Structural and mechanistic insights into VEGF ...

Structural and mechanistic insights into VEGF receptor 3 ligand binding and activation Veli-Matti Leppänena,1, Denis Tvorogova,1, Kaisa Kiskob, Andrea E. Protab, Michael Jeltscha, Andrey Anisimova, Sandra Markovic-Muellerb, Edward Stuttfeldb,c, Kenneth N. Goldied, Kurt Ballmer-Hoferb, and Kari Alitaloa,2 a

Wihuri Research Institute and Translational Cancer Biology Program, Institute for Molecular Medicine Finland and Helsinki University Central Hospital, Biomedicum Helsinki, University of Helsinki, 00014 Helsinki, Finland; bLaboratory of Biomolecular Research, Paul Scherrer Institute, CH-5232 Villigen PSI, Switzerland; cStructural Biology and Biophysics, Biozentrum, University of Basel, CH-4056 Basel, Switzerland; and dCenter for Cellular Imaging and NanoAnalytics, Biozentrum, University of Basel, CH-4056 Basel, Switzerland Edited by Napoleone Ferrara, University of California at San Diego, La Jolla, CA, and approved June 26, 2013 (received for review January 23, 2013)

Vascular endothelial growth factors (VEGFs) and their receptors (VEGFRs) are key drivers of blood and lymph vessel formation in development, but also in several pathological processes. VEGF-C signaling through VEGFR-3 promotes lymphangiogenesis, which is a clinically relevant target for treating lymphatic insufficiency and for blocking tumor angiogenesis and metastasis. The extracellular domain of VEGFRs consists of seven Ig homology domains; domains 1–3 (D1-3) are responsible for ligand binding, and the membrane-proximal domains 4–7 (D4-7) are involved in structural rearrangements essential for receptor dimerization and activation. Here we analyzed the crystal structures of VEGF-C in complex with VEGFR-3 domains D1-2 and of the VEGFR-3 D4-5 homodimer. The structures revealed a conserved ligand-binding interface in D2 and a unique mechanism for VEGFR dimerization and activation, with homotypic interactions in D5. Mutation of the conserved residues mediating the D5 interaction (Thr446 and Lys516) and the D7 interaction (Arg737) compromised VEGF-C induced VEGFR-3 activation. A thermodynamic analysis of VEGFR-3 deletion mutants showed that D3, D4-5, and D6-7 all contribute to ligand binding. A structural model of the VEGF-C/VEGFR-3 D1-7 complex derived from small-angle X-ray scattering data is consistent with the homotypic interactions in D5 and D7. Taken together, our data show that ligand-dependent homotypic interactions in D5 and D7 are essential for VEGFR activation, opening promising possibilities for the design of VEGFR-specific drugs. signal transduction

| receptor tyrosine kinase

V

EGFs stimulate angiogenesis and lymphangiogenesis via VEGF receptors (VEGFRs) in endothelial cells. VEGF-A signaling is mediated predominantly through activation of VEGFR-2, resulting in sprouting of blood vessels from preexisting vasculature (1). In contrast, VEGFR-1 seems to have an inhibitory role by sequestering VEGF-A and thereby preventing its interaction with VEGFR-2 (2). On the other hand, VEGFR-3 plays an indispensable role in lymphangiogenesis (3). VEGFRs are involved in various pathological conditions, including solid tumor growth, tumor metastasis, and vascular retinopathies (4, 5). VEGF-C and VEGF-D compose a VEGFR-3–specific subfamily of VEGFs. They are produced with large N- and C-terminal propeptides and gain activity toward VEGFR-3 and VEGFR-2 on proteolytic processing (reviewed in ref. 5). VEGFR-3 maturation involves proteolytic cleavage of the extracellular domain (ECD) in D5 (6–8). Both VEGF-C and VEGFR-3 also interact with the coreceptor neuropilin-2 (9). Loss of the Vegfc gene results in embryonic lethality owing to a lack of lymphatic vessel formation (10), whereas mutations that interfere with VEGFR-3 signaling have been associated with hereditary lymphedema, and mice deficient in Vegfr3 die in utero due to abnormal development of the blood vasculature (11, 12). VEGFR-3 and its heterodimers with VEGFR-2 are also important for sprouting angiogenesis and vascular network formation (13–15). VEGFRs are type V receptor tyrosine kinases (RTKs) composing a family of three transmembrane receptors, with the ECDs consisting of seven Ig homology domains (D1–D7). For ligand 12960–12965 | PNAS | August 6, 2013 | vol. 110 | no. 32

binding, VEGFRs use predominantly D2, with D3 providing additional binding affinity (16–18). In addition, D1 is required for VEGFR-3 ligand binding, but the exact role of this domain remains elusive (19, 20). VEGFRs and the closely related type III RTKs are activated by ligand-induced dimerization of the extracellular domain, followed by tyrosine autophosphorylation of the intracellular kinase domain to generate downstream signaling (21). As a prototype for these families, crystal structures of the Mast/stem cell growth factor receptor KIT ECD revealed ligandinduced homotypic interactions in D4 (22). Electron microscopy (EM), small-angle X-ray scattering (SAXS), and functional assays have established that VEGFR-2 D4-7 is essential for receptor activation by facilitating dimerization through homotypic contacts formed between D4-5 and D7 (23–26). In line with these findings, the crystal structure of a VEGFR-2 D7 homodimer reveals a pair of salt bridges in the juxtaposed loops between strands E and F (E-F loop), which are dispensable for receptor dimerization but not for receptor activation (24). Similar mutations in the E-F loop in D4 of VEGFR-2 do not affect VEGF-A-induced receptor activation (24), suggesting unique interactions in D4-5. Blockage of VEGFR function has been shown to inhibit tumor angiogenesis, lymphangiogenesis, and metastasis in several mouse models (4, 5). Identified inhibitors include antibodies that block ligand/receptor binding. We recently reported an antibody against VEGFR-3 D5 that did not block ligand binding but did inhibit VEGFR-3 homodimer and VEGFR-3/VEGFR-2 heterodimer formation and signal transduction (8). Similarly, an antibody and designed ankyrin repeat proteins (DARPins) targeted against D4-7 of VEGFR-2 were found to inhibit ligand-induced receptor activation, but not dimerization (26, 27), suggesting that such binders represent a new generation of highly specific inhibitors of VEGFR signaling. In this study, we provide the structural basis of ligand binding to D1-2 of VEGFR-3 and define a unique role of D4-5 for VEGFR dimerization and activation. Using receptor mutants, we show that homotypic interactions in D5 and D7 are essential for VEGFR-3 activation. Based on our data, we suggest a general mechanism for VEGFR dimerization and activation by ligand-induced homotypic interactions in D5 and D7.

Author contributions: V.-M.L., D.T., K.K., M.J., K.N.G., K.B.-H., and K.A. designed research; V.-M.L., D.T., K.K., A.E.P., A.A., S.M.-M., and E.S. performed research; V.-M.L., D.T., K.K., A.E.P., A.A., S.M.-M., and E.S. analyzed data; and V.-M.L., K.B.-H., and K.A. wrote the paper. The authors declare no conflict of interest. This article is a PNAS Direct Submission. Data deposition: The atomic coordinates and structure factors have been deposited in the Protein Data Bank, www.pdb.org (PDB ID codes 4BSJ and 4BSK). 1

V.-M.L. and D.T. contributed equally to this work.

2

To whom correspondence should be addressed. E-mail: Kari.Alitalo@helsinki.fi.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. 1073/pnas.1301415110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1301415110

Structure of the VEGF-C/VEGFR-3 D1-2 Complex. We expressed ma-

ture VEGF-C (C137A) and VEGFR-3 D1-2 in insect cells and crystallized the complex in space group I23 (a, b, c = 166.7 Å). The complex structure was solved at 4.2-Å resolution by molecular replacement using multiple isomorphous replacement with anomalous scattering (MIRAS) phases, and was refined to a crystallographic R value of 33.4% and a free R value of 37.1% (Fig. 1 and Table S1). The complex with WT VEGF-C was crystallized in the same crystal form, and cross-phasing at 6-Å resolution indicated identical packing and complex formation (Fig. S1A). The overall architecture of the VEGFR-3 complex is very similar to the previously reported VEGFR-1 (16, 28, 29) and VEGFR-2 (18, 30) structures except, that in the VEGFR-3 complex structure, D1 is resolved as well. VEGF-C binding is limited to D2, with D1 protruding away from VEGF-C. Residues 28–134 of D1 are largely disordered, and we could build only the core of the apparent I-type Ig domain using KIT D1 (PDB ID code 2E9W) as a model. Our structure confirmed the presence of a disulfide bridge between Cys51 and Cys111 and N-glycosylation of Asn33 and Asn104. Residues 135–224 of D2 represent a smaller I-type Ig domain with a disulfide bridge between Cys158 and Cys206 and N-glycosylation of Asn166. Confronted with the low resolution, we could not define interactions at the binding interface. The structure clearly shows packing of the VEGF-C N-terminal helix against strand E and the C/C′ hairpin loop of D2. To confirm that these interactions are functional, we generated a double mutant of VEGF-C, D123A/Q130A, for VEGFR-3 binding studies (Fig. S1B). Our structure shows that loop 2 (L2) interacts with the loop connecting strands E and F of D2 and L3 with strands A′ and G.

The VEGF-C/VEGFR-3 D1-2 and VEGF-C/VEGFR-2 D2-3 complex structures can be aligned with an rmsd of 2.1 Å for 336 Cαatoms (Fig. S2 A–C). The N-terminal helix of VEGF-C in the VEGFR-3 complex is bent ∼9° toward D2, but L1–L3 adopt approximately the same conformations as in the VEGFR-2 complex. Structure of VEGFR-3 D4-5. We expressed VEGFR-3 D4-5 in insect cells and crystallized the purified protein in space group P3121 (a, b = 133.3 Å; c = 48.9 Å). The structure of D4-5 was determined at 2.5-Å resolution using single isomorphous replacement with anomalous scattering (SIRAS) phases, and was refined to a crystallographic R value of 21.0% and free R value of 25.3% (Fig. 1, Table S1, and Fig. S3 A and B). D4, consisting of residues 330–419, is a small I-type Ig domain. Structural comparison shows that it is very similar to KIT D4; both domains can be aligned with an rmsd of 1.6 Å for 82 residues. However, unique to VEGFR-3 D4, strand D is missing, and strand C is connected directly to a short strand E on the opposite side of the molecule. D5, consisting of residues 420– 553, is a larger I-type Ig domain characterized by a single 310 helical turn between strands A and A′, long strands D and E, and a long C-D loop of approximately 35 residues. Unlike D4, D5 has low homology to the corresponding D5 of KIT, with an rmsd of 2.6 Å for 74 residues. The overall temperature factor (B-factor) for D4 is far larger for D4 than for D5 (69.0Å2 vs. 55.1Å2; Fig. S3C). D4 has no disulfide bridges, whereas D5 has a buried disulfide bridge between the two sheets and an exposed disulfide bridge between Cys466 and Cys486 in the C-D loop, in which proteolytic cleavage occurs between Arg472 and Ser473 (7). The crystallized VEGFR-3 D4-5 appears to be only partially cleaved (Fig. S3D). Residues 470–483 in the C-D loop were disordered and thus were omitted from the final model. VEGFR-3 D4-5 has an overall extended structure connected by the linker peptide, a salt bridge between Glu344 and Lys539, and a hydrogen bond between Glu391 and Tyr448 (Fig. S3E). VEGFR-3 D4-5 Homodimer and Homotypic Interactions in D5. The D4-5 structure revealed homotypic interactions in D5, covering a solvent-accessible area of ∼570 Å2 per chain, but not in D4, where in contrast, Arg378 and Lys387 in the adjacent D4 face each other, which may lead to domain repulsion (Fig. S3F). The dimeric D4-5 molecules have a V-shaped arrangement at an angle of ∼45° and C and N termini at distances of approximately 14 Å and 65 Å, respectively (Fig. S3 F and G). The homotypic contacts in D5 between the sheets comprised of strands A′, B, D, and E are mediated by the helical protrusion between strands A and A′ occupying the cavity between the curved D-E loop and strands A and A′ in the other chain (Fig. 2A). The interactions are centered on the fully conserved Thr446 and Lys516, which create hydrogen bonds with the backbone atoms of Ser430 and Glu428 in strand A′ of the other chain (Fig. 2 B and C). His425, Glu428, and Glu509 contribute additional ionic interactions. Val518 and Ser430, and Thr446 and Ala429 form van der Waals contacts. The VEGFR-3 D4-5 dimer structure agrees with the architecture of our previously reported ligand/VEGFR-2 D2-3 complex structures (PDB ID codes 2X1X, 3V6B, and 3V2A). Superposition of the VEGF-C/VEGFR-2 D2-3 and the VEGFC/VEGFR-3 D1-2 structures on the KIT/SCF complex (PDB ID code 2E9W) creates a model of a VEGF/VEGFR D1-5 complex (Fig. 1 and Fig. S2C). In this model, the N termini of VEGFR-3 D4 (Asn330) are located within a few Ångstroms of the C termini of VEGFR-2 D3 (Glu326, the VEGFR-3 Glu329 counterpart).

Fig. 1. Crystal structures of the VEGF-C/VEGFR-3 D1-2 complex and the homodimer of VEGFR-3 D4-5. Shown are surface and cartoon representations with the two chains of VEGFR-3 in slate blue or yellow and the two chains of VEGF-C in orange or light orange. The VEGFR-3 complex, the D4-5 homodimer and our previous VEGF-C/VEGFR-2 D2-3 complex (PDB code 2X1X) were superimposed to the KIT/SCF complex (PDB code 2E9W). VEGFR-2 D3 is in gray. VEGFR Ig domains 1–5 (D1–D5), VEGF-C loops 1–3 (L1–L3), and the N-terminal helix (αN) are labeled. Glycan moieties are shown as spheres.

Leppänen et al.

Homotypic Interactions in D5 and D7 Are Essential for Ligand-Induced VEGFR-3 Activation. To determine the functional significance of the

D5 and D7 interfaces in VEGF-C–induced VEGFR-3 activation, we generated a D5 double mutant, T446E/K516A (5EA), a D7 single mutant, R737A (7A) (24), and a D5/D7 triple mutant, T446E/K516A/R726 (5EA7A), of VEGFR-3 (Fig. 3A). To do so, we expressed WT VEGFR-3 and the mutant constructs in HEK293 cells and analyzed VEGF-C–induced VEGFR-3 PNAS | August 6, 2013 | vol. 110 | no. 32 | 12961

BIOCHEMISTRY

Results

region and the main body represents the dimerized membraneproximal domains. Using the available VEGFR-2 and VEGFR-3 crystal structures and a homology model for D6, we prepared a model of the dimeric, glycosylated VEGF-C/VEGFR-3 ECD complex by applying rigidbody refinement of the SAXS data. The models with the best fit revealed strong bending in domains D3-5 that would bring VEGFC closer to the plasma membrane (Fig. 5 C–E and Fig. S5C). We also visualized the complex by negative-staining EM. The data agree with our structural models and clearly identify D1 protruding away from the complex (Fig. 5F and Fig. S6 A and B). Comparison of selected 2D projections of the D1-5 complex model with the EM structures also revealed a strong similarity with ligand-bound D2-5, although the membrane-proximal part was only partially resolved.

Fig. 2. VEGFR-3 D5 homotypic interactions are centered to the conserved residues Thr446 and Lys516. (A) Homotypic interactions in D5. The two chains are shown in cartoon form, color-coded as in Fig. 1. Thr446, Glu509, Lys516, and their counterparts in strands A and A’ are shown as sticks. Hydrogen bonds are shown as red dashed lines. (B) D5, with conservation pattern in cyan through dark red for variable to conserved amino acids. Evolutionary rates of human, mouse, chicken, and zebrafish VEGFR sequences were were plotted using the ConSurf Web server. Thr444, Thr446, Glu426, and Lys516 compose a highly conserved patch on the D5 surface. (C) Representative amino acid sequence alignment from B. Residues involved in homotypic interactions are colored.

activation using receptor immunoprecipitation, followed by anti-phosphotyrosine Western blot analysis (Fig. S4A). To quantify tyrosine autophosphorylation of VEGFR-3, we compared the level of phosphorylation of the band with an apparent Mr of 120 kDa on Western blots, corrected for VEGFR-3 expression level. This band represents the fully mature form of the receptor displayed on the cell surface (Fig. 3B). Our results indicate that the combination of D5 and D7 mutations (5EA7A) impairs VEGF-C–induced VEGFR-3 activation. We performed a similar analysis in transfected porcine aortic endothelial (PAE) cells stably expressing the constructs (Fig. 3C and Fig. S4B). In these cells, both the D5 double mutant and D7 single mutant demonstrated reduced VEGFR-3 activity; however, effective VEGFR-3 blocking was obtained only with the triple mutant 5EA7A, suggesting the mutual importance of D5 and D7 interactions for VEGFR-3 activation.

Discussion Here we present structural data describing ligand binding to VEGFR-3 and propose a model for receptor activation. The structural data comprise crystal structures of a VEGF-C/VEGFR-3 D1-2 complex and the VEGFR-3 D4-5 homodimer, as well as an analysis of the complex by SAXS and EM. Our findings, complemented by a thermodynamic analysis of VEGF-C binding to VEGFR-3 and a cellular analysis of receptor constructs specifically mutated in D5 and D7, clearly show that the VEGFR-3 interactions observed in the crystal structures are functionally relevant for ligand-mediated dimerization and activation. The VEGF-C/VEGFR-3 D1-2 complex structure provides a reliable model of the overall architecture of the ligand–receptor complex. Comparisons with the VEGFR-1 and VEGFR-2 complex structures show that all three VEGFRs share a conserved interface in D2 for ligand binding. VEGF-D has an extended N-terminal helix responsible for high-affinity binding and activation of VEGFR-3 (20), and we show here that mutation of the N-terminal Asp123 and Gln130 of VEGF-C decreases receptor binding, supporting the idea that the helix is an important

Membrane-Proximal Domains of VEGFR-3 Provide Increased VEGF-C Affinity. To explore the effect of the membrane-proximal domains

of VEGFR-3 on VEGF-C binding, we generated C-terminal truncations of the soluble, monomeric VEGFR-3 ECD and measured VEGF-C binding by isothermal titration calorimetry (ITC). Our data show that VEGF-C binding to VEGFR-3 was enthalpically and entropically favorable, and that the presence of D3 and membraneproximal D4-7 increased VEGF-C affinity (Fig. 4). In addition, deletion of D6-7, D4-5, and D3 gave rise to a gradual increase in binding enthalpy (ΔH), suggesting that these domains form additional interactions with the ligand in the case of D3, or ligandinduced dimerization in the case of D4-7. In addition, the D5 mutant demonstrated decreased affinity and a larger ΔH (Fig. S4 C and D). SAXS and Single-Particle EM Analysis of the VEGF-C/VEGFR-3 ECD Complex. To confirm our structural models, we analyzed VEGFR-3

D1-7 and its complex with VEGF-C using multiangle laser light scattering (MALS) and SAXS (Fig. 5 A and B and Fig. S5 A and B). Our data show that the VEGFR-3 ECD is monomeric in solution and that dimerization is ligand-dependent. The SAXS-derived ab initio model of the complex revealed a bent, T-shaped molecular envelope, suggesting that the protrusions at the top of the molecules represent the D1 domains in the ligand-binding 12962 | www.pnas.org/cgi/doi/10.1073/pnas.1301415110

Fig. 3. Homotypic interactions in D5 and D7 are mutually important for ligand-induced VEGFR-3 activation. (A) Schematic presentation of the WT VEGFR-3, 5EA, 7A, and 5EA7A mutants of VEGFR-3. (B) Biotinylation of cell surface-expressed VEGFR-3 isoforms. Biotinylated PAE-VEGFR-3 cell lysates were immunoprecipitated with streptavidin beads and blotted for VEGFR-3 and HSC70. Total lysates of PAE and PAE-VEGFR-3 cells were used as controls. (C) PAE cells stably expressing the VEGFR-3 constructs were stimulated with VEGF-C, and the cell lysates were immunoprecipitated with anti–VEGFR-3 and analyzed for VEGFR-3 autophosphorylation (pY) and expression (R3) by Western blot analysis. The pY/R3 ratio was quantified based on the ∼120kDa band representing the fully processed VEGFR-3 (6, 7).

Leppänen et al.

Fig. 4. Thermodynamic analysis of VEGF-C binding to VEGFR-3. Calorimetric titrations of VEGF-C to the monomeric VEGFR-3 ECD (D1-7) and its domain deletion mutants D1-2, D1-3 and D1-5 are shown. The enthalpy change (ΔH), dissociation constant (Kd), and stoichiometry (N) of the ITC assays are indicated.

BIOCHEMISTRY

determinant of VEGFR-3 ligand specificity (Figs. S1B and S4C). Additional structural elements of VEGF-C, including loops L1, L2, and L3, are involved in VEGFR-3 binding as well. L1 likely interacts with D3 (18, 30).

Our data also provide insight into D1. Despite its importance in VEGFR-3 ligand binding (19, 20), D1 protrudes away from VEGF-C and has no connections with it. The bent conformation of the D1-2 module is very similar to that observed in KIT, platelet-derived growth factor receptor beta, and macrophage colony-stimulating factor 1 receptor (Fig. S2C) (22, 31, 32). So far, KIT seems to be the only type III/V RTK that directly uses D1 for ligand binding (22). D1 may have a positive impact on D2 structural stability, as demonstrated by the poor resolution of the first strand of D2 (βA) in the multiple VEGFR-2 D2-3 structures, which lack D1 (18, 30). On the other hand, VEGFR-3 D1 has large disordered regions in the ligand complex, suggesting that it may interact with other binding partners in vivo, such as neuropilin-2 or the large N- and C-terminal propeptides of VEGF-C and VEGF-D. VEGFR-3 maturation involves proteolytic cleavage in D5 (6, 7). This cleavage also occurred in the insect cell-expressed constructs, although the crystallized D4-5 was mainly uncleaved (Figs. S3D and S5A). The structure of the VEGFR-3 D4-5 homodimer suggests that proteolytic cleavage did not introduce significant changes in the D5 conformation, because it occurred in the C-D loop, which is disordered in the Cys466/Cys486 disulfide bridge. This loop is not involved in VEGFR-3 dimerization, leaving the function of this processing step elusive. Nonetheless, our structure suggests that the disulfide bridge restricts the flexibility of this loop, thereby increasing the stability of D5. Calorimetric data for VEGF-C binding to our VEGFR-3 ECD mutant constructs indicate direct interaction of D3 with the ligand. In contrast, the membrane-proximal domains D4-5 and D6-7 modulate ligand binding indirectly. The contribution of D4-5 suggests that the homotypic interactions in D5 are energetically favorable. D6-7 does not increase VEGF-C affinity further; however, the decrease in ΔH indicates additional homotypic receptor contacts. This explains why mutations in D5 and D7 showed an additive inhibitory effect on VEGF-C–stimulated receptor phosphorylation in both HEK293 and PAE cells. The mutant and the ITC data agree with the observed interactions in D5 and with the putative salt bridges between Arg737 and Asp742 (human VEGFR-3 numbering) in VEGFR-3 D7. Taken together, the data indicate that the observed homotypic interactions in the membrane-proximal domains of VEGFRs are functionally highly relevant, as has been shown previously for VEGFR-2 (24, 26) and here for D5 and D7 of VEGFR-3.

Fig. 5. Characterization of the VEGF-C/VEGFR-3 D1-7 complex in solution and in EM. (A) MALS analysis of the VEGF-C/VEGFR-3 D1-7 complex and VEGFR-3 D1-7. (B) The SAXS-derived distance distribution functions and averaged ab initio shape reconstructions of VEGFR-3 D1-7 and the VEGF-C/VEGFR-3 D1-7 complex. (C) Rigid-body models before and after refinement against the SAXS data were aligned using VEGF-C and are shown in cartoon form. The calculated (red) and experimental scattering curves (black) are compared. (I) The symmetrical model before refinement. (II) A representative model of the refinement with limited movement: D123C dimer–linker–D45 dimer–linker–D67 dimer. (III) Representative model of the refinement with increased movement: D12C dimer–linker–D3-D3–linker–D4-D4–linker–D5 dimer–linker–D67 dimer (SI Methods). (D) Rigid-body models from C aligned using the membrane-proximal D6-7 in vertical orientation and shown in a surface representation. (E) Overlay of the ab initio model of the complex from B and the final rigid-body model from C. (F) Representative class averages of the negative stain EM analysis of the VEGF-C/VEGFR-3 ECD complex. (Scale bar: 10 nm.)

Leppänen et al.

PNAS | August 6, 2013 | vol. 110 | no. 32 | 12963

VEGFRs are structurally highly similar and also signal as heterodimers (14, 33). Similar to the conserved D7 interface, the key residues involved in D5 dimerization (Thr446 and Lys516) are fully conserved, suggesting that similar D5 interactions exist in other VEGFRs. D4-directed homotypic interactions have been suggested previously based on the ligand-induced crossover in the middle of the EM structure of the VEGF-A/VEGFR-2 ECD complex (23). Considering the low resolution of the 2D projections of the EM data, these data are compatible with the involvement of D5 in receptor dimerization, which we report here for VEGFR-3. A previous binding study indicated that both D4 and D5 increase the affinity of VEGFR-2 for VEGF-A. Compared with D1-3, D1-5 showed 86-fold greater affinity and D1-4 showed 16-fold greater affinity (34). Similarly, both D4 and D4-5 of VEGFR-1 have been shown to enhance ligand-induced receptor dimerization in cross-linking experiments (35). Thus, the results of the VEGFR-1– and VEGFR-2–binding studies support a role for D5 in VEGFR dimerization, although additional interactions in D4 cannot be ruled out. D4 may have an additional functional role, as suggested by previous studies showing that a swap of D4 with another domain, but not mutation of the D4 E-F loops facing each other, impaired ligand-induced VEGFR-2 activation (24, 26). D4 may facilitate proper orientation of the ligand-binding D1-3 in the active complex. D4 in VEGFRs and in type III RTKs lacks the intradomain disulfides (24), which have been shown to improve Ig domain stability (36). Compared with D5, the lack of disulfides in VEGFR-3 D4 is reflected in the higher overall B-factor level (Fig. S3C), suggesting increased domain flexibility in D4. This may be essential for the bending around D3-5 observed in the SAXS data. D3-4 and D4-5 hinge-like motions were also observed in the ligand-induced activation of KIT (22). Based on high-resolution structural data for KIT, a model for the mechanism of ligand-induced activation of type III/V RTKs has been proposed (22). With this model, together with the available structures for VEGFR-2 and VEGFR-3, it is now possible to construct a similar model for VEGFR activation. Ligand-induced KIT dimerization promotes reorientation of D4-5, enabling lateral D4–D4 interactions essential for receptor activation (22). This seems to be a common mechanism in type III RTK activation (37–40). Previous studies found similarities in the ligand-binding modes for type III and IV RTKs and almost identical homotypic interactions in VEGFR-2 D7 and KIT D4 (18, 24, 32). The VEGFR-3 structures shown here indicate additional similarities to type III RTKs; however, our results also demonstrate some unique properties of these receptors that affect the mechanism of activation. Despite the lack of homotypic interactions in VEGFR-3 D4, the arrangement of D4 in KIT and VEGFR-3 dimers is similar, with the E-F loops facing each other (Fig. S3). KIT D4-5 shows an almost parallel orientation in the ligand-bound receptor dimer, whereas VEGFR-3 D4-5 is twisted, resulting in separation of D4 and formation of unique D5 interactions (Fig. 6A). Because VEGFR activation also requires homotypic interactions in D7, the two additional Ig domains of VEGFRs give rise to a more complex, two-step mechanism for ligand-induced dimerization and receptor activation. This may be functionally relevant by allowing tight control of receptor activation. The subunit interactions in the D7 homodimer of VEGFR-2 cover a solvent-accessible area of only 480 Å2 per chain, and D7 has been shown to be monomeric in solution, indicating that the interactions are of low affinity (24). The dimerization interface in D5 is also relatively small (∼570 Å2 per chain), and the interface is dominated by ionic interactions. Like VEGFR-2 ECD, VEGFR-3 ECD is monomeric in solution, clearly showing that D5 and D7 are not capable of dimerizing the receptor in the absence of ligand (Fig. 5) (25). The restricted mobility of the receptors in the transmembrane domain may further aid the formation of such contacts in active receptor dimers. A recent EM study revealed that the VEGFR-2 ECD is flexible and exists in multiple conformations in ligand-induced dimerization 12964 | www.pnas.org/cgi/doi/10.1073/pnas.1301415110

(23). Similarly, comparison of the available ligand/VEGFR-2 D2-3 complex structures found large variations in the D2-3 twist angles, indicating a D2-3 hinge-like rigid body motion (18, 30). Taken together with our data, these findings suggest a general mechanism for VEGFR activation in which ligand-induced D2-3 reorientation facilitates homotypic interactions in D5 and D7, resulting in specific positioning of the transmembrane and intracellular kinase domains in active receptor dimers (Fig. 6B). Interestingly, our SAXS data revealed strong bending of the receptor ECD complex in solution, which would bring VEGF-C closer to the plasma membrane when bound to the cell surface-expressed VEGFR-3. This bending, along with the distorted symmetry of receptor dimers, might be relevant for coreceptor binding. VEGFRs also demonstrate differences in ligand-induced activation. D2 is the major ligand-binding domain, and the presence of D3 results in increased affinity in VEGF-A binding to VEGFR2 and in placental growth factor binding to VEGFR-1 (17, 28). Consistent with the D2-3 reorientation, VEGFR-3 D3 also increases VEGF-C binding, whereas D4-5 and D6-7 further increase VEGFR-3 ligand-binding affinity. In contrast, the homotypic interactions in VEGFR-2 D4-7 are energetically unfavorable (30), suggesting that D3’s contribution to ligand binding is essential for VEGFR-2 dimerization and activation. VEGFR-1 D4-7 has only a small positive effect on ligand binding (16, 28). VEGFR signaling has emerged as a key target for inhibiting tumor growth and metastasis by blocking tumor vascularization. Current therapeutics inhibiting VEGFR signaling include tyrosine kinase inhibitors, ligand traps, and antibodies blocking ligand binding to VEGFRs (4, 5). However, as is evident from previous studies, VEGFR and type III RTK activation requires additional specific homotypic interactions between the membraneproximal domains (22–26, 38). Antibodies blocking these homotypic interactions are promising tools for therapeutic modulation of VEGFR activity (8, 26, 27). The experiments reported herein introduce a mechanism for VEGFR dimerization and activation via D5 and explain the efficacy of the VEGFR-3 D5-targeted antibodies (8). Our data suggest that the ligand-induced homotypic interactions in D5 and D7 are essential for VEGFR activation, providing strategies for the design of specific VEGFR inhibitors for use in combination with current anti-angiogenic inhibitors.

Fig. 6. The mechanism of ligand-induced VEGFR dimerization and activation. (A) Comparison of the ligand-induced dimerization of D1-5 of type III/V RTKs. The KIT/SCF complex (22) and the model of the VEGFR-3/VEGF-C complex are shown as surface representations in two orientations. SCF dimer is colored in magenta and in light magenta, and the two chains of KIT, VEGF-C, and VEGFR3 are color-coded as in Fig. 1. (B) A proposed model of the ligand-induced dimerization and activation of VEGFRs. D1-2 represents the major ligandbinding unit. Ligand-induced D2-3 reconfiguration facilitates homotypic interactions in D5 and D7 that together are important for VEGFR activation. SAXS data indicates bending of the VEGF-C/VEGFR-3 complex around D3-5.

Leppänen et al.

Such an approach with two HER2 antibodies has proven successful in the treatment of patients with HER2-positive metastatic breast cancer (41).

(Table S1). The sites were identified, and the phases were refined using autoSHARP (42). The structures were refined using PHENIX (43). Crystallographic details are provided in SI Methods.

Methods

SAXS Data Collection and Analysis. The SAXS data were collected at SLS beamline cSAXS, and the data were processed using the ATSAS program package (44). The ab initio shape reconstructions were computed using DAMMIF, and rigid-body modeling was done using SASREF. Details of SAXS data collection and analysis are provided in SI Methods.

Binding Assays. Calorimetric titrations of VEGF-C (C137A) to the monomeric VEGFR-3 deletion mutants were performed with a MicroCal VP-ITC calorimeter, as described in SI Methods. The data were processed using Origin 7.0 (MicroCal).

Negative-Staining EM and Image Analysis. For the VEGF-C/VEGFR-3 D1-7 complex, data were acquired using a Philips CM10 transmission electron microscope equipped with an LaB6 filament. For projection analysis, 2,916 particles of the complex were selected from 487 images and classified into 30 classes. The EM analysis is described in detail in SI Methods.

Crystallization and Structure Determination. The VEGF-C/VEGFR3-D12 complex crystals were grown over a reservoir solution of 1.0–1.5 M ammonium sulfate (pH 8.5–9.5). Heavy-atom derivatives were prepared by soaking the crystals with methylmercury acetate, potassium tetrachloroplatinate, or hexatantalum tetradecabromide. VEGFR-3 D4-5 crystals were grown over a reservoir solution of 20–25% PEG 400 (pH 7.5–8.5), and a heavy-atom derivative was obtained with methylmercury acetate. Complete datasets to 4.2-Å and 2.5-Å resolution were collected from the VEGF-C/VEGFR-3 D1-2 complex and the VEGFR-3 D4-5 crystals, respectively, at beamline X06SA at the Swiss Light Source (SLS). Anomalous data on the heavy-atom derivatives were collected at SLS beamlines X06SA and X06DA

ACKNOWLEDGMENTS. We thank Tapio Tainola and Seppo Kaijalainen for excellent technical assistance, Dr. John H. Missimer for SAXS data evalution, and the staff at European Synchrotron Radiation Facility beamline ID29 and staff at Swiss Light Source beamlines X06SA, X06DA, and cSAXS for assistance and access to their facilities. This work was supported by grants from the Academy of Finland, the Sigrid Juselius Foundation, the LouisJeantet Foundation, the European Research Council (ERC-2010-AdG-268804TX-FACTORS), Swiss National Science Foundation (Grant 31003A-130463), Oncosuisse (Grant OC2 01200-08-2007), NOVARTIS Stiftung für medizinischbiologische Forschung (Grant 10C61), and the Swiss Initiative for Systems Biology (SystemsX.ch; grant CINA).

1. Koch S, Tugues S, Li X, Gualandi L, Claesson-Welsh L (2011) Signal transduction by vascular endothelial growth factor receptors. Biochem J 437(2):169–183. 2. Ferrara N (2004) Vascular endothelial growth factor: Basic science and clinical progress. Endocr Rev 25(4):581–611. 3. Tammela T, Alitalo K (2010) Lymphangiogenesis: Molecular mechanisms and future promise. Cell 140(4):460–476. 4. Ferrara N, Kerbel RS (2005) Angiogenesis as a therapeutic target. Nature 438(7070): 967–974. 5. Alitalo K (2011) The lymphatic vasculature in disease. Nat Med 17(11):1371–1380. 6. Pajusola K, et al. (1994) Signalling properties of FLT4, a proteolytically processed receptor tyrosine kinase related to two VEGF receptors. Oncogene 9(12):3545–3555. 7. Lee J, et al. (1996) Vascular endothelial growth factor-related protein: A ligand and specific activator of the tyrosine kinase receptor Flt4. Proc Natl Acad Sci USA 93(5):1988–1992. 8. Tvorogov D, et al. (2010) Effective suppression of vascular network formation by combination of antibodies blocking VEGFR ligand binding and receptor dimerization. Cancer Cell 18(6):630–640. 9. Kärpänen T, et al. (2006) Functional interaction of VEGF-C and VEGF-D with neuropilin receptors. FASEB J 20(9):1462–1472. 10. Kärkkäinen MJ, et al. (2004) Vascular endothelial growth factor C is required for sprouting of the first lymphatic vessels from embryonic veins. Nat Immunol 5(1): 74–80. 11. Kärkkäinen MJ, et al. (2000) Missense mutations interfere with VEGFR-3 signalling in primary lymphoedema. Nat Genet 25(2):153–159. 12. Dumont DJ, et al. (1998) Cardiovascular failure in mouse embryos deficient in VEGF receptor-3. Science 282(5390):946–949. 13. Tammela T, et al. (2008) Blocking VEGFR-3 suppresses angiogenic sprouting and vascular network formation. Nature 454(7204):656–660. 14. Nilsson I, et al. (2010) VEGF receptor 2/-3 heterodimers detected in situ by proximity ligation on angiogenic sprouts. EMBO J 29(8):1377–1388. 15. Tammela T, et al. (2011) VEGFR-3 controls tip to stalk conversion at vessel fusion sites by reinforcing Notch signalling. Nat Cell Biol 13(10):1202–1213. 16. Wiesmann C, et al. (1997) Crystal structure at 1.7-Å resolution of VEGF in complex with domain 2 of the Flt-1 receptor. Cell 91(5):695–704. 17. Fuh G, Li B, Crowley C, Cunningham B, Wells JA (1998) Requirements for binding and signaling of the kinase domain receptor for vascular endothelial growth factor. J Biol Chem 273(18):11197–11204. 18. Leppänen VM, et al. (2010) Structural determinants of growth factor binding and specificity by VEGF receptor 2. Proc Natl Acad Sci USA 107(6):2425–2430. 19. Jeltsch M, et al. (2006) Vascular endothelial growth factor (VEGF)/VEGF-C mosaic molecules reveal specificity determinants and feature novel receptor binding patterns. J Biol Chem 281(17):12187–12195. 20. Leppänen VM, et al. (2011) Structural determinants of vascular endothelial growth factor-D receptor binding and specificity. Blood 117(5):1507–1515. 21. Lemmon MA, Schlessinger J (2010) Cell signaling by receptor tyrosine kinases. Cell 141(7):1117–1134. 22. Yuzawa S, et al. (2007) Structural basis for activation of the receptor tyrosine kinase KIT by stem cell factor. Cell 130(2):323–334. 23. Ruch C, Skiniotis G, Steinmetz MO, Walz T, Ballmer-Hofer K (2007) Structure of a VEGF-VEGF receptor complex determined by electron microscopy. Nat Struct Mol Biol 14(3):249–250. 24. Yang Y, Xie P, Opatowsky Y, Schlessinger J (2010) Direct contacts between extracellular membrane-proximal domains are required for VEGF receptor activation and cell signaling. Proc Natl Acad Sci USA 107(5):1906–1911.

25. Kisko K, et al. (2011) Structural analysis of vascular endothelial growth factor receptor-2/ligand complexes by small-angle X-ray solution scattering. FASEB J 25(9): 2980–2986. 26. Hyde CA, et al. (2012) Targeting extracellular domains D4 and D7 of vascular endothelial growth factor receptor 2 reveals allosteric receptor regulatory sites. Mol Cell Biol 32(19):3802–3813. 27. Kendrew J, et al. (2011) An antibody targeted to VEGFR-2 Ig domains 4-7 inhibits VEGFR-2 activation and VEGFR-2-dependent angiogenesis without affecting ligand binding. Mol Cancer Ther 10(5):770–783. 28. Christinger HW, Fuh G, de Vos AM, Wiesmann C (2004) The crystal structure of placental growth factor in complex with domain 2 of vascular endothelial growth factor receptor-1. J Biol Chem 279(11):10382–10388. 29. Iyer S, Darley PI, Acharya KR (2010) Structural insights into the binding of vascular endothelial growth factor-B by VEGFR-1(D2): Recognition and specificity. J Biol Chem 285(31):23779–23789. 30. Brozzo MS, et al. (2012) Thermodynamic and structural description of allosterically regulated VEGFR-2 dimerization. Blood 119(7):1781–1788. 31. Chen X, Liu H, Focia PJ, Shim AH, He X (2008) Structure of macrophage colonystimulating factor bound to FMS: Diverse signaling assemblies of class III receptor tyrosine kinases. Proc Natl Acad Sci USA 105(47):18267–18272. 32. Shim AH, et al. (2010) Structures of a platelet-derived growth factor/propeptide complex and a platelet-derived growth factor/receptor complex. Proc Natl Acad Sci USA 107(25):11307–11312. 33. Cudmore MJ, et al. (2012) The role of heterodimerization between VEGFR-1 and VEGFR-2 in the regulation of endothelial cell homeostasis. Nat Commun 3:972. 34. Shinkai A, et al. (1998) Mapping of the sites involved in ligand association and dissociation at the extracellular domain of the kinase insert domain-containing receptor for vascular endothelial growth factor. J Biol Chem 273(47):31283–31288. 35. Barleon B, et al. (1997) Mapping of the sites for ligand binding and receptor dimerization at the extracellular domain of the vascular endothelial growth factor receptor FLT-1. J Biol Chem 272(16):10382–10388. 36. Carl P, Kwok CH, Manderson G, Speicher DW, Discher DE (2001) Forced unfolding modulated by disulfide bonds in the Ig domains of a cell adhesion molecule. Proc Natl Acad Sci USA 98(4):1565–1570. 37. Verstraete K, et al. (2011) Structural insights into the extracellular assembly of the hematopoietic Flt3 signaling complex. Blood 118(1):60–68. 38. Yang Y, Yuzawa S, Schlessinger J (2008) Contacts between membrane proximal regions of the PDGF receptor ectodomain are required for receptor activation but not for receptor dimerization. Proc Natl Acad Sci USA 105(22):7681–7686. 39. Elegheert J, et al. (2011) Extracellular complexes of the hematopoietic human and mouse CSF-1 receptor are driven by common assembly principles. Structure 19(12): 1762–1772. 40. Ma X, et al. (2012) Structural basis for the dual recognition of helical cytokines IL-34 and CSF-1 by CSF-1R. Structure 20(4):676–687. 41. Swain SM, et al. (2013) Pertuzumab, trastuzumab, and docetaxel for HER2-positive metastatic breast cancer (CLEOPATRA study): Overall survival results from a randomised, double-blind, placebo-controlled, phase 3 study. Lancet Oncol 14(6):461–471. 42. Vonrhein C, Blanc E, Roversi P, Bricogne G (2007) Automated structure solution with autoSHARP. Methods Mol Biol 364:215–230. 43. Adams PD, et al. (2010) PHENIX: A comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr D Biol Crystallogr 66(Pt 2):213–221. 44. Petoukhov MV, Svergun DI (2007) Analysis of X-ray and neutron scattering from biomacromolecular solutions. Curr Opin Struct Biol 17(5):562–571.

Leppänen et al.

PNAS | August 6, 2013 | vol. 110 | no. 32 | 12965

BIOCHEMISTRY

Protein Expression and Purification. Human VEGF-C, residues 103–215; C137A mutant of VEGF-C; human VEGFR-3 Ig D1-2, D1-3, D1-5, and D1-7; and the mutant 5EA in D1-5 were expressed in Sf21 insect cells using the baculovirus system and purified as described in SI Methods.

Supporting Information Leppänen et al. 10.1073/pnas.1301415110 SI Methods Protein Expression and Purification. For crystallization, human

VEGF-C and the C137A mutant of human VEGF-C (residues 103–215) and human VEGF receptor (VEGFR)-3 domain (D) 4-5 (residues 330–553) were cloned into the pFastBac1 baculovirus expression vector (Invitrogen) with a C-terminal His tag. VEGFR-3 domains 1–2 (residues 1–229) were also cloned into the pFastBac vector with a C-terminal factor Xa cleavage site and an Fc tag (human IgG1). Recombinant baculovirus was produced in Sf21 insect cells in serum-free Insect-XPRESS medium (Lonza) supplemented with 50 μg/mL gentamycin (Sigma-Aldrich) at 26 °C. For protein expression, Sf21 cells were infected with the corresponding recombinant baculovirus at high multiplicity, and at 3 d postinfection, the supernatant was harvested by centrifugation. VEGFR-3 D4-5 was extracted by Ni2+charged chelating Sepharose (GE Healthcare), after which the resin was washed in PBS containing 15 mM imidazole and eluted with 0.4 M imidazole. Finally, VEGFR-3 D4-5 was purified by gel filtration on a Superdex 200 column (GE Healthcare) in 10 mM MES and 0.1 M NaCl (pH 6.0). The VEGF-C/VEGFR-3 D1-2 complex was purified by protein A sepharose (GE Healthcare) from the mixture of expression media, followed by Fc tag removal and gel filtration on the Superdex 200 column in 10 mM Hepes and 0.15 M NaCl (pH 7.5). For binding assays, domains 1 and 2 (D1-2; residues 1–229), 1–3 (D1-3, residues 1–329), 1–5 (D1-5, residues 1–553), and 1–7 (D1-7, residues 1–776) of human VEGFR-3 were cloned into the pFastBac baculovirus expression vector with the C-terminal factor Xa cleavage site and an Fc tag (human IgG1). The D5 double mutant (T446E/K516A; 5EA) of VEGFR-3 D1-5 was cloned into the same vector. All of the receptor constructs were expressed in Sf21 insect cells using baculovirus expression and were purified by a protein A-Sepharose (GE Healthcare) affinity step, followed by Fc tag removal and gel filtration on a Superdex 200 column in Hepes-buffered saline (HBS). Affinity Measurements. Isothermal calorimetric titrations of human VEGF-C (C137A) to the soluble monomeric VEGFR-3 deletion mutants D1-2, D1-3, D1-5, and D1-7 and the D5 double mutant 5EA of VEGFR-3 D1-5 were performed at 25 °C using a MicroCal VP-ITC calorimeter. To control for heat dilution effects, all of the protein buffers were adjusted to HBS at pH 7.5 by gel filtration. The receptor constructs were used in the calorimeter cell at concentrations of 5–8 μM, and the VEGF-C ligand in the syringe was used at a concentration of ∼0.10 mM. Data were processed with MicroCal Origin 7.0 software. Crystallization and Structure Determination. For crystallization, VEGFR-3 D4-5 was concentrated to 9 mg/mL, and the buffer [10 mM MES (pH 6.0) and 100 mM NaCl] was supplemented with 0.1% (vol/vol) P8340 protease inhibitor mixture (SigmaAldrich) and 0.01% (wt/vol) NaN3. Crystallization conditions were screened using the sitting-drop vapor-diffusion technique. VEGFR-3 D4-5 crystals were grown for several days at room temperature over a reservoir solution of 0.1 M phosphate buffer (pH 7.5–8.5) and 25% PEG 400 (wt/vol). The final drops were prepared by manually mixing 1–2 μL of the reservoir solution and 1–2 μL of the protein solution. The hexagonal crystals belong to space group P3121 (a, b = 133.3 Å; c = 48.9 Å) with one monomer per asymmetric unit. For heavy-atom derivative data collection, the crystals were soaked in 1% (vol/vol) of saturated solution of methyl mercury Leppänen et al. www.pnas.org/cgi/content/short/1301415110

acetate (pH 8.0) for 1 h. For data collection at 100 K, the crystals were frozen in liquid nitrogen in the reservoir solution supplemented with 20% (vol/vol) glycerol. The VEGF-C/VEGFR-3 D1-2 complex was initially crystallized using WT VEGF-C without any modifications, but subsequently, larger crystals with better diffraction were obtained with the Cys137Ala mutant of VEGF-C by lysine methylation (1) of the whole complex. Typically, the complex was concentrated to 5–10 mg/mL in HBS supplemented with 0.1% (vol/vol) P8340 protease inhibitor mixture (Sigma-Aldrich) and 0.01% (wt/vol) NaN3, and crystals were grown over a reservoir solution of 0.1 M Bis-Tris buffer (pH 8.5–9.5) and 1.0–1.5 M ammonium sulfate at room temperature. The final crystals belong to the cubic space group I23 (a, b, c = 166.7 Å), with a single chain each of VEGF-C and VEGFR-3 D1-2 per asymmetric unit. For heavy-atom derivative data collection, the crystals at pH 9.0 were soaked in 1% (vol/vol) of saturated solution of methylmercury acetate for 1 h, in 0.1 mM potassium tetrachloroplatinate for 12 h, or in 2 mM hexatantalum tetradecabromide for 12 h. For data collection at 100 K, the crystals were frozen in liquid nitrogen in the reservoir solution supplemented with 20% (vol/vol) glycerol. Complete datasets to 2.5-Å and 3.5-Å resolution were collected from single native VEGFR-3 D4-5 and Hg-derivative crystals, respectively, at beamline X06SA at the Swiss Light Source (SLS) in Villigen, Switzerland (Table S1). Data were processed with XDS (2) and the CCP4 suite of programs. The VEGFR-3 D4-5 structure was solved by single isomorphous replacement with anomalous scattering (SIRAS) phasing using autoSHARP (3). Automated model building in Arp/warp (4) was used to generate an initial model and improve the phases. The model was completed by iterative refinement in PHENIX (5) and model building in Coot (6). A subset of 5% of the diffraction data was omitted from refinement for calculating the free R factor (Rfree). The final model contains residues 330–469 and 484–553, N-terminal aspartate and histidine residues from the linker, a C-terminal histidine from the His-tag, two N-linked glucosamine moieties, and 21 solvent molecules. A long loop in D5 (residues 470–483) had poor density, and the atoms were omitted from refinement. The side chain atoms of Arg378 and Lys387, facing each other in the adjacent D4 domains in the homodimer, had poor density and could not be built into the model with confidence. The final VEGFR-3 D4-5 model comprises residues 330–553. The majority (96.7%) of these residues are within the most favored region of the Ramachandran plot. All molecular graphics were generated with PYMOL (http:// www.pymol.org/). Data for the VEGF-C/VEGFR-3 D1-2 complex crystal with WT VEGF-C were collected at beamline ID-29 at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France (Table S1). Native data to 4.2-Å resolution from the lysinemethylated complex crystals with VEGF-C C137A mutant were collected at SLS beamline X06SA, and anomalous data from the heavy-atom derivatives were collected at SLS beamline X06DA (Table S1). Data were processed with XDS (2), and heavy-atom sites were identified and phases refined by multiple isomorphous replacement with anomalous scattering (MIRAS)-phasing using autoSHARP (3). The complex structure was solved using VEGF-C (PDB ID code 2X1X), VEGFR-1 D2 (PDB ID code 1FLT), and KIT D1 (PDB ID code 2E9W) as search models in phased molecular replacement in MOLREP (7). The model was completed by iterative cycles of model building in Coot (6) and refinement in PHENIX (5). 1 of 9

For the low-resolution refinement, the molecular replacement search models were used as a reference model and strict geometry restraints and TLS refinement were applied. The final VEGF-C model contains residues 115–213 and N-linked glucosamine moieties in Asn175 and Asn205. The final VEGFR-3 D1-2 model contains residues 28–76, 90–113, 124–210, and 213–223 and N-linked glucosamine moieties in Asn33, Asn104, and Asn166. The asymmetric unit contains only one chain each of VEGF-C and VEGFR-3 D1-2, and the full complex with a 2:2 stoichiometry is generated by a twofold crystallographic axis. All molecular graphics were generated with PYMOL. Homology Modeling. The model of the dimeric VEGF-C/VEGFR-3 D1-5 complex (Fig. 1) was constructed by combining the VEGFR-3 crystal structures with a homology model of D3. First, the VEGF-C/VEGFR-2 D2-3 complex structure (PDB ID code 2X1X) and the homodimer of VEGFR-3 D4-5 were superimposed with perpendicularly oriented KIT/SCF complex. Next, the VEGF-C/VEGFR-3 D1-2 complex and a homology model of D3 were superimposed with the VEGFR-2 complex to obtain a model of the VEGF-C/VEGFR-3 D1-5 complex (Fig. 1). For rigid-body refinement against the small-angle X-ray scattering (SAXS) data (Fig. 5 C and D), a VEGF-C/VEGFR-3 D1-7 complex was constructed as follows. A homology model of the homodimer of VEGFR-2 D7 was superimposed with the VEGFR-3 D4 dimer, and two copies of a D6 model were superimposed with the two D3 domains in the D1-5 complex model. Then the D6-7 module was moved below D5 with a perpendicular translation. D3-4 and D6-7 thus adopted KIT D3-4–like interdomain cavities. Similar cavities were observed in the EM study of the VEGF-A/ VEGFR-2 extracellular domain (ECD) complex (8). VEGFR-3 D3 and D7 homology models were obtained starting from the corresponding VEGFR-2 domains (PDB ID codes 2X1X and 3KVQ, respectively) in Swiss-Model (9). A VEGFR-3 D6 homology model based on an IgG2 domain (PDB ID code 1AXT) was obtained using the Phyre server (10). SAXS. The SAXS experiments were conducted at the SLS

beamline cSAXS using a wavelength of λ = 1.0 Å. The scattered intensities were recorded using a Pilatus 2M detector (Dectris) in the scattering vector range 0.008 < q < 0.3 1/Å, where q is the length of the scattering vector, defined as q = 4π sinθ/λ, with 2θ the scattering angle. The scattering vector range was calibrated using a silver behenate sample, and a standard BSA sample was measured as a quality control. All of the samples were ultracentrifuged before the experiments. The samples were measured in 1-mm quartz capillaries with a 0.01-mm wall thickness (Hilgenberg). The data were recorded in a series of 20 scans along the capillary, with each scan consisting of 10 exposures of 0.5 s separated by 0.4 mm. A buffer dataset was collected before each protein sample dataset in the same capillary, to enable accurate background subtraction. At least three concentrations from each protein sample were measured. The buffer and protein scans were radially averaged, checked for radiation damage or other changes in the intensities, and averaged. The buffer background was subtracted from the protein scattering using in-house MATLAB scripts. SAXS data modeling was done using ATSAS version 2.3.1 (11). The pair-distance distribution function, P(r), was calculated using Autoporod and used as input for the ab initio shape reconstruction program DAMMIF. Twenty independent DAMMIF models were reconstructed, aligned, averaged, and filtered. Rigid-body refinement of the dimeric VEGF-C/VEGFR-3 D1-7 complex model was conducted using the SASREF program. Interdomain movement was enabled by introducing linkers with a maximum length of 7 Å between the domains in the symmetric starting model. Several different models with different numbers of linkers were generated. At least five independent runs were made Leppänen et al. www.pnas.org/cgi/content/short/1301415110

with each model. The results shown in Fig. 5 are from runs conducted using a complex model with high mannose N-glycan moieties (PDB ID code 1GYA) modeled into the known N-glycosylation sites in VEGF-C and VEGFR D1-5 and to the putative sites in VEGFR-3 D6-7. Modeling without sugars resulted in similar models but higher χ2 values (Fig. S5). Final χ2 values were calculated from the model coordinates by CRYSOL, taking into account the hydration shell surrounding the proteins in solution. Negative-Stain EM and Image Analysis. The VEGF-C/VEGFR-3 D1-7 complex samples were prepared for EM analysis using a conventional negative-staining protocol (12). In brief, 5 μL of protein sample was adsorbed to a glow-discharged carbon-coated copper grid, washed with two drops of deionized water, and stained with one drop of 2% (wt/vol) uranyl acetate. A protein concentration of 3 μg/mL was used for the grid preparation. The samples were imaged at room temperature with a Philips CM10 electron microscope equipped with an LaB6 filament and operated at an acceleration voltage of 80 kV. Images of specimens were recorded with a side-mounted Veleta 2k × 2k CCD camera (Olympus Soft Imaging Solutions) using low-dose procedure at a magnification of 92,000×. The X3D display program (13) was used to select particles. For projection analysis, 2,916 particles of VEGF-C/VEGFR-3 D1-7 complex were interactively selected from 487 images. The particles were windowed into 70 × 70 pixel images, stacked, centered, subjected to rotationally invariant k-means alignment, and classified into 30 output classes using the SPIDER program suite (14). For comparison of the VEGF-C/VEGFR-3 D1-5 complex and the VEGF-C/VEGFR-3 EM structure, the model of the VEGFC/VEGFR-3 D1-5 complex was converted into density volume filtered to 25-Å resolution using the Bsoft software package (15). The volume was then used to calculate projections at an angular interval of 15° with the SPIDER image processing suite (14). The projections with the most similar features to the experimental 2D averages were selected (Fig. S6B). Density volumes and ribbon diagrams were displayed and oriented with the UCSF Chimera package (16). VEGFR-3 Mutagenesis and Cell Culture. The VEGFR-3 D5 T446E/ K516A (5EA) mutation with and without the VEGFR-3 D7 R737A (7A) mutation was introduced into pMX/VEGFR-3streptagIII and pcDNA3.1/VEGFR-3 expression vectors. The desired mutations were confirmed by nucleotide sequencing. Retroviruses were produced in 293GPG cells (17). Porcine aortic endothelial (PAE) cells were transfected with the retroviruses and human embryonic kidney epithelial 293 (HEK293T) cells using chemical transfection. The cells were maintained in DMEM containing 10% (vol/vol) fetal serum and grown in a humidified atmosphere with and 5% (vol/vol) CO2 at 37 °C. Cell Surface Biotinylation. PAE-VEGFR-3 cells were starved overnight in serum-free medium and then treated with Sulfo-NHSLC-Biotin (Thermo Fisher Scientific) in accordance with the manufacturer’s recommendations. After the treatment, the cells were lysed in 1% (vol/vol) Nonidet P-40 in PBS with protease inhibitors, and biotinylated proteins were precipitated using streptavidin-agarose resin (Thermo Fisher Scientific) and analyzed by SDS-PAGE and Western blot analysis with rabbit antihuman VEGFR-3 antibodies (Santa Cruz Biotechnology). Western Blot and Immunoprecipitation Analyses. For immunoprecipitation and Western blot analysis of phosphotyrosines, the PAE cells expressing WT and mutant VEGFR-3 were lysed in 1 mL of PLCLB lysis buffer [150 mM NaCl, 5% (vol/vol) glycerol, 1% (vol/vol) Triton X-100, 1.5 M MgCl2, and 50 mM Hepes; pH 7.5] supplemented with 1 mM vanadate, 2 mM phenylmethylsulphonyl 2 of 9

fluoride, 2 μg/mL leupeptin, and 0.07 U/mL aprotinin. Cleared lysates were incubated with 2 μg of mouse monoclonal 9D9F9 antibody against VEGFR-3 for 2 h, after which the immunocomplexes were captured using protein G-Sepharose, washed three times in the PLCLB buffer, and separated in 7.5% (wt/vol) SDS-PAGE. After blotting of the proteins to nitrocellulose mem-

branes and blocking in 5% (wt/vol) BSA, the membranes were probed with monoclonal phosphotyrosine antibodies (0.5 μg/mL) and visualized by chemiluminescence (Pierce) using HRP-coupled antibodies (Dako). For autophoshorylation and VEGFR-3 expression quantification, the band area density was quantified with ImageJ software (National Institutes of Health).

1. Walter TS, et al. (2006) Lysine methylation as a routine rescue strategy for protein crystallization. Structure 14(11):1617–1622. 2. Kabsch W (2010) Xds. Acta Crystallogr D Biol Crystallogr 66(Pt 2):125–132. 3. Vonrhein C, Blanc E, Roversi P, Bricogne G (2007) Automated structure solution with autoSHARP. Methods Mol Biol 364:215–230. 4. Langer G, Cohen SX, Lamzin VS, Perrakis A (2008) Automated macromolecular model building for X-ray crystallography using ARP/wARP version 7. Nat Protoc 3(7): 1171–1179. 5. Adams PD, et al. (2010) PHENIX: A comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr D Biol Crystallogr 66(Pt 2):213–221. 6. Emsley P, Cowtan K (2004) Coot: Model-building tools for molecular graphics. Acta Crystallogr D Biol Crystallogr 60(Pt 12 Pt 1):2126–2132. 7. Vagin A, Teplyakov A (2010) Molecular replacement with MOLREP. Acta Crystallogr D Biol Crystallogr 66(Pt 1):22–25. 8. Ruch C, Skiniotis G, Steinmetz MO, Walz T, Ballmer-Hofer K (2007) Structure of a VEGF-VEGF receptor complex determined by electron microscopy. Nat Struct Mol Biol 14(3):249–250. 9. Arnold K, Bordoli L, Kopp J, Schwede T (2006) The SWISS-MODEL workspace: A Web-based environment for protein structure homology modelling. Bioinformatics 22(2):195–201.

10. Kelley LA, Sternberg MJ (2009) Protein structure prediction on the Web: A case study using the Phyre server. Nat Protoc 4(3):363–371. 11. Petoukhov MV, Svergun DI (2007) Analysis of X-ray and neutron scattering from biomacromolecular solutions. Curr Opin Struct Biol 17(5):562–571. 12. Ohi M, Li Y, Cheng Y, Walz T (2004) Negative staining and image classification: Powerful tools in modern electron microscopy. Biol Proced Online 6:23–34. 13. Conway JF, et al. (1997) Visualization of a 4-helix bundle in the hepatitis B virus capsid by cryo-electron microscopy. Nature 386(6620):91–94. 14. Frank J, et al. (1996) SPIDER and WEB: Processing and visualization of images in 3D electron microscopy and related fields. J Struct Biol 116(1):190–199. 15. Heymann JB, Belnap DM (2007) Bsoft: Image processing and molecular modeling for electron microscopy. J Struct Biol 157(1):3–18. 16. Pettersen EF, et al. (2004) UCSF Chimera—a visualization system for exploratory research and analysis. J Comput Chem 25(13):1605–1612. 17. Ory DS, Neugeboren BA, Mulligan RC (1996) A stable human-derived packaging cell line for production of high-titer retrovirus/vesicular stomatitis virus G pseudotypes. Proc Natl Acad Sci USA 93(21):11400–11406.

Fig. S1. (A) WT and the C137A mutant of VEGF-C show identical complex formation with VEGFR-3 D1-2. Shown is the VEGF-C/VEGFR-3 D1-2 complex structure in the electron density map calculated with the WT VEGF-C data (Table S1) at 6-Å resolution and contoured at 1σ. The density-modified map was calculated with the program DM (Cowtan 1994) using the MIRAS phases. VEGF-C receptor-binding epitopes (L1–L3, αN) and D2 of VEGFR-3 are labeled. The connections in electron density between VEGF-C αN and VEGFR-3 D2 are indicated by red arrows. The electron density map calculated using the WT VEGF-C data are consistent with the refined structure with the C137A mutant of VEGF-C. (B) Thermodynamic analysis of the D123A/Q130A double mutant of VEGF-C binding to VEGFR-3 D1-7 showing the enthalpy change (ΔH), dissociation constant (Kd), and stoichiometry (N) of the ITC assay.

Leppänen et al. www.pnas.org/cgi/content/short/1301415110

3 of 9

Fig. S2. Comparison of the VEGFR-3 and VEGFR-2 complexes with VEGF-C. (A) Top views of the VEGF-C/VEGFR-3 D1-2 complex structure and the published VEGF-C/VEGFR-2 D2-3 complex (PDB ID code 2X1X). The structures are shown as ribbon diagrams, with the VEGFR-3 complex color-coded as in Fig. 1, VEGFR-2 in magenta, and VEGFR-2 -bound VEGF-C in cyan. Compared with the VEGFR-2 complex, the N-terminal helix of VEGF-C in the VEGFR-3 complex is bent ∼9° toward D2. (B) Side view of A. (C) Comparison of the KIT/SCF complex (green/magenta; PDB ID code 2E9W) and the VEGF-C complexes with VEGFR-2 and VEGFR-3.


4 of 9

Fig. S3. Crystal structure of the VEGFR-3 D4-5 homodimer. (A) Cartoon representation of the D4-5 structure, with β-strands labeled where applicable. Asnlinked glucosamine moieties and the Asn residues are colored in red and shown as sticks. (B) Homodimer of D4-5 with the two chains in surface or cartoon representations. Color-coding is as in Fig.1. (C) Comparison of the overall temperature factors (B-factors) of D4 and D5. Shown is a Pymol representation of the D4-5 backbone with backbone B-factor values correlated with tube radius and rainbow coloring. The overall B-factors are given in parentheses. The Cα atoms of the fully conserved Thr446 and Lys516 are shown as red spheres. (D) SDS-PAGE (4–20%) analysis of the protein in VEGFR-3 D4-5 crystals under reducing conditions with silver staining. The apparent N- and C-terminal fragments arising from the proteolytic cleavage of D5 are indicated. (E) Close-up of the D4/D5 interface. Glu344, Lys539, Glu391 and Tyr448 are shown as sticks, and the hydrogen bond and salt bridge are shown as a red dashed line. (F) The homodimer of D4-5 tilted toward the viewer and Arg378 and Lys387 shown in a ball-and-stick representation. Symmetry-related Arg378 and Lys387 face each other, resulting in putative domain repulsion. (G) Comparison of D4-5 dimerization in KIT (PDB ID code 2E9W) and VEGFR-3. KIT and VEGFR-3 D4-5 were superimposed in Coot (6); the dimers are shown as a surface representation from the bottom view. VEGFR-3 is color-coded as in B, and the two chains of KIT are in gray. The two chains of KIT D4-5 run almost parallel, whereas the D4-5 in the two chains of VEGFR-3 are twisted. The C termini are separated by ∼29 Å in VEGFR-3 D5, compared with 14 Å in KIT. In addition, the C termini of VEGFR-3 D5 are rotated by ∼28° degrees compared with those of KIT.


5 of 9

Fig. S4. Homotypic interactions in VEGFR-3 D5 and D7 are important for receptor activation and ligand binding. (A) HEK293 cells transiently expressing WT VEGFR-3, the 5EA double mutant, the 7A single mutant, and the 5EA7A triple mutant were stimulated with 50 ng/mL VEGF-C for 10 min, after which the cell lysates were immunoprecipitated with anti–VEGFR-3 and analyzed for VEGFR-3 autophosphorylation (pY) and VEGFR-3 expression (R3) by Western blot analysis. The pY/R3 ratio was quantified based on the ∼120-kDa band (arrow), representing the fully processed VEGFR-3. (B) PAE cells stably expressing wt VEGFR-3 and the 5EA7A triple mutant were stimulated with VEGF-C as indicated, and the cell lysates were immunoprecipitated with anti-VEGFR-3 and evaluated by Western blot analysis as in A. (C) Summary of the enthalpy change (ΔH), entropy change (ΔS), dissociation constant (Kd), and stoichiometry (N) of the VEGF-C binding to VEGFR-3 domain deletion and the D5 mutant in ITC experiments. (D) Calorimetric titrations of VEGF-C to the monomeric 5EA mutant of VEGFR-3 D1-5.


6 of 9

Fig. S5. (A) SDS-PAGE (4–20%) analysis of the purified VEGFR-3 D1-7 in the VEGF-C complex (R3/C) and alone. Because of the proteolytic cleavage in D5, VEGFR-3 D1-7 runs as two bands of approximately 70 kDa and 45 kDa under reducing conditions. (B) SAXS data for VEGFR-3 D1-7 (blue) and the VEGF-C/ VEGFR-3 D1-7 complex (black). Parameters in the table were calculated from the SAXS data. Values are averages between the different concentrations, and the errors are SDs. The molecular weight estimate was calculated from the excluded volume (molecular weight = volume/2). The values for VEGF-C are from Kisko et al. (1). (C) Rigid-body refinement without N-linked glycosylation. Typical runs for the same models as in Fig. 5C are shown. Symmetric model (Top). Model with limited movement (Middle): D123C dimer–linker–D45 dimer–linker–D67 dimer. Model with increased movement (Bottom): D12C dimer–linker–D3-D3– linker–D4-D4–linker–D5 dimer–linker–D67 dimer.

1. Kisko K, et al. (2011) Structural analysis of vascular endothelial growth factor receptor-2/ligand complexes by small-angle X-ray solution scattering. FASEB J 25(9):2980–2986.


7 of 9

Fig. S6. (A) Negative-staining EM analysis of the VEGF-C/VEGFR-3 D1-7 complex. A gallery of 30 2D class averages from image classification is shown. (Scale bar: 10 nm.) (B) Comparison of the model of the VEGF-C/VEGFR-3 D1-5 complex shown in Fig. 1 and the VEGF-C/VEGFR-3 ECD EM class averages. Shown are comparisons of representative class averages with selected 2D projections and with the corresponding 3D-volume models of the VEGF-C/VEGFR-3 D1-5 complex model. The volumes were filtered at 25-Å resolution. The first class average on the left is a mirror image of the corresponding class average in Fig. 5F. (Scale bars: 10 nm.)


8 of 9

Table S1. Data collection and refinement statistics VEGFR-3 D4-5

Data collection Beamline Space group Cell dimensions a, b, c, Å α, β, γ, ° Resolution, Å Rsym I/sI Completeness, % Redundancy Heavy-atom sites Refinement Resolution, Å No. of reflections Rwork/Rfree No. of atoms Protein Glycan Water B-factors Protein Glycan Water Ramachandran plot, % Favored Allowed Outliers rmsd Bond lengths, Å Bond angles,°

VEGF-C/VEGFR-3 D1-2 complex

Native

Hg derivative

Native

Native (C137A)

Pt derivative

Hg derivative

Ta derivative

X06SA (SLS) P3121

X06SA (SLS) P3121

ID29 (ESRF) I23

X06SA (SLS) I23

X06DA (SLS) I23

X06DA (SLS) I23

X06DA (SLS) I23

133.3, 133.3, 48.9 90, 90, 120 40–2.50 (2.65–2.50)* 4.8 (132.3) 28.1 (2.2) 99.8 (99.7) 9.9 (10.3)

133.6, 133.6, 48.5 90, 90, 120 50–3.67 (3.90–3.67) 10.8 (35.0) 18.9 (6.6) 99.3 (96.1) 8.8 (8.3) 4

164.0, 164.0, 164.0 90, 90, 90 20–6.0 (6.33–6.0) 8.5 (104.6) 16.8 (2.4) 96.6 (98.6) 6.6

166.7, 166.7, 166.7 90, 90, 90 30–4.20 (4.40–4.20) 5.1 (150) 20.9 (2.1) 99.6 (100) 9.9 (10.5)

162.9, 162.9, 162.9 90, 90, 90 40–6.30 (6.71–6.30) 5.9 (126.3) 34.5 (1.7) 99.3 (97.1) 19.3 (10.3) 3

164.3, 164.3, 164.3 90, 90, 90 40–6.47 (6.86–6.47) 9.3 (95.5) 31.0 (4.1) 100 (100) 23.0 (21.4) 5

165.3, 165.3, 165.3 90, 90, 90 40–5.44 (5.77–5.44) 4.0 (123.1) 41.6 (3.2) 98.8 (98.2) 27.9 (26.4) 1

40–2.50 17,525 21.0/25.3

20–4.20 5,721 33.4/37.1

1673 28 21

1790 84 0

61.0 81.5 49.2

133.7 182.6

96.7 3.3 0

90.4 9.2 0.4

0.006 1.06

0.004 1.032

*Values in parentheses are for the highest-resolution shell.


9 of 9