Deconvolution of a Complex Target Using DNA Aptamers*

In vitro selection of single-stranded nucleic acid aptamers from large random sequence libraries is now a straightforward process particularly when screening with a single target molecule. These libraries contain considerable shape diversity as evident by the successful isolation of aptamers that bind with high affinity and specificity to chemically diverse targets. We propose that aptamer libraries contain sufficient shape diversity to allow deconvolution of a complex mixture of targets. Using unfractionated human plasma as our experimental model, we aim to develop methods to obtain aptamers against as many proteins as possible. To begin, it is critical that we understand how aptamer populations change with increasing rounds of in vitro selection when using complex mixtures. Our results show that sequence representation in the selected population changes dramatically with increasing rounds of selection. Certain aptamer families were apparent after only three selection rounds. Two additional cycles saw a decline in the relative abundance of these families and the emergence of yet another family that accounted for more than 60% of sequences in the pool. To overcome this population convergence, an aptamer-based target depletion method was developed, and the library screen was repeated. The previous dominant family effectively disappeared from the selected populations but was replaced by other aptamer families. Insights gained from these initial experiments are now being applied in the creation of second generation plasma protein screens and also to the analysis of other complex biological targets.

In vitro selection of single-stranded nucleic acid aptamers from large random sequence libraries is now a straightforward process particularly when screening with a single target molecule. These libraries contain considerable shape diversity as evident by the successful isolation of aptamers that bind with high affinity and specificity to chemically diverse targets. We propose that aptamer libraries contain sufficient shape diversity to allow deconvolution of a complex mixture of targets. Using unfractionated human plasma as our experimental model, we aim to develop methods to obtain aptamers against as many proteins as possible. To begin, it is critical that we understand how aptamer populations change with increasing rounds of in vitro selection when using complex mixtures. Our results show that sequence representation in the selected population changes dramatically with increasing rounds of selection. Certain aptamer families were apparent after only three selection rounds. Two additional cycles saw a decline in the relative abundance of these families and the emergence of yet another family that accounted for more than 60% of sequences in the pool. To overcome this population convergence, an aptamer-based target depletion method was developed, and the library screen was repeated. The previous dominant family effectively disappeared from the selected populations but was replaced by other aptamer families. Insights gained from these initial experiments are now being applied in the creation of second generation plasma protein screens and also to the analysis of other complex biological targets.
It is becoming increasingly apparent that our understanding of the complexity of many biological processes is limited by studies that focus on just a few molecules. This realization is largely due to the recent development of technologies that allow simultaneous measurement of thousands of different molecules. Information obtained from this type of assay not only puts a scale (number of genes or proteins) to the system under study but allows more accurate classification of biological states. Clearly, more detailed insights into a biological system or process can be gained if the number and identity of components is known (1,2). Arraybased platforms for profiling gene expression have already provided important new insights into many areas of biology (3)(4)(5)(6). In particular the assignment of genotypic and phenotypic signatures to various biological states is not only useful diagnostically but has also increased our understanding of disease progression (7)(8)(9).
Although analysis of nucleic acids has and will continue to be important, extension of the same technology to proteins is eagerly anticipated.
The task overall is the same: identify all of the components of the system and develop ways to measure their concentration. The technical challenge, however, is considerably more difficult because of the enormous shape diversity present in the proteome (10). This multiplicity of shapes is increased further when consideration is given to how the immediate chemical environment can modify three-dimensional structure, for example, whether the protein is part of a complex or subject to some form of activation. Current approaches to defining the composition of complex mixtures of proteins commonly involve state-of-the-art mass spectrometry coupled with some form of chromatography. Quantitation of individual proteins is largely being addressed using mass spectrometry or by shape recognition via antibodies (11). These approaches are limiting either because they are expensive and require a high level of technical expertise or because deriving the required antibodies is slow.
Various attempts have been made to use random shape libraries and in vitro selection methods to study complex biological targets. In certain cases this approach has led to the identification of functional agonists or antagonists (12,13), whereas in others it has provided a method to gain insight into the molecular complexity of the biological target (14 -17). Such libraries contain considerable shape diversity in a form that can be readily accessed. Typically the library is subjected to some form of affinity partitioning against a target molecule, and the small amount of bound material is recovered and amplified. Specific target-binding species are enriched by repeating cycles of in vitro selection until a predetermined end point is reached. In a majority of cases the selection is stopped when sequence analysis shows that the selected pool contains multiple representations of a particular motif. Typically this motif will bind to a single target within the complex mixture (18 -22). Used in this way, in vitro selection simply produces the equivalent of an antibody with the advantage that all manipulations occur in vitro, and the desired reagent is obtained in a shorter period of time.
Rather than focus on just one or a few targets in a complex biological mixture, there is no reason why most if not all proteins could be identified using random sequence libraries. This would, however, require a library screening strategy that balances a requirement for obtaining shapes that exhibit high affinity binding against the observed narrowing of the population toward a single target with increasing rounds of in vitro selection. To achieve this balance it is necessary to understand how certain experimental variables influence the composition and target binding affinity of in vitro selected populations and how those populations change with increasing rounds of selection.
Arguably one of the most important variables that determines the success of an in vitro selection experiment is the fraction of shape space present in a library (Ref. 23 and references therein). Although rough approximations can be made for the amount of sequence diversity in a library, current knowledge does not allow reliable estimates of the corresponding shape diversity. The influence of a number of other variables has been modeled both experimentally (49) and mathematically both for single targets and mixtures of targets (24 -26). In practice, enrichment for high affinity binders has usually involved a progressive decrease in target concentration or extended washing prior to elution of the bound population from the target. The diversity of selected pools has been followed either by sequencing or some form of fingerprinting technique but only in the context of deciding when to terminate the screen rather than to understand in vitro selection dynamics. Much still needs to be done to more thoroughly understand the in vitro selection process, particularly when applied to complex mixtures.
It is well known that single-stranded nucleic acids adopt unique sequence-dependent three-dimensional shapes (27,28). This property has been exploited to create shape libraries that have yielded ligands that bind with high affinity and specificity to a variety of molecules across a broad range of size and chemical composition (29,30). A number of novel diagnostics and therapeutics have been isolated from these aptamer libraries, further indicating the significant amount of shape diversity they possess (31)(32)(33)(34)(35)(36). We have initiated studies to establish whether there is enough shape diversity in a nucleic acid aptamer library to allow recognition of each component of a complex biological mixture. Human plasma was chosen as a model complex mixture because it contains a large number of proteins present at concentrations that span 6 orders of magnitude and because its composition can easily be changed by chromatography (37). The latter property is important because the high concentration of albumin has plagued other efforts to define the plasma proteome, and we needed to be able to remove it if required (38,39). Using a straightforward in vitro selection protocol, we isolated aptamers that bound to several different plasma proteins. We followed the relative abundance of aptamer families with increasing rounds of in vitro selection to develop an understanding of the selection process particularly in terms of the number of targets identified. Insights gained from these initial results led to the design of a modified library screening protocol that yielded a different selected population profile and dynamic. Binding affinities were determined at different stages of the selection process for several aptamers that bound to prothrombin. High affinity aptamers were present in the population after only four cycles of selection.

EXPERIMENTAL PROCEDURES
Oligonucleotide Library and Reagents-DNA oligonucleotide template library 5Ј-AGCTCAGAATAAACGCTCAA-N 45 -TTCGACAT-GAGGCCGGATC-3Ј, unmodified and modified (5Ј-biotin or 5Ј-phosphate) primers (forward, 5Ј-GGGAGCTCAGAATAAACGCTCAA-3Ј; reverse, 5Ј-GATCCGGCCTCATGTCGAA-3Ј) were produced by solid phase chemical synthesis (MWG Biotech, Ebersberg, Germany). Because of variability in the quality of oligonucleotide synthesis, fulllength product was purified on 8% denaturing polyacrylamide gels. DNA was recovered from gels by electroelution and ethanol-precipitated. This single-stranded material was made double-stranded by a standard fill-in reaction using T7 DNA polymerase, and library complexity was estimated from the yield of double-stranded DNA (40). Larger amounts of this pool of double-stranded DNA were created by polymerase chain reaction. Typically, 2 ϫ 10 14 double-stranded DNA molecules were amplified in a 34.8-ml reaction (for a 0.1-ml reaction: 20 pM forward and reverse primers, 200 M dNTPs, 0.5 unit Taq polymerase (New England Biolabs, Beverly, MA)). A total of five cycles (94°C 30 s, 58°C 10 s, and 72°C 30 s) were performed. Unless otherwise specified, all reagents were obtained from Sigma.
Coupling Plasma Proteins to CH-Sepharose-Human blood was obtained by venipuncture into vacuette tubes containing 5 mM EDTA (Greiner Bio-One, Frickenhausen, Germany). Platelet poor plasma was isolated by centrifugation at 2500 ϫ g for 15 min and then dialyzed (Spectra/Por MWCO 3500; Spectrum Laboratories Inc., CA) against 137 mM NaCl, 2.7 mM KCl, 8 mM Na 2 HPO 4 , 1.15 mM KH 2 PO 4 , 5 mM MgCl 2 , pH 7.4 (1ϫ BB). The plasma proteins were coupled to CH-Sepharose 4B according to the manufacturer's instructions (Amersham Biosciences). The quantity of protein covalently bound to Sepharose was determined using a bicinchoninic acid binding assay (Pierce). Typically 70% of plasma proteins were coupled at a density of 17.5 g protein/mg of Sepharose.
In Vitro Selection Protocol-Library screens were initiated using 4.5 g of single-stranded DNA (ssDNA) 3 including 20 fmol of 5Ј-32 P-labeled ssDNA (5 ϫ 10 3 CPM/fmol ssDNA) as tracer. Single-stranded DNA in 1ϫ BB (1 M) was heat-denatured for 10 min at 68°C and cooled to room temperature. To remove matrix-binding aptamers the library was incubated for 90 min with 20 mg of CH-Sepharose that had been blocked with Tris-HCl. Unbound ssDNA was partitioned from the Sepharose by spin filtration (1000 ϫ g for 30 s) with a SPIN-X spin filter (Corning Costar, Acton, MA). Recovered ssDNA was then incubated with 20 mg of plasma-coupled Sepharose for 90 min at room temperature. Unbound ssDNA was again removed by spin filtration, and the beads were washed five times with 0.5 ml of 1ϫ BB. Bound ssDNA was recovered by incubating beads in 7 M urea, 5 mM EDTA, pH 8, at 80°C for 5 min followed by ethanol precipitation. Recovered ssDNA was amplified (94°C 30 s, 58°C 10 s, and 72°C 30 s) with forward and phosphorylated reverse primers using Taq polymerase (New England Biolabs). Excess primer was digested with exonuclease I (2 units/ml), and then ssDNA was generated using exonuclease (12.5 units/ml) (New England Biolabs). For subsequent cycles the amount of aptamer pool used for selections was reduced 5-fold.
Cloning and DNA Sequence Analysis-Amplified double-stranded DNA from selected pools was cloned into pGEM T-Easy (Promega) and sequenced using a Beckman CEQ TM 8000 automated DNA sequencer (Beckman Coulter, Fullarton, CA). For alignments, a software program written in Python (www.python.org/) and running the SPLASH pattern recognition algorithm (41) was used (Biolateral, Sydney, Australia). Sequences that conformed to a minimum pattern length of 8 bases with a minimum of 6 non-wild card bases in the pattern were obtained.
Protein Purification and Identification-Single-stranded DNA for individual aptamers was generated by PCR using a 5Ј-biotinylated forward primer. Biotinylated ssDNA (10 pmol) was thermally folded and incubated with 25 l of dialyzed plasma in a final volume of 100 l for 90 min at room temperature. Aptamer-protein complexes were then captured on streptavidin-Sepharose (Amersham Biosciences). The complexes were washed four times with 0.5 ml of 1ϫ BB containing 0.05% Tween 20. The beads were resuspended in Laemmli loading buffer and heated at 90°C for 3 min, and eluates were electrophoresed on 6% SDS-PAGE gels. The proteins were visualized by silver staining (42).
Matrix-assisted laser adsorption ionization mass spectroscopy of aptamer bead-bound protein and peptide mass fingerprinting analysis was performed by the Australian Proteomics Facility (Macquarie University, Sydney, Australia).
Surface Plasmon Resonance Measurements-Biomolecular interaction analysis was performed using a Biacore 2000 biosensor (Biacore, Uppsala, Sweden). Sensor chips (CM5 research grade) and reagents for covalent coupling were purchased from Biacore. The running buffer was 1ϫ BB containing 0.005% Tween 20. For kinetic studies, streptavidin in 10 mM sodium acetate, pH 4.5, at 5 g/ml was covalently coupled to the carboxymethylated dextran matrix on all four flow cells using amine coupling reagents N-ethyl-NЈ-(3-diethylamino propyl)carbodiimide and N-hydroxysuccinimide according to the preset software specifications (Biacore AB). Typically, 450 -500 resonance units of streptavidin was immobilized. To flow cells 1 and 2, saturating amounts of linker oligonucleotide (1 M in 1ϫ BB), which is complementary to 5Ј primer-binding site of the aptamer (5Ј-CGTTTATTCTGAGCTCCCbiotin-3Ј), were injected at 30 l/min for 30 s. This process was repeated for flow cells 3 and 4 using a linker oligonucleotide that was complementary to the 3Ј primer-binding site (5Ј-biotin-GATCCGGGCCT-CATG-3Ј). Aptamers (200 nM in 1ϫ BB) were then injected over the biotinylated linker/streptavidin complex on flow cell 2 and 4 at a flow rate of 5 l/min for 5 min. In some cases duplex formation was linkerspecific in that some aptamers would only form a duplex with either the 5Ј or 3Ј linker oligonucleotide. The experiments were performed on flow cells where 20 -50 resonance units of linker/aptamer partial duplex was formed.
For kinetic studies, a concentration series of prothrombin in running buffer was injected at a flow rate of 100 l/min over the linker and linker/ aptamer flow cells at 20°C. The association and dissociation phases were each monitored for 3 min, and the flow cells were regenerated with a pulse of 0.05% SDS. To determine error margins, two experiments were performed for each aptamer. This included stripping the aptamer from the flow cell with a pulse of 10 mM NaOH and then reforming the duplex and repeating the same concentration series of prothrombin.
When analyzing the data, the signals generated from injections of prothrombin over the linker only reference flow cell and those generated from injection of running buffer over the aptamer flow cell were subtracted from sensorgrams obtained for prothrombin injected over the aptamer flow cell. Rate constants (k a and k d ) were obtained using global data fitting algorithms assuming a Langmuir (1:1) binding isotherm (BIAevaluation V3.1 software; Biacore). The equilibrium dissociation constant (K D ) was determined from the ratio k d /k a .
Prothrombin Depletion-A 5Ј-biotinylated prothrombin aptamer (F1.06; see Fig. 3 for sequence) truncated to remove primer-binding sites was used. Plasma (60 l), dialyzed against 1ϫ BB, was mixed with 100 pmol of biotinylated oligonucleotide in a volume of 100 l for 2 h at room temperature. Prothrombin-aptamer complexes were captured using streptavidin-Sepharose (Amersham Biosciences) and, recovered by spin filtration. Three cycles of aptamer mediated depletion were performed and the captured prothrombin together with known quantities of bovine serum albumin were analyzed on 8% SDS-PAGE. To quantitate prothrombin extraction and check for the presence of other plasma proteins, the gels were stained with SYPRO Orange (Molecular Probes, Eugene, OR) and viewed on a Typhoon FluorImager (Molecular Dynamics). Prothrombin-depleted plasma was subsequently coupled to CH-Sepharose, and an in vitro selection was performed as described above.

RESULTS
Aptamer Library Screening Strategy-In light of the large number of variables that could influence the outcome of an aptamer library screen, we deliberately chose the simplest selection protocol possible that included using a library with a moderately sized random sequence region. We anticipated that results from such a screen would indicate whether more sophisticated strategies were needed and what form they might take.
A library of ϳ10 14 single-stranded DNA molecules was incubated with unfractionated plasma proteins coupled to activated Sepharose beads. The quantity of beads and therefore the concentration of target remained constant throughout the screen as did the washing and elution conditions. A trace amount of 32 P-labeled aptamer pool was included at each cycle of selection to monitor enrichment of specific binding sequences. Targets for individual aptamers were identified by incubating plasma with 5Ј-biotinylated aptamer, trapping the complex on streptavidin-Sepharose, and then analyzing the bound protein by SDS-polyacrylamide gel electrophoresis. Protein identification by this method was limited by the resolving capacity of the gel and the sensitivity of the silver staining method used. Despite these limitations, this format was considered the best option given the large number of aptamers to be assayed.
Dynamics of Aptamer Library Screening against a Complex Target-The DNA aptamer library contained a random sequence region of 45 nucleotides flanked by two 20-nucleotide regions for amplification. The amount of library used in a screen is a important variable, and we chose what corresponded to an average value (4.5 g) obtained from the literature. It is common to quote a figure for the sequence complexity rather than a mass, but in our view the former is impossible to estimate accurately because of template bias that manifests itself at several steps during the creation of the library. A relatively modest amount of target was used (300 g), especially considering the abundance of albumin. We reasoned that if albumin was going to dominate the screen, then increasing the amount of target would not significantly increase our chances of isolating aptamers that bound to other proteins. A total of 13 cycles of selection were performed, and the amount of bead-bound aptamer was measured after each cycle. Significant binding above background was detected after cycle five, and this increased to a total of 12% at the last cycle (Fig. 1).
Aliquots of selected pools from cycles 3-12 were cloned, randomly chosen colonies were picked, and their inserts were sequenced. The resultant data were then analyzed with the SPLASH pattern algorithm to identify common sequence motifs. Homology between different aptamers suggests they might bind to the same target and is an indirect measure of the degree of enrichment obtained at any stage of the in vitro selection process. In addition to searching for multiple representations of a motif at any cycle of selection, we also followed individual sequences with increasing cycles of selection to see whether their relative frequency in the population changed.
The level of sequence diversity observed within aptamer families was highly variable, with some families exhibiting less than 5% sequence variation between clones, whereas others were characterized by small highly conserved consensus motifs (data not shown). Representative aptamer sequences for all aptamers families are included in TABLE ONE.
The first observation we make is that the number of aptamers that either show no homology to any other member of their pool or were FIGURE 1. Monitoring in vitro selection cycles. Specific binding of aptamers to plasma proteins (squares) was monitored by including tracer quantities of [␥-32 P]ATP end-labeled ssDNA in the aptamer population at each cycle. The amount of radioactivity in 5% of the recovered ssDNA was determined by liquid scintillation counting and expressed as a percentage of the input counts/min for each cycle. Background binding (triangles) was determined by a negative selection step performed for every cycle with Sepharose that had been reacted with Tris-HCl instead of plasma. The graph shows that only background binding was detected up to cycle 5. A progressive increase in binding above background was observed from cycle 6 until the selection was terminated. OCTOBER 7, 2005 • VOLUME 280 • NUMBER 40 present only once declined dramatically from cycle 3 to cycle 5 ( Fig. 2A). This trend indicated that diversity within the selected aptamer population declined rapidly and was apparent after a relatively small number of selection cycles. The decline in sequence diversity of pools was mirrored by an increase in the appearance of aptamer families. This is most obvious if the relative frequency of the different aptamer families is plotted against selection cycle number. We have split this data between two graphs because the abundance of two families F1 and F2 masks all other families. Family F1 was clearly the dominant family from the time it first appeared through to the last selection cycle (Fig. 2B). There is a suggestion, however, from pool 12 that the F1 population might be declining, whereas F2, which emerged late in the screen, was increasing. Other families (F15, F16, F17, and F18) with some homology to F1 are also included in this graph, but their frequency remains low. Examination of the data for other families immediately shows the large difference in their abundance relative to F1. Moreover, F1 was not the first family to appear. F12, F13, and F14 were multiply represented in the population from cycle 3 (Fig. 2C). Not surprisingly they represented a very small percentage of the total number analyzed but indicated that the selection process was enriching for certain aptamers even at this very early stage. The distribution of sequences changed significantly between cycles 3 and 4. Only one of the families (F12) identified in cycle 3 was represented in cycle 4, and although its frequency had increased, it was not the most abundant family. Examination of aptamer family frequencies at later stages of the selection revealed complex dynamics. Some fami-lies appeared and then disappeared (F13 and F22), whereas others only emerged toward the end of the experiment (F2 and F3).

In Vitro Selection Using Complex Targets
Protein Target Identification-Computer analysis of nucleotide sequences predicts relationships between different aptamers. To see how informative this analysis was, we sought to identify protein targets for many of the aptamers isolated in the screen. The target assay was designed to allow rapid analysis of some 500 aptamers. We made an assumption that most of the aptamers isolated in the screen would bind to abundant or moderately abundant proteins. The silver staining protocol used could easily detect 10 ng of albumin, and so a single concentration of plasma (25 l) was used in each binding reaction. If we assume that an aptamer traps the majority of its target present in the plasma sample, then we should detect any protein present at a concentration equal to or greater than 400 ng/ml. Results from these experiments showed that aptamer binding was very specific with little or no background (representative data in Fig. 3). In many cases aptamers bound single proteins, but in some cases two or more proteins were present, suggesting binding to one subunit of a complex. It should be noted that all of the proteins in our experiments should be in their native conformation except where attachment to the Sepharose beads alters their structure. In this first round of assays, protein targets were identified for 17 of the 23 aptamer families previously defined on the basis of unique sequence motifs (TABLE ONE). Six families that failed to yield a detectable target were then reassayed using five times more plasma. This resulted in the identification of protein targets for families F5 and F8. It became increasingly apparent when performing these experiments that aptamers bound to their targets with very high specificity. A single step of aptamer-mediated affinity chromatography typically yielded proteins that by gel analysis were 85-90% pure, which in the case of prothrombin represents a 400-fold purification from crude plasma. These results suggested to us that amino acid sequence information could be obtained directly from protein bound to aptamer bearing beads. In a proof of principle experiment, we used matrix-assisted laser adsorption ionization time-of-flight mass spectrometry and peptide mass fingerprinting to identifying protein targets for F1, prothrombin (72 kDa), and F3, C3 complement (180 kDa).
Several families (F2 and F15-18) appeared to bind a protein that comigrated on SDS-polyacrylamide gels with the protein recognized by F1. A Western blot was probed with a monoclonal antibody to prothrombin to establish whether these aptamers were binding to the same protein. All of the bands were positive, thereby confirming their identity (data not shown). This result indicates that greater than 70% of the aptamers present in the cycle 5 pool bound to prothrombin.
Aptamer Target Affinity and Cycles of Selection-Much of the in vitro selection method is based on the assumption that higher affinity aptamers should progressively replace lower affinity aptamers as the number of selection cycles increase. Our isolation of a large number of prothrombin binding aptamers provided us with an opportunity to determine whether this assumption was correct under the conditions of our experiment.
Aptamers assigned to F1 and F15-18 show a region of high homology characterized by a series of G-rich repeats. Sequences flanking this motif are variable, allowing classification of F1 into subfamilies. G-rich regions are also present in F2, but there is little homology to the more common motif present in the other families (Fig. 4).
The binding kinetics of a subset of these aptamers to prothrombin were determined by surface plasmon resonance. Sensor chips were prepared to allow multiple aptamers to be tested on the same sensor flow cell. This was achieved by immobilizing via a streptavidin linkage, a biotinylated oligonucleotide (18-mer) complementary to either the 5Ј or 3Ј primer-binding site of each aptamer. Individual aptamers were hybridized to the linker oligonucleotide forming a partial duplex that was stable upon prothrombin binding and chip regeneration. A 100-fold dilution series of pure prothrombin was injected over the partial duplex, and the resulting sensorgrams were fitted to kinetic models assuming a 1:1 binding interaction. Experimental and mathematically derived sensorgrams were essentially superimposable, supporting the 1:1 model (Fig. 5).
Rate constants (k a and k d ) and equilibrium dissociation constants for a number of aptamers together with their frequency in the selected populations are presented in TABLE TWO. To help interpret these data, we have assigned aptamers to one of four classes based on calculated rate constants and visual examination of sensorgrams. Aptamers that appeared at early cycles of selection and remained in the population until the last cycle had fast association rates and slow dissociation rates (class I). Aptamers present at early cycles of selection but then disappearing in later populations belonged to either class II or class III. Only one aptamer (F2) was assigned to class IV (very slow dissociation rate), and it appeared at the very end of the selection. In general we found that aptamer target association rates varied over a narrow range, whereas the   OCTOBER 7, 2005 • VOLUME 280 • NUMBER 40 range of dissociation rates was significantly greater. It is noteworthy that aptamers with subnanomolar dissociation constants were present when this family first appeared in the selected population.

In Vitro Selection Using Complex Targets
Selection Dynamics after Dominant Target Depletion-Our observation of the emergence of dominant aptamer families in a screen using a mixture of targets is consistent with what has been reported by others (18,19,21). How this occurs remains a puzzle given the concentration range and diversity of targets. Regardless of the mechanism, it is a problem that has to be overcome when trying to isolate aptamers against as many targets as possible.
In practice there are two obvious ways to prevent narrowing of the selected population because of dominant aptamer families. One is to focus on a more thorough analysis of early rounds of selection before the aptamer population becomes skewed. The problem with early round pools is the high frequency of low or nonspecific binding aptamers. The other is to deplete the complex mixture of targets recognized by dominant aptamer families. Our results showed that aptamers bind their targets with high affinity and specificity, making them ideal reagents for this purpose (Fig. 3).
To explore the effectiveness of specific target depletion and how this might influence the dynamics of aptamer isolation, we made use of the prothrombin-binding aptamer F1.06. Plasma was depleted of prothrombin by simple aptamer-mediated affinity chromatography and then coupled to activated Sepharose. The aptamer library screen was then performed as before, except that the selection was stopped after eight cycles. Individual aptamers from cycles 5 and 8 were sequenced, and their targets were identified by SDS gel electrophoresis.
The prothrombin-depleted plasma screen was performed twice to see whether the changes we observed were reproducible. The data show a dramatic decline in the number of prothrombin-binding aptamers isolated, such that they were absent in the cycle 8 population of both screens. It is important to note that in these experiments a low level of prothrombin was still present in the depleted plasma. It would appear from these results that there is no need to remove it completely.
The two prothrombin-depleted plasma screens gave qualitatively the same results (Fig. 6, B and C). The same aptamer families were isolated in both screens, but their relative representation in the selected populations differed. This is not surprising given the stochastic nature of the in vitro selection process. Interestingly, these families appeared earlier in the selection with the depleted plasma than they did in the whole plasma screen In the absence of prothrombin, it would appear that other aptamer families now begin to dominate the screen. In one experiment it was the C3 complement binding family (F3), and in the other it was F5. Whether this is solely due to the stochastic nature of the selection process or some other mechanism remains to be determined. It is also noteworthy that two new aptamer families (F30 and F31) not isolated in the whole plasma screen were present in the enriched populations of the depleted plasma screens. . Sequence alignments of prothrombin-binding aptamers. The sequences were aligned using the SPLASH pattern recognition algorithm. The boxed sequences define those conserved in all aptamers. The aptamer family designations are listed. Dashes in the sequence have been introduced to facilitate sequence comparisons. FIGURE 5. Kinetic analysis of aptamers binding to prothrombin. Surface plasmon resonance was used to measure the association and dissociation of aptamers across a range of prothrombin concentrations. Aptamers anchored to sensor chips via a linker oligonucleotide were exposed to prothrombin for 4 -6 min and then washed with buffer to measure dissociation. Representative sensorgrams showing the experimentally derived responses (black) and the computational fits, assuming a 1:1 aptamer to prothrombin binding interaction (red), are shown. Aptamers were assigned to different classes (Roman numerals) based on the assessment of sensorgrams and calculated rate constants. The concentration series used are as follows: F1.01, 25, 6.5, 2.5, 1, and 0.25 nM for A; F15, 25, 10, 5, 2, and 0.5 nM for B; F16, 10, 2.5, 1, 0.5, and 0.1 nM for C; and F2, 2.5, 1, 0.5, and 0.1 nM for D.

DISCUSSION
Our goal is to develop methods that extend aptamer library screening beyond the single target paradigm. Human plasma was chosen as a model for our study because of its molecular complexity and the practical challenges it presents (reviewed in Ref. 37). A number of examples have been reported where a complex target has been used for aptamer library screening (12-14, 18, 19, 21, 43-47). All of these studies followed traditional screening protocols developed for use with single pure targets where cycles of in vitro selection continue until some predetermined functional end point is reached. Because each study had a single end point there was no attempt and indeed no need to understand the dynamics of the selection process. In stark contrast, our study has as many end points as there are proteins in plasma. It was therefore critical that we obtain a detailed understanding of how the selection process works to establish whether our goal was achievable.
Using unfractionated human plasma as a target, we isolated and analyzed the binding characteristics of a large number of aptamers. Aptamers were assigned to families based on sequence motif homologies, and their protein targets were identified by SDS gel electrophoresis. Using these simple methods, we successfully identified aptamers that bound to 14 different plasma proteins. This appears a low number given the variety of potential targets in plasma, but a detailed analysis of aptamer populations at different times during the selection process revealed the reason for this result. A family of aptamers that bound prothrombin emerged midway through the screen and dominated all of the selected populations thereafter. The emergence of dominant aptamer families is the desired end point when screening libraries with single targets. However, for our purposes, narrowing the diversity of selected populations beyond a certain point is counterproductive, particularly when a family appears and then disappears with additional cycles of selection (e.g. F12).
It is a widely held view that increasing rounds of in vitro selection yield aptamers with progressively higher target binding affinities. We tested this proposal by measuring the binding characteristics of prothrombin aptamers isolated at different times during the selection process. Surprisingly, aptamers with subnanomolar binding affinities were isolated after only four cycles of selection, that is, when they first appeared in the selected population. A single aptamer family (F2) with a higher binding affinity for the same protein only appeared toward the very end of the screen. Our data suggest that large numbers of selection cycles might not be needed to obtain aptamers with high binding affinity. We need to extend these binding measurements to other aptamer families isolated in this study to see how typical this observation is.
There are several reports of in vitro selections where aptamer families  In Vitro Selection Using Complex Targets OCTOBER 7, 2005 • VOLUME 280 • NUMBER 40 have emerged to dominate a population that upon further characterization has been shown to recognize sites not present on the intended target. In many cases these aptamers bind to the support used to immobilize the desired target. In others, the intended target is present as part of a mixture that contains molecules that are preferred targets. Attempts to overcome this problem have focused on depleting the aptamer population of sequences that rapidly accumulate. In practice, solutions involve some form of negative selection and range from simply passing aptamer populations over the underivatized target support to more complicated sequence specific enzymatic methods (48). Alternatively, attempts have been made to block ligand-binding sites with nonamplifiable sequences isolated in an initial in vitro selection experiment. The success of this approach depends on a number of factors including the binding affinity of the blocking ligand and the amount of target in the mixture. In addition it has been found that blocking one site on a target does not prevent the isolation of ligands to other sites on the same protein (16). In our case we considered it more efficient to deplete the target population than to focus on the aptamer population. Our data showed that this strategy was effective as selection dynamics changed significantly as did the representation of different aptamer families within selected pools. Of note, new aptamer families were identified, and other aptamer families that previously represented only a small fraction of the selected pool now began to dominate. Although the target depletion strategy employed in this study has some advantages, it also has some limitations. First, it can only be applied to soluble molecules and not to complex systems such as cell surfaces. Second, if the dominant target is part of a protein complex, then other proteins could be depleted indirectly via their association with the complex. Based on current information we are unable to say why dominant aptamer families arise. From our work and that of others it would appear that target concentration is not a major factor. It is possible there is something special about the shape or surface of a protein analogous to the concept of immunogenicity that might differentiate one protein from another. Alternatively, the appearance of dominant families might have nothing to do with the target but might be determined by the relative abundance of a particular shape in the unselected aptamer library.
Ours is the first study we are aware of that has examined whether the traditional in vitro selection strategy is appropriate for deconvoluting a complex mixture. We conclude based on the data reported in this manuscript that it probably is not. Our results have identified a number of areas where further work is required to better understand how certain variables influence the outcome of an in vitro selection experiment. We need to establish why certain aptamer families come to dominate selected populations and how their influence on subsequent cycles of selection can be minimized. What role does target concentration within a complex mixture play in successful aptamer isolation? How many high affinity binding aptamers are present in populations obtained in the early rounds of selection? Answers to these and several other questions arising from this study should lead to the development of more efficient and effective library screening protocols.