Quantitative selection and parallel characterizationof aptamersMinseon Choa,b,1, Seung Soo Ohb,1, Jeff Niec,2, Ron Stewartc, Michael Eisensteina,b, James Chambersd,Jamey D. Marthd,e, Faye Walkerf, James A. Thomsonc,d,g,3, and H. Tom Soha,b,3

Departments of aMechanical Engineering, dMolecular, Cellular, and Developmental Biology, fChemistry, and bMaterials Department, University of California,Santa Barbara, CA 93106; cMorgridge Institute for Research, Madison, WI 53715; eSanford-Burnham Medical Research Institute, La Jolla, CA 92037; andgDepartment of Cell and Regenerative Biology, School of Medicine and Public Health, University of Wisconsin-Madison, Madison, WI 53706

Contributed by James A. Thomson, August 23, 2013 (sent for review February 8, 2013)

Aptamers are promising affinity reagents that are potentially wellsuited for high-throughput discovery, as they are chemically syn-thesized and discovered via completely in vitro selection processes.Recent advancements in selection, sequencing, and the use of mod-ified bases have improved aptamer quality, but the overall processof aptamer generation remains laborious and low-throughput. Thisis because binding characterization remains a critical bottleneck,wherein the affinity and specificity of each candidate aptamer aremeasured individually in a serial manner. To accelerate aptamerdiscovery, we devised the Quantitative Parallel Aptamer SelectionSystem (QPASS), which integrates microfluidic selection and next-generation sequencing with in situ-synthesized aptamer arrays,enabling simultaneous measurement of affinity and specificity forthousands of candidate aptamers in parallel. After using QPASS toselect aptamers for the human cancer biomarker angiopoietin-2(Ang2), we in situ synthesized arrays of the selected sequences andobtained equilibrium dissociation constants (Kd) for every aptamerin parallel. We thereby identified over a dozen high-affinity Ang2aptamers, with Kd as low as 20.5 ± 7.3 nM. The same arrays enabledus to quantify binding specificity for these aptamers in parallel bycomparing relative binding of differentially labeled target and non-target proteins, and by measuring their binding affinity directly incomplex samples such as undiluted serum. Finally, we show thatQPASS offers a compelling avenue for exploring structure−functionrelationships for large numbers of aptamers in parallel by couplingarray-based affinity measurements with next-generation sequenc-ing data to identify nucleotides and motifs within the aptamer thatcritically affect Ang2 binding.

Since the invention of hybridoma monoclonal antibodies (1),affinity reagents that bind specifically to target molecules

have become a cornerstone of modern biotechnology. The de-mand for high-quality affinity reagents has increased sharply asresearchers continue to discover remarkable diversity within theproteome, with many closely related but functionally distinctprotein variants arising through processes such as alternativesplicing and posttranslational modification (2, 3). Unfortunately,conventional methods of generating monoclonal antibodies (e.g.,hybridoma production) are inherently low-throughput, and thereis a pressing need for alternative technologies that can generatehigh-quality affinity reagents to meet this growing demand in ahigh-throughput and economical manner (4). Toward this end,nucleic acid aptamers represent a compelling class of affinityreagents. Aptamers are chemically synthesized and their discoveryis performed in vitro rather than relying on in vivo biological pro-cesses, making them potentially well suited for high-through-put discovery (5, 6). Furthermore, aptamers with high affinityand specificity have been previously reported for a wide rangeof molecular targets including proteins (7, 8), small molecules (9),and cell surface markers (10, 11), and chemically modified andnonnatural nucleic acids have introduced an expanded repertoireof functional groups and structural conformations that can fur-ther improve binding affinity and specificity (12).

Recent advancements in selection and sequencing techniqueshave greatly increased the efficiency of aptamer discovery. Forexample, novel separation methodologies such as capillary elec-trophoresis (13, 14) and microfluidic devices (15–17) can iso-late high-affinity aptamers after a minimal number of selectionrounds. This increased efficiency is achieved through tightercontrol over selection stringency by applying rigorous washing toisolate aptamers with slow off-rates or by using minimal targetquantities to establish a highly competitive binding environment.Furthermore, next-generation sequencing (NGS) has enabledanalysis of far greater numbers of aptamer sequences than thoseobtained via standard bacterial cloning techniques. Traditionalselection techniques depend on the convergence of the aptamerpool toward a relatively small number of sequences, such thatconsensus motifs can be obtained from a few hundred sequences.NGS can analyze millions of sequences, thus identifying enrichedsequences much earlier in the selection process; this reducesthe number of selection rounds required, and enables theidentification and elimination of artifacts that typically ariseover repeated rounds of PCR or during the cloning process(18–20). Despite these advances, the generation of high-qualityaptamers remains a time-consuming and low-throughput pro-cess. This is because aptamer binding is still characterized ina serial fashion, with the affinity and specificity of each candidateaptamer measured individually. Thus, to truly accelerate aptamer

Significance

The comprehensive functional mapping of the human pro-teome will require access to high-quality affinity reagents thatspecifically bind to their respective proteins with high affini-ties. Unfortunately, currently available antibodies can onlytarget a small fraction of the proteome, and their affinity andspecificity can vary considerably for each protein. Thus there isan urgent need for novel technologies capable of generatingalternative, synthetic affinity reagents in a scalable and eco-nomical manner. Toward this end, we report a unique screeningsystem (termed the “Quantitative Parallel Aptamer SelectionSystem”) that can accelerate discovery of high-quality aptamerreagents by enabling simultaneous measurements of bindingaffinity (Kd) and specificity for thousands of aptamers in parallel.

Author contributions: M.C., S.S.O., and H.T.S. designed research; M.C., S.S.O., and J.C.performed research; M.C., S.S.O., J.N., R.S., J.C., J.D.M., and J.A.T. contributed new re-agents/analytic tools; M.C., S.S.O., J.N., R.S., J.C., J.D.M., J.A.T., and H.T.S. analyzed data;and M.C., S.S.O., M.E., F.W., and H.T.S. wrote the paper.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.1M.C. and S.S.O. contributed equally to this work.2Present address: Division of Biomedical Statistics and Informatics, Department of HealthSciences Research, Mayo Clinic College of Medicine, Rochester, MN 55905

3To whom correspondence may be addressed. E-mail: tsoh@engr.ucsb.edu or jthomson@morgridgeinstitute.org.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1315866110/-/DCSupplemental.

18460–18465 | PNAS | November 12, 2013 | vol. 110 | no. 46 www.pnas.org/cgi/doi/10.1073/pnas.1315866110

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

discovery, it is necessary to alleviate this bottleneck by devisinga strategy that enables binding characterization of large numbersof aptamers in a massively parallel manner.As a solution to this problem, we describe an integrated

aptamer discovery platform that enables simultaneous mea-surement of affinity and specificity for thousands of aptamers(Fig. 1). Our Quantitative Parallel Aptamer Selection System(QPASS) integrates microfluidic aptamer selection (M-SELEX)and NGS to identify potential high-affinity aptamers, which arethen synthesized in situ on an aptamer array for binding charac-terization. Previous studies have used aptamer arrays to explorethe relationship between sequence and function for existingaptamers (21, 22). Here we integrated selection and NGS todiscover unique aptamers by measuring the binding affinityand specificity of a large numbers of aptamers in parallel.After performing four rounds of M-SELEX and NGS analysiswith our model target, human angiopoietin-2 (Ang2), we usedour aptamer arrays to simultaneously measure equilibriumdissociation constants (Kd) for ∼1,000 candidate aptamers in par-allel, identifying six high-affinity Ang2 aptamers with Kd < 30 nM.In addition, these aptamer arrays also allowed us to quantify therelative binding specificity of our aptamers for Ang2 relative toother, nontarget proteins via a two-color fluorescence measure-ment scheme. Finally, we used our arrays to generate binding datain undiluted serum, enabling us to specifically identify aptamerswith optimal binding characteristics in complex, clinical samples.

ResultsAptamer Library Design and Microfluidic Selection. We chose Ang2as a model target because it is an important mediator of angio-genesis and a biomarker of colon, prostate, and breast cancers(23, 24). The monomeric Ang2 used in this experiment is 66 kDain size, with an isoelectric point of 5.5 (25). Previously, 2′ fluoro-modified RNA aptamers for Ang2 have been reported by theSullenger group (7), but no DNA aptamer has been reported. Tobegin the selection, we immobilized Ang2 onto carboxylic acid-functionalized magnetic beads via ethyl(dimethylaminopropyl)carbodiimide (EDC)−NHS coupling reaction. To verify that theconformation of Ang2 was preserved on the magnetic beads, weconfirmed that binding of monoclonal anti-Ang2 antibody wasunaffected by immobilization (SI Appendix, Fig. S1). We mini-mized nonspecific binding of DNA to the bead surface by tuningthe hydroxyl passivation layer, as previously described (15).Each member of our aptamer library consists of a 40-nt ran-

dom core region flanked by two 20-nt primer-binding regionsfor PCR amplification (Methods). We performed four rounds of

M-SELEX using the micromagnetic separation (MMS) devicepreviously developed by our group (16), which allowed us touniformly control the selection and washing stringency (15–18).Although the MMS device is capable of performing highlystringent washing, we used moderate selection conditions for thisexperiment. This is because higher stringency selection generallyresults in convergence toward an aptamer pool in which a smallnumber of candidate aptamer sequences predominate, whereasthe goal of this experiment was to maintain greater sequencediversity to demonstrate massively parallel characterization oflarge numbers of aptamers (see Methods for a description ofselection conditions). After each round (R) of selection, wePCR-amplified the enriched aptamers from each pool (R1−R4)and enzymatically digested the double-stranded amplicons togenerate single-stranded DNAs for subsequent rounds of selec-tion (16). To monitor enrichment progress, we collected an ali-quot from each pool to measure the bulk Kd via a bead-basedfluorescence assay (Methods). We found that the initial library,R1, R2, and R3 pools exhibited minimal binding, but observeda noticeable increase in the affinity of the R4 pool (SI Appendix,Fig. S2). As described in the literature, pools from early roundsof selection are dominated by nonspecific and low-affinity binders,and the average affinity of the pool subsequently increases in laterrounds as these sequences are purged (26). Accordingly, by R4,the high-affinity sequences were becoming prominently enriched,making it possible to begin identifying candidate aptamers suit-able for further characterization.

Sequencing and Aptamer Array Design. To quantitatively identifyaptamer sequences that were enriched in each round of M-SELEX,we sequenced the R1−R4 pools with the Illumina GenomeAnalyzer IIx (27). Briefly, we PCR-amplified the pools at anoptimum cycle number determined by pilot PCR, and thenused the single-read Chip-Seq DNA Sample Prep Kit (Illumina)to prepare double-stranded products for sequencing. This yielded∼3 × 107 sequences from each pool (SI Appendix, Table S1). Forquality control purposes, we filtered out sequences of incorrectlength or those containing mismatches of more than one basewithin the primer-binding regions. After filtering, we trimmed theprimer-binding regions, leaving only the 40-nt core aptamer se-quence for downstream analysis. Many different approacheshave been described in the literature to discriminate high-af-finity aptamers from undesired background sequences arisingfrom selection biases. These methods include enrichment fold(18), repeating motif (8), and copy number (19, 20) analysis.We chose copy number analysis to test our capacity to directlydistinguish high-affinity aptamers from background sequencesvia parallel binding measurements. We therefore rank-orderedall sequences from each pool (R1−R4) based on their copy number,and used this information to identify subsets of putative high-affinityaptamers for further characterization.We devised an array-based strategy to simultaneously charac-

terize the affinities and specificities of candidate aptamers in par-allel. We fabricated a custom aptamer chip (Agilent Technologies;Methods) containing eight identical arrays of 15,000 features. Eacharray incorporated the 235 most highly represented aptamer can-didates from each of the four selection rounds as well as 65 controlsequences (SI Appendix, Table S2), with each sequence repre-sented by three copies to enable triplicate measurement. Agilentarrays can accommodate sequences of up to 60 nt in length, but theaptamers we selected were 80 nt in length, including 20-nt forwardand reverse primer-binding regions. We therefore excised theprimer-binding regions, which further allowed us to effectivelyscreen for aptamers that retain their function without thosesequences. To provide relief from the array surface (21), wetested every aptamer with four different linker types in thearray (SI Appendix, Table S3). We achieved best performancewith the poly-T linker, and these are the results reported below.

Fig. 1. Overview of the QPASS platform. (Left) Starting with ∼1 nmol ofrandom library (∼6 × 1014 sequences), we selected aptamers that bind Ang2in four rounds of M-SELEX. (Center) We sequenced the enriched pool fromeach round with the Illumina GAIIx NGS instrument, obtaining ∼3 × 107

sequences that we subsequently analyzed for copy number and sequencehomology. (Right) Based on these data, we generated a chip consisting ofeight identical in situ-synthesized aptamer arrays containing ∼15,000 aptamers,in which each of the 235 top candidate aptamers from every selection roundand numerous control sequences were all represented in triplicate. We usedthese identical arrays to measure the binding affinity and specificity of everycandidate aptamer in parallel, in both buffer and undiluted serum.

Cho et al. PNAS | November 12, 2013 | vol. 110 | no. 46 | 18461

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

Parallel Measurement of Affinity and Specificity. Our chip contains120,000 in situ-synthesized features. To use this chip to measurethe binding affinity of each candidate aptamer in parallel, wedivided the chip into eight identical arrays and incubated eacharray with a different concentration of fluorescently labeled Ang2.Then, we obtained the fluorescence signal from every singleaptamer feature at each different Ang2 concentration (Fig. 2A) toconstruct a binding isotherm, which allowed us to derive Kd valuesfor every sequence simultaneously. Specifically, we incubatedeach array with 5, 10, 25, 50, 75, 100, 150, or 200 nM Alexa Fluor647-labeled Ang2. After washing and drying, we used an arrayscanner to measure the fluorescence intensity (excitation = 649nm, emission = 666 nm) from every feature. We averaged thetriplicate signals from each aptamer candidate and used these datato calculate Kd values. We assumed a Langmuirian binding iso-therm and used the equation Y = Bmax × X/(Kd + X) (28), where Yis the net fluorescence intensity at each concentration, X is theconcentration of fluorescently labeled Ang2, and Bmax is the netfluorescence intensity at saturation. We discarded sequenceswhose Bmax was less than double the background and thus iden-tified 60 aptamers (SI Appendix, Table S4).To illustrate how the fluorescence data from each array cor-

relate with target binding, we present graphically rearrangedfluorescence images for the six aptamers with the lowest Kds(Fig. 2B), and the associated binding isotherms are plotted inFig. 2C. All six aptamers displayed high affinities, with Kds < 30nM; R4-002 exhibited the highest binding affinity, with a Kd of20.5 ± 7.3 nM. We also validated the measured affinities ofthese six aptamers by quantifying Kd with an alternative,magnetic bead-based fluorescence assay (Methods). This assayyielded similar Kd values to those obtained from the array,with Kd measurements differing only by 1.0–2.1-fold (SI Appendix,Fig. S3).

QPASS also makes it possible to monitor the evolution ofaptamer affinity such that potential “winners” can be identi-fied in early rounds of selection before the pool has convergedto a small number of sequences. For instance, after incubatingthe array with 50 nM fluorescently labeled Ang2, we plottedthe fluorescence signal from the 235 sequences with highestcopy numbers from each pool (Fig. 2D). As expected, multiplesequences from R4 displayed high fluorescence intensities,while almost every sequence from R1−R3 showed negligiblebinding to Ang2. We randomly selected six sequences fromR1−R3 and measured their binding to Ang2, which verified theirminimal affinity (SI Appendix, Fig. S4). Interestingly, the onenotable exception was a high-affinity sequence from R3 (R3-006)that turned out to be identical with R4-002, the highest-affinityaptamer, indicating that this particular sequence had already be-come enriched within three rounds of low-stringency selection.Importantly, our arrays can be used to identify aptamers that

exhibit optimal binding in complex sample conditions. As proofof principle, we identified aptamers with high affinity for Ang2 inundiluted FBS. Specifically, we spiked 50 nM Alexa Fluor 647-labeled Ang2 into FBS and incubated the sample on the array for1 h at room temperature. To ensure a high signal-to-noise ratio,we then disassembled the chip and rinsed it overnight (Methods).After background subtraction, we found eight aptamers (R4-002,-033, -039, -065, -086, -145, -201, and -227) with high affinity inserum (Fig. 3A). Interestingly, all 8 belonged to the group of 18aptamers that displayed the highest target affinities in buffer(Fig. 2D). We therefore believe that our aptamer arrays canreadily be used to identify aptamers that perform optimally ina variety of biological samples and assay conditions, an importantconsideration given that aptamer affinity and specificity can varyconsiderably depending on sample properties (e.g., salt concen-tration, pH, and temperature) (29).

Fig. 2. Parallel affinity measurement using aptamer arrays. (A) Fluorescence micrograph of eight identical arrays incubated with different concentrations ofAlexa Fluor 647-labeled Ang2. (B) Spot images of six aptamer features from R4 after binding with varying concentrations of Alexa Fluor 647-labeled Ang2. Kd

values were obtained from these scanned images. (C) Equilibrium binding curves for the six R4 aptamers with highest affinity for Ang2. Aptamer R4-002exhibited the lowest Kd value (Kd = 20.5 ±7.3 nM). (D) Fluorescence intensities obtained from the top 235 aptamer candidates from each selection round,sorted by copy number from highest to lowest. Fluorescence intensities were extracted from the scanned image of the array incubated with 50 nM Alexa Fluor647-labeled Ang2. Fluorescence intensities are the average of triplicate measurements. Eighteen aptamers from R4 showed a strong binding signal; however,all molecules from R1−R3 showed negligible binding with the exception of a single sequence (R3-006) that was also isolated in R4 (R4-002).

18462 | www.pnas.org/cgi/doi/10.1073/pnas.1315866110 Cho et al.

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

Finally, we used our aptamer arrays to quantify the specificityof these aptamers for Ang2 relative to other nontarget proteinsin parallel, using a two-color fluorescence scheme. As an ex-ample, we quantified the specificity of our Ang2 aptamers withBSA, the most abundant plasma protein (Fig. 3B). We incubatedthe aptamer array for 1 h at room temperature with 50 nM AlexaFluor 647-labeled Ang2 spiked into a solution of 1 μM AlexaFluor 555-labeled BSA. After washing and drying, we imagedthe array using two different excitation wavelengths: 649 nm forAlexa Fluor 647 (Ang2), and 532 nm for Alexa Fluor 555 (BSA).Although the BSA concentration was 20-fold higher, Ang2 (Fig.3B, Left) generated a far stronger fluorescence signal than BSAfor all aptamers (Fig. 3B, Right). The relative fluorescence inten-sities from Ang2 and BSA are plotted in Fig. 3C. From these data,we calculated each aptamer’s specificity in terms of its fluorescenceratio (i.e., Ang2 (F647)/BSA (F555)) and identified 11 aptamers(R4-002, -033, -039, -048, -086, -110, -119, -145, -159, -201, and-227) that showed the highest specificity (SI Appendix, Fig. S5).In this way, our aptamer arrays offer the capacity to quantifyspecificity in parallel and with far greater ease than conventionalmethods, which require collection of individual binding measure-ments for each aptamer.

DiscussionIn this work, we describe a technology platform that acceleratesdiscovery of high-quality aptamers by integrating microfluidicselection and NGS analysis with aptamer arrays that enable

parallel characterization of affinity and specificity for largenumbers of sequences. We show that the QPASS platformovercomes the limitations inherent to conventional aptamerdiscovery approaches. Microfluidic selection enables the useof extremely small amounts of target with controlled and re-producible imposition of selection pressures. NGS allows us toanalyze vastly larger number of molecules (∼107 sequences) com-pared with conventional cloning-based methods (∼102 sequences),such that putative high-affinity aptamers that become enrichedover the course of selection can be identified earlier and withoutfully converging the enriched pool. This is important, becauseincreasing the number of selection rounds introduces and ampli-fies undesirable biases that may confound the discovery of target-specific sequences (20, 30). Most importantly, the aptamer arraysmake it possible to perform quantitative measurement of bothaffinity and specificity for thousands of aptamer candidatessimultaneously, even in complex samples such as undiluted serum.We have experimentally verified that QPASS is not only morerapid than conventional SELEX but also generates higher-qualityaptamers. We performed 15 rounds of SELEX for Ang2 usingnitrocellulose membranes. We obtained 94 sequences (SI Ap-pendix, Methods), and determined that the Kd of the best aptamerwas 85.6 ± 22.0 nM (SI Appendix, Fig. S6). In contrast, QPASSgenerated an aptamer with ∼400% higher affinity (Kd = 20.5 ±7.33 nM) within four rounds.Moreover, the NGS data coupled with the affinity data obtained

from QPASS enable bioinformatics analyses that can identify

Fig. 3. Measurement of aptamer specificity andaffinity in the presence of nontarget proteins. (A)Measurement of aptamer affinity in undiluted FBS.We quantified fluorescence intensities of our top235 R4 aptamer candidates after binding to 50 nMAlexa Fluor 647-labeled Ang2 in undiluted serum.After subtracting background fluorescence, eightaptamer features (R4-002, -033, -039, -065, -086,-145, -201, and -227) exhibited high fluorescenceintensities, indicating that these aptamers maintaintheir affinity to Ang2 in FBS. (B) Quantitative spec-ificity measurement using a two-color scheme. Wetested the specificity of the R4 aptamers by bindingagainst Alexa Fluor 647-labeled Ang2 in the presenceof an excess of BSA labeled with Alexa Fluor 555.Compared with Ang2 (Left), we observed minimalsignal from BSA binding (Right). (C) Quantitativecomparison of target versus nontarget binding. Thefluorescent signals from Ang2 (red) in the presenceof 20-fold excess BSA (green) closely resemble thepatterns observed for binding against Ang2 in bufferwithout BSA.

Cho et al. PNAS | November 12, 2013 | vol. 110 | no. 46 | 18463

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

structure and sequence elements that are critical for targetbinding. As an example, we analyzed a consensus group con-sisting of 25 sequences from the R4 pool, which included ourhighest-affinity aptamer (R4-002) but generally exhibited abroad range of affinities with Kds ranging from 20.5 to 152.1nM (SI Appendix, Table S5). We used these binding and se-quence data to explore how one- or two-nucleotide differencesin the consensus motif affect the overall affinity of the aptamer.To do so, we used mfold software (31) to predict the secondarystructure of R4-002 and identified a protruding stem and loopregion that contains highly conserved sequence elements,implying that these regions may be critical determinants ofaptamer affinity. We subsequently found specific bases in thesestem and loop regions that appear to be important for bindingaffinity, and have graphically summarized how base changes atthese positions affect overall binding affinities in SI Appendix,Fig. S7. Notably, two thymines (T) at the 20th and 26th posi-tions within the stem-loop region were especially critical foraffinity, and their mutation to guanine (G) resulted in a sev-enfold decrease of binding affinity. In this way, QPASS offersan effective means of linking sequence variations with changesin binding affinity, a capability that could be extended to high-throughput investigations of other nucleic acid structure−functionrelationships. Although previous work with immobilized aptamershas shown that qualitative changes in fluorescence intensity on thearray are commensurate with differences in binding affinity (21,22), we were concerned that aptamers identified in our array maynot perform equally well in free solution. We therefore validatedthe affinity of our aptamers using a bead-based fluorescent assay.As shown in SI Appendix, Fig. S3, the Kd values obtained for thesefree aptamers were in close agreement with those obtained withthe array, demonstrating that the array can accurately predictbinding performance in solution.We have identified a number of opportunities for further im-

proving the performance of QPASS in the future. For one, wefound that the affinities of the selected aptamers were compro-mised by the limitations of the array format, which required us toeliminate primer-binding regions from the selected molecules.These DNA segments are known to play an important role inaptamer folding (32), and we confirmed the importance of thesesegments in four of our most highly represented sequences fromR4. Three of these four aptamers exhibited higher affinities whenmeasured as full-length 80-nt molecules relative to just the 40-ntcore segment (SI Appendix, Fig. S8). On the other hand, elimi-nating the primer-binding sequences from the aptamers on ourarray allowed us to identify the shorter, core sequences that con-tribute to binding. Nevertheless, implementation of a shorter op-timized library or array synthesis platforms compatible with longersequences should lead to discovery of higher-affinity aptamers.Although we have demonstrated the capacity of QPASS to mea-sure aptamer specificity with BSA as a nontarget protein, futureimplementation of QPASS would benefit greatly from in-corporation of counterselection steps with closely related proteinsto maximize aptamer specificity and eliminate cross-recognition ofstructurally similar nontarget molecules. In the absence of suchcounterselection, we tested the specificity of our aptamers bymeasuring their affinity for angiopoietin-1 (Ang1), and ob-served notable binding to this closely related protein. We willtherefore be implementing additional counterselections in fu-ture iterations of QPASS. In parallel, we believe that our multi-color measurement scheme could be readily extended to a widerange of nontarget molecules (including mixtures) for more ex-tensive analysis of specificity.A key advantage of QPASS lies in the fact that because

binding measurements are performed using an in situ-synthesizedarray, the time and labor required for binding characterizationremains relatively constant regardless of the number of moleculesbeing analyzed. As an initial proof of concept, we synthesized

relatively small 15,000-feature arrays, but custom DNA arrayswith more than 1,000,000 features are already commerciallyavailable at reasonable cost to researchers, and the economicadvantages of scaling will continue to accrue as feature densitygrows. This feature of QPASS architecture will make it in-herently more cost-effective for multiplexing and characterizingaptamers for multiple targets in the future, relative to conven-tional serial processes. Just as large-scale integrated circuits havebenefited from parallel fabrication processes in accordance withMoore’s law (33), we believe parallel architectures for analysisand characterization will be a critical step for accelerating affinityreagent discovery to meet the growing demands of the post-genomic era.

MethodsMicrofluidic SELEX Targeting Human Ang2. We immobilized Ang2 (R&D Sys-tems) on the surface of micrometer-sized magnetic beads. M-270 carboxylicacid Dynabeads (Invitrogen) were activated with EDC and NHS, and targetprotein was immobilized after activation by following the manufacturer’sprocedure. Immobilized Ang2 proteins were quantified using the Nano-Orange protein quantification kit (Invitrogen).

Each member of the single-stranded DNA (ssDNA) random library includes40 randomized nucleotides flanked by two 20-nt primer-binding sequencesfor PCR (5′-AGCAGCACAGAGGTCAGATG-[40N]-CCTATGCGTGCTACCGTGAA-3′).The library was synthesized by Integrated DNA Technologies. Ang2-coatedmagnetic beads were washed with Ang2 binding buffer (20 mM Hepes,150 mM NaCl, 1 mM MgCl2, 1 mM CaCl2, pH 7.4) before each selection. Atotal of 2 × 107 protein-coated beads were used in the first round (R1), 4 × 106

beads for the second round (R2), and 1 × 106 beads for the third (R3) andfourth (R4) rounds. The ssDNA library (∼1014 molecules) was denatured byheating at 95 °C for 10 min, and cooled down to room temperature. ThisssDNA library was incubated with magnetic beads in Ang2 binding buffer for1 h at room temperature. After incubation, we diluted the complex solutioninto large volumes of Ang2 wash buffer (20 mM Hepes, 150 mM NaCl, 1 mMMgCl2, 1 mM CaCl2, 0.001% Tween-20, pH 7.4) with dilution factors of 40 (R1:1mL), 200 (R2:10 mL) or 400 (R3 and R4:20 mL). The beads were then trappedvia magnetic particle concentrator (Invitrogen) for R1 and via MMS chip forR2−R4. The diluted samples were loaded into the chip at a flow rate of 100mL/h to continuously separate protein-bound aptamers from unbound andweakly bound DNAs. Aptamer-bound beads were collected, and elutedssDNAs were amplified by PCR using forward and phosphorylated reverseprimers. The ssDNA was generated for the next round of selection by lambdaexonuclease (New England Biolabs) digestion. After four rounds of selection,each selection pool was tested for binding affinity for Ang2 with a magneticbead-based fluorescence assay.

Magnetic Bead-Based ELISA. Two-and-one-half microliters of Ang2-immobi-lized magnetic beads and 2.5 μL of Tris-coated magnetic beads were washedthree times with phosphate buffered saline with Tween-20 (PBST) buffer (10mM Na2HPO4, 1.8 mM KH2PO4, 137 mM NaCl, 2.7 mM KCl, 0.05% Tween-20,pH 7.4). Beads were blocked with 2% (wt/vol) BSA in PBST buffer on a ro-tator for 1 h at room temperature. After washing each bead solution threetimes with PBST, we split the solutions into two tubes. We incubated eachaliquot with 2 μg/mL of anti-Ang2 antibody solution (R&D Systems) in 0.5%BSA with PBST buffer, along with one untreated tube containing Ang2-immobilized magnetic beads as a negative control, for 30 min on a rotator atroom temperature. After three washes with PBST, we incubated the beadsolutions with HRP-conjugated secondary antibody (0.8 mg/mL, JacksonImmunoResearch Laboratories) diluted between 1:500 and 1:2,500, and in-cubated for 30 min on a rotator at room temperature. All bead solutionswere washed three times with PBST and then transferred to new tubes. Weadded 100 μL of 2,2′-azino-bis(3-ethylbenzothiazoline-6-sulphonic acid)(ABTS, Invitrogen) to each tube and incubated for 5–60 min for color de-velopment. To stop the color reaction, we added 100 μL of ABTS stop so-lution and transferred all samples to an ELISA plate to obtain readings foreach sample in duplicate. We used a microplate reader (Tecan) to read theOD405 for each well.

Binding Affinity Measurements for Selected Pools with a Magnetic Bead-BasedFluorescence Assay. We tested the binding affinity of the library and eachselection pool (R1−R4) for Ang2 using a fluorescence binding assay. We in-troduced carboxyfluorescein (FAM) labeling via PCR with FAM-labeled for-ward primers, and diluted the labeled library and R1–4 DNA pools to various

18464 | www.pnas.org/cgi/doi/10.1073/pnas.1315866110 Cho et al.

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

different concentrations ranging from 0 to 100 nM. These pools were heatedto 95 °C for 10 min, and then immediately cooled down in an ice bath. Two ×106 Ang2-coated magnetic beads were washed three times in Ang2 bindingbuffer and incubated with the various DNA pools at room temperature for1 h. To remove unbound DNA, the beads were washed three times withAng2 wash buffer. Bound DNAs were then eluted from the beads by heatingat 95 °C for 10 min. Released DNAs were quantified by fluorescence mea-surement using a microplate reader (excitation = 490 nm, emission = 520nm). Dissociation constants (Kd) were calculated by nonlinear fitting analysis.

Aptamer Array Design and Synthesis. We designed and ordered custom 8 ×15,000 DNA microarrays through the custom microarray program from Agi-lent. Each slide consisted of eight identical arrays of 15,000 individual features.The array design was based on the aptamer candidates identified from se-quencing. The 235 aptamer candidates with the most copies from each roundwere incorporated into the array design, along with 65 negative control DNAmolecules including primer repeats and aptamers specific for other targets(human α thrombin and PDGF-BB) (SI Appendix, Table S2). Each sequence wassynthesized in triplicate, with each molecule anchored by either a 3′ poly-T20linker or one of three other alternative linkers (SI Appendix, Table S3).

Fluorescence Labeling of Proteins. Ang2 was labeled with Alexa Fluor 647(Invitrogen) dye according to the manufacturer’s protocol. Briefly, 1 mg/mLof Ang2 (in 100 mM NaHCO3, pH 8.3) was incubated with 3 μL of Alexa Fluor647 succinimidyl ester (7.94 nmol/μL) for 1 h at room temperature. To sep-arate labeled protein from unreacted dye, we used a spin column and thesupplied purification resin with Ang2 binding buffer. The eluted conjugateswere analyzed using a NanoDrop Technologies ND-1000 spectrophotometer.We obtained a ∼95% yield of conjugate, with an 8.75 degree of labeling.The labeled protein was stored at 4 °C, protected from light. BSA (BSA) waslabeled with Alexa Fluor 555 dye according to same protocol. We obtaineda ∼95% yield of conjugate, with a degree of labeling of 9.35.

Binding and Specificity Experiments on Aptamer Array. Slides were assembledwith eight-well gaskets in an Agilent hybridization chamber for blocking and

sample incubation. Each gasket was filled with 40 μL of blocking or samplesolution. Before binding Alexa Fluor 647-labeled Ang2 to the array, weblocked the array surface for 1 h at room temperature with blocking buffer(10 mg/mL casein, 0.1% Tween-20, 10 mM Na2HPO4, 1.8 mM KH2PO4,2.7 mM KCl, 137 mM NaCl, 1 mM MgCl2, pH 7.4). Slides were disassembled innanopure H2O and then dried by centrifugation (130 × g for 2 min usinga swing bucket rotor). Different concentrations of labeled Ang2 (5, 10, 25,50, 75, 100, 150, and 200 nM) were diluted in blocking buffer and incubatedon the arrays for 1 h at room temperature. For specificity tests, 1 μM AlexaFluor 555-labeled BSA was mixed with 50 nM Alexa Fluor 647-labeled Ang2and then added to the array. Slides were disassembled in washing buffer(0.1% Tween-20, 10 mM Na2HPO4, 1.8 mM KH2PO4, 2.7 mM KCl, 137 mMNaCl, 1 mM MgCl2, pH 7.4) and rinsed three times in washing buffer withdecreasing amounts of Tween-20 (0.1%, 0.01%, and 0.001% of Tween-20).After a final wash with nanopure H2O, slides were dried by centrifugation(1,500 rpm for 2 min using a swing bucket rotor).

For binding experiments in undiluted serum, 50 nM Alexa Fluor 647-labeledAng2 was spiked into 100% FBS (Invitrogen) and then incubated on the arraysfor 1 h at room temperature. Slides were disassembled in washing buffer with0.1% Tween-20 and rinsed overnight. After a final wash with nanopure H2O,slides were dried by centrifugation (1,500 rpm for 2 min using a swingbucket rotor).

Slides were scanned using a Perkin-Elmer Proscan array HT reader at 5-μmresolution. Data were extracted from images using ImaGene 7.5 software(BioDiscovery), and dissociation constants were calculated by computationalnonlinear least squares fitting in a Microsoft Excel worksheet (28).

ACKNOWLEDGMENTS. Microfabrication was carried out in the Nanofabri-cation Facility at University of California, Santa Barbara. We are grateful forthe financial support from the Otis Williams Foundation, The CharlotteGeyer Foundation, California Institute of Regenerative Medicine, NationalInstitutes of Health (U01 HL099773, U54 DK093467, CA71932, and DK48247),and the Institute of Collaborative Biotechnologies through the Army Re-search Office (W911NF-09-0001).

1. Köhler G, Milstein C (1975) Continuous cultures of fused cells secreting antibody ofpredefined specificity. Nature 256(5517):495–497.

2. Stoevesandt O, Taussig MJ (2007) Affinity reagent resources for human proteomedetection: Initiatives and perspectives. Proteomics 7(16):2738–2750.

3. Wark KL, Hudson PJ (2006) Latest technologies for the enhancement of antibodyaffinity. Adv Drug Deliv Rev 58(5-6):657–670.

4. Ramachandran N, Srivastava S, Labaer J (2008) Applications of protein microarrays forbiomarker discovery. Proteomics Clin Appl 2(10-11):1444–1459.

5. Keefe AD, Pai S, Ellington A (2010) Aptamers as therapeutics. Nat Rev Drug Discov9(7):537–550.

6. Nimjee SM, Rusconi CP, Sullenger BA (2005) Aptamers: An emerging class of thera-peutics. Annu Rev Med 56:555–583.

7. White RR, et al. (2003) Inhibition of rat corneal angiogenesis by a nuclease-resistantRNA aptamer specific for angiopoietin-2. Proc Natl Acad Sci USA 100(9):5028–5033.

8. Bock LC, Griffin LC, Latham JA, Vermaas EH, Toole JJ (1992) Selection of single-stranded DNA molecules that bind and inhibit human thrombin. Nature 355(6360):564–566.

9. Huizenga DE, Szostak JW (1995) A DNA aptamer that binds adenosine and ATP.Biochemistry 34(2):656–665.

10. Lupold SE, Hicke BJ, Lin Y, Coffey DS (2002) Identification and characterization ofnuclease-stabilized RNA molecules that bind human prostate cancer cells via theprostate-specific membrane antigen. Cancer Res 62(14):4029–4033.

11. McNamara JO II, et al. (2006) Cell type-specific delivery of siRNAs with aptamer-siRNAchimeras. Nat Biotechnol 24(8):1005–1015.

12. Gold L, et al. (2010) Aptamer-based multiplexed proteomic technology for biomarkerdiscovery. PLoS ONE 5(12):e15004.

13. Mendonsa SD, Bowser MT (2004) In vitro evolution of functional DNA using capillaryelectrophoresis. J Am Chem Soc 126(1):20–21.

14. Berezovski M, et al. (2005) Nonequilibrium capillary electrophoresis of equilibriummixtures: A universal tool for development of aptamers. J Am Chem Soc 127(9):3165–3171.

15. Lou X, et al. (2009) Micromagnetic selection of aptamers in microfluidic channels. ProcNatl Acad Sci USA 106(9):2989–2994.

16. Oh SS, et al. (2011) Improving aptamer selection efficiency through volume dilution,magnetic concentration, and continuous washing in microfluidic channels. Anal Chem83(17):6883–6889.

17. Ahmad KM, et al. (2011) Probing the limits of aptamer affinity with a microfluidicSELEX platform. PLoS ONE 6(11):e27051.

18. Cho M, et al. (2010) Quantitative selection of DNA aptamers through microfluidic se-lection and high-throughput sequencing. Proc Natl Acad Sci USA 107(35):15373–15378.

19. Hoon S, Zhou B, Janda KD, Brenner S, Scolnick J (2011) Aptamer selection by high-throughput sequencing and informatic analysis. Biotechniques 51(6):413–416.

20. Schütze T, et al. (2011) Probing the SELEX process with next-generation sequencing.PLoS ONE 6(12):e29604.

21. Katilius E, Flores C, Woodbury NW (2007) Exploring the sequence space of a DNAaptamer using microarrays. Nucleic Acids Res 35(22):7626–7635.

22. Fischer NO, Tok JB, Tarasow TM (2008) Massively parallel interrogation of aptamersequence, structure and function. PLoS ONE 3(7):e2720.

23. Tsigkos S, Koutsilieris M, Papapetropoulos A (2003) Angiopoietins in angiogenesisand beyond. Expert Opin Investig Drugs 12(6):933–941.

24. Hu B, Cheng S-Y (2009) Angiopoietin-2: Development of inhibitors for cancer therapy.Curr Oncol Rep 11(2):111–116.

25. Kim I, et al. (2000) Characterization and expression of a novel alternatively splicedhuman angiopoietin-2. J Biol Chem 275(24):18550–18556.

26. Wang JP, Rudzinski JF, Gong Q, Soh HT, Atzberger PJ (2012) Influence of targetconcentration and background binding on in vitro selection of affinity reagents. PLoSONE 7(8):e43940.

27. Mardis ER (2008) Next-generation DNA sequencing methods. Annu Rev GenomicsHum Genet 9:387–402.

28. Kemmer G, Keller S (2010) Nonlinear least-squares data fitting in Excel spreadsheets.Nat Protoc 5(2):267–281.

29. Hianik T, Ostatná V, Sonlajtnerova M, Grman I (2007) Influence of ionic strength, pHand aptamer configuration for binding affinity to thrombin. Bioelectrochemistry70(1):127–133.

30. Zimmermann B, Gesell T, Chen D, Lorenz C, Schroeder R (2010) Monitoring genomicsequences during SELEX using high-throughput sequencing: Neutral SELEX. PLoS ONE5(2):e9169.

31. Zuker M (2003) Mfold web server for nucleic acid folding and hybridization pre-diction. Nucleic Acids Res 31(13):3406–3415.

32. Legiewicz M, Lozupone C, Knight R, Yarus M (2005) Size, constant sequences, andoptimal selection. RNA 11(11):1701–1709.

33. Moore GE (1965) Cramming more components onto integrated circuits. Electronics38:114–117.

Cho et al. PNAS | November 12, 2013 | vol. 110 | no. 46 | 18465

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021