cox2002-automated_selection_of_aptamers_against_protein_targets_translated_in_vitro_from_gene_to_aptamer...

Automated selection of aptamers against proteintargets translated in vitro: from gene to aptamerJ. Colin Cox1, Andrew Hayhurst2, Jay Hesselberth3, Travis S. Bayer1, George Georgiou1,2

and Andrew D. Ellington1,3,*

1Institute for Cellular and Molecular Biology, 2Department of Chemical Engineering and 3Department of Chemistryand Biochemistry, University of Texas at Austin, Austin, TX 78712, USA

Received May 31, 2002; Revised July 18, 2002: Accepted August 8, 2002

ABSTRACT

Reagents for proteome research must of necessitybe generated by high throughput methods. Apta-mers are potentially useful as reagents to identifyand quantitate individual proteins, yet are currentlyproduced for the most part by manual selection pro-cedures. We have developed automated selectionmethods, but must still individually purify proteintargets. Therefore, we have attempted to select apta-mers against protein targets generated by in vitrotranscription and translation of individual genes. Inorder to speci®cally immobilize the protein targetsfor selection, they are also biotinylated in vitro. As aproof of this method, we have selected aptamersagainst translated human U1A, a component of thenuclear spliceosome. Selected sequences demon-strated exquisite mimicry of natural bindingsequences and structures. These results not onlyreveal a potential path to the high throughput gener-ation of aptamers, but also yield insights into theincredible speci®city of the U1A protein for itsnatural RNA ligands.

INTRODUCTION

The development of in vitro selection methods has potentiatedthe identi®cation of nucleic acid-binding species (aptamers)and catalysts (aptazymes) that are proving increasingly usefulin therapeutic and diagnostic applications (1±3). However, asinformation continues to accrue about the sequences oforganismal genomes and the composition of organismalproteomes, it will be necessary to increase the throughput ofaptamer selection methods in order to either quickly identifyaptamers against novel targets or to generate aptamers againstentire proteomes or metabolomes.

To this end, we have previously developed methods for theautomated selection of aptamers (4±6). We initially estab-lished a rudimentary set of chemistries and robotic manipu-lations that facilitated the basic execution of nucleic acidselection and ampli®cation (5). The original automatedsystem was used for the selection of aptamers againstisolated oligonucleotide or protein targets. For example, the

non-nucleic acid-binding protein lysozyme was immobilizedon streptavidin beads and anti-lysozyme aptamers that hadreasonable af®nities (Kd » 31 nM) for the cognate protein andthat inhibited enzymatic function were selected (6). The sameprocedures were applied in short order to a variety of otherproteins, including CYT-18 (a tyrosyl-tRNA synthetase fromthe mitochondria of the fungus Neurospora crassa), MEK1 (ahuman MAP kinase kinase) and Rho (a transcriptionaltermination protein from the archaebacterium Thermotogamaritima). Each of the aptamers generated by automatedselection procedures have demonstrated high speci®city andaf®nity for their targets (Kd values in the pico- to mid-nanomolar range) (4). The current automated selection systemcan yield aptamers against virtually any compatible targetwithin 3 days.

However, automated selection methods are still too lowthroughput to contemplate the acquisition of aptamers againstall of the protein targets in a proteome. One of the primarylimitations on the method is that target proteins must beindividually acquired, most frequently by puri®cation from anoverproducer strain. To avoid this bottleneck, we have nowattempted to generate aptamers against protein targets thathave been directly transcribed and translated on the roboticworkstation. Blackwell and Weintraub previously demon-strated that manual selections with double-stranded DNAlibraries could be carried out against protein targets translatedin vitro (7).

An initial selection was carried out against translated U1Aprotein, a component of the human spliceosome. The wild-type sequences that U1A binds to have been characterized ingreat detail (8±10) and the anti-U1A aptamers that weregenerated by automated selection methods mimic the wild-type sequences in every important detail. These results suggestthat automated selection methods may soon prove capable ofdeciphering speci®c interactions between multiple differentnucleic acid-binding proteins and cellular RNA sequences.

MATERIALS AND METHODS

Oligonucleotides, genes and plasmids

The N-terminal RNA-binding domain of U1A was synthe-sized from 15 oligonucleotides using a PCR-based assemblymethod (11). The sequence of U1A was derived fromGenBank accession no. M60784, modi®ed to include the

*To whom correspondence should be addressed. Tel: +1 512 471 6445; Fax: +1 512 471 7014; Email: [email protected]

ã 2002 Oxford University Press Nucleic Acids Research, 2002, Vol. 30 No. 20 e108

same mutations (Y31H and Q36R) that were present in theprotein crystallized by Oubridge et al. (12). The assembledgene was sub-cloned into pET28a, yielding a gene encoding afusion protein with an N-terminal 63 histidine tag. This wasplaced into an expression vector to create pJH-hisU1A. Thesequence of the ®nal construct was veri®ed.

The assembled U1A gene fragment was used to generate atemplate for in vitro transcription and translation (TnT)reactions. Speci®cally, `splice-overlap extension' (SOE) PCR(13) was used to attach sequences necessary to direct highlevel expression and quantitative biotinylation of U1A duringin vitro coupled transcription±translation reactions (Fig. 1).

The primers used during these experiments are described inTable 1.

The 5¢ region for SOE PCR is ~150 bp in length. The T7gene 10 promoter and translation initiation region (TIR) areincluded to enhance transcript stability and translation (14,15).Protein synthesis is then initiated with a 24 amino acid biotinacceptor peptide sequence (`biotag', MAGGLNDIFEAQKIE-WHEDTGGSS) (16). A megaprimer containing both the TIRand the biotag was generated using primer AHX10, primerAHX47 and the Escherichia coli biotinylation and expressionplasmid pDW363 (17). The 3¢ region is ~1050 bp in length.The birA gene is followed by the major transcription

Figure 1. Scheme for in vitro transcription, translation and biotinylation. Splice-overlap extension was used to assemble 5¢ and 3¢ fragments onto genes ofinterest. The 5¢ fragment contained a T7 promoter and biotin protein ligase (BPL) recognition sequence (`biotag'). The 3¢ fragment contained the gene forBPL (birA) and a T7 terminator. Transcription generated a dicistronic mRNA that when translated yielded a `biotagged' gene of interest and functional BPL.Free biotin was covalently attached to the gene of interest by BPL and the tagged protein was puri®ed from the reaction by capture with strepavidin.

e108 Nucleic Acids Research, 2002, Vol. 30 No. 20 PAGE 2 OF 14

terminator signal for T7 RNA polymerase to enhancetranscript stability (14). A megaprimer containing birA andthe transcription terminator was generated using primerAHX33, primer AHX29 and pDW363. The assembled U1Agene was PCR ampli®ed with primers (U1A.biotag andU1A.t0) that added overlap regions complementary to bothmegaprimers. The 3¢ assembly junction creates an opaltermination codon followed by a functional RBS for birA(17). SOE assembly was carried out with the two `rescue'primers AHX31 and AHX30.

Initial in vitro transcription, translation andbiotinylation experiments

The maltose-binding protein (MBP) gene with the N-terminalbiotin acceptor peptide from pDW363 (17) was PCR ampli®edusing primers AHX10 and AHX32. The ®nal productcontained a suitable ribosome-binding site for birA after thembp termination codon. Standard PCR and SOE wereperformed using YieldAce DNA polymerase (Stratagene, LaJolla, CA), which outperformed other polymerases in terms ofyield, especially with SOE. Primers were gel puri®ed to ensureef®cient SOE. In order to determine whether endogenousbiotin protein ligase (BPL) was suf®cient to biotinylate proteintargets, the mbp gene was separately merged with twodifferent birA variants (18). The ®rst construct was generatedusing primers AHX29 and AHX55 and encoded a short, ®veamino acid fragment of the birA gene (MKDNT) followed bythe T7 major terminator. The second construct was generatedusing primers AHX33 and AHX29 and encoded a full-lengthbirA gene. The ®nal SOE products were ampli®ed usingprimers AHX30 and AHX31. DNA was puri®ed using aQIAquick PCR puri®cation kit (Qiagen, Valencia, CA).

For the in vitro transcription and translation reaction, ~1 mgof template was introduced into 50 ml T7 PROTEINscript-PRO reactions (Ambion, Austin, TX) supplemented with10 mM D-biotin. Reactions were incubated at 37°C for 60 min.

A control reaction without template was carried out simul-taneously. Reactions were centrifuged (~13 000 g, 15 min,4°C). The supernatants were removed and designated `solublefraction' while the pellets were resuspended in 50 ml of waterand designated `insoluble fraction'. Samples (10 ml) werecombined with Laemmli sample buffer (10 ml), boiled andloaded onto 10% Laemmli gels (19). Rainbow molecularweight markers (Amersham Pharmacia Biotech, Piscataway,NJ) were employed as standards.

Following semi-dry transfer to Immobilon P PVDF(Millipore, Bedford, MA), the membranes were blockedovernight with ~25 ml of PBS containing 2% BSA (w/v). Onemembrane was probed with a 1:10 000 dilution (v/v) of rabbitanti-MBP serum (New England Biolabs, Beverly, MA)followed by a 1:30 000 dilution of anti-rabbit IgG±horseradish peroxidase (HRP) (Bio-Rad, Hercules, CA) todetect MBP. The other membrane was probed with a 1:100 000dilution of avidin-conjugated HRP (Sigma-Aldrich, St Louis,MO) to detect biotinylated products. Sera probes wereincubated with blots for 30±60 min. Blots were washed in100 ml of PBS + 0.1% Tween-20, four times for 10 min eachtime. Blots were further washed with 2 3 100 ml of PBS for10 min each. Development was for a few seconds using aSuperSignal West Pico chemiluminescence kit (Pierce,Rockford, IL) and signals were captured on X-ray ®lm.

Having established that co-translation of BPL was neces-sary to ensure ef®cient biotinylation of the MBP target, wethen applied the methodology to three other genes: chlor-amphenicol acetyltransferase (cat), neomycin phosphotrans-ferase II (neo) and b-lactamase (amp). The cat gene wasampli®ed using primers AHX51 and AHX52 with plasmidpHELP1 (available from the authors, derived frompACYC184) as template (20). The neo gene was ampli®edusing primers AHX49 and AHX50 with plasmid pIMS102(available from the authors, derived from pNEO) as template.The amp gene was ampli®ed using AHX53 and AHX54 with

Table 1. Primers listed in Materials and Methods

Priming sequences are shown 5¢®3¢.

PAGE 3 OF 14 Nucleic Acids Research, 2002, Vol. 30 No. 20 e108

plasmid pUC19 as template. SOE PCR was used to fuse eachof the genes to the T7 promoter, biotag, birA and terminator.The dicistronic templates were puri®ed and employed inin vitro transcription, translation and biotinylation reactions asabove.

Protein targets were assayed for biotinylation. The solubleportions of reactions were applied to P6 desalting chromatog-raphy columns (Bio-Rad) to remove free biotin and thencombined with 50 ml of glycerol to ensure precipitation-freestorage at ±20°C. A high-binding ELISA plate (Corning, NewYork, NY) was coated overnight at 4°C with 1 mg/mlNeutravidin (100 ml; Pierce) in 50 mM sodium bicarbonatebuffer, pH 9.6. The Neutravidin-coated plate and an uncoatedcontrol plate were blocked overnight in ~350 ml of PBS with2% BSA (block). The TnT reactions in glycerol (10 ml) wereserially diluted by factors of two in block and distributed ontoboth Neutravidin and control plates (100 ml ®nal volumes).After 1 h, plates were washed three times with ~350 ml of PBS+ 0.1% Tween-20 and then twice with plain PBS. Primaryantisera speci®c for the gene products were applied in 100 mlof block for 1 h. Anti-cat, anti-neo and anti-amp antisera(Eppendorf±5 prime, Boulder, CO) were used at 1:1000dilutions in 100 ml of PBS; anti-mbp antisera (New EnglandBiolabs) was used at a 1:10 000 dilution in 100 ml ofPBS. Following washing, anti-rabbit IgG±HRP (Bio-RadLaboratories) was applied at a 1:3000 dilution in 100 ml for1 h. Following washing, the plates were developed for 10 minwith 100 ml of o-phenylenediamine (Sigma-Aldrich) at0.4 mg/ml. o-Phenylenediamine was originally dissolved in50 mM phosphate citrate buffer pH 5.0, with 4 ml of 30% H2O2

added per 10 ml. Color development was stopped with 50 ml of2±3 M sulfuric acid and the OD was measured at 492 nm on anELISA plate reader (Bio-Tek, Winooski, VT). The blank platewas subtracted from the experimental plate to correct forbackground binding.

In vitro transcription and translation of U1A

The U1A±birA construct was transcribed and translated in vitrousing the RTS 100 E.coli HY kit (Roche Diagnostics, Basel,Switzerland). In order to generate suf®cient protein for theentire course of an automated selection experiment a quad-ruple sized reaction (200 ml) was carried out starting with 2 mgof DNA template. Biotin (14 mM ®nal concentration) was alsoadded at the start of the combined TnT reaction. The reactionwas incubated at 30°C for 6 h, followed by incubation at 4°Cfor ~4 h. Unincorporated biotin was removed by passing theentire reaction though a Bio-Spin 6 chromatography column(Bio-Rad). The puri®ed, biotinylated protein sample wasincubated with 2.4 mg of strepavidin-coated magneticDynabeads (Dynal, Oslo, Norway) in the presence of 13selection binding buffer (13 SBB = 20 mM Tris pH 7.5,100 mM KOAc, 5 mM MgCl2) for 15 min. Beads werethoroughly washed ®ve times in 500 ml of selection buffer toremove any non-biotinylated protein.

Protein expression and biotinylation were veri®ed byelectroblotting and western blot analysis using strepavidin-conjugated alkaline phosphatase as a reporter for biotinylatedproteins (Fig. 2). Aliquots of the TnT reaction (5 ml) wereplaced in 1 ml of b-mercaptoethanol (b-ME), 10 ml of waterand 4 ml of 53 SDS±PAGE loading dye (250 mM Tris±HClpH 6.8, 50% glycerol, 2.5% SDS, 142 mM b-ME, 0.05%

bromophenol blue) and heat-denatured by boiling for 5 min.Half of the denatured sample (10 ml) was loaded onto a 12%acrylamide (29:1 acrylamide:bisacrylamide) denaturingprotein gel on a MiniPROTEAN 3 gel rig (Bio-Rad). Thegel was run until the bromophenol blue dye reached thebottom (~30 min at 200 V). The gel was electroblotted ontoTrans-Blot nitrocellulose (Bio-Rad) in a Mini Trans-BlotElectrophoretic Transfer Cell (Bio-Rad) in electroblottingbuffer (10 mM NaHCO3, 3 mM NaCO3, 30% methanol,pH 9.9) for 90 min at 350 mA. The blot was blocked with 13Tris-buffered saline and Tween-20 (TBST) for 30 min at roomtemperature with gentle agitation. Next, the blot was incu-bated with 3 ml of alkaline phosphatase-conjugated strepavidin(Promega, Madison, WI) in fresh 13 TBST buffer (15 ml) for30 min with mild agitation. Unbound alkaline phosphatasewas removed with three washes with 13 TBST (15 ml, 5 mineach). Finally, the blot was brie¯y rinsed with deionized waterand placed in 15 ml of Western Blue Stabilized Substrate foralkaline phosphatase (Promega) for 1±5 min.

In vitro selection

Automated in vitro selection was carried out as previouslydescribed (4,6). In short, a RNA pool that contained 30randomized positions (N30) (4,6,21) was used as a startingpoint for selection. The RNA was incubated in 13 SBB withbiotinylated protein immobilized on 200 mg of streptavidinbeads (suf®cient to bind 4±8 3 1012 biotinylated antibodymolecules). RNA±protein complexes were ®ltered from freeRNA and weakly bound species were removed by washingwith 4 3 250 ml of 13 SBB. Bound RNA molecules wereeluted in water at 97°C for 3 min. RNA for additional roundsof selection was generated by reverse transcription, PCR andin vitro transcription (using primers 41.30 and 24.30). All ofthe reactions were carried out sequentially on a Beckman-Coulter (Fullerton, CA) Biomek 2000 automated laboratoryworkstation that had been specially modi®ed and programmedto interface with a PCR machine (MJ Research, Waltham,MA) and a ®ltration plate device (Millipore).

Figure 2. Detection of translated, biotinylated U1A. Expression of biotinyl-ated U1A was assayed by western blot analysis. Biotinylated protein sizestandards are in the ®rst lane. The second lane shows a no template control,while the third and fourth lanes show controls with gene products that werenot biotinylated. The ®fth lane contains an aliquot from an in vitro tran-scription and translation reaction with the biotagged U1A gene. The arrowindicates the translated, biotinylated U1A gene product. The ~20 kDa bandseen in all control and experimental lanes is thought to be biotin carboxy-carrier protein (BCCP), the single biotinylated protein in E.coli.


Eighteen rounds of selection were performed againstbiotinylated U1A. In the ®rst 12 rounds of selection 20 cyclesof PCR were carried out, while in the last six rounds thisnumber was decreased to 16 cycles to prevent overampli®ca-tion of the selected pool; these parameters had been empiric-ally determined during previous automated selectionexperiments (6). In the ®rst 12 rounds of selection the washbuffer was 13 SBB. In the last six rounds of selection thestringency of the selection was increased by progressivelyincreasing the monovalent salt concentration, as detailed inTable 2.

The progress of selection was monitored every six rounds(6, 12 and 18) by placing [a-32P]-radiolabeled ribo- anddeoxyribonucleotides in the ampli®cation reactions andresolving ampli®cation products by gel electrophoresis. Theautomated protocol included a provision to archive aliquots ofthe reverse transcription (10 ml, 10% of the total reaction) andin vitro transcription (10 ml, 10% of the total reaction)reactions. After the automated protocol had run its course, thealiquots (in standard stop dye) were run on 8% acrylamide±7 M urea (19:1 acrylamide:bisacrylamide) denaturing gels.Decade RNA Markers (Ambion) were used as size standards.The gels were visualized using a PhosphorImager SI(Amersham Pharmacia Biotech). In addition, single pointbinding assays were carried out with equimolar concentrationsof protein and RNA samples (50 nM), as described below (seealso Fig. 5).

High throughput sequencing

Aliquots (1 ml of a 50 ml archive) of RT±PCR reactions from agiven round of selection were further ampli®ed and thenligated into a thymidine-overhang vector (TA Cloning Kit;Invitrogen). Templates for sequencing reactions were gener-ated from individual colonies via 50 ml colony PCR reactions(22) with standard M13 sequencing primers ¯anking theinsertion site of the thymidine-overhang vector (primers M13forward and M13 reverse). The PCR products were puri®ed

away from primers, salt and enzyme using a MultiScreen96-PCR clean-up plate (Millipore). Aliquots of the colony PCRreactions (5 ml) were developed on a 4% agarose gel to verifythe insertion of aptamers into vectors.

Cycle sequencing reactions were performed using a CEQDTCS Quick Start Kit (Beckman-Coulter) and the vendor'smodi®ed M13 sequencing primer (primer ±47 seq). Reactionassembly and cycling conditions were performed largely asdescribed in the vendor's instructions. Approximately100 fmol of puri®ed colony PCR products were used astemplates and reactions were performed with half of themaster mix concentration recommended in the instructions inorder to minimize reagent use (4 ml of master mix in a 20 mlsequencing reaction, rather than 8 ml). Unincorporated dyewas removed by size exclusion chromatography, as describedin Beckman-Coulter Technical Application InformationBulletin T-1874A (http://www.beckman.com/Literature/BioResearch/T-1874A.pdf). Brie¯y, 45 ml of dry SephadexG50 (Amersham-Pharmacia Biotech) was placed into eachwell of a MultiScreen HV plate using a MultiScreen 45 mlColumn Leader (Millipore). The Sephadex chromatographyresin was allowed to hydrate in 300 ml of water for 3 h at roomtemperature. After incubation, the resin was packed bycentrifugation for 5 min at 1100 g. The columns were rinsedonce with 150 ml of water and packed again at the same speedfor the same time. The 20 ml sequencing reactions were loadedonto the tops of the columns using a multichannel micro-pipettor and spun again for 5 min at 1100 g. Puri®ed sampleswere recovered from a CEQ Sample microplate (Beckman-Coulter) that had been placed below the chromatographyplates before centrifugation. Recovered samples were driedunder vacuum at room temperature. Pellets were resuspendedin 40 ml of deionized formamide and developed on a CEQ2000XL eight channel capillary DNA sequencer (Beckman-Coulter). Aptamer secondary structures were predicted usingRNAstructure 3.6 by Mathews et al. (23).

Cellular expression of U1A

The plasmid pJH-hisU1A was transformed into BL21 cells(Stratagene). A 1 ml starter culture grown from a singleplasmid was used to inoculate 50 ml of fresh LB. The culturewas grown to an OD600 of ~0.6 at 37°C and protein expressionwas induced by the addition of IPTG to a ®nal concentration of840 mM. The induced culture was grown for an additional 3 hat 37°C. Cells were pelleted at 5000 g and lysed in B-PER(Pierce), 10 U DNase (Invitrogen) and 10 mM MgCl2 in a totalvolume of 5 ml. After incubation at room temperature for15 min, cellular debris was pelleted at 13 000 g. Thesupernatant was further puri®ed by nickel-chelation chroma-tography. IMAC resin (2 ml; Amersham-Pharmacia Biotech)was equilibrated with 4 ml of water, followed by 4 ml ofcharging buffer (50 mM NiCl2). The column was equilibratedwith 10 ml of protein binding buffer (PBB; 13 PBB = 50 mMTris pH 7.5, 100 mM NaCl, 5 mM imidazole). Clari®edsupernatant (~15 ml) was loaded and the column was washedwith 4 ml of 13 PBB and 15 ml of wash buffer (50 mM TrispH 7.5, 100 mM NaCl, 50 mM imidazole). The U1A proteinwas eluted by the addition of 3 ml of elution buffer (50 mMTris pH 7.5, 100 mM NaCl, 500 mM imidazole); 500 mlfractions were collected for gel analysis. Fractions containingsigni®cant amounts of U1A were pooled and dialyzed in

Table 2. Conditions for theautomated selection of anti-U1Aaptamers

See Materials and Methods for details.The binding ability of the pool wasassayed at 0, 6, 12 and 18 rounds.


50 mM Tris pH 7.5, 100 mM NaCl with four buffer exchangesof 500 ml every 3 h at 4°C in order to remove imidazole fromthe preparation. Finally, the absorbance at 280 nm was used todetermine protein concentration (extinction coef®cient =5442 M±1 cm±1).

Binding constants

Plasmids containing individual aptamers were used to gener-ate transcription templates via PCR. Transcription reactionswere carried out with an AmpliScribe T7 RNA transcriptionkit (Epicentre, Madison, WI) according to the manufacturer'sinstructions, except that incubation was at 42°C rather than37°C. Aptamers were puri®ed on denaturing polyacrylamidegels (24), dephosphorylated and radiolabeled with [g-32P]ATP(25). Radiolabeled RNA was extracted with phenol:chloro-form (1:1) and unincorporated nucleotides were removedusing size exclusion spin columns (Princeton Separations,Adelphia, NJ).

Nitrocellulose ®lter binding assays were employed todetermine the dissociation constants of aptamer±proteincomplexes (25). A standard protocol was automated usingthe Biomek 2000 automated laboratory workstation and amodi®ed Minifold I ®ltration manifold (Schleicher & Schuell,Keene, NH). RNA samples in 13 SSB were thermallyequilibrated at 25°C for 30 min. The RNA concentration forbinding reactions was kept constant at a ®nal concentration of200 pM while the concentration of U1A ranged from 1 mMdown to 17 pM (11 different concentrations). Equal volumes(60 ml) of RNA and U1A were incubated together for 30 min atroom temperature in 13 SBB. The binding reactions (100 ml)were ®ltered through a sandwich of Protran pure nitrocellulose(Schleicher & Schuell) and Hybond N+ nylon transfermembrane (Amersham Pharmacia Biotech) that had beenassembled in the modi®ed Minifold I vacuum manifold. The®lters were washed three times with 125 ml of 13 SBB. Theamount of radiolabeled RNA captured from a reaction onto thenitrocellulose membrane was quanititated using a Phosphor-Imager SI (Amersham Pharmacia Biotech). The log of U1Aconcentration was plotted against the amount of RNA bound.Data were ®tted to the equation y = (a´b)/(b + x) + C, where Cis the fraction of RNA bound to the nitrocellulose at zeroprotein concentration, b is the maximum fraction bound, x isthe fraction of RNA bound to U1A and a is the dissociationconstant for the RNA±protein complex. Assays were per-formed in triplicate and standard deviations were calculated.The 21 nt synthetic RNA (AAUCCAUUGCACUCCGGA-UUU) previously employed in structural studies of the U1Aprotein was used as a positive control (12).

RESULTS AND DISCUSSION

From gene to biotinylated protein

While we have previously automated the in vitro selection ofaptamers that target proteins (4±6), our methods are currentlylimited to puri®ed proteins. Ultimately, the need to purifyprotein targets individually would drastically reduce the speedwith which aptamers could be generated against proteomesand would therefore reduce the utility of aptamers as reagentsfor proteome analysis. Therefore, we have sought to increaseselection throughput by generating protein targets via in vitro

transcription and translation. Moreover, in order to manipulateprotein targets during automated selection, we have attemptedto introduce a biotin tag during the in vitro synthesisprocedure.

A number of kits were available for the in vitro transcriptionand translation of genes. Ultimately, we found that the RocheRTS 100 E.coli HY kit worked well with the automatedselection procedures we had previously established. In order tobiotinylate translated proteins, templates were modi®ed(Fig. 1) so that translated proteins would contain an N-ter-minal peptide tag (MAGGLNDIFEAQKIEWHEDTGGSS)that was an ef®cient substrate for BPL, the product of the birAgene (26±28). The e-amino group of the single lysine residuein the BPL recognition peptide becomes covalently linked tobiotin (29). Biotin ligase can either be co-translated or addedas a separate reagent.

Existing BPL present in E.coli S30 extracts provedinsuf®cient to ef®ciently biotinylate target proteins.Therefore, the MBP gene (malE) was expressed in tandemwith a downstream birA cistron. Following gel separation, ananti-MBP antibody was used to con®rm that roughly equalamounts of MBP were synthesized. However, whenavidin±HRP was used as a probe it was apparent that onlythe dicistronic mbp±birA template directed ef®cient biotinyl-ation. The band at ~20 kDa that stains intensely withavidin±HRP is E.coli biotin carboxyl-carrier protein(BCCP), which is present in the S30 extracts used for in vitrotranslation (Fig. 2). MBP and three other biotinylated proteintargets were immobilized in Neutravidin-coated ELISA wellsand detected with cognate sera (Fig. 3). These results indicatedthat in vitro translated and biotinylated proteins could fold intonative structures that present epitopes similar to those foundin vivo. The amounts of in vitro translation and biotinylationreactions necessary to saturate microwells for selectionexperiments proved to be relatively small (1% of reactionvolume), indicating that this procedure should be useful formultiplex formats, such as the acquisition of aptamers againstmultiple proteins in an organismal proteome.

We have previously carried out selections against chemical-ly biotinylated target proteins (4,6). Unfortunately, chemicalbiotinylation generally generates a population of moleculeswith differing amounts of biotinylation at different conjuga-tion sites, typically a-amino groups on a protein. Thus,multiple different epitopes are presented during selection. Abiotinylation tag obviates this problem and a relativelyhomogeneous set of epitopes should be present during theselection. Additionally, chemical biotinylation may block anactive or allosteric site of the protein, while speci®cbiotinylation at the N-terminus is less likely to interfere withfunction.

In vitro selection of aptamers that bind to translated,biotinylated U1A

As in our previous automated selection experiments, biotinyl-ated protein targets are loaded onto streptavidin beads andincubated with RNA libraries. Bound RNA molecules aresieved from unbound by ®ltration; the use of beads facilitatesrobotic manipulation. The beads are directly transferred to athermal cycler and RNA is prepared for the next round ofselection by a combination of reverse transcription, PCR andin vitro transcription. The entire selection procedure can be


iteratively carried out using a Beckman-Coulter Biomek 2000automated laboratory workstation that has been speciallymodi®ed and programmed to interface with a thermal cycler.The selection robot can carry out six cycles of selection perday; a typical selection is done in 2±3 days. This represents anincrease in speed of at least an order of magnitude relative tomanual selection methods, and up to eight targets can behandled in parallel by the Biomek 2000. Overall, thesemethods allow us to go from a gene sequence to high af®nitybinding reagents for the protein product of that gene in under1 week.

As an initial target, we attempted to select aptamers againsta RNA-binding domain of the U1A protein. U1A is a memberof the U1 snRNP, a component of the nuclear spliceosome thatparticipates in processing of pre-mRNA splicing (30). The ®rst96±100 amino acids of U1A are responsible for bindinghairpin II of U1 snRNA in the splicing complex (31). ItsN-terminal portion, containing one of the two RNP domains ofU1A, has been shown to maintain the binding characteristicsand sequence speci®city of the full-length protein (9,10). Mostimportantly, U1A has previously been targeted by in vitroselection experiments (8) and thus we could be assured that itwould likely generate aptamers.

As the selection progressed, the DNA and RNA poolsproduced at each round were examined by denaturing gelelectrophoresis (Fig. 4). The lengths of the DNA and RNAmolecules were remarkably uniform, especially given that thenumber of cycles and incubation times used for DNA andRNA ampli®cation were pre-set. The consistent sizes ofampli®cation products are in part a result of our prioroptimization of PCR for automated selection procedures(4±6). The stringency of the selection was increased byincreasing the salt concentration from 100 mM during the ®rst12 rounds (2 days) of selection to 300 mM at round 13. It is

interesting to note that the amount of RNA that is ampli®edfollowing this round seems to drop, just as though the increasein stringency reduced the amount of pool that survives theselection. A further increase in stringency was introduced atround 15 by increasing the salt concentration to 600 mM.

Filter binding assays were employed to detect signi®cantchanges in the af®nities of selected pools for U1A (Fig. 5).The initial pool showed signi®cant binding to U1A, likely dueto electrostatic interactions with the highly charged targetprotein (25,32) (+7.2 at pH 7). Binding assays were thereforealso carried out at higher salt concentrations (1 M) to reducethis non-speci®c background binding signal. Six rounds ofautomated selection yielded no signi®cant increase in apparentaf®nity, but a further six yielded an appreciable increase insignal (1.7-fold increase in binding signal even in the presenceof high salt). The ®nal six rounds, carried out underincreasingly stringent conditions, yielded additional improve-ments in pool binding (3.1-fold increase in binding signal evenin the presence of high salt). All binding assays wereperformed using a his-tagged U1A isolated from E.coli, ratherthan the biotin-tagged U1A generated by in vitro translationand biotinylation. The fact that binding did not appear to bedependent upon the type of tag present suggested that bindingwas speci®c for the RNP. It is interesting that no aptamerswere isolated against BCCP, given that this protein was alsoalways present as a target. It is likely that RNP is a better targetfor selection than BCCP. When `mock' selections werecarried out without the gene for U1A, aptamers against BCCPwere isolated (data not shown).

Anti-U1A aptamers resemble natural binding sequences

The natural RNA ligands of the U1A RNP have been wellcharacterized. Hairpin II of U1 snRNA is a short stemtopped by a G:C base pair and a loop with the sequence

Figure 3. Detection of in vitro expressed and biotinylated products. Four constructs were investigated for their ability to bind to neutravidin-coated microwellsafter in vitro expression and biotinylation: b-lactamase (amp), chloramphenicol acetyltransferase (cat), aminoglycoside phosphotransferase (neo) and maltose-binding protein (mbp). Puri®ed, translated and biotinylated products were serially diluted and probed with their respective antisera and again with anti-rabbitIgG±HRP. Signal intensity is measured after incubation of the bound complexes with o-phenylenediamine at 492 nm.


5¢-AUUGCACUCC (Fig. 6A). The bold heptamer sequencehas been shown to be critical for binding to U1A; for instance,G9 makes hydrophobic contacts with Arg52 and hydrogenbonding interactions with Arg52, Leu49, Asn15 and Lys50 ofU1A (33,34). The other six bases form similar hydrophobic,hydrogen bonding and stacking interactions with other peptidebackbone and side chain positions in the U1A protein (12,34).The last three bases (UCC) do not physically interact with theprotein and act primarily as a spacer to close the loop(8,12,35,36). In addition, the U1A protein interacts in a similarmanner with the 3¢-UTR of its own mRNA (37,38),autoregulating its own expression by inhibiting polyadeny-lation and cleavage (12,39,40) (Fig. 6B). However, in thisinstance the canonical heptamer sequence is presented as aninternal loop (Box II), rather than a hairpin loop. Followingbinding of one U1A monomer to the Box II internal loop, asecond U1A monomer is recruited to bind to an adjacentinternal loop containing a variant of the canonical heptamer,AUUGUAC (Box I).

Aptamers from rounds 6, 12 and 18 were cloned andsequenced (Fig. 7). Aptamer sequences that contained thecanonical heptamer were seen at the sixth round of selection:8% (1 of 12 clones) contained a perfect match to the heptamer,while 42% (5 of 12 clones) had imperfect matches to thissequence (Fig. 7A). After another six rounds of selection thesequence diversity of the pool was further reduced and mostindividuals (79%, 11 of 14 clones) possessed the heptamer ornear perfect matches (14%, 2 of 14 clones; Fig. 7B). Furtherselection with increasing stringency yielded a pool where 87%of sequences (26 of 30 clones) contained the heptamer and the

remaining 13% (4 of 30 clones) had near perfect matches(Fig. 7C). No sequences lacked perfect or near perfect matchesin the ®nal pool.

Based on similar sequences found outside of the conservedheptamer, anti-U1A aptamers could be clustered into twosequence families. Family 1 makes up 33% of the ®nal pool,and can be further subdivided into ®ve clonal sub-families, allwith the consensus sequence UGRACAUUGCACUACG.Family 2 comprises 23% of the ®nal population of the pooland is largely clonal.

Predictions of the secondary structures of individualaptamers are consistent with the hypothesis that Family 1mimics the wild-type U1 snRNA hairpin II, while Family 2 isinstead structurally similar to the 3¢-UTR of U1A mRNA(Fig. 6C and D). It should be noted that aptamers that are not inFamily 1 or Family 2 may still generally resemble the wild-type ligands. For example, many of the aptamers listed as`other perfect matches' (Fig. 7C) can readily be folded to formhairpin II-like structures.

Interestingly, all residues that have previously been shownby structural or functional analyses to be important forhydrophobic, stacking or hydrogen bonding interactions wereconserved in the selected aptamers. Conversely, residues thatare known to not be important for interactions with U1A werenot conserved. The degree to which features of natural ligandshave been recapitulated by in vitro selection is remarkable. AC:G base pair closes the loop in the wild-type U1 snRNAhairpin and has been shown to be crucial for binding anarginine residue in U1A. Disruption of this base pair and thisinteraction eliminates U1 snRNA binding to U1A (9,41). All

Figure 4. Nucleic acid products during automated U1A selection. Radioactive reaction products from RT±PCR and transcription reactions during automatedselection were resolved on denaturing polyacrylamide gels. RNA (80 nt) and DNA (99 nt) products are indicated by green and blue arrows, respectively. Theround numbers are listed above the lanes.


aptamers in Family 1 are predicted to form a typical stem±loopstructure closed by the requisite C:G base pair. Conversely,the remaining base pairs in the stem interact non-speci®callywith positively charged residues in U1A (12,33,39). Becauseof this the identity of the base pairs in the stem have beenfound to be irrelevant to U1A binding and, in fact, noparticular base pairs predominate in the predicted stemstructures of the Family 1 aptamers. Similarly, the last threeresidues (UCC) of the 10 residue hairpin loop of the wild-typeU1 snRNA do not physically interact with the protein (12).Family 1 aptamers all have three or four dissimilar (non-conserved) residues in these same positions. The variability inoverall loop length was expected, given that it has previouslybeen shown that the insertion of a single residue into the loophas little effect on protein binding (35). Conversely, no loopsshorter than 10 residues were seen. In this regard, it haspreviously been shown that a nine residue loop lacking one ofthe three otherwise non-conserved residues has 1000-fold lessaf®nity for U1A. The removal of another residue decreases

binding an additional 10-fold, while the creation of a sevenresidue loop ultimately decreases the Kd by 100 000-fold (35).

In the wild-type 3¢-UTR the canonical AUUGCAC andloop-closing C:G base pair again form hydrogen bonds andhydrophobic interactions with U1A (33,34) and, as statedbefore, Family 2 aptamers retain all critical sequence andstructural features. In addition, in the 3¢-UTR a singleadenosine residue (A24) on the opposite side of the internalloop has been shown to form essential hydrogen bonding andstacking interactions with serine, valine and arginine residuesin U1A (33,34). Aptamers in Family 2 can in fact be folded soas to present a single adenosine residue opposite the conservedheptamer.

The wild-type RNA ligands of the U1A protein are knownto bind with high af®nity (Kd values of 10±11±10±8 M,depending on pH and monovalent and divalent salt concen-trations) (35). Under our assay conditions, the minimized, 21 ntwild-type U1 snRNA hairpin II structure forms a complexwith U1A with a Kd of 11 nM. Most aptamers were found to

Figure 5. Progress of automated U1A selection. Binding assays are shown for rounds 0, 6, 12 and 18 of the selection. The ®rst column (light blue) shows thefraction of RNA captured in a standard binding assay. The second column (dark blue) shows the fraction of RNA captured when a high salt wash was used toremove non-speci®cally bound RNA.


have Kd values in the low nanomolar range (Table 3). Family 1aptamers, those most akin to hairpin II, have Kd values as lowas 4.5 nM. Interestingly, at least one member of Family 1,clone 09, has a much higher Kd value (69 nM). This aptamer isalso predicted to form a hairpin-like structure, just as the otherfamily members are. The fact that its Kd deviates signi®cantlysuggests that subtle changes in the presentation of theheptamer motif can be extremely important for binding.Family 2 aptamers bind U1A slightly worse than Family 1,

with the majority aptamer forming a complex with a Kd of17 nM. As expected, sequences with `near perfect' motifsbound much less well than sequences that conformed to thenatural ligands.

The selection of the natural binding sequences was notcompletely unexpected. A previous manual selection had beenperformed against the U1A protein by Tsai et al. (8). Theauthors carried out selection experiments starting from threerandomized pools. The ®rst two pools contained the wild-type

Figure 7. (Following page) Sequences of anti-U1A aptamers. Sequences of selected clones from (A) round 6, (B) round 12 and (C) round 18 are shown. Thestatic (priming) regions are shaded light gray, while the randomized region is in black text. Perfect matches to the U1A snRNA heptamer are highlighted ingray and bold red; near perfect heptamer sequences are highlighted in gray. In (C), the C:G base pair that closes the U1 snRNA hairpin II loop (Family 1)and that precedes Box II of the 3¢-UTR U1 mRNA (Family 2) is shown in blue. The adenosine found in the internal loop region of the 3¢-UTR U1 mRNA(Family 2) is shown in green.

Figure 6. U1A binding elements. (A) U1 snRNA hairpin II; (B) the 3¢-UTR of U1 mRNA. The heptamer sequence critical for recognition by U1A is high-lighted with a black bar. The 3¢-UTR of U1 mRNA (B) contains a second heptamer sequence (Box I) that contains a base change (AUUGUAC) at the ®fthnucleotide. The critical C:G base pair that closes stem II and that precedes Box II is in bold. The adenosine found in the internal loop region of the 3¢-UTR isalso in bold. (C) Aptamer Family 1, drawn in a manner similar to (A). (D) Aptamer Family 2, drawn in a manner similar to (B). Residues contributred by theprimer-binding sites are shown as dark yellow.


OF 14 Nucleic Acids Research, 2002, Vol. 30 No. 20 e108

U1 snRNA stem structure topped by a randomized loop regionof 10 or 13 nt. The third pool contained 25 randomized nt¯anked by primer sequences, similar to our N30 pool. Both theconstrained and unconstrained pools generated aptamerscontaining the canonical heptamer sequence. How-ever, the aptamers generated from the third pool by thesemanual selection procedures exhibited poor binding to U1Aprotein in gel shift experiments (100-fold less af®nity thanwith wild-type ligand). In keeping with this observation, themanually selected aptamers did not mimic the wild-typesequences as well as those generated by automated selection;for example, they did not contain the C:G base pair closing theU1 snRNA loop. Similarly, none of the aptamers selected fromthe unconstrained pool was predicted to fold into either U1snRNA-like or U1A 3¢-UTR mRNA-like structures. Whileadditional rounds of manual selection might of course haveimproved the selected populations, these would have beenprohibitively tedious. The fact that multiple rounds ofautomated selection can be carried out within days allowsthe sequences, structures and af®nities of aptamers to be ®nelytuned.

Prospects for automated selection with in vitro generatedprotein targets

Overall, the results presented here illustrate a strategy for thegeneration of multiple aptamers against a large number ofprotein targets and for the acquisition of sequence andstructural data relating to RNA±protein interactions. Basedon the throughputs available with automated selection, thegeneration of aptamers against the translation products of all

genes within an average bacterial genome could be completedwithin a course of months. Many of the selected aptamerswould of course be useful as reagents alone, for example forlabeling or detecting (1±3) key proteins or even the expressionpro®le of the entire proteome. Nonetheless, the generality ofthese methods for multiple different classes of proteinsremains to be seen.

Our results also con®rm that it may be possible to usein vitro selection to decipher speci®c interactions betweenmultiple different nucleic acid-binding proteins and theircorresponding cellular RNA sequences. In vitro selection haspreviously been used to identify or de®ne a variety of naturalnucleic acid-binding sequences (42). For example, an in vitroselection experiment that targeted the E.coli methioninerepressor generated DNA aptamers that contained an octamersequence found in natural DNA-binding sites (43). Likewise,in vitro RNA selection experiments have generated aptamersto T4 DNA polymerase that both recreate a natural bindingelement and generate a new element that possesses the sameKd as the wild-type (44). Similar ®ndings have been observedduring selections against phage MS2 coat protein, phage R17coat protein and the E.coli special elongation factor SelB(45±47).

From this vantage, it is interesting to note that our selectionproduced two aptamer families that closely corresponded tothe two natural RNA-binding elements of U1A. The fact thatthe natural sequences and structures were extracted from acompletely random sequence library is remarkable and hasadditional important implications for the speci®city of inter-actions between U1A and its natural targets. One measure ofspeci®city is what set of sequences a given protein willpreferentially recognize. In the case of U1A, the naturalligands are apparently globally optimal, at least within thecontext of a sequence space of roughly 30 residues. Indeed, theU1A protein can even ®nely distinguish between ligands thatsuper®cially resemble one another in terms of criticalsequences and structures, since different Family 1 memberscan have Kd values that vary by an order of magnitude. Theseresults imply both that the U1A protein has structured itsRNA-binding domain in such a way that it can rejectsequences and structures that are not exactly like the naturalligands, and that the natural ligands have been derived by asearch of a sequence space that is much larger than the size ofthe genomes in which they are ensconced. This is one of the®rst examples of natural optimality in a sequence space of thissize.

While other methods for ascertaining natural nucleic acid-binding sequences are also possible, including genomicSELEX (48±50), in which libraries are generated directlyfrom natural sequences, the use of randomized libraries alsoallows the identi®cation of unnatural sequences that may haveimproved af®nities and greater utility as laboratory reagents.For example, in vitro selection against the Rev protein ofHIV-1 yielded aptamers that were similar to the wild-typeRev-binding element (RBE), but bound to Rev 10 times better(51). In the current instance, despite the fact that the naturalRNA sequences have been exquisitely optimized by naturalselection, automated selection has apparently improved onwild-type af®nities by a factor of approximately 2.4 (4.5versus 11 nM).

Table 3. Dissociation constants ofcomplexes formed with selectedsequences

Numbers under families are clonenumbers from round 18. Binding assayswere performed in triplicate to determinestandard deviations.


ACKNOWLEDGEMENTS

We thank Dr David Waugh (NCI±Frederick) for the gift ofpDW363. This work was funded by grants from the BeckmanFoundation, Robert A. Welch Foundation (F-1393), ArmyResearch Of®ce, (MURI DAAD19-99-1-0207) and NIH-Rev(AI36083-09).

REFERENCES

1. Jayasena,S.D. (1999) Aptamers: an emerging class of molecules thatrival antibodies in diagnostics. Clin. Chem., 45, 1628±1650.

2. Rajendran,M. and Ellington,A.D. (2002) Selecting nucleic acids forbiosensor applications. Comb. Chem. High Throughput Screen., 5,263±270.

3. Wilson,D.S. and Szostak,J.W. (1999) In vitro selection of functionalnucleic acids. Annu. Rev. Biochem., 68, 611±647.

4. Cox,J.C., Rajendran,M., Riedel,T., Davidson,E.A., Sooter,L.J.,Bayer,T.S., Schmitz-Brown,M. and Ellington,A.D. (2002) Automatedacquisition of aptamer sequences. Comb. Chem. High ThroughputScreen., 5, 289±299.

5. Cox,J.C., Rudolph,P. and Ellington,A.D. (1998) Automated RNAselection. Biotechnol. Prog., 14, 845±850.

6. Cox,J.C. and Ellington,A.D. (2001) Automated selection of anti-proteinaptamers. Bioorg. Med. Chem., 9, 2525±2531.

7. Blackwell,T.K. and Weintraub,H. (1990) Differences and similarities inDNA-binding preferences of MyoD and E2A protein complexes revealedby binding site selection. Science, 250, 1104±1110.

8. Tsai,D.E., Harper,D.S. and Keene,J.D. (1991) U1-snRNP-A proteinselects a ten nucleotide consensus sequence from a degenerate RNA poolpresented in various structural contexts. Nucleic Acids Res., 19,4931±4936.

9. Jessen,T.H., Oubridge,C., Teo,C.H., Pritchard,C. and Nagai,K. (1991)Identi®cation of molecular contacts between the U1 A small nuclearribonucleoprotein and U1 RNA. EMBO J., 10, 3447±3456.

10. Hall,K.B. (1994) Interaction of RNA hairpins with the human U1AN-terminal RNA binding domain. Biochemistry, 33, 10076±10088.

11. Stemmer,W.P., Crameri,A., Ha,K.D., Brennan,T.M. and Heyneker,H.L.(1995) Single-step assembly of a gene and entire plasmid from largenumbers of oligodeoxyribonucleotides. Gene, 164, 49±53.

12. Oubridge,C., Ito,N., Evans,P.R., Teo,C.H. and Nagai,K. (1994) Crystalstructure at 1.92 AÊ resolution of the RNA-binding domain of the U1Aspliceosomal protein complexed with an RNA hairpin. Nature, 372,432±438.

13. Horton,R.M., Hunt,H.D., Ho,S.N., Pullen,J.K. and Pease,L.R. (1989)Engineering hybrid genes without the use of restriction enzymes: genesplicing by overlap extension. Gene, 77, 61±68.

14. Mertens,N., Remaut,E. and Fiers,W. (1996) Increased stability of phageT7g10 mRNA is mediated by either a 5¢- or a 3¢-terminal stem-loopstructure. Biol. Chem., 377, 811±817.

15. Olins,P.O., Devine,C.S., Rangwala,S.H. and Kavka,K.S. (1988) The T7phage gene 10 leader RNA, a ribosome-binding site that dramaticallyenhances the expression of foreign genes in Escherichia coli. Gene, 73,227±235.

16. Schatz,P.J. (1993) Use of peptide libraries to map the substrate speci®cityof a peptide-modifying enzyme: a 13 residue consensus peptide speci®esbiotinylation in Escherichia coli. Biotechnology, 11, 1138±1143.

17. Tsao,K.L., DeBarbieri,B., Michel,H. and Waugh,D.S. (1996) A versatileplasmid expression vector for the production of biotinylated proteins bysite-speci®c, enzymatic modi®cation in Escherichia coli. Gene, 169,59±64.

18. Sambrook,J. and Russell,D.W. (2001) Molecular Cloning: A LaboratoryManual, 3rd Edn. Cold Spring Harbor Laboratory Press, Cold SpringHarbor, NY.

19. Laemmli,U.K. (1970) Cleavage of structural proteins during theassembly of the head of bacteriophage T4. Nature, 227, 680±685.

20. Hayhurst,A. and Harris,W.J. (1999) Escherichia coli skp chaperonecoexpression improves solubility and phage display of single-chainantibody fragments. Protein Expr. Purif., 15, 336±343.

21. Bell,S.D., Denu,J.M., Dixon,J.E. and Ellington,A.D. (1998) RNAmolecules that bind to and inhibit the active site of a tyrosinephosphatase. J. Biol. Chem., 273, 14309±14314.

22. Clackson,T.D., GuÈssow,D. and Jones,P.T. (1991) Characterizingrecombinant clones directly by PCR screening. In Rickwood,D. andHames,B.D. (eds), PCR 1. Oxford University Press, Oxford, UK, pp.210±212.

23. Mathews,D.H., Sabina,J., Zuker,M. and Turner,D.H. (1999) Expandedsequence dependence of thermodynamic parameters improves predictionof RNA secondary structure. J. Mol. Biol., 288, 911±940.

24. Ellington,A.D. (1997) Preparation and analysis of DNA. In Ausubel,F.,Brent,R., Kingston,R.E., Moore,D.D., Seidman,J.G., Smith,J.A. andStruhl,K. (eds), Short Protocols in Molecular Biology, 3rd Edn. JohnWiley & Sons, New York, NY, pp. 2-43±2-44.

25. Jhaveri,S.D. and Ellington,A.D. (2000) Combinatorial methods innucleic acid chemistry. In Beaucage,S.L., Bergstrom,D.E., Glick,G.D.and Jones,R.A. (eds), Current Protocols in Nucleic Acid Chemistry. JohnWiley & Sons, New York, NY, pp. 9.3.1±9.3.25.

26. Chapman-Smith,A. and Cronan,J.E.,Jr (1999) The enzymaticbiotinylation of proteins: a post-translational modi®cation of exceptionalspeci®city. Trends Biochem. Sci., 24, 359±363.

27. Saviranta,P., Haavisto,T., Rappu,P., Karp,M. and Lovgren,T. (1998)In vitro enzymatic biotinylation of recombinant fab fragments through apeptide acceptor tail. Bioconjug. Chem., 9, 725±735.

28. Cull,M.G. and Schatz,P.J. (2000) Biotinylation of proteins in vivo andin vitro using small peptide tags. Methods Enzymol., 326, 430±440.

29. McAllister,H.C. and Coon,M.J. (1966) Further studies on the propertiesof liver propionyl coenzyme A holocarboxylase synthetase and thespeci®city of holocarboxylase formation. J. Biol. Chem., 241,2855±2861.

30. De Guzman,R.N., Turner,R.B. and Summers,M.F. (1998) Protein-RNArecognition. Biopolymers, 48, 181±195.

31. Scherly,D., Boelens,W., van Venrooij,W.J., Dathan,N.A., Hamm,J. andMattaj,I.W. (1989) Identi®cation of the RNA binding segment of humanU1 A protein and de®nition of its binding site on U1 snRNA. EMBO J.,8, 4163±4170.

32. Binkley,J., Allen,P., Brown,D.M., Green,L., Tuerk,C. and Gold,L. (1995)RNA ligands to human nerve growth factor. Nucleic Acids Res., 23,3198±3205.

33. Allain,F.H., Howe,P.W., Neuhaus,D. and Varani,G. (1997) Structuralbasis of the RNA-binding speci®city of human U1A protein. EMBO J.,16, 5764±5772.

34. Allain,F.H., Gubser,C.C., Howe,P.W., Nagai,K., Neuhaus,D. andVarani,G. (1996) Speci®city of ribonucleoprotein interaction determinedby RNA folding during complex formulation. Nature, 380, 646±650.

35. Williams,D.J. and Hall,K.B. (1996) RNA hairpins with non-nucleotidespacers bind ef®ciently to the human U1A protein. J. Mol. Biol., 257,265±275.

36. Katsamba,P.S., Myszka,D.G. and Laird-Offringa,I.A. (2001) Twofunctionally distinct steps mediate high af®nity binding of U1A proteinto U1 hairpin II RNA. J. Biol. Chem., 276, 21476±21481.

37. Boelens,W.C., Jansen,E.J., van Venrooij,W.J., Stripecke,R., Mattaj,I.W.and Gunderson,S.I. (1993) The human U1 snRNP-speci®c U1A proteininhibits polyadenylation of its own pre-mRNA. Cell, 72, 881±892.

38. van Gelder,C.W., Gunderson,S.I., Jansen,E.J., Boelens,W.C.,Polycarpou-Schwarz,M., Mattaj,I.W. and van Venrooij,W.J. (1993) Acomplex secondary structure in U1A pre-mRNA that binds twomolecules of U1A protein is required for regulation of polyadenylation.EMBO J., 12, 5191±5200.

39. Nagai,K., Oubridge,C., Ito,N., Avis,J. and Evans,P. (1995) The RNPdomain: a sequence-speci®c RNA-binding domain involved inprocessing and transport of RNA. Trends Biochem. Sci., 20, 235±240.

40. Jovine,L., Oubridge,C., Avis,J.M. and Nagai,K. (1996) Two structurallydifferent RNA molecules are bound by the spliceosomal protein U1Ausing the same recognition strategy. Structure, 4, 621±631.

41. Nagai,K., Oubridge,C., Jessen,T.H., Li,J. and Evans,P.R. (1990) Crystalstructure of the RNA-binding domain of the U1 small nuclearribonucleoprotein A. Nature, 348, 515±520.

42. Conrad,R.C., Symensma,T.L. and Ellington,A.D. (1997) Natural andunnatural answers to evolutionary questions. Proc. Natl Acad. Sci. USA,94, 7126±7128.

43. He,Y.Y., Stockley,P.G. and Gold,L. (1996) In vitro evolution of the DNAbinding sites of Escherichia coli methionine repressor, MetJ. J. Mol.Biol., 255, 55±66.

44. Tuerk,C. and Gold,L. (1990) Systematic evolution of ligands byexponential enrichment: RNA ligands to bacteriophage T4 DNApolymerase. Science, 249, 505±510.


45. Xu,W. and Ellington,A.D. (1996) Anti-peptide aptamers recognize aminoacid sequence and bind a protein epitope. Proc. Natl Acad. Sci. USA, 93,7475±7480.

46. Schneider,D., Tuerk,C. and Gold,L. (1992) Selection of high af®nityRNA ligands to the bacteriophage R17 coat protein. J. Mol. Biol., 228,862±869.

47. Hirao,I., Spingola,M., Peabody,D. and Ellington,A.D. (1998) The limitsof speci®city: an experimental analysis with RNA aptamers to MS2 coatprotein variants. Mol. Divers., 4, 75±89.

48. Shtatland,T., Gill,S.C., Javornik,B.E., Johansson,H.E., Singer,B.S.,Uhlenbeck,O.C., Zichi,D.A. and Gold,L. (2000) Interactions of

Escherichia coli RNA with bacteriophage MS2 coat protein: genomic

SELEX. Nucleic Acids Res., 28, e93.49. Singer,B.S., Shtatland,T., Brown,D. and Gold,L. (1997) Libraries for

genomic SELEX. Nucleic Acids Res., 25, 781±786.50. Gold,L., Brown,D., He,Y., Shtatland,T., Singer,B.S. and Wu,Y. (1997)

From oligonucleotide shapes to genomic SELEX: novel biologicalregulatory loops. Proc. Natl Acad. Sci. USA, 94, 59±64.

51. Giver,L., Bartel,D., Zapp,M., Pawul,A., Green,M. and Ellington,A.D.

(1993) Selective optimization of the Rev-binding element of HIV-1.

Nucleic Acids Res., 21, 5509±5516.


cox2002-automated_selection_of_aptamers_against_protein_targets_translated_in_vitro_from_gene_to_aptamer...

Documents