high-affinity vegf antagonists by oligomerization of a minimal sequence vegf-binding domain

High-Affinity VEGF Antagonists by Oligomerization of a MinimalSequence VEGF-Binding DomainJames E. Stefano,*,† Julie Bird,† Josephine Kyazike,† Anthony Wai-Ming Cheng,∥ Ekaterina Boudanova,†

Markryan Dwyer,§ Lihui Hou,† Huawei Qiu,† Gloria Matthews,‡ Michael O’Callaghan,⊥

and Clark Q. Pan†

†Protein Engineering, ‡Preclinical Orthopaedics, and §New Biologics, Genzyme Corporation (A Sanofi Company), Framingham,Massachusetts, United States∥Department of Pathology, Duke University Medical Center, Durham, North Carolina, United States⊥4s3Bioscience Inc., Medford, Massaschusetts, United States

*S Supporting Information

ABSTRACT: Vascular endothelial growth factor (VEGF)neutralizing antagonists including antibodies or receptorextracellular domain Fc fusions have been applied clinicallyto control angiogenesis in cancer, wet age-related maculardegeneration, and edema. We report here the generation ofhigh-affinity VEGF-binding domains by chemical linkage of thesecond domain of the VEGF receptor Flt-1 (D2) in severalconfigurations. Recombinant D2 was expressed with a 13 a.a.C-terminal tag, including a C-terminal cysteine to enable itsdimerization by disulfide bond formation or by attachment todivalent PEGs and oligomerization by coupling to multivalentPEGs. Disulfide-linked dimers produced by Cu2+ oxidation ofthe free-thiol form of the protein demonstrated picomolaraffinity for VEGF in solution, comparable to that of a D2-Fc fusion (sFLT01) and ∼50-fold higher than monomeric D2,suggesting the 26 a.a. tag length between the two D2 domains permits simultaneous interaction of both faces of the VEGFhomodimer. Extending the separation between the D2 domains by short PEG spacers from 0.35 kD to 5 kD produced a modest∼2-fold increase in affinity over the disulfide, thus defining the optimal distance between the two D2 domains for maximumaffinity. By surface plasmon resonance (SPR), a larger (∼5-fold) increase in affinity was observed by conjugation of the D2monomer to the termini of 4-arm PEG, and yielding a product with a larger hydrodynamic radius than sFLT01. The higheraffinity displayed by these D2 PEG tetramers than either D2 dimer or sFLT01 was largely a consequence of a slower rate ofdissociation, suggesting the simultaneous binding by these tetramers to neighboring surface-bound VEGF. Finally, disulfide-linked D2 dimers showed a greater resistance to autocatalytic fragmentation than sFLT01 under elevated temperature stress,indicating such minimum-sequence constructs may be better suited for sustained-release formulations. Therefore, theseconstructs represent novel Fc-independent VEGF antagonists with ultrahigh affinity, high stability, and a range of hydrodynamicradii for application to multiple therapeutic targets.

■ INTRODUCTION

Vascular endothelial growth factor (VEGF) serves the role as apotent angiogenic agent as well as having the capacity forincreasing vascular permeability,1,2 thereby increasing theperfusion of serum into tissue. It has been a primary therapeutictarget for suppressing angiogenesis in cancer therapy, wet age-related macular degeneration (AMD), and diabetic retinopathy,conditions whose pathology is directly related to vessel growth.The most common form, VEGF A, is found in multiple splicevariants, although the principal form is 165 amino acids inlength (VEGF165).3−5 VEGF165 has affinity for the heparinsulfate proteoglycan (HSPG) component of the extracellularmatrix due to a heparin binding domain(s) in the C-terminalportion,6,7 which is critical for its mitogenic potential.8 The

angiogenic activity of VEGF is associated with the binding ofVEGF receptor 2 (VEGF R2, KDR, or Flk), a tyrosine kinasewhich induces endothelial cell growth with an EC50 in the sub-nanomolar range.4,5 However, VEGF shows a higher (pM)affinity for a second receptor, VEGF R1 (or Flt-1), which isrequired for normal blood vessel formation during develop-ment, but whose tyrosine kinase activity is nonessential.9 As aresult, Flt-1 is thought to play a role in vivo largely in acting as adecoy to modulate VEGF availability. Two approvedbiotherapeutics (bevacizumab and ranibizumab) comprising

Received: June 7, 2012Revised: November 6, 2012Published: November 26, 2012

Article

pubs.acs.org/bc

© 2012 American Chemical Society 2354 dx.doi.org/10.1021/bc300301m | Bioconjugate Chem. 2012, 23, 2354−2364

pubs.acs.org/bc

an IgG and its affinity-matured Fab fragment against VEGFhave been approved for cancer and wet age-related maculardegeneration (AMD), respectively.Strategies to exploit the natural function of Flt-1 in regulating

VEGF by the construction of soluble Flt-1 receptor decoys havefocused on defining the minimal sequence required for high-affinity VEGF binding. Davis-Smyth demonstrated, through aseries of deletions and swaps with other PDGF receptor familydomains, that deletion of domain 2 (D2) obliterated binding toVEGF, but inclusion of Flt flanking domains 1 and 3 wererequired to replicate the high (pM) affinity of the full-lengthectodomain.10 While such constructs demonstrated efficacy inpreclinical models, high doses were required,11and theseconstructs demonstrated poor pharmacokinetic profiles.12

Moreover, continuous low-level expression of Flt(D1-D3) byadenoviral gene therapy was associated with morbidity andascites formation.13 The crystal structure of a VEGF fragmentwith a monomeric Flt-1 D2 fragment showed two D2 domainsinteracting with each of the outside faces of the VEGFhomodimer, but monomeric D2 was reported to have ∼100-fold lower binding than the complete receptor ectodomain. Alow picomolar affinity was observed only with a monomericD1-D3 fragment or a dimeric (D2-D3) Fc fusion.14 An Fcfusion protein including the second domain of VEGF R1 andthe homologous third domain of VEGF R2 (VEGF Trap,aflibercept) was subsequently found to circumvent the toxicityand PK issues with Flt(D1-D3)12 and was approved for AMDand is currently in clinical trials for macular edema. Morerecently, the reported requirement for both D2 and D3 for fullVEGF binding activity was challenged by the finding that aforced homodimer, containing only the D2 domain of Flt-1fused to human IgG Fc through a 9Gly linker (sFLT01),demonstrated high-affinity binding comparable to two- andthree-domain constructs.15 A spacer sequence in addition to aCH3 domain which provided for formation of a noncovalentdimer were found to be required for high-affinity (pM) binding,suggesting that simultaneous binding to both chains of theVEGF homodimer is required for high affinity. A clinical trial ofthis molecule in gene therapy for AMD is in progress.All VEGF antagonists currently in the clinic utilize an IgG Fc

domain to provide for a long circulating half-life intended toensure reasonable target exposure by systemic administration.However, systemic exposure to all antiVEGF agents isassociated with a number of potentially mechanism-relatedside effects including hypertension, complications in woundhealing, intestinal perforations, and nephrotic syndrome,16,17 asubset of which may also have a component related to the Fceffector function. These side effects may restrict the use of thesemolecules from otherwise potentially treatable indicationslinked to VEGF activity, but which would require long-termadministration. This class includes osteoarthritis, wheresubchondral bone damage may induce the expression ofVEGF, which would elicit serum effusion into the joint, apossible direct source of pain. In addition, angiogenesis andgrowth of demyelinated nerve in the subchondral bone in thepresence of VEGF also suggests VEGF may be a usefultherapeutic target in this context.18 The means to restrictpotential side effects for this and other indications may benefitfrom the targeted delivery of the antiVEGF reagents to criticalsites of action such as the tumor periphery or the joint. Here,we investigate the quaternary structure requirements for high-affinity VEGF binding for Flt-1 domain 2 (D2) throughdimerization of the minimal sequence structure as well as

conjugation to poly(ethylene glycol)s as a strategy forprolonging circulating half-life. Conjugation to multiarmPEGs was performed to assess possible avidity effects and toprovide a potential platform for targeted delivery. Finally, thesuitability of two compositions for sustained-release formula-tion was assessed by the resistance to thermal stress.

■ EXPERIMENTAL PROCEDURESExpression and Purification of D2. Oligonucleotides

corresponding to amino acids 133−226 of VEGF R1 (Flt-1)isoform 1 (SwissProt P17948) with a Kozak sequence(GCCACCATGG) and the Flt-1 signal peptide sequencewere synthesized and Gateway-cloned into a modifiedcommercially available vector containing a CMV immediateearly enhancer-promoter (pCEP4, Invitrogen). HEK293-EBNAmonolayers were transfected using Lipofectamine-2000 andconditioned medium harvested after 3 and 6 days, pooled, andstored at −80 °C. The medium was concentrated ∼10-fold bycentrifugal ultrafiltration (10 kDa MWCO) and applied to a 1mL antiProtein C (HPC4) mAb affinity column (Roche11815024001) equilibrated with 150 mM NaCl, 5 mM CaCl2,50 mM Tris pH 7.2 (EB) at 0.5 mL/min. The column waswashed with 20CV EB and protein eluted with 10CV EB + 5mM EDTA without CaCl2. The eluate was concentrated bycentrifugal ultrafiltration and stored at −80 °C.

Preparation of Conjugates. Homobifunctional and 4-armmaleimide PEGs were obtained commercially (ThermoScientific, Rockfield, IL; Jenchem, Allen TX; NOF Corp.,White Plains, NY). Purified D2 in 25 mM NaCl, 2 mM EDTA,and 25 mM Na phosphate pH7 was reduced for 1 h with 2 mMTCEP at 25 °C, desalted by a single dilution and centrifugalultrafiltration on Vivaspin 4 5 kDa MWCO filters at 4 °C(Corning) prewashed with water. The product was reacted for4 h in molar excess over PEG maleimide groups as described inthe text and purified by SEC over a Superdex 200 column (G.E.Healthcare) using PBS as the mobile phase at 0.4 mL/min.Peaks were collected using A214, concentrated by centrifugalultrafiltration, and flash frozen at −80 °C. Concentrations weredetermined by an ultra micro BCA method using BSA as astandard, with the final A562 measured using a 720 nm referencewavelength on a Nanodrop 1000 spectrophotometer (ThermoScientific). Assays were performed in triplicate. All values werewithin the standard curve and showed a c.v. of <10%.

Production of D2 Dimers through Forced Oxidation.For preparing D2 dimers by forced oxidation, the affinitycolumn eluate was buffer exchanged into PBS, concentrated to1 mg/mL, and treated with 2 mM TCEP (BondBreaker,Thermo Scientific) for 2 h at 25 °C. The product was bufferexchanged into 25 mM Tris HCl pH 7.5 using centrifugaldiafiltration on Vivaspin 4 (5 kDa MWCO), adjusted to 0.5mg/mL, and treated with 0.2 mM CuCl2 for 1 h at 25 °C.EDTA was added to 2 mM and the product desalted with Trisbuffer and then exchanged into PBS by centrifugal diafiltration.The concentration was determined by A280 using E = 0.91.

VEGF Binding by Surface Plasmon Resonance.Recombinant human VEGF165 (R&D Systems, Minneapolis,MN, USA) was covalently immobilized to the surface of a CM5sensor chip using standard amine coupling chemistry on aBiacore 3000 instrument (GE Healthcare, Piscataway, NJ,USA). Human VEGF165 (293-CF, R&D Systems) wasreconstituted at 50 μg/mL in 10 mM sodium acetate pH 4.5,then diluted to 10 or 20 μg/mL and injected over the activatedchip surface at 5 μL/min to achieve the desired surface

Bioconjugate Chemistry Article

dx.doi.org/10.1021/bc300301m | Bioconjugate Chem. 2012, 23, 2354−23642355

densities. Three densities were achieved for VEGF (∼130,∼500, and ∼4000 RU). For a reference surface, the fourthflowcell was activated and quenched without protein. To followkinetics, each sample was serially diluted 3-fold from 10 to 0.37nM in 10 mM HEPES pH 7.4, 150 mM sodium chloride, 3 mMEDTA, 0.005% Surfactant P20 (HBS-EP, GE Healthcare,Piscataway, NJ), and flowed over the reference and VEGF-coupled surfaces for 4 min at 30 μL/min, followed by 5 min ofbuffer alone for dissociation. Samples were analyzed induplicate. Each surface was regenerated with two 30 s injectionsof 10 mM glycine HCl pH 2.0 at 60 μL/min. Biosensor datawere processed using Scrubber2 software (BioLogic SoftwarePty Ltd., Campbell, AU). Kinetic constants were obtained byglobal fit to the sensorgram data using a 1:1 binding modelbased on the molar concentration of conjugate.VEGF Binding by Competition ELISA. Briefly, human

VEGF165 (293-VE, R&D Systems, Minneapolis, MN) wasdiluted to 20 pM in assay diluent (M199 Medium, 5% FBS, 1×penicillin/streptomycin) and mixed with an equal volume (150μL) of serially diluted sample in assay diluent and incubatedovernight with gentle shaking at room temperature. UnboundVEGF in the reaction was then determined by ELISA using animmobilized monoclonal antiVEGF for capture and apolyclonal antiVEGF-HRP conjugate for detection (HumanVEGF Quantikine ELISA DVE00, R&D Systems) using the kitVEGF as standard and analysis using SpectraMax software.Sigmoidal fits to the free VEGF values were then derived usingPrism (GraphPad Software, Inc., La Jolla, CA) to determineEC50 values. The reduction in free VEGF was assumed toreflect binding to conjugate in a 1:1 ratio.

■ RESULTSOligonucleotides encoding 107 amino acids including aminoacids 9−109 of the D2 domain (a.a. 133−226 of Flt-1) with aHPC4 C-terminal appendage for affinity purification weresynthesized and expressed in HEK293 cells. The resultingconstruct as shown in Figure 1a was isolated by affinitychromatography on an antiHPC4 tag antibody column. Theproduct (expected 12304 Da) migrated as a band above the 22kDa marker on the gel (Figure 1b). However, MALDI analysis(Figure 1c) showed a predominant peak at 16.6 kDa with asmaller amount of 32.8 kDa product, suggesting the 25 kDa(apparent MW) band represents glycosylated monomer.Consistent with this, treatment of the product with PNGaseF resulted in a doublet migrating around the expected 12 kDaon SDS PAGE (data not shown). Edman degradation alsoshowed a predominant amino terminal RPFV sequenceconsistent with cleavage of the signal at the expected site.The presence of a C-terminal cysteine was indicated by theability of the purified product to dimerize by oxidation (seebelow). The molecular weight by MALDI (16.6 kDa) wasconsistent with the presence of glycans of approximately 2.1kDa at each of the two potential N-linked glycosylation sites,comprising 26% of the total molecular weight of the expressionproduct likely explaining the anomalous mobility by SDSPAGE. Initial results showed that the product was active inbinding immobilized VEGF165, suggesting the C-terminalHPC4 tag did not strongly effect binding.When expressed at larger scale, SDS PAGE showed the

predominant product as a doublet near 21 kDa, but in addition,a significant amount of a second doublet migrating at ∼40kDaapp (Figure 2a). Upon reduction, the upper bandsdisappeared with an increase in intensity of the lower,

suggesting the upper represent dimers coupled throughdisulfide bonds. The persistence of the doublet upon reductionmay reflect heterogeneity in glycosylation. By SEC (Figure 2b,blue trace), two peaks (1, 2) were observed, corresponding tothe upper and lower doublet. Reduction of the HPC4 eluate(red trace) yielded only peak 2, which contained the ∼21kDaapp doublet material by SDS PAGE (2b, right lane 2).A fusion protein sFLT01, in which a D2 domain is separated

from the IgG1 Fc by a Gly9 linker was found to bind VEGF165with picomolar affinity.15 The distance between the C-terminiof two D2 domains in a crystal structure of D2 in complex withthe VEGF165 homodimer14 suggested that simultaneousbinding to both VEGF subunits may be achieved by thislinker. Thus, it seemed likely that D2 monomers linked througha comparably sized 13aa C-terminal HPC4 tag might also beable to bind the VEGF with high affinity. A preparation ofdimer was purified by SEC of the affinity column eluate (Figure2b) to yield dimer and monomer fractions (Figure 2b right).Alternatively, the dimer/monomer mixture in the eluate wasconverted to dimer through a reduction−oxidation approach.As found with other expressed proteins, the free cysteines were“capped” with disulfide-linked glutathione or cysteine and thusunable to dimerize directly by oxidation (not shown). Theaffinity-purified protein was therefore first reduced with lowlevels of TCEP followed by oxidation with Cu2+ to yield dimerin nearly pure form at high yield (Figure 2c). We reasoned thatthe 26 amino acid span provided by the two HPC4 tags andcysteines, if fully extended, would be sufficient to cover the ∼55Å distance between the C-termini of the D2 domains bound toVEGF as seen in the crystal structure. The effect of PEG linkerson the affinity for VEGF dimer was assessed, since PEGylationfrequently affects target binding. Thus, dimers were preparedby coupling the D2 terminal thiols through bis-maleimide PEGsof varying lengths. A 0.4 kDa PEG only 3 ethylene oxide units

Figure 1. DNA construct and expression of the Flt-1 D2 domain(amino acids 133−226 of Flt-1) with a C-terminal HPC4 tag andcysteine. (a) Diagram of the expression construct and amino acidsequence. The expected amino terminal RPFV was confirmed byEdman sequencing (not shown). (b) Nonreducing SDS PAGE of theHPC4-purified product, which migrates as a broad band just above the22 kDa marker. (c) MALDI analysis of the purified product showing apredominant species of 16.6 kDa with a second minor peak of ∼33kDa, likely reflecting glycosylation and the presence of some dimerizedprotein in conditioned medium.



in length (“BM(PEG)3″, 0.35 kDa), yielded an SEC profile(Figure 3a, blue trace) indistinguishable from that of thedisulfide-linked dimer (purple). From the low abundance ofmonomer after reduction by SDS PAGE (Figure 3b), weconcluded that this preparation largely contained dimerscoupled through PEG rather than disulfide linkages. Thesimilar SEC retention time (RT) for the shortPEG anddisulfide-linked dimers also suggests the PEG does not producea significant change in the solution conformation. In contrast,with larger bis-maleimide PEG linkers (2 kDa, 5 kDa), theincrease in the linker MW resulted in a significant reduction inretention time (Figure 3a). For the 2 kDa PEG spacer, this RTshift was similar to that between the monomer and dimer (cfpeaks 2 and 3 vs 3 and 4), suggesting the PEG conjugate is inan extended orientation in solution. Increasing the length of thespacer to 5 kDa resulted in a further RT shift (peak 1).As a control, monomeric D2 was also obtained either directly

from conditioned medium by SEC of HPC4 affinity columneluates (Figure 3a, peak 4 purple trace) or from the late-elutingSEC peak at the end of conjugation reactions run in D2 excess

(Figure 3a peak 4 red, green, blue)). Similar to many proteinsisolated from conditioned medium where solution-accessiblecysteines are capped with a disulfide-linked glutathione orcysteine, monomer isolated from conditioned medium showedalmost undetectable amounts of dimer (Figure 3d, “D2 (CM)”)and migrated on nonreducing SDS PAGE as a doublet ofslightly higher apparent MW than reduced monomer recoveredfrom a PEGylation reaction such as shown for conjugation to 5kDa PEG (Figure 3d, lane “5 kDa”), consistent with thepresence of a cap preventing oxidative dimerization. Asexpected, monomer isolated from conjugation reactionsshowed a tendency toward dimerization (Figure 3d, “5 kDa”).Each of the SEC fractions was assayed for VEGF binding

activity by surface plasmon resonance (SPR, Figure 4a), whichshowed very slow dissociation rates for the dimers but rapid

Figure 2. Identification and isolation of disulfide-linked dimers. (a)Nonreducing and reducing SDS PAGE of HPC4 affinity-purified D2.D2(CM), HPC4-purified protein from conditioned medium. (b) SECprofile of D2 from conditioned medium before (blue) and after (red)TCEP treatment. (Right) Silver-stained nonreducing SDS PAGE ofpeak fractions 1 and 2. (c) Preparation of disulfide-linked D2 dimer bycopper oxidation of TCEP-reduced HPC4-affinity-purified protein. D2(CM): HPC4-purified protein from conditioned medium. TCEP:reduced with TCEP and desalted. Cu++: reduced, desalted, andoxidized with CuCl2.

Figure 3. Preparation and characterization of PEG-linked D2 dimers.HPC4-purified D2 was reduced with TCEP and then reacted withmolar excess of 2 kD or 5 kDa maleimide PEGs. (a) SEC profiles ofPEGylation reaction products. Red, 5 kDa PEG linker conjugate;green, 2 kDa linker; blue, BM(PEG)3 linker; magenta, D2 fromconditioned medium containing a mixture of disulfide-linked dimerand monomer. Peaks 1−3 are presumed dimers based on SDS PAGEanalysis (see b,c,d). (b) Reducing SDS PAGE of SEC fractions fromD2 conjugation to BM(PEG)3 linker. Numbers above the lanes referto SEC peaks shown in (a). Note that only minor amounts of materialfrom the dimer peak 3 ran as monomer upon reduction, indicatingalmost all was thioether linked through the PEG. (c) Reducing SDSPAGE of PEG conjugate peaks 1 (5 kDa PEG) and 2 (2 kDa PEG).Note that each of these dimer fractions also contained minor amountsmigrating between monomer and the main PEGylated product whosegel mobility was PEG-dependent. These likely represent PEGylatedmolecules in which a single D2 is linked to each PEG. (d)Nonreducing SDS PAGE of peak 4 of the D2 monomer isolatedfrom conditioned medium (C2 CM) or recovered from 5 kDa PEGconjugation reaction after coupling (5 kDa).



dissociation of monomers bound to immobilized VEGF. Theresults of these analyses are summarized in Table 1. All of thePEG-linked dimers showed slightly higher affinities thansFLT01 as a consequence of a 2- to 3-fold increase in therate of binding VEGF on the chip (ka). Of these, theBM(PEG)3-linked dimer showed the largest increase (2.4-fold). The off-rates (kd) among all of the constructs as well asthe fusion protein sFLT01 were very similar ((2.5 ± 0.21) ×10−4 s−1), suggesting that all of the complexes with VEGF onthe chip involved an identical set of interactions. A minor lossin complex stability was observed for dimer recovered from theconjugation reaction (“-S-S-linked D2 dimer CR”, Table 1),possibly reflecting partial reduction of the single internal D2disulfide. The conjugates were also assayed for VEGF bindingin solution by incubation with VEGF followed by ELISA toquantitate the amount of unbound VEGF.12,15 Incubation of 10pM VEGF with picomolar concentrations of either sFLT01 orany of the D2 dimers produced nearly quantitative depletion offree VEGF (Figure 4b). Surprisingly, the EC50 values insolution were about 20-fold lower than the KD values obtainedby SPR and close to those expected for an irreversible reactionwith 1:1 stoichiometry. A similar result has been reported forexpressed sFLT01 in conditioned medium.15 Despite thisextremely low EC50, the sigmoidal shape and slopes of thecurves were consistent with single-component equilibria withKD values in the picomolar range (Table 1).By SPR, monomers either isolated from conditioned medium

or recovered from conjugation reactions showed a 2.5- to 4-foldfaster on-rate than dimers, consistent with lower MW, but alsoshowed a 70- to 80-fold faster rate of dissociation, consistent

with a loss of half of the VEGF-binding determinants (Figure 4and Table 1). Monomer isolated from conditioned mediumshowed the lowest affinity for VEGF (20 nM) but within 2-foldof excess monomer recovered from conjugation reactions (11nM). Consistent with this result, no detectable binding insolution by ELISA was observed with up to 0.15 nM monomerobtained directly from conditioned medium. (Higher levelswere not tested.) In those experiments, however, some bindingwas observed with unconjugated monomer recovered fromcoupling reactions at the end of incubation. However, sincesuch preparations contain free thiols and frequently containdetectable dimer at low levels by nonreducing SDS PAGE(Figure 3d, 5 kDa; Figure 5a, lane 6:1), the possibility thatdimer produced by air oxidation was responsible for thebinding could not be excluded (data not shown).The lack of a strong effect of PEG linker length on the

affinity for VEGF suggested that higher MW multivalentpolymers might show a still higher affinity through an avidityeffect from the greater number of D2 domains. Thus, a pair of4-arm maleimide-terminated PEGs of 10 kDa and 20 kDa wereused as scaffolds to generate multivalent conjugates. As shownin Figure 5a, reaction of reduced D2 with a 4-arm 10 kDa PEGat a 1:1 molar D2/PEG ratio consumed virtually all of themonomer (at 21 kDaapp) and produced a somewhat complexset of products near 40 kDa and higher MW. As the D2/PEGmolar ratio was increased to 6:1 (1.5:1 D2 to PEG maleimidegroups), some remaining unconjugated D2 in both monomerand dimer forms (∼20 kDaapp and ∼40 kDaapp) were seen with

Figure 4. Binding of D2 dimers linked through disulfides or short PEGspacers to VEGF on a solid support (Biacore) and in solution. (a)Biacore sensorgrams. Samples were passed over a 500RU VEGF165chip. Note that the response is directly related to the MW, whichdiffers among the conjugates (sFLT01, 75 kDa; D2 dimers, 33−38kDa). Tan: Reduced Monomer: unconjugated fraction of D2monomer after coupling in excess to 5 kDa bis-maleimide PEG. (b)Solution phase binding to VEGF165. Conjugates were coincubatedwith 10 pM VEGF165 overnight followed by determination of the free(unbound) VEGF concentration by ELISA.

Table 1. Kinetic and Equilibrium Values for Binding VEGFon a Solid Support and in Solution

Biacoresolutionphase

sampleka (×10

5

M−1 s−1)kd (×10

−4

s−1)kD

(nM)aEC50(pM)

D2 Monomer, CMb 10.2 202 19.8 NDc

D2 Monomer, CRd 15.5 171 11.1 e

-S−S-Linked D2 Dimer,CMf

6.0 2.5 0.42 8.4

-S−S-Linked D2 Dimer,CRg

6.9 3.7 0.53 7.1

-S−S-Linked D2 Dimer,Cu2+h

- - - 8.3

BM(PEG)3-linked D2Dimer

9.1 2.5 0.27 4.0

2 kDa PEG-linked D2Dimer

7.6 2.8 0.37 5.8

5 kDa PEG-linked D2Dimer

8.1 2.2 0.27 5.8

10 kDa 4-arm PEG-linkedTetramer

7.8 0.6 0.08 3.7

20 kDa 4-arm PEG-linkedTetramer

6.3 0.7 0.11 4.8

sFlt01 3.8 2.5 0.66 4.6aEquilibrium values determined from binding kinetics on ∼500RUVEGF-loaded chips. The KD values for dimer and multimericconjugates were dependent on the VEGF density on the chip (seeFigure 8 and text). bMonomer isolated from conditioned medium.cNo binding detected within the range of the assay (≤0.15 nM).dUnconjugated monomer recovered from a conjugation reaction.eWeak and variable binding. This species is susceptible to spontaneousoxidative dimerization. fDisulfide-linked dimer isolated from con-ditioned medium. gDisulfide-linked dimer byproduct of a conjugationreaction. hDimer generated by copper oxidation of monomer reducedwith TCEP.



a concomitant formation of a species with an apparent MW of∼100 kDa. The appearance of similar ladder patterns for otherPEG conjugates (not shown) led us to assign the band(s)formed at 1:1 D2/PEG as shown (Figure 5a): The broad bandjust above the 40 kDa disulfide dimer as the 4-arm PEG with asingle D2 coupled, 60 kDa two, 80 kDa three, and 100 kDa fourD2, respectively. These apparent MWs are reasonably close tothat expected for the coupling of increasing numbers of D2monomers of ∼20 kDaapp to a 10 kDa PEG with an apparentMW of 20 kDa. As expected from the 4 arms present on eachPEG, products of higher MW were not observed. SEC of the 10kDa PEG product at 6:1 mol/mol (Figure 5b) showed threemajor peaks with the earliest (1) containing mostly the 100kDaapp and 80 kDaapp products by nonreducing SDS PAGE(Figure 5c, lane 2), peak 3 an ∼40 kDa product consistent withdisulfide-linked dimer (Figure 5c, lane 3), and peak 4, an ∼20kDa species consistent with reduced D2 monomer (Figure 5c,lane 4). The RT for peaks 3 and 4 were identical forconjugations with either PEG, whereas the RT of the leadingpeak varied with the PEG MW, consistent with 1 and 2 beingthe PEG conjugates and 3 and 4 as unconjugated D2 dimer andmonomer, respectively. The background smear of peaks 1 and 2fractions seen by silver staining (lane 2) but only weakly visiblewith Coomassie staining the reaction products (Figure 5a, lane

“6:1”) may reflect some MW and chain length heterogeneity inthe PEG produced by random EtO polymerization on atetravalent core, as well as possible D2n=3 isomers. Note thatSEC fractionation of the crude reaction mixture yieldedmaterial with reduced polydispersity (cf., Figure 5c, lanes 2and Rxn). Similar results were obtained for the 20 kDa PEGconjugate (not shown). As shown in Figure 6, the SEC

retention times of these conjugates were significantly shorterthan the those of the PEG-linked dimers (Figure 3), as well asof two fusion proteins in which the D2 is linked either to thehuIgG1 Fc (sFLT01) or the CH3 domain (sFLT02),15

suggesting they will have substantially better PK than thedimer constructs.Each of the SEC fractions was recovered and analyzed by

SPR using immobilized VEGF165 (Figure S1, Table 1). Withthe 10 kDa PEG conjugate, binding to VEGF showed similarkinetics and final saturation levels as sFLT01 (Figure S1) andall of the conjugates demonstrated high-affinity binding (Table1). However, although the 4-arm PEG tetramers with twice asmany D2 showed a higher affinity of the magnitude expected byan increase in the number of D2 on the kinetics of binding, theincrease was instead largely attributable to the ∼4-fold decreasein off-rate, suggesting a greater number of contacts with VEGFon the chip surface. The dimer with a 5 kDa PEG spacershowed a comparable rate of binding but a 3.6-fold faster off-rate than the 10 kDa 4-arm PEG tetramer in which 5 kDa ofPEG separates each D2 pair. In contrast, in solution, the 4-arm10 kDa PEG tetramer showed an only slightly higher affinitythan sFLT01 (EC50 = 3.7 vs 4.7 pM) and the 20 kDa PEGconjugate was no different than sFLT01 (Table 1). As with thedimers, there was a large discrepancy between the KD values bySPR and the EC50 values by solution-phase binding (80- to 200-fold).Particularly with the multiarm PEG conjugates, the apparent

affinity (1/KD) by SPR was dependent upon the VEGF loadingon the chip and increased with higher VEGF densities (Figure7). With the 20 kDa PEG conjugate, a 5-fold increase ofimmobilized VEGF resulted in a 10-fold increase in theapparent affinity. The increase was less for the 10 kDa 4-arm

Figure 5. Conjugation of D2 to 4-arm PEGs. (a) SDS PAGE ofreaction mixtures initiated with 1:1 and 6:1 (mol/mol) D2 monomerto 4-arm 10 kDa PEG. Assignments for the reaction products areshown. PEG bearing a single D2 (PEG-D2 dash in figure) migratesslightly more slowly than the disulfide-linked D2 dimer (S−S-dimer).This species in the reaction products at high D2 excess likely reflectsreoxidation of unconjugated monomer during conjugation. (b) SECprofiles of conjugation products with 10 kDa and 20 kDa PEGs. (c)Silver-stained nonreducing SDS PAGEof peak SEC fractions from a 10kDa PEG conjugation. Peak 2 comprises a mixture of twopredominant PEGylated forms with apparent MW of ∼80 kDa and∼100 kDa, likely corresponding to PEG coupled to three and four D2,respectively (see panel a). Peak 3 contained a predominant product of∼38 kDa (disulfide-linked dimer), produced by oxidation, and peak 4,remaining unreacted D2 monomer added in excess to the reaction.Slight cross-contamination of dimer and monomer isolates is apparent.

Figure 6. Comparison of tetrameric PEG conjugates to dimeric Ig Fcfusions by SEC. The profiles for all conjugates are of unpurifiedreaction end products. Annotations indicate the MW of the PEGlinker. The flow rate was 0.5 mL/min with absorbance measured at214 nm. Red and blue: tetrameric PEG conjugates. Green and gray:sFLT01, a D2-Gly9-Fc fusion; and sFLT02, a D2-Gly9-CH3 fusion.15

The total MWs without glycosylation are as follows: 20 kDa PEGconjugate, 69 kDa; 10 kDa PEG conjugate, 59 kDa; sFLT01, 75 kDa;sFLT02: 47 kDa; D22 (dimer), 26.6 kDa; D2, 12.3 kDa. A portion ofthe 20 kDa PEG conjugate profile has been removed for clarity.



PEG conjugate and smallest with the disulfide-linked dimer.This effect resulted a change in rank order of affinity for the 4-arm conjugates. In contrast, D2 monomer showed nosignificant change in affinity with VEGF loading. For all ofthe PEG conjugates, the affinity increase with increasing VEGFdensity was entirely due to the reduction in off-rate. ForsFLT01, the rate of association (ka) increased only 16% andwas not significantly different (<5%) than the dimer and PEGconjugates even with a 5-fold increase in VEGF. At very highloading, extremely slow off-rates and mass transport issuesduring binding, as suggested by a lack of curvature in thebinding reaction profile, precluded an accurate determination ofKD values (data not shown). These results are consistent withavidity effects arising from simultaneous interaction of theconjugates with adjacent VEGF molecules on the chips, whichwould increase with decreasing distance between VEGF.The high affinity for VEGF of the small dimer conjugates

suggested the possibility of a broader application of theseconstructs to other applications such as localized delivery byincorporation into sustained-release depots. However, theformulation processes pose a significant challenge to proteinstability, since most were designed for small moleculetherapeutics and typically employ conditions and reagentswhich have a significant potential to induce proteindenaturation. Consistent with this, internal efforts to formulatesFLT01 for sustained localized release were complicated byfragmentation and loss of VEGF binding activity. The minimalsequence D2 dimer (32 kDa) is within the MW range ofproteins reported to be successfully incorporated intosustained-release depots. The higher yield of D2 dimergenerated through a more direct process of reduction/oxidationenabled an assessment of the suitability by testing its thermalstability under comparable conditions. Thus, sFLT01 and theD2 dimer generated by copper oxidation were incubated in PBSpH 7.2 at 45 °C for extended periods in the presence ofprotease inhibitors and EDTA. As shown by reducing SDSPAGE (Figure 8a), sFLT01 showed a distinct cleavage product(arrowhead) and a diffuse band around 21 kDa (bracket),which accumulated over 36 days. In addition, higher MWcovalent aggregates were also visible. By comparison, the D2dimer showed no obvious fragmentation and only slightaccumulation of a higher MW aggregate (Figure 8b). Since itwas possible that an inhibitor-resistant protease in the sFLT01

was responsible for degradation, dimer was also coincubatedwith the sFLT01 at elevated temperature. After 28 days,cleavage products corresponding to sFLT01 were still observedwithout detectable fragmentation of D2 (data not shown). Theaffinity of the molecules for VEGF was also assessed bysolution-phase binding (Figure 8c). Both molecules showedloss of activity, but the initial rate of loss was significantly fasterfor sFLT01. After 14 days at 45 °C, sFLT01 had lost ∼30%activity, while the D2 dimer showed less than 10% reduction.

■ DISCUSSIONThese results demonstrate that D2 dimerization through avariety of linkages is sufficient and necessary for very highaffinity binding to VEGF165 in the picomolar range, and thatmultimerization through conjugation to polymers yieldsmolecules with comparable hydrodynamic radii as Fc fusionswhile retaining extremely high affinity. The different conjugateconfigurations explored in this study are shown diagrammati-cally in Scheme 1. These results demonstrate that the couplingof this minimal-sequence entity to such scaffolds provides ameans to select the optimal hydrodynamic radius andrepresents a viable strategy for balancing the opposingrequirements for prolonged PK and efficient tissue penetrationfor different therapeutic applications. The low MW of thedisulfide-linked D2 dimer may be optimal for generating

Figure 7. Effect of chip loading on apparent affinity by SPR. Bindingaffinities were assessed at three immobilized VEGF levels. Theapparent affinity constants (Ka) of the constructs at two densities (100,500 RU) are shown. The 20 kDa construct showed the greatest changein apparent Ka with chip loading. D2 monomer showed no significantchange in Ka over the same range.

Figure 8. Forced degradation study of sFLT01 and D2 disulfide-linkeddimer. Samples (0.25 mg/mL) were incubated in PBS + EDTA andprotease inhibitors at 45 °C for up to 36 days; then, aliquots werewithdrawn, stored at −80 °C, and analyzed by SDS PAGE. (a)Stability of sFLT01 to elevated temperature. Arrow: primarydegradation product. (b) Stability of D2 dimer produced by Cu2+

oxidation. (c) Effect on VEGF-binding activity in solution. Bindingassays were performed on samples from a different stability study butwhich showed a similar kinetics of degradation as in (a) and (b): (○)D2 Cu2+ Dimer; (□) sFLT01.



sustained-release formulations for localized delivery, thusextending the application of these antiangiogenics to a broaderrange of indications such as osteoarthritis. In preliminarystudies, we have found that these dimers demonstratesignificantly greater stability than the sFLT01 Fc fusion toextended incubation, a stressor which mimics the effects of thedepot formulation process (Li, unpublished results). Inaddition, conjugation to multivalent scaffolds may, throughavidity effects, show superiority over dimeric constructs forneutralizing VEGF bound to extracellular matrix as illustratedby the higher-affinity binding to VEGF coupled to the dextranpolymer on the chip surface. These multivalent scaffolds alsoprovide a potential platform for the targeted delivery of VEGFantagonists through the co-conjugation of targeting moieties. Abispecific mAb with antiHer2 combined with an antiVEGFdomain is currently in preclinical development.20 However, themultiarm conjugation approach presented here could besuperior by providing a second D2 domain to produce thehigh affinity of the bivalent interaction, and to more effectivelysequester VEGF to a level below the EC50 for cell proliferation.Requirement for Dimerization and the Effects of

Spacers and the HPC4 Tag. Our results confirm previousindications15 that the D2 sequence is sufficient to interact withVEGF165 with high affinity, but only in a dimer form thatprovides the ability to simultaneously interact with bothbinding faces in the VEGF homodimer. This conclusion wasarrived at by the finding that both a linker element and a CH3domain to provide for the formation of noncovalent dimersduring expression of Fc fusion constructs were required forhigh-affinity binding to VEGF. By in vitro reconstruction, thepresent work directly demonstrates that physical linkage of twodomains is required for a low picomolar EC50 interaction.Previously, the requirement for a linker element was inferredfrom molecules which lacked linkages showing either poor

VEGF neutralizing activity or a substantial (∼100-fold)reduction in binding.15 We have extended these observationsusing synthetic spacers to examine the effect of increaseddistance between the D2. Only a modest effect of increasing thelength of the spacer provided by the HPC4 tag to a 5 kDa PEGon VEGF binding was observed, suggesting that two disulfide-linked 13aa HPC4 tags in the dimer is sufficient to span thedistance between the D2 C-termini when bound to VEGF andprovide for a high-affinity interaction. The distance in thecrystal structure between the C-termini of the two D2 incomplex with a VEGF8−109 homodimer (1flt) is ∼55 Å 14

allowing for an additional unresolved terminal Thr in one chain,which could be easily spanned by the tag in fully extended form(80−90 Å at 3.4 Å/residue). The indistinguishable stability ofVEGF complexes with sFLT01 and the HPC4 dimers asreflected in the rate of dissociation (Table 1) suggests that asignificant interaction between either linker and VEGF isunlikely. However, the addition of a PEG spacer (BM(PEG)3,0.35 kDa) to the HPC4-dimer produced an increase in affinityby solution-phase assay or SPR, where the effect was seen duesolely to a consequence of a faster rate of association. This isconsistent with fewer degrees of freedom for binding VEGF forthe molecule containing the disulfide bridge. Although only 55Å separates the C-termini of D2 in the crystal structure, theyare not in direct sight of each other. Thus, an increase in spacerlength by PEG insertion may increase the probability offorming a bidentate complex when the linker is forced awayfrom the VEGF surface. Model building to obtain a lowest-Tm-fold (Pep-Fold)21 of the HPC4 tag suggests the peptide mayform a relatively condensed structure and that its length maynot be sufficient to permit the two D2 to simultaneouslyinteract at their binding sites on the distal opposing faces ofVEGF (not shown). However, the distance could be spannedby insertion of a small PEG (18 Å fully extended), even if thetag were partially folded. It is also worth noting that with theVEGF fragment the crystal structure, which lacked the C-terminal 56 amino acids of VEGF165, the C-terminus extendsinto the groove at the bottom face of the dimer and impingeson the path between the two D2 C-termini,14 raising thepossibility that this portion of VEGF165 could interfere withbinding. A closer examination of the linker length dependencemay be useful toward further understanding of the dispositionof the two D2 in the complex.

Effect of PEG Size on Affinity. Surprisingly, an increase inthe size of the PEG linker from 0.35 kDa to 5 kDa had anegligible further effect on binding as detected by SPR and onlya slight decrease in the EC50 for solution binding (Table 1) inspite of a significant increase in the conjugate hydrodynamicradius (Figure 3). An increase in the spacer size beyond thatnecessary for high-affinity complex formation might also beexpected to slow the rate of binding by lowering the probabilityof both D2 simultaneously interacting with the VEGF dimer,similar to the often-seen effect of PEGylation on binding othermolecules. However, the stability of the initial complex of oneD2 to a VEGF subunit, as seen with the D2 monomer−VEGFinteraction (kd = 0.02 s−1; equivalent to t1/2 = 35s; Table 1,Figure 4) is likely sufficient, as a consequence of the high localconcentration of the second D2, to virtually guarantee thesubsequent formation of the bidentate complex. Thus, evenallowing for a substantial effect of PEG viscosity on thediffusion rate, the t1/2 for binding of the second D2 could easilybe so short as to produce a negligible effect of PEG size oneither the observed rate of binding or EC50. Alternatively, it is

Scheme 1. Diagrammatic Representation of the Molecules inthe Studya

aCM: conditioned medium containing a mixture of capped monomerand disulfide-linked dimer forms. TCEP treatment reduces both thecap and linking disulfide to yield a free thiol form (center). Copperoxidizes the reduced product to produce a highly enriched dimerfraction without the use of SEC. Reaction of the free thiol form of theD2 with either 2-arm or 4-arm maleimide PEG yields PEG-linkeddimers and tetramers, respectively. The structures of D2 and VEGFare taken from Weismann14 and displayed at low resolution. The whitepatches indicate the interface residues identified in the crystalstructure. The pink and blue portions of the VEGF indicate the twosubunits. The structure of the HPC4 tag (blue rod) is unknown. Theorientation of D2 bound to VEGF places the HPC4 tag linkerstructure on the bottom face of VEGF as shown. The size of the PEGis for representational purposes only.



possible that the D2 domains have a weak affinity for eachother to form inactive complexes, for example, through anonspecific interaction of their nearly flat hydrophobic VEGF-binding faces. PEG may solvate exposed hydrophobic residues,blocking the nonproductive interaction and thus effectivelyincreasing the availability for binding VEGF. Longer PEGswould be more effective at this. However, since a longer PEGwould also lower the availability for binding VEGF throughsteric hindrance, the combination of these factors may producea negligible net effect with increasing PEG size. Of course, sucha nonspecific interaction would be expected to strongly inhibitVEGF binding with shorter linkers such as BM(PEG)3 or thesFLT01 sequence, but below a certain size, the linkers may betoo short to allow the D2 to freely interact in the most stableorientation. In the full receptors, Flt-1 and KDR, suchinterdomain interactions would be prevented by evolution ofa higher-order structure for the receptor.SPR vs Solution-Based Assays. The lack of an effect of

PEG length in the dimers encouraged us to examine still largerPEG spacer, including HMW multiarm PEGs, which might alsoprovide additional avidity effects. The increase in affinity forVEGF by the 4-arm PEG conjugates as measured by SPR (6- to8-fold) was of the magnitude expected from the 6-fold increasein the number of pairs of D2 in the conjugate which areavailable for interaction with VEGF. However, all of the affinityincrease was found to be due instead to a decrease in off-rate,unexpected for the binding of a single VEGF dimer by a pair ofD2 domains since each VEGF homodimer can at mostaccommodate only two D2. It is likely that the higher affinityof these conjugates instead reflects the simultaneous binding ofneighboring VEGF dimers on the chip. This is consistent withthe observed drop in the apparent KD with increasing VEGFdensity on the chip, which was more pronounced forconjugates with wider spacing between the D2 domains andcompletely absent for the D2 monomer. The largest effect wasseen with the 20 kDa tetramer, where the stimulation was >10-fold with a 5-fold increase in VEGF density (Figure 7). As seenwith the dimers, the affinities determined by SPR were typically∼100-fold lower than those measured by the solution-phasebinding ELISA-based assay (Table 1). These data suggest thatall of the contacts on the VEGF homodimer that are availablein solution are not available when immobilized on a chip,perhaps due to interference of binding by the linkage to one ofthe two subunits. In this view, the much higher affinity insolution reflects the avidity associated with binding of bothVEGF subunits by separate D2 domains in the conjugate, whichis partially blocked in the case of immobilized VEGF.Consistent with this, in experiments where sFLT01 wasbound to the chip surface and VEGF was in the mobilephase, a much higher affinity in the picomolar range wasobserved (data not shown). A similar effect on avidity as aconsequence of ligand/receptor orientation has been reportedfor another dimeric molecule, TGFβ, where re-engineering thelinkage to the chip replicated the high affinity associated withthe naturally occurring binding to two separate receptordomains.22 On the basis of the density of carboxylic acidgroups on the dextran surface coating available for coupling D2,it is likely that a single bond links the VEGF dimer to thesurface, leaving one unmodified subunit. Thus, binding of themonomeric D2 to VEGF by SPR may provide an accuratepicture of the interaction of D2 with a single subunit of theVEGF dimer, and as expected, there was a complete absence ofbinding of monomer up to the highest concentration in the

solution-phase assay (0.15 nM) consistent with the KDobserved by SPR (20nM). Thus, excluding the Kd values forthe dimers and tetramers obtained by SPR as largely reflectingavidity for multiple surface-bound VEGF, the difference inaffinities for the monomer vs the multimers is >4000-fold,strongly supporting the use of multimeric conjugates for thesequestration of low concentrations of VEGF.Because of the restricted access to all of the VEGF contacts

on the chip, SPR may be limited in its ability to detect anyacceleration in binding compared to single VEGF homodimersas a consequence of the increase in D2 binding pairs such as inthe 4-arm conjugates. However, the solution-phase assay alsoshowed only a minor increase in affinity for these conjugates. Itis likely that this assay significantly underestimates the actualaffinity increase as a consequence of the extremely highaffinities of the conjugates for VEGF. The limited sensitivity ofthe ELISA assay (∼1 pM VEGF) required the use of VEGFconcentrations close to and exceeding the EC50 of the VEGFbinders. Under such conditions, the excess VEGF could depletethe available free binder to shift the equilibrium and observedEC50 upward from the actual KD.

19 Consistent with this,preliminary results with a higher sensitivity assay permitting theuse of lower VEGF concentrations yielded EC50 values forsFLT01 in the 1−2 pM range (data not shown).). Thus, theeffect of multimerization on affinity may be underestimated byboth assays. Nonetheless, for the purpose of reducing freeVEGF levels, given the picomolar affinities of these constructs,binding of VEGF particularly by the PEG multimers may beconsidered to be nearly irreversible.

Thermal Stability of the Disulfide-linked Dimer. Thesimilarly high affinity displayed by the low MW disulfide-linkeddimer suggested a broader application of this syntheticapproach to VEGF antagonism in applications such as localizeddelivery, where a reduction in systemic exposure would bedesirable to minimize side effects. This issue would beparticularly keen in the case of long-term administration ofVEGF antagonists for indications such as OA pain. However, amajor challenge for the development of sustained-release depotformulations has been protein stability, and success in the clinichas been reported only in a single instance of a low MWprotein (<25 kDa),23 but not with larger proteins. Indeed, ourefforts to formulate the 80 kDa sFLT01 for sustained releasehave been complicated by fragmentation and loss of VEGFbinding activity. A similar pattern of fragmentation wasobserved by extended incubation at elevated temperatures,that are similar to those used in forced degradation studies forbiotherapeutics. Several lines of evidence indicate that this isnot due to low-level protease contamination or deamidation,suggesting it may instead arise from stress on the D2 structure(Li et al., Genzyme Corporation, unpublished). It can beinferred from its greater thermal stability that the smalldisulfide-linked dimer may be better suited to these stresses offormulation as well as providing the long-term stability requiredfor this application. This greater stability compared to sFLT01is somewhat surprising considering the larger degrees offreedom expected for an unstructured glycine sequence as insFLT01 compared to the HPC4 tag. The lesser effect of theVEGF surface density on binding the dimer by SPR (Figure 7)suggests that the D2 domains are effectively closer togetherthan in sFLT01, even though this molecule shows a slightlylower apparent affinity for VEGF in solution. Thus, it ispossible that the HPC4 tag partially interferes with thesimultaneous binding of both D2 to the VEGF dimer subunits.



The tag may similarly constrain the hydrophobic D2 interfacesfrom interacting with each other in the absence of VEGF, anevent which could distort the structure and make it prone tofragmentation. The site of fragmentation in D2 is close to thehydrophobic binding interface (not shown). Interestingly,PEGylation of sFLT01 was found to also increase its thermalstability (J.S., unpublished) suggesting that the PEG spacerelements such as employed in the multimeric conjugatesdescribed here may also provide additional benefit for theapplication of those molecules to such formulations.

■ CONCLUSIONS

These data demonstrate VEGF binders with equally high (pM)affinity as Fc receptor domain fusions can be obtained throughdisulfide linked dimerization and oligomerization of a minimalsequence Flt-1 D2 domain on PEG scaffolds. Higher affinitiesfor binding VEGF bound to a biopolymer matrix were obtainedthrough avidity effects achieved through multivalent conjuga-tion of D2 to the termini of multiarm PEGs. These conjugates,particularly the 4-arm 20 kD PEG D2 tetramer with a total ∼85kDa, may possess sufficient hydrodynamic radius to displaycomparable PK/PD profile as existing VEGF antagonists thatare Fc-based without the potential undesirable side effectsassociated with Fc-mediated effector function. The potential toco-conjugate targeting moieties may provide a means togenerate targeted, high-affinity VEGF sequestering moleculesto minimize possible side effects associated with systemicVEGF antagonism. In addition, the very small D2 homodimers(∼35 kDa) linked through minimal-structure elements wereshown to achieve similar high affinities but with greater thermalstability than the D2-Fc fusion (∼80 kD). These novelconjugates may be more suitable for certain applications suchas tissue penetration and sustained-release formulations.

■ ASSOCIATED CONTENT

*S Supporting InformationAdditional figure. This material is available free of charge via theInternet at http://pubs.acs.org.

■ AUTHOR INFORMATION

Corresponding Author*E-mail: [email protected]. Phone: 508-270-2271. Fax:508-872-9080.

NotesThe authors declare the following competing financialinterest(s): All authors except A.W.M.C and M.O. areemployees of Genzyme Corporation.

■ ACKNOWLEDGMENTS

This study was in part supported by a scholarship to A.W.M.C.from the Osteoarthritis Research Society International. Wewould like to thank Dan Li and Jon Kingsbury for usefuldiscussions and Bob Mattaliano for a critical reading of themanuscript.

■ ABBREVIATIONS

CV, column volume; PEG, poly(ethylene glycol); TCEP, tris-carboxyethylphosphine; PAGE, polyacrylamide gel electro-phoresis; SPR, surface plasmon resonance

■ REFERENCES(1) Connolly, D. T., Heuvelman, D. M., Nelson, R., Olander, J. V.,Eppley, B. L., Delfino, J. J., Siegel, N. R., Leimgruber, R. M., and Feder,J. (1989) Tumor vascular permeability factor stimulates endothelialcell growth and angiogenesis. J. Clin. Invest. 84, 1470−8.(2) Connolly, D. T., Olander, J. V., Heuvelman, D., Nelson, R.,Monsell, R., Siegel, N., Haymore, B. L., Leimgruber, R., and Feder, J.(1989) Human vascular permeability factor. Isolation from U937 cells.J. Biol. Chem. 264, 20017−24.(3) Harper, S. J., and Bates, D. O. (2008) VEGF-A splicing: the keyto anti-angiogenic therapeutics? Nat. Rev. Cancer 8, 880−887.(4) Robinson, C. J., and Stringer, S. E. (2001) The splice variants ofvascular endothelial growth factor (VEGF) and their receptors. J. CellSci. 114, 853−65.(5) Byrne, A. M., Bouchier-Hayes, D. J., and Harmey, J. H. (2005)Angiogenic and cell survival functions of vascular endothelial growthfactor (VEGF). J. Cell Mol. Med. 9, 777−94.(6) Gitay-Goren, H., Soker, S., Vlodavsky, I., and Neufeld, G. (1992)The binding of vascular endothelial growth factor to its receptors isdependent on cell surface-associated heparin-like molecules. J. Biol.Chem. 267, 6093−8.(7) Fairbrother, W. J., Champe, M. A., Christinger, H. W., Keyt, B. A.,and Starovasnik, M. A. (1998) Solution structure of the heparin-binding domain of vascular endothelial growth factor. Structure 6,637−48.(8) Keyt, B. A., Berleau, L. T., Nguyen, H. V., Chen, H., Heinsohn,H., Vandlen, R., and Ferrara, N. (1996) The carboxyl-terminal domain(111−165) of vascular endothelial growth factor is critical for itsmitogenic potency. J. Biol. Chem. 271, 7788−95.(9) Hiratsuka, S., Minowa, O., Kuno, J., Noda, T., and Shibuya, M.(1998) Flt-1 lacking the tyrosine kinase domain is sufficient for normaldevelopment and angiogenesis in mice. Proc. Natl. Acad. Sci. U. S. A.95, 9349−54.(10) Davis-Smyth, T., Chen, H., Park, J., Presta, L. G., and Ferrara, N.(1996) The second immunoglobulin-like domain of the VEGFtyrosine kinase receptor Flt-1 determines ligand binding and mayinitiate a signal transduction cascade. EMBO J. 15, 4919−27.(11) Gerber, H. P., Kowalski, J., Sherman, D., Eberhard, D. A., andFerrara, N. (2000) Complete inhibition of rhabdomyosarcomaxenograft growth and neovascularization requires blockade of bothtumor and host vascular endothelial growth factor. Cancer Res. 60,6253−8.(12) Holash, J., Davis, S., Papadopoulos, N., Croll, S. D., Ho, L.,Russell, M., Boland, P., Leidich, R., Hylton, D., Burova, E., Ioffe, E.,Huang, T., Radziejewski, C., Bailey, K., Fandl, J. P., Daly, T., Wiegand,S. J., Yancopoulos, G. D., and Rudge, J. S. (2002) VEGF-Trap: aVEGF blocker with potent antitumor effects. Proc. Natl. Acad. Sci. U. S.A. 99, 11393−8.(13) Ferrara, N., Chen, H., Davis-Smyth, T., Gerber, H. P., Nguyen,T. N., Peers, D., Chisholm, V., Hillan, K. J., and Schwall, R. H. (1998)Vascular endothelial growth factor is essential for corpus luteumangiogenesis. Nat. Med. 4, 336−40.(14) Wiesmann, C., Fuh, G., Christinger, H. W., Eigenbrot, C., Wells,J. A., and de Vos, A. M. (1997) Crystal structure at 1.7 A resolution ofVEGF in complex with domain 2 of the Flt-1 receptor. Cell 91, 695−704.(15) Pechan, P., Rubin, H., Lukason, M., Ardinger, J., DuFresne, E.,Hauswirth, W. W., Wadsworth, S. C., and Scaria, A. (2009) Novel anti-VEGF chimeric molecules delivered by AAV vectors for inhibition ofretinal neovascularization. Gene Ther. 16, 10−6.(16) Cohen, M. H., Gootenberg, J., Keegan, P., and Pazdur, R.(2007) FDA drug approval summary: bevacizumab (Avastin) plusCarboplatin and Paclitaxel as first-line treatment of advanced/metastatic recurrent nonsquamous non-small cell lung cancer.Oncologist 12, 713−8.(17) Freeman, H. J. (2012) Colitis associated with biological agents.World J. Gastroenterol. 18, 1871−4.(18) Walsh, D. A., McWilliams, D. F., Turley, M. J., Dixon, M. R.,Franses, R. E., Mapp, P. I., and Wilson, D. (2010) Angiogenesis and



http://pubs.acs.org

mailto:[email protected]

nerve growth factor at the osteochondral junction in rheumatoidarthritis and osteoarthritis. Rheumatology (Oxford) 49, 1852−61.(19) Hulme, E. C., and Trevethick, M. A. (2010) Ligand bindingassays at equilibrium: validation and interpretation. Br. J. Pharmacol.161, 1219−37.(20) Bostrom, J., Yu, S. F., Kan, D., Appleton, B. A., Lee, C. V.,Billeci, K., Man, W., Peale, F., Ross, S., Wiesmann, C., and Fuh, G.(2009) Variants of the antibody herceptin that interact with HER2 andVEGF at the antigen binding site. Science 323, 1610−4.(21) Maupetit, J., Derreumaux, P., and Tuffery, P. (2010) A fastmethod for large-scale de novo peptide and miniprotein structureprediction. J. Comput. Chem. 31, 726−38.(22) De Crescenzo, G., Pham, P. L., Durocher, Y., and O’Connor-McCourt, M. D. (2003) Transforming growth factor-beta (TGF-beta)binding to the extracellular domain of the type II TGF-beta receptor:receptor capture on a biosensor surface using a new coiled-coil capturesystem demonstrates that avidity contributes significantly to highaffinity binding. J. Mol. Biol. 328, 1173−83.(23) Cleland, J. L., and Jones, A. J. (1996) Stable formulations ofrecombinant human growth hormone and interferon-gamma formicroencapsulation in biodegradable microspheres. Pharm. Res. 13,1464−75.



high-affinity vegf antagonists by oligomerization of a minimal sequence vegf-binding domain

Documents