Molecular Simulation of DNA β-Sheet and β-Barrel Structures on Graphite and Carbon Nanotubes

Molecular Simulation of DNA �-Sheet and �-Barrel Structures on Graphite and CarbonNanotubes

Daniel Roxbury,† Suresh Manohar,† and Anand Jagota*,†,‡

Department of Chemical Engineering and Bioengineering Program, Lehigh UniVersity,Bethlehem, PennsylVania 18015

ReceiVed: June 4, 2010; ReVised Manuscript ReceiVed: June 25, 2010

It has recently been discovered that certain short DNA sequences recognize specific carbon nanotubes (CNTs),allowing a mixture to be sorted into individual types. A novel �-sheet and �-barrel secondary DNA motif hasbeen proposed as the structural basis for this recognition. In this study, using molecular simulation, weinvestigate a class of DNA structures that can be formed by interstrand hydrogen bonding, their stability inplanar and barrel forms, and whether they can form the basis for CNT recognition. We show how a libraryof DNA �-barrel structures can be built from base-dimer tiling units. Various combinations of the (GT)n

family of sequences have been studied in greater detail, both as adsorbed in planar form to graphite and aswrapped helically on a CNT surface. We find that G-quartet formation brings stability to the �-sheet, whilediameter and chirality matching between the proposed DNA �-barrel and core CNT appears to stabilize anordered hybrid structure.

1. Introduction

It is well-known that the secondary structure of biopolymerssuch as proteins and nucleic acids is a principal determinant oftheir function via interactions with other biopolymers. The DNAdouble helices1 and proteins with R-helix, �-sheet, and �-barrel2

motifs are well-known examples. As in these examples, second-ary structure is usually stabilized by noncovalent interactions.When biopolymers encounter material surfaces, especially thoseof nanomaterials, they can adopt novel, sometimes ordered,conformations.

Our interest is in the interactions between DNA and carbonnanotubes. Single-stranded DNA (ssDNA) and carbon nanotubes(CNTs) form a stable hybrid that renders the latter water-dispersible3 and has enabled successful separation by length anddiameter.3–5 It has recently been shown that certain short ssDNAsequences recognize specific CNTs, thus allowing their separa-tion from a mixture by ion exchange chromatography.6 Highselectivity strongly suggests that an ordered structure is formedby the recognition sequence around a particular CNT. Novel�-sheet and �-barrel secondary structures for DNA stabilizedboth by hydrogen bonding and by base stacking onto thesubstrate have been proposed as the structural basis forrecognition.6,7 Furthermore, well-defined charge densities for(GT)30-CNT hybrids7 measured using capillary electrophoresisalso suggest a well-ordered DNA structure. In this paper, usingmolecular simulations, we explore the structure and stability ofthese novel DNA motifs.

To build models for ordered structures based on oligomericDNA adsorption, it is useful to consider first structures formedby DNA bases themselves. Several studies have been conductedto understand the adsorption of DNA bases onto solid surfacesthat have relevance to our study.8–11 The adsorption isothermsfor different bases at the graphite-water interface show that

strength of adsorption decreases in the order9 G > A > T > Cdue to their tendency to form self-assembled monolayers.9–11

The aromatic ring structures of DNA bases are known tostack on an aromatic surface due to overlapping π-orbitals,12,13

and often the bases are laterally stabilized by base-basedimer formation14–16 through interbase cyclic hydrogen bond-ing, which is further stabilized by π-bond cooperativity. Allthe dimers considered in this work have a minimum of twohydrogen bonds.1,17 Homobase dimers in a centrosymmetricconfiguration are favored because the purines and pyrimidineshave large dipole moments which are canceled in an antiparallelconfiguration.1,16 Centrosymmetric dimers can pack into a flatsheet stabilized by hydrogen-bonding interactions betweenadjacent dimers. Several high-resolution AFM and STM imagesof bases adsorbed onto a substrate reveal the above picture.11,15,18,19

Apart from dimers, guanine, adenine, and mixtures of adenine/thymine, guanine/uracil, or guanine/cytosine are known to formsupramolecular structures at the liquid/solid interface.11,14,15,18,20–22

Guanine-rich telomeric DNA in human chromosomes is ofsignificant interest because of the recombination and degradationprotection that G-quartet structures confer.23–25

What ordered structures can surface-adsorbed ssDNA mol-ecules form? Some studies have suggested the formation ofordered structures of DNA on CNTs, but little else is knownabout them.3,4,26,27 (It is known that ssDNA strands with contourlength much greater than their Kuhn length generally adsorbonto a flat surface as random coils.) The existence of someordered structures is highly plausible given the evidence of CNTrecognition by certain short strands of ssDNA and the demon-strated propensity of bases to organize into 2D structures.Indeed, one may view the class of ordered ssDNA structuresas a subset of those permitted by association of free bases withthe additional constraints that arise because bases are cova-lently attached via a sugar-phosphate backbone. In general,both ssDNA and dsDNA adsorb onto hydrophobic substrates,28–30

ssDNA adsorbing stronger than dsDNA.31 We propose thatssDNA molecules readily adsorb to hydrophobic surfaces, suchas graphite and CNT, to form hybrids through π-stacking

* To whom correspondence should be addressed. Phone: (610) 758-4396.E-mail: [email protected].

† Department of Chemical Engineering.‡ Bioengineering Program.

J. Phys. Chem. C 2010, 114, 13267–13276 13267

10.1021/jp1051497 2010 American Chemical SocietyPublished on Web 07/19/2010

interactions between the DNA base and the surface and byinterbase hydrogen bonding, while the negatively chargedphosphate backbone is exposed to water.

Molecular simulations of a single short ssDNA strand on aCNT showed disordered structures with nonhelical loop structurerepresenting the global free energy minimum,32,33 suggestingthe necessity of multiple short strands to form an orderedstructure. The main contributors to the free energy of the DNA/CNT hybrid were shown to be a competition between adhesiveinteractions between DNA and CNT and electrostatic repulsionbetween charges on the DNA backbone.12 Interestingly, thepredominant feature in the minimum free energy conformationsof (GT)7 is one in which DNA bases alternate on either side ofthe backbone, which minimizes steric hindrance.32 We now shiftour focus to the assumption that multiple strands are interactingon graphite and/or nanotube surfaces. From previously statedevidence, we assume that bases on antiparallel strands have thepropensity to form hydrogen-bonded dimers. We show that withthese constraints the DNA strands can form an ordered inter-strand hydrogen-bonded 2D sheet structure on a flat substratesuch as graphite, which can then be folded in several ways into3D-barrel structures with discrete diameters. A plausible mech-anism for the recognition of CNTs by ssDNA is then thematching of their diameters and chirality. The ordered structureswe propose, sheets and barrels, are analogous to secondarystructures of proteins, �-sheets and �-barrels, respectively. Webegin by describing the procedure for constructing the orderedstructures on graphite and CNTs. For a particular DNAsequence, (GT)n, using molecular dynamics (MD) simulationswith explicit water molecules, we study the stability of ordered�-sheet structures on graphite and the factors governing stability.To examine our hypothesis that recognition comes frommatching the DNA barrel and CNT, we perform MD simulationsof several DNA �-barrel structures around a (6,5) CNT.

2. Ordered �-Sheet and �-Barrel Structures

2.1. Construction of �-Sheets and �-Barrels. This sectionintroduces the new class of ordered DNA structures, �-sheetsand �-barrels. A step-by-step procedure, using the (GT)n

sequence as an illustration, is presented that explains theconstruction of the periodic, hydrogen-bonded sheet and barrelstructures. As mentioned in the Introduction, single adsorbedshort ssDNA chains on a substrate maximize stacking andminimize steric hindrance by adopting conformations in whichbases alternate from one side to the other of the DNAbackbone.32 We have found that two such strands placedadjacent and antiparallel to each other on a surface can forminterstrand hydrogen-bonded 2D sheets.6,7 This hydrogen-bondedstructure can be extended to form a periodic 2D sheet structureon graphite as illustrated in Figure 1a. It shows antiparallel (GT)n

strands arranged so that interstrand guanines can form hydrogenbonds. This ordered structure can be extended indefinitely ineither planar direction. Following a particular roll-up vector, c,one can create a DNA �-barrel structure by hydrogen-bondinginterstrand thymines (Figure 1b,c). It is clear that only certain,discretized, values of c are allowed, which means that theresulting barrels will have distinct diameters and helical pitches(Figure 1f,g).

2.2. Tiling Units. The example described in the previoussection shows how antiparallel ssDNA sheets can be hydrogen-bonded into an extended sheet and then rolled into a barrel. Toexamine systematically the variety of ways in which suchstructures can be constructed, we adopt as tiling units elementsof the set of hydrogen-bonded base dimers, each attached

covalently to a sugar-phosphate backbone. Figure S1 (Sup-porting Information) shows the four bases (Gua, Ade, Thy, andCyt) with possible hydrogen bond donor/acceptor pairs. Thesefour bases can be combined into 28 possible hydrogen-bondingdimers with at least two hydrogen bonds each.1 Because weare interested in extended periodic structures, we impose theadditional constraint that the backbones be parallel or antiparallelto each other and in the same plane as the bases. In this work,we consider structures in which the base alternates to the leftand right of the backbone, although structures that violate thiscondition can also be constructed using the tiling units.Examining all possible dimers, the following picture emerges.

(1) Only non-Watson-Crick base pairing is possible.Watson-Crick base pairing is inconsistent with parallelbackbones.

(2) Neighboring strands must be antiparallel.(3) Along a given strand, the O4′ atoms in the sugar rings

alternately point up and down as the base alternates from oneside to the other.

(4) The glycosyl sugar-base linkages on the same strandshould all be in either syn or anti configuration. (“Syn” vs “anti”configurations designate one of two ranges of torsion angle thatthe sugar-base glycosyl linkage can assume.1 We follow thesame reference for nomenclature of atoms in the DNA structure.)

With these constraints, only 10 of the 28 hydrogen-bondingpossibilities survive (Figure 2a). The nomenclature is based onthat of Saenger’s planar nucleic acid dimers;1 e.g., GT32au refersto a dimer with hydrogen bonding between guanine and thymine,with the “3” referring to the third documented GT configuration,the superscript “2” to two hydrogen bonds, and “a” to the antiorientation of the glycosyl torsion. A rotation of the backboneswith respect to the base by 180° about the glycosyl bond resultsin a new allowed configuration, doubling the number ofconfigurations to 20 (Figure S2b, Supporting Information).Further possibilities can emerge if one considers the placementof the O4′ oxygen with respect to the plane on which the strandadsorbs (Figure S2a). However, whether new overall configura-tions emerge depends on the composition, which will beapparent as we develop the example of (GT)n. Table 1 displaysparameter values for the 10 aforementioned tiling units (the full40 units can be found in the Supporting Information, Table S1).Notice that there are significant differences in the distancebetween adjacent strands but that the angle θ does not varymuch between dimers.

3. (GT)4 Sheets on Graphite

3.1. Creating DNA �-Sheets. Using the 40 allowed tilingunits, one can create many sheet structures. We regard this asproviding a set of starting structures that can be studied usingmolecular simulation. We find that the order implied by tilingwith dimers can evolve into more complex structures, e.g., bythe formation of G-quartets, A-quartets, and AT-quartets.14,15,34

These increase the number of hydrogen bonds and are knownto be stable structures.21 In this study we focus attention on thesequence (GT)n, which is one of a small set of special CNT-recognizing sequences. Using MD simulations, we have studiedthe equilibrium structure and stability of a few sheet and barrelstructures formed by this sequence. It is a daunting task to studyall the sheet structures through MD, and hence, a coarse grainmodel or procedure will probably be needed to predict stablesheet structures similar to the algorithms that can predict theprotein �-sheets.35,36 We restrict our attention in this section tosimulations on two or four (GT)4 strands. Instead of using MDalone to explore the entire space of conformations, we have

13268 J. Phys. Chem. C, Vol. 114, No. 31, 2010 Roxbury et al.

chosen to study selected regions of conformational space, bystarting each simulation with an ordered sheet structure. Thus,the simulations answer the question of relative stability ofordered structures, avoiding the computationally less tractableissue of whether they will assemble spontaneously fromarbitrarily arranged starting conditions.

Figure 2c shows one type of sheet structure that can be madeusing (GT)n strands. The dimension of the tiling unit is specifiedby two vectors and the angle between them. From the dimen-sions of tiling units, one can calculate the dimension of the sheetunit cell given by vectors a and b with magnitudes a and b andthe angle between them (see the Supporting Information, sectionS3).

3.2. Detailed Study of (GT)4 Sheet Structures. For (GT)n

strands, there are four ways in which bases G and T can formhydrogen bonds that comply with the conditions imposed foran ordered structure. Accounting for the syn/anti conformationsand up/down orientation of the oxygen atom in the sugar ring,there are 16 possible tiling units (4 GG32, 4 GG12, 4 TT62, and4 GT32 units). The 16 tiling units on a backbone can combinein several ways to form different ordered sheet structures whichcan be divided into two broad categories: sheets with homobasepairing (HM sheets) and sheets with heterobase pairing (HT

sheets). The HM sheets have GG dimers on one side of thebackbone and TT dimers on the other side, giving rise to 32sheet structures (4 GG32 × 4 TT62 and 4 GG12 × 4 TT62).The constraint on the backbone (adjacent oxygen atoms pointup and down; all bases on the same strand are in anti or synconfiguration) reduces the number of HM sheets by a factor of4, i.e., to eight HM sheets. Similarly, for the case of HT sheetswith GT base pairing on either side of the backbone, the initialnumber of sheets, 16 (4 GT32 × 4 GT32), is reduced to 4 sheets.This number is further reduced to two because the sheet formedfrom GT32au on one side and GT32ad on the other side isidentical to the sheet formed from GT32ad on one side andGT32au on the other side. Hence, a total of 10 sheet structurescan be formed by the (GT)n sequence.

We further focus our attention on a few of these 10 structuresthat are expected to be preferred, on the basis of previousknowledge of base-sugar orientation and base adsorption ontosurfaces. Ordered structures formed by nucleotides on graphiteshow that centrosymmetric hydrogen-bonded dimers are pre-ferred in homobase pairing.14 This implies that the GG12 dimer,being noncentrosymmetric, is less preferred and therefore thefour sheets with the GG12 dimer have not been studied. Further,in pyrimidines, the anti conformation is dominant, while in

Figure 1. (a) Four (GT)8 strands in a sheet configuration shown with GG and TT interstrand hydrogen bonding. Guanine bases are shown in blueand thymine in yellow. Connecting via hydrogen bonds the thymine labeled “O” to thymines labeled “1”-“7” gives seven distinct roll-up vectors.(b) Side view of (GT)8 sheets. (c) Extended (GT)n ribbon showing one specific roll-up vector (white dashed line). The roll-up progression is shown(d, e), resulting in the final DNA �-barrel structure (f, g).

DNA Structures on Graphite and CNTs J. Phys. Chem. C, Vol. 114, No. 31, 2010 13269

purines, both anti and syn conformations are equally preferred.1

We restrict our study to sheets with all bases in anti conforma-tions. This leaves us with three sheet structures formed from

(1) GG32au and TT62ad (sheet 1), (2) GG32ad and TT62au (sheet2), and (3) GT32ad and GT32au (sheet 3) tiling units. Apartfrom the MD simulations of these three sheets, we conductedtwo control simulations: (i) a sheet constructed from GG32auand TT62ad tiling units in the absence of graphite and (ii) asheet made from GG32au and TT62ad tiling units with alternatestrands removed. The control simulations were carried out tounderstand the stabilizing role of graphite and hydrogen bondingin ordered structures.

3.3. MD Results for (GT)4 Sheets. Figure 3 shows the MDresults for sheets 1 and 3 (also see Figure S4 in the SupportingInformation). The result for sheet 2 (Figure S4b) was similarto that for sheet 1, indicating that the orientation of oxygenatoms O4′ in the sugar ring does not have a major influence onthe sheet structure. The reason may be that there is an equalnumber of bases with O4′ pointing toward and away fromgraphite in both cases. Note that the starting minimized (Figure

Figure 2. (a) Ten possible interbase hydrogen-bonding dimer configurations are allowed with given constraints. (b) Each dimer is represented asa parallelogram tiling unit and characterized by the distance between the O3′ (purple) and O5′ (red) atoms and an angle θ. (c) (GT)4 sheets ongraphite employing GG32au and TT62ad tiling units. In this case, the final “d” refers to O4′ pointing “down”, toward the substrate, and “u” refersto it pointing “up”, away from the substrate. (d) Example of the (GT)n sheet structure with three tiling units (upon repeating), GG32ad, TT62au, andGT2au, showing that numerous sheet structures are possible even for a simple sequence such as (GT)n.

TABLE 1: Average Parameters for the 10 DNA DimerTiling Units Shown in Figure 2a in an “Anti-Down”Configuration, Minimized on Graphite in a Vacuuma

tilingunit

av θ(deg)

av O3′-O5′distance (Å)

tilingunit

av θ(deg)

av O3′-O5′distance (Å)

AA42ad 85.4 19.0 TT62ad 80.9 16.6AA12ad 81.3 15.9 CC32ad 85.7 15.9GG32ad 87.2 18.6 CC42ada

AA22ad 82.6 17.1 AC42ad 86.6 17.4GG12ad 83.3 14.5 GT32ad 83.3 14.8

a Note that the CC42 value is not reported as this configurationminimizes to CC32 on graphite. See the Supporting Information forthe parameters of all 40 configurations.

13270 J. Phys. Chem. C, Vol. 114, No. 31, 2010 Roxbury et al.

3a) and the equilibrated (Figure S4a) sheet structures did nothave G-quartets. Structures obtained after 2 ns (Figure S4a) and4 ns (Figure 3a) show the formation of G-quartets, which giveextra stability to this sheet structure. At 2 ns (Figure S4a), thetop two strands look disordered with a few hydrogen bondsbroken, but at 4 ns (Figure 3a), they find their hydrogen-bondingpartners, forming a G-quartet. (G-quartets are customarilystabilized further by a positive ion, say Na+. However, we didnot have excess ions in our simulation.) Sheet 3 reveals adifferent picture. Many bases have broken their hydrogen bonds,and the sheet appears to be disordered (Figure 3b). Theformation of a G-quartet seems to be an important featurestabilizing the ordered sheet structure. Therefore, from amongthe three sheets, those with GG32au/TT62ad and GG32ad/TT62autiling units are more stable than the sheets with GT32au/GT32adtiling units.

3.4. Analysis of Results. In this section, we analyze thestability of the ordered sheet structure formed by the (GT)4

sequence. A parameter that describes order is the conformationalspace accessible to the bases of DNA that increases with theentropy of the bases. We expect that an ordered structure willhave lower entropy and therefore its components will beconfined to a smaller region. To obtain the accessible space forbases in the presence and absence of graphite or hydrogen-bonding partners, we track the locations of their centroids duringdynamics simulations in a plane parallel to the graphite substrate.Parts a and b of Figure 4 show the base centroid positions forsheets 1 and 3 obtained between 4 and 5 ns of dynamics andfor two control simulations, respectively. It is evident fromvisual inspection of Figure 4a, the structure with GG/TT tilingunits, that the bases remain quite confined to a small area. Incomparison, the structure with GT tiling units, Figure 4b, hasconsiderably larger fluctuations. This comparison more quan-titatively supports the observation that the formation of G-quartets, by increasing the number of hydrogen-bonding pos-sibilities, provides considerable additional stability to the

structure. This role of interstrand hydrogen bonding is confirmedby the observation that, if alternate strands are removed fromthe simulations, therefore removing stabilization by interstrandhydrogen bonds, the fluctuations of the remaining bases increasesignificantly (Figure 4c). Finally, if we retain all the ssDNAstrands but remove the graphite layer, we find that the structurequickly loses its order (Figure 4d). Therefore, we conclude thatboth a substrate on which to adsorb strongly and hydrogenbonding between strands are required to stabilize �-sheetstructures. Within the possibilities permitted by arrangementsof the dimers introduced in the previous section, certainstructures achieve considerably greater stability by formingadditional hydrogen-bonding possibilities, e.g., by arrangingneighboring guanines into a quartet. To quantify the accessiblearea for the bases, we analyze the distribution of base locationsto calculate a characteristic constraining radius, r*. Large r*values correspond to large accessible areas, resulting in adisordered structure, and vice versa. (See the SupportingInformation, sections S5 and S6.) For all three sheets that containstabilizing complementary strands and are adsorbed on asubstrate, the values of r* for thymine and guanine bases are4 ( 0.25 and 3.6 ( 0.2 Å, respectively (mean ( standard error).There is a small but significant difference between guanine andthymine bases because in several cases the extra hydrogen bondson the former provided extra constraints. The visual differencein order between sheet 1 (Figure 4a) and sheet 2 (Figure 4b) ismore due to overall drift of the strands in the latter case thandue to decreased constraints over a short 500 ps time period.In contrast, the constraining radii r* for sheet 1 when alternatestrands are removed, the structure thus lacking hydrogen-bonding strands, for thymine and guanine are 6.4 ( 0.23 and7.2 ( 0.37 Å, respectively. This comparison confirms theconclusion that inter- and intrastrand hydrogen bonding isimportant to maintain order in the sheet structures.

As a second measure of relative stability of the sheet structure,we track the number of hydrogen bonds. The criteria for the

Figure 3. Snapshots from the molecular dynamics of (GT)4 sheetstructures with GG32au and TT62ad tiling units (sheet 1). G-quartetsare highlighted by white ovals. (b) Corresponding results for GT32auand GT32ad tiling units.

Figure 4. Base centroid positions for the (GT)4 sheet structures between4 and 5 ns of dynamics: (a) sheet 1 with GG32au and TT62ad tilingunits, (b) sheet 3 with GT32au and GT32ad tiling units, (c) sheet 1lacking alternate hydrogen-bonding strands between 0.5 and 1.5 ns,(d) sheet 1 lacking a graphite surface between 0.5 and 1.25 ns.

DNA Structures on Graphite and CNTs J. Phys. Chem. C, Vol. 114, No. 31, 2010 13271

existence of a hydrogen bond (A-H · · ·B) between the donorpair (A-H) and the acceptor (B) are that the distance betweenH and B should be less than 3.5 Å and the angle AHB shouldbe between 140° and 180°. Sheets 1 and 2, with GG32au/TT62adand GG32ad/TT62au tiling units, respectively, have a signifi-cantly greater number of hydrogen bonds compared to sheet 3with GT32au/GT32ad tiling units (Figure 5). This confirms ourreasoning that sheets 1 and 2 with favorable centrosymmetrichydrogen bonding and G-quartet configuration are more stablethan sheet 3. From the study of ordered structures on planargraphite we therefore conclude that (i) confinement to a planeby strong adsorption and interstrand hydrogen bonding are bothneeded to stabilize ordered �-sheet structures, (ii) the cen-trosymmetric hydrogen-bonding configuration is preferred, (iii)sheets formed by GG32au/TT62ad and GG32ad/TT62au tilingunits are more stable than the sheets with GT32au/GT32ad tilingunits, and (iv) G-quartet structures form spontaneously, con-tributing additionally to the stability of the ordered sheets.

4. (GT)30 �-Barrels on CNTs

In this section we examine the hypothesis that recognitionof CNTs by �-barrels is based on matching diameter andchirality between the two by conducting MD simulations on anumber of different �-barrels constructed from the same �-sheeton a given (6,5) CNT.

4.1. Creating DNA �-Barrels. Starting with the 2D sheetstructure, a barrel can be obtained by rolling the sheet in thedirection of the roll-up vector, c, and connecting two equivalentpoints on the periodic sheet, for example, two equivalentphosphorus atoms. The magnitude of the roll-up vector becomesthe circumference of the barrel. Discrete values for c implydiscrete diameter barrels. Figure 1f shows one such barrelcreated by rolling in the direction shown in Figure 1c. The vectorc can be written as a linear combination of unit cell parametersof the sheet, a and b (Figure 2c), as

where p and q are integers. The magnitude of the vector c is

Similar to the chirality of nanotubes, we can define thechirality of the barrels on the basis of the roll-up direction andunit cell parameters. The angle between the roll-up vector, c,and the unit cell vector b, θch, is given as

to which we add angle R to obtain the true chirality of the�-barrel (see Figure 2c). The diameter of the barrel, dbarrel, isfound by dividing eq 2 by π, and the diameter of its availableinner space is calculated by subtracting from it 2 times the vdWdistance, bvdw, between the base and the CNT (bvdw ≈ 3.5 Å):

4.2. Molecular Dynamics of (GT)30 Barrel Structures. Wecarried out MD simulations with a number of �-barrel structuresbased on (GT)30 sheets wrapped around a (6,5) CNT. Of the 10possible (GT)n sheets, we have chosen the GG-TT hydrogen-bonding scheme because of the greater stability it exhibited inMD simulations on graphite. In this study, we limit ourselvesto the (6,5) chirality nanotube because of its abundance inSWeNT CoMoCAT nanotubes. We have examined sevendifferent roll-up vectors (Figure 1a). Note that all of the rollingoccurred “out of the page” so that the O4′ atoms in the guaninesugar rings all pointed inward, or toward the CNT. Rolling “intothe page” would form a second set of seven. For these sevenbarrels, we ask the following: Choosing one of the (GT)n sheetsand a single type of CNT, what is the effect of choosing differentroll-up vectors, each one of which yields a different radius andchirality? Specifically, do the molecular simulations support thenotion that �-barrels that match the core CNT better are moreordered and stable? As in the case of simulations on planarsheets, we have chosen to study selected regions of conforma-tional space, by starting each simulation with an ordered barrelstructure. Thus, we view the simulations as providing an answerto the question of whether a particular putative ordered barrelstructure is stable compared to another or to a negative control.Specifically, the simulations do not address the computationallyless tractable issue of whether barrels will assemble spontane-ously from arbitrarily arranged starting conditions.

4.3. Results for (GT)30 �-Barrels. Figure 6 shows three ofthe seven structures at the beginning and after 6 ns of MDsimulation. These seven are numbered sequentially 1-7, withdecreasing �-barrel diameters. The largest diameter, dtube, is 12.8Å, while the smallest is 3.6 Å, compared to 7.56 Å for the core(6,5) CNT. Figure S10 and Table S4 (Supporting Information)show that the fraction of hydrogen bonds, one of the parameterswe use as indicative of an ordered structure, decreases quicklyand appears to have converged to a stable value in all cases.For two of the cases, numbers 5 and 7, we have carried out thesimulation for 20 ns, and the results for the fraction of hydrogenbonds are shown in Figure 7. From these results we concludethat the 6 ns simulations are of sufficiently long duration toanswer the question we have posed about the relative stabilityof the starting structures we are comparing.

Qualitatively, it can be seen that in all of the cases the DNAstrands remain adsorbed onto the nanotube. Closer inspectionshows that many of the T-T hydrogen bonds have broken, butin the more stable structures the G-G H-bonds largely remainintact. The tight-fitting case, with a starting barrel diameter of3.6 Å, quickly becomes disordered. (To create DNA/CNThybrids with �-barrels smaller than the actual CNT diameter,we first created a CNT with smaller diameter by reducingcarbon-carbon bond distances. This allowed the CNT to beinserted into the �-barrel, after which the C-C bond distanceswere slowly increased and the structure was continuously

Figure 5. Fraction of hydrogen bonds as a function of the dynamicstime for sheet 1 formed by GG32ad/TT62au tiling units (red solid line),sheet 2 formed by GG32au/TT62ad tiling units (black dotted line), andsheet 3 formed by GT32au/GT32ad tiling units (blue dashed line).

c ) pa + qb (1)

c ) √(pa)2 + (qb)2 - 2pqab cos φ (2)

θch ) sin-1(pa sin(π - θ)c ) ) sin-1(qa sin θ

c ) (3)

dtube ) dbarrel - 2bvdw (4)

13272 J. Phys. Chem. C, Vol. 114, No. 31, 2010 Roxbury et al.

minimized.) The loose-fitting case, 12.8 Å diameter, also losesorder, although that is not so apparent from a visual inspection.The intermediate case shown in Figure 6, with a barrel diameterof 7.7 Å, fits the CNT well and appears to maintain an orderedstructure. We observed that when the �-barrel and CNTdiameters are highly mismatched, the barrel responds to theimposed strain by deforming axially. In the process, interstrandhydrogen bonds are broken. It is possible that, given sufficientsimulation time, one configuration could transform into anotherby re-forming hydrogen bonds, effectively changing its roll-upvector, c.

We conducted two control simulations to understand thestabilizing role of interstrand hydrogen bonding and of the CNTcore. Parts a and b of Figure 8 show configuration 5 after 6 nsof dynamics simulation with and without a CNT core, respec-tively. Parts c and d of Figure 8 show base centroid positionsfor the structures at different times, reported as a function ofthe angle θ around the circumference as one traverses downthe axial length of the nanotube. From Figure 8b we observethat the barrel structure changes shape considerably and henceis no longer stable, absent the CNT core. Figure 8c clearly showsthat, with the stabilizing influence of the CNT core, basecentroids remain in an ordered arrangement, which is lost whenthe CNT is removed (Figure 8d). Similar to the case of sheetson graphite, we conducted a second control simulation in whichone of the two (GT)30 strands was removed. Again, we foundthat removal of interstrand hydrogen bonds in this manner resultsin a large increase of disorder (data not shown).

Like in the sheet structures, for the more stable configurations(e.g., numbers 4 and 5) we observe the spontaneous formationof G-quartets. The flexibility of the backbone allows enoughfreedom so that this well-documented structure can emerge. A

particular instance is highlighted by the white oval in Figure 9for configuration 5. Such quartet formation doubles the numberof hydrogen bonds holding the barrel together by adding anextra four intrastrand hydrogen bonds per quartet. We suggestthat the formation of such quartets is key to maintaining stabilityof the structures over long periods of time. The version of thebarrel studied here consists of two helical grooves, oneconsisting of guanines and the other of thymines. Note that sincethe backbone deforms to accommodate the formation of theG-quartets in the guanine groove, thymines in the neighboringgroove are pulled away from each other, forcing their respectivehydrogen bonds to break.

For a more quantitative analysis of the MD simulations, wecomputed both the number of remaining hydrogen bonds(normalized by the starting value) and a constraining radius,r*, as described in section 3.4. In Figure 10 the green squaresrepresent the (6,5) CNT, which is dominant in CoMoCAT andis the one studied here. Figure 10 also shows the seven barrelconfigurations as circles. The diameter of each circle is inverselyproportional to r*; i.e., larger circles represent more confinedand hence more stable DNA barrel structures. Each circle iscolored on a gradient from blue to red, proportional to thenumber of H-bonds; i.e., structures represented by redder circleshave a greater number of hydrogen bonds and are more stable.Therefore, an ordered DNA barrel structure is represented by alarger and redder circle. It is apparent from Figure 10 that thereis a strong correlation between stability and order on one handand the “distance” from the (6,5) CNT on the other. In otherwords, the DNA barrels (4 and 5) that better match the (6,5)CNT diameter and chirality (closer to green squares) are morestable and ordered (larger and redder circles). The dashed linerepresents the locus of chirality and barrel diameter that wouldbe achieved using average unit cell parameters and serves as aguide for the eye to observe the trend in discrete barreldiameters.

Several other facts are interesting. The allowed barreldiameters are in the correct range to match CNTs but are well-separated from each other, which allows discriminative one-to-one matching. The DNA barrels are quite flexible, and thestarting diameter and chirality can likely adjust somewhat. Theresults shown in Figure 10 suggest that both diameter matchingand chirality matching play a role, thus providing a plausiblemechanism for chirality-specific recognition.

The main findings of this section are (i) the CNT substrate,as well as interstrand hydrogen bonding, is necessary to stabilizethe proposed DNA barrel structures, (ii) recognition by DNA�-barrels of CNTs appears to require matching of both diameterand chirality, and (iii) the formation of quartet structurescontributes a significant additional stabilizing influence.

Figure 6. Three of the seven (numbers 1, 5, and 7 in Figure 1) DNA/CNT hybrid �-barrels before (left) and after (right) 6 ns of dynamicssimulation. The water box and Na+ counterions have been removed for ease of viewing. From top to bottom, the figure shows relatively loose (12.8Å), fitting (7.7 Å), and tight (3.6 Å) �-barrel starting diameters, respectively.

Figure 7. Fraction of hydrogen bonds remaining during two 20 nslong simulations.

DNA Structures on Graphite and CNTs J. Phys. Chem. C, Vol. 114, No. 31, 2010 13273

5. Conclusions

Using molecular simulations, we have examined the proposedstructural basis for the recently discovered ability of certainssDNA sequences to recognize specific carbon nanotubes.Molecular dynamics simulations support the idea that stablessDNA strands can form �-sheet and �-barrel structures whenadsorbed onto a surface such as graphite or a carbon nanotube.We have identified a set of base-dimer tiling units that can beused to form extended, ordered, DNA sheet structures. In thiswork, we have studied the (GT)n sequence in some detail. Wefind that, by the formation of G-quartets, sheets with separateGG and TT tiling units are significantly more stable than thosewith GT tiling units.

Using the stable sheet form, we have simulated seven differentDNA barrels around a (6,5) CNT core. We find that both theCNT core and interstrand hydrogen bonds are needed for a stableand ordered DNA �-barrel. The results also show that bettermatching of diameter and chirality with the core CNT confershigher stability and order on the DNA �-barrels. This findingtherefore supports the hypothesis that chirality-specific recogni-tion of CNTs by DNA proceeds by matching of the DNA

�-barrel diameter and chirality with those of the core CNT. Theproposed ordered DNA models provide a structural basis forCNT recognition in CNT purification via ion-exchangechromatography.

6. Molecular Simulation Methodology

All structures were first created in Materials Studioworkplace37 and converted into formats suitable for otherprograms. Strands of two-dimensional (GT)n sheets wereformed antiparallel to each other and minimized in vacuumat 0 K on a 2D hexagonal sheet of sp2-hybridized carbonusing the CHARMM (version 35) force field program.38 (The

Figure 8. MD simulation of the �-barrel structure, configuration 5, (a) with and (b) without a CNT core. Centroid positions are given for a 100ps time frame (c) with and (d) without a CNT core.

Figure 9. MD simulation showing development of the G-quartetstructure on a nanotube, highlighted by the white oval.

Figure 10. Chiral angle vs diameter for GG-TT DNA �-barrels(circles) (6,5) CNTs (squares). The diameter of each circle is inverselyproportional to r*; i.e., larger circles represent more confined and hencemore stable DNA barrel structures. Each circle is colored on a gradientfrom blue to red, proportional to the number of H-bonds; i.e., structuresrepresented by redder circles have a greater number of hydrogen bondsand are more stable. Therefore, an ordered DNA barrel structure isrepresented by a larger and redder circle. The numeric values are foundin the Supporting Information.

13274 J. Phys. Chem. C, Vol. 114, No. 31, 2010 Roxbury et al.

atoms in the sheet were constrained against motion by aspring of strength 13.9 N m-1.) For �-sheet simulations, awater box was created around the entire structure with asufficient number of Na+ ions added to balance the chargecreated by the negatively charged phosphates on the DNAbackbone. The structures were then heated and equilibratedover a period of 20 ps each using the NAMD (version 2.7b2)molecular modeling package.39 For �-barrel simulations, thearomatic carbon sheet was first removed and the remainingDNA sheet was folded around a circumferential axis into a3D barrel structure. A (6,5) chirality carbon nanotube wasinserted into the DNA barrel, and the full structure wasminimized in CHARMM (again constraining the CNT atoms).A water box (40 × 40 × 300 Å) was created to envelop thehybrid structure, and charge-neutralizing Na+ counterionswere added. Heating and equilibration similar to thosedescribed above were applied to the �-barrel using the NAMDmolecular modeling program. All configurations were thenready for dynamic simulation (graphite and CNT carbonatoms remained constrained). Each configuration was run forat least 6 ns with a time step of 1 fs and a data recordinginterval of 10 ps.

Acknowledgment. This work was supported in part by theNational Science Foundation under Grant CMS-0609050. Wegratefully acknowledge many useful discussions with Drs.M. Zheng and X. Tu. This research was supported in part bythe National Science Foundation through TeraGrid resourcesprovided by the Texas Advanced Computing Center (TACC)under grant number [TG-MCB100049]. We specificallyacknowledge the assistance of Mr. Chris Hempel and Dr.Hang Liu.

Supporting Information Available: Detailed informationon the full 40 DNA tiling units and a derivation for �-barrelparameters, snapshots of initial and final configurations frommolecular dynamics simulations for different simulatedconfigurations, and results of analysis of MD simulations,including hydrogen bond and constraining potential proper-ties, for both DNA �-sheet and �-barrel structures. Thisinformation is available free of charge via the Internet athttp://pubs.acs.org.

References and Notes

(1) Saenger, W. Principles of Nucleic Acid Structure; Springer-Verlag:New York, 1984; pp 134-135.

(2) Branden, C.; Tooze, J. Introduction to Protein Structure; GarlandPublishing: New York, 1999.

(3) Zheng, M.; Jagota, A.; Semke, E. D.; Diner, B. A.; Mclean, S. R. L.;Richardson, R. E.; Tassi, N. G. DNA-assisted dispersion and separation ofcarbon nanotubes. Nat. Mater. 2003, 2, 338–343.

(4) Zheng, M.; Jagota, A.; Strano, M. S.; Santos, A. P.; Barone, P.;Chou, S. G.; Diner, B. A.; Dresselhaus, M. S.; Mclean, S. R. L.; Onoa,G. B.; Samsonidze, G. G.; Semke, E. D.; Usrey, M.; Walls, D. J. Structure-based carbon nanotube sorting by sequence-dependent DNA assembly.Science 2003, 302 (5650), 1545–1548.

(5) Huang, X.; Mclean, S. R. L.; Zheng, M. High-resolution lengthsorting and purification of DNA-wrapped carbon nanotubes by size-exclusion chromatography. Anal. Chem. 2005, 77 (19), 6225–6228.

(6) Tu, X. M.; Manohar, S.; Jagota, A.; Zheng, M. DNA sequencemotifs for structure-specific recognition and separation of carbon nanotubes.Nature 2009, 460 (7252), 250–253.

(7) Khripin, C. Y.; Manohar, S.; Zheng, M.; Jagota, A. Measurementof electrostatic properties of DNA-carbon nanotube hybrids by capillaryelectrophoresis. J. Phys. Chem. C 2009, 113 (31), 13616–13621.

(8) Sowerby, S. J.; Holm, N. G.; Petersen, G. B. Origins of life: Aroute to nanotechnology. Biosystems 2001, 61 (1), 69–78.

(9) Sowerby, S. J.; Cohn, C. A.; Heckl, W. M.; Holm, N. G. Differentialadsorption of nucleic acid bases: Relevance to the origin of life. Proc. Natl.Acad. Sci. U.S.A. 2001, 98 (3), 820–822.

(10) Sowerby, S. J.; Edelwirth, M.; Heckl, W. M. Self-assembly at theprebiotic solid-liquid interface: Structures of self-assembled monolayersof adenine and guanine bases formed on inorganic surfaces. J. Phys. Chem.B 1998, 102 (30), 5914–5922.

(11) Tao, N. J.; Shi, Z. Monolayer guanine and adenine on graphite inNaCl solution: A comparative STM and AFM study. J. Phys. Chem. 1994,98 (5), 1464–1471.

(12) Manohar, S.; Tang, T.; Jagota, A. Structure of homopolymer DNA-CNT hybrids. J. Phys. Chem. C 2007, 111 (48), 17835–17845.

(13) McCarthy, B.; Coleman, J. N.; Czerw, R.; Dalton, A. B.; Panhuis,M. I. H.; Maiti, A.; Drury, A.; Bernier, P.; Nagy, J. B.; Lahr, B.; Byrne,H. J.; Carroll, D. L.; Blau, W. J. A microscopic and spectroscopic study ofinteractions between carbon nanotubes and a conjugated polymer. J. Phys.Chem. B 2002, 106 (9), 2210–2216.

(14) Edelwirth, M.; Freund, J.; Sowerby, S. J.; Heckl, W. M. Molecularmechanics study of hydrogen bonded self-assembled adenine monolayerson graphite. Surf. Sci. 1998, 417 (2-3), 201–209.

(15) Mamdouh, W.; Dong, M. D.; Xu, S. L.; Rauls, E.; Besenbacher,F. Supramolecular nanopatterns self-assembled by adenine-thymine quartetsat the liquid/solid interface. J. Am. Chem. Soc. 2006, 128 (40), 13305–13311.

(16) Sowerby, S. J.; Stockwell, P. A.; Heckl, W. M.; Petersen, G. B.Self-programmable, self-assembling two-dimensional genetic matter. OriginsLife EVol. Biosphere 2000, 30 (1), 81–99.

(17) Jeffrey, G. A.; Saenger, W. Hydrogen Bonding in BiologicalStructures; Springer-Verlag: New York, 1991.

(18) Mamdouh, W.; Kelly, R. E. A.; Dong, M. D.; Kantorovich, L. N.;Besenbacher, F. Two-dimensional supramolecular nanopatterns formed bythe coadsorption of guanine and uracil at the liquid/solid interface. J. Am.Chem. Soc. 2008, 130 (2), 695–702.

(19) Sowerby, S. J.; Edelwirth, M.; Reiter, M.; Heckl, W. M. Scanningtunneling microscopy image contrast as a function of scan angle inhydrogen-bonded self-assembled monolayers. Langmuir 1998, 14 (18),5195–5202.

(20) Xu, S. L.; Dong, M. D.; Rauls, E.; Otero, R.; Linderoth, T. R.;Besenbacher, F. Coadsorption of guanine and cytosine on graphite: Orderedstructure based on GC pairing. Nano Lett. 2006, 6 (7), 1434–1438.

(21) Otero, R.; Schock, M.; Molina, L. M.; Laegsgaard, E.; Stensgaard,I.; Hammer, B.; Besenbacher, F. Guanine quartet networks stabilized bycooperative hydrogen bonds. Angew. Chem., Int. Ed. 2005, 44 (15), 2270–2275.

(22) Gu, J. D.; Leszczynski, J. Structures and properties of the planarG ·C ·G ·C tetrads: Ab initio HF and DFT studies. J. Phys. Chem. A 2000,104 (31), 7353–7358.

(23) Hackett, J. A.; Feldser, D. M.; Greider, C. W. Telomere dysfunctionincreases mutation rate and genomic instability. Cell 2001, 106 (3), 275–286.

(24) Parkinson, G. N.; Lee, M. P. H.; Neidle, S. Crystal structure ofparallel quadruplexes from human telomeric DNA. Nature 2002, 417 (6891),876–880.

(25) Phillips, K.; Dauter, Z.; Murchie, A. I. H.; Lilley, D. M. J.; Luisi,B. The crystal structure of a parallel-stranded guanine tetraplex at 0.95angstrom resolution. J. Mol. Biol. 1997, 273 (1), 171–182.

(26) Johnson, R. R.; Johnson, A. T. C.; Klein, M. L. Probing the structureof DNA-carbon nanotube hybrids with molecular dynamics. Nano Lett.2008, 8, 69–75.

(27) Cathcart, H.; Nicolosi, V.; Hughes, J. M.; Blau, W. J.; Kelly, J. M.;Quinn, S. J.; Coleman, J. N. Ordered DNA wrapping switches onluminescence in single-walled nanotube dispersions. J. Am. Chem. Soc. 2008,130, 12734–12744.

(28) Brett, A. M. O.; Chiorcea, A. M. Atomic force microscopy of DNAimmobilized onto a highly oriented pyrolytic graphite electrode surface.Langmuir 2003, 19, 3830–3839.

(29) Paquim, A. C.; Oretskaya, T. S.; Brett, A. M. O. Adsorption ofsynthetic homo- and hetero-oligodeoxynucleotides onto highly orientedpyrolytic graphite: Atomic force microscopy characterization. Biophys.Chem. 2006, 121 (2), 131–141.

(30) Paquim, A. C.; Oretskaya, T. S.; Brett, A. M. O. Atomic forcemicroscopy characterization of synthetic pyrimidinic oligodeoxynucleotidesadsorbed onto an HOPG electrode under applied potential. Electrochim.Acta 2006, 51 (24), 5037–5045.

(31) Cardenas, M.; Braem, A.; Nylander, T.; Lindman, B. DNAcompaction at hydrophobic surfaces induced by a cationic amphiphile.Langmuir 2003, 19, 7712–7718.

(32) Johnson, R. R.; Kohlmeyer, A.; Johnson, A. T. C.; Klein, M. L.Free energy landscape of a DNA-carbon nanotube hybrid using replicaexchange molecular dynamics. Nano Lett. 2009, 9 (2), 537–541.

(33) Martin, W.; Zhu, W.; Krilov, G. Simulations study of noncovalenthybridization of carbon nanotubes by single-stranded DNA in water. J. Phys.Chem. B 2008, 112 (50), 16076–16089.

(34) Davis, J. T. G-quartets 40 years later: From 5′-GMP to molecularbiology and supramolecular chemistry. Angew. Chem., Int. Ed. 2004, 43(6), 668–698.

DNA Structures on Graphite and CNTs J. Phys. Chem. C, Vol. 114, No. 31, 2010 13275

(35) Cheng, J.; Randall, A. Z.; Sweredoski, M. J.; Baldi, P. SCRATCH:A protein structure and structural feature prediction server. Nucleic AcidsRes. 2005, 33, W72-W76.

(36) Cheng, J. L.; Baldi, P. Three-stage prediction of protein beta-sheetsby neural networks, alignments and graph algorithms. Bioinformatics 2005,21, I75–I84.

(37) Materials Studio (R), Accelrys Inc., San Diego, CA.

(38) CHARMM, version 35b2, Cambridge, MA.(39) Phillips, J. C.; Braun, R.; Wang, W.; Gumbart, J.; Tajkhorshid, E.;

Villa, E.; Chipot, C.; Skeel, R. D.; Kale, L.; Schulten, K. Scalable moleculardynamics with NAMD. J. Comput. Chem. 2005, 26, 1781–1802.

JP1051497

13276 J. Phys. Chem. C, Vol. 114, No. 31, 2010 Roxbury et al.