Crystal Structure of the Orf Virus NZ2 Variant of Vascular Endothelial Growth Factor-E

Mammalian vascular endothelial growth factors constitute a family of polypeptides, vascular endothelial growth factor (VEGF)-A, -B, -C, -D and placenta growth factor (PlGF), that regulate blood and lymphatic vessel development. VEGFs bind to three types of receptor tyrosine kinases, VEGF receptors 1, 2, and 3, that are predominantly expressed on endothelial and some hematopoietic cells. Pox viruses of the Orf family encode highly related proteins called VEGF-E that show only 25-35% amino acid identity with VEGF-A but bind with comparable affinity to VEGFR-2. The crystal structure of VEGF-E NZ2 described here reveals high similarity to the known structural homologs VEGF-A, PlGF, and the snake venoms Vammin and VR-1, which are all homodimers and contain the characteristic cysteine knot motif. Distinct conformational differences are observed in loop L1 and particularly in L3, which contains a highly flexible GS-rich motif that differs from all other structural homologs. Based on our structure, we created chimeric proteins by exchanging selected segments in L1 and L3 with the corresponding sequences from PlGF. Single loop mutants did not bind to either receptor, whereas a VEGF-E mutant in which both L1 and L3 were replaced gained affinity for VEGFR-1, illustrating the possibility to engineer receptor-specific chimeric VEGF molecules. In addition, changing arginine 46 to isoleucine in L1 significantly increased the affinity of VEGF-E for both VEGF receptors.

sis, diabetic retinopathy, arthritis, malignant cell growth, some neurodegenerative diseases, such as amyotrophic lateral sclerosis (1,2), and a placental insufficiency, preeclampsia (3). The mammalian VEGFs 5 are among the major mediators of angiogenesis and belong to a gene family that includes VEGF-A, -B, -C, and -D (reviewed in Ref. 4) and placenta growth factor (5). Due to alternative splicing and posttranslational processing, VEGFs are generated in a number of functionally distinct isoforms (reviewed in Refs. 4 and 6). Orf viruses encode reading frames for highly homologous proteins collectively called VEGF-E (7)(8)(9), and similar proteins, termed VEGF-F, were found in some snake venoms (10 -13). All members of the VEGF family are secreted as cysteine-cross-linked dimeric glycoproteins.
VEGF-E family members display an extraordinary degree of * This work was supported in part by Swiss National Foundation Grant 3100A0-100204 (to K. B.-H.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. The atomic coordinates and structure factors (code 2GNN) have been deposited in the Protein Data Bank, Research Collaboratory for Structural Bioinformatics, Rutgers University, New Brunswick, NJ (http://www.rcsb.org/). □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental Figs. S1 and S2. 1 Supported by a fellowship from the Paul Scherrer Institut. 2 Both contributed equally to this work. 3 Supported by Grant 3100AO-100204 from the Swiss National Science Foundation. 4  sequence variation (29,30) yet show only 25-35% amino acid identity with VEGF-A. They lack a heparin binding domain, but some variants retain binding to neuropilin-1 (8,9). The viral VEGFs are potent mitogens stimulating proliferation of human endothelial cells in vitro and vascularization of sheep skin in vivo with potencies equivalent to VEGF-A (25). In addition, transgenic mice overexpressing the NZ7 variant of VEGF-E showed increased vascularization in subcutaneous tissue without producing the edematous lesions typically present on the skin of VEGF-A transgenic mice (31).
To understand the molecular basis of receptor specificity of VEGF-E, extensive functional studies were carried out in which the residues important for interaction with VEGFR-2 were tentatively established. Kiba et al. (32) showed that exchanging the region encompassing loops L1 and L3 of the VEGF-E variant NZ7 by the corresponding loops from PlGF or VEGF-A strongly reduced the activity of this viral VEGF implying specific interactions between L1 and L3 with VEGFR-2 (32). However, despite extensive mutagenesis and molecular modeling, receptor specificity of VEGF-E is not understood at the molecular level, and no chimeric VEGF-E homologs have been described so far that gain the ability to bind VEGFR-1.
Here we report the crystal structure of full-length Orf virus VEGF-E NZ2, a representative of the more than 20 VEGF-like molecules expressed by pox viruses. This molecule folds into the same overall structure as VEGF-A, PlGF, VR-1, and Vammin with the characteristic cysteine knot motif. We also present functional data demonstrating that the combined structural epitope formed by loops L1 and L3 dictates receptor specificity.

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Purification of VEGF-E-The template DNA was a kind gift from A. A. Mercer (University of Otago, New Zealand). The sequence encoding residues 20 -133 of the NZ2 variant Orf virus VEGF-E was amplified by PCR using the following primers: forward primer, 5Ј-GCTGAATTCGACAGCAA-CACGAAAGGGATGG-3Ј; reverse primer, 5Ј-GATGATGGTC-GACGCGGCGTCTTCTGGGCGG-3Ј. The PCR fragment was digested with EcoRI and SalI, purified, and ligated into the corresponding pPICZ␣A (Invitrogen) vector for expression in the methylothrophic yeast, Pichia pastoris X-33. The open reading frame encodes a hexahistidine tag at the 3Ј-end. Cloning and selection of VEGF-E expressing transformants was performed as described earlier (33). Cells were grown in 1 liter of buffered glycerol complex medium for 16 h at 30°C. Upon induction of protein synthesis with 1% methanol, cells produced and secreted VEGF-E for 70 h at 28°C. Methanol was added every 24 h to a final concentration of 1% (v/v).
VEGF-E-containing cell-free medium was centrifuged and filtered. The concentrated supernatant was incubated with endoglycosidase F overnight at 30°C and subsequently applied to an Ni 2ϩ -nitrilotriacetic acid column (GE Healthcare). The bound protein was eluted with 250 mM imidazole, pooled, and chromatographed on a Superdex TM 75 column (GE Healthcare). Native VEGF-E eluted as a homogenous peak that corresponded to a molecular size of 30.3 Ϯ 0.2 kDa as determined by static light scattering (n ϭ 2). The identity of the protein was checked by immunoblotting, by reducing and non-reducing SDSpolyacrylamide electrophoresis, by amino-terminal sequencing, and by matrix-assisted laser desorption time-of-flight mass spectrometry.
Cloning, Expression, and Purification of Soluble VEGFR-2 Domains-A covalently linked dimer of the extracellular part of VEGFR-2 was obtained by carboxyl-terminal fusion of the extracellular domain of VEGFR-2 to the leucine zipper motif of the yeast transcription factor GCN4 containing a disulfide bond-promoting cysteine residue (34,35). The template DNA for the zipper motif was a gift from M. O. Steinmetz (PSI, Villigen, Switzerland). The sequence encoding the zipper was amplified by PCR, generating a 187-bp fragment containing the linker downstream from the sequence encoding the extracellular seven Ig-like domains of VEGFR-2 and the leucine zipper of GCN4, followed by a sequence encoding a tobacco etch virus protease cleavage site, a His 6 tag and the sequence downstream of the multiple cloning site in the pcDNA3 vector (Invitrogen). The primers used were as follows: forward primer, 5Ј-GTGCCCAGGAAAAGACGAACTTCGGA-TAACAGTGCGAAGACAAAGTTGAAGAACTGCTGTC-3Ј; reverse primer, 5Ј-GGTGACACTATAGAATAGGGCCCTTA-GTATCAGTGATGGTGATGGTGATGCTGGAAGTAGAG-GTTCTCACCAACCAGTTTTTTCAGACG-3Ј. Boldface characters represent the cysteine mutation. In a second PCR, the amplified fragment was inserted by site-specific mutagenesis (36) into pcDNA3 carrying the sequence encoding the extracellular domain of VEGFR-2.
Soluble VEGFR-2 domains were expressed as covalently linked dimers in transiently transfected HEK293T cells grown in Dulbecco's modified Eagle's medium supplemented with 6 mM sodium butyrate and 0.1% fetal bovine serum. After 72 h, the culture supernatant was harvested and loaded onto a Ni 2ϩnitrilotriacetic acid affinity column (GE Healthcare). The column was washed, and the His 6 -tagged protein was eluted with 250 mM imidazole. Further purification was achieved by gel filtration on Superdex TM 200 HR (GE Healthcare). The identity of the protein was verified by SDS-PAGE, immunoblotting, amino-terminal sequencing, and molecular weight determination by static light scattering.
Radioiodination-125 I labeling of canine recombinant VEGF-A 165 and human PlGF-1 was performed as described previously (37). Canine VEGF-A is one amino acid shorter than the human isoform but functionally identical; we follow the numbering of the human protein (38). The specific activity of the proteins was 60,000 -70,000 cpm/ng.
Competition Assays-PAE cells overexpressing VEGFR-1 or VEGFR-2 were grown overnight to subconfluence in 12-well (for EC 50 determination) or 24-well tissue culture plates and incubated for 4 h on ice with 0.1 nM 125 I-VEGF-A 165 or 125 I-PlGF-1 in the presence or absence of various concentrations of cold competitor. Data points were determined in duplicates (for EC 50 determination) or in triplicates. Unbound radioactive protein was eliminated by three washes with minimal essential medium, Hepes-buffered saline, 25 mM HEPES, without l-glutamine (Amimed, Allschwil, Switzerland) containing 0.25% bovine serum albumin. Cells were lysed in 1 M NaOH, and radioactivity was determined in a Beckman ␥-counter. Data were analyzed with Prism 4.0 software (GraphPad Software, Inc.), and EC 50 values were determined using nonlinear regression. S.D. values are expressed as a 95% confidence interval of the fitted EC 50 .
Surface Plasmon Resonance-All experiments were performed on a Biacore 3000 apparatus (Biacore Life Sciences) at the Functional Genomics Center of the University of Zürich. Proteins were dissolved in HBS-EP buffer (10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA, and 0.005% surfactant P20) at 20°C. 1500 response units (RU) of dimeric soluble VEGFR-2 were immobilized on a CM5 chip by amine coupling. As a reference, a blank surface was prepared. The analytes were injected at concentrations ranging from 0.025 to 1.6 nM at a flow of 30 l/min. Regeneration of the surface was achieved with two 1-min pulses of 4 M MgCl 2 . Data were evaluated by double referencing using the program BIAevaluation 4.1 (Biacore Life Sciences). k on and k off were simultaneously fitted assuming a 1:1 Langmuir binding mode. For VEGF-E L3 , a drifting base line was included in the calculations, which led to a better fit of the data. Without this correction, 2 was 0.134, k on was 2.96 ϫ 10 6 M Ϫ1 s Ϫ1 , k off 9.3 ϫ 10 Ϫ4 s Ϫ1 , and the K d was 314 pM.
Crystallization, Data Collection, and Structure Determination-Crystals were grown in hanging drops using 40 mg/ml protein solution and 0.6 M ammonium sulfate, 3% polyethylene glycol 4000, 0.1 M sodium citrate, pH 5, and 0.3% benzamidine as precipitant as described. 6 Crystals belong to the space group P4 1 22 (a ϭ b ϭ 98.7 Å, c ϭ 240.2 Å) and contain two VEGF-E dimers in their asymmetric unit. Data were collected at beamline X06SA of the Swiss Light Source (Paul Scherrer Institut, Villigen, Switzerland) using a MAR165 CCD detector. Data collection and structure determination have been described in detail. 6 Buried surface areas were calculated using ArealMol (11) by subtracting the total surface of the interacting molecules from the sum of the surfaces of the individual molecules. All figures representing protein structures were made with PyMol software (DeLano Scientific, San Carlos, CA).

Structure Determination of VEGF-E-
The crystallized protein represents the mature full-length VEGF-E variant NZ2, encompassing residues 20 -133 with a carboxyl-terminal hexahistidine tag. Amino-terminal sequencing of the purified protein confirmed correct proteolytic cleavage by Kex2 in P. pastoris. The protein was functional as judged by its capacity to form covalent homodimers in solution, to bind VEGFR-2 with high affinity, and to induce receptor activation in VEGFR-2expressing cells (data not shown and supplemental Fig. 2). The structure was determined by x-ray crystallography using the sulfur-SAD method. 6 The model was built into electron density maps calculated from experimental phases using Moloc (39) and refined with Refmac 5.2 (40). The crystallographic R-factor for the final model and all available data between 50 and 2.3 Å is 22.3%, and the corresponding free R-factor for 5% of the data (2687 reflections) is 24.7%. The model has good geometry with small root mean square (r.m.s.) deviations from ideal values for bond lengths and bond angles. PROCHECK analysis (41) showed no residues in disallowed regions in the Ramachandran plot. Statistical figures are given in Table 1.
The asymmetric unit of the VEGF-E crystals contains two homodimers (named AB and CD) that are related by noncrystallographic 2-fold symmetry. The VEGF-E residues in the structure are numbered according to the numbering of VEGF-A. The final model comprises residues A11-A82, A90 -A109, B13-B107, C11-C41, C44 -C82, C91-C110, and D14 -D106. Residues A84 -A89, C42-C43, and C82-C90 are disordered. Furthermore, two N-acetylglucosamines (Asn 75B and Asn 75D ), three benzamidines, three chloride and two sulfate ions, eight glycerol molecules, and two Tris molecules are present. In each monomer, Pro 49 is in the cis-configuration. The additional carboxyl-terminal residues originating from the His 6 tag are disordered in all monomers.
VEGF-E consists of a homodimer that is covalently linked by two intermolecular disulfide bonds between Cys 51 and Cys 60 . The two monomers are related by a noncrystallographic 2-fold axis perpendicular to the plane of the ␤ sheets. Each monomer contains a central antiparallel ␤ sheet formed by two pairs of twisted antiparallel ␤ strands, named ␤ 1 , ␤ 2 , ␤ 3 , and ␤ 4 , with the characteristic cysteine knot described also for other related growth factors (11,(42)(43)(44)(45)(46). The knot consists of an eight-residue ring formed by the backbone atoms of residues 57-61 and 102-104 and two intramolecular disulfide bridges, Cys 57 -Cys 102 and Cys 61 -Cys 104 , and a third disulfide bridge (Cys 26 -Cys 68 ) that passes almost perpendicularly through the center of the ring (Fig. 1A). Each VEGF-E monomer contains an amino-terminal ␣ helix and three solvent-accessible loop regions, L1, L2, and L3, connecting strands ␤ 1 to ␤ 2 , ␤ 2 to ␤ 3 , and ␤ 3 to ␤ 4 , respectively. The segment that connects strand ␤ 1 to ␤ 2 contains a single turn of an ␣ helix (␣2). As observed in the other cysteine knot structures, ␤2 and ␤4 are connected by only one main chain hydrogen bond between Met 55 and His 99 . Asn 75 in the monomers B and D is glycosylated, whereas the A and C monomers lack the N-acetylglucosamine moiety.
The Noncrystallographic Dimer-Dimer Interface-The two homodimers in the asymmetric unit superimpose with an r.m.s. deviation of ϳ0.5 Å (181 C ␣ atoms) (Fig. 1, A and B). The interface across the noncrystallographic 2-fold axis of the two dimers features extensive hydrophobic, polar, and ionic contacts with a total buried surface area of ϳ2400 Å 2 . Residues involved in dimer-dimer formation are located on the aminoterminal helix ␣1 and on the loop segments L1, L2, and L3. The core of the interface consists of a hydrogen bond network formed by the side chains of Asn 48B , Asn 48D , Asn 89B , and Asn 89D and 11 water molecules (Fig. 1C). Several hydrophobic contacts (Leu 21A -Leu 81D , Leu 21C -Leu 81B , Trp 17A -Met 91B , Trp 17C -Met 91D , Pro 49B , and Pro 49D ) cover the hydrogen bond network like a lid. At the bottom of the interface one benzamidine molecule is stacked between the -planes of two salt bridges, Arg 46D -Glu 64A and Arg 46B -Glu 64C . Two additional benzamidine molecules form essential crystal contacts through -stacking and hydrogen bond interactions at either end of the interface. The rest of the interface is formed by hydrogen bonds (Fig. 1C). Of all of the 15 residues involved in the dimer-dimer contacts, only three are strictly conserved among the different VEGF homologs, namely Cys 26 , Pro 49 , and Cys 104 . Whether the tetrameric structure observed in the crystal has any biological significance or simply reflects a crystallization artifact remains unclear at this point. Gel filtration analysis of the purified protein yields an apparent molecular mass of ϳ48 kDa, which is close to the expected molecular mass of a tetramer, whereas analysis by static light scattering clearly shows the molecular weight of a dimer of 30 kDa.
Structural Relationship between VEGF-E and Homologs-The structure of the core region of VEGF-E is very similar to those published for VEGF-A, PlGF, Vammin, and VR-1 in their free and liganded states and can be superimposed with an r.m.s. deviation of ϳ1.0 Å (112 C ␣ atoms). Conformational differences are observed in loops L1, L2, and L3 ( Fig. 2A). The segment from Val 35 to Gln 45 , which includes the single turn helix ␣2 and loop L1, deviates from the conformations observed in all other known structures with the exception of the PlGF/VEGFR-1 domain 2 structure.
Loop L2 is highly conserved among all VEGF homologs with variations in only three residues. All molecules that bind VEGFR-2 have either Asn or Thr at position 62; in PlGF, this position is occupied by Gly. Interestingly, in the structure of VEGF-E, we observed a peptide flip at Asn 62 , which has previously been observed in the crystal structures of unliganded VEGF-A, Vammin, and VR-1, but not in the PlGF structures. These differences are at the putative interaction site with domain 3 of receptor 2. Without structural information for such a ligand-receptor complex, it remains unclear why loop L2 plays such a crucial role in binding to receptor 1. Compared with all other homologs, loop L3 of VEGF-E appears intrinsically flexible due to the high glycine and serine content but is ordered in monomers B and D at the dimer-dimer interface. Its conformation differs remarkably from those in the other VEGF family proteins, and the relative orientation of L1 and L3 is not fixed in VEGF-E. In all available crystal structures, L1 and L3 are connected by an extensive hydrogen bond network and by hydrophobic contacts, suggesting that these loops form a structural entity essential for receptor binding. In VEGF-E monomers B and D, L1 is stabilized through a single hydrogen bond between the side chains of Gln 45 and Gln 92 . O⑀1 of Gln92 is further hydrogen-bonded to the main chain amide of Ala 83 , whereas the tips of loops L1 and L3 are not in contact with each other. In monomers A and C, the two side chains are too far apart for optimal hydrogen bonding, since L3 is solvent-exposed and disordered in both chains (Fig. 2B). In PlGF, the relative orientation between these loops is maintained through an extensive hydrogen bond network that involves the side chains of Lys 82 , Glu 42 , and His 45 and the backbone of Glu 42 and His 45 (Fig. 2B).
Receptor Binding Properties of VEGF-E/PlGF Chimeric Proteins-Whereas loop L2 is highly conserved among all VEGF homologs and is therefore unlikely to determine receptor specificity, the orientation of the loops L1 and L3 of VEGF-E differs significantly from that of VEGF-A and PlGF. We therefore investigated the role of these loops in receptor binding, constructing chimeric VEGF-E variants in which the sequence 42 LTSQR 46 in loop L1 was replaced by the corresponding sequence EVEHM of PlGF and in which 82 GASGSGS 88 in loop L3 was replaced with KIRSGDR of PlGF and a mutant where both loops were replaced (Fig. 4). The loop segments were chosen based on the differences observed in the superimposed structures of VEGF-E and PlGF. In addition, we made the mutant VE R46I by exchanging Arg 46 with the corresponding amino acid in VEGF-A, Ile. This mutant was constructed to assess the importance of the previously postulated salt bridge between Arg 46 and Glu 64 that was suggested to block binding of VEGF-E to VEGFR-1 (29). Dimer formation of VEGF-E NZ2 mutants was determined by SDS-PAGE under reducing and nonreducing conditions after the proteins were deglycosylated with endoglycosidase F (supplemental Fig. 1). The monomers had an apparent M r in the range of 16 -19 kDa, and the dimers ranged from 25 to 29 kDa. These results show that all mutants assume the expected disulfide-bonded dimeric structure.
Competitive binding assays were performed on PAE cells expressing either VEGFR-1 or -2 but lacking neuropilin-1. Binding of radiolabeled VEGF-A 165 to VEGFR-2 was competed with a 1000-fold excess of cold VEGF-E but not with the loop mutants VE L1 , VE L3 , or VE L13 or PlGF-1. This shows that exchanging either one or both of these loops abolished binding to VEGFR-2. Interestingly, competition with VE R46I  50 values were determined with Prism 4.0 software using nonlinear regression. C, binding affinities were determined by surface plasmon resonance using a BIAcore 3000 apparatus. Dimeric extracellular VEGFR-2 domains were immobilized on a CM5 chip, and analytes were injected in concentrations ranging from 0.025 to 1.6 nM. k on and k off were simultaneously fitted to kinetic measurements using the program BIAevaluation 4.1. For the affinity constants of VEGF-E L3 , a drifting base line was included in the calculation. was more efficient than with the native protein, indicating that this viral variant is not fully optimized for receptor binding (Fig. 3A).
Quantitative competition assays were performed with cold ligands over a concentration range of 6 orders of magnitude. Cold VEGF-A 165 competed binding of the radiolabeled ligand with an EC 50 of 1.0 nM (0.9 -1.1 nM; 95% confidence interval); for VEGF-E and VE R46I , values of 80.6 (71.7-90.6) nM and 16.2 (14.7-17.8) nM were calculated, respectively (Fig. 3D). Substitution of Arg 46 to Ile in L1 of VEGF-E resulted in a 5-fold decrease of the EC 50 . However, none of these VEGF-E variants was as potent as VEGF-A 165 in displacing the radioligand from VEGFR-2 on PAE cells. To study the differences in affinity of VEGF-A 165 and VEGF-E in more detail, we determined the k off and k on (supplemental Fig. 2) with surface plasmon resonance. Soluble receptor proteins consisting of the seven extracellular Ig-like domains of VEGFR-2 fused to a leucine zipper motif were produced in HEK293T cells, and such dimeric receptor proteins were then used to study the kinetics of ligand binding. As expected, VE L3 bound only weakly to the receptor, and unambiguous determination of the K d was not possible (Fig.  3C). The K d values for VEGF-E (432 pM) and VE R46I (221 pM) differed 2-fold, indicating that a more stable complex is formed with VE R46I . This was attributed to the smaller k off rate of VE R46I . The higher affinity of VEGF-A 165 resulted from the combined effect of a higher k on and a lower k off rate, resulting in a rise of the K d to 56 pM. These data show that Ile 46 , which is a critical determinant for high affinity binding of VEGF-A to VEGFR-2 (47), also contributes to binding of VEGF-E to this receptor.
We also performed competition experiments with 125 I-PlGF-1 on PAE cells expressing VEGFR-1. VEGF-A 165 , PlGF-1, VE L13 , and VE R46I showed specific binding (Fig. 3B), whereas single loop replacement in VEGF-E completely abrogated binding to VEGFR-1. At 100 nM, VE R46I and VE L13 showed equal potential to displace 125 I-PlGF-1 from the receptor, whereas at lower concentrations, we observed clear differences. With an EC 50 of 4.2 (3.7-4.7) nM, VE L13 binding to VEGFR-1 was more than 8-fold better than that of VE R46I (EC 50 34.8 (31.9 -38.0) nM; Fig. 3E). PlGF-1 was competed with an EC 50 of 0.17 (0.16 -0.19) nM. These results underscore the importance of L1 and L3 in determining receptor specificity and demonstrate that subtle structural changes in the ligand-receptor interface drastically alter receptor binding and specificity of VEGF-E. The data also show that VEGF-E binds VEGFR-1, although with very low affinity. This finding has not been reported before, and its functional implications remain at present unclear.

DISCUSSION
Here we show the crystal structure of the full-length biologically active VEGF-E variant isolated from the Orf virus NZ2. VEGF-E NZ2 adopts a similar structure as the previously reported VEGF homologs, with significant structural differences only in loops L1 and L3 that presumably determine binding specificity to VEGF receptors. With our study, there are now five crystal structures of VEGFlike molecules available in the free form and two structures in complex with the minimal ligand-binding domain of VEGFR-1. Alanine mutagenesis of VEGFs identified the critical residues for high affinity receptor binding (47,48), but based on all of the available data to date, no conclusive structural model for receptor specificity of the various VEGF isoforms could be drawn. Fig. 4, A and B, shows the critical residues determining receptor binding of VEGF-A, PlGF, and VEGF-E NZ7. These residues presumably also regulate receptor interaction of VEGF-E NZ2, Vammin, and VR-1 (29,32,48). The three loop regions, L1 (magenta), L2 (slate), and L3 (orange), that determine the interaction with VEGFR-1 and -2 (48) are marked. The sequences highlighted in yellow and in blue (Fig. 4A) are the amino acids tentatively determined in earlier studies as the sites responsible for ligand binding to VEGFR-1 and VEGFR-2, respectively. Mercer et al. (29) proposed a very similar structural model for receptor binding of VEGF-A and VEGF-E despite extensive sequence variation among these molecules. In this model, a salt bridge across a groove formed between Arg 46 and Glu 64 in VEGF-E NZ2 was proposed to prevent binding to VEGFR-1. This 6.5-Å-wide groove connecting L1 with L2 in all VEGF dimers is believed to interact with the region linking domains 2 and 3 of VEGFR-1 and to position receptor domain 3 to contact L1 and L3 of VEGF (Fig. 5) (49). In our VEGF-E structure, the salt bridge postulated by Mercer et al. (29) is only observed across the dimer-dimer interface between two neighboring homodimers (Fig. 1C). The different conformations observed for Arg 46 and Glu 64 in the four independent VEGF-E monomers in the asymmetric unit suggests, however, that this salt bridge might also form in the VEGF-E homodimer. Our mutant VE R46I showed improved binding to both VEGF receptors, suggesting that Ile 46 indeed directly regulates receptor interaction. The walls of the groove in VEGF-A are formed by Asp 63 and Glu 64 on one side and Phe 36 , Ile 43 , and Ile 46 on the other side, whereas the floor is formed by Asp 34 and Ser 50 (Fig. 5). Mutation of the two corresponding residues Asp 63 and Glu 64 in PlGF-1 to alanine gave rise to a mutant defective for VEGFR-1 binding (50). Similarly, in VEGF-A, loop L2 with Asp 63 , Glu 64 , and Glu 67 was a critical determinant for high affinity binding to VEGFR-1 (48). Interestingly, the overall structure of this groove is also conserved in the recently reported crystal structures of Vammin and VR-1 with the exception of Asp 34 (Pro in Vammin, Ser in VR-1), yet both molecules exclusively bind to VEGFR-2 (11). Moreover, in VEGF-E NZ2, Asp 63 and Glu 64 , which form one side of this groove, are conserved in comparison with VEGF-A and PlGF, but this molecule does not bind to VEGFR-1. This may result from the structural changes on the opposite side of the groove formed by Ser 36   (highlighted in green in Fig. 5). The binding data of the mutant VE R46I show that receptor binding affinity of such viral VEGF-E variants can be further optimized.
VEGF-E, where L1 was replaced by the sequence from PlGF, did not bind VEGFR-2, in agreement with the work of Kiba et al. (32), who generated similar chimeras with the NZ7 variant of VEGF-E. These authors also showed that the VEGF-A mutants I43A and I46A bound VEGFR-2 with reduced affinity. Loop L3 shows virtually no sequence homology in all VEGFs but is crucial for receptor binding, as shown by our mutant VE L3 , which binds neither VEGFR-1 nor -2. The most interesting finding is that VE L13 , where both loops L1 and L3 were exchanged by sequences from PlGF, shows drastically reduced binding to VEGFR-2 yet interacts, although with lower affinity than VEGF-A or PlGF, with VEGFR-1. VE L13 is, to our knowledge, the first artificially engineered VEGF mutant where such minimal sequence replacement established new receptor specificity, documenting the progress in protein engineering technology in the past decade. Our data are reminiscent of earlier published work where structure-based design of receptor antagonists and of ligands with improved receptor specificity have been described (51)(52)(53). Similar approaches have been described for the design of enzymes with new substrate specificity or functionality (54 -56).
In summary, our data clearly suggest that loop L2 with Asp 63 , Glu 64 , and Glu 67 (His 67 in PlGF) is essential for binding to VEGF receptors, whereas the distinct orientation of L1 and L3 determines VEGF receptor specificity. The particular combination of structural motifs defined by these loops might be central for receptor specificity. Receptor-ligand complexes analyzed from co-crystals of VEGFR-2 bound to various isoforms of VEGF are needed for a final assessment of receptor specificity of VEGF family proteins and will be crucial for the development of receptor-specific VEGF antagonists.