2′-Fluoropyrimidine RNA-based Aptamers to the 165-Amino Acid Form of Vascular Endothelial Growth Factor (VEGF165)

Vascular endothelial growth factor (VEGF) has been implicated in the pathological induction of new blood vessel growth in a variety of proliferative disorders. Using the SELEX process (systematic evolution of ligands byexponential enrichment), we have isolated 2′-F-pyrimidine RNA oligonucleotide ligands (aptamers) to human VEGF165. Representative aptamers from three distinct sequence families were truncated to the minimal sequence capable of high affinity binding to VEGF (23–29 nucleotides) and were further modified by replacement of 2′-O-methyl for 2′-OH at all ribopurine positions where the substitution was tolerated. Equilibrium dissociation constants for the interaction of VEGF with the truncated, 2′-O-methyl-modified aptamers range between 49 and 130 pm. These aptamers bind equally well to murine VEGF164, do not bind to VEGF121 or the smaller isoform of placenta growth factor (PlGF129), and show reduced, but significant affinity for the VEGF165/PlGF129 heterodimer. Cysteine 137 in the exon 7-encoded domain of VEGF165 forms a photo-inducible cross-link to a single uridine residue in each of the three aptamers. The aptamers potently inhibit the binding of VEGF to the human VEGF receptors, KDR and Flt-1, expressed by transfected porcine aortic endothelial cells. Furthermore, one of the aptamers is able to significantly reduce intradermal VEGF-induced vascular permeability in vivo.

The growth of new blood vessels, or angiogenesis, is an essential physiological response to increased demand for nutrients and the accumulation of metabolic end products. In normal physiological processes such as wound healing and the formation of corpus luteum and endometrium, angiogenesis is tightly regulated by positive and negative signals. In several disease states, however, overactive angiogenesis contributes to advancement of disease (1,2). Vascular endothelial growth factor (VEGF), 1 also known as vascular permeability factor, has recently emerged as a central positive regulator of angiogenesis. VEGF displays activity as an endothelial cell mitogen and chemoattractant in vitro (3)(4)(5) and induces vascular permeability and angiogenesis in vivo (4, 6 -8). VEGF and its two tyrosine kinase receptors, Flt-1 and Flk-1/KDR, are essential during embryonic development for the differentiation of endothelial cell precursors and the formation of a vascular network (9 -12). VEGF is secreted as a disulfide-linked homodimer that occurs in four isoforms (121, 165, 189, and 206 amino acids) that derive from alternatively spliced forms of a common mRNA (4,13). The two larger isoforms are cell matrix-associated as a consequence of their high affinity for heparin, while the smaller isoforms are more readily diffusible (13). VEGF 165 also binds heparin, while VEGF 121 does not (13). The role of different isoforms of VEGF in various biological contexts remains to be fully elucidated.
There is now substantial evidence that VEGF induces angiogenesis in several pathological settings. VEGF is secreted by a wide variety of cancer cell types and promotes the growth of tumors by inducing the development of tumor-associated vasculature (6, 7, 14 -18). Inhibition of VEGF function has been shown to limit both the growth of primary experimental tumors as well as the incidence of metastases in immunocompromised mice (19 -22). Elevated VEGF expression is correlated with several forms of ocular neovascularization that often lead to severe vision loss, including diabetic retinopathy (23), retinopathy of prematurity (24), and macular degeneration (25). VEGF may also play a role in inflammatory disorders such as rheumatoid arthritis (26) and psoriasis (27). Thus, agents that specifically inhibit VEGF may have great utility in combatting a variety of human diseases for which few effective treatments are presently available.
Nucleic acids, as a function of their primary structure, can fold into complex three-dimensional shapes with a great diversity of binding specificities. Using the SELEX (systematic evolution of ligands by exponential enrichment) process, oligonucleotides may be efficiently isolated from enormous randomized libraries of RNA, DNA, or modified nucleic acids that bind with high affinity and high specificity to various molecular targets (28,29). The method has been used to isolate ligands for proteins, peptides, carbohydrates, and small organic molecules (reviewed in Ref. 30). Such oligonucleotide ligands, termed "aptamers" (29), can be highly potent antagonists of enzyme catalysis or of specific protein-protein interactions (30). The potential utility of aptamers as therapeutic or diagnostic agents is considerably enhanced by chemical modifications that lend resistance to nuclease attack. In particular, substitution at the 2Ј-position of ribonucleotides with 2Ј-amino (2Ј-NH 2 ), 2Ј-fluoro (2Ј-F), or a variety of 2Ј-O-alkyl moieties confers resistance to ribonucleases that utilize the 2Ј-OH group for cleavage of the adjacent phosphodiester bond (31,32).
We have previously described the use of the SELEX process to identify RNA (33) and 2Ј-NH 2 -pyrimidine RNA aptamers to VEGF 165 (34). The incentive for performing SELEX experiments with 2Ј-F-pyrimidine RNA libraries, described in this report, was essentially 2-fold: first, we hoped to obtain nuclease-resistant aptamers that bind to VEGF with higher affinities than the 2Ј-NH 2 -pyrimidine-based aptamers. 2Ј-NH 2 modifications have been observed to decrease the stability of model DNA/DNA, RNA/ RNA, and RNA/DNA duplexes (35,36), while substitution of 2Ј-F in model duplexes dramatically increases their thermal stability (32,37,38). If 2Ј-NH 2 groups increase the conformational flexibility of oligonucleotides in general, the entropic cost of binding may limit the affinity of aptamers derived from 2Ј-NH 2 -pyrimidine RNA libraries (39). In contrast, 2Ј-F-pyrimidine aptamers may adopt more rigid conformations and, thus, may exhibit higher binding affinities for their targets. Second, apart from possible advantages related to binding affinity, the chemical synthesis of aptamers derived from 2Ј-F-pyrimidine libraries is considerably more economical. The coupling efficiency of 2Ј-F-pyrimidine phosphoramidites during oligonucleotide synthesis is greater than that of 2Ј-NH 2 -pyrimidine phosphoramidites and the 2Ј-F groups do not require protection/deprotection steps.
Here we report that VEGF aptamers isolated from 2Ј-Fpyrimidine RNA libraries generally display higher affinities for VEGF than do the 2Ј-NH 2 -pyrimidine RNA aptamers isolated previously (34). For three representative aptamers, the minimal sequence required for high affinity binding to VEGF is encoded in 23-29 nucleotides and all but two of the 2Ј-OHpurine positions can be substituted with 2Ј-O-methyl-(2Ј-OMe-) purines with only modest decreases in binding affinity. The minimal, substituted aptamers bind specifically to VEGF 165 with affinities between 49 and 130 pM and show no detectable binding affinity for VEGF 121 or the shorter isoform of placenta growth factor (PlGF 129 ), a protein with 53% homology to VEGF (40). The aptamers bind to the heterodimers of VEGF 165 and PlGF 123 , but with reduced affinities. A site of photo-cross-linking between each of the aptamers and VEGF 165 was mapped to Cys 137 in the carboxyl-terminal exon-7-encoded domain. In vitro, the 2Ј-F-pyrimidine-, 2Ј-OMe-purine-substituted VEGF aptamers inhibit the binding of VEGF 165 to both the human Flt-1 and KDR VEGF receptors expressed on porcine aortic endothelial cells. Furthermore, one of the aptamers blocks VEGF induction of vascular permeability as measured in the Miles assay (7), and thus shows potential utility as an inhibitor of VEGF-mediated effects in vivo.

Materials
Recombinant human VEGF 165 purified from the insect cell line Sf21 was purchased from R & D Systems (Minneapolis, MN) as a carrier-free lyophilized powder. The protein was resuspended in phosphate-buffered saline (PBS) to a concentration of 10 M and stored at Ϫ20°C in small aliquots until use. Aliquots were stored at 4°C for up to 4 weeks after thawing. Sf21-expressed mouse VEGF 164 , and Escherichia coli-expressed human VEGF 121 , VEGF 165 /PlGF 129 heterodimer, and PlGF 129 were also purchased from R & D Systems as carrier-free, lyophilized preparations.
Oligonucleotides were purchased from Operon Technologies, Inc. (Alameda, CA), or were synthesized in our laboratories using an Applied Biosystems Model 394 oligonucleotide synthesizer according to optimized protocols. The covalent coupling of polyethylene glycol (PEG) to aptamers was accomplished by synthesis of an oligonucleotide bearing a primary amine at the 5Ј-end using a trifluoroacetyl-protected pentylamine phosphoramidite, followed by reaction with 40-kDa PEG N-hydroxysuccinimide ester (Shearwater Polymers, Huntsville, AL). 2Ј-F-and 2Ј-OMe-ribonucleotide phosphoramidites were prepared by JBL Scientific, Inc. (San Luis Obispo, CA) for NeXstar Pharmaceuticals. 2Ј-F-pyrimidine nucleotriphosphates were also purchased from JBL. 2Ј-OH-purine nucleotriphosphates and deoxynucleotriphosphates were from Pharmacia Biotech (Piscataway, NJ). [␣-32 P]ATP and [␥-32 P]ATP were obtained from NEN Life Science Products (Boston, MA).

Methods
The SELEX Protocol-DNA oligonucleotide template libraries (5Ј-TAATACGACTCACTATAGGGAGGACGATGCGG(N 30 or 40 )CAGAC-GACTCGCCCGA-3Ј, where N ϭ any nucleotide) were prepared by chemical synthesis ("30N7" and "40N7"). Italicized nucleotides at the 5Ј-end of each template correspond to the T7 RNA polymerase promoter sequence. Oligonucleotide primers (5Ј-TCGGGCGAGTCGTCTG-3Ј ("3N7") and 5Ј-TAATACGACTCACTATAGGGAGGACGATGCGG-3Ј ("5N7")) were also synthesized for use in template amplification and reverse transcription. Double-stranded DNA templates were prepared by annealing primer 3N7 to the 30N7 or 40N7 libraries and extending the primer using Klenow DNA polymerase (New England Biolabs, Beverly, MA) at 37°C or avian myeloblastosis virus reverse transcriptase (Life Sciences, Inc., St. Petersburg, FL) at 45°C. We reasoned that the higher temperature of incubation used for the avian myeloblastosis virus reverse transcriptase reaction would facilitate complete extension through highly structured template oligonucleotides. 1 nmol of each library was transcribed using T7 RNA polymerase (Enzyco, Inc., Denver, CO) in the presence of 1 mM each of 2Ј-OH-(ATP and GTP), 3 mM each of 2Ј-F-(CTP and UTP), and 50 Ci of [␣-32 P]ATP. RNAs were purified from denaturing (7 M urea) polyacrylamide gels by excising and crushing the gel slice containing the RNA and soaking it for several hours or overnight in 2 mM EDTA. Approximately 5 nmol of RNA were obtained from each transcription.
The SELEX process of affinity selection followed by amplification of the selected pool has been described in detail (41). In brief, one round of selection and amplification was performed as follows: VEGF was mixed with a 5-or 10-fold excess of 32 P-radiolabeled RNA in PBS with 1 mM MgCl 2 (PBSM) (30N7 and 40N7 libraries) or in Tris-buffered saline, 1 mM MgCl 2 , 1 mM CaCl 2 (TBSMC) (30N7 library only). After incubation at 37°C for 15 min, the mixtures were passed through 0.45-m Type HA nitrocellulose filters (Millipore, Bedford, MA) to collect complexes of VEGF with RNA. The fraction of input RNA bound was monitored by measuring the radioactivity bound to the filter. RNAs were eluted from the filters by incubation in a 2:1 mixture of phenol (pH 7), 7 M urea. After precipitation from the aqueous phase, RNAs were annealed to primer 3N7 and reverse transcribed using avian myeloblastosis virus reverse transcriptase. The resultant cDNAs were amplified with 15 cycles of the polymerase chain reaction using the 3N7 and 5N7 primers and Taq DNA polymerase (Perkin-Elmer). Transcription of the polymerase chain reaction product yielded a new library enriched for sequences with affinity for VEGF. At round 4, the binding of the RNA libraries to nitrocellulose filters without added VEGF substantially increased in all three selected RNA pools. To deplete the pools of filter-binding RNAs, rounds 5 and 6 were performed with an alternative scheme for partitioning VEGF-bound RNAs from unbound molecules: after incubation of the 32 P-radiolabeled RNA pool with VEGF, each mixture was applied to an 8% polyacrylamide nondenaturing gel and run at 10 W for 45-60 min at 4°C. VEGF-RNA complexes migrated above the unbound RNA in this system and were visualized by autoradiography. For these rounds, selected RNAs were purified by the crush and soak method, as described above. The concentrations of RNA and protein were decreased in concert (from approximately 10 Ϫ7 M in the first selection to approximately 10 Ϫ12 M in the last round) as the affinity of the enriched pool for VEGF improved. After 12 rounds of selection and amplification, individual molecules in the selected pools were cloned using the pCR-Script Direct Cloning kit from Stratagene (La Jolla, CA). Plasmids were purified using the alkaline lysis method (PERFECTprep Plasmid DNA kit, 5 Prime 3 3 Prime, Inc., Boulder, CO) and sequences of the cloned regions were obtained using the Dye Terminator Cycle Sequencing kit available from Perkin-Elmer. Fluorescent sequencing ladders were read in the laboratory of Dr. Brian Kotzin, National Jewish Hospital, Denver, CO. Sequences were grouped into families and aligned by eye and with the aid of software designed at NeXstar Pharmaceuticals. 2 Measurement of Binding Affinities-Aptamers radiolabeled during transcription by incorporation of ␣-32 P-labeled nucleotriphosphates, or after synthesis using [␥-32 P]ATP and T4 polynucleotide kinase (New England Biolabs), were incubated in low concentration (typically less than 70 pM) with varying concentrations of VEGF or other proteins at 37°C for 15-20 min. Incubations were in TBS, PBS, or HEPES-buffered saline (HBS), pH 7.4, with or without divalent cations. Samples were passed through prewashed 0.45-m nitrocellulose filters followed by a 5-10-ml wash with binding buffer. Filters were immersed in scintillant and the radioactivity counted to quantitate the fraction of RNA bound at each protein concentration. The binding of individual aptamers was often biphasic in nature, consistent with a model in which two species that do not interconvert on the time scale of the experiment bind to a single site on VEGF with different affinities. Equations that describe the fraction of RNA bound as a function of K d and the total concentrations of RNA and protein (both measurable quantities) have been described for both monophasic and biphasic binding behavior (42). Because the concentrations of RNA used in these experiments were near the K d values of the aptamers and were too low to determine accurately, the least squares fitting of the data points to the binding equations was performed with the RNA concentration set to a negligibly low value. In making this assumption, we ensured that the binding affinities reported here are, at worst, underestimates of the actual values.
Affinity Selection of Aptamer Fragments-Ten pmol of internally radiolabeled transcripts of high affinity VEGF aptamers were partially digested with S7 nuclease (Boehringer Mannheim) to generate a mixture of radiolabeled fragments. One-tenth of the fragmented RNA was incubated with 10 pM VEGF in 45 ml of binding buffer, prior to filtration through nitrocellulose. Selected fragments recovered from the filter were run out on a high resolution denaturing polyacrylamide gel next to a lane loaded with the unselected fragment pool. The smallest selected bands were individually purified from the gel and further labeled at their 5Ј-ends with T4 polynucleotide kinase to increase their specific activity. One-half of the sample was annealed to a cDNA of the original transcript and extended to the end of the template using Sequenase DNA polymerase (U. S. Biochemical Corp., Cleveland, OH). Comparison of the migration of the purified fragment and its extension product to a standard sequencing ladder was used to determine the probable size and position of the selected fragment within the original transcript. Synthetic oligonucleotides corresponding in sequence to the affinity selected fragments were prepared to verify that the truncated aptamer retained affinity for VEGF.
2Ј-OMe Substitution-2Ј-F-pyrimidine oligonucleotides corresponding to truncated VEGF aptamer sequences were chemically synthesized using a 1:2 mixture of 2Ј-OMe-purine:2Ј-OH-purine phosphoramidites at five or six purine positions. Because 2Ј-OMe-nucleoside phosphoramidites couple with higher efficiency, the actual ratio of 2Ј-OMe-purine to 2Ј-OH-purine incorporated at each substituted position was roughly 3:1. The sequences of the oligonucleotides are shown below, with the substituted purine positions underlined. U and C represent 2Ј-F-uridine and 2Ј-F-cytidine, unless otherwise indicated. All oligonucleotides were synthesized using commercial sources of controlled pore glass beads, and thus, bear an additional 2Ј-OH-nucleotide at their 3Ј-ends. The sequences are as follows: t22.29-OMe1, GACGAUGCGGUAGGA-AGAAUUGGAAGCGC(U-2ЈOH); t22.29-OMe2, GACGAUGCGGUAG-GAAGAAUUGGAAGCGC(U-2Ј-OH); t22.29-OMe3, GACGAUGCGGU-AGGAAGAAUUGGAAGCGC(U-2ЈOH); t22.29-OMe4, GACGAUGCG-GUAGGAAGAAUUGGAAGCGC(U-2ЈOH); t2.31-OMe1, GGCGAACC-GAUGGAAUUUUUGGACGCUCGCC(U-2ЈOH); t2.31-OMe2, GGCGA-ACCGAUGGAAUUUUUGGACGCUCGCC(U-2ЈOH); t2.31-OMe3, GGCGAACCGAUGGAAUUUUUGGACGCUCGCC(U-2ЈOH); t44.29-OMe1, GCGGAAUCAGUGAAUGCUUAUACAUCCGC(U-2ЈOH); t44.29-OMe2, GCGGAAUCAGUGAAUGCUUAUACAUCCGC(U-2ЈOH); t44.29-OMe3, GCGGAAUCAGUGAAUGCUUAUACAUCCGC-(U-2ЈOH). Each oligonucleotide was radiolabeled at the 5Ј-end and incubated with VEGF at 320, 60, and 20 pM concentration. The mixtures were filtered through nitrocellulose and bound RNAs were collected by incubation of the filter in 2:1 phenol, pH 7, 7 M urea. Selected RNAs were collected from the aqueous phase by ethanol precipitation. The selected RNAs, along with aliquots of each unselected, radiolabeled oligonucleotide, were subjected to partial alkaline hydrolysis by incubation at 90°C in 50 mM sodium carbonate buffer, pH 9, for 11 min. Hydrolyzed samples were applied to a 20% polyacrylamide, 7 M urea gel. Radioactive bands were visualized using a Fuji Fujix BAS 1000 PhosphorImager and the intensity of bands corresponding to hydrolysis at individual purine positions was quantitated using MacBAS software, version 2.0. Band intensities were normalized to the total intensity in the lane to correct for variability in sample loading. Band intensity ratios were determined for each purine position by dividing the normalized band intensity in the affinity selected sample by the normalized band intensity at the same position in the unselected sample. Values for the two or three oligonucleotides where a particular purine was not substituted were averaged to obtain a baseline band intensity ratio. Band intensity ratios for substituted positions that fell well below or above the baseline value provided a qualitative indication of positions that show bias for or against 2Ј-OMe substitution.
Temperatures of Melting (T m )-Approximately 10 g of aptamer were diluted in 2.5 ml of degassed buffer. Absorbance at 260 nm was monitored in a Varian Cary spectrophotometer relative to a buffer reference as the temperature of the sample was raised from 10 or 20°C to 90 or 95°C at a rate of 1°/min. T m values were determined by fitting the data to a mathematical model (43) in which each aptamer is assumed to occupy one of two states (folded or unfolded). The baseline absorbance of the folded and unfolded states is assumed to be linear with changes in temperature. Six parameters describe the mathematical model, including the slope and intercept of the upper and lower linear baselines and values for the ⌬H and ⌬S of the folded to unfolded transition. The temperature at which ⌬G ϭ 0 (T m ) was calculated from the fitted values for ⌬H and ⌬S. T m values for aptamers t22.23, t22-OMe, t2.29, and t2-OMe were determined in PBS. For t44.27 and t44-OMe, HBS with 1 mM EGTA was used for an initial determination; after cooling, CaCl 2 or MgCl 2 was added to 2 mM final concentration and the T m was measured again.
Binding Rate Constants-A small amount (typically less than 1 pmol) of 5Ј-radiolabeled aptamers were incubated with 1 nM VEGF at 37°C in 1 ml of buffered saline supplemented with divalent cations. At time 0, 50 l were filtered through nitrocellulose to determine the fraction of RNA bound to protein, then an excess (100 or 500 nM final concentration) of unlabeled aptamer was added in a volume of 2-4 l and 50-l aliquots were filtered at time points thereafter. Filters were counted in scintillant to determine the amount of radiolabeled RNA still bound to VEGF at each time point. The data, plotted as fraction of RNA bound (f) versus time, were fit to a first order rate equation, f (t) ϭ f 0 e Ϫ(kd)t ϩ b, where f 0 is the fraction of RNA bound at time 0, k d is the dissociation rate constant, and b is the residual binding of radiolabeled RNA to the filter at infinite time. Association rate constants (k a values) were calculated from the measured k d and K d values according to the equation, k a ϭ k d /K d .
Photo-cross-linking of Aptamers to VEGF-Truncated VEGF aptamers were synthesized with 2Ј-OH, 5-iodo-U (5-I-U) in place of 2Ј-F-U at one position in the molecule. All possible substituted oligonucleotides were prepared for each of the aptamers. The substituted aptamers were initially screened for their capacity to form a cross-link to VEGF 165 : a trace amount of 5Ј-end-radiolabeled oligonucleotide was incubated with 0.1 or 1 M VEGF at 37°C in binding buffer, then exposed to pulses of 308 nm monochromatic light in a 1-cm path length cuvette using a XeCl excimer laser (175 mJ/pulse, 20 pulses/s). The cuvette was positioned 50 cm from a lens of 10-cm focal length. Aliquots were removed from the cuvette after 0, 500, 2000, 5000, and 7500 pulses and the samples were applied to an 8% polyacrylamide, 7 M urea gel to separate the more slowly migrating cross-linked VEGF-aptamer complexes from free radiolabeled RNA. The efficiency of cross-linking was estimated for each aptamer from a PhosphorImage of the gel. Aptamers with no 5-I-U substitutions cross-linked with an efficiency of 1% or less. Substituted oligonucleotides varied in efficiency from 1 to 33%. For the aptamers with the highest cross-linking efficiency, cross-linked material was prepared at a preparative scale for trypsin or chymotrypsin digestion and peptide sequencing. 2 nmol of 5-I-U-substituted aptamer were mixed with 2 nmol of VEGF in 2-ml volume and the mixture was irradiated as described above with 5000 pulses of light. The irradiated sample was precipitated and resuspended in 0.5 M Tris⅐HCl, pH 7.5, 8 M urea, 2 mM EDTA and reduced and alkylated according to published procedures (44), with some modifications. Dithiothreitol was added to 10 mM final concentration and the sample was incubated at 37°C for 1 h. Iodoacetamide was added to 12 mM final concentration and incubation was continued at 37°C for 1 h. Alkylation of sulfhydryls in the protein allows a more reliable assignment of cysteine residues during automated peptide sequencing. The reduced and alkylated cross-linked complex was precipitated and resuspended in 15 l of 1% SDS. 15 l of 1 M Tris⅐HCl (pH 8.5 for trypsin digestions or pH 7.8 for chymotrypsin digestions) were added and trypsin or chymotrypsin (sequencing grade, Boehringer Mannheim) dissolved in 1 mM HCl was added in equal mass to the VEGF. The sample was brought to 150 l with water and incubations were performed at room temperature or 37°C for several hours. Typically, an additional 5 g of enzyme was added and incubation was continued overnight. The peptide-RNA complexes were separated on a 20% polyacrylamide, 8 M urea gel and electroblotted to a polyvinylidene difluoride membrane. Alternatively, the major digestion product was excised from the gel, extracted by incubation in 2.5 M ammonium acetate overnight at 37°C, and precipitated with ethanol. Samples were resuspended by boiling in a small volume of 1% SDS before diluting to 0.05% SDS final concentration and applied to a polyvinylidene difluoride membrane using a ProSorb cartridge (Perkin-Elmer) according to the manufacturer's recommendations. Samples were submitted to Dr. David McCourt at Midwest Analytical, Inc., St. Louis, MO, for automated peptide sequencing.
Receptor Binding Inhibition-Porcine aortic endothelial (PAE) cells transfected with either the flt-1 or kdr VEGF receptor gene have been described previously (45). Cells were cultured in Ham's F-12 (Life Technologies, Inc. or Biochrom) supplemented with penicillin/streptomycin and 10% fetal calf serum (Life Technologies, Inc.). Confluent cells in 24-well tissue culture plates (approximately 400,000 cells/well) were washed twice with ice-cold Ham's F-12 medium, 1 mg/ml bovine serum albumin, then incubated for 2 h on ice in 0.5 ml of the same medium containing 35,000 cpm/ml 125 I-labeled VEGF (Amersham) and varying concentrations of aptamer or control oligonucleotide. The cells were washed 3 times with ice-cold binding medium, then lysed in 20 mM Tris⅐HCl, pH 7.5, 1% Triton X-100, 10% glycerol. Cell lysates were counted in a ␥-counter. 125 I-VEGF binding in the absence of competitor was taken as the maximum signal; nonspecific binding of the radiolabeled protein was determined by incubation in the presence of 50 ng/ml unlabeled VEGF. Estimates for the maximum and minimum values of each competition curve, expressed as a percentage of maximum signal, were obtained by fitting the data to a competition equation that describes mutually exclusive binding of two species to the same target (46). The concentration of aptamer inhibiting 50% of the maximum signal above background (IC 50 ) was determined from the fitted curve.
Dermal Vascular Permeability Assay-The ability of the minimal 2Ј-OMe-modified aptamers to attenuate VEGF-induced changes in the permeability of the dermal vasculature (Miles assay) was performed as described previously (7) with minor modifications. Briefly, adult female guinea pigs (3/study) were anesthetized with isoflurane and the hair on the dorsal and lateral back areas was removed with clippers. Evans Blue dye (2.5 mg/guinea pig) was administered intravenously. Injection solutions (PBS, VEGF, aptamers, and anti-VEGF monoclonal antibody) were prepared Ն30 min in advance, co-mixed where indicated, with final concentrations as shown. Each solution was then injected intradermally (duplicate injections/guinea pig; 40 l/site) in a randomized manner in a grid pattern drawn on the clippered area. Guinea pigs were allowed to recover from anesthesia and were sacrificed by CO 2 exposure 30 min after completion of the intradermal injections. The skin was then harvested, trimmed free of subcutis, and transilluminated. Images were captured and analyzed using a color CCD camera (Hitachi Denshi KP-50U, Japan) and Image-Pro Plus software (Version 3.1, Media Cybernetics, Silver Spring, MD). Each skin sample was normalized for intensity with each injection site analyzed for optical density and the area involved. The resultant vascular permeability indices were averaged for duplicate spots on the same animal, then normalized to the average signal obtained with VEGF alone on that animal (% of control). The mean and S.E. were calculated from the normalized data for three guinea pigs (n ϭ 3) for all aptamer-containing samples and for six animals (n ϭ 6) for injection of PBS alone.

RESULTS
Selection of Aptamers-Aptamers to VEGF 165 were isolated in three separate SELEX experiments from 2Ј-F-pyrimidine RNA libraries containing 30 or 40 random nucleotides. Selections were performed in PBS supplemented with 1 mM MgCl 2 (30N and 40N libraries) or in TBS with 1 mM MgCl 2 and 1 mM CaCl 2 (30N library only). Approximately 1 nmol of RNA was included in the first selection cycle of each experiment. After 10 cycles, the affinity between VEGF and each RNA pool had improved approximately 1000-fold relative to the starting pools (data not shown). As no further improvement in binding affinity was observed after two additional cycles, individual members of the 12th round pools were cloned and sequences were determined for about 50 isolates from each selection.
Of a total of 143 clones analyzed, 75 sequences differing by more than one nucleotide were obtained. 46 of these sequences were grouped into three major families based on conserved primary structure motifs (Table I). The remaining sequences could be grouped into minor families with five or fewer members or were orphan sequences that were not obviously related to any other sequence. Ligands containing the primary structure motif defined by Families 1 and 2 were enriched in all three affinity selections. Family 1 ligands share a strongly conserved sequence (boldface in Table I) flanked by variable regions. Although possible base pairing interactions may be identified for some of the ligands, generally between the 5Ј-fixed sequence and nucleotides on the 3Ј-side of the conserved sequence motif, no predicted secondary structure common to all or most of the ligands is evident. In contrast, members of Family 2 share the ability to form a short base paired stem (underlined in Table I) enclosing a conserved, discontinuous sequence motif. With the exception of the closing A/U base pair, the sequence identity of bases in the putative stem regions is not conserved. Such co-variation of bases that conserves secondary rather than primary structure supports the existence of the putative stem and suggests that this structure may be important for the high affinity conformation of this family of VEGF aptamers. The conserved primary structure motifs of Family 1 and Family 2 are similar: most ligands include the sequence 5Ј-GAAN (3)(4) UUGG-3Ј. Indeed, two members of Family 2 (VP40.9 and VP40.14) could as easily be grouped with Family 1 sequences. Members of both families are thus likely to form similar complexes with VEGF.
A third family of ligands were enriched only in the selections performed in TBSMC (Family 3, Table I). In addition to a highly conserved primary structure motif (boldface, Table I), the members of this family also share a predicted secondary structure: 3-10 nucleotides 3Ј of the conserved sequence region show base pairing complementarity to nucleotides in the 5Ј-fixed region adjacent to the selected sequence region (underlined in Table I). For three of the ligands (VT30.29, VT30.44, and VT30.54), the 5Ј-side of the putative stem extends one additional base into the variable sequence region. Our efforts to define a minimal high affinity sequence derived from this family (described below) were guided by the strong conservation of this motif.
The affinities of the individual RNA ligands for VEGF were determined as described under "Experimental Procedures." With few exceptions, the ligands showed very high affinity for VEGF, with most K d values ranging between 5 and 50 pM.
Minimal Aptamers-To identify the minimal sequence elements that confer high affinity binding to VEGF, we used both a biochemical approach and predictions based on conserved secondary structure motifs to derive a high affinity truncated aptamer from one member of each sequence family (asterisks in Table I). For clone VP30.22 from Family 1, which shares no obvious secondary structure potential with other members of the group, an initial prediction for a minimal sequence was made by mapping the ends of a purified, affinity-selected fragment of the full-length aptamer (data not shown and see "Experimental Procedures"). The sequence of the 29-nucleotide high affinity fragment (5Ј-GACGAUGCGGUAGGAAGAAUU-GGAAGCGC-3Ј) encompassed the conserved sequence motif, and an oligonucleotide corresponding to the selected fragment (t22.29; truncated aptamer derived from clone VP30.22, 29 nucleotides in length) showed an approximately 3-fold loss in binding affinity for VEGF (K d ϭ 60 pM) relative to the full-length ligand. Further truncation at the 3Ј-end of this molecule caused a precipitous loss in affinity. In contrast, up to 6 additional nucleotides could be removed from the 5Ј-end with little or no effect on binding affinity. The resultant 23-nucleotide aptamer, t22.23 (Table I, gray shading), included all of the consensus primary structure motif and bound to VEGF with a K d of 90 pM. For Families 2 and 3, the conserved regions of base complementarity (indicated by underlining in Table I and  Similarly, the truncated aptamer t44.29 (5Ј-GCGGAAUCAGU-GAAUGCUUAUACAUCCGC-3Ј) and a shorter, 27-nucleotide molecule lacking the G/C base pair at the base of the putative stem retained an affinity for VEGF equivalent to that of the full-length molecule (K d ϭ 10 pM). However, further truncation of the stem to yield a 25-nucleotide aptamer increased the K d to 60 pM. Based on these data, 29-and 27-nucleotide minimal sequence aptamers (t2.29 and t44.27) were chosen for clones VP30.2 and VT30.44, respectively (Table I, gray shading).

2Ј-OMe Substitution at
Purines-Substitution at the 2Ј-OH positions of RNA oligonucleotides by 2Ј-OMe improves their stability against nucleases present in a variety of biological fluids (34,47), and, like 2Ј-F-modified nucleotides, allows for more efficient chemical synthesis of aptamers because the 2Ј-OMe group does not need to be protected. Unfortunately, 2Ј-OMe-modified nucleoside triphosphates are not generally accepted as substrates by RNA polymerases under standard reaction conditions. However, 2Ј-OMe-purines may be incorpo-  165 Oligonucleotide ligands to VEGF 165 were isolated in three separate SELEX experiments. Individual clones were isolated and sequenced and the sequences grouped into families based on shared primary structural motifs. The name of each ligand indicates the target (V ϭ VEGF), the selection buffer (P ϭ PBS; T ϭ TBS), the length of the randomized region in the library (30 or 40 nucleotides) and the clone number (following the decimal). The frequency with which a sequence appeared among the clones analyzed is indicated in parentheses; sequences that differed by only one nucleotide were attributed to polymerase chain reaction mutations of a common precursor and were grouped together with the variable base indicated in the sequence by the appropriate symbol (Y ϭ U or C). The fixed sequences common to all ligands are shown in lower case letters at the top. For individual clones the sequence of the variable region is shown in upper case. For some ligands, fixed region sequences in lower case are appended to the variable region sequence where they contribute to a conserved sequence motif or possible secondary structures. Underlining indicates regions of potential base pairing. The value of K d for the high affinity binding fraction of each ligand is shown (N.D., not determined). One ligand in each family was selected for further analysis (asterisks). Gray boxes indicate the minimal high affinity aptamer derived from each sequence. rated at any position in a specific oligonucleotide by chemical synthesis. We and others 3 have observed that high affinity RNA ligands generally accept a high percentage of 2Ј-OMepurine substitutions with little or no loss of affinity for the target protein (34,48). To identify those purine positions in aptamers VP30.22, VP30.2, and VT30.44 where 2Ј-OMe substitution is compatible with high affinity binding to VEGF, syntheses of the truncated aptamers t22.29, t2.31, and t44. 29 were prepared in which five or six purines at a time were partially substituted with the modified nucleotide. Affinity selection of each partially substituted library was used to isolate those molecules that retained substantial affinity for VEGF. In such an affinity selected pool, positions that do not tolerate substitution are biased for 2Ј-OH and thus show higher sensitivity to hydrolysis by alkali relative to the same position in the unselected library. 5Ј-Radiolabeled unselected and affinity selected pools were partially hydrolyzed by alkali and the products were displayed on a high resolution polyacrylamide gel. A representation of the relative band intensity at each purine position is shown in Fig. 1. Open circles indicate the average "baseline" band intensity ratio for each purine position determined from those libraries where the position was unmodified. Filled circles represent relative band intensities where the position was partially 2Ј-OMe-substituted. A position that displays a strong preference for 2Ј-OH may be easily identified from these plots by a filled circle that falls well above the baseline value for that position (34). For t22.29, G4 and A6 showed substantial bias for 2Ј-OH in the affinity selected pool, as did A5 and G20 in t2.31, and A4 and A5 in t44.29 (Fig. 1). Note that the numbering of the nucleotide positions is adjusted such that the first nucleotide of the minimal truncated sequence, described above, is position one.
While the foregoing analysis identifies those positions at which the 2Ј-OMe substitution is likely to result in reduced affinity, one cannot predict from these data how simultaneous modification of all other purines will affect the binding affinity. This analysis was conducted using the minimal sequence aptamers described above. Substitution of all but two purines with 2Ј-OMe-purines in the minimal ligand t22.23 (t22-OMe) had little or no impact on its binding affinity for VEGF (K d ϭ 72 pM, Table II). Aptamer t2-OMe, corresponding to the minimal aptamer t2.29 substituted at all purines except A5 and G20, showed a 3.5-fold increase in K d relative to the unsubstituted molecule (130 pM compared with 40 pM), while t44-OMe (t44.27 substituted at all but the adjacent residues A4 and A5) showed a 5-fold increase, from 10 to 49 pM (Table II). Thus, the effect of the combined 2Ј-OMe substitutions on the K d values of the minimal aptamers was either neutral or slightly unfavorable. Further substitution at either or both of the remaining 2Ј-OH positions in each of the aptamers resulted in at least a 10-fold decrease in binding affinity.
Divalent Cation Dependence of Aptamer Binding-Aptamers in Families 1 and 2 were selected in the presence of magnesium cations while Family 3 aptamers were selected in a buffer containing both magnesium and calcium. Since divalent cations may stabilize RNA structures by binding within a specific pocket formed by the RNA or through nonspecific interaction with the phosphodiester backbone, we asked whether magnesium and/or calcium were required for the high affinity binding of representative aptamers to VEGF. The affinities of t22-OMe and t2-OMe (from Families 1 and 2, respectively) were unchanged in the presence or absence of supplemental divalent cations or the chelating agent EDTA (data not shown). How-ever, for Family 3 ligands, as represented by t44-OMe in Fig. 2, calcium was absolutely required for high affinity binding to VEGF. Binding was dramatically reduced (K d Ͼ 10 Ϫ7 ) when divalent cations in the binding buffer were replaced with EGTA. The addition of excess MgCl 2 to the binding buffer depleted of divalent cations resulted in no improvement in binding affinity. In contrast, CaCl 2 , in 2-fold molar excess over EGTA, fully restored binding activity. Similar binding behavior was observed for the unmodified aptamer t44.29 and for other members of the sequence family (data not shown).
Thermal Denaturation Properties of Aptamers-T m , the inflection point of the thermal denaturation profiles of nucleic acids, is often used as an indicator of the thermodynamic stability of the structured conformation. A higher T m suggests that more energy is required to disrupt the folded conformation and that a greater proportion of molecules will occupy the folded state at 37°C, the temperature at which binding affinities are measured. The T m values determined for the minimal FIG. 1. Effect of 2-OMe-purine substitutions on binding. 5Ј-Radiolabeled libraries were prepared for each of three truncated aptamers in which five or six 2Ј-OH-purine positions were partially 2Ј-OMesubstituted, as described under "Experimental Procedures." Each library was incubated with VEGF, and substituted oligonucleotides bound by the protein were collected on nitrocellulose filters. The selected pool and the starting unselected library were partially hydrolyzed by alkali and the products were displayed on a high resolution polyacrylamide gel. The "band intensity ratio" corresponds to the normalized PhosphorImage signal obtained from hydrolysis at a particular purine position in the selected pool divided by the signal obtained for the same position in the unselected library. Filled circles represent band intensity ratios where the position was partially 2Ј-OMe substituted in the library; open circles show the average band intensity ratio Ϯ S.D. for all libraries in which the position was unsubstituted. Filled circles that fall well above the range for a particular position are indicative of a bias for 2Ј-OH (against 2Ј-OMe) in the affinity selected pool. The numbering of the nucleotide positions in each aptamer is such that the first nucleotide of the minimal aptamer sequence is position one. length aptamers were 48°C for t22.23, 63°C for t2.29 (both measured in PBS), and 57°C for t44.27 (measured in HBS with 1 mM EGTA and 2 mM CaCl 2 ). These T m values are somewhat higher (for t22.23) or considerably higher (for t2.29 and t44.27) compared with those observed for 2Ј-NH 2 -pyrimidine aptamers to VEGF (34) and other targets. 4 The T m values determined for the three 2Ј-OMe-substituted aptamers were slightly higher than those of the unsubstituted aptamers (Table II), consistent with the increase in thermal stability observed generally for 2Ј-OMe substitution in model duplexes (32,37) and aptamers (34). The measured values were 49°C for t22-OMe, 66°C for t2-OMe, and 62°C for t44-OMe. Family 3 ligands show an absolute dependence on the presence of calcium for high affinity binding to VEGF. Since calcium might affect the stability of a particular solution conformation of these aptamers, we asked whether the observed thermal stability of aptamer t44-OMe was significantly different in the absence of calcium. In HBS with 1 mM EGTA only, the T m was 59°C; thus, the presence of calcium appeared to cause a slight increase of 3°C in the thermal stability of the aptamer. However, in the presence of HBS with 1 mM EGTA and 2 mM MgCl 2 , a buffer in which high affinity binding to VEGF was not observed (Fig. 2), the T m was identical to that measured in HBS/EGTA/CaCl 2 (62°C). Thus, the specific effect of calcium, as compared with magnesium, on the high affinity interaction of t44-OMe with VEGF cannot be observed by comparison of the thermal stability of the aptamer in the presence of either divalent cation.
Binding Rate Constants for Substituted Aptamers-Dissociation rate constants (k d ) were determined for each of the three 2Ј-OMe-substituted aptamers by following the loss of a preformed complex between radiolabeled aptamer and VEGF upon the addition of a large excess of unlabeled aptamer. t22-OMe showed the fastest rate of dissociation with a k d of 0.012 s Ϫ1 , corresponding to a t1 ⁄2 of 60 s (Table II). t2-OMe and t44-OMe showed slightly slower rates of dissociation (k d ϭ 0.0042 and 0.0074 s Ϫ1 , or t1 ⁄2 values of 170 and 90 s, respectively. Association rate constants (k a ), calculated from the dissociation rate constant and the equilibrium dissociation constant (k a ϭ k d / K d ), ranged from 3 ϫ 10 7 to 2 ϫ 10 8 M Ϫ1 s Ϫ1 (Table II). Such rapid rates of association suggest that the binding interaction between these aptamers and VEGF may be nearly diffusionlimited, and are similar to the association rate constants determined for aptamers derived to other targets (49) or from other libraries (34,42).
Specificity of Aptamers-The oligonucleotides described here were selected based on their affinities for VEGF 165 . All three minimal 2Ј-OMe-substituted aptamers bind to human VEGF 165 and its mouse homologue (VEGF 164 ) with comparably high affinity (Fig. 3). No binding, or minimal binding, was observed for the three aptamers with up to 100 nM VEGF 121 , PlGF 129 , a protein with 53% homology to VEGF (40), or reduced VEGF 165 (Fig. 3). Because the affinity of the aptamers for PlGF was very low, we tested the ability of each ligand to recognize a single subunit of VEGF in the context of a VEGF/PlGF heterodimer. Although the binding affinity of t22-OMe and t44-OMe for the VEGF/PlGF heterodimer was lower compared with the VEGF 165 homodimer, an appreciable fraction of high affinity binding remained (K d ϭ 750 and 430 pM, respectively). In contrast, aptamer t2-OMe bound the heterodimer with considerably lower affinity (K d ϭ 40 nM) (Fig. 3).
Photo-cross-linking Aptamers to VEGF 165 -To probe for sites of close contact between VEGF and the three minimal aptamers, oligonucleotides corresponding in sequence to t22.23, t2.31, and t44.29 were synthesized in which one 2Ј-F-U position at a time was substituted with 2Ј-OH-, 5-iodo-U (5-I-U). Cross-linking may occur when a photo-induced uridinyl radical generated from the 5-iodo-substituted base reacts with an amino acid in close proximity within the bound complex (50,51). While the effect of most single 5-I-U substitutions on the affinity of the aptamers for VEGF was small, some substitutions did result in lower binding affinity for VEGF. 5-I-U substitution at U14 or U15 in aptamer t22.23, at U18 or U19 in t2.31, and at U6 in aptamer t44.29 caused a significant loss in binding affinity and/or significantly reduced the fraction of RNA that bound with higher affinity. The lower binding affinity displayed by these aptamers was acceptable in these experiments because 4 B. Feistner and S. Gill, unpublished observations.

TABLE II
Binding parameters and T m values of 2Ј-OMe-substituted minimal aptamers Truncated ligands t22.23, t2.29, and t44.27 were chemically synthesized with all but two 2Ј-OH-purine positions (boldface) substituted by 2Ј-OMe-purines. All three aptamers (t22-OMe, t2-OMe, and t44-OMe) were terminated by a 3Ј-3Ј-linked deoxythymidine (not shown). Asterisks mark the positions where 5-iodo-U substitution permits a photo-induced cross-link to VEGF 165 . Values for K d and k d represent average (ϮS.E.) for five or six determinations or three to five determinations, respectively. k a values were calculated from the relationship k a ϭ k d /K d . Errors for k a calculations were determined using the delta method. T m values were determined in PBS for aptamers t22-OMe and t2-OMe and in HBS with 1 mM EGTA and 2 mM CaCl 2 for t44-OMe.

2Ј-Fluoropyrimidine RNA-based Aptamers to VEGF 165
the cross-linking was performed at relatively high concentrations of VEGF and oligonucleotide. 5Ј-End-radiolabeled aptamers were incubated with VEGF and the mixtures were exposed to pulses of 308 nm monochromatic light. A slowly migrating cross-linked complex was separated from free RNA by electrophoresis through a denaturing polyacrylamide gel. Complex formation required the presence of VEGF and was inhibited by the presence of an excess of high affinity unsubstituted VEGF aptamer, but not by an excess of oligonucleotide with low affinity for VEGF (for example, Fig. 4a). The most efficient cross-linking was observed for 5-I-U14 substitution in t22.23 (15%), 5-I-U18 substitution in t2.31 (34%), and 5-I-U14 substitution in t44.29 (10%). Less efficient cross-linking was observed for several of the aptamers substituted at other positions, while less than 1% cross-linking was observed with the unsubstituted oligonucleotides. The highest efficiency cross-linked complex for each aptamer was generated on a preparative scale and digested with trypsin or chymotrypsin. The products were separated by electrophoresis and the major cross-linked fragment was eluted from the gel and submitted for automated peptide sequencing. Trypsin digestion of VEFG-aptamer complexes derived from either t22.23 (5-I-U14) or t2.31 (5-I-U18) yielded the major peptide sequence, _SCK, where the underscore corresponds to a blank in the sequencing data. Tryspin cleavage after Lys 136 and Lys 140 in VEGF would yield a peptide of sequence CSCK (Fig. 4b). These data were thus consistent with the formation of a photo-inducible cross-link between aptamer t22.23 (5-I-U14) or t2.31 (5-I-U18) and Cys 137 of VEGF. However, a second minor peptide was also present in both sequences in 5-10-fold lower molar yield that mapped to the amino-terminal domain of VEGF. To clarify these results, cross-linked VEGF was also digested with chymotrpysin and the major fragment was purified and sequenced. The sequence, VQDPQTCK_SCKN, mapped to residues 129 through 141 of VEGF, where the blank residue indicated by the underscore again corresponds to Cys 137 (Fig. 4b). Peptide sequence beyond Asn 141 was not obtained. Minor peptides present in these sequences were smaller chymotryptic fragments of the major sequence or were fragments of chymotrypsin itself. Aptamer t44.29 (5-I-U14) was cross-linked to VEGF and digested with chymotrypsin only. The sequence of the major cross-linked chymotryptic fragment again corresponded to residues 129 through 141 of VEGF. In this case, both Cys 137 and Ser 138 were not detected in the peptide sequence; however, since serine gave a relatively weak signal in all sequences obtained, we believe these data are interpreted most simply as a cross-link at Cys 137 and a poorly yielding sequencing cycle at Ser 138 . Thus, all three aptamers appear to bind to VEGF in a manner which brings the aptamer surface in close proximity to Cys 137 .
VEGF Receptor Binding Inhibition-Previous selections for oligonucleotide ligands to VEGF yielded molecules that bound with high affinity to the growth factor and inhibited its binding toVEGF receptors expressed by human umbilical vein endothelial cells (33,34). Two VEGF receptors, Flt-1 (fms-like tyrosine kinase) and KDR (kinase insert domain-containing receptor), have been identified on human vascular endothelial cells (52,53). Site-specific mutations in the VEGF protein differentially impact its association with Flt-1 or KDR, suggesting that different regions of VEGF form close contacts with one or the other receptor (54,55). We assessed the capacity of the minimal 2Ј-OMe-substituted aptamers to inhibit VEGF binding to each of the receptors individually. PAE cells transfected with either the flt-1 or kdr human VEGF receptor gene (45) were incubated with 125 I-labeled VEGF 165 in the absence or presence of increasing concentrations of the three truncated, 2Ј-OMe-substituted aptamers. Cell-associated VEGF decreased, in each case, with increasing concentration of the aptamer (Fig. 5). Higher concentrations of aptamer were required to inhibit VEGF binding to PAE/Flt-1 cells compared with PAE/KDR cells, in accord with the reported higher binding affinity of Flt-1 for VEGF compared with KDR (16). In general, the concentration of each aptamer required to inhibit 50% of the 125 I-labeled VEGF binding above background (IC 50 ) correlated with the affinity of each ligand for VEGF. IC 50 values for aptamer competition with the Flt-1 receptor ranged from 5 ϫ 10 Ϫ11 to 3 ϫ 10 Ϫ10 M, compared with 2 ϫ 10 Ϫ7 M for a sequence-scrambled analog of t44-OMe (scr-t44-OMe, K d ϭ 350 nM for binding to VEGF). 50% inhibition of VEGF binding to KDR required about 2 or 3 ϫ 10 Ϫ12 M for aptamers t22-OMe and t44-OMe, respectively, 6 ϫ 10 Ϫ11 M for t2-OMe, and 5 ϫ 10 Ϫ8 M for scr-t44-OMe. None of the three minimal aptamers at 1 nM concentration inhibited the binding of 125 I-platelet-derived growth factor-BB to PAE cells expressing the platelet-derived growth factor ␤-receptor.
Vascular Permeability Assay-The Miles assay (7) offers a simple and rapid means of monitoring the ability of various compounds to inhibit the activity of VEGF in vivo. In this FIG. 3. Specificity of aptamer binding. The affinity of aptamers t22-OMe (upper panel), t2-OMe (middle panel), and t44-OMe (lower panel) for VEGF and related proteins described in the text were determined as described under "Experimental Procedures." Measurements were performed in PBSM for t22-OMe and t2-OMe and in HBSMC for t44-OMe. h, human; m, murine; red, reduced with dithiothreitol. assay, intradermal injection of VEGF in adult guinea pigs induces a rapid increase in the permeability of dermal microvessels that may be monitored by quantitating the leakage of intravascular Evans Blue dye into the skin. Co-injection of VEGF with an excess of neutralizing anti-VEGF antibody reduces the dye leakage to a level equivalent to that upon injection of PBS alone. Preincubation of VEGF with 1 or 0.1 M of each 2Ј-OMe-substituted aptamer showed varying degrees of inhibition of the vascular permeability response (Fig. 6). Relative to the vascular leakage observed with VEGF alone, aptamer t22-OMe inhibited 37% of the response at 1 M and 13% at 0.1 M; however, only the data for the lower concentration of aptamer meets the p Ͻ 0.05 criterion for statistical significance. Aptamer t2-OMe showed no inhibition in this assay. t44-OMe inhibited the response by 58% at 1 M and 48% at 0.1 M. The sequence-scrambled control oligonucleotide, scr-t44-OMe, showed no significant inhibitory activity. Aptamer t44-OMe is thus the most effective antagonist of VEGF-induced vascular permeability. The moderate degree of inhibition observed with t44-OMe was substantially enhanced by conjugating the aptamer to 40-kDa PEG. The addition of 40-kDa PEG at the 5Ј-end of t44-OMe resulted in a slight apparent reduction (ϳ4-fold) in binding affinity to VEGF (data not shown) but a marked enhancement in inhibitory activity in the Miles assay (Fig. 6). The 40-kDa PEG-t44-OMe conjugate inhibited 83% of VEGF-induced vascular permeability at 0.1 M, while a conjugate of 40-kDa PEG to the scr-t44-OMe control oligonucleotide showed no inhibition at the same concentration. DISCUSSION The choice of oligonucleotide library has an enormous impact on the outcome of a SELEX experiment. Without exception, selections performed with the same target but with nucleic acid libraries with different 2Ј-moieties have yielded families of aptamers with distinctly different sequences, (33,34,56,57). This may be explained, in part, by differences in the size and FIG. 4. Photo-cross-links of 5-I-Usubstituted aptamers to VEGF. 5Ј-End radiolabeled 5-I-U-substituted aptamers were incubated with VEGF and exposed to pulses of 308 nm monochromatic light. Cross-linked complexes were separated from free aptamer by electrophoresis in a denaturing polyacrylamide gel. In a, the specificity of the observed complex is demonstrated for ligand t22.23 (5-I-U14). A trace amount of radiolabeled aptamer was incubated in buffer alone or with 1 nM VEGF in the presence or absence of 10 nM unlabeled competitor oligonucleotide. Competitors were either the VEGF-specific aptamer, t22.23 (t22 in lane 3) or a control 2Ј-F-pyrimidine oligonucleotide not selected for binding to VEGF. The cross-linked complex is observed only when VEGF is present in the incubation mixture (compare lanes 1 and 2) and its formation is inhibited only in the presence of an oligonucleotide that can compete for the high affinity aptamer-binding site on VEGF (compare lanes 3 and 4). Photo-degradation products of the 5-I-Usubstituted aptamer are indicated. In b, a portion of the carboxyl-terminal sequence of VEGF 165 with the numbering of the amino acids in the mature, secreted protein is shown above with the amino acid that participates in cross-linking to three 5-I-U-substituted aptamers circled. The chymotryptic (upper bar) and tryptic (lower bar) fragments of VEGF that remain cross-linked to the aptamers after digestion are indicated below the amino acid sequence. The dashed line indicates the point at which the sequence data were no longer reliable. Below, schematic representations of the three aptamers are shown where the U that participates in cross-link formation is circled. The nucleotide position of the cross-linking U within the minimal aptamer sequence is indicated. Similarities in the consensus primary structures of aptamers t22.23 and t2.31 are boxed for emphasis. Nucleotides shown in white shadowed letters at the base of the stems in aptamers t2.31 and t44.29 are deleted in the minimal aptamer sequences shown in Table II. hydrogen bonding properties of various 2Ј-substituents. In addition, the 2Ј-substituent on ribose affects its propensity to adopt a 2Ј-endo or 3Ј-endo pucker (58) which, in turn, influences the tertiary structure of an oligonucleotide: the A-and B-form helices seen for RNA and DNA, respectively, provide a familiar example. For these reasons, it is not surprising that the VEGF aptamers derived previously from RNA (33) or 2Ј-NH 2 -pyrimidine RNA libraries (34) and those isolated from 2Ј-Fpyrimidine libraries, described here, are not obviously related.
Aptamers with more stable structures may be expected to have a lower entropic barrier to binding and may therefore have higher affinity for their targets than aptamers that are conformationally unstable in solution (39). In model duplexes, a single 2Ј-NH 2 substitution has been observed to cause profound destabilization of the helix (35,36), while 2Ј-F-or 2Ј-OMe-modified oligomers generally show enhanced helix stabilities (32,37,59). The VEGF aptamers derived from 2Ј-Fpyrimidine-modified RNA libraries described here were almost exclusively of very high affinity, with K d values in the range of 10 Ϫ11 M. Three of the aptamers could be truncated significantly to a minimal length ranging from 23 to 29 nucleotides. Despite their small size, the T m values for the three minimal aptamers, prior to modification of the 2Ј-OH-purines, ranged between 48 and 63°C. The affinities of the 2Ј-F-pyrimidine VEGF aptamers were indeed higher than most of the 2Ј-NH 2 -pyrimidine ligands isolated previously (34), and, while there are insuffi-cient data to make a systematic comparison, the three truncated 2Ј-F-pyrimidine aptamers showed substantially higher thermal stabilities than the 2Ј-NH 2 -pyrimidine-based aptamers (34). A comparison of 2Ј-F-and 2Ј-NH 2 -pyrimidine aptamers has also been made for ligands to keratinocyte growth factor (56), P-selectin, 4 and interferon-␥ (57). With some exceptions (57), the aptamers derived from the 2Ј-F-pyrimidine libraries displayed higher affinities for their target proteins. Together, these data support the notion that oligonucleotide libraries with intrinsically higher thermal stabilities may be more likely to yield aptamers with very high affinities for a molecular target. Relevant to this idea is the recent x-ray crystallographic evidence that affinity maturation of a hapten-binding antibody was accompanied with a reduction in conformational change in the antibody combining site upon antigen binding (60).
The VEGF 165 spliced transcript includes exon 7 which encodes a basic, carboxyl-terminal domain that mediates much of the heparin binding activity of the protein (61,62). VEGF 121 , which lacks the exon 7-encoded domain, does not bind to heparin, and is about 100-fold less potent than VEGF 165 in inducing a mitogenic response in endothelial cells (62,63). The three truncated, 2Ј-OMe-substituted oligonucleotides show no binding to VEGF 121 at up to 100 nM concentration. The aptamers described here are thus useful reagents for further delineating the role of VEGF 165 relative to VEGF 121 .
All three aptamers can form a cross-link to VEGF 165 at residue Cys 137 within the exon 7-encoded domain. Combined with the absence of binding to VEGF 121 , these data suggest that the basic domain encoded by exon 7 contributes key contacts to the binding site for each of the aptamers. Structural information for this region of the VEGF protein has not been obtained (55) but we hypothesize from the cross-linking data FIG. 5. VEGF receptor binding inhibition. 125 I-Labeled VEGF (0.5 ng/ml) was incubated in the absence or presence of increasing concentrations of aptamers or a control oligonucleotide with porcine aortic endothelial cells expressing either the Flt-1 (upper panel) or KDR (lower panel) human VEGF receptor. Counts/min of cell associated radioactivity is plotted against the concentration of competing oligonucleotide for each of the three aptamers described in the text (t22-OMe, t2-OMe, and t44-OMe) and for a control oligonucleotide (scr t44-OMe) that corresponds to a sequence-scrambled analog of ligand t44-OMe. The points are the average of three separate experiments; less than 10% deviation was observed in any value. Cell associated radioactivity in the absence of competitor was 1479 cpm for PAE/Flt-1 cells and 3431 cpm for PAE/KDR cells. Residual binding in the presence of 50 ng/ml unlabeled VEGF was 456 cpm for PAE-Flt-1 cells and 1236 cpm for PAE/ KDR cells.

FIG. 6. Inhibition of VEGF-induced vascular permeability.
Adult female guinea pigs were injected intravascularly with Evans Blue dye followed by intradermal injection of VEGF alone (20 nM) or VEGF co-mixed with aptamer at 1 or 0.1 M concentration. Leakage of the dye into the skin at the injection site as a result of VEGF-induced vascular permeability was quantitated from a digital image of the trans-illumininated skin. Data for each sample are expressed as a percentage of the signal obtained with VEGF alone. PBS alone provided a control for VEGF-independent optical density associated with the injection site. PEG oligonucleotides were conjugated to 40-kDa polyethylene glycol at the 5Ј-end. Asterisks indicate those samples where the average value was significantly different from 100% (p Ͻ 0.05, one-sample t tests). The standard error of three determinations is shown for all oligonucleotide containing samples or of six determinations for PBS alone.
that Cys 137 is solvent-accessible and that each of the aptamers binds to a similar site on the VEGF protein that is very near this residue. It is interesting to note in this context that Soker et al. (64) have found Gys 137 to be essential for the binding of exon 7-encoded peptides to the recently identified VEGF 165specific receptor (now also known as neuropilin-1 (65)). We assume that the site of cross-linking within each aptamer occurs at the position of 5-I-U substitution. For all three aptamers, the cross-linking site corresponds to a residue that is highly conserved among the members of each sequence family. As noted previously, Family 1 aptamers (represented by t22.23) and Family 2 aptamers (represented by t2.31) share a common conserved sequence pattern, 5Ј-GAAN (3)(4) UUGG-3Ј. The first U in this sequence corresponds to the site of cross-linking in both aptamers. This observation further supports the notion that the members of both sequence families share a related tertiary structure and interact similarly with VEGF.
Placenta growth factor (PlGF) is a recently described cytokine whose amino acid sequence shows homology to that of VEGF (53% identity) (40). Both proteins share a conserved pattern of eight cysteine residues with platelet-derived growth factor that defines a tertiary structure common to all three proteins (40,55). PlGF binds to Flt-1 and, like VEGF, elicits migration of monocytes and tissue factor expression in monocytes and endothelial cells in culture (66). However, unlike VEGF, PlGF is a very weak mitogen for endothelial cells (66). Consistent with the very high specificity that is generally displayed by SELEX-derived aptamers, no binding was seen to PlGF homodimers. The recombinant PlGF used in these experiments corresponds to the smaller, 129-amino acid isoform of the protein (67). It is possible that the larger isoform, PlGF 152 , that includes a short, highly basic, carboxyl-terminal domain, may bind with higher affinity to the aptamers.
Heterodimers between PlGF and VEGF have recently been isolated from the supernatants of tumor-derived cell lines and have been shown to be potent endothelial cell mitogens (68,69). We therefore tested whether the aptamers selected for binding to VEGF 165 display cross-reactivity with heterodimers of PlGF 129 and VEGF 165 . The affinities of the three aptamers were consistently lower for the heterodimer relative to VEGF 165 , with K d values of 790 pM, 40 nM and 430 pM for ligands t22-OMe, t2-OMe and t44-OMe, respectively. This corresponds to a reduction in affinity for the heterodimer compared with the VEGF 165 homodimer of 11-, 308-, and 9-fold, respectively. The fact that the homodimer of VEGF has two potential binding sites for an aptamer while the VEGF/PlGF heterodimer has only one can account for at most a 2-fold difference in the observed K d . Since all of the observed binding may be presumed to derive from contacts to the VEGF subunit in the heterodimer, these data suggest that the high affinity binding site for each of the aptamers involves contributions from both subunits in the VEGF homodimer. Alternatively, the VEGF subunit may adopt a slightly different conformation, or critical contact sites may be buried or blocked, in the heterodimer. Thus, despite the similarity in the binding sites of the aptamers suggested by the cross-linking data, each aptamer clearly forms a unique set of contacts with VEGF that affect its behavior in assays of heterodimer binding, inhibition of VEGF binding to its receptors and inhibition of VEGFinduced vascular permeability.
In addition to PlGF, other proteins with considerable homology to VEGF have been described, such as VEGF-B and VEGF-C (also called VEGF-related protein or VRP) (70 -72), and since the amino acid conservation extends to the eight cysteine residues involved in inter-and intra-subunit disulfide bonds, VEGF may form heterodimers with these or other mem-bers of this protein family. Indeed, VEGF and VEGF-B are capable of forming heterodimers (70). Furthermore, partial digestion of VEGF 165 by plasmin transiently generates a VEGF 165 / VEGF 110 heterodimer whose mitogenic potency is intermediate between the undigested and fully digested (VEGF 110 /VEGF 110 ) species (62). The ability of aptamers such as t44-OMe to bind with appreciable affinity to VEGF 165 (and, by inference, inhibit its activity) in various heterodimeric or partially digested forms may have important functional consequences.
Human umbilical vein endothelial cells express two major VEGF receptors: Flt-1 is expressed at a few thousand copies per cell and binds VEGF with higher affinity (K d ϭ 1-20 pM), while KDR shows a lower affinity for VEGF (K d ϭ 50 -770 pM) but is more abundant on the cell surface (16,52,53). Experiments using PAE cells transfected with the flt-1 or kdr receptor gene have suggested that KDR is the primary transducer in endothelial cells of VEGF-mediated signals related to changes in cell morphology and mitogenicity (41). More recently it was demonstrated that VEGF (or PlGF) signaling through Flt-1 induces cell migration in monocytes and secretion of tissue factor in both endothelial cells and monocytes (73,74). Sitedirected mutagenesis studies of VEGF 165 or of a VEGF  construct (which lacks the exon 7-encoded domain) have identified discrete regions in the protein where charged residue-toalanine mutations resulted in significantly reduced binding of the mutant protein to Flt-1 or KDR receptor fusion proteins (54,55). The recently described crystal structure of VEGF (8 -109) bound to domain 2 of Flt-1 has further delineated many of the contacts in the high affinity VEGF-VEGF receptor complex (75). The VEGF aptamers described here inhibit the binding of VEGF 165 to both receptors expressed on PAE cells. It is likely that the three aptamers make contacts with VEGF outside of the basic carboxyl-terminal domain, particularly in light of the fact that the oligonucleotides are nearly one-fifth as large as the homodimeric protein. Thus, regions of VEGF critical for binding to each of the receptors may be blocked by a bound aptamer. Furthermore, while the relative contribution of KDRand Flt-1-mediated signaling to angiogenesis in vivo requires further clarification, these data suggest that the minimal 2Ј-OMe-modified aptamers should potently inhibit activities induced by either signaling pathway.
Since the minimal 2Ј-OMe-modified aptamers were not easily differentiated on the basis of length, degree of 2Ј-OMe substitution, affinity for VEGF, kinetics of binding, or the receptor binding inhibition properties, we used the Miles assay to screen the three ligands for their relative capacity to inhibit the vascular permeability response induced by VEGF in vivo. One of the aptamers, t2-OMe, showed no activity in this assay relative to a control oligonucleotide. This aptamer was also somewhat less potent in inhibiting the binding of 125 I-labeled VEGF to receptor-expressing cell lines, in agreement with its somewhat lower affinity for VEGF. In contrast, a small decrease in permeability was observed with aptamer t22-OMe and significant inhibition was observed with t44-OMe. A VEGF concentration of 20 nM was required to obtain a reproducibly robust vascular permeability response in our hands. This exceeds by several orders of magnitude the concentration of VEGF used to monitor VEGF receptor binding and must account, in part, for the much higher concentration of aptamer required to inhibit VEGF-induced vascular permeability. While the relative inhibitory activity of the ligands roughly correlates with their affinities for VEGF, the differences in the K d values are small and this seems an unlikely explanation for their differing behavior in the Miles assay. The dissociation rates of all three aptamers are comparable and fairly rapid so that diffusion away from the injection site may limit opportunities for rebinding of the aptamers to VEGF. Consistent with this notion is the observation that conjugation of polyethylene glycol to one of the aptamers dramatically improves its inhibitory capacity, an effect that may derive from the relatively slower diffusion rate of the higher molecular weight conjugate.
Aptamers are chemically synthesized and can be readily derivatized with a wide variety of functional groups (e.g. PEG) to modulate their properties in vivo, including plasma residence time and biodistribution. We are currently testing the 2Ј-F-pyrimidine, 2Ј-OMe-purine-substituted aptamers conjugated to a variety of chemical moieties for inhibitory activity in in vivo models of angiogenesis.