Structure and Dynamics of Supercoil-StabilizedDNA Cruciforms

Luda S. Shlyakhtenko1, Vladimir N. Potaman2, Richard R. Sinden2

and Yuri L. Lyubchenko1*

1Departments of Biology andMicrobiology, Arizona StateUniversity, TempeAZ 85287-2701, USA2Institute for Bioscience andTechnology, Department ofBiochemistry and BiophysicsTexas A&M UniversityHouston, TX 77030, USA

Understanding DNA function requires knowledge of the structure oflocal, sequence-dependent conformations that can be dramatically differ-ent from the B-form helix. One alternative DNA conformation is thecruciform, which has been shown to have a critical role in the initiationof DNA replication and the regulation of transcription in certain systems.In addition, cruciforms provide a model system for structural studies ofHolliday junctions, intermediates in homologous DNA recombination.Cruciforms are not thermodynamically stable in linear DNA due tobranch point migration, which makes their study using many biophysicaltechniques problematic. Atomic Force Microscopy (AFM) was applied tovisualize cruciforms in negatively supercoiled plasmid DNA. Cruciformsare seen as clear-cut extrusions on the DNA ®lament with the lengths ofthe arms consistent with the size of the hairpins expected from a 106 bpinverted repeat. The cruciform exists in two different conformations, anextended one with the angle of ca. 180� between the hairpin arms and acompact, X-type conformation, with acute angles between the hairpinarms and the main DNA strands. The ratio of molecules with the differ-ent conformations of cruciforms depends on ionic conditions. In thepresence of high salt or Mg cations, a compact, X-type conformation ishighly preferable. Remarkably, the X-conformation was highly mobileallowing the cruciform arms to adopt a parallel orientation. The structureobserved is consistent with a model of the Holliday junction with aparallel orientation of the exchanging strands.

# 1998 Academic Press

Keywords: inverted repeats in DNA; cruciform conformation; DNAsupercoiling; Holliday junction; Atomic Force Microscopy*Corresponding author

Introduction

Local sequence-dependent DNA conformationscan be dramatically different from the canonicalB-form helix. One class of alternative DNA confor-mations is cruciforms, which can form in DNAsequences with a 2-fold symmetry (invertedrepeats or palindromes; Gierer, 1966). Cruciformformation requires negative DNA supercoiling andsuf®cient supercoiling exists in cells that cruciformscan exist in vivo (reviewed by Sinden, 1994).Inverted repeats are present in many functionallyimportant genome regions. Speci®cally, invertedrepeats are associated with origins of replication inplasmids and in many prokaryotic and eukaryotic

viruses (Kornberg & Baker, 1992). An involvementof cruciform in the regulation of gene expressionhas been suggested based on the requirement ofinverted repeat symmetry for transcription of somegenes (Glucksmann et al., 1992). Another motiv-ation in the study of cruciforms is their putativeequivalence to Holliday junctions (Lilley &Kemper, 1984). The latter type of DNA structure isan intermediate in a homologous DNA recombina-tion (Sigal & Alberts, 1972; Kowalczykowski,1994), and the cruciform structure might be themost appropriate system for modeling Hollidayjunctions.

Inverted repeats under physiological conditionscan adopt cruciform structures which are stabilizedby unrestrained negative DNA supercoiling(reviewed by Sinden, 1994). The formation of openregions at the loops and possible base unstackingat the junction make the existence of cruciforms in

Abbreviations used: AFM, Atomic Force Microscopy;AP-mica, mica functionalized with3-aminopropyltriethoxy silane.

Article No. mb981855 J. Mol. Biol. (1998) 280, 61±72

0022±2836/98/260061±12 $30.00/0 # 1998 Academic Press

linear DNA thermodynamically less favorable(reviewed by Vologodskii, 1992), and they arenever observed in stable form in linear DNA mol-ecules. The details of the secondary structure ofcruciforms have been elucidated utilizing enzy-matic and chemical techniques. Several bases at theloops are easily accessible to chemical probes andto single strand-speci®c enzymes, indicative ofunpaired and unstacked bases (Sinden, 1994;Palecek, 1992). The bases at the junction are muchless reactive, but they can be attacked by chemicalprobes at elevated temperatures or under con-ditions of low salt (McClellan & Lilley, 1987;Voloshin et al., 1989). Several studies using differ-ent physical chemical techniques have analyzedthe structure of stable, immobile four-way DNAjunctions (Cooper & Hagerman, 1987; Lilley &Clegg, 1993; Seeman & Kallenbach, 1994; Lilley,1997). However, there has been little attempt toanalyze the spatial geometry of cruciforms. Inlarge part, this is due to the instability of cruci-forms in small linear DNA molecules. Cruciformsare not thermodynamically stable in linear DNAdue to branch point migration. Moreover, stablecruciforms in relatively large negatively super-coiled DNA molecules are not amenable to studiesby traditional structural biology methods, such ashigh resolution NMR and crystallography. Electronmicroscopy has been used to visualize cruciformstructures, but this system involved totally palin-dromic supercoiled circular DNA, in which thesize of cruciforms varied considerably ( Mizuuchiet al., 1982).

Since the early 1990s another visualization tech-nique, Atomic Force Microscopy (AFM; Binniget al., 1986), has become available for imagingDNA and nucleoprotein complexes (Bustamanteet al., 1994; Hansma & Hoh, 1994; Lyubchenkoet al., 1995). We have developed a sample prep-aration method permitting the deposition and ima-ging of the sample in a broad range ofenvironmental conditions (Lyubchenko et al., 1992,1993a,b, 1995). This method has been recentlyapplied to structural studies of supercoiled DNA(Lyubchenko & Shlyakhtenko, 1997) to demon-strate directly that the overall structure of super-coiled DNA dramatically depends on ionicconditions. Here we report a visualization of cruci-forms in supercoiled DNA. The AFM data demon-strate directly and unambiguously two differentcruciform conformations, a compact X-type and anextended one. The capability of AFM to operate inaqueous solutions allowed us to detect a surpris-ingly high conformational ¯exibility of cruciforms.In particular, an X-type conformation was verymobile, allowing the hairpin arms to movebetween the orthogonal and almost parallel orien-tations. The geometry of cruciforms with parallelhairpin arms is structurally equivalent to Hollidayjunctions with a parallel orientation of exchangingstrands.

Results

AFM imaging of cruciforms in supercoiled DNA

We used plasmid pUC8F14C DNA containing a106-bp F14C inverted repeat with an expectedcruciform arm length of 53 bp (Sinden, 1994;Kochel & Sinden, 1988). We prepared DNAsamples with different superhelical densitiesincluding those suf®cient for the cruciform tran-sition in all of the topoisomers ( Kochel & Sinden,1988). The key in the AFM sample preparation pro-cedure is the use of the functionalized substrate(AP-mica) which permits one to obtain high resol-ution AFM images of DNA by a one step depo-sition from a very wide range of ionic conditions(Lyubchenko et al., 1992, 1993a,b; Lyubchenko &Shlyakhtenko, 1997). This procedure is highlyreproducible and allows one to perform imaging inair or in aqueous solutions (Lyubchenko et al.,1992, 1993b; Lyubchenko & Shlyakhtenko, 1997).

The AFM images of the sample prepared bydeposition from a low salt solution (TE buffer) areshown in Figure 1. A fraction of DNA topoisomerswith two to six supercoils remaining after the tran-sition ((Lk ÿ Lk0) � ÿ12 to ÿ18) was used in theexperiments shown in Figure 1A. The main featureof these images is the presence of rather long extru-sions indicated by arrows. These extrusions aregenerally of two types; those with either extendedor acute geometry of cruciform arms. In theextended geometry, both arms lie almost in oneline, so that the angle between them is ca. 180(molecules 2 and 4). Extrusions of this type werefound in 25(�5)% of molecules. Molecules inwhich the cruciform arms form a rather acuteangle (X-type geometry, molecules 1, 3, 5, 6) withthe DNA strand being sharply bent were the mostcommon, ca. 75(�5)%. Zoomed images of themolecules with linear extrusions and those withthe X-type geometry are shown as insets (i) and(ii), respectively. The size of the arms measured forextended extrusions is 15 to 20 nm, in full agree-ment with the expected length of the cruciformhairpins containing 53 bp (18 nm for B-helix DNAgeometry).

At lower superhelical density ((Lk ÿ Lk0) � ÿ7 toÿ13), molecules that appeared to be relaxed circleswith extrusions in extended and X-type confor-mations were observed; two examples are shownin Figure 1B and C, respectively. According totwo-dimensional gel electrophoresis data, thetransition of the 106-bp inverted repeat into thecruciform structure takes place in a very narrowrange of (Lk ÿ Lk0) (from ÿ9 to ÿ10), so the cruci-form-containing plasmids of this sample should betopologically relaxed molecules. Crucforms willpersist in these molecules, since cruciforms arestable in DNA with very low superhelical density(Sinden et al., 1983). The AFM observations ofrelaxed DNA molecules containing cruciformsagree with electrophoretic data, which support thehypothesis that the extrusions visualized with

62 Structure of Cruciforms

Figure 1. AFM images of pUC8F14C plasmid DNA obtained by imaging in air. The cruciforms are indicated witharrows and numbered. A, For these studies a sample, which contained a set of topoisomers with the linking numberdifference Lk from ÿ12 to ÿ18, was used. The insets (i) and (ii) are the zoomed images of extended and X-type con-formations of the cruciforms, respectively. B and C, These images were obtained for the sample with lower supercoil-ing density ((Lk ÿ Lk0) from ÿ7 to ÿ13).

Structure of Cruciforms 63

AFM are, in fact, the cruciform arms of the 106 bppalindrome.

Effect of ionic conditions on conformation ofthe cruciform

To study the effect of ionic strength on the cruci-form conformation, DNA was deposited from highsalt solutions (TE buffer plus 200 mM NaCl). Thedata for the sample with (Lk ÿ Lk0) � ÿ12 to ÿ18are shown in Figure 2A, where unambiguouslyidenti®ed cruciforms are indicated with arrows.A small scale image with three DNA molecules isshown in the inset. Under these high ionic con-ditions the cruciform conformation was predomi-nantly the X-type. Only a few DNA moleculescontained cruciforms in the extended conformation(e.g. molecule 2). The data for a highly supercoiledDNA sample ((Lk ÿ Lk0) � ÿ18 to ÿ24) also indi-cate that the conformation of the cruciform ismainly X-type (Figure 2B). Similar results wereobtained for the sample prepared in TE buffer plus10 mM MgCl2. Thus, cations stabilize the X-typecruciform structure rather than an extended one.These results are generally consistent with theeffect of cations on geometry of immobile four-wayjunctions (reviewed by Cooper & Hagerman, 1987;Lilley & Clegg, 1993; Seeman & Kallenbach, 1994;Lilley, 1997): the junctions change from anextended conformation into the X-type geometryupon increasing concentrations of Mg or Na ions.

To compare the geometric characteristics of theX-type cruciform at different ionic conditions, wemeasured the angles between the hairpins andbetween the main DNA strands for the imagesobtained under low salt conditions (TE buffer),

high salt conditions (200 mM NaCl), and in thepresence of Mg cations. The data in Table 1 showthat there was only a very weak effect of ionic con-ditions on the X-type geometry of the cruciform:the mean values for the angles are rather insensi-tive to a high ionic strength or the presence of Mgcations. A slightly higher value for the anglebetween the main strand DNA arms under lowsalt conditions may be due to a more extendedconformation of supercoiled DNA in low salt sol-utions as found in our earlier work (Lyubchenko &Shlyakhtenko, 1997). In addition, there is a veryhigh variability of the inter-arm angles (see, forexample, molecules 2 and 3 in Figure 2B). Theseangles in different molecules vary in such a broadrange that the molecules with almost parallelorientation of the hairpin arms are present in ahigh proportion. These data suggest that the armsof the cruciform move in a rather broad range, sothe high variability of the angles between the hair-pin arms may represent snapshots of a highlymobile structure. To observe the dynamics of cruci-forms directly, we performed AFM imaging insolution omitting the step of sample drying (AFMin situ; Lyubchenko & Shlyakhtenko, 1997). In thiscase, DNA molecules are loosely attached to thesurface of a weakly cationic AP-mica, that allowsDNA fragments between anchoring points to retaina quasi-solution mobility (Lyubchenko &Shlyakhtenko, 1997).

The direct observation of dynamics ofthe cruciform

The DNA sample was prepared in TE buffer,because at these ionic conditions we observed both

Figure 2. AFM images of the samples deposited at high ionic conditions (TE buffer plus 200 mM NaCl). Images in Awere obtained for the sample with (Lk ÿ Lk0) from ÿ12 to ÿ18. AFM data for highly supercoiled sample ((LkÿLk0)from ÿ18 to ÿ24) are shown in B. The cruciforms are indicated with arrows. The scale bar for the insets represents125 nm.

64 Structure of Cruciforms

the extended and X-type geometry cruciform con-formation. DNA solution was injected into the ima-ging cell and the images of the molecules bound tothe surface were taken (Figure 3). The images Aand B are six scans from a series of scans over thesame area with three DNA molecules containingcruciforms of different geometries. Molecule 1(Figure 3A) contains the cruciform in an extendedconformation which remained unchanged in thenext image (Figure 3B). The cruciform confor-mation of molecule 2 is of the X-type with theca. 70� angle between the arms (Figure 3A). Uponrescanning, however, the arms dramatically chan-ged their orientation becoming almost parallel toeach other as shown in Figure 3B. The dataobtained for six consecutive scans are shown inFigure 3C. Each of the three zoomed regions of theinitial images shows the dynamics of one speci®ccruciform in individual columns. The images in the®rst column show that the extended conformationof cruciform 1 did not change during the entireobservation period. On the contrary, the X-typeconformation of the cruciform 2 was very mobile.The hairpin arms of cruciform 2 were initiallyquite far away from each other (with an angle ofca. 100), but then they moved (frames c and d) and®nally adopted a parallel orientation (frame f). Thebehavior of cruciform 3 was quite different.Initially, the angle between the hairpins was ca. 60�,

then the arms became parallel (frames c and d)and moved apart again at the end of the exper-iment (frames e and f). Thus, the X-type structureof cruciforms is very mobile, which explains alarge variability of the angle values shown inTable 1. A relatively low mobility of the mainDNA strands in experiments in liquid can beexplained by their stronger binding to the surfacecompared with the short hairpin arms. Unlike theX-type conformation, the extended cruciform geo-metry is less dynamic. We measured the anglesbetween the hairpins and the main DNA strands.The statistics for the opposite (vertical) acuteangles, 76(�11) and 79(�10) , indicate that the hair-pin arms of extended cruciforms have a very lowmobility. This conclusion is consistent with theAFM in situ data (Figure 3C) showing that cruci-form 1 did not change its shape during the entireobservation period.

Discussion

The geometry of the cruciforms on the surfaceand in solution

The results we obtained demonstrate clearly thata local alternative DNA structure, such as thecruciform, can be visualized with AFM. The cruci-form arms can lie on opposite sides of the mainstrand which is almost straight in the vicinity of

Table 1. The angles between the hairpin arms and the main DNA strands of the cruciform in X-type conformation

The angle between the The angle between the The number ofhairpin arms main strands molecules

Buffer solution (degrees � SD) (degrees � SD) measured

TE-buffer 71 � 36 83 � 38 38TE-buffer plus 10mM MgCl2 74 � 38 66 � 34 32TE buffer plus 200 mM NaCl 67 � 37 59 � 40 33

Figure 3(A and B) (legend on page 66)

Structure of Cruciforms 65

Figure 3. AFM images of non-dried samples of pUC8F14C plasmid DNA taken by scanning in TE buffer. A and Bare two consecutive scans over �500 mm�500 mm area. The molecules with three different conformations of the cruci-form are indicated with arrows and numbered. The time difference between images A and B is ten minutes. C, Thecruciforms in motion: consecutive scans of the fragments of molecules 1, 2 and 3 zoomed around only one cruciform;the time is indicated on the right.

66 Structure of Cruciforms

the cruciform (e.g. Figure 1B). We called this geo-metry of the cruciform extended conformation.However, more often we observed another orien-tation of the hairpin arms with rather acute anglesbetween them (e.g. Figure 1C). The main DNAstrands for this conformation of the cruciform areseverely bent. This is a compact X-conformation.This geometry, which is stabilized by cations, isvery dynamic allowing the hairpin arms to adoptalmost parallel orientation. Extended conformationis less mobile and the hairpin arms prefer anorthogonal orientation. The supercoil-stabilizedcruciform may adopt these two different confor-mations in solution as well. However, there maybe an alternative interpretation of the data, that acontinuum of conformations of the cruciform existsin solution and the two types of structuresobserved with the AFM are a consequence of depo-sition of a complex 3D-structure of the cruciformonto a plane. The geometry of immobilized cruci-form may depend on virtual orientation of themolecule relative to the surface before the depo-sition, so the two structures observed would corre-spond to two different projections of the cruciformon the plane. There may also be a selection processthat would lead to preferential binding to the sur-face of the cruciform in particular orientations ofthe arms. A close look at characteristics of the sur-face may help differentiate between these twoalternatives.

Throughout this work we used AP-mica as asubstrate for AFM studies. The AP-mica hasspeci®c surface characteristics that should be con-sidered in understanding the behavior of DNAmolecules and the cruciforms in particular at thesurface-liquid interface and their interaction withthe surface. As discussed in our previous publi-cations (Lyubchenko et al., 1992, 1993b, 1995;Lyubchenko & Shlyakhtenko, 1997), AP-mica is apositively charged surface with uniform and com-paratively low surface charge density. There arethree main consequences of these characteristics ofAP-mica. First, the immobilization of DNA on AP-mica is due to electrostatic interactions betweenprotonated amino groups of the substrate and thenegatively charged DNA backbone. Second, due touniform distribution of the charge on AP-micathere is no preferred orientation of DNA strandswhile they approach and bind to the surface.Third, owing to low surface charge density ofAP-mica the geometry of immobilized supercoiledDNA molecules is close to two-dimensional projec-tions of unperturbed DNA molecules onto a plane(Lyubchenko & Shlyakhtenko, 1997). These fea-tures of AP-mica were crucial for unambiguousdemonstration of the effect of ionic strength on thegeometry of supercoiled DNA. Moreover, we alsofound that the shape of supercoiled DNA obtainedat low and high ionic conditions are strikinglysimilar to computer simulated results obtained forsupercoiled DNA molecules of similar lengths(Vologodskii & Cozzarelli, 1994, 1996). These fea-tures of AP-mica are important prerequisites for

application of this technique to AFM studies on thestructure of supercoil-stabilized cruciforms.

Now we consider the adsorption onto AP-micaof the cruciform with a random orientation of thearms. Due to electrostatic character of DNA-surfaceinteraction both arms (of the same length) shouldbind to the surface with the same ef®ciency. Inaddition, owing to the uniform distribution of ami-nopropyl groups on AP-mica, there should be noselectivity on binding of the arms with speci®crelative orientation. So all types of the cruciformconformations should be present on the surface.However, because of ¯atness of the AP-micasurface the conformations of the cruciform that arenearly planar should have higher binding af®nitiesthan non-¯at ones. This selectivity effect dependson the strength of the arm-surface interactionwhich is a function of ionic strength. At low ionicstrength, electrostatic interaction may be suf®cientto hold the cruciform with X-type orientation ofthe arms despite its unfavorable geometry (at leastone of the arms looks away from the surface). Anincrease of ionic strength will weaken the inter-action between the arms, such that, at speci®c saltconcentration, the forces between the arms and thesurface may be insuf®cient to hold the cruciformwith unfavorable X-type conformations. At suchionic conditions only conformations with a highnumber of contacts between the surface and thecruciform remain stable. From that point of viewextended ¯at conformations of the cruciform ratherthan non-¯at X-type conformations should havehigh af®nity to the substrate. In other words, theincrease of the salt concentration should lead to theincrease of the ratio of extended conformations ofthe cruciform. So there are two main consequencesof the alternative model. First, the cruciforms withrandom orientations of the arms should beobserved with AFM. Second, the ratio of extendedconformations should increase as ionic strengthincreases. We will show below that our AFMobservations do not support these expectations.

First, not all orientations of the hairpin armshave been observed. For example, X-conformationswith antiparallel orientation of the arms relative tothe main DNA strands (see Figure 4) are very rareevents (see p. 64). A more striking example is thegeometry of the cruciform in extended confor-mations. In the majority of cases the hairpin armstend to be perpendicular to the main strand(Figure 1). Second, the dependence of the ratio ofextended conformations as a function of ionicstrength is opposite to that expected from thealternative model: the number of cruciforms inextended conformations drops considerably atelevated salt conditions.

At the same time, both sets of evidence favorour interpretation, namely that these two types ofconformations of the cruciform, the compactX-type structure and an extended conformation,exist in solution. It is important that this interpret-ation as well as the stabilization of X-type confor-mation of the cruciform in high salt solutions is in

Structure of Cruciforms 67

accord with the data obtained for a rather similarsystem, immobile four-way junctions (Cooper &Hagerman, 1987; Eis & Millar, 1993; Lilley &Clegg, 1993; Seeman & Kallenbach, 1994; Lilley,1997; Miick et al., 1997). Moreover, the AFM dataindicate that the X-conformation has a strong bendin the main DNA strand, and this AFM obser-vation is in remarkable coincidence with early gelretardation data (Gough & Lilley, 1985; Zheng &Sinden, 1988).

It is important to mention, however, the struc-tures observed with the AFM are projections onthe plane of a non-¯at three-dimensional cruciformwhich may be distorted during the deposition ontothe surface. In addition, it is formally possible thatprojections of different 3D-structures may be verysimilar and indistinguishable from the AFM, there-fore other conformations of cruciforms in additionto the extended and the X-conformation may existin solution. It is also likely that the X-conformationis indeed a family of conformers with rather closeenergies, so transition between these structures can

occur in solution at ambient conditions. Exper-imental evidence for the existence of a set differentconformations of immobile four-way junctions wasobtained recently by Miick et al. (1997).

The dynamics of supercoil-stabilized cruciforms

An interesting feature of the cruciforms emergesfrom their localization in the supercoil. As seen inFigure 2, the X-type cruciforms tend to be localizedat the loops of the plectonemic superhelix. Asshown earlier, intrinsically bent DNA regions pre-fer an apical localization in supercoiled DNA(Laundon & Grif®th, 1988). Our data suggest thatcruciforms, like bent DNA sequences, are elementsthat can dictate a spatial geometry to a DNAsuperhelix. The compact X-type structure stabilizedby cations was observed for immobile four-wayjunctions (Cooper & Hagerman, 1987; Lilley &Clegg, 1993; Seeman & Kallenbach, 1994; Lilley,

Figure 4. The cruciform and two different conformations of recombination DNA intermediates with parallel (B) andantiparallel (D) orientation of exchanging strands. The schemes of the cruciform in X-type conformation found here(B, C) and the conformation of immobile four-way junctions (D, E) are drawn according to Seeman & Kallenbach(1994).

68 Structure of Cruciforms

1997) and seems electrostatically reasonable at siteswhere DNA molecules make a U-turn.

We also found that cruciforms are dynamicstructures. According to the data obtained fordried samples, the mean value for the anglebetween the arms of the X-type cruciform is �70�and the standard deviation for this parameter is ca.50%. High mobility of the cruciform is also similarto that of immobile four-way junctions studiedby different methods in solution (Cooper &Hagerman, 1987; Eis & Millar, 1993; Lilley, 1997;Miick et al., 1997). However, the AFM allowed usto obtain both, the numbers for the mean value ofthe angle between the arms and the variability ofthis parameter. The angles between the hairpinarms are very close to that for between the mainstrands which is demonstrated by statistical anal-ysis (Table 1) and is seen in Figures 1 and 2: thepairwise values of the hairpin arm ± main DNAstrand angles in one molecule are very similar. Inmost cases these inter-arm angles for one moleculeare almost identical, indicating that the arms movein a coordinated manner. It is reasonable tosuggest that there is an interaction (stacking)between the cruciform helices at the base of thefour-way junction. This assumption is in accordwith data on chemical probing of cruciforms whichindicates that the bases at the junction are muchless reactive than those at the hairpin loops(Voloshin et al., 1989). At the same time, our exper-iments in liquid (Figure 3) show that the inter-action between the arms is not strong enough toprevent the movement of the hairpin arms if longmain arms are ®xed at the surface. In this regard,comparatively low mobility of extended confor-mations of the cruciform seems quite surprising.We have found that the hairpin arms prefer anorthogonal orientation, so acute angles betweenthe hairpin arms and the main strands are �80�and variability of this parameter is ca. 15%. Thesedata obtained for dried samples are consistent withour observations performed in solution (Figure 3).Fully extended conformation of the cruciform with�90� between the adjacent arms is a highly sym-metric structure, which can be thermodynamicallyquite stable. Note in this regard that favorable con-ditions for observation of extended conformationsof the cruciform is low ionic strength. So electro-static repulsion of the nucleotides at the joint andthe arms themselves may be two important factorsthat stabilize the fully extended con®guration andthus limit its mobility. We also cannot exclude thatfully extended con®guration of the cruciformbecause of its ¯atness has high af®nity to the sur-face, so this selectivity may lead to an elevatedproportion of this speci®c conformation of the cru-ciform on the surface in comparison with theirpopulation in solution.

An alternative explanation for the high variabil-ity of the inter-arm angles for X-type conformationmust be also considered. That is, in solution thecruciform adopts a set of interchangable confor-mations of the X-type family differing in the inter-

arm angles. Each conformer has a mobility com-parable with that of the extended conformation,but two-dimensional projection of the entire familywould give a broad distribution of the angles thatcould be interpreted as a high mobility of oneX-conformation. However, in the experiments inliquid spontaneous movement of the hairpin armswas observed over a very broad range (Figure 3),thus the barrier between these conformations mustbe so low that it can be overcome by a thermalmotion of the arms. Therefore, from this point ofview, the X-type conformers are indistinguishableand uni®cation of them under one X-type confor-mation is a rather natural ®rst approximation.

The conformation of cruciforms and itsrelation to DNA recombination and regulationof DNA replication and transcription

As mentioned above, the supercoil-inducedcruciform can be considered a putative model ofthe Holliday junction. First, the symmetry of thesequence of the cruciform by de®nition permits thebranch migration. Second, genomic DNA is orga-nized in topological domains with unrestrainedsupercoiling (Sinden, 1994) and topoisomerases areinvolved at the steps of homologous pairing andDNA strand transfer (Kowalczykowski, 1994),suggesting that DNA supercoiling is an importantfactor for the DNA recombination.

Currently existing models for recombination ofDNA are shown schematically in Figure 4. Inaddition to extended conformation of the cruciform(A) two different orientations of the arms ((B) and(D)) that correspond respectively to parallel (B)and antiparallel (D) orientations of exchangingarms are possible (Sigal & Alberts, 1972; Seeman &Kallenbach, 1994; Sobell, 1974). The model (B) witha parallel orientation of exchanging strandssuggested by Sigal & Alberts (1972) easily explainsthe strand exchange by a branch migration mech-anism. According to Table 1 (column 1), the inter-arm angles 1 and 2 are rather acute (seeFigure 4(C)). The molecules with obtuse angles 1and 2, as shown in Figure 4(E), were very rareevents in the AFM images, and molecules with clo-ser distances between the hairpins and the DNAstrands were not observed. So the AFM data areconsistent with the model (B) of geometry of cruci-forms which is equivalent to the Holliday junctionwith a parallel orientation of exchanging strands. Itis surprising that a seemingly similar type of struc-ture, an immobile four-way junction, can adopt thegeometry (D) which corresponds to antiparallelstrand orientation (Figure 4(E); Cooper &Hagerman, 1987; Lilley & Clegg, 1993; Seeman &Kallenbach, 1994; Lilley, 1997). Although antiparal-lel and parallel states of immobile junctions arethermodynamically very close ( Lu et al., 1991) andthe barrier between the two conformations is nothigh (Srinivasan & Olson, 1994), only an antiparal-lel strand orientation in immobile junctions hasbeen observed (Cooper & Hagerman, 1987; Lilley

Structure of Cruciforms 69

& Clegg, 1993; Seeman & Kallenbach, 1994; Lilley,1997; Eis & Millar, 1993). The migration of thecrossover site of antiparallel conformation of thejunction requires considerable conformationalchanges, but once the helical arms are aligned in aparallel manner, the branch migration proceedswith relatively little energetic costs (Srinivasan &Olson, 1994). The nature of a supercoiled topologi-cal domain and the propensity for bent DNAs suchas a cruciform (Gough & Lilley, 1985; Zheng &Sinden, 1988) to orient themselves at the tip of aloop (Figure 2) strongly favor the parallel geome-try. Probably, the constraints placed on the struc-ture of the Holliday junction by topology andsupercoiling outweigh a small effect of an energydifference in the parallel and antiparallel structuresof four-way junctions. Thus, DNA supercoilingand the topological domain organization inchromosomes may dictate the structure of a Holli-day junction.

A number of other biological implications followfrom the features described for cruciforms. Theability of cruciforms and four-way DNA junctions(Cooper & Hagerman, 1987; Lilley & Clegg, 1993;Seeman & Kallenbach, 1994; Lilley, 1997;Srinivasan & Olson, 1994; Eis & Millar, 1993;Petrillo et al., 1988) to adopt different confor-mations might be functionally important. TheseDNA structures are the targets for different typesof proteins such as recombination enzymes(Parsons et al., 1995; Yu et al., 1997), topoisomerases(Roca et al., 1993), and architectural proteins suchas HMG1 and histones H1 and H5 (Timsit &Moras, 1996), so that the conformational ¯exibilityof DNA is an important factor for accommodatingsuch a plethora of proteins. For example, the varia-bility of the cleavage patterns of the same four-way junctions for resolving enzymes from differentspecies (Lilley, 1997) is easily explained by adynamic character of the four-way junction.Another example is the interaction of four-wayjunctions with the E. coli RuvA, B and C (Parsonset al., 1995; Yu et al., 1997). An extended confor-mation of the four-way junction (Figure 4(A)) isthe target for RuvA and RuvB proteins at theinitial step of the recombination process, but the®nal step, resolving the junction with RuvC,requires a compact X-type conformation (Lilley &Clegg, 1993).

There are other known examples of the role ofcruciforms in biology. The formation of supercoil-induced cruciforms changes the level of unrest-rained torsional stress in topological domains ofthe genome, which could in¯uence gene regulationor replication. The ability of cruciforms to movemay facilitate their interaction with speci®c pro-teins, and this may be important for regulation ofDNA replication (Noirot et al., 1990) and transcrip-tion (Glucksmann et al., 1992), where roles for cru-ciforms have been demonstrated. AFM has alreadybeen successfully applied to structural studies ofcomplexes of linear DNA with a number of nucleo-protein complexes (Bustamante et al., 1994;

Hansma & Hoh 1994; Lyubchenko et al., 1995).Now that the structural information is availablefrom AFM studies of supercoiled DNA, the tech-nique might be applied to complexes of super-coiled DNA with the key proteins of transcription,replication and recombination.

Materials and Methods

DNA preparation

Plasmid pUC8F14C with a 106 bp inverted repeat(Sinden, 1994) was grown in E. coli strain HB101 andisolated by cell lysis with 1% (v/v) Brij 58, 0.4% (w/v)deoxycholate, followed by CsCl centrifugation. Topoi-somers of particular superhelical densities were gener-ated by incubating supercoiled plasmid with atopoisomerase extract from HeLa cells in the presence ofdifferent concentrations of ethidium bromide (Kochel &Sinden, 1988). The superhelical densities of closed circu-lar molecules (about 80% of DNA molecules in topoi-somer fractions) were estimated by 1.5% (w/v) agarosegel electrophoresis in the presence of 3, 9 and 25 mg/mlchloroquine phosphate. To determine the superhelicaldensity required for cruciform formation, a mixture oftopoisomers (about 0.25 mg each fraction) was separatedin a 1.5% agarose two-dimensional gel (15 cm � 15 cm)at 25�C, 4 V/cm using Topo TAE buffer (40 mM Tris-HCl, 5 mM sodium acetate, 1 mM EDTA, pH 8.3) for the®rst dimension and the same buffer containing 30 mg/mlchloroquine phosphate for the second dimension.

AFM procedure

Throughout this work only mica functionalized with3-aminopropyltriethoxy silane (AP-mica) was used asa substrate (Lyubchenko et al., 1992, 1993a,b, 1995;Lyubchenko & Shlyakhtenko, 1997). For imaging in air,10 ml of the DNA (0.5mg/ml) in TE buffer was placedonto pieces of AP-mica for two minutes, thoroughlyrinsed with deionized water (ModuPure Plus, Continen-tal Water System Corp., San Antonio, TX) and argon-dried. NanoProbe TESP (Digital Instruments, Inc., SantaBarbara, CA) tips were used for imaging in air. Thetypical tapping frequency was 240 to 280 kHz; scanningrate was 1.97 Hz. For imaging in liquid, Si3N4, 100 mmlong, NanoProbe tips mounted on a glass tip holder fortapping mode (Digital Instruments, Inc., Santa Barbara,CA) were used. The tip was manually approached tothe AP-mica surface mounted on the sample stage on themicroscope until the distance between the tip and thesurface was 20 to 50 mm. The DNA solution (25 to 50 ml)was injected into a gap between the mica surface and thetip holder and the tip was brought into a tapping rangeunder control of the instrument. The scanning par-ameters were as follows: frequency 8 to 9 kHz, scanningrate 1.97 Hz. The images were taken in the topographicmode. A NanoScope III MultiMode system (DigitalInstruments, Inc., Santa Barbara, CA) operated intapping mode was used for these experiments.

Analysis of the AFM data

NIH image 1.60 software was used for measurementsof the angles between the arms. Statistical analysis of thedata was performed by using Kaleidagraph 3.0.1. soft-ware. The number of the molecules used for measure-ments of the angles for X-type conformation of the

70 Structure of Cruciforms

cruciforms are indicated in Table 1 (the last column).Similar dataset (31 molecules) was used for measuringthe angles between arms for extended conformations ofthe cruciform.

Acknowledgments

This work was supported in part by grants GM 54991from NIH and Digital Instruments, Inc. We are gratefulto G. Manning, W. Olson, N Seeman and V. Zhurkin forreading the manuscript and valuable comments.

References

Binnig, G., Quate, C. F. & Gerber, C. H. (1986). Atomicforce microscope. Phys. Rev. Letters, 56, 930±933.

Bustamante, C., Erie, D. A. & Keller, D. (1994). Bio-chemical and structural applications of scanningforce microscopy. Curr. Opin. Struct. Biol. 3, 750±760.

Cooper, J. P. & Hagerman, P. J. (1987). Gel electrophor-etic analysis of the geometry of a four-way junction.J. Mol. Biol. 198, 711±719.

Eis, P. S. & Millar, D. P. (1993). Conformational distri-butions of a four-way DNA junction revealed bytime-resolved ¯uorescence resonance energy trans-fer. Biochemistry, 32, 13852±13860.

Gierer, A. (1966). Model for DNA and protein inter-action and the function of the operator. Nature, 212,1480±1481.

Glucksmann, M. A., Markiewicz, P., Malone, C. &Rothman-Denes, L. R. (1992). Speci®c sequencesand a hairpin structure in the template strand arerequired for N4 virion RNA polymerase promoterrecognition. Cell, 70, 491±500.

Gough, G. W. & Lilley, D. M. J. (1985). DNA bendinginduced by cruciform formation. Nature, 313, 154±156.

Hansma, H. G. & Hoh, J. (1994). Biomolecular imagingwith the atomic force microscopy. Annu. Rev. Bio-phys. Biochem. Struct. 23, 115±139.

Kochel, T. & Sinden, R. R. (1988). Analysis of tri-methylpsoralen photoreactivity to Z-DNA providesa general in vivo assay for Z-DNA: analysis of thehypersensitivity by (GT)n B-Z junctions. Biotechni-ques, 6, 532±543.

Kornberg, A. & Baker, T. A. (1992). DNA Replication,Freeman, New York.

Kowalczykowski, S. C. (1994). In vitro reconstituion ofhomologous recombination reactions. Experimentia,50, 204±215.

Laundon, C. H. & Grif®th, J. D. (1988). Curved helixsegments can uniquely orient the topology of super-twisted DNA. Cell, 52, 545±549.

Lilley, D. M. J. (1997). All change at Holliday junction.Proc. Natl Acad. Sci. USA, 94, 9513±9515.

Lilley, D. M. J. & Clegg, R. M. (1993). Parallel and anti-parallel Holliday junctions differ in structure andstability. Annu. Rev. Biophys. Biomol. Struct. 22, 299±328.

Lilley, D. M. J. & Kemper, B. (1984). Cruciform-resolvaseinteractions in supercoiled DNA. Cell, 36, 413±422.

Lu, M., Guo, Q., Seeman, N. C. & Kallenbach, . N. R.(1991). Parallel and antiparallel Holliday junctionsdiffer in structure and stability. J. Mol. Biol. 221,1419±1432.

Lyubchenko, Y. L. & Shlyakhtenko, L. S. (1997). Visu-alization of supercoiled DNA with atomic forcemicroscopy. Proc. Natl Acad. Sci. USA, 94, 496±501.

Lyubchenko, Y. L., Gall, A. A., Shlyakhtenko, L. S.,Harrington, R. E., Jacobs, B. L., Oden, P. I. &Lindsay, S. M. (1992). Atomic force microscopy ima-ging of dsDNA and RNA. J. Biomol. Struct. Dynam.10, 589±606.

Lyubchenko, Y. L., Oden, P. I., Lampner, D. & Lindsay,S. M. (1993a). Atomic Force Microscopy of DNAand bacteriophage in air, water and propanol: roleof adhesion forces. Nucl. Acids Res. 21, 1117±1123.

Lyubchenko, Y. L., Shlyakhtenko, L. S., Harrington,R. E., Oden, P. I. & Lindsay, S. M. (1993b). Atomicforce microscopy of long DNA: imaging in air andunder water. Proc. Natl Acad. Sci. USA, 90, 2137±2140.

Lyubchenko, Y. L., Jacobs, B. L., Lindsay, S. M. &Stasiak, A. (1995). Atomic force microscopy ofnucleoprotein complexes. Scann. Microsc. 9, 705±727.

McClellan, J. A. & Lilley, D. M. J. (1987). A two-stateconformational equilibrium for alternating (A-T)nsequences in negatively supercoiled DNA. J. Mol.Biol. 197, 707±721.

Miick, S. M., Fee, R. S., Millar, D. P. & Chazin, W. J.(1997). Crossover isomer bias is the primarysequence-dependent property of immobilized Holli-day junctions. Proc. Natl Acad. Sci. USA, 94, 9080±9084.

Mizuuchi, K., Mizuuchi, M. & Gellert, M. (1982). Cruci-form structures in palindromic DNA are favored byDNA supercoiling. J. Mol. Biol. 156, 229±243.

Noirot, P., Bargonetti, J. & Novick, R. P. (1990).Initiation of rolling-circle replication in pT181 plas-mid: initiator protein enhances cruciform extrusionat the origin. Proc. Natl Acad. Sci. USA, 87, 8560±8564.

Palecek, E. (1992). Probing of DNA structure in cellswith osmium tetraoxide-2. 20-bipyridine. MethodsEnzymol. 212, 305±318.

Parsons, C. A., Stasiak, A., Bennett, R. J. & West, S. C.(1995). Structure of a multisubunit complex thatpromotes DNA branch migration. Nature, 374, 375±378.

Petrillo, M. L., Newton, C. J., Cunningham, R. P., Ma,R. I., Kallenbach, N. R. & Seeman, N. C. (1988). Lig-ation and ¯exibility of four-arm DNA junctions. Bio-polymers, 27, 1337±1352.

Roca, J., Berger, J. M. & Wang, J. C. (1993). On the sim-ultaneous binding of eukaryotic DNA topoisome-rase II to a pair of double-stranded DNA helices.J. Biol. Chem. 268, 14250±14255.

Seeman, N. C. & Kallenbach, N. R. (1994). DNAbranched junctions. Annu. Rev. Biophys. Biomol.Struct. 23, 53±86.

Sigal, N. & Alberts, B. (1972). Genetic recombination:the nature of a crossed strand-exchange betweentwo homologous DNA molecules. J. Mol. Biol. 71,789±793.

Sinden, R. R. (1994). DNA Structure and Function, pp.134±178, Academic Press, San Diego.

Sinden, R. R., Broyles, S. S. & Pettijohn, D. E. (1983).Perfect palindromic lac operator DNA sequenceexists as a stable cruciform structure in supercoiledDNA in vitro but not in vivo. Proc. Natl Acad. Sci.USA, 80, 1797±1801.

Structure of Cruciforms 71

Sobell, H. M. (1974). Concerning the stereochemistry ofstrand equivalence in genetic recombination. InMechanisms in Recombination (Grell, R. F., ed.),pp. 433±438, Plenum Publishing Corp., New York.

Srinivasan, A. R. & Olson, V. K. (1994). Computermodels of DNA four-way junctions. Biochemistry,33, 9389±9404.

Timsit, Y. & Moras, D. (1996). Cruciform structures andfunctions. Qaurt. Rev. Biophys. 29, 279±307.

Vologodskii, A. V. (1992). Topology and Physics of Circu-lar DNA, pp. 31±153, CRC Press, Boca Raton.

Vologodskii, A. V. & Cozzarelli, N. R. (1994). Annu. Rev.Biophys. Biomol. Struct. 23, 609±643.

Vologodskii, A. V. & Cozzarelli, N. R. (1996). Effect ofsupercoiling on the juxtaposition and relative orien-tation of DNA sites. Biophys. J. 70, 2548±2556.

Voloshin, O. N., Shlyakhtenko, L. S. & Lyubchenko, Y. L.(1989). Localization of melted regions in supercoiledDNA. FEBS Letters, 243, 377±380.

Yu, X., West, S. C. & Egelman, E. H. (1997). Structureand subunit composition of the RuvAB-Hollidayjunction complex. J. Mol. Biol. 266, 217±222.

Zheng, G. & Sinden, R. R. (1988). Effect of base compo-sition at the center of inverted repeated DNAsequences on cruciform transition in DNA. J. Biol.Chem. 263, 5356±5361.

Edited by I. Tinoco

(Received 21 November 1997; received in revised form 7 April 1998; accepted 8 April 1998)

72 Structure of Cruciforms