Novel Peptides Selected to Bind Vascular Endothelial Growth Factor Target the
Receptor-Binding Site
Wayne J. Fairbrother,
‡
Hans W. Christinger,
‡
Andrea G. Cochran,
‡
Germaine Fuh,
‡
Christopher J. Keenan,
‡
Clifford Quan,
§
Stephanie K. Shriver,
‡
Jeffrey Y. K. Tom,
§
James A. Wells,
‡,|
and Brian C. Cunningham*
,‡,|,
Departments of Protein Engineering and Bioorganic Chemistry, Genentech, Inc., 1 DNA Way,
South San Francisco, California 94080
ReceiVed August 11, 1998; ReVised Manuscript ReceiVed October 20, 1998
ABSTRACT: Peptides that inhibit binding of vascular endothelial growth factor (VEGF) to its receptors,
KDR and Flt-1, have been produced using phage display. Libraries of short disulfide-constrained peptides
yielded three distinct classes of peptides that bind to the receptor-binding domain of VEGF with micromolar
affinities. The highest affinity peptide was also shown to antagonize VEGF-induced proliferation of primary
human umbilical vascular endothelial cells. The peptides bind to a region of VEGF known to contain the
contact surface for Flt-1 and the functional determinants for KDR binding. This suggests that the receptor-
binding region of VEGF is a binding “hot spot” that is readily targeted by selected peptides and supports
earlier assertions that phage-derived peptides frequently target protein-protein interaction sites. Such
peptides may lead to the development of pharmacologically useful VEGF antagonists.
Vascular endothelial growth factor (VEGF)
1
is a primary
modulator of vascular neogenesis, angiogenesis, and vessel
permeability (1, 2). VEGF contains two identical polypeptide
chains that are linked covalently by a pair of disulfide bonds
(3, 4). The hormone exists as five isoforms having 121, 145,
165, 189, or 206 residues per monomer (3-7). The different
isoforms share a common N-terminal receptor-binding
domain of 115 residues/monomer, while the longer forms
(VEGF
165
, VEGF
189
, and VEGF
206
) have the same C-terminal
50 residues which constitute a heparin-binding domain (8,
9). VEGF induces dimerization of the tyrosine kinase
receptors KDR [kinase insert domain containing receptor (10,
11)] or Flt-1 [fms-like tyrosine kinase-1 (12)], which
stimulates mitogenesis of vascular endothelium (13, 14)or
organizational effects on the vasculature (15, 16), respec-
tively. Receptor dimerization occurs as a consequence of the
symmetry of the VEGF molecule, which presents a pair of
identical binding sites located at the two poles of the VEGF
receptor-binding domain (17-20).
In addition to its normal physiological role, VEGF is
associated with numerous pathogenic states, including cancer,
rheumatoid arthritis, diabetic retinopathy and psoriasis;
development of VEGF antagonists is therefore clinically
attractive (21). Indeed, neutralizing monoclonal antibodies
(22, 23) and SELEX-derived RNA molecules (24) that target
VEGF, suppress tumor growth that is dependent on vascu-
larization of adjacent normal tissue (25). Humanized neutral-
izing antibodies have been shown to interact with VEGF near
the KDR and Flt-1 binding sites (18, 26), while the RNA
inhibitors are thought to interact specifically with the heparin-
binding domain (9, 24).
Multicopy display of random peptides on f1 filamentous
phage particles has allowed the efficient selection of novel
peptide ligands (27). We have used this technique to identify
small disulfide-constrained peptides that are capable of
blocking the interaction of VEGF with its receptors. The
peptides have substantially lower molecular weights than the
previously identified antibody or RNA antagonists and bind
independently of the heparin-binding domain to a region of
the receptor-binding domain common among the VEGF
isoforms found in vivo. The peptides are thus potentially
useful leads for the further development of VEGF antagonist
molecules.
MATERIALS AND METHODS
Expression and Purification of VEGF. The receptor-
binding domain of human VEGF (residues 8-109; VEGF
8-109
)
was overexpressed in Escherichia coli inclusion bodies,
purified, and refolded as described previously (28). For NMR
experiments,
15
N-labeled VEGF
11-109
was overexpressed in
E. coli as a His-tagged protein, purified, refolded, and cleaved
specifically, as described (29). Both VEGF constructs used
in the present study have been shown by phage ELISA to
have KDR-IgG binding affinities equivalent to that of
VEGF
1-109
(18).
Construction of Random Peptide Libraries. A Gene VIII
phagemid vector (30) was used to create phagemid libraries
for polyvalent display of random peptides. The vector
encodes the stII signal sequence followed by the peptide,
and a (Gly)
3
-Ser-(Gly)
3
-Ala spacer links the peptide to
the N-terminus of the Gene VIII coat protein. All peptide
* To whom correspondence should be addressed. Fax: (650) 556-
8824. E-mail: bcc@sunesis-pharma.com.
‡
Department of Protein Engineering.
§
Department of Bioorganic Chemistry.
|
Present address: Sunesis Pharmaceuticals, Inc., 3696 Haven Ave.,
Suite C, Redwood City, CA 94063.
1
Abbreviations: BSA, bovine serum albumin; DTT, dithiothreitol;
Flt-1, fms-like tyrosine kinase-1; HUVEC, human umbilical vein
endothelial cells; KDR, kinase insert domain-containing receptor;
VEGF, vascular endothelial growth factor.
17754 Biochemistry 1998, 37, 17754-17764
10.1021/bi981931e CCC: $15.00 1998 American Chemical Society
Published on Web 12/04/1998

sequences included a fixed pair of cysteine residues. Eight
different libraries were constructed as described (30). Indi-
vidual libraries had typical diversities of 5 × 10
8
and gave
a combined diversity consisting of about 4 × 10
9
unique
sequences.
Selection of VEGF-Binding Phage. A pool of seven
X
(i)
CX
(j)
CX
(k)
libraries (library A) and a X
4
CX
2
GPX
4
CX
4
library (library B) (Figure 1) were independently sorted for
VEGF binders as follows: in the first and second rounds of
sorting, biotinylated VEGF
8-109
(prepared by treatment of
VEGF with 1.5 molar equivalents of NHS-SS-Biotin
[sulfosuccinimidyl 2-(biotinamido) ethyl-1,3-dithiopropi-
onate; Pierce]) was bound to neutravidin-coated Maxisorp
plates (Nunc) and subsequently treated with Blocker Casein
Buffer (Pierce). Phage libraries were added in blocking buffer
and the plates were gently shaken for 1-2 h at room
temperature. Unbound phage were rinsed away and the
remaining bound phage were eluted by a 5 min treatment
with 75 mM DTT, which cleaves the disulfide linker and
releases immobilized VEGF. Eluted phage were then propa-
gated in 20 volume equivalents of log phase XL-1 blue cells
(Stratagene) for the next round of sorting. In the third and
fourth rounds of sorting, conditions were modified to prevent
the selection of casein or neutravidin binders. VEGF
8-109
was directly coated onto microtiter plates and blocked with
0.5% BSA. Bound phage were eluted with 10 mM HCl (in
100 mM NaCl and 0.3% BSA) instead of DTT because no
reducible cross-link was present. The isolated phage were
then neutralized with 0.05 vol of 1 M Tris (pH 8) and
propagated for the next round of selection. Specific binding
was monitored by measuring the number of phage that eluted
from VEGF-coated microtiter plate wells versus the number
eluted from identical wells that lacked the VEGF target.
Peptide Optimization by Soft Randomization. In contrast
to the entirely random libraries (excluding the fixed cysteine
residues) used to select the initial VEGF-binding peptides,
libraries intended to evolve these sequences further were
randomized only partially at each position. This process,
referred to herein as “soft randomization”, retains an overall
sequence similarity to the parent peptide while introducing
a fixed level of mutations from which improved sequences
can be selected. Since the mutations occur over the entire
sequence, soft randomization offers the advantage of enabling
the selection of multiple cooperative mutations as well as
simple point mutations. Additionally, the mutational fre-
quency can be adjusted as necessary to accommodate
different peptide lengths and library sizes. In this case, the
mutagenesis primers were synthesized using an 80:7:7:7
mixture (i.e., 80% original base and ∼7% each of the other
three bases) for each base in the randomized codons. This
translates to an amino acid mutation frequency of about 40%
at each residue position in the peptide sequence. Libraries
for soft randomization were prepared by site-directed mu-
tagenesis of a Gene III monovalent phage-display vector (31),
yielding 0.5 × 10
8
, 0.75 × 10
8
, and 2.0 × 10
8
independent
peptide sequences for classes 1, 2, and 3 (see Results for
definition of the peptide classes), respectively. The libraries
were sorted against directly coated VEGF
8-109
. Casein block
was used for the initial two rounds of sorting, and BSA block
was used in subsequent rounds. In each case, the phage-
binding buffer used was the same as the blocking buffer.
For the fourth and fifth rounds of sorting, an additional
chasing step was employed to select for phagemid with
slower off-rates. After rinsing away nonbinders, the plates
were incubated for 3-16 h in the presence of 100 nM
competing KDR to prevent the rebinding of dissociated
phage. Another rinse step removed these phage before the
bound phage were eluted with acid.
Peptide Optimization by Tailored Randomization. Libraries
for tailored randomization were prepared by site-directed
mutagenesis of the monovalent phage display vector de-
scribed above. Mutagenesis primers were synthesized using
equimolar mixtures of specific combinations of bases that
were chosen to encode amino acids observed in the soft
randomization selectants, while introducing as few additional
residues as possible. The peptide-encoding portions of the
oligonucleotides were GRG GAM CKG TGG TGC TTC
SAM GGT CCT SKT GMA TGG GTG TGT KGG KDK
VWW RDT (class 1), AGG GGC TGG GTG GAG ATC
TGC GMA KCG GAC GWM WDS GGG MDW TGC STG
AVT GRA SSA NNS (class 2), and VKC DDG NNS RMM
TGC GAC RTT VHT RKA ATG TGG GWG TGG SAS
TGC TTC GHA NKT KTM (class 3) (IUPAC nucleotide
codes). These libraries contained 2.6 × 10
7
, 1.4 × 10
7
, and
1.3 × 10
8
independent clones for classes 1, 2, and 3,
respectively. The first and second rounds of sorting were
performed as described in the previous section. The third
through seventh rounds of sorting were performed against
progressively lower concentrations of solution-phase bioti-
nylated VEGF
8-109
. After 2 h, bound phage were captured
with magnetic streptavidin beads (Strept. MagneSphere,
Promega). The beads were rinsed 6 times, and the phage
were eluted with DTT as described earlier.
Peptide Synthesis. Peptides were synthesized using stan-
dard 9-fluorenylmethoxycarbonyl (Fmoc) protocols, cleaved
off the resin with 5% triisopropylsilane in trifluoroacetic acid
(TFA), oxidized by potassium ferricyanide, and purified by
reversed-phase HPLC. Masses of each peptide were verified
by electrospray mass spectrometry.
Measurement of Binding Affinities. Affinities of peptides
for VEGF were measured by BIAcore (BIAcore Inc.), using
either a competitive binding assay or kinetic analysis of
peptide binding directly to immobilized VEGF. For the
competitive binding assay, monomeric human KDR (domains
1-3) and a heterodimeric VEGF variant with one binding
site per dimer (hV-1) were expressed, refolded, and purified
as described previously (20). Approximately 4000 refractive
index units (RUs) of KDR(1-3) were coupled to a CM5
FIGURE 1: Twenty-residue peptide-phage libraries used for selection
of VEGF-binding peptides.
Peptide Inhibitors of VEGF-Receptor Binding Biochemistry, Vol. 37, No. 51, 1998 17755