A single-molecule barcoding system using nanoslitsfor DNA analysisKyubong Jo*†, Dalia M. Dhingra*†, Theo Odijk‡, Juan J. de Pablo§, Michael D. Graham§, Rod Runnheim*†, Dan Forrest*†,and David C. Schwartz*†¶

*Laboratory for Molecular and Computational Genomics, Laboratory of Genetics, and Biotechnology Center, University of Wisconsin,425 Henry Mall, Madison, WI 53706; †Department of Chemistry, University of Wisconsin, 1101 University Avenue, Madison, WI 53706;‡Complex Fluids Theory, Faculty of Applied Sciences, Delft University of Technology, Julianalaan 67, 2628 BC Delft, The Netherlands;and §Department of Chemical and Biological Engineering, University of Wisconsin, 1415 Engineering Drive, Madison, WI 53706

Communicated by David E. Housman, Massachusetts Institute of Technology, Cambridge, MA, December 14, 2006 (received for review November 12, 2006)

Molecular confinement offers new routes for arraying large DNAmolecules, enabling single-molecule schemes aimed at the acqui-sition of sequence information. Such schemes can rapidly advanceto become platforms capable of genome analysis if elements of anascent system can be integrated at an early stage of development.Integrated strategies are needed for surmounting the stringentexperimental requirements of nanoscale devices regarding fabri-cation, sample loading, biochemical labeling, and detection. Wedemonstrate that disposable devices featuring both micro- andnanoscale features can greatly elongate DNA molecules whenbuffer conditions are controlled to alter DNA stiffness. Further-more, we present analytical calculations that describe this elonga-tion. We also developed a complementary enzymatic labelingscheme that tags specific sequences on elongated molecules withindescribed nanoslit devices that are imaged via fluorescence reso-nance energy transfer. Collectively, these developments enablescaleable molecular confinement approaches for genome analysis.

DNA labeling � genomics � nanofabrication � polymer confinement

Approaches using single molecules are poised to radicallyalter the way most biological researchers conceive, perform,

and analyze experiments. The use of single molecules as analytesrepresents absolute miniaturization and fosters new measure-ment schemes, allowing for a more comprehensive analysis. Incomparison with traditional bulk measurement schemes, exper-iments using single molecules are direct and rapid, promising thegeneration of thousands of data points in a short time period. Assuch, the entry of single-molecule analytes into mainstreamgenomic investigation requires the development of high-throughput platforms, which in turn must effectively fuel newbiological insights. To meet this challenge, integration of exper-imental or analytical components into working systems is acritical barrier because developments that simply identify newsingle-molecule phenomena or create new greatly miniaturizeddevices are merely the first few steps in this process. This impliesthat research efforts truly intent on impacting the genomicsciences will likely leverage single-molecule effects through theintegration of robust devices with equally robust biochemicaland detection schemes.

Almost paradoxically, many schemes for the analysis of singleDNA molecules, such as sequencing, employ amplification steps(1, 2) using genomic DNA substrates. These steps, althoughobviating traditional clone libraries, also eliminate the principleadvantage of single-molecule techniques for which measure-ments describe individual molecules, free from any conse-quences of an ensemble. Thus, new single-molecule approachesaimed at genomic analysis require a high level of precision andthroughput before they become tools for research.

When large genomic DNA molecules are the primary analyte,how random coils are unraveled for high-resolution detection ofmarkers (typically by fluorescence microscopy) as well as thepresentation techniques employ a variety of schemes: devices

that direct f luid flows for surface deposition of molecules (3, 4);agarose gel matrices (5, 6) or engineered nanopillar arrays (7, 8),which enable electrokinetically driven reptation or threading ofDNA molecules; and, most recently, direct nanoconfinement (9,10), where molecules are trapped in elongated configurations.During the last decade, nano- and microfabrication techniqueshave catalyzed these developments, allowing researchers tofabricate devices boasting complex geometries in both hard andsoft materials. Traditionally, nano- and microscale fabricationsfor biological research have used silicon or quartz wafers as anadaptation from microelectromechanical systems (11), but therehave been problems in integrating these materials with biologicalor genomic applications. Traditional device fabrication is tech-nologically demanding and time-consuming, requiring expensivelithographic equipment (e.g., electron beam lithography for eachnanoscale device) in a clean-room facility and later requiringcomplicated sealing approaches. Additionally, although electro-kinetic control is very common for biomolecule manipulation,the semiconductor attributes of silicon wafers favor quartzfabrication due to its insulatory properties. However, toilsomeand expensive device fabrication from quartz necessitates reuse,leading to contamination issues (12). Although nanoscale fea-tures are more rapidly rendered through molding techniques(13), the inability to produce a large number of disposabledevices within a typical laboratory environment (14) has greatlyhindered their use in high-throughput genomic applications.

To overcome the limitations of hard materials, soft lithogra-phy based on elastomer replicas (14, 15) has been developed,augmenting traditional lithography for biological applicationsusing microfluidics (12, 16, 17). Soft lithography uses molds(masters) created from silicon wafers patterned with photoresistfor producing disposable elastomeric replicas (14), most com-monly from poly(dimethylsiloxane) (PDMS), ideally suited forhigh-throughput applications. Unfortunately, PDMS elasticityportends mechanical instability, especially for nanostructures of�100 nm, which often disappear by collapsing against f latsubstrates (18).

Given these concerns, our goal was to develop a high-throughput system for genome analysis by using disposabledevices offering effective nanoscale geometries sufficient for thepresentation of large elongated DNA molecules. The idea was to

Author contributions: K.J., D.M.D., and D.C.S. designed research; K.J. and D.M.D. per-formed research; K.J., T.O., R.R., and D.F. contributed new reagents/analytic tools; K.J.,D.M.D., T.O., J.J.d.P., M.D.G., and D.C.S. analyzed data; and K.J., D.M.D., T.O., and D.C.S.wrote the paper.

The authors declare no conflict of interest.

Abbreviations: PDMS, poly(dimethylsiloxane); ddNTP, dideoxyribonucleotide; TE, Tris–EDTA.

¶To whom correspondence should be sent at the * address. E-mail: dcschwartz@facstaff.wisc.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0611151104/DC1.

© 2007 by The National Academy of Sciences of the USA

www.pnas.org�cgi�doi�10.1073�pnas.0611151104 PNAS � February 20, 2007 � vol. 104 � no. 8 � 2673–2678

BIO

PHYS

ICS

create the basis for a successor to the established opticalmapping system (19, 20), which spans entire genomes throughoverlapping ordered restriction maps created from individualgenomic DNA molecules. Our motivation for the work describedhere was advancement of whole-genome restriction mappingfrom a tool for discovery, or sequence validation, to a means foranalysis of human populations and cancer genomes. We solvedthe conundrum of how to imbue features in disposable PDMSdevices with nanoconfinement capabilities that closely parallelthose found in devices with features of �100 nm and requiringtraditional lithographic approaches.

By considering DNA as a polyelectrolyte, its enlargement interms of physical measures of size hinges on a polymer’s per-sistence length, commonly �50 nm (21). Physical properties ofDNA chains are derived from treatments modeling them asworm-like coils (22), where they exhibit both local rigidity andlong-range flexibility. From the theoretical work of Odijk,Skolnick, and Fixman (23, 24), we know that ionic strength is animportant factor governing intrachain electrostatic repulsionaffecting the persistence length of worm-like polyelectrolytecoils, estimated from the Debye–Huckel screening length. Thisview was experimentally confirmed for large DNA molecules byBaumann et al. (21), who found that the persistence length ofDNA molecules inversely varied with ionic strength, in goodagreement with the theory of Odijk, Skolnick, and Fixman.Consequently, we reasoned that low-ionic-strength conditionswould sufficiently increase DNA persistence length, allowing forlarger channels to be used for polymer confinement. Thus, whencomparable length scales are achieved (persistence length andchannel dimensions), polymer confinement regimes shift, en-abling extensive elongation of polymer chains (25) within chan-nel dimensions readily supported by standard PDMS fabricationtechniques. Accordingly, we report our findings showing thatDNA molecules are stretched up to 60% of their polymercontour length in disposable PDMS devices having 100-nm �1-�m channels under low-ionic-strength conditions; remarkably,these results are comparable with those previously obtainedunder standard buffer conditions using 30- � 40-nm channels(fused silica) employing nanoimprint or electron beam lithog-raphy (26).

Although low-ionic-strength buffers enable DNA elongationin larger nanoslits readily made of PDMS, the avoidance ofbiochemically meaningful salt concentrations causes problemsfor most DNA enzymes used for genome analysis. However,molecular confinement and DNA modification enzymes (e.g.,restriction endonucleases) are not necessarily incompatiblewhen used within standard �100-nm fabrications, as recentlydemonstrated by Riehn et al. (27). Instead, issues arise regardingthe scalability of their device as a viable platform for genomicanalysis due to the requirement that molecules must be contin-uously imaged for discernable biochemical events, greatly di-minishing potential throughput. As such, we developed a single-molecule labeling scheme obviating these concerns whileoffering distinct advantages for robust detection and integrationwithin a system for genome analysis.

Conventional hybridization techniques are not suited formarking discrete DNA molecules because presentation ap-proaches require intact, double-stranded DNA molecules afterprocessing for analysis. Because optical mapping successfullyemploys restriction enzymes for reliable placement of sequence-specific markers onto individual molecules, we reasoned that anew class of endonucleases nicking at specific sites (28) wouldalso confidently mark molecules but without unwanted double-strand cleavage. Because nicks cannot be directly discerned byfluorescence microscopy, we label these sites by nick-translationof DNA molecules in bulk solution using fluorochrome-labelednucleotides. We then counterstain DNA backbones with thebis-intercalator YOYO-1 before imaging. Specificity is enhanced

by using ligase and dideoxynucleotide blocking steps, greatlydiminishing labeling of preexisting random nicks. Finally, incor-porated fluorochrome labels support FRET detection, whichsimplifies data acquisition by requiring one laser for excitationof DNA backbones (donor) and labels (acceptor).

Here, we report a series of interlocking developments andfindings to potentiate a device design based on physical modi-fication of large single DNA molecules through simple alter-ations of solution ionic strength. This developmental stancefosters creation of usable systems for advancing genome analysis.We demonstrate proof of principle with physical maps of BACs.

Results and DiscussionMultiscale Fabrication Facilitates DNA Loading and Elongation. Pre-viously, we used soft lithography techniques to fabricate amicrochannel device designed to load and deposit large DNAmolecules onto charged glass surfaces (3). Because this devicehas proven robust for routine optical mapping, we have adaptedit here for capillary preloading of newly incorporated nanoslits(100 nm � 1 �m; Fig. 1). These PDMS devices are fabricated byfollowing standard rapid prototyping procedures (12, 17) butwith the inclusion of a dry-etch step, creating 100-nm-highfeatures (Fig. 1), used because it persists through cycles of replicamolding, fostering large-scale fabrication of devices. A photore-sist pattern of microchannels is then overlaid, creating nanoslitand microchannel features on the same silicon wafer. Thenanoslit geometry engenders nanoscale confinement conditionswhile employing simple relief-based fabrication techniques, andwider channel entrances offer simplified loading and less clog-ging. Because the relaxed dimensions of our molecules (� and T4bacteriophage DNA, Rg � 0.55 and 1.13 �m, respectively) aregreater than the slit dimensions, they are excluded from entering(29) by simple diffusion after capillary loading; an electricpotential (70 V) transports relaxed DNA coils to nanoslitentrances for subsequent elongation.

Fig. 2 shows images of electrokinetically loaded � (48.5 kb), T4(166 kb), and Escherichia coli genomic DNA molecules. For

Fig. 1. Microchannel–nanoslit device design and loading scheme. (A) Plexi-glas slide (25.4 � 76.2 mm) with a rectangular opening to which a glasscoverslip window (18 � 18 mm) is affixed with wax. The PDMS device is bondedto this window and immersed in buffer for electrokinetic loading via theindicated electrodes. Before buffer immersion, DNA is pipetted into micro-channels for capillary loading. (B) Illustration (top view) shows nanoslits(diagonal, 100 nm high � 1 �m wide) overlaid with microchannels (horizontal,3 �m � 100 �m wide). (C) Cartoon depicts relaxed and stretched DNA mole-cules occurring during electrokinetic loading within microchannels andnanoslits. (D) Photograph of silicon master mold bearing photoresist andetched features; arrows show the path of DNA molecules taken through themicrochannel and nanoslit features. (E) Scanning electron micrograph of thesilicon master shows a single nanoslit mold feature before the photoresistoverlay, conferring microchannel features. (Scale bar, 300 nm.) The upperimage shows many such nanoslit features spaced 4 �m apart (center-to-center). (Scale bar, 10 �m.)

2674 � www.pnas.org�cgi�doi�10.1073�pnas.0611151104 Jo et al.

elongation, a molecule undergoes electrophoresis within a mi-crochannel toward a nanoslit entrance, where it proceeds toenter then stretch. An E. coli genomic DNA molecule (Fig. 2 A)shows a contour length exceeding that of a nanoslit so that itsends appear relaxed within the microchannels. Here, the largestE. coli genomic DNA fragment spans the 105-�m-long nanoslitand is qualitatively sized at 432 kb, assuming a stretch of 0.60 (seethe next section); this stretch may be increased through tensionexerted by the relaxed ends. Accordingly, the correspondingfluorescence intensity profile shows that this molecule is wellstretched without detectable local folding back of DNA seg-ments, also known as hairpins (30). The total size of thismolecule, accounting for the relaxed portions, may exceed 1megabase. Such stretching is contrasted with the moleculescompletely within the nanoslits that exhibit folding, as evidencedby heightened values in their f luorescence intensity profiles.

Odijk (30) has put forward a theory of hairpins in various typesof nanochannels (circular, square, and rectangular). The B-formof DNA remains intact so that the bending energy is purelyelastic (see Eq. 3 below). The confines of a nanochannel cut offorientational and translational degrees of freedom, an entropiceffect that tightens hairpins, forcing them toward the central axisof the channel. Because of this effect, which increases the freeenergy of the hairpin markedly, the distance between hairpinsmay reach tens or hundreds of micrometers, even for a nanoslitof the dimensions used here.

Low-Ionic-Strength Effects on DNA Elongation. Diminished ionicstrength conditions increase DNA intrachain electrostatic re-pulsion (23, 24), with a concomitant increase of the polymerpersistence length, observable within a nanoslit as DNA stretch-ing. We study these effects by using dilutions of the same workingbuffer, Tris–EDTA (1� TE; 10 mM Tris base/1 mM EDTA, pH8.0, titrated with HCl; ionic strength, 8.4 mM), performed withinthe same slit geometry (100 nm � 1 �m) for all experimentspresented here. In addition to the TE buffer system, sodiumphosphate and diluted New England Biolabs buffer 4 (Materialsand Methods) were evaluated, and each demonstrated DNAstretching within nanoslits at low apparent ionic strengths (datanot shown). Fig. 2 B and C show typical images of T4 and � DNA(polymer contour lengths, 74.5 and 21.8 �m, respectively)stretched within nanoslits under low-ionic-strength conditions(0.05� and 0.01� TE), with average apparent lengths of 40.9 �8.4 �m and 13.0 � 2.4 �m, respectively. As a direct comparison,our value for the � DNA is nearly equivalent to the findings ofReisner et al. (26), who used 30- � 40-nm channels and TBEbuffer (0.5� TBE � 45 mM Tris base/1 mM EDTA/45 mM boricacid). Fig. 3 shows a plot of DNA elongation as a function of TEdilution (1/TE) reported as stretch, which is defined as theaverage apparent length, X, divided by the dye adjusted polymer

contour length, L (e.g., L � 21.8 �m for � DNA); a stretch of1.0 indicates complete elongation.

This plot (Fig. 3) shows that DNA stretch is size-independentand inversely related to salt concentration or ionic strengthaccording to persistence length calculations using Eq. 1. Thepersistence length (P) is related to the ionic strength (I) by anexpression derived by Odijk, Skolnick, and Fixman (23, 24) andinterpreted by Baumann et al. (21):

P � Po � Pel � Po �1

4�2lB� Po � 0.324I�1Å , [1]

where Po is the nonelectrostatic intrinsic persistence length dueto base stacking, Pel is the electrostatic persistence length due tointrachain repulsion, ��1 is the Debye–Huckel screening length,and lB is the Bjerrum length (�7 Å in water). Although the valueof Po varies according to environmental conditions (21), here wetake Po to be 50 nm. Accordingly, we see that stretch plateaus ata 20� dilution (effective ionic strength, 0.45 mM), with a

Fig. 2. Gallery of fluorescence micrographs shows stretched and relaxed DNA molecules within the nanoslit device after electrokinetic loading; images weretaken a few minutes after the electric field was shut off. (A) A large E. coli DNA molecule spans across the 105-�m-long nanoslit (0.01� TE buffer) showing relaxedends (circled) within abutting microchannels. (B) T4 DNA (166 kb) molecules; 0.05� TE. (C) � DNA (48.5 kb) molecules; 0.01� TE. Green lines demarcatenanoslit–microchannel interfaces; blue indicates a nanoslit, and yellow lines show integrated fluorescence intensity profiles revealing folded ends (B and C,arrows). Relaxed molecules within the microchannel regions appear as diffuse, partly out-of-focus, fluorescent balls, whereas stretched molecules present aslong, linear objects. (Scale bars, 20 �m.)

Str

etch

(X

/L)

Per

sist

ence

Len

gth

(nm

)

1/TE

A B

C

0 20 40 60 80

100

0

300

200

400

0.2

0.0

0.6

0.4

0.8

Fig. 3. DNA stretch varies with TE concentration. (A) Ionic strength variedthrough successive 5-, 10-, 15-, 20-, 50-, and 100-fold dilutions of 1� TE buffer(ionic strength, 8.4 mM). The reciprocal TE buffer concentrations vs. thestretch of � (black square) and T4 (red circle) DNA are plotted along with thecalculated persistence length (Eq. 1) using ionic strengths determined fromdilutions (solid line with open circles). A dotted line continues the calculatedpersistence length for ionic strengths at �0.5 mM, accounting for uncertain-ties associated with very low ionic strength. A dashed blue line represents afitted curve using Eq. 12. The stretch is defined by apparent length (X) dividedby the polymer contour length (L) of YOYO-1-stained DNA. Each data pointrepresents measurements from 50 to 300 molecules; error bars show standarddeviations on these means. (B and C) Fluorescence images (a combination offive separate experiments) show T4 DNA (B) and � DNA (C) at five different TEdilutions, 1.002� (0.9980), 0.102� (9.80), 0.0520� (19.2), 0.0220� (45.5), and0.0120� (83.3), with the corresponding dilution factors shown in parentheses.(Scale bar, 10 �m.)

Jo et al. PNAS � February 20, 2007 � vol. 104 � no. 8 � 2675

BIO

PHYS

ICS

corresponding persistence length of 122 nm, a value greater thanour nanoslit height (100 nm). Although we interpret observedDNA stretching to be strictly a function of persistence lengthdetermined at a given salt concentration, such lengths may notbe accurate due to additional low-ionic-strength effects that weare not taking into account here, including local DNA melting,particularly at AT-rich regions, or experimental limitations, forexample, carbonate dissolution or contamination, which altersionic strength at very low salt concentrations. A dotted line isused for describing alterations of persistence length at �0.05�TE (20� dilution) in Fig. 3.

Polymer Elongation Regimes Under Confinement. DNA elongation isgreatly enhanced when the scale of confinement geometryapproaches a value comparable to the radius of gyration (29, 31,32) or persistence length (24, 26). Previously, Brochard and deGennes (29) developed a scaling argument explaining polymerstretching under confinement within a channel, where a self-avoiding chain is considered as a series of blobs provided thatchannel dimensions are much larger than the persistence length(D �� P) and the polymer contour length is greater than D (L�� D). In their treatment, the stretch (X/L) of a polymer(persistence length P and molecule width w) in a channel (widthD1 and height D2) is given by X/L � (wP/D1D2)1/3 (10, 29, 32, 33);however, this argument is not valid for highly stretched polymers(e.g., X/L � 0.5). Instead, for highly stretched polymer chainsunder nanoconfinement, a different scaling argument (25) treatspolymer stretching within a nanochannel as a long chain de-flected by the walls of a nanochannel with the stretch given by

X�L � cos�; ��2 � D�P�2/3. [2]

Here, we wish to estimate the numerical coefficient. We intro-duce an analytical treatment based on a rescaling of the har-monic potential to the hard wall repulsions experienced by aworm-like chain confined in a slit geometry, followed by com-parison with our experimental observations (Fig. 3). The regimeof strong elongation is complicated because of the formation ofhairpins (30). Because of this, the crossover between the tworegimes is nontrivial and not understood at present.

Calculation of Stretch Within a Slit Regime. A rigorous analyticalcalculation is impossible because of the hard wall boundaryconditions previously mentioned in ref. 25. Nevertheless, anexact computation is feasible for a worm-like chain confined ina harmonic potential, V(x) (34). The Hamiltonian H in onedimension may be written as

HkBT

�12

P �0

L

ds� d2xds2� 2

�12

b �0

L

dsx2, [3]

where the first term is the scaled bending energy of the chain andthe second is V[x(s)], which simulates the confinement. Theposition of the chain is x(s) at contour distance s from one end.Statistical mechanics applied to Eq. 3 may be shown to lead toa Gaussian distribution for the worm (as L 3 �)

Gx� �1

�2 exp��x2

d2� ; �x2 �12

d2. [4]

In this way, we may eliminate the dummy variable b. From theorientation– translation distribution derived by Burkhardt (35),one may obtain (in one dimension)

�x2 � 2�3/2b�3/4P�1/4 �12

d2f b3/4 � 2�1/2d�2P�1/4, [5]

��2 � 2�3/2b�1/4P�3/4 � 2�4/3 �dP�

2/3

, [6]

and

Ftot�kBT � 2�12 � b

P�1/4

f Fconf �3

25/2 � bP�

1/4

. [7]

One has to subtract the energy from the external field to retainonly the configurational free energy Fconf, which is entropic inorigin [Fconf � (3/4)Ftot; for a detailed discussion, see appendix1 in ref. 36]. After eliminating b, we obtain

Fconf �3

28/3 d�2/3P�1/3kBT . [8]

Burkhardt (34) also computed Fconf for a worm in a hard slit, butnumerically,

Fconf � A rec

kBTP1/3 A�2/3 � B�2/3� , [9]

where Arec � 1.1036.

Rescaling. If we now assume that a hard wall may be simulated bya Gaussian, then we must identify d via the respective freeenergies (Eqs. 8 and 9) in two dimensions x, y:dx, dy. For instance,in the x direction, we have

dx2/3 � � 3

28/3Arec� A2/3. [10]

Hence, we then insist on

��x2 �

316Arec

� AP�

2/3

. [11]

Accordingly, we arrive at the following approximate expressionfor the relative elongation:

XL

� �cos� � 1 �12

��2 � 1 �12

��x2 �

12

��y2

� 1 � 0.085 � � AP�

2/3

� � BP�

2/3 , [12]

where A � 1,000 nm and B � 100 nm in our experimentalcondition. Of course, as �cos� diverges further away from unity,it becomes more approximate because we assume there is noback-folding. We note that the numerical coefficient in Eq. 12 isremarkably low. It implies that the formation of hairpins must bedifficult even when A and B are of the order of the persistencelength. This corroborates the computations of the high hairpinfree energies under the same conditions (30). A fit of Eq. 12shows good agreement with the experimental observations inFig. 3, discounting TE buffer dilutions below 1/TE � 20, wheremeasured stretch values plateau, as attributed to the previouslystated experimental conditions associated with very low ionicstrength effects. We also point out that the effect of theelectrostatic interaction between the DNA and the negativelycharged walls of the nanoslit in our expansion for the stretch Eq.12 is quite negligible. Although the Debye scaling length at1/TE � 20 is a substantial 15 nm long and should result indecreasing the height B to an effective value of �70 nm, thewidth A is altered little, if at all. The stretch is then predicted toincrease by a few more percentage points because of electro-statics, which is well within the experimental margin of errordisplayed in Fig. 3A.

2676 � www.pnas.org�cgi�doi�10.1073�pnas.0611151104 Jo et al.

Equilibration of DNA Stretch. DNA molecules electrokineticallymove through the microchannels as relaxed coils until approach-ing a nanoslit entrance. There, as one end of a molecule entersthe nanoslit, it transiently elongates, adopting a conformation wedescribe as a single-ended dumbbell. The dumbbell progressivelydisappears as it threads into the nanoslit, adopting a nascent,fully confined conformation while approaching equilibrium afterthe field is shut-off. Fig. 4 shows a plot of length vs. time for DNAmolecules reaching such equilibrium, where the inset imagesshow dynamic shrinking and unfolding of T4 DNA molecules.These experiments used New England Biolabs buffer 4 at a1,000� dilution, resulting in a stretch of 0.58 (42.8 � 3.6 �m; 81molecules) (data not shown). As seen in Fig. 4, DNA moleculesexpectedly enter nanoslits with different conformations andreach equilibrated forms via different processes. For example,the apparent equilibration time constants for these molecules are7.4 and 13.7 s. These findings bear interesting semblance to arecent report in which Mannion et al. (37) investigated DNAequilibration within nanochannels under relatively high saltbuffer conditions (5� TBE; 445 mM Tris borate/10 mM EDTA).In their 100- � 100-nm nanochannels, T4 DNA molecules areinitially stretched up to 40 �m and exponentially contract downto 26.4 �m (X/L � 0.23) with a relaxation time constant of 9.3 s,whereas in our 100-nm � 1-�m nanoslits, DNA molecules underlow salt conditions start at �25 �m and then stretch up to �40�m (X/L � 0.55).

DNA Barcoding via Fluorochrome Labeling of Specific Nick Sites. Fig.5 describes a molecular barcoding approach compatible with lowsalt conditions producing sequence-specific landmarks detect-able by fluorescence microscopy. Importantly, all steps areperformed in ‘‘test tubes’’ and then loaded into the nanoslitdevice after adjustment of ionic strength. Here, a nickingenzyme, Nb.BbvCI, cleaves only cognate sites on single strandsof double-stranded molecules, made detectable by nick transla-tion using fluorochrome-labeled nucleotides. Because preexist-ing DNA nicks would produce spurious signals, such sites are

repaired or disabled by using T4 ligase and polymerase incor-poration of dideoxyribonucleotides (ddNTPs) before labeling.Because nick translation efficiently incorporates fluorochrome-labeled nucleotides, crossing of mobilized nick sites on comple-mentary DNA strands occurs and produces double-strandbreaks. We attenuate this action by the continued presence ofddNTPs in the labeling reaction mix, thus limiting the number ofnucleotides incorporated per nick site through chain termina-tion. This step also controls the size of fluorescent punctates,which would otherwise expand into each other if nucleotideincorporation were unchecked, thus diminishing the number ofdiscrete markers.

FRET Imaging of Barcoded Molecules in Nanoslits. Because labeledDNA molecules are stained with the intercalating dye YOYO-1,FRET operates with incorporated fluorochrome labels (AlexaFluor 647; FRET acceptor). Fig. 5 shows FRET detection ofbarcode features, mapping their spacing (measured in kilobasepairs) by using integrated fluorescence intensity measurementsof intervening YOYO-1 signals (intervals). Averaging of suchbarcodes from multiple molecules produces ‘‘unity-based maps’’(4) that are compared against in silico barcodes constructed fromsequence information. There is good agreement with the in silicoreferences constructed from sequence information [supportinginformation (SI) Fig. 6]. Comparisons show an average absoluterelative error of 2.79% and a pooled standard deviation of 3.34kb, demonstrating development of a DNA barcoding system (foradditional analysis, see SI Text).

Conclusion. We have shown that DNA stretch is possible inrelatively large-scale nanoslits by using low-ionic-strength buff-ers. Such polymer behavior was considered by using an analyticaltreatment explicitly dealing with slit geometries that engendermolecular nanoconfinement. These insights were leveragedthrough the synergetic effects of low-ionic-strength buffers and

Fig. 4. T4 DNA relaxation within nanoslits after electrokinetic loading. Plotsshow the relaxation kinetics (0.4 mM ionic strength, New England Biolabsbuffer 4) (see Materials and Methods) of two DNA molecules [the first (black)enters nanoslits in a folded state, and the second (red) is not significantlyfolded] gauged by apparent length measurements as a function of time; 0 scorresponds to the electric field being shut off. Six images of molecule 1 showevolution of fluorescence intensity profiles (blue traces) echoing conforma-tional and length changes obtained at �5, 0, 14, 21, 27, and 37 s, with arrowsindicating putatively folded regions flagged by increased fluorescence inten-sities. We interpret this analysis as indicating that molecule 1 enters a nanoslithighly folded, with both ends tucked in. After the field is shut off, folded armsappear, signaled by the fluorescence intensity profiles, and continue unfold-ing up to �40 s, until the entire molecule appears fully equilibrated at 50 s.Interestingly, at 21 s, the molecule shrinks (also at �5 s), then proceeds to relaxin an exponential fashion; curve fits (gray lines) of molecules 1 and 2 show timeconstants of 7.4 and 13.7 s, respectively. For further details, see SI Movie 1.

Fig. 5. Molecular barcoding scheme and maps of nicked, fluorochrome-labeled BAC molecules imaged by fluorescence microscopy. (Inset) First, DNAligation and nick translation with ddNTPs obviate inherent nicks beforebarcoding; second, Nb.BbvCI places site-specific nicks on these molecules; andlast, E. coli DNA polymerase I incorporates fluorochrome-labeled deoxyribo-nucleotides F-dCTP and F-dUTP (Alexa Fluor 647-aha-dCTP, dUTP) into nicksites. Fluorescence images of labeled DNA molecules (pseudocolored DNAbackbones are green and FRET imaged punctates red) compared with ex-pected labeling patterns (measured in kilobases) from sequence, and unity-based maps constructed from analyzed molecules (Materials and Methods)are shown for BAC79 (113.7 kb) (A), BAC150 (116.8 kb) (B), and BAC614 (82.5kb) (C). Yellow arrows orient nick translation on DNA strands. DNA moleculeswere stretched in 0.01� TE buffer within described nanoslits. (Scale bar,10 �m.)

Jo et al. PNAS � February 20, 2007 � vol. 104 � no. 8 � 2677

BIO

PHYS

ICS

the design of disposable PDMS devices bearing nanoslits forDNA analysis through codevelopment of a compatible labelingscheme for barcoding single molecules.

The correlation of increasing DNA elongation with decreasingionic strength is explained by the electrostatic repulsive forcewithin a persistence length resulting from the Debye screeninglength. By minimizing the constraints of confinement dimen-sions required for elongating DNA molecules and instead fo-cusing on salt conditions, PDMS nanostructures become usefultools for future genomic and polymer physics studies. Forgenomic applications, a stretch of 0.60 is capable of providingvaluable data fueling biological investigations, as was previouslyobtained with the optical mapping system (3). As such, futureefforts will leverage fully automated imaging, processing, andbarcode construction (preliminary system created; data notshown) for the comprehensive assessment of errors, allowing forfull integration with new algorithms created for assembling mapsthat span entire genomes.

Materials and MethodsNanoslit Preparation. Fabrication of PDMS devices used standardrapid prototyping procedures (12). A master wafer of nanoslits(100 nm high, 1 �m wide, and 5 mm long) was dry-etched andoverlaid with microchannels (3 �m high, 100 �m wide, and 10mm long) made of negative photoresist; replica devices wereproduced by curing PDMS onto such master wafers (for details,see SI Text).

DNA Sample Preparation and Loading. DNA samples [1 ng/�l �DNA (New England Biolabs, Ipswich, MA), 0.78 ng/�l T4 DNA(Waco Chemicals, Waco, TX), and E. coli DNA (38)], contained0.25 �M YOYO-1, from 1� to 0.01� TE buffer, 4% (vol/vol)2-mercaptoethanol, and 0.1% (wt/vol) POP6 (Applied Biosys-tems, Foster City, CA) to suppress electroendosmosis (10). Inaddition to TE buffer, sodium phosphate buffer (pH 7.9; a 10:93

mixture of 10 mM NaH2PO4/10 mM Na2HPO4) and NewEngland Biolabs buffer 4 with 20 mM EDTA were also used.YOYO-1-stained DNA molecules were loaded into the micro-channels via capillary action and then entered the nanoslits byusing an applied electrical field (70 V) (Fig. 1) with platinumelectrodes inserted into the reservoirs.

Microscopy and Image Processing. The microscopy setup for singlecolor imaging and imaging flattening processes for shadingcorrection is as reported in ref. 3 (for details, see SI Text).

DNA Barcoding. The circular DNA molecules BAC79, BAC150,and BAC614 were linearized with FseI or SpeI (New EnglandBiolabs). Preexisting nicks were repaired by using 2 units of T4DNA ligase (1 mM ATP) at 16°C for 2 h, with a total volume of17.5 �l of New England Biolabs buffer 4 or New England Biolabsbuffer 2. The mix was then heat-inactivated at 65°C for 10 min.Endonuclease-free-grade E. coli DNA polymerase I (10 units)(Roche Applied Sciences, Indianapolis, IN) and added ddNTPsat 0.2 �M each (Amersham Biosciences, Piscataway, NJ) blockedremaining nicks at 37°C for 30 min (total volume, 40 �l). Thelabeling reaction mix [3 units of Nb.BbvCI (New EnglandBiolabs)/2 �M Alexa Fluor 647-aha-dCTP/2 �M Alexa Fluor647-aha-dUTP (Invitrogen, Carlsbad, CA)/20 �M dATP/20 �MdGTP/1 �M dCTP/1 �M dTTP] was then added, and the reactionwas incubated for 30 min at 37°C. Enzymes were digested with 100ng/�l proteinase K in 0.1% (wt/vol) N-lauroylsarcosine for 3 h at50°C. Samples were diluted 4,000-fold, or buffer conditions wereadjusted by dialysis against 2 liters of 0.01� TE buffer solutionovernight with a microdispodialyzer (Spectrum Laboratories,Rancho Dominguez, CA) at 4°C.

We thank S. Zhou, K. Potomousis, A. Ramme, K. Kounovsky, G.Ananiev, T. Durfee, D. Frisch, N. Hermersmann, G. Plunkett III, R.Roberts, and S. Y. Xu for assistance. This work was supported byNational Institutes of Health Grant 5R01HG000225 and National Sci-ence Foundation Grant NSEC DMR-0425880.

1. Margulies M, Egholm M, Altman WE, Attiya S, Bader JS, Bemben LA, BerkaJ, Braverman MS, Chen YJ, Chen ZT, et al. (2005) Nature 437:376–380.

2. Shendure J, Porreca GJ, Reppas NB, Lin XX, McCutcheon JP, RosenbaumAM, Wang MD, Zhang K, Mitra RD, Church GM (2005) Science 309:1728–1732.

3. Dimalanta ET, Lim A, Runnheim R, Lamers C, Churas C, Forrest DK, dePablo JJ, Graham MD, Coppersmith SN, Goldstein S, et al. (2004) Anal Chem76:5293–5301.

4. Jing J, Reed J, Huang J, Hu X, Clarke V, Edington J, Housman D, Anan-tharaman TS, Huff EJ, Mishra B, et al. (1998) Proc Natl Acad Sci USA95:8046–8051.

5. Schwartz DC, Koval M (1989) Nature 338:520–522.6. Guo XH, Huff EJ, Schwartz DC (1992) Nature 359:783–784.7. Volkmuth WD, Austin RH (1992) Nature 358:600–602.8. Chan EY, Goncalves NM, Haeusler RA, Hatch AJ, Larson JW, Maletta AM,

Yantz GR, Carstea ED, Fuchs M, Wong GG, et al. (2004) Genome Res14:1137–1146.

9. Cao H, Yu ZN, Wang J, Tegenfeldt JO, Austin RH, Chen E, Wu W, Chou SY(2002) Appl Phys Lett 81:174–176.

10. Tegenfeldt JO, Prinz C, Cao H, Chou S, Reisner WW, Riehn R, Wang YM,Cox EC, Sturm JC, Silberzan P, et al. (2004) Proc Natl Acad Sci USA101:10979–10983.

11. Manz A, Miyahara Y, Miura J, Watanabe Y, Miyagi H, Sato K (1990) SensActuators B 1:249–255.

12. Effenhauser CS, Bruin GJM, Paulus A, Ehrat M (1997) Anal Chem 69:3451–3457.

13. Chou SY, Krauss PR, Renstrom PJ (1995) Appl Phys Lett 67:3114–3116.14. Whitesides GM, Ostuni E, Takayama S, Jiang XY, Ingber DE (2001) Annu Rev

Biomed Eng 3:335–373.15. Kumar A, Biebuyck HA, Abbott NL, Whitesides GM (1992) J Am Chem Soc

114:9188–9189.16. Delamarche E, Bernard A, Schmid H, Michel B, Biebuyck H (1997) Science

276:779–781.

17. Duffy DC, McDonald JC, Schueller OJA, Whitesides GM (1998) Anal Chem70:4974–4984.

18. Odom TW, Thalladi VR, Love JC, Whitesides GM (2002) J Am Chem Soc124:12112–12113.

19. Armbrust EV, Berges JA, Bowler C, Green BR, Martinez D, Putnam NH,Zhou SG, Allen AE, Apt KE, Bechner M, et al. (2004) Science 306:79–86.

20. Zhou S, Herschleb J, Schwartz DC (2007) New Methods for DNA Sequencing(Elsevier, Amsterdam).

21. Baumann CG, Smith SB, Bloomfield VA, Bustamante C (1997) Proc Natl AcadSci USA 94:6185–6190.

22. Schellman JA (1974) Biopolymers 13:217–226.23. Skolnick J, Fixman M (1977) Macromolecules 10:944–948.24. Odijk T (1977) J Polym Sci B Polym Phys 15:477–483.25. Odijk T (1983) Macromolecules 16:1340–1344.26. Reisner W, Morton KJ, Riehn R, Wang YM, Yu ZN, Rosen M, Sturm JC, Chou

SY, Frey E, Austin RH (2005) Phys Rev Lett 94:196101.27. Riehn R, Lu MC, Wang YM, Lim SF, Cox EC, Austin RH (2005) Proc Natl

Acad Sci USA 102:10012–10016.28. Heiter DF, Lunnen KD, Wilson GG (2005) J Mol Biol 348:631–640.29. Brochard F, de Gennes PG (1977) J Chem Phys 67:52–56.30. Odijk T (2006) J Chem Phys 125:1–8.31. Jendrejack RM, Schwartz DC, Graham MD, de Pablo JJ (2003) J Chem Phys

119:1165–1173.32. Turban L (1984) J Phys (Paris) 45:347–353.33. Odijk T (1979) Biopolymers 18:3111–3113.34. Burkhardt TW (1997) J Phys A Math Gen 30:L167–L172.35. Burkhardt TW (1995) J Phys A Math Gen 28:L629–L635.36. Ubbink J, Odijk T (1999) Biophys J 76:2502–2519.37. Mannion JT, Reccius CH, Cross JD, Craighead HG (2006) Biophys J 90:4538–

4545.38. Lin J, Qi R, Aston C, Jing J, Anantharaman TS, Mishra B, White O, Daly MJ,

Minton KW, Venter JC, et al. (1999) Science 285:1558–1562.

2678 � www.pnas.org�cgi�doi�10.1073�pnas.0611151104 Jo et al.