Identification of vascular endothelial growth factor determinants for binding KDR and FLT-1 receptors. Generation of receptor-selective VEGF variants by site-directed mutagenesis.

Vascular endothelial growth factor (VEGF) expression in various cell types is induced by hypoxia and other stimuli. VEGF mediates endothelial cell proliferation, angiogenesis, vascular growth, and vascular permeability via the endothelial cell receptors, kinase insert domain-containing receptor (KDR)/fetal liver kinase 1 (Flk-1) and FLT-1. Alanine-scanning mutagenesis was used to identify a positively charged surface in VEGF that mediates binding to KDR/Flk-1. Arg82, Lys84 and His86, located in a hairpin loop, were found to be critical for binding KDR/Flk-1, while negatively charged residues, Asp63, Glu64, and Glu67, were associated with FLT-1 binding. A VEGF model based on PDGFb indicated these positively and negatively charged regions are distal in the monomer but are spatially close in the dimer. Mutations within the KDR site had minimal effect on FLT-1 binding, and mutants deficient in FLT-1 binding did not affect KDR binding. Endothelial cell mitogenesis was abolished in mutants lacking KDR affinity; however, FLT-1 deficient mutants induced normal proliferation. These results suggest dual sets of determinants in the VEGF dimer that cross-link cell surface receptors, triggering endothelial cell growth and angiogenesis. Furthermore, this mutational analysis implicates KDR, but not FLT-1, in VEGF induction of endothelial cell proliferation.

VEGF 1 is a specific mitogen for vascular endothelial cells in vitro and a potent angiogenic and vascular permeability enhancing factor in vivo (1)(2)(3)(4)(5). VEGF, also known as vascular permeability factor, is expressed in response to hypoxia and other stimuli by a variety of differentiated cells (1, 2, 6 -14). VEGF expression has also been found in numerous human and rodent transformed cells (3,4,(15)(16)(17)(18)(19). VEGF is encoded by a single gene, but it exists in four isoforms of 121, 165, 189, and 206 amino acids due to alternative mRNA splicing (20,21). The two low molecular weight forms, VEGF 121 and VEGF 165 , are secreted as soluble factors, while the other higher molecular weight forms, VEGF 189 and VEGF 206 , are secreted but remain bound to extracellular matrix (22). Of the four isoforms, VEGF 165 is the most abundantly expressed splice variant.
VEGF 165 is a heparin binding glycoprotein with a single glycosylation site (at Asn 75 ) and is secreted as a homodimer of approximately 45 kDa (3,17). VEGF 165 can be cleaved by plasmin to yield VEGF 110 , which is equipotent to VEGF 121 with respect to mitogenic activity on endothelial cells (22,23). The amino acid sequence of VEGF 110 exhibits limited but significant homology (17%) with PDGF. All eight cysteine residues, which are involved in intra-or interchain disulfides, are conserved among these growth factors. A cDNA sequence encoding a similar protein was identified by screening a human placental cDNA library (24). Placental growth factor (PLGF), with 47 and 18% sequence homology with VEGF and PDGF, respectively, is recognized as a member of the PDGF/VEGF family of growth factors.
Two tyrosine kinases, fms-like tyrosine kinase (FLT-1) and the kinase insert domain-containing receptor (KDR/Flk-1), have been identified as high affinity VEGF receptors (25)(26)(27). The murine homolog of KDR, fetal liver kinase 1 (Flk-1) has 85% homology with the human VEGF receptor (28). These receptors are localized on the cell surface of various endothelial cell types (29,30). The FLT-1 and KDR/Flk-1 genes encode tyrosine kinase receptors, characterized by an extracellular domain containing seven immunoglobulin-like domains and a split tyrosine kinase intracellular domain (31). In many aspects, the VEGF receptors are structurally and functionally similar to the PDGF receptors, which have five immunoglobulin-like repeats in the extracellular domain (32,33). In addition to VEGF, PLGF has also been demonstrated to bind FLT-1 with high affinity, but not to KDR/Flk-1 (34,35). VEGF is a potent endothelial cell mitogen in vitro, while PLGF is 3 orders of magnitude less potent at inducing endothelial cell proliferation (4,(15)(16)(17)(18). However, PLGF can significantly potentiate the in vitro and in vivo activity of low concentrations of VEGF (34).
To gain a better understanding of the biological activity of VEGF, we began an analysis of structure/activity relationships using site-directed mutagenesis of VEGF. In the current study, we report on the mutational analysis of VEGF and identify the receptor binding determinants for KDR/Flk-1 and FLT-1. Furthermore, the epitope mapping of a neutralizing monoclonal antibody is also described. We used the strategy of charged-toalanine scanning mutagenesis to evaluate the role of various sites on VEGF in receptor binding. To confirm the results of alanine scanning, we have also introduced novel glycosylation sites in VEGF to inhibit receptor binding with additional carbohydrate. We report that KDR and FLT-1 receptors bind to different sites on VEGF 165 , which may serve to dimerize tyrosine kinase receptors resulting in endothelial mitogenesis, proliferation, and initiation of angiogenesis and vasculogenesis.
Site-directed Mutagenesis and Expression of VEGF Variants-Mutants were prepared using the Muta-Gene Phagemid in vitro mutagenesis kit according to the method of Kunkel and co-workers (36,37). A plasmid vector pRK5 containing cDNA for VEGF 165 isoform was used for mutagenesis and transient expression. The pRK5 vector is a modified pUC118 vector and contains a cytomegalovirus enhancer and promoter (38,39). The mutagenized DNA was purified using the Qiagen Plasmid Midi kit tip 100, and the sequence of the mutations was verified using Sequenase version 2.0 kit. The mutated DNA was analyzed by restriction enzyme digestion as described by Sambrook et al. (40). Transient transfection of human fetal kidney "293" cells was performed in 6-well plates using the modified calcium phosphate precipitate method as described previously (42)(43)(44). 2 Briefly, approximately 1.2 ϫ 10 6 cells were incubated overnight at 37°C in the presence of 15 g of precipitated DNA. Cell culture supernatant was replaced with serum-free medium, and cell monolayers were incubated for 72 h at 37°C. Conditioned media (3 ml) was harvested, centrifuged, aliquoted, and stored at Ϫ70°C.
Quantitation of VEGF 165 Variants by ELISA-A radioimmunometric assay described previously (45) was adapted for the quantitation of VEGF mutants. Individual wells of a 96-well microtiter plate were coated with 100 l of a 3 g/ml solution of an anti-VEGF 165 polyclonal antibody in 50 mM sodium carbonate buffer, pH 9.6, overnight at 4°C. The supernatant was discarded, and the wells were washed 4 times with PBS containing 0.03% Tween 80. The plate was blocked in assay buffer (0.5% BSA, 0.03% Tween 80, 0.01% Thimerosal in PBS) for 1 h (300 l/well) at 25°C and then washed. Diluted samples (100 l) and VEGF 165 standard (ranging from 0.1 to 10 ng/ml) were added to each well and incubated for 1 h at at 25°C. The supernatant was discarded, and the wells were washed. Anti-VEGF murine monoclonal antibody 5F8 solution (100 l at 1 g/ml) was added, and the microtiter plate was incubated for 1 h at 25°C. Supernatant was discarded, the plate was washed, and horseradish peroxidase-conjugated goat IgG specific for murine IgG (100 l) at 1:25000 dilution was immediately added to each well. The plate was incubated for 1 h at 25°C, after which the supernatant discarded, and the wells were washed and developed with orthophenylenediamine (0.04%), H 2 O 2 (0.012%) in 50 mM citrate phosphate buffer, pH 5 (100 l), and then incubated in the dark at 25°C for 10 min. The reaction was stopped by adding 50 l of 4.5 N H 2 SO 4 to each well, and the absorbance was measured at 492 nm on a microplate reader (SLT Labs). The concentrations of VEGF 165 variants were quantitated by interpolation of a standard curve using nonlinear regression analysis. For purposes of comparison, a second ELISA was developed that utilized a dual monoclonal format. The assay was similar to the above described ELISA, except a neutralizing monoclonal antibody (mAb A4.6.1) was used to coat the microtiter plates (46).
Immunoblotting of VEGF Mutants-Aliquots of conditioned cell media (16 l) containing VEGF or VEGF mutant (ϳ10 ng) were added to 5 ϫ SDS sample buffer (4 l) and heated at 90°C for 3 min prior to loading on SDS gradient (4 -20%) gels. Gels were electrophoresed and transferred to nitrocellulose paper in a Bio-Rad tank blotter containing BufferEZE with 0.1% SDS for 90 min at 250 mA at 25°C. Transferred immunoblots were blocked in PBS overnight at 4°C with 1.0% BSA, 0.1% Tween 20 (blocking buffer). A solution containing five murine anti-VEGF mAbs (A.4.6.1, 5C3, 5F8, 4D7, and 2E3) was prepared with 2 g/ml of each mAb in blocking buffer and used as primary antibody. Immunoblots were incubated with primary antibody for 4 h at 25°C and then washed 3 times for 10 min in blocking buffer at 25°C. 125 I-Labeled Protein A (10 4 cpm/ml in blocking buffer) was incubated with the immunoblots for 1 h at 25°C. Immunoblots were washed 3 times for 10 min in blocking buffer at 25°C and then dried on filter paper and placed on Kodak X-Omat film with two intensifying screens at Ϫ70°C for 3 days.
Preparation of 125 I-Labeled VEGF 165 -Radiolabeling of CHO-derived VEGF 165 was prepared using a modification of chloramine T catalyzed iodination (47). In a typical reaction, 10 l of 1 M Tris-HCl, 0.01% Tween 20 at pH 7.5 was added to 5 l of sodium iodide-125 (0.5 mCi, 0.24 nmol) in a capped reaction vessel. To this reaction, 10 l of CHO-derived VEGF 165 (10 g, 0.26 nmol) was added followed by 10 l of 1 mg/ml chloramine T. After 60 s, iodination was terminated with sodium metabisulfite (20 l, 1 mg/ml). Radiolabeled VEGF was isolated on a PD-10 column (Sephadex G-25) that was pre-equilibrated with 0.5% BSA, 0.01% Tween 20 in PBS. Typically, the specific radioactivity of the iodinated VEGF was 26 Ϯ 2.5 Ci/g, which corresponded to one 125 I/two molecules of VEGF 165 dimer.
VEGF 165 Receptor Binding Assay-The assay was performed in 96well immunoplates (Immulon-1); each well was coated with 100 l of a solution containing 10 g/ml of rabbit IgG anti-human IgG (Fc-specific) in 50 mM sodium carbonate buffer, pH 9.6, overnight at 4°C. After the supernatant was discarded, the wells were washed 3 times in washing buffer (0.01% Tween 80 in PBS). The plates were blocked (300 l/well) for 1 h in assay buffer (0.5% BSA, 0.03% Tween 80, 0.01% Thimerosal in PBS). The supernatant was discarded, and the wells were washed. A mixture was prepared with conditioned media containing VEGF 165 mutants at varying concentrations (100 l) and 125 I-radiolabeled VEGF 165 (ϳ5 ϫ 10 3 cpm in 50 l), which was mixed with VEGF receptor-IgG chimeric protein, FLT-1 IgG, or KDR-IgG (3-15 ng/ml, final concentration, 50 l) in micronic tubes. An irrelevant antibody such as humanized anti-HER 2 IgG (IgG1 subclass) was used as a control for nonspecific binding of radiolabeled VEGF. Aliquots of these solutions (100 l) were added to precoated microtiter plates and incubated for 4 h at 25°C. The supernatant was discarded, the plates were washed, and individual wells were counted by ␥ scintigraphy (LKB model 1277). The competitive binding between unlabeled VEGF 165 (or VEGF 165 mutants) and 125 I-radiolabeled VEGF 165 to the FLT-1 or KDR receptors were plotted and analyzed by four parameter fitting (Kaleidagraph, Abelbeck Software). The apparent dissociation constant for each VEGF mutant was estimated from the concentration required for 50% inhibition (IC 50 ). Nonspecific binding to an unrelated antibody was approximately 2% of maximal binding under similar conditions as binding to VEGF receptor-IgG fusion proteins.
Assay for Vascular Endothelial Cell Growth-The mitogenic activity of VEGF variants was determined by using bovine adrenal cortical endothelial cells as target cells as described previously (1). Briefly, cells were plated sparsely (7000 cells/well) in 12-well plates and incubated overnight in Dulbecco's modified Eagle's medium with 10% calf serum, 2 mM glutamine, and antibiotics. The medium was exchanged the next day, and VEGF or VEGF mutants diluted in culture media from 100 ng/ml to 10 pg/ml were layered in duplicate onto the seeded cells. After 5 days at 37°C, the cells were dissociated with trypsin and quantified using a Coulter counter.
Molecular Modeling-A model of VEGF was generated with the Insight II and Discover software from Biosym Technologies (San Diego, CA). The model of VEGF was based on the crystal structure of PDGFb solved by Oefner et al. (50). The sequences of VEGF and PDGF were aligned to determine which residues were to be replaced or deleted (see Fig. 1). The side chains of six amino acids were substituted with those residues found in VEGF (i.e. Asp 63 , Glu 64 , Glu 67 , Arg 82 , Lys 84 , and His 86 ). 2 M. Jordan, C. Koehne, and F. M. Wurm, manuscript in preparation.

Comparison of VEGF, PLGF, and PDGF Sequences-We
have previously described studies that localize the receptor binding determinants for FLT-1 and KDR/Flk-1 in the aminoterminal (1-110) dimer of VEGF (23). Plasmin catalyzes the cleavage of the carboxyl-terminal, heparin-binding region (111-165), releasing the VEGF 110 dimer, which displays bioactivity in the endothelial cell growth assay and in the Miles permeability assay (22). We compared the receptor binding region of VEGF (i.e. 1-110) with homologous proteins, PLGF, PDGFa, and PDGFb (Fig. 1). The sequences were aligned with respect to the eight cysteines shared by these proteins. Six cysteines form intrachain disulfides, and two cysteines participate in interchain covalent bonds (48,49). Two gaps, inserted in the VEGF and PLGF sequences, are located in external loops based on the crystal structure of PDGFb dimer (50). VEGF 110 shares 47, 15, and 19% sequence identity and 63, 24, and 28% similarity with PLGF, PDGFa, and PDGFb, respectively (51). Inspection of sequence similarity and divergence among these growth factors offers little insight as to the receptor binding determinants. We undertook the functional mapping of VEGF by site-directed mutagenesis.
Clustered Charged-to-Alanine Scan Mutagenesis-30 mutants of VEGF 165 were constructed by site-directed mutagenesis where groups of between one and four neighboring charged amino acids (Arg, Lys, His, Asp, and Glu) were replaced with alanine. Table I lists the specific amino acid substitutions for each mutant and indicates the mean residue number for the position of the mutation(s). For a given mutant, this number is the average of the altered positions. Plasmid DNA encoding these mutants was transiently transfected in human 293 kidney cells, and the amount of VEGF in the conditioned cell media was determined using two VEGF-specific immunochemical assays. In Fig. 2, the results of a polyclonal/monoclonal ELISA are compared with those obtained with a dual monoclonal assay. In the poly-/monoclonal assay, affinity-purified polyclonal antibody reacted with multiple epitopes, while the monoclonal antibody 5F8 is specific for determinants in the carboxyl-terminal, heparin-binding region (111-165) of VEGF. In contrast, the dual monoclonal ELISA utilized neutralizing and non-neutralizing monoclonal antibodies (mAbs A4.6.1 and 5F8, respectively). The use of two immunochemical detection methods assisted in the accurate determination of mutant VEGF concentration in conditioned media. For most VEGF mutants, the results of two immunochemical analyses were in good agreement, with transient expression levels ranging from 0.2 to 2 g/ml of VEGF antigen in the conditioned media. Nearly all VEGF mutants were expressed with variable yield for repetitive transfections, with the notable exception of the R56A mutant of VEGF. No immunopositive protein was detected with the R56A mutation despite reconstruction of the variant and numerous transfection attempts. It is interesting that arginine is strictly conserved at position 56 in VEGF, PLGF, and PDGF, suggesting that this amino acid plays a vital role in structural integrity and/or native protein folding (Fig.  1). Significantly, mutations in the region 82-86 were consistently underquantitated in the dual monoclonal ELISA compared with those results obtained with the poly-/monoclonal assay, indicating that the epitope recognized by the neutralizing monoclonal antibody, A4.6.1, includes this determinant in VEGF. The single amino acid substitution, R82A yields a mutant of VEGF exhibiting almost complete loss of immunochemical reactivity with mAb A4.6.1 (Fig. 2). The monoclonal immunoreactivity was abolished by the triple alanine mutation, R82A/K84A/H86A and by an extraglycosylation mutation, R82N/I83L/K84S, as indicated in Fig. 2 by the mutants with mean residue numbers 84 and 83, respectively. Partial loss of mAb A4.6.1 immunoreactivity was also observed with the combined mutant H90A/E93A VEGF (mean residue number 91.5). These data suggest that the epitope of a neutralizing monoclonal antibody is localized to a region of VEGF including amino acids 82-93.
Nonreducing, SDS-PAGE Analysis of VEGF Mutants-A representative set of transiently transfected supernatants (from 293 cells) containing approximately 10 -20 ng of VEGF or VEGF mutant were analyzed by nonreducing SDS-PAGE (Fig. 3). The gels were transferred and blotted as described under "Experimental Procedures," using a mixture of five monoclonal anti-human VEGF 165 antibodies. Autoradiography of the immunoblots indicated a major band at 45 kDa for wild type and mutant forms of VEGF. This immunopositive protein band co-migrated with purified, dimeric VEGF 165 derived from CHO cells. For some mutants of VEGF, and for 293 cell-derived wild-type VEGF (but not VEGF derived from CHO cells), there appeared an additional minor band at approximately 70 kDa. Apparent molecular weights for all charged-to-alanine replacement mutants, as indicated by SDS-PAGE, were equivalent to that observed for wild-type VEGF 165 derived from 293 or CHO cells. There was no indication of degraded forms of VEGF that would yield lower molecular weight species as has been observed for plasmin cleavage of VEGF (22,23).
VEGF Binding to KDR Receptor Is Primarily Mediated by Arg 82 , Lys 84 , and His 86 -The binding of VEGF mutants to soluble KDR-IgG was evaluated by competitive displacement of 125 I-labeled VEGF 165 in the absence or presence of heparin. The list of mutations is given in Table I. The results for 27 charged-to-alanine scan mutants of VEGF in studies of binding to KDR-IgG are shown in Fig. 4, plotted with respect to the position of the mutation(s). Wild-type VEGF 165 expressed in 293 cells and CHO cells were equivalent with respect to displacement of 125 I-labeled VEGF 165 in KDR binding. The concentrations required to achieve half-maximal inhibition (IC 50 ) were 31 and 29 pM for 293-or CHO-derived VEGF 165 , respectively (n ϭ 8 replicates each). The IC 50 values for wild-type VEGF were not significantly different in the absence versus the presence of 10 g/ml heparin.
Many of the mutant proteins exhibited binding comparable with wild-type VEGF. In fact, the binding to KDR for 19 out of 25 alanine scan mutants was similar to that of wild-type VEGF; the average IC 50 for mutants with wild-type phenotype was 29 Ϯ 18 pM (n ϭ 19). Three mutants (E42A, E44A, and D63A) exhibited an apparent 4 -6-fold increased binding to KDR receptor compared with wild-type VEGF. Since this effect (for two of these mutants) was observed for FLT-1 binding as well, the potential exists for under quantitation of selected mutants by the polyclonal/monoclonal ELISA. However, the most significant effect on binding was observed with the R82A, K84A, H86A mutant of VEGF, which exhibited 1000-fold decreased affinity for the KDR receptor in the absence of heparin, relative to that of wild-type VEGF (Fig. 4A). Interestingly, in the presence of heparin, the binding of this triple mutant was only 10-fold decreased compared with that of wild-type VEGF. These results are consistent with VEGF binding to KDR as a function of two sites of interaction, a heparin-independent site in the 1-110 dimer and a heparin-dependent binding site in the 111-165 domain. When the binding of VEGF 165 to KDR is mediated entirely by the 1-110 region (in the absence of heparin) the mutations at 82, 84, and 86 severely compromise the binding of VEGF to KDR.
To evaluate the relative contribution of individual residues, single amino acid substitution mutants of VEGF were constructed. The single mutations, R82A, K84A, and H86A were found to display more modest decreases with respect to KDR binding (Table I). R82A VEGF exhibited wild-type KDR binding, while K84A VEGF and H86A VEGF were approximately 5.5-and 1.8-fold decreased in binding compared with that of wild-type VEGF, respectively. While the 84 position of VEGF was most dominant in the triple alanine mutant, the combination of mutations at 82, 84 and 86 clearly exhibited a synergistic effect on the interaction with the KDR receptor. In addition to the major KDR binding determinant, a minor site was observed in the 63-67 region. The triple mutant, D63A/E64A/ E67A VEGF, was 2.4-and 3.0-fold reduced in binding to KDR in both the presence and absence of heparin. The single amino acid mutant E64A VEGF exhibited 7-fold decreased affinity for KDR, which implicated this site more than neighboring mutations in this region, i.e. D63A and E67A. Although modest in comparison to the major effects observed with R82A/K84A/H86 VEGF, the most potent effects with single alanine replacement of charged amino acids were observed for E64A VEGF and K84A VEGF.
VEGF Binding to FLT-1 Receptor Involves Interaction with Asp 63 , Glu 64 , and Glu 67 -As was observed for KDR binding, most of the alanine scan mutants of VEGF bound FLT-1 with similar affinity as wild-type VEGF (Fig. 4B). The IC 50 values for wild-type VEGF were 22 Ϯ 8 and 15 Ϯ 8 pM in the absence and presence of heparin, respectively (n ϭ 13). Analysis of alanine scan VEGF mutants indicated two sites of interaction with FLT-1 that co-localized with the KDR binding determi- nants. A major site for FLT-1 binding involves the 63-67 region of VEGF as indicated by the 30-fold reduction in affinity with D63A/E64A/E67A VEGF in the absence of heparin. This is in contrast to the results with KDR, which indicated that mutations in the 63-67 region of VEGF exhibited only modest effects on KDR binding. The major site of KDR interaction (82-86 region) yielded only minor effects with respect to FLT-1 binding. Additional mutational sites at the carboxyl terminus were associated with minor effects on FLT-1 binding. The relative roles of major and minor receptor binding sites are reversed for FLT-1 in comparison with that for KDR. The Effect of Glycosylation on Receptor Binding-An unglycosylated form of VEGF was constructed, expressed in 293 cells, and visualized by SDS-PAGE and immunoblotting (Fig.  3). This mutant, N75A VEGF, appeared to have a lower molecular weight consistent with the lack of glycosylation at position 75. The binding of N75A VEGF was indistinguishable from that of wild-type VEGF for both KDR and FLT-1. For the wild-type protein, N-linked carbohydrate at Asn 75 does not appear to play a role in mediating VEGF receptor binding.
We inserted potential neoglycosylation sites to observe the effects of carbohydrate addition at or near putative receptor binding sites. Surface accessible sites were considered optimal in exterior loops or turns as predicted on the basis of the crystal structure of PDGFb dimer (50). One such site (42-44 region) was selected as a control since no receptor binding determinants were identified in this region by charged-to-alanine scanning mutagenesis. The neocarbohydrate site in E42N/E44S VEGF was apparently glycosylated, as indicated by the increased molecular weight observed on SDS-PAGE immunoblots (Fig. 3). The N-linked carbohydrate at position 42 did not interfere with binding to KDR or FLT-1 receptors as indicated by IC 50 values of 15 and 13 pM, respectively.
Potential Glycosylation Site at Position 82 Results in Severely Decreased KDR Binding-The extent of extraglycosylation was not apparent on the immunoblot for R82N/I83L/K84S VEGF (Fig. 3). Although the R82N/I83L/K84S mutation had minimal effect on apparent molecular weight, the effect on KDR binding was quite significant. R82N/I83L/K84S VEGF exhibited only partial displacement of the labeled VEGF in KDR binding (Fig. 5A). The half-maximal inhibitory concentration for R82N/ I83L/K84S VEGF was estimated to be 10,000 pM. However in the presence 10 g/ml heparin, the relative affinity of R82N/ I83L/K84S VEGF for KDR was 50-fold decreased compared with that of wild-type VEGF. Interestingly, this putative extraglycosylation mutant exhibited normal affinity for FLT-1 (Fig. 5B). Mutations in the 82-86 region (R82A, K84A, H86A, and R82N/I83L/K84S) confer significantly decreased interaction with KDR and normal binding to FLT-1.
Extraglycosylation at Position 64 Decreases FLT-1, but Not KDR Binding-E64N/L66S VEGF was observed as a faint band with apparent increased molecular weight on SDS-PAGE (Fig.  3). E64N/L66S VEGF exhibited approximately 45-fold decreased binding to FLT-1 in the presence or absence of heparin, respectively (Fig. 5B, and Table I). This mutant having FLT-1-specific effects exhibited modestly decreased binding with KDR receptor. The relative binding of D63A/E64A/E67A VEGF and E64N/L66S VEGF to soluble KDR was approximately 3and 5-fold decreased, respectively. These changes are small in comparison with their effects on FLT-1 binding, where this region is the major binding determinant.
VEGF Mutants with Decreased KDR Receptor Binding Are Weak Endothelial Cell Mitogens-Mitogenic activities of VEGF and mutants of VEGF were determined using bovine adrenal cortical capillary endothelial cells. Wild-type VEGF, derived from 293 cells or CHO cells, induced half-maximal proliferation at 28 Ϯ 10 pM (n ϭ 6) and 16 Ϯ 8 pM (n ϭ 9), respectively. Conditioned cell media from mock-transfected 293 cells did not induce endothelial cell proliferation. The half-maximally effective concentrations (EC 50 ) for most VEGF mutants were similar to wild-type VEGF (Table I). Significant effects on endothelial cell proliferation were observed with mutations in the 82-86 region. The mitogenic potency of R82A/K84A/H86A VEGF was decreased 20-fold compared with wild-type VEGF. Proliferation by R82N/I83L/K84S VEGF was reduced to such an extent that 50% of maximal growth was not achieved at the highest concentration tested (Fig. 6). To quantitate the potency of R82N/I83L/K84S VEGF, we compared the concentration of the mutant required to achieve 20% of maximal VEGF-induced stimulation. The difference in EC 20 values for wild-type VEGF and R82N/I83L/K84S VEGF (4 pM versus 230 pM, respectively) indicated 60-fold reduced potency for the mutant with a neoglycosylation site in the region specific for KDR binding. The effect of these mutations on endothelial cell growth is consistent with the KDR binding data. The affinity of R82A/K84A/ H86A VEGF and R82N/I83L/K84S VEGF with soluble KDR (in the presence of heparin) was reduced 10-and 50-fold, respectively, compared with that observed with wild-type VEGF. Since endothelial cells in vitro express surface and matrix associated heparan sulfate-containing proteoglycans (53), it is appropriate to compare the mitogenic response of endothelial cells to VEGF or VEGF mutants with the binding data for soluble VEGF receptors in the presence of heparin. Taken together, the mutational analysis of VEGF by alanine scanning and extra-glycosylation provide strong evidence that binding to KDR receptors on endothelial cells is a triggering event for the induction of proliferation observed with VEGF.
VEGF Mutants with Decreased FLT-1 Receptor Binding Are Fully Active Endothelial Cell Mitogens-Alanine scan substitutions in the 63-67 region of VEGF were shown to have normal binding to KDR and decreased binding to FLT-1 (Figs. 4 and 5). Triple and single alanine mutants (D63A/E64A/E67A VEGF, D63A VEGF, E64A VEGF, and E67A VEGF) were evaluated for induction of endothelial cell growth. All of these mutants exhibited mitogenic potency similar to that of wildtype VEGF (Fig. 6 and Table I). The mutant with a putative extraglycosylation site in the 63-67 region; E64N/L66S VEGF also exhibited normal activity on endothelial cells. These data reinforce the observation that FLT-1-deficient mutants of VEGF induce endothelial cell proliferation similar to wild-type VEGF. Furthermore, these data suggest that VEGF binding to FLT-1 receptors on endothelial cells is unrelated to mitogenesis and proliferation. This mutational analysis has identified VEGF variants that are relatively selective for KDR or FLT-1 receptors. The data in this report suggest an electrostatic component of VEGF-receptor interaction, such that the determinants for KDR and FLT-1 include predominantly positive or negatively charged regions of VEGF, respectively.

DISCUSSION
KDR and FLT-1 receptor binding determinants on VEGF, identified by site-directed mutagenesis, were localized on a model of VEGF based on the crystal structure of a homologous protein, PDGFb. VEGF and PDGFb display 28% amino acid sequence homology with conserved alignment of eight cysteines within the receptor binding domain (i.e. 1-110 of VEGF). The crystal structure of PDGFb dimer has been elucidated at a 3.0-Å resolution (50). A model of the three-dimensional structure of PDGFb dimer is shown (see Fig. 7), with the appropriate amino acids of VEGF highlighted that are involved in receptor binding as identified by mutational analysis.
The polypeptide chain of each monomer is folded into two twisted anti-parallel pairs of ␤-strands (shown as green ribbons, Fig. 7). The amino termini of the PDGFb dimer were not resolved, and the structure begins at residue 17 in VEGF numbering. Three intramolecular disulfide bonds are arranged in a knotted configuration, shown in yellow for Cys 26 -Cys 68 , Cys 57 -Cys 102 , and Cys 60 -Cys 104 (VEGF numbering, see Ref. 48 for PDGF disulfides). Two intermolecular disulfide bonds form the covalent linkage of the homodimeric structure via cysteines at positions 51 and 60. Dimerization leads to an elongated structure with three surface loops that cluster at each end of the molecule. Oefner and co-workers (50) suggest that these loops would probably form receptor recognition sites for PDGF. The surface loops, denoted as I, II, and III in PDGFb, include the corresponding regions of VEGF from 36 to 46, 61 to 68, and 84 to 87, respectively (VEGF numbering). Loop I is not resolved in the PDGFb structure and is denoted by a gap in the peptide strand from residues 37 to 45 (VEGF numbering). Loops II and III in VEGF are involved in receptor recognition as indicated by the binding studies with mutants D63A/E64A/E67A VEGF and R82A/K84A/H86A VEGF, respectively. Interestingly, the loop II residues Asp 63 , Glu 64 , and Glu 67 that mediate VEGF binding predominantly to FLT-1, form a negatively charged surface at one end of each monomer (residues 63, 64, and 67 shown in red, see Fig. 7). In contrast, the loop III associated residues, Arg 82 , Lys 84 , and His 86 , which predominantly mediate binding to KDR, cluster to form a positively charged surface at the other end of the VEGF monomer (residues 82, 84, and 86 shown in blue). These oppositely charged surface loops are at distal ends of the monomer, but they are in close proximity in the dimeric form of VEGF. Although the receptor binding regions of loop II and III predominantly mediate binding of FLT-1 and KDR, respectively, there is indication that KDR interacts with residues in both loops. For example, the VEGF mutant with a potential extraglycosylation site at position 64 exhibits reduced affinity for KDR as well as for FLT-1 receptors (ϳ6-and 40fold, respectively). Altogether, the functional analysis of VEGF by site-directed mutagenesis indicates that the VEGF dimer exhibits receptor binding regions at both ends of the protein.
Two sets of receptor binding sites on VEGF interact predominantly with FLT-1 or KDR. These sites, shown schematically as A and B in Fig. 8, display binding to KDR and FLT-1, respectively. Site A, composed of basic amino acids (including Arg 82 , Lys 84 , and His 86 ) predominantly mediates the interaction of VEGF with KDR. It is interesting to note that in recent studies, a cationic series of amino acids in a pentapeptide sequence of PDGF (Val 158 -Arg 159 -Lys 160 -Lys 161 -Pro 162 , PDGF numbering) has been demonstrated to play a critical role in ligand binding to the PDGF␣ receptor (52). This sequence is perfectly conserved between PDGFa and PDGFb, but is not found in either VEGF or PLGF (see Fig. 1). The cationic sequence is located in the loop III region of the PDGF structure, which corresponds to amino acids 80 -90 of VEGF. PDGF and VEGF appear to share a requirement for basic amino acids in the loop III region (site A) in binding to PDGF␣ receptor and KDR, respectively. These data do not exclude significant con-  (50). Shown as an oval ribbon diagram is the polypeptide backbone in green, and disulfides in yellow. The ribbon begins with residue 17 (VEGF numbering) and extends to residue 109 with a gap in sequence due to unresolved structure (from 37 to 45). The side chains of acidic amino acids involved in FLT-1 receptor binding are shown in red for Asp 63 , Glu 64 , and Glu 67 . The side chains of basic amino acids involved in KDR receptor binding are shown in blue for Arg 82 , Lys 84 , and His 86 . PDGFb has three additional amino acids; Arg, Lys, Lys (shown in gray) that are conserved in PDGFa, but not VEGF or PLGF (see Fig. 1). tributions by noncharged amino acids, such Gln 79 , Ile 80 , Met 81 , Ile 83 , and/or Pro 85 .
The B site of VEGF, composed of acidic residues (including Asp 63 , Glu 64 , and Glu 67 ) is shown in Fig. 8 as weakly interacting with the KDR receptor. This second site, however, appears to be directly involved with the binding of VEGF to FLT-1. LaRochelle et al. (54), utilizing site-directed mutagenesis of PDGF, have described similar findings with respect to mapping the ␣/␤ receptor binding determinants. Mutations of Asn 136 and Arg 137 in PDGF, which correspond to Asp 63 and Glu 64 of VEGF (see Fig. 1), have been shown to be essential for the ␣/␤ receptor specificity of PDGFa and PDGFb. In the present study, the effect of mutations within the loop II of VEGF (site B) did not strongly affect the binding of VEGF 165 to FLT-1 as compared with the effect of loop III mutations (site A) on KDR binding. The alanine replacement of Asp 63 , Glu 64 , and Glu 67 resulted in 30-fold decreased affinity with FLT-1, whereas the R82A/K84A/H86A mutant exhibited 1000-fold decreased binding to KDR (see Table I). Noncharged amino acids such as Asn 62 and Leu 66 may also contribute to the interaction of VEGF with FLT-1. These data suggest additional site(s) of FLT-1 interaction, possibly in the carboxyl-terminal domain, residues 111-165 of VEGF. Amino acids in site A appear not to interact with the FLT-1 receptor, as indicated by the wild-type phenotype associated with R82A/K84A/H86A and R82A/I83A/ K84A mutations.
The 2-fold symmetry of PDGFbb localizes the three loop regions on either end of the molecule (50). The evidence of receptor binding sites at both ends of the dimer is consistent with the growth factors of the PDGF/VEGF family exhibiting bivalency. Fretto et al. (55) have established that PDGF is bivalent with respect to PDGF receptor binding (55). Bivalent binding of VEGF provides a mechanism for dimerization of the KDR and/or FLT-1 tyrosine kinase receptors on endothelial cells. Consistent with cross-linking of tyrosine kinase receptors followed by tyrosine phosphorylation, VEGF has been demonstrated to stimulate tyrosine phosphorylation of a 200-kDa membrane protein from endothelial cells (56,57). It remains to be seen if KDR and FLT-1 can form heterodimeric receptors on endothelial cells, as has been postulated by Waltenberger et al. (57). The concept of heterodimerization of receptor tyrosine kinases is well documented, particularly in the case of ␣/␤ heterodimers of PDGF receptors, which differentially bind PDGFa and PDGFb (58). The recently reported finding of VEGF/PLGF heterodimers adds another layer of complexity with respect to various forms of VEGF and PLGF, which have differential specificity for FLT-1 and/or KDR (59).
In the present study, we describe VEGF mutants that are relatively selective for KDR or FLT-1 receptors. These mutants, with differential receptor specificity, will provide molecular tools to probe the relative functional significance of KDR and FLT-1 receptors. The role of KDR and FLT-1 in stimulating endothelial cell proliferation has been addressed using the VEGF mutants. R82A/K84A/H86A and R82N/I83L/K84S mutants of VEGF are significantly decreased in both KDR binding and endothelial cell growth assays (Figs. 5 and 6 and Table I). Furthermore, we have observed that the FLT-1 deficient mutants (D63A, E64A, E67A, and E64N/L66S forms of VEGF) display full activity with respect to endothelial cell proliferation. These results are consistent with the lack of endothelial growth in response to PLGF, which is specific for FLT-1 but does not bind KDR (34). These data, with the present study, correlate the mitogenic and proliferative effects of VEGF with KDR binding and subsequent receptor phosphorylation. In another study, it was shown that KDR-expressing cells respond to VEGF stimulation with marked changes in cell morphology, actin organization, chemotaxis, and mitogenicity, whereas FLT-1 expressing cells lacked these responses to VEGF (57). The significance of the regulation of endothelial function by receptor tyrosine kinases is unclear, especially with respect to the role of FLT-1 in angiogenesis. The pivotal role of both FLT-1 and KDR/Flk-1 in development has been clearly demonstrated by the recent reports on mice deficient in these endothelial-specific receptor kinases. Homozygous mutations that inactivate the FLK gene result in failure of blood island formation and vasculogenesis, whereas homozygous mutations that inactivate the FLT-1 gene result in defective assembly of endothelial cells into tubes (41,60). Both mutations result in intrauterine death at day 8.5. Clearly, both FLT-1 and KDR/ Flk-1 are critical for growth and development of normal vascular endothelium. We anticipate further analysis of the biological significance of KDR and FLT-1-mediated responses to VEGF in vitro and in vivo through the use of receptor-selective VEGF mutants. FIG. 8. VEGF displays different receptor binding sites for KDR and FLT-1. This schematic diagram provides a functional description of the two receptor epitopes identified by mutational analysis. The A site, composed of basic residues (region 82-86), mediates the binding to KDR. The B site, composed of acidic residues (region 63-67), mediates the interaction with FLT-1 receptor. This diagram illustrates bivalent epitopes at opposite ends of VEGF, suggesting a possible mechanism for ligand-induced dimerization of receptors.