Engineered ligand‐based VEGFR antagonists with increased receptor binding affinity more effectively inhibit angiogenesis

Abstract Pathologic angiogenesis is mediated by the coordinated action of the vascular endothelial growth factor (VEGF)/vascular endothelial growth factor receptor 2 (VEGFR2) signaling axis, along with crosstalk contributed by other receptors, notably αvβ3 integrin. We build on earlier work demonstrating that point mutations can be introduced into the homodimeric VEGF ligand to convert it into an antagonist through disruption of binding to one copy of VEGFR2. This inhibitor has limited potency, however, due to loss of avidity effects from bivalent VEGFR2 binding. Here, we used yeast surface display to engineer a variant with VEGFR2 binding affinity approximately 40‐fold higher than the parental antagonist, and 14‐fold higher than the natural bivalent VEGF ligand. Increased VEGFR2 binding affinity correlated with the ability to more effectively inhibit VEGF‐mediated signaling, both in vitro and in vivo, as measured using VEGFR2 phosphorylation and Matrigel implantation assays. High affinity mutations found in this variant were then incorporated into a dual‐specific antagonist that we previously designed to simultaneously bind to and inhibit VEGFR2 and αvβ3 integrin. The resulting dual‐specific protein bound to human and murine endothelial cells with relative affinities of 120 ± 10 pM and 360 ± 50 pM, respectively, which is at least 30‐fold tighter than wild‐type VEGF (3.8 ± 0.5 nM). Finally, we demonstrated that this engineered high‐affinity dual‐specific protein could inhibit angiogenesis in a murine corneal neovascularization model. Taken together, these data indicate that protein engineering strategies can be combined to generate unique antiangiogenic candidates for further clinical development.

potency, however, due to loss of avidity effects from bivalent VEGFR2 binding. Here, we used yeast surface display to engineer a variant with VEGFR2 binding affinity approximately 40-fold higher than the parental antagonist, and 14-fold higher than the natural bivalent VEGF ligand.
Increased VEGFR2 binding affinity correlated with the ability to more effectively inhibit VEGFmediated signaling, both in vitro and in vivo, as measured using VEGFR2 phosphorylation and Matrigel implantation assays. High affinity mutations found in this variant were then incorporated into a dual-specific antagonist that we previously designed to simultaneously bind to and inhibit VEGFR2 and a v b 3 integrin. The resulting dual-specific protein bound to human and murine endothelial cells with relative affinities of 120 6 10 pM and 360 6 50 pM, respectively, which is at least 30-fold tighter than wild-type VEGF (3.8 6 0.5 nM). Finally, we demonstrated that this engineered high-affinity dual-specific protein could inhibit angiogenesis in a murine corneal neovascularization model. Taken together, these data indicate that protein engineering strategies can be combined to generate unique antiangiogenic candidates for further clinical development.

| I N T R O D U C T I O N
Protein ligands and receptors have been used as the basis for a number of successful biotherapeutics. As examples, etanercept, an Fc-fusion of tumor necrosis factor receptor 2, was approved for treatment of rheumatoid arthritis 1 ; aflibercept (VEGF-Trap), an Fc-fusion of VEGFR1 and VEGFR2 extracellular domains, was approved for treatment of pathologic angiogenesis 2,3 ; and recombinant TRAIL (TNF-related apoptosisinducing ligand) is under investigation for oncology applications. 4 Despite these successes, natural ligands or receptors often lack required attributes of a potent therapeutic such as desired target affinity or specificity, or optimal functional activity. In these cases, proteins with altered properties can be generated via directed or combinatorial engineering methods. 5 Examples include engineered ligands with altered receptor binding profiles, 6 receptors engineered to possess ultrahigh affinity to their cognate ligand, 7 engineered ligands with improved cell trafficking, 8 or receptor agonists engineered to function as antagonists. 9 VEGF and its principal receptor, VEGFR2, have generated interest for their central role in pathologic angiogenesis, 10  ziv-aflibercept/aflibercept (Zaltrap/Eylea). While the development of these agents underscores the clinical utility of VEGF/VEGFR2 inhibition, it has also highlighted several challenges, including acquired resistance to therapy and limited efficacy in certain disease states and patient subsets. 11,12 At the same time, a wealth of accumulated evidence has established that pathologic angiogenesis is mediated by the coordinated action of a number of other receptors, including platelet derived growth factor receptor, Tie receptor, and a V b 3 integrin receptor. [13][14][15] These findings have spurred the development of molecules with improved pharmacological properties, in particular, ones that can target a broader set of ligand-receptor interactions responsible for mediating pathologic angiogenesis. 11,16 Previous studies have explored modifying the natural VEGF ligand to alter its function from a receptor agonist to that of a receptor antagonist. VEGF is a homodimeric protein that mediates endothelial cell growth, proliferation, and neovascularization through activation of the receptor tyrosine kinase VEGFR2 (Figure 1a). 17 A VEGF homodimeric ligand binds to two molecules of VEGFR2, leading to receptor dimerization and autophosphorylation, and activation of intracellular signaling pathways, including PI3K, Src, Akt, and ERK. 18 The concept of converting VEGF into an antagonist of VEGFR2 signaling was first explored by introduction of mutations that generated a monomeric form of the receptor, 19 or that disrupted one pole of the VEGF/VEGFR2 binding interface, preventing dimerization and activation. 20,21 In another example, key amino acids involved in VEGFR2 recognition were mutated in VEGF (chain 1: E64R, chain 2: I46R), and the two subunits in the resulting heterodimer were connected via a 14-amino acid linker, thereby creating a single-chain VEGF (scVEGF) construct. 22 Combination of both mutations on one pole of scVEGF abolished binding of one copy of VEGFR2; this scVEGF variant was found to inhibit the mitogenic effects of wild-type VEGF protein on endothelial cells. 22 In all of these examples, the monovalent VEGF ligand that resulted from these protein engineering efforts bound significantly weaker to VEGFR2 compared to the natural bivalent growth factor ligand due to loss of avidity effects, limiting the antagonistic potency of these inhibitors, and hence their clinical utility.
Motivated by the extensive biochemical cross-talk between VEGFR2 and a V b 3 integrin receptors that mediates pathologic angiogenesis 14,16,[23][24][25][26] (Figure 1a), we previously described the conversion of scVEGF into a dual-specific antagonist that targets both VEGFR and a V b 3 integrin. 27 In this work, we introduced four point mutations (chain 1 F17A, E64A; chain 2 I46A, I83A) on one VEGFR2 binding pole to create a variant (termed scVEGFmut) that retained VEGFR2 binding on the opposite pole, but was not capable of inducing receptor dimerization and activation ( Figure 1b). Next, we introduced an engineered integrin-binding loop into the mutated VEGF pole to generate a protein that simultaneously bound both VEGFR and a V b 3 integrin. This dualspecific protein was more a potent antagonist than the parent scVEGFmut, however, it has limited therapeutic potential as an inhibitor as its affinity is only comparable to that of the natural VEGF ligand. Here, we utilize protein engineering to address these limitations, first by generating a more potent ligand-based antagonist by increasing VEGFR2 binding affinity (Figure 1c). Mutations identified in these screens that conferred increased VEGFR2 binding were then added to our previous VEGFR/a V b 3 integrin antagonist (Figure 1d). The resulting high-affinity dual-specific protein binds to human endothelial cells with a relative affinity of 120 6 10 pM, which is at least 30-fold higher affinity than wild-type VEGF (3.8 6 0.5 nM). Additionally, the dual-specific protein bound murine endothelial cells with a relative affinity of 360 6 50 pM, and inhibited pathologic neovascularization in an in vivo model. Residues from both chains of the VEGF homodimer interact with VEGFR2. (b) Single-chain VEGF antagonist (scVEGFmut) has one VEGFR2 binding site mutated, preventing a second receptor molecule from binding, thereby blocking activation. (c) Single-chain VEGF affinitymatured antagonist contains mutations that enable it to bind more tightly to its target receptor and demonstrates more potent inhibition of VEGFR2 activation. (d) Single-chain VEGF affinity-matured dual-specific antagonist (scVEGFdual-specific) contains mutations that confer higher affinity binding to VEGFR2 and an engineered a v b 3 integrin binding loop on the opposite pole of the molecule. We previously showed that dual-specific proteins that bind both VEGFR2 and a v b 3 integrin more potently inhibit angiogenic processes compared to mono-specific proteins 27

| Engineering a ligand-based antagonist with highaffinity binding to VEGFR2
The mutations we previously introduced to generate the monovalent protein scVEGFmut (chain 1: F17A, E64A; chain 2: I46A, I83A) decreased its affinity for VEGFR2 due to loss of avidity effects ( Figure 1b). Thus, we sought to identify variants with increased binding affinity by screening libraries of scVEGFmut proteins displayed on the yeast cell surface against soluble VEGFR2 extracellular domain (VEGFR2-ECD). Libraries containing random mutations in scVEGFmut were created via error-prone PCR using nucleotide analogs, with an average mutation frequency from 0.2 to 2%. 28 Table 1). Significantly, each of these consensus mutations could be rationalized based on the existing crystal structure of the VEGF-VEGFR2 complex and complementary alanine-scanning mutagenesis studies at the same interface. 30,31 The consensus mutations were located either within or adjacent to the intact VEGF-VEGFR2 binding interface 30 (Figure 2e). Every isolated variant contained one or more of the three mutations, chain 1 H86Y, Q87R, and Q89H, representing altered size and electrostatic changes in the primary VEGF-VEGFR2 binding loop. 30,31 In prior mutagenesis studies, mutation of residues 82-84 led to a several 100-fold decrease in binding affinity, while the H86A mutation only led to an approximately twofold decrease; mutation of residues Q87 and Q89 was not reported. 31 Chain 1 E44, which was mutated to Gly in several clones, lies in a separate loop, but also within the VEGF-VEGFR2 binding interface. The E44A mutation led to a fivefold increase in apparent affinity of VEGF for VEGFR2, 31 potentially explaining the occurrence of the similar E44G mutation in several clones. Chain 2 D63N, which was found in eight of the nine unique clones also lies within the VEGF-VEGFR2 binding interface, 30,31 Another common mutation, chain 1 F36L (Figure 2e), is not directly at the VEGF-VEGFR2 binding interface, but is close spatially 30 and presumably stabilizes the receptor-bound conformation.

| Engineered proteins bind with higher affinity to cell surface-expressed VEGFR2
We selected variants mA, mE, and mJ for further study as they each contained many of the individual consensus mutations, and collectively, they constituted a set of proteins that allowed maximal sampling of mutations that were not common across clones (Supporting Information Table 2). These clones, along with scVEGFmut, were recombinantly expressed in Pichia pastoris (see Section 4 and Supporting Information Figure 1). We measured the binding affinity of the scVEGF variants against porcine aortic endothelial (PAE) cells that were stably  Information Table 2); thus, we speculated that these mutations might be extraneous or could potentially confer undesirable bivalent VEGFR2 binding.

| scVEGFmE exhibits high thermal and storage stability
We next examined the stability of scVEGFmE in two distinct, but complementary contexts: thermal stability and stability upon storage. Thermal melts were tracked by circular dichroism spectroscopy. We were unable to obtain complete thermal denaturation of scVEGFmE, even at 958C, demonstrating the high stability of this protein (Supporting Information Figure 2). As a further test of stability, scVEGFmE was heated to 508C, then after cooling its binding affinity to PAE-KDR cells was determined and compared to protein that was not challenged thermally. Both proteins exhibited binding affinities that were indistinguishable (mE: 0.64 6 0.01 nM, mE-heated: 0.72 6 0.01 nM) confirming the thermal stability of this protein (Supporting Information Figure 3). Next, we tested the stability with respect to long-term storage by comparing the binding affinity of the same preparation of scVEGFmE tested 24 months apart and stored in phosphate buffered saline without any stabilizing agents at 2-88C. Again, the binding affinities at the two-time points were indistinguishable demonstrating remarkable stability on storage (data not shown).

| Correlation of increased VEGFR2 affinity and inhibition of VEGF-mediated activity
We measured the effects of scVEGFmE and scVEGFmut on VEGFmediated cell signaling in human umbilical vein endothelial cells 2.6 | scVEGFmE7I: A dual-specific agent that binds VEGFR2 and a v b 3 integrin with high affinity Significant cross-talk and synergy exists between VEGFR2 and a v b 3 integrin in the context of pathologic angiogenesis. 14,23,24 The coordi-nated signaling mediated by these two receptors is necessary for the angiogenic cascade which includes endothelial cell proliferation, controlled remodeling of the extra-cellular matrix, and cell adhesion and migration 16 (Figure 1a). As VEGFR2 and a v b 3 integrin possess a reciprocally stimulatory relationship, 25,26 more potent antiangiogenic effects should be observed on blocking both receptors (Figure 1d). Indeed, in a previous study, co-administration of mono-specific inhibitors of VEGFR2 and a v b 3 integrin showed more complete inhibition of angiogenesis in a mouse model compared to modest inhibition for singleagent treated groups. 33 Leveraging these biochemical insights, we previously engineered a first-in-class dual-specific protein, scVEGF7I, which was comprised of scVEGFmut and an additional introduced epitope that enabled simultaneously binding to VEGFR2 and a v b 3 integrin. 27 In this prior study, scVEGF7I showed superior efficacy in inhibiting angiogenic processes relative to mono-specific inhibitors of VEGFR2 and a v b 3 integrin. 27 However, the relatively modest binding affinity of this engineered dual-specific antagonist for human endothelial cells (single digit nM) offers opportunities for further improvement. To achieve this goal we introduced the seven amino acid mutations identified in scVEGFmE into the scVEGF7I protein. The resulting dual-specific protein, scVEGF-mE7I, was expressed and purified in a similar manner as described above.
We showed that scVEGFmE7I bound to HUVECs with an apparent affinity of 120 6 10 pM, which is 30-fold tighter than wild-type VEGF (scVEGFwt) and at least 330-fold tighter than the parental scVEGFmut antagonist (Figure 5a, Supporting Information Figure 4). In comparison, scVEGFmE bound to HUVECs with an apparent affinity of We next examined the thermal stability of scVEGFmE7I by circular dichroism spectroscopy. As before, we were unable to obtain complete thermal denaturation of scVEGFmE7I, even at 958C (Supporting Information Figure 2), demonstrating the high stability of this protein that was comparable to the scVEGFmE.

| Engineered high-affinity dual-specific antagonist inhibits angiogenesis in vivo
We next tested the ability of scVEGFmE7I to inhibit angiogenesis in an in vivo model of neovascularization. For these studies, we chose a mouse model for corneal neovascularization that has been previously used to evaluate the in vivo efficacy of angiogenesis inhibitors. [37][38][39] The target tissue in this model expresses both VEGFR2 and a v b 3 integrin receptor. 40 For therapeutic delivery, a saline control (n 5 3) or varying concentrations of mE7I (n 5 5 for each concentration) were formulated within a pellet that is implanted into the cornea. A stimulatory growth factor was included in each pellet to promote angiogene-sis. Six days after pellet implantation, the resulting neovascular area was quantified by imaging (Supporting Information Figure 5). At all doses tested, scVEGFmE7I implanted within each pellet was able to significantly inhibit neovascularization in this model relative to the saline control group (p < .001 for all treated groups) (Figure 6a). Immunofluorescent staining confirmed the expression of VEGFR2 and a v b 3 integrin on vasculature in tissue isolated from the positive control (Figure 6b,c). Sham pellets containing saline and no growth factor showed that pellet implantation did not induce angiogenesis (Supporting Information Figure 5D). Further, the therapy did not appear to be acutely toxic at the doses administered as evidenced by measurement of body mass and daily external evaluation of the treated eye (Supporting Information Figure 5). As our previous work extensively compared the in vitro and in vivo efficacy of dual-specific and mono-specific VEGFR2 or a v b 3 integrin targeting agents with different receptor binding affinities, we did not repeat these controls here, and simply sought to confirm that scVEGFmE7I could inhibit angiogenesis in a corneal neovascularization model as a next step for clinical translation.

| C O NC LU S I O N S
This study highlights a strategy for engineering natural ligands as potential therapeutic leads for further clinical development. A caveat of this approach is that converting a bivalent, homodimeric ligand into a monomeric antagonist results in a significant decrease in receptor binding affinity due to loss of avidity effects, ultimately limiting therapeutic efficacy. The engineered ligand-based antagonists scVEGFmE and scVEGFmE7I bind to human endothelial cells with significantly higher affinities compared to wild-type VEGF. This affinity enhancement is important for the ability of such inhibitors to effectively outcompete natural stimulating ligands in a disease setting. While we used scVEGFmut for affinity maturation studies, we could have alternatively started with our first-generation dual-specific variant scVEGF7I to arrive at a Multi-specific protein therapeutics have generated great interest in drug development as they offer opportunities for: (a) improved therapeutic efficacy, (b) lower/less frequent dosing, and (c) lower risk of systemic exposure and off-target effects. [41][42][43] In particular, targeting VEGFR2 and a V b 3 integrin using the high-affinity dual-specific angiogenesis inhibitor described in this work has the potential to reduce disease burden, and benefit patient subgroups that are either unresponsive or develop resistance to current therapies.

| Preparation of scVEGF constructs and libraries
The scVEGF constructs were prepared as described 27 and cloned into the pCT yeast display plasmid. Library DNA containing random mutations was generated from scVEGFmut using error-prone PCR and homologous recombination as described previously. 28,44 Briefly, a range of mutation frequencies (0.2-2%) was obtained using Taq polymerase

| Recombinant protein production and characterization
Protein production was performed using the Multi-Copy Pichia Expression Kit (Invitrogen), using the pPIC9K plasmid (Invitrogen) and the P.
pastoris GS115 yeast strain (Invitrogen). An N-terminal FLAG epitope tag and a C-terminal hexahistidine tag were included as handles for cell binding studies and protein purification, respectively. Endo H f (New England Biolabs) was used to remove N-linked glycosylation, 22  For the scVEGFmE7I construct tested in the corneal neovascularization study, we incorporated two modifications. First, the original 14 amino acid linker connecting the two VEGF chains in the scVEGF scaffold (GSTSGSGKSSEGKG) 22 was replaced by a longer linker comprised of (G 4 S) 4 . This replacement was guided by the insight that the maximum distance that can be bridged by the original linker (40 Å) is very close to the distance between the C-terminus of monomer A and the N-terminus of monomer B (38 Å) of the homodimeric VEGF. 46 Second, we removed the N-terminal FLAG epitope tag to eliminate potential in vivo artifacts originating from the inclusion of this sequence.

| VEGFR2 phosphorylation assays
VEGFR2 phosphorylation assays were carried out following a previously described protocol, 47 with suitable modifications as described below.
Briefly, subconfluent HUVECs were cultured in growth factor-and A vial of 10 lg basic fibroblast growth factor (bFGF; Invitrogen) was combined with 10 ll sterile PBS. Carafate (4 mg) was then added and the solution was briefly vortexed. A 12-15% Hydron solution (polyhydroxyethylmethacrylate, a slow release polymer) was added and mixed. Once a uniform solution was obtained, the solution was spread evenly onto a grid evenly using a cell scraper. Pellets were allowed to dry for 2-3 min, then ejected from the grid into a sterile petri dish and stored at 48C. Pellets containing scVEGFmE7I were prepared identically except the appropriate concentration of engineered protein was used in place of saline.
Sham pellets were prepared identically except that bFGF was omitted.
bFGF was used as the angiogenic inducer as it has been shown to induce VEGF expression in this and other models. 50