Tenascin-C aptamers are generated using tumor cells and purified protein.

Tenascin-C (TN-C) is an extracellular matrix protein that is overexpressed during tissue remodeling processes, including tumor growth. To identify an aptamer for testing as a tumor-selective ligand, SELEX (systematic evolution of ligands by exponential enrichment) procedures were performed using both TN-C and TN-C-expressing U251 glioblastoma cells. The different selection techniques yielded TN-C aptamers that are related in sequence. In addition, a crossover procedure that switched from tumor cell to purified protein selections was effective in isolating two high-affinity TN-C aptamers. When targeting tumor cells in vitro, the observed propensity of naive oligonucleotide pools to evolve TN-C aptamers may be due to the abundance of this protein. In vivo, TN-C abundance may also be well suited for aptamer accumulation in the tumor milieu. A size-minimized and nuclease-stabilized aptamer, TTA1, binds to the fibrinogen-like domain of TN-C with an equilibrium dissociation constant (K(d)) of 5 x 10(-9) m. At 13 kDa, this aptamer is intermediate in size between peptides and single chain antibody fragments, both of which are superior to antibodies for tumor targeting because of their smaller size. TTA1 defines a new class of ligands that are intended for targeted delivery of radioisotopes or chemical agents to diseased tissues.

Tenascin-C is a very large (Ͼ1 ϫ 10 6 Da) hexameric glycoprotein that is located primarily in the extracellular matrix (ECM). 1 TN-C is expressed during fetal development, wound healing, tumor growth, atherosclerosis and psoriasis, suggesting a role for this protein in tissue remodeling processes (reviewed in Refs. 1 and 2; see also Refs. [3][4][5]. As judged by Western blotting and immunohistochemical staining (6 -16), TN-C levels in tumors are significantly higher than in normal tissue. Further, TN-C levels are predictive of local tumor recurrence and are correlated with invasiveness and distant metastasis (17)(18)(19), although these findings remain controversial. Tumor metastases can also express TN-C (10,20). In addition to localization in tumor stroma, TN-C can be associated with tumor vascular structures (21)(22)(23)(24) and may promote angiogenesis through interaction with the integrin ␣ v ␤ 3 (25). Because of the abundance of TN-C in tumor stroma and its association with angiogenesis, high-affinity TN-C ligands may be clinically useful tumor-targeting agents. In fact, radiolabeled antibodies to TN-C are currently being evaluated in glioblastoma patients (26,27) with significant responses to treatment in a phase II study (28).
Aptamers are typified by high affinity and specificity for their cognate proteins (reviewed in Refs. 29 -31) and can be considered as oligonucleotide analogs of antibodies. However, as nucleic acids, aptamers are fundamentally distinct from antibodies. In having small size (8)(9)(10)(11)(12)(13)(14)(15) relative to antibodies (150 kDa), aptamers are candidates for rapid tumor penetration and blood clearance. These are useful attributes for noninvasive diagnosis of disease (32) and may provide advantages over antibodies and fragments thereof, which demonstrate slower tissue penetration and clearance rates. To identify an aptamer for investigation of tumor-targeting and blood clearance properties, we describe herein a SELEX process to identify TN-C aptamers and then focus attention on a single aptamer, TTA1.
The SELEX process uses large (10 14 -10 15 sequences) oligonucleotide pools to identify binding species, i.e. aptamers, to a variety of purified molecular targets. In addition to generating aptamers against purified proteins/small molecules, SELEX technology can generate aptamers to cells (33) and tissues. 2 The advantages of complex targets include freedom from the need to define and purify a molecular target, and presentation of proteins in native folding and glycosylation states. For complex SELEX experiments, identifying optimal selection conditions is theoretically possible (34) but remains a challenging task. In contrast, selection against purified protein allows ready experimental manipulation to achieve optimal enrichment of high-affinity aptamers (35) and requires no deconvolution to identify the cognate protein (or lipid, oligosaccharide, nucleic acid, etc.). Relative to cells and tissues, purified proteins often exhibit lower nonspecific binding of nucleic acids, and therefore selections proceed more rapidly. Because each has advantages, we elected to use both purified protein and cells as target sources to obtain TN-C ligands.
A previous SELEX experiment targeting U251 glioblastoma cells identified a DNA aptamer that binds to tenascin-C, 3 demonstrating that TN-C is a selectable target on U251 cells. The ssDNA aptamer displays greatly reduced binding affinity at * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
§ To whom correspondence should be addressed: SomaLogic, 1775 38th St., Boulder, CO 80301. Tel.: 303-625-9039; Fax: 303-545-2525; E-mail: bhicke@somalogic.com. 1 The abbreviations used are: ECM, extracellular matrix; SELEX, systematic evolution of ligands by exponential enrichment; TN-C, tenascin-C; ssDNA, single-stranded DNA; SPR, surface plasmon resonance; RT-PCR, reverse transcription-PCR. physiological temperatures, 4 perhaps because these initial cell SELEX experiments were performed at 4°C. Thus the aptamer has relatively low affinity (K d ϳ100 nM) at 37°C and, being composed of DNA, is susceptible to nuclease activity in vivo. These features render the DNA aptamer unsuitable for in vivo applications. To identify aptamers for use as tumor-targeting agents, we undertook SELEX experiments at 37°C using a nuclease-stabilized 2Ј-F pyrimidine oligonucleotide library.

EXPERIMENTAL PROCEDURES
Cells, Proteins, and Oligonucleotides-U251 cells, derived from a human glioblastoma, were obtained from the National Cancer Institute-Frederick Cancer Research Facility Tumor Repository and cultured in RPMI 1640 (Life Technologies, Inc.) ϩ 10% fetal bovine serum (Summit, Ft. Collins, CO) to 90% confluency on tissue culture-treated polystyrene. Viability was checked using trypan blue staining. Human TN-C (Chemicon, Temecula, CA; other sources were inferior in purity and activity) was stored as frozen aliquots at Ϫ80°C. After thawing, preparations could be stored for at least 2 months at 4°C without loss of aptamer binding activity. Purified recombinant tenascin-C fragments (36) were obtained from Harold P. Erickson, Duke University. Human and bovine serum albumin (fraction V) were obtained from Sigma. DNA oligonucleotides were obtained from Operon (Alameda, CA), and aptamer synthesis was performed at NeXstar Pharmaceuticals as described (37). Specialty phosphoramidites were obtained from Glen Research (Sterling, VA; (CH 2 CH 2 O) 6 ϭ Spacer 18; hexylamine; 3ЈdT polystyrene support), JBL Laboratories (San Luis Obispo, CA; 2Ј-F C and 2Ј-F U), or Proligo (Boulder, CO; PAC-protected 2Ј-OCH 3 -G, rG, and rA). 5Ј-amine-containing oligonucleotides were conjugated to succinimidyl biotin (Pierce) in 30% dimethyl formamide, 200 mM sodium borate, pH 9.0, at 25°C for 15 min and purified on polyacrylamide gels. Alternatively, aptamer transcripts were 5Ј-biotinylated using a method described elsewhere. 5 In brief, transcription of DNA templates is initiated with a 5-fold excess of a modified GMP over GTP in order to efficiently place a unique biotin on the 5Ј end of the transcript. The modified GMP bears a hexyl amine moiety on the 5Ј position that has been conjugated to biotin using the succinimidyl chemistry described above.
SELEX Procedures-These procedures were generally performed as described (38). To prepare the initiating random library, doublestranded transcription templates were prepared by Klenow fragment extension of 40N7a ssDNA: 5Ј-TCGCGCGAGTCGTCTG(40N)CCGCA-TCGTCCTCCC-3Ј. The reverse complement of this sequence is the "sense" strand, representing the fixed sequences that span the random regions shown for individuals in Fig. 5. This was done using the 5N7 primer: 5Ј-TAATACGACTCACTATAGGGAGGACGATGCGG-3Ј, which contains the T7 polymerase promoter (underlined). The 32 P-body labeled library was prepared with T7 RNA polymerase; all transcription reactions were performed in the presence of 2Ј-F pyrimidine nucleotides and 2Ј-OH purine nucleotides. For cell selections, U251 cells were grown to 90% confluence on 150-mm-diameter tissue culture plates and washed three times with 10 ml of binding buffer (Dulbecco's phosphatebuffered saline with MgCl 2 and CaCl 2 (Life Technologies, Inc.) and 0.05% bovine serum albumin). 1500 pmol of 32 P body-labeled library was then incubated with the cells in 10 ml of binding buffer for 45 min with gentle shaking. Unbound oligonucleotides were removed using seven washes (10 min each) of 10-ml binding buffer, typically removing Ͼ99% of input radioactivity. A final 20 -40-min wash of 5 ml included 10 mM EDTA that caused U251 cells to detach. Cells were pelleted (5 min at 300 ϫ g) and the supernatant removed (the EDTA elution). Plates were then treated with 1 ml of Trizol (Life Technologies, Inc.), and the remaining cells/extracellular matrix were removed using a cell scraper. To form the final Trizol elution, pelleted cells from the EDTA elution were added to the Trizol extraction from the plate. After the first round of selection, the EDTA elution pool and Trizol elution pools were kept separate. In the Trizol arm, EDTA-sensitive aptamers were eluted and discarded before Trizol elution. All washes and incubations were at 37°C in binding buffer. For EDTA elutions, the sample was extracted three times with a 1:1 mixture of phenol/chloroform and once with chloroform. To recover additional radioactivity, organic phases were back-extracted with 100 l of 10 mM Tris, pH 7.5, and an additional volume of chloroform. Nucleic acid was then precipitated twice using 2 M NH 4 OH and 2.5 volumes of ethanol. For the Trizol elution, an additional 0.2 volume of chloroform was added to facilitate phase separa-tion. The aqueous phase was then extracted twice with phenol:chloroform and once with chloroform and then treated with 5-10 g of RNase A for 10 min at 37°C to degrade contaminating cellular RNA (the aptamers are resistant to RNase A by virtue of 2Ј-F-modified pyrimidines). Organic phases were back-extracted as described for the EDTA elution. To precipitate nucleic acids, 0.25 volume of 0.8 M sodium citrate, 1.2 M sodium chloride was added along with 0.25 volume of isopropanol. Reverse transcription (RT)-PCR and transcription were performed as described (38). Two synthetic primers, 5N7 (see above) and 3N7a (5Ј-TCGCGCGAGTCGTCTG-3Ј), were used for RT-PCR. To monitor aptamer pool complexity, renaturation rates were measured as described (39).
Plate SELEX Procedures-These procedures were performed as described (40). For each round, 96-well Lumino plates (Labsystems, Needham Heights, MA) were coated for 2 h at room temperature with 200 l of Dulbecco's phosphate-buffered saline containing tenascin-C. Control wells lacked tenascin-C in this initial coating step. After being coated, the wells were blocked using HBSMC buffer (20 mM HEPES, pH 7.4, 137 mM NaCl, 1 mM CaCl 2 , 1 mM MgCl 2 , and 1 g/liter human serum albumin) for rounds 1-5. For rounds 6 -8, wells were blocked with HBSMC buffer containing 1 g/liter casein (I-block; Tropix). This switch of blocking agent was performed to decrease background binding of aptamer pools to the plate surface. Binding and wash buffer consisted of HBSMC buffer containing 0.1% Tween 20. The aptamer pool was diluted into 100 l of binding buffer and incubated for 30 min at 37°C in the protein-coated wells. After binding, six washes of 200 l each were performed. The wells were then emptied and placed on top of a 95°C heat block for 5 min ("heat elution"). Standard avian myeloblastosis virus (AMV)-reverse transcriptase reactions (50 l) were performed at 48°C directly in the well, and the reaction products were utilized for standard PCR and transcription reactions. Cloning and sequencing used standard procedures.
Binding Assays-To measure aptamer pool binding, U251 cells were grown to confluence in 12-well tissue culture plates (Falcon 3047, Becton Dickinson). After the cells were washed with binding buffer, 32 P body-labeled aptamer pools were incubated with cells in binding buffer for 40 min at 37°C. Unbound radioactivity was removed by aspiration, and two 10-s washes were performed. Trizol was used to collect bound cpm, which were quantitated by liquid scintillation counting.
For SPR, aptamer pools were 5Ј-biotinylated (described above) and immobilized to a streptavidin-containing surface (SA chip, BIACORE 2000, Biacore AB, Uppsala, Sweden) at a level of ϳ1000 response units. Running buffer was HBSMC containing 0.005% Tween 20 (P20, Biacore AB). A reference flow cell for each experiment consisted of a random sequence oligonucleotide pool. Kinetic constants for TN-9 binding to TNfbg (the bacterially expressed fibrinogen-like domain) were determined using standard methods (41,42). Nitrocellulose filter partitioning assays were performed as described (43). Briefly, 32 P end-labeled oligonucleotides at 0.5 ϫ 10 Ϫ10 M were incubated with increasing concentrations of TN-C in HBSMC buffer ϩ 0.01% (w/v) human serum albumin at 37°C for 15 min. Reactions were then filtered over nitrocellulose, and bound cpm were quantitated. Data were fit to obtain binding constants as described (44).

RESULTS
To identify aptamers to TN-C, a tripartite SELEX experiment was carried out as diagrammed in Fig. 1. In the first arm, purified TN-C was used. The second arm consisted of selection against a TN-C-expressing glioblastoma cell line, U251. This arm was subdivided into EDTA and Trizol "elutions" to recover bound aptamers. A third arm was a crossover from cell selections onto purified protein selections.
A SELEX Experiment Using Purified TN-C-Human TN-C was adsorbed to polystyrene 96-well microtiter plates. To initiate selections, a random aptamer pool consisting of 10 14 oligonucleotides was generated using RNA polymerase. The oligonucleotides contained 2Ј-F pyrimidines and 2Ј-OH purines with a 40-nucleotide random sequence region flanked by fixed sequences for RT-PCR. Selections were performed according to Drolet et al. (40), essentially consisting of protein-oligonucleotide incubations, washes to remove unbound oligonucleotides, and RT-PCR amplification of the bound oligonucleotides. The amounts of protein with each well in a 96-well plate and the amount of input RNA are indicated in Table I. A qualitative assessment of PCR amplification indicated that background binding of the RNA pools to polystyrene without associated tenascin-C ("no protein" control) was increasing through the initial five rounds. At round 6, the blocking agent was switched from human serum albumin to casein, which resulted in dramatically decreased aptamer pool binding to the no protein control wells. Progress was quantitated by measuring the affinity of 32 P-labeled aptamer pools for TN-C using a nitrocellulose filter capture assay (45). After five rounds, a slight improvement in binding was evident. Coincident with the switch in blocking agent, the amount of TN-C binding in the aptamer pool rose dramatically in round 6. By round 8, affinity had increased at least 1000-fold to an equilibrium dissociation constant (K d ) of 3 ϫ 10 Ϫ9 M. As no further affinity improvement was evident in the subsequent round, selection was deemed complete at round 8.
A SELEX Experiment Using Tumor Cells-A second experiment used human U251 glioblastoma cells as the target source.
These cells construct an ECM containing abundant TN-C (46). Cells were grown to confluence and incubated with 10 14 sequences of a random oligonucleotide pool (identical to that described above) at 37°C for 1 h. After extensive washing, a final wash buffer containing 10 mM EDTA was applied to elute EDTA-sensitive aptamers. Because nucleic acid structures and nucleic acid-protein interactions often utilize divalent cations, it was expected that EDTA would elute a subset of cell-bound aptamers. The cells were solubilized, nucleic acids were extracted using Trizol™, a reagent that combines chaotropic denaturation of proteins with organic extraction of nucleic acids, and then the remaining aptamers were collected. Thus the EDTA served to elute a subset of bound aptamers, and the subsequent Trizol elution collected all remaining aptamers along with cellular RNAs. Aptamers from both EDTA and Trizol elutions were amplified by RT-PCR and transcribed, closing the first round of this SELEX experiment. Unlike the purified protein SELEX experiment, cell and input RNA concentrations remained constant throughout nine rounds of selection (Table I).
The progress of the cell selections was monitored by measuring the binding of radiolabeled aptamer pools to U251 cells. To analyze the EDTA elution SELEX experiment, Fig. 2A compares binding of a control aptamer pool to rounds 3, 5, and 9. The control aptamer pool bound the cells detectably, and binding was saturable. Relative to this nonspecific binding, rounds 3, 5, and 9 showed progressively increasing binding.
Similar to the EDTA elution pools, the Trizol pools showed increased binding compared with a random aptamer pool (Fig.  2B). The T9 (Trizol round 9) pool showed less apparent binding than the T5 pool. This was due to increased binding to the polystyrene surface (data not shown). This outcome suggests FIG. 1. Tripartite SELEX experiment for TN-C aptamers. A random sequence 2Ј-F pyrimidine RNA library was incubated with either purified tenascin-C or tenascin-C-expressing U251 cells. The iterative SELEX procedure was carried out to identify aptamers. Protein and two-tumor cell SELEX procedures were kept separate, as defined by the dotted line. The protein SELEX procedure was carried out by hydrophobic adsorption of TN-C to polystyrene 96-well plates, whereas the cell-based procedure used adherent U251 glioblastoma cells. Two methods of eluting bound aptamers from cells were used, EDTA and Trizol. A crossover SELEX experiment was also performed, symbolized by the horizontal lines crossing from Cell SELEX to Protein SELEX. Here, the ninth round cell SELEX aptamer pools were selected for two rounds against purified tenascin-C. The total numbers of rounds performed are indicated on the left. RNA and protein input For the purified protein selections, protein input into each well represents the quantity of protein incubated with each well for adsorption, which was then incubated in buffer with the indicated quantity of RNA. Decreases in protein and RNA input occurred as the pool affinities improved. For the tumor cell SELEX experiment, U251 glioblastoma cells were grown to confluence in tissue culture plates for each round. that Trizol-eluted aptamers bound to the polystyrene surface, directing selective pressure away from cell binding and toward polystyrene binding. The cell binding analysis demonstrated pool evolution toward U251 binding. However, this analysis did not fully evaluate the progression of the tumor cell SELEX experiment; this is because a pool of low-affinity ligands for an abundant protein would show higher cell binding than a pool of high-affinity ligands for a rare protein. For many applications, the latter pool is desirable. Therefore we employed another measure of progression, aptamer pool complexity, which can be estimated by measuring nucleic acid renaturation rates (C o t analysis) (39). Decreasing pool complexity serves as a proxy for convergence upon a high-affinity solution. The C o t analysis predicted that the E9 pool would contain ϳ100 different oligonucleotide sequences, whereas the T9 pool would contained ϳ50,000 sequences (data not shown). Taken together, the cell binding and C o t analyses indicated that EDTA was more effective than Trizol in driving pool convergence toward cell binding.
To determine whether the U251 aptamer pools contain TN-C aptamers, binding was investigated using a surface plasmon resonance (SPR) assay. Aptamer pools were biotinylated at the 5Ј terminus and immobilized, via streptavidin, onto the surface of a biosensor chip. TN-C binding was then measured by SPR. Specific binding of TN-C to aptamer pools P8 and E9 was evident, with P8 showing significantly higher binding (Fig. 3). A faint signal was detected in the T9 pool; a low signal was also detected using the filter binding assay described below. Quantitative measurements were not possible because of the large mass of TN-C (Ͼ 1 mDa), which slows diffusion through the surface matrix. A further hindrance to quantitation is the hexameric structure of TN-C, which likely causes slow dissociation from the surface because of multivalent interactions. In addition, it was not possible to couple active TN-C to the matrix. The SPR assay using soluble TN-C therefore served a qualitative role; we observed low binding to the tumor cell Trizol arm, increased binding to the tumor cell EDTA arm, and the highest binding to the purified protein arm.
A Crossover SELEX Experiment Using Tumor Cells and Purified TN-C-To enrich TN-C aptamer representation in the tumor cell aptamer pools, a crossover SELEX experiment was performed as diagrammed in Fig. 1. The two cell-selected pools were subjected to two rounds of protein selection (Table I), generating pools E9P2 (E9P2 ϭ 9 rounds of EDTA elution from cells and 2 rounds of protein selection) and T9P2. Affinities for TN-C were then determined using a nitrocellulose filter binding assay (Fig. 4). For comparison, the P8 pool was included. This analysis indicated that two rounds of crossover selection on TN-C improved the affinity of E9 by 50-fold. For the Trizol arm, affinity rose from undetectable to 2 ϫ 10 Ϫ9 M in two rounds. Remarkably, just two rounds of crossover selection were required to enrich the high-affinity aptamers that were rare in each tumor cell aptamer pool.
Isolation and Sequencing of High-affinity TN-C Ligands-To analyze the content of selected pools, five aptamer pools were cloned and sequenced. Aptamers could be grouped into the three families shown in Fig. 5. Family I members were found in the P8, E9, and T9 pools. These sequences are related through the consensus sequence GACNYUUCN 1-3 GCAYC and have affinities for TN-C ranging from 20 to 100 ϫ 10 Ϫ9 M. The T9 pool contains many different family I sequences, consistent with the high sequence complexity predicted by C o t analysis (data not shown). Family II members are related through the consensus sequence CGU(C)GCC(G)A. Consistent with their overrepresentation in the P8 and E9P2 pools, family II aptamers have the highest affinities for TN-C. Although family II aptamers from the crossover SELEX procedure (E9P2 clones) are highly related to family II aptamers obtained using purified protein (TN clones), they arose from distinct sequence lineages. This indicates that cell-derived aptamers did not arise from cross-contamination during the selections. Family III sequences are less common but can also have high affinity for TN-C. Thus, two distinct target sources, cells and protein, identified TN-C ligands that are sequence-related.
Trends in Selection for High Affinity-When aptamer pool affinities and family representations were tabulated, trends in TN-C aptamer selection became evident. For example, lower affinity pools (P6, E9, T9) contain more family I members than family II members and vice versa for the higher affinity pools (P8 and E9P2, Table II). This trend is observed for both cell and protein SELEX experiments. Family I and unrelated sequences ("Others") predominate in early, low-affinity populations, whereas family II aptamers emerge later and have higher affinity for TN-C. As discussed elsewhere (34,35), these trends in selection could be due to higher prevalence of family I aptamers in the random sequence starting pool.
Returning to individual clones, a single family I clone, E9-3, dominates the E9 pool at 24% of the sequences identified (Fig.  5). This overrepresentation indicates that E9 pool had converged upon a small number of sequences. A family II aptamer, TN-9, represented 22% of the P8 pool, whereas family II aptamers E9P2-1 and E9P2-2 together represented 56% of the E9P2 pool. Thus, family II aptamers dominated the advanced pools and displayed high affinity for TN-C with K d s of 1-10 ϫ 10 Ϫ9 M.
Two Aptamer Epitopes on TN-C-TN-C is a multidomain protein. It contains epidermal growth factor-like repeats, fibronectin type III repeats, and a fibrinogen-related globular domain at the C terminus (47). To determine the dominant epitope, if any, for the aptamer pools, bacterially expressed TN-C domains (36) were tested for binding. Transcribed aptamer pools were 5Ј-biotinylated and immobilized onto a streptavidin surface. Binding of TN-C and recombinant domains was then detected by SPR (Fig. 6). An existing ssDNA aptamer, GB41, 3 bound to full-length TN-C as did pool P8 (Fig.  6A). Slow dissociation from the surface was observed, most likely because of the large mass and multivalency of TN-C. We then examined three different fragments of TN-C (36) for aptamer and aptamer pool binding. TNfnA-D comprises a series of fibronectin type III repeats; no binding was observed to TNfnA-D (Fig. 6B). TNfn3-5, a different series of fibronectin repeats, bound to ssDNA aptamer GB41 but not to the P8 pool we describe here. TNfn3-5 dissociates very rapidly from GB41 at 37°C (Fig. 6C), indicating poor binding at physiologic temperature. In contrast to the ssDNA aptamer, the P8 pool bound to the C-terminal fibrinogen-like domain, TNfbg (Fig. 6D). Individual aptamers from families I, II, and III each bound to immobilized TNfbg in the SPR assay (data not shown). Association and dissociation rate constants of aptamer TN-9 for TNfbg were 10 5 M Ϫ1 s Ϫ1 and 10 Ϫ3 s Ϫ1 , respectively. However, the k on and k off are considered estimations, as the SPR experiments were not designed to ensure rigorous measure of these rate constants. The K d derived from these estimates, at 10 ϫ 10 Ϫ9 M, is 10-fold greater than the K d that was measured using a filter binding assay. In summary, the bacterially expressed domains permit identification of aptamer binding regions on TN-C. To date there are two aptamer epitopes, one within the type III fibronectin repeats and one within the C-terminal fibrinogenlike globule.
Aptamer Binding Specificity and Species Cross-reactivity-We next examined the cross-reactivity of a human TN-C aptamer toward mouse TN-C. Because mouse TN-C was not available in purified form, a modified enzyme-linked immunosorbent assay was developed in which TN-C was captured onto a surface and incubated with an aptamer. First, antitenascin antibodies were immobilized onto polystyrene 96-well microtiter plates. The wells were then incubated with the tissue culture supernatant of mouse (3t12) or human (U251) cell lines, capturing TN-C onto the surface. Capture was confirmed using a second monoclonal antibody. Aptamer TTA1 (described below), a truncated version of aptamer TN-9, was chosen to test species cross-reactivity. TTA1 was 5Ј-biotinylated and incubated with captured TN-C at increasing concentrations. After rapid washing, bound TTA1 was quantified. Fig. 7 indicates that half-maximal binding to human TN-C occurred at ϳ20 ϫ 10 Ϫ9 M, similar to the measured K d of 5 ϫ 10 Ϫ9 M. In contrast, half-maximal binding to mouse TN-C occurred at ϳ400 ϫ 10 Ϫ9 M, a 20-fold reduction relative to the human protein. Binding of the control aptamer, TTA1.NB, to the captured TN-C was significantly decreased relative to the binding of TTA1 (Fig. 7). These data indicate that, despite 93% sequence identity between the human and mouse TNfbg domains, aptamer TTA1 has high specificity for human TN-C. Further tests of specificity were conducted by incubating the biotinylated aptamer with adherent cells that express either human or mouse TN-C. We again observed greatly diminished binding of TTA1 to cells expressing mouse TN-C (data not shown). In this case, the cells have developed an ECM, indicating the specificity of TTA1 for a single protein within the native ECM. In general, aptamers selected for binding to a particular protein exhibit low crossreactivity toward unrelated proteins and even have low crossreactivity toward highly related proteins. For example, a Pselectin aptamer displays 10,000 -100,000-fold selectivity for Pversus L-and E-selectin (48).
A final measure of specificity indicates that TTA1 does not bind appreciably to a wide range of extracellular matrix proteins. The radiolabeled aptamer binds to tumor tissue that expresses TN-C but does not bind to tissue that lacks TN-C or to tumors that do not express human TN-C. 6 Aptamer TTA1: Size Minimization and Further Nuclease Stabilization-To prepare an aptamer for in vivo experimentation, we made three alterations: size minimization; further stabilization against nuclease activity; and incorporation of a bioconjugation handle for addition of biotin, fluorescent tags, 6 B. Hicke, unpublished observations. FIG. 4. Binding of aptamer pools to Tenascin-C: filter partitioning assay. Aptamer pool binding was measured using a binding assay (45) in which protein-bound 32 P-RNA are partitioned from unbound 32 P-RNA by filtration through nitrocellulose. Tenascin-C binding of cell-selected pools (X, random aptamer pool; open circles, Trizol round 9 (T9); filled circles, E9) was compared with tenascin-C binding after two rounds of crossover selections onto purified protein (squares with cross, Trizol R9/Protein R2 (T9P2); cross, E9P2) or eight rounds of purified protein selections (diamond, P8). The E9/P2, T9/P2, and P8 data sets were fit using a biphasic binding equation, and the E9 pool was fit using a monophasic binding equation (44). Pool K d s reported in Table II refer to the high-affinity component of these binding fits. radiometal chelators, etc. Ideally, such changes maintain the affinity and specificity of the aptamer.
A cloned aptamer sequence is typically 70 -80 nucleotides in length. For efficient chemical synthesis, it is desirable to identify the minimal high-affinity aptamer sequence. This can be initiated by determining the maximum permissible truncations at the 5Ј and 3Ј termini. Using described techniques (49) on aptamer TN-9, we found that no nucleotides could be removed from the 5Ј terminus, and 16 nucleotides could be removed from the 3Ј terminus. This exercise produced aptamer TN-9.4, which has K d ϭ 2 ϫ 10 Ϫ9 M for TN-C (Fig. 8).
To identify extraneous nucleotides that reside within the 3Ј and 5Ј termini required for high-affinity binding to TN-C (e.g. internal loops that do not contribute to the protein-oligonucleotide binding interaction), an RNA secondary structure prediction algorithm (50, 51) 7 was utilized. TN-9.4 and its analogous family II sequences, TN-7.4 and TN-21.4, were each subjected to the algorithm. A predicted structure common to all three aptamers was a three-way junction that places the conserved CGUCGCC element at the center of the junction. Of the predicted stems, the distal portion of the second stem did not appear conserved in sequence or length, suggesting that it is dispensable for high affinity binding to TN-C. By chemically synthesizing a series of deletions in this stem, we found that 17 nucleotides could be replaced by a non-nucleotide spacer, (CH 2 CH 2 O) 6 , with no affinity loss (TN-9.6, Fig. 8). Although consistent with the three-way junction predicted by the algorithm, this internal deletion does not rule out other potential structures. Together, the truncation and internal deletion analyses enabled trimming of a 55-nucleotide sequence to 39 nucleotides and a small spacer. Such a size reduction is critical for efficient chemical synthesis of the aptamer. As shown in Fig.  8B, the size-minimized species has K d ϭ 5 ϫ 10 Ϫ9 M, a 5-fold loss in affinity relative to the full-length aptamer.
Pyrimidine positions, as opposed to purines, are the primary source of nucleolytic instability in plasma. Therefore the pyrimidines are protected from nuclease activity by 2Ј-F groups incorporated during the SELEX procedure. Further stabilization can be achieved by converting purines to 2Ј-OCH 3 purines (52). This occurs after selections, because the addition of 2Ј-OCH 3 -modified purines to the existing SELEX procedure causes inefficient transcription. To identify purines that could be converted to 2Ј-OCH 3 without loss of function, aptamer TN-9.4 was divided into five sectors based on the putative three-stem junction structure (Fig. 9). In general, sectors were chosen so that 2Ј-OCH 3 substitutions occurred on one strand of a putative helix, to reduce any helical distortion caused by the substitutions. All purines in each sector were synthetically 7 Found on the Web at bioinfo.math.rpi.edu/ϳzukerm/.

FIG. 5. Sequences and affinities of YN-C aptamers.
Aptamers from three SELEX procedures (protein, cell, cell/protein crossover) were grouped into families based on sequence similarity. For simplicity, the fixed sequences at the 5Ј and 3Ј ends have been omitted. Affinities for TN-C were measured by a nitrocellulose filter binding assay. % of pool, aptamer representation in the pool from which it was cloned; P8, purified protein SELEX procedure eighth round; E9, U251 A EDTA elution ninth round; T9, Trizol elution ninth round.

TABLE II Family frequencies in aptamer pools
Five sequence sets are tabulated: selection using protein (P6 ϭ protein sixth round, etc; P8), U251 EDTA elution (E9) and Trizol elution (T9) arms, and cells/protein crossover (E9/P2; E9 ϩ 2 rounds protein). By visual inspection, sequences were grouped into one of four families or were considered as unrelated ("Others"). Elution, method used for eluting bound aptamer during the SELEX protocol. "K d for TN-C" refers to the affinity of the aptamer pool. substituted with 2Ј-OCH 3 purines, and affinity for TN-C was measured (Table III). In sectors 2 and 4, binding affinity decreased 20-fold and Ͼ1000-fold, respectively, because of the 2Ј-OCH 3 substitutions. We then identified which of the nine purines in sectors 2 and 4 were responsible for affinity loss upon 2Ј-OCH 3 substitution; in the context of complete substitution in sectors 1, 3, and 5, individual purines were substituted and the aptamers tested for affinity (Table III). In so doing, we found that substitution at four of the nine Gs caused loss of affinity: G 9 , G 28 , G 31 , and G 34 . Finally, the size minimization and 2Ј-OCH 3 -substitution data were combined to synthesize aptamer TTA1 (Fig. 10), a 39-mer. A nonbinding control aptamer, TTA1.NB, was also synthesized. TTA1.NB has a 5-nucleotide deletion; it does not bind TN-C at concentrations up to 1 ϫ 10 Ϫ6 M (data not shown). TTA1 bears a 5Ј amine for bioconjugations and a 3Ј-3Ј cap for exonuclease protection. Pyrimidines contain 2Ј-F modifications, and purines contain 2Ј-OCH 3 modifications (except the four obligatory 2Ј-OH Gs and a fifth purine, G 1 , that also remained 2Ј-OH). In summary, TTA1 is a nuclease-stabilized 39-nucleotide aptamer that binds human TN-C tightly, with a K d of 5 ϫ 10 Ϫ9 M.

DISCUSSION
In this work, our goal was to identify a physiologically active, nuclease-stabilized aptamer that could be tested for tumor targeting capability in vivo. A previous tumor cell SELEX experiment yielded a ssDNA aptamer that binds to a major extracellular matrix component, tenascin-C. However, selection at 4°C resulted in an aptamer that binds moderately well at 4°C (K d ϳ 100 ϫ 10 Ϫ9 M) 6 but poorly at physiological temperatures (K d Ͼ 1 ϫ 10 Ϫ6 M). In addition, DNA is not sufficiently stabilized against nuclease activity to which bloodborne nucleic acids are exposed. Nevertheless, the cell selections had identified a protein that is of significant interest as a marker for tissue remodeling processes including tumor formation. We therefore performed new cell selections and protein selections using a 2Ј-F pyrimidine library at 37°C.
In the tumor cell SELEX experiment, EDTA elution was superior to Trizol elution for enrichment of cell-binding aptamers. Because nucleic acid structures and nucleic acid-protein interactions often require divalent cations, we reasoned that low EDTA concentrations could selectively elute a subset of cell-or ECM-bound aptamers. In contrast, the chaotropic agent contained in Trizol removes oligonucleotides indiscriminately from cells, ECM, polystyrene, etc. We found that Trizol elution enriched a population of polystyrene binding sequences that hindered the enrichment of cell-binding aptamers. C o t analysis predicted far higher sequence complexity in the ninth round Trizol pool than in the ninth round EDTA pool. This prediction was borne out by sequence analysis that showed that the E9 pool had repeated sequences, whereas the Trizol pool had no duplicates in 50 analyzed clones. Thus, EDTA elution was superior to Trizol in reducing background binding to polystyrene and in creating an aptamer pool that had converged on a set of cell-binding aptamers.
In the tumor cell SELEX experiment, TN-C was a dominant protein for aptamer selection. There are at least two explanations for this result. First, aptamers bound to membrane targets are candidates for internalization and subsequent degradation by nucleases, and would not be recovered during selections. A second more likely explanation for preferential selection of TN-C aptamers is target abundance. TN-C is present at high (1-10 M) concentrations in the ECM (16,46). 8 In a random oligonucleotide pool, any single aptamer is present at a vanishingly small concentration (ϳ10 -20 M). Therefore, concentration of the aptamer target determines the rate and extent of aptamer binding in initial rounds of selection, favoring abundant proteins as targets. The TN-C M concentration in the ECM is sufficient to drive the binding of single-copy aptamers in an initial round of the SELEX process and is perhaps sufficient to serve as a dominant target over other less abundant proteins. Furthermore, extended washes were used to remove unbound oligonucleotides. High-affinity aptamers, including the TN-C aptamer described here, typically have k off values of ϳ 10 Ϫ3 s Ϫ1 (44,48,53), or t1 ⁄2 values of 2-10 min. After 50 min of washing, ϳ97% of an aptamer with t1 ⁄2 ϭ 10 min would be washed away unless the target protein is at sufficiently high concentration for rebinding to occur. Therefore, two factors, extensive washing and the rarity of individual aptamers in a random pool, may each bias cell selections toward abundant proteins. These factors may account for the propensity of two different U251 cell SELEX experiments to isolate tenascin-C aptamers.
For the identification of tumor-targeting aptamers, tumor cells have strengths and weakness relative to purified protein.
Using cells enables the presentation of epitopes in their native state and also requires no knowledge of a potential target. The cell SELEX procedure can lead to the identification of new target proteins or the new appreciation of a known protein as an aptamer target. Indeed, this occurred with the initial U251 cell selections that identified an ssDNA aptamer against TN-C, as well as the identification of a new 42-kDa trypanosome protein (54). A disadvantage of cell selections is that slow convergence of cell SELEX pools is caused by higher background binding of nucleic acids to cells (55) than to purified proteins. Furthermore, selection for high affinity may be compromised because concentration of the target protein(s) is unknown. This is illustrated by the experiments in which TN-C ligands dominated the cell selections at round 9, but pool affinity was poor at K d ϭ 100 ϫ 10 Ϫ9 M. This pool was then exposed to purified protein in the crossover procedure. By applying selection pressure for affinity, the K d for TN-C improved 50-fold in two rounds and isolated two dominant sequences. Therefore high-affinity aptamers were present, but not evident, in the E9 tumor cell pool. By extension, the E9 pool may contain a variety of low abundance/high-affinity aptamers to additional tumor cell proteins of interest. To summarize, cell selections may guide new target identification in unbiased fashion and can be complemented by crossover procedures (or de novo SELEX experiments) to select for high-affinity binding to a protein of interest.
The repeated emergence of TN-C as a target suggests that abundant ECM proteins may be generally well suited as targets for cell-based aptamer selections. It is possible that abundant ECM proteins may also be well suited for aptamer-based tissue targeting in vivo. However, this hypothesis is currently difficult to test; a key difference between in vitro and in vivo selection is that after intravenous injection, oligonucleotides must run a gauntlet of nucleases and blood clearance mechanisms before gaining access to a target tissue, decreasing library complexity significantly. Further experiments using in vivo aptamer selection are needed to address these issues. A truncated aptamer that can be nuclease-stabilized and modified at will has considerable advantages in vivo over the aptamer pools typically used for selections. Therefore aptamers designed for in vivo tissue targeting can be readily derived using protein-and cell-based selection techniques, with appropriate post-selection modifications. As tissue-and organismbased SELEX procedures mature, we suggest that abundant ECM proteins may be useful surrogates for identifying tissuetargeting aptamers.
To characterize the binding and epitopes of family I, II, and III tenascin-C aptamers, we used an SPR assay. Each ligand family binds to the C-terminal fibrinogen-like domain, TNfbg. Although experimental conditions were not optimized to determine the kinetics of aptamer binding (41), the k on and k off for aptamer TN-9 were ϳ10 5 M Ϫ1 s Ϫ1 and 10 Ϫ3 s Ϫ1 , respectively. This corresponds to a K d of 1 ϫ 10 Ϫ8 M as compared with our equilibrium measurement of K d ϭ 2 ϫ 10 Ϫ9 M using a nitrocellulose filter partitioning assay. TNfbg is one of two basic domains of an otherwise acidic protein that has an overall pI predicted to be Ͻ5. Notably, a previous U251 cell SELEX experiment 3 isolated a TN-C aptamer that binds the other basic region, which consists of fibronectin type III repeats 3-5. The two SELEX experiments differed in selection temperature (37 versus 4°C) and oligonucleotide library (2Ј-F pyrimidine/ 2Ј-OH purine versus 2Ј-H), but it is not clear how these differences caused the current SELEX experiments to preferentially identify aptamers for the fibrinogen-like domain. It is known that distinct aptamers can bind different epitopes on a singledomain protein (56), and these data extend the observation by 8 C. K. Lynott, data not shown. FIG. 7. Aptamer binding to human and mouse tenascin-C. Antibodies to human TN-C (HxBO6) and mouse tenascin-C (mTN12) were immobilized to the surface of a 96-well plate and then incubated with medium from tenascin-C expressing human cells (U251 glioblastoma) or mouse cells (3t12 fibroblast). Binding or nonbinding aptamer TTA1 was 5Ј-biotinylated and incubated with the wells at increasing concentrations. The wells were washed, and bound aptamer was quantified using a streptavidin-alkaline phosphatase conjugate with chemiluminescent detection. Values for the nonbinding aptamer (open circles, U251 cell supernatant; open squares, 3t12 cell supernatant) were subtracted from values for the binding aptamer, represented as specifically bound aptamer (filled circles, U251 cell supernatant; filled squares, 3t12 cell supernatant).
indicating that conditions can also alter the dominant aptamerbinding domain on a large, multidomain protein.
Because reduced size may lead to increased tissue penetration rates (32) and will lead to more efficient chemical synthesis, we focused on size reduction of TN-9. Notably, an internal deletion was identified with the aid of an RNA structure prediction algorithm. This algorithm predicted that TN-9.4 could form a three-stem junction, among other possibilities, as a secondary structure. In comparing four family II aptamers, the algorithm revealed a potentially variant stem. We found that a deletion within this stem removed 17 nucleotides but left highaffinity binding intact. The resulting aptamer, TN-9.6, has a 5-fold reduced affinity relative to the full-length aptamer. Importantly, TN-9.6 is 17 nucleotides, or ϳ6 kDa, smaller than TN-9.4. The reduced size of TN-9.6 could increase tissue penetration rates and cause it to clear from the blood more rapidly than TN-9.4, which is an advantage for in vivo imaging.
To further stabilize the aptamer against nuclease activity, purines were substituted with 2Ј-OCH 3 groups. A combinato-rial approach for identifying purine positions that tolerate 2Ј-OCH 3 substitution has been described (49). In that study (49), a library of partially 2-OCH 3 -substituted aptamer molecules was synthesized. A selection experiment then separated binding from nonbinding species. The 2Ј-OCH 3 substitution pattern of the binding species was then identified by base hydrolysis, indicating purines that tolerate the substitution with retention of aptamer function. In our present work, the aptamer was sectored into five parts, and all purines in each sector were substituted, followed by affinity analysis. We found affinity reductions in two of five sectors and analyzed the contribution of individual nucleotides to affinity within these two sectors. In the end, 15 of the 19 purines could be substituted with 2Ј-OCH 3 without significant effect on affinity. Compared with the combinatorial 2Ј-OCH 3 substitution technique, a disadvantage of this sectoring approach is in the quantity of oligonucleotides that must be synthesized. The advantages are found in 1) obtaining direct binding data on a substituted species, rather than inferring effects on affinity from the selection experiment; and in 2) simpler oligonucleotide syntheses and binding experiments. Neither method requires knowledge of aptamer structure. After size reduction, 2Ј-OCH 3 substitution, 3Ј capping, and incorporation of a 5Ј amine, the synthetic aptamer TTA1 has a K d ϭ 5 ϫ 10 Ϫ9 M for binding to human TN-C, which is only a 5-fold reduction in affinity from the parent aptamer.
Taken together, the expression pattern of TN-C, demonstrated binding to integrins, and demonstrated adhesive activities (1) suggest that TN-C may play an active role in tissue remodeling processes. Such processes include tumorigenesis, angiogenesis, atherosclerosis, and wound healing. In particular, recent data implicate the fibrinogen-like domain in binding to the integrin ␣ v ␤ 3 (25), a critical protein for angiogenesis. The aptamer described here binds tightly to the fibrinogen-like domain of TN-C and therefore has potential application in investigating the role of TN-C in tissue remodeling processes.
TTA1 is a size-minimized and nuclease-stabilized aptamer that binds with high affinity to tenascin-C, an abundant extracellular matrix protein that is overexpressed during tissue remodeling. The potential clinical advantages of aptamers for tissue targeting have been discussed (32). These advantages include high affinity and specificity, small size, amenity to chemical modification to alter biodistribution, and pharmacokinetics, and rapid tissue penetration. TTA1 can be conjugated to a variety of radioisotope chelators, fluorescent dyes, and FIG. 9. Sectoring of TN-9.4 for 2-OCH 3 purine substitution. For identification of purine positions that must remain 2Ј-OH for high affinity tenascin-C binding, aptamer TN-9.4 was arbitrarily divided into five sectors. Shown is a secondary structure predicted by an RNA folding algorithm. Sectors were chosen such that 2Ј-OCH 3 purine substitutions (sequences and affinities are shown in Table III) would occur on only one strand of a putative helix.

-purine-substituted species
To identify purines that cannot be substituted with 2Ј-OCH 3 purines, that aptamer was sectored into five areas as shown in Fig. 9. Aptamers were synthesized in which all purines in a sector were substituted with 2Ј-OCH 3 purines. Each aptamer was tested for tenascin-C affinity using a nitrocellulose filter partitioning assay. Having identified sectors showing affinity loss upon substitution, individual purines within the affected sectors were studied within the context of complete substitution in sectors 1, 3, and 5. The sum of these data was incorporated into a maximally 2Ј-OCH 3 -substituted aptamer (in the context of the sizeminimized 39-mer) to form TTA1, shown in Fig. 10 Fig. 9.
FIG. 10. Aptamer TTA1: modifications and putative secondary structure. All pyrimidines are 2Ј-F and all purines are 2Ј-OCH 3 except the Gs, marked by arrowheads, which are 2Ј-OH. (CH 2 CH 2 O) 6 is a phosphoramidite linker joining G 9 with C 27 . Nucleotides 10 -26 have been deleted. The oligonucleotide is synthesized with a 3Ј-3Ј linkage and a 5Ј primary amine incorporated by phosphoramidite coupling. TTA1.NB is a nonbinding sequence that is identical to TTA1 except that it contains a deletion in which CCCUG-3Ј-3Ј-T is replaced by 3Ј-3Ј-T. K d for tenascin-C ϭ 5 ϫ 10 Ϫ9 M. biologically active moieties. Thus modified, TTA1 can be tested for targeted delivery to the extracellular matrix of tumors and/or atherosclerotic lesions.