSP␤ exhibited similar binding affinities to immobilized RON ectodomain variants of increasing length (Sema, SP, SPI 1 , and SPI 4 ), indicating that only the RON Sema contributes to the binding affinity between MSP␤ and RON ectodomain (40). MSP␣␤ SPR binding experiments to the immobilized RON ectodomain constructs produced the same result. 3 This is consistent with the lack of strong binding between MSP␣ and RON ectodomain (33,34,40).
The association of single-chain MSP␤ with SPI 1 was sufficiently tight to yield crystals of the complex, but our attempts to obtain crystals of two-chain MSP␤ (cleaved at Arg 483 -Val 484 ) with either single-or two-chain SPI 1 were unsuccessful. In retrospect, the uncleaved regions of MSP␤ and SPI 1 are involved in crystal contacts, and the cleavages in these loops might have prevented the formation of these crystal-packing interactions. Table 1 summarizes the crystallographic data for the SPI 1 -MSP␤ structure. The RON SPI 1 model includes residues Gln 28 -Glu 683 . No electron density is associated with RON residues 25-27 at the N terminus; residues 358 -360 of Sema; and residues 582-583, 598 -602, 621-633, and 646/647-651 of IPT 1 , and these are omitted from the model. The electron density map revealed N-glycosylation at one of the five predicted sites on SPI 1 , enabling model building of a GlcNAc␤-(1,4)GlcNAc unit at Asn 488 (Sema) of one molecule in the asymmetric unit and a Man␤(1,4)GlcNAc␤(1,4)GlcNAc of the second Asn 488 in the asymmetric unit. For the free RON crystal structure, the cleavages of 17 N-terminal amino acids and the C-terminal half of IPT 1 occurred under the crystallization condition (39). When bound to MSP␤, both of these regions remained intact ( Fig. 2A), providing the first view of the RON IPT 1 domain and its spatial relationship to the Sema and PSI domains. An interdomain disulfide bond between the N-terminal Cys 29 of Sema and Cys 590 of IPT 1 tethers the Sema, PSI, and IPT 1 domains and restricts domain flexibility ( Fig. 2A). The intact N-terminal polypeptide meanders adjacent to the ␤-strands 6D and 6E of Sema, disrupting the hydrogen bond interactions between these ␤-strands, as observed previously in the free RON SP structure (39). In contrast, MET lacks an equivalent interdomain Cys 29 -Cys 590 disulfide bond despite the amino acid sequence conservation (Cys 26 and Cys 548 in the MET numbering system) (59). Instead, Cys 26 is disordered, and Cys 548 is unpaired in the MET SPI 2 /InlB complex structure (59). Niemann (25) had suggested an alternative interdomain disulfide bond between Cys 26 and the non-conserved Cys 800 of MET IPT 3 . Because RON IPT 3 does not have an equivalent cysteine, this Cys 29 -Cys 590 disulfide bond is proposed as the physiological interdomain linkage for RON. In addition, the cysteine residue pattern differs in the distinct extrusion regions of RON and MET Semas, resulting in two disulfide bonds in RON and one disulfide bond in MET (Fig. 1A). The remaining 22 cysteine residues of RON and MET SPI 1 form conserved intradomain disulfide linkages (39). In the free RON SP structures, the ␤4D-␤4DЈ Sema loop containing the proteolytic maturation site is disordered in both intact and cleaved proteins (39). This ␤4D-␤4DЈ loop adopts a defined conformation in the complex, stabilized by crystal contacts. The physiological relevance of this loop conformation is uncertain because of its involvement in crystal contacts and the introduction of mutations that replaced the furin-specific sequence by a thrombin cleavage sequence.
Comparison of the free and MSP␤-bound structures reveals that the RON PSI motif is oriented differently with respect to the Sema (Fig. 2B). This may be significant because PSI motifs typically serve as linkers that orient the flanking domains for where F o and F c are the observed and calculated structure factors, respectively. d R free is computed from 1,889 randomly selected reflections that were omitted from the refinement.
interactions with different partner proteins (60). Previously, rigid body rotations of PSI with respect to Sema have been noted in MET SP-HGF␤ and MET SPI 2 -InlB structures, in which the PSI relative orientations differed by ϳ60° (59). Rigid body rotation analysis of the two RON structures, using the program DynDom (61), showed a ϳ45°rotation around an effective hinge axis running parallel to residues 525-527 connecting the Sema and PSI domains (Fig. 2B). This domain motion resulted in the closure of the Sema-PSI interface, doubling the buried surface area upon closure from ϳ385 Å 2 in the free RON SP structure to ϳ820 Å 2 . The 17 N-terminal residues, previously missing in the RON SP structure, also contribute to the Sema-PSI contacts ( Fig. 2A). In light of the conformational restriction imposed by the Cys 29 -Cys 590 interdomain disulfide bond and the contacts between the N-terminal region of Sema and IPT 1 , it remains to be seen whether and how the PSI motif still functions in the mechanism as a hinge that induces tyrosine phosphorylation at the RON cytoplasmic kinase domain following MSP binding to the ectodomain. RON IPT 1 belongs to the early Ig-like (E-set) IPT/TIG domain superfamily (Figs. 1A and 2C). Members of this superfamily usually mediate protein-protein interactions. The six core ␤-strands of IPT 1 , arranged in the order ABE and CFG, form an antiparallel two-layer ␤ sandwich. Despite their low (20%) amino acid sequence identity, DALI analysis (62) revealed that the RON and MET IPT 1 are the closest structural homologs (PDB code 2UZX, Z ϭ 9.5, RMSD ϭ 2.5 Å, for 80 paired C␣ atoms) (Fig. 2C). Superposition of the Semas of RON SPI 1 and MET SPI 1 structures showed an ϳ16-Å shift in the positions of the respective IPT 1 domains (Fig. 2D). These structural differences may contribute to their functional specificities. For example, the region surrounding the ␤IBЈ-␤IBЉ hairpin loop of MET IPT 1 is the primary ligand binding site for the bacterial invasion protein, InlB, which uses MET as a specific cell surface receptor (59). The analogous RON IPT 1 region differs in both sequence and size from that of MET in that it is larger than MET by 21 amino acids, which are inserted into three loops (Lys 625 -Asp 634 in ␤ID-␤IE, Gly 651 -Thr 653 in ␤IE-␤1F, and Pro 663 -Val 670 in ␤IF-␤IG) (Fig. 1A). Of these loops, the first two loops are structurally disordered, but the ␤IF-␤IG loop is ordered and interacts with the PSI motif, burying ϳ430 Å 2 surface area. The biological activities associated with several RON splice variants suggest that RON IPT 1 modulates proteinprotein interactions. Of the four RON IPT domains, IPT 1 is most frequently subjected to alternative splicing and proteolysis events, profoundly affecting RON signaling activities (12). When the entire IPT 1 domain is deleted, the resulting RON⌬160 splice variant spontaneously dimerizes on the cell membrane and gains a constitutive phosphorylation activity (35). The RON⌬160 ectodomain lacks the interdomain Cys 29 -Cys 590 disulfide bond and therefore may be more flexible and able to adopt a conformation that allows dimerization without bound MSP. We had proposed that the ligand-independent dimerization of RON⌬160 may be mediated via the Sema/Sema interface, previously observed in the free RON SP structure (39). In addition, the RON⌬110 splice variant, which comprises only part of the IPT 1 followed by IPT 2-4 and the cytoplasmic kinase domain, also exhibits constitutive transphosphorylation activity (11,12). Moreover, the RON E5/6in splice variant, encoding a 20-amino acid insertion in the IPT 1 domain, introduces another level of functional regulation by proteolysis. This variant requires MSP binding for activation, but cleavage within the inserted region generates the constitutively active RON⌬110. Together, these constitutively active RON variants suggest that IPT 1 plays a role in regulating ligand-dependent dimerization of the receptor.
Structure of MSP␤-For consistency, we follow the structural unit assignments previously used to describe the structure of two-chain MSP␤, which included the 19-amino acid linker region cleaved at Arg 483 -Val 484 (41). As the two-chain MSP␤, the single-chain MSP␤ adopts the classic chymotrypsin-like serine protease fold ( Fig. 2A). There is no electron density for the N-terminal residues of the ␣␤ linker residues 465-467, residues 545-548, and the entire L8 loop (residues 608-615/616, including the N-glycosylation site at Asn 615 ; Fig. 1B). Superposition of the single-and two-chain MSP␤ structures reveals only minor conformational changes, primarily in loop regions (L4, L5, L10, L11, and L13), yielding RMSD of 0.7 Å for 205 paired C␣ atoms (Fig. 2E). However, there is a dramatic conformational change associated with the proteolytic cleavage at Arg 483 -Val 484 , resulting in the rearrangement of the 19-residue ␣␤ linker region. The N-terminal residue Val 484 , generated from the cleavage at the Arg 483 -Val 484 peptide bond, inserts into a pocket buried under the L8 loop (41). By contrast, the  OCTOBER 24, 2014 • VOLUME 289 • NUMBER 43 intact linker in the single-chain MSP␤ is fully solvent-exposed and interferes with the placement of the L8 loop, leading to its disorder (Fig. 2E).

Structural Studies of RON-MSP Complex
RON SPI 1 -MSP␤ Interface-There are two RON SPI 1 -MSP␤ complexes in the asymmetric unit, and the alignment of the two copies yielded an RMSD of 0.17 Å for 861 paired C␣ atoms. The complementing receptor-ligand binding interface spans the ␤3A-␤3B, ␤3C-␤3D, and ␤4C-␤4D loops of Sema and the ␣1, L4, L6, L10, L11, and L13 regions of MSP␤ (Figs. 1 (A and B) and  3 (A and B)). Using the program PISA (48), the average buried surface areas of the two complexes in the asymmetric unit are ϳ898 and ϳ874 Å 2 for Sema and MSP␤, respectively. The total buried surface area of ϳ1770 Å 2 engages 26 or 28 residues of the 2 Sema molecules in the symmetric unit and 25 or 26 residues of the 2 MSP␤ molecules. The molecular contacts include multiple salt bridges and hydrogen bonds as well as hydrophobic and van der Waals interactions (Fig. 3A). The RON Sema-MSP␤ interface involves the MSP␤ Arg 683 residue previously identified by the site-directed mutagenesis studies as essential for RON receptor recognition (34). It is also consistent with the prediction by Carafoli et al. (41) based on the homology to the MET SP-HGF␤ structure (58).
The degenerate serine protease active site cleft of MSP␤ comprises the center of the receptor recognition surface, with the protruding ␤3A-␤3B hairpin loop of Sema inserted into the MSP␤ cleft (Fig. 3A). Two MSP␤ arginine residues (Arg 521 and Arg 683 ) in the ␤-barrel subdomains flanking the serine protease cleft are embedded in the receptor-ligand interface (Fig. 3A).  (34), supporting our conclusion that the RON-MSP interface in the crystal structure is physiologically relevant. Moreover, the NH 2 group of RON Gln 193 interacts with the carboxylate groups of MSP Glu 644 as well as with the backbone carbonyl of MSP Arg 639 (the latter is not shown). These molecular contacts suggest that the buried Gln 193 of RON-Sema plays a crucial role in ligand recognition. In addition to interacting with MSP Arg 683 , the carboxylate group of RON Glu 190 forms a salt bridge with the guanidinium group of MSP Arg 639 , located at the interface periphery (Fig. 3A). Finally, an aromatic interaction occurs between the side chains of two histidine residues, RON His 424 and MSP His 528 . These histidines are probably uncharged because the crystals were obtained at pH 8.5. Their imidazole groups stack face-to-face at the periphery of the Sema-MSP␤ interface, similar to interactions found in other crystal structures (63).
Structural Basis for Receptor-Ligand Specificity in RON-MSP␤ and MET-HGF␤-Despite the common recognition surfaces, the RON-MSP␤ and MET-HGF␤ interfaces differ in details, which explains the unique specificity and lack of crossreactivity of these binding partners. The locations of the receptor-ligand interfaces in RON SPI 1 -MSP␤ and MET SP-HGF␤ are approximately the same (Fig. 3, B-D). Both interfaces bury a total of ϳ1700 Å 2 of surface area involving ϳ50 amino acids. The local density for RON-MSP␤ and MET-HGF␤ complexes is also similar at ϳ37, calculated using the program ProFace (49). This value falls within the local density values of 42 Ϯ 6, reported for specific protein-protein interfaces (64,65). Superposition of the MSP␤ and HGF␤ in the two complex structures highlights the differences (RMSD of 0.97 Å for 182 paired C␣ atoms) (Fig. 3E). The most striking feature is the projections of the ␤3A-␤3B hairpin loops of RON and MET Sema into their respective ligands. Due to the different length and functionality of the amino acids, the RON ␤-hairpin (colored magenta in Fig.  3E) projects more deeply into the MSP␤ cleft than the corresponding MET ␤-hairpin into the HGF␤ cleft (colored green). The deeper projection of the RON ␤-hairpin loop may be attributed to its smaller amino acids (Gly 192 and Gln 193 ) compared with those on the MET ␤-hairpin (Asp 190 and Arg 191 ).
The key discriminating amino acids on the respective ligands are MSP␤ Gln 568 and HGF␤ Asp 578 (Fig. 3E). MSP␤ Gln 568 would clash with MET Arg 191 , whereas MET Arg 191 forms a salt bridge with the shorter Asp 578 of its own ligand, HGF␤. Conversely, if the MET ␤-hairpin were to adopt the same conformation as RON ␤-hairpin, the side chains of MET Asp 190 and Arg 191 , which are larger than their RON counterparts (Gly 192 and Gln 193 ), would clash with MSP␤ Gln 568 and the backbone and side chain of Arg 639 .
Additionally, MSP␤ contains two more residues (Ser 526 -Cys 527 ) in its L4 loop when compared with the same loop of HGF␤ (Fig. 1B). The Cys 527 in MSP␤ forms a disulfide bond with the Cys 562 on ␤6, whereas such a disulfide bond is absent in HGF␤. Consequently, the L4 loops of MSP and HGF adopt entirely different conformations. The MSP␤ L4 loop conformation enables the stacking of the MSP␤ His 528 against RON His 424 (Fig. 3A) and an interaction of the backbone amide of Cys 527 with the carboxylate group of RON Glu 289 (not shown). This imidazole ring stacking may also be a RON-MSP selectivity determinant because RON His 424 is located on the ␣Ex2 helix of the extrusion region (residues 371-429). The structural integrity of the RON extrusion region is maintained by two adjacent disulfide bonds (Cys 385 -Cys 407 and Cys 386 -Cys 422 ) (Fig. 1A). By contrast, the extrusion loop of MET is partially disordered in both MET-HGF␤ and MET-InlB structures (58,59). Thus, the different folds adopted by respective extrusion regions of the MET and RON structures suggest distinct functional roles (39,58).
Structure-based sequence alignments of MSP␤, HGF␤, and plasmin showed that MSP␤ contains two clusters of triple arginine residues in the L10 (Arg 637 , Arg 639 , Arg 641 ) and L13 (Arg 683 , Arg 687 , Arg 689 ) loops (41). The authors proposed these arginine-rich regions as the specificity determinants of RON-MSP recognition. Of the 6 arginine residues, only Arg 683 on MSP␤ L13 is fully embedded in the interface (Fig. 3E), yet Arg 683 is unlikely to be a specificity determinant because the HGF␤ counterpart is also an arginine (Arg 695 ). MSP Arg 687 and Arg 689 are located remotely from the RON-MSP interface. The  OCTOBER 24, 2014 • VOLUME 289 • NUMBER 43

JOURNAL OF BIOLOGICAL CHEMISTRY 29955
caveat is that MSP L13, including Arg 687 and Arg 689 , is involved in an interaction with a symmetry-related RON Sema that generates an entirely different Sema-MSP␤ interface. This crystal contact also involves the intact ␤4D-␤4DЈ maturation loop of RON and the uncleaved linker region of MSP␤, and therefore might not reflect interactions within the physiological complex. In addition, because the L13 loop is located on the same face of MSP␤ as the ␣␤ linker, it may mediate interactions between the ␣ and ␤ domains of MSP rather than interaction with RON. Nevertheless, the possibility of conformational transition of these arginine residues in the L13 loop upon binding to RON receptor in solution cannot be ignored. For L10 arginine residues, Arg 637 is conserved in HGF (Arg 647 ). Arg 639 (Lys 649 in HGF) forms a salt bridge with RON Glu 190 (Val 188 in MET) and may be involved in ligand-receptor selectivity. Arg 641 of MSP␤ interacts with the backbone oxygen of RON Gly 192 in one com-plex of the asymmetric unit but is disordered in the second complex, suggesting that this is not a key interaction.

RON-MSP Interaction in Solution-
The AUC studies complement the crystallographic studies by investigating whether the protein partners can form complexes with a stoichiometry higher than 1:1, as observed in the crystal structure. Although crystals were only obtained with RON SPI 1 and MSP␤, the receptor-ligand interactions in solution were characterized with both MSP␤ and full-length MSP␣␤. Analogous studies performed with MET and HGF showed a 2:2 MET SP-HGF␣␤ stoichiometry in solution (66). Likewise, the MET SP-HGF␤ crystal structure exhibited only a 1:1 complex (58).
The SV and SE experiments revealed that single-chain MSP␤, two-chain MSP␤, MSP␣, pro-MSP, MSP␣␤, and RON SPI 1 exist predominantly as monomers in solution (Fig. 4,  A-H). The c(s) distribution profile of each protein showed a The dashed lines correspond to the SV profiles of free proteins. Inset, the s w (c) isotherm derived by integration of c(s) profiles. Fits for a 1:1 heterodimer association were calculated with hydrodynamic constraints (s-value of 6.2 S for the complex). The calculated K d value from the nonlinear least square analysis shown in the inset was 0.28 M, which reflects data from two independent sets of experiments, distinguished by squares and circles. J, SE profiles of 1 M equimolar SPI 1 /MSP␤ mixture with a best fit RMSD of 0.0039 AU, collected at 8,000-, 12,000-, 16,000-, and 21,000-rpm rotor speeds. Solid lines, calculated global best fit distributions using an A ϩ B 7 AB model with mass conservation. The c(s) distributions were normalized by dividing all c(s) values by the total absorbance present in the sample. All SE profiles were globally analyzed using a single species of interaction system with mass conservation. The best fits are shown as black solid lines through the experimental data. The combined residuals in AU from the same cell at different rotor speeds are shown below the plot. major symmetric peak with experimental weight average sedimentation coefficient (s 20,w ) that was consistent with the calculated value (Table 2) (53,57). Moreover, the SE profiles of free RON and MSP proteins were best fitted by a monomeric species model, confirming the SV results (Table 2). With the exception of MSP␤, small amounts of higher order aggregates were detected in these samples (4 -7% of the RON SPI 1 at ϳ7.3-8.8 S, ϳ9% MSP␣ at ϳ6 -8 S, and ϳ3-7% pro-MSP and MSP␣␤ at ϳ8.5-9.5 S). The amount of aggregates was independent of protein concentrations, indicating that they are probably irreversibly associated oligomers (data not shown). MSP␣ exhibited a broader sedimentation boundary with an experimental f/f 0 Յ 1 (Fig. 4C), characteristic of protein heterogeneity (67). Yet a monomer model best fits the SE profiles of 10 M MSP␣ (Fig. 4D), whereas a monomer-dimer model yielded a poorer fit with a low molecular mass of 46,470 Da (data not shown). Pro-MSP and two-chain MSP␣␤ displayed Յ0.2 S unit difference in their s 20,w values ( Table 2), suggesting only a limited conformational change from the proteolytic maturation event. This conclusion is supported by identical elution profiles of pro-MSP and MSP␣␤ from the Superdex 200 HR size exclusion column, presumably due to similar radii of gyration. 3 By contrast, a difference of ϳ0.8 S unit and a 30 -44-Å increase in the radius of gyration were observed for the closed and intermediate open forms of plasminogen, a homologue of MSP and HGF (68). Interestingly, unlike a monomeric MSP␣␤, the fulllength HGF␣␤ readily forms dimers and tetramers and exists only as monomers in the presence of 1 M NaCl (69). A preliminary SV analysis of 3.3 M RON SPI 4 (108,971 Ϯ 366 Da) also revealed a monomeric protein with an s 20,w of 5.5 S (f/f 0 ϭ 1.3-1.5), comparable with the value of 5.2 S obtained for the 101.5-kDa MET SPI 4 monomer (59).
The stoichiometry of the SPI 1 -MSP␤ complex was examined by SV using a 1 M single-chain MSP␤ in the presence of 0.12-5.0 M SPI 1 . At excess MSP␤, the c(s) distributions showed only two peaks corresponding to free MSP␤ and the 1:1 SPI 1 -MSP␤ complex at 6.1 S (Fig. 4I), in agreement with the structure-based s 20,w of 6.22 S. In the presence of excess SPI 1 , the 6.1 S peak shifted gradually toward the free SPI 1 peak, indicating a relatively fast dissociation (k off Ͼ 0.001/s) of the SPI 1 -MSP␤ com-plex. The weight average sedimentation coefficient s w (c) isotherm of SPI 1 -MSP␤ was obtained by integrating the 3-6.5 S peaks based on the mass-balance conservation (Fig. 4I, inset). A nonlinear least squares analysis of s w (c) using a heteroassociation model (A ϩ B 7 AB) gave an equilibrium dissociation constant (K D ) of ϳ0.28 M with a fixed S AB of 6.2 S. Analyses of SE profiles of SPI 1 /MSP␤ mixtures confirmed the SV results (Fig. 4J), in that they were also best fit globally to the same model of a 1:1 complex with K D of ϳ0.15 M. The differences in K D values derived from the SE and SV experiments are within experimental error (70).
The c(s) distributions for the biologically active MSP␣␤ and SPI 1 showed a major species at 7.7-8.15 S (Fig. 5A), consistent with the calculated s 20,w of 8.1 S for a 1:1 SPI 1 -MSP␣␤ complex ( Table 2). However, ϳ3-8% of the total signal in these experiments resolved as 9.5-10.5 S species (Fig. 5A), which may correspond to a higher state of receptor-ligand association. By contrast, the SPI 1 -MSP␤ samples did not reveal any higher order species under similar protein concentrations (Fig. 4I). A complementary SE experiment of SPI 1 -MSP␣␤ association was conducted to determine the stoichiometry of this higher order protein complex (Fig. 5, B-D). Initial analysis of the data (2 M equimolar) showed a poor fit to a simple (A ϩ B 7 AB) model (Fig. 5E). Therefore, the SE profiles were analyzed using two more complex models wherein the reactants reversibly associate to form a complex with either a 2:1 (2A ϩ B 7 AB ϩ A 7 A 2 B) or 2:2 (2A ϩ 2B 7 2AB 7 (AB) 2 ) stoichiometry, where A corresponds to SPI 1 and B corresponds to MSP␣␤. These models were considered probable based on knowledge of the stoichiometry of receptor-ligand complexes involved in other signal transduction pathways. The analyses yielded much better fits with either model compared with the 1:1 association model as evidenced by the distribution of residuals (Fig. 5, C-E). The dissociation constants for the 2:1 association model yielded similar dissociation constants of K d1 ϳ0.2-0.3 M and K d2 ϳ0.02-0.9 M for the 1:1 and 2:1 adducts, respectively (ranges obtained from three independent experiments). For the 2:2 SPI 1 -MSP␣␤ model, the K d1 and K d2 values were ϳ0.1-0.2 and ϳ2-36 M, respectively. Both models gave K d1 values that were consistent with the SPI 1 -MSP␤ dissociation constant. However, in contrast to the 2:1 association, the binding affinity of the 2:2 species is at least 10-fold weaker than that of the 1:1 species, consistent with the SV experiments showing predominantly the 1:1 species and only minor higher oligomeric species. Species population analysis supports the conclusion that the 2:2 complex comprises the high oligomeric species because this model predicts that, as observed by SV (Fig. 5A), the 1:1 species predominates over the entire experimental concentration range (Fig. 5F). In contrast, the alternative 2:1 association model predicts that the populations of the 1:1 and 2:1 species change with protein concentration (Fig. 5G), which is not supported by the SV experiments.
The simplest interpretation of the combined SPI 1 -MSP␤ and SPI 1 -MSP␣␤ ultracentrifugation experiments is that the ␣-chain of MSP␣␤ mediates the dimerization of the RON receptor. The weak binding affinity of the 2:2 SPI 1 -MSP␣␤ complex may be physiologically relevant because of the transient nature of the signal transduction process.
Conclusion-Comparison between the crystal structures of RON SPI 1 -MSP␤ and MET SP-HGF␤ explains the origin of receptor-ligand selectivity. Despite their identical domain architecture and their 45% amino acid sequence identity, it has been known for many years that MSP and HGF exhibit strikingly distinct binding properties to their respective receptors. Pro-MSP does not bind to RON and MSP␣ binds at best weakly to RON. Only activated MSP binds to RON, interactions that are mediated primarily if not solely by the ␤-chain. In contrast, pro-HGF and HGF␣ bind to MET with high affinities, although the bound pro-HGF does not activate MET (71,72). Sequence alignment reveals that the interdomain linker regions of the ␣-chains of MSP and HGF vary in length and in amino acid sequences, enough to allow different orientations of the respective N domain and four Kringle domains. Thus, one would expect the spatial arrangement of the N domain and Kringle domains in pro-MSP to hinder the interaction between MSP␤ and RON Sema, hindrance that is removed upon proteolytic conversion into MSP␣␤.
The solution properties of single-chain pro-MSP and MSP␣␤ indicated that both forms retained similar overall dimensions, suggesting that the conformational transition accompanying the MSP maturation is subtle. In contrast, the small angle x-ray scattering and electron microscopy studies of HGF revealed a compact pro-HGF molecule and an elongated, biologically active HGF␣␤ (66). Because pro-HGF binds to MET, it is tempting to speculate that the protein undergoes conformational transition to adopt the elongated shape in the presence of MET, which enables ligand-receptor binding. However, the binding to MET is insufficient for function, and receptor activation requires the proteolytic cleavage, which may be accompanied by additional structural adjustments.
The AUC results showed that MSP␣␤ could facilitate the dimerization of soluble RON SPI 1 -MSP␣␤ complex with 2:2 stoichiometry. This is similar to the dimerization of MET induced by HGF␣␤ binding, proposed to occur via a ligandmediated interface (66). The molecular details for the HGF␣␤induced MET dimerization are not yet fully resolved because the 2:2 complex seen in the combined small angle x-ray scattering and electron microscopy studies was observed only with MET SP and not with MET SPI 4 . Perhaps the MET must be anchored to the membrane for an appropriate orientation of its  ). B, SE profiles of a 4.5 M SPI 1 and 3.8 M MSP␣␤ mixture collected at 8,000-, 12,000-, and 18,000-rpm rotor speeds at 4°C and analyzed globally using the 2:1 association model described under "Results and Discussion," with mass conservation, which yielded K d1 ϭ 0.45 M, K d2 ϭ 0.35 M, an overall reduced 2 ϭ 0.85, and RMSD ϭ 0.005 AU. C, the combined residuals for a fit to a 2:1 association model. D, the combined residuals for a fit to a 2:2 association model described under "Results and Discussion," which yielded K d1 of 0.16 M and K d2 ϭ 13.8 M, 2 ϭ 1.01, RMSD ϭ 0.005 AU. E, the combined residuals for a fit to a 1:1 association model described under "Results and Discussion," which yielded K d ϭ 0.0004 M, 2 ϭ 3.23, RMSD ϭ 0.009 AU. F, SPI 1 -MSP␣␤ species distributions calculated as a function of total protein concentrations using the 2:2 association model with K d1 ϭ 0.16 M and K d2 ϭ 13.8 M. G, SPI 1 -MSP␣␤ species distributions calculated as a function of total protein concentrations using the 2:1 association model with K d1 ϭ 0.16 M and K d2 ϭ 0.12 M.
IPT domains that promotes dimerization. Nevertheless, the best encasement model generated for the 2:2 MET SP-HGF␣␤ involved the N and K 1 domains of HGF␣ (66). Currently, we do not know whether MSP␣ alone mediates the formation of a 2:2 RON-MSP␣␤ complex and, if so, which of the MSP␣ domains are involved. At the least, functional studies of MSP␣ domain mutants suggest that the MSP K 1-2 domains or the K 2 domain alone may be involved in receptor dimerization (32). In other words, the MSP mutants lacking these domains lost cellular activities, whereas mutants missing the single N, K 1 , K 3 , or K 4 domains still retained some biological activities.