J. Mol. Biol. (1996) 257, 1052–1069

Solution Structure of a Pyrimidine·Purine·PyrimidineTriplex Containing the Sequence-specific IntercalatingNon-natural Base D 3

Edmond Wang 1, Karl M. Koshlap 1, Paul Gillespie 2, Peter B. Dervan 2

and Juli Feigon 1*

We have used NMR spectroscopy to study a pyrimidine·purine·pyrimidine1Department of ChemistryDNA triplex containing a non-natural base, 1-(2-deoxy-b-D-ribofuranosyl)-and Biochemistry and4-(3-benzamido)phenylimidazole (D3), in the third strand. The D3 base hasMolecular Biology Institute

University of California been previously shown to specifically recognize T·A and C·G base-pairs viaLos Angeles, CA 90095, USA intercalation on the 3' side (with respect to the purine strand) of the target

base pair, instead of forming sequence-specific hydrogen bonds. 1H2Division of Chemistry and resonance assignments have been made for the D3 base and most of theChemical Engineering non-loop portion of the triplex. The solution structure of the triplex wasCalifornia Institute of calculated using restrained molecular dynamics and complete relaxationTechnology, Pasadena matrix refinement. The duplex portion of the triplex has an over-all helicalCA 91125, USA structure that is more similar to B-DNA than to A-DNA. The three

aromatic rings of the D3 base stack on the bases of all three strands andmimic a triplet. The conformation of the D3 base and its sequence specificityare discussed.

7 1996 Academic Press Limited

Keywords: DNA; triplex; NMR; 2DNMR; triple helix; triple strand;*Corresponding author intercalator

Introduction

Canonical nucleic acid triplexes are formed by thebinding of a single strand in the major groove of ahomopurine·homopyrimidine duplex (Felsenfeldet al., 1957; Sun & Helene, 1993; Moser & Dervan,1987; Wells et al., 1988). Two general motifs fortriplexes exist, parallel and anti-parallel, which aredefined by their third strand orientation relative tothe homopurine strand in the Watson–Crick pairedtarget duplex. In parallel triplexes, the third strandis homopyrimidine and binds parallel to thehomopurine strand of the duplex. A protonated Cin the third strand binds to a G in the duplex viaHoogsteen base-pairing forming a C+·G·C triplet.

Similarly, a T in the third strand recognizes an A inthe duplex forming a T·A·T triplet (de los Santoset al., 1989; Le Doan et al., 1987; Moser & Dervan,1987; Rajagopal & Feigon, 1989a,b). In anti-paralleltriplexes, the third strand is homopurine or amixture of purines and thymines, and bindsanti-parallel to the homopurine strand of theduplex. The G·C and A·T base-pairs are recognizedby a G and A (or T), respectively, forming G·G·C andA·A·T (or T·A·T) triplets via reverse Hoogsteenbase-pairing (Beal & Dervan, 1991; Broitman et al.,1987; Lipsett, 1964; Radhakrishnan et al., 1991).

The possibility of targeting genes with highsequence specificity makes triplexes potentiallyversatile biochemical tools. Triplexes have been usedfor gene regulation (Cooney et al., 1988; Degols et al.,1994; Grigoriev et al., 1992; Maher, 1992), DNAisolation and detection (Ji et al., 1994; Olivas &Maher, 1994) and DNA modification (Havre et al.,1993; Luebke & Dervan, 1991). While most DNA-binding proteins are limited to recognition of a fewbase-pairs of DNA, triplex formation can easilyencompass sequences longer than ten base-pairswith a mismatch discrimination comparable to thatof duplex formation (Greenberg & Dervan, 1995;Mergny et al., 1991; Roberts & Crothers, 1991).Triplexes could potentially be designed to recognize

Abbreviations used: D3, 1-(2-deoxy-b-D-ribofuranosyl)-4-(3-benzamido)phenylimidazole; DTAtriplex, d(AGATAGAACCCCTTCTATCTTATATCTD3-TCTT; NOE, nuclear Overhauser enhancement; NOESY,NOE spectroscopy; P.COSY, purged correlationspectroscopy; HOHAHA, Homonuclear-Hartmann-Hahn; HOENOE, HOHAHA edited NOESY; HMQC,Heteronuclear multiple quantum coherencespectroscopy; RMSD, root-mean-squared deviation;FPLC, fast protein liquid chromatography; VDW, vander Waals.

0022–2836/96/151052–18 $18.00/0 7 1996 Academic Press Limited

Solution Structure of a D3-containing DNA Triplex 1053

Figure 1. (a) Chemical structure of the D3 base with theatom numbering scheme. The torsion angles N3A-C4A-C1B-C2B (u1), C2B-C3B-N-C (u2) and N-C-C1C-C2C (u3)are indicated. (b) Folding schematic drawing of the DTAtriplex with the residue numbering scheme used in thetext. The Watson–Crick base-pairs are indicated by filledcircles, the Hoogsteen base-pairs are indicated by opencircles, and the protonated C Hoogsteen base-pairs areindicated with a +.

NMR spectroscopy demonstrated that this D3-con-taining oligonucleotide forms an intramoleculartriplex, with a CCCC and a TATA loop connect-ing the Watson–Crick and the Hoogsteen-pairedstrands, respectively (Koshlap et al., 1993) (Fig-ure 1(b)). This study also revealed that instead ofhydrogen-bonding to the Watson–Crick base-pair,the D3 base intercalates between its target T·Abase-pair and the adjacent 3' T·A·T triplet (i.e 3' withrespect to the purine strand). Here we present the1H NMR assignments and solution structure of theDTA triplex. The structure reveals that intercalationof the D3 base is readily accommodated and that theD3 base mimics a triplet. The over-all helicalstructure is more similar to B-DNA than to A-DNA.The effect of the D3 base intercalation in a triplexstructure is discussed, and the results are comparedwith those from fiber diffraction and NMR-derivedtriplex structures.

Results

1H NOESY spectrum of exchangeableresonances of DTA

The imino region of a NOESY spectrum of theDTA triplex in water at 1°C is shown in Figure 2.This spectrum exhibits features typical of pre-viously characterized intramolecular triplexes, e.g.imino proton resonances corresponding to bothWatson–Crick and Hoogsteen hydrogen-bondedbase-pairs are observed (Sklenar & Feigon, 1990)(Figure 2(a)). Crosspeaks from the downfield-shifted amino protons, characteristic of protonatedcytosine bases, are also observed (Figure 2(b)).Assignment of the spectrum followed protocolspreviously developed in our laboratory (Macayaet al., 1991, 1992b; Rajagopal & Feigon, 1989a;Sklenar & Feigon, 1990) and are given in Table 1.Analysis of the NOESY spectrum confirms thatthe DTA oligonucleotide forms an intramoleculartriplex. In Figure 2(a), the Watson–Crick imino-imino sequential connectivities are indicated abovethe diagonal and the Hoogsteen imino-iminosequential connectivities are indicated below thediagonal. There is an unexpected break in thesequential connectivities between T4 and T16. Noimino-imino crosspeaks are observed to C+26 due tochemical exchange with water during the mixingtime. However, the C+26 imino was readilyidentified by the crosspeaks to its own aminoresonances (Figure 2(b)). T27 and T25 wereidentified on the basis of some of the intra-tripletNOE crosspeaks characteristic of T·A·T triplets, e.g.both the Watson–Crick and the Hoogsteen T iminoprotons have crosspeaks to the central A aminoresonances (Sklenar & Feigon, 1990). Of particularinterest are the crosspeaks observed between iminoand D3 protons. As is indicated in Figure 2(b), theT16 imino proton of the DTA triplex has a crosspeakto the D3 H3/5C resonance. Crosspeaks are alsoobserved between the T29 imino resonance and theD3 H5B and H6B protons.

a single site in an entire genome (Strobel & Dervan,1991; Strobel et al., 1991). Unfortunately, recognitionby the third strand is limited to only two of the fourpossible base-pairs. To some extent a T·A base-paircan be recognized by a G (Griffin & Dervan, 1989)and possibly a C (Belotserkovskii et al., 1990), anda C·G base-pair can be recognized by a T (Yoonet al., 1992), but triplexes containing these non-canonical triplets are less stable. Also, since a C andT already recognize a G·C and A·T base-pair,respectively, sequence specificity is lost whenemploying such alternative triplets.

The non-natural deoxyribonucleoside, 1-(2-de-oxy-b-D-ribofuranosyl)-4-(3-benzamido)phenylimi-dazole (D3) (Figure 1(a)) was designed to recognizea C·G base-pair by forming specific hydrogenbonds (Griffin et al., 1992). However, chemicalfootprinting studies found that when the D3 base isincorporated into a pyrimidine oligonucleotide itrecognizes both T·A and C·G base-pairs (Griffinet al., 1992). Moreover, recognition was found to bestrongly dependent on the identity of the 3'neighboring triplet (i.e. 3' with respect to the purinestrand) but only weakly dependent on the identityof the 5' neighboring triplet (Griffin et al., 1992;Kiessling et al., 1992). In order to investigate thebinding mode of the D3 base targeted to aWatson–Crick T·A base-pair, the oligonucleotidewith the sequence d(AGATAGAACCCCTTC-TATCTTATATCTD3TCTT), (DTA triplex) was syn-thesized (the bases in the two loops are underlinedand the D3 base and the target T·A base-pair are inbold). This was designed to be similar to sequencesthat had previously been shown to fold into stableintramolecular triplexes (Macaya et al., 1991c, 1992b;Sklenar & Feigon, 1990; Wang et al., 1992). A pre-liminary study using one- and two-dimensional

Solution Structure of a D3-containing DNA Triplex1054

Figure 2. 1H 2D NOESY spectrum of the DTA triplexin 90%H2O/10%2H2O at 1°C with a mixing time of100 ms. (a) The imino-imino region. The Watson–Crickimino sequential connectivities are indicated abovethe diagonal, and the Hoogsteen imino sequentialconnectivities are indicated below the diagonal. Themissing sequential crosspeak between the T4 and T16iminos is indicated by a box and connected by brokenlines. (b) The imino-aromatic, amino region. Thecrosspeaks from the T16 and T29 iminos to the D328aromatic protons are indicated. The C+26 imino-aminocrosspeaks are also indicated. The NOESY spectrum wasacquired with a sweep width of 12048 Hz in bothdimensions, 2048 and 300 complex points in t2 and t1,respectively, and 64 scans per t1 block. In t2 600 pointswere apodized using a 60° phase-shifted squared sinebellwith a skew factor of 0.8; the baseline was flattenedwith a first-order polynomial. In t1 300 points wereapodized using a 60° phase-shifted squared sinebell witha skew factor of 0.8, and then zero-filled to 2048 complexpoints.

Assignments of thenon-exchangeable resonances

Regions of a two-dimensional 2H2O NOESYspectrum of the DTA triplex at 25°C are shown inFigures 3 and 4. Except for the D3 base, assign-ment of the DTA triplex could largely be done aspreviously described for intramolecular triplexesusing NOESY, P.COSY, HOHAHA and HOENOEspectra (Macaya et al., 1992b). Assignments wereobtained for all of the base, H1', H2',2" and H3'resonances in the triplex, except for the nucleotidesin the TATA loop and several of the nucleotides inthe two triplets at the 5' end (with respect to thepurine strand) of the triplex. The assignments aregiven in Table 1.

The assignments of the D3 base protons wereobtained by analysis of the data from P.COSY,NOESY and 1H-13C HMQC NMR experiments. Inthe P.COSY spectrum (Marion & Bax, 1988; Mueller,1987) of the triplex (data not shown), four unusualcrosspeaks are observed in the H1'-H1' region,which are absent in the P.COSY spectra of othertriplexes. These crosspeaks form two spin systemsof three protons each. In the NOESY spectrum, thesame crosspeaks are observed with additionalcrosspeaks between the outer protons within eachspin system (Figure 3). These two sets of spins mustbe attributed to protons of rings B and C of the D3

base. Protons H2A and H5A were identified by theiraromatic-aromatic sequential connectivity to T29H6(Figure 3, peaks a and b), and confirmed by a weakcrosspeak between the two. In the aromatic-methylregion of the NOESY, there is a proton that has acrosspeak to only T4Me and was identified as H2B(Figure 4, peak e). This, as well as other D3 protonresonance assignments, was supported by a 1H-13CHMQC spectrum (Bax et al., 1983; data not shown).If the relative orientation of rings A and B of D328were opposite from what is depicted in Figure 1(a),one would also expect to see an NOE crosspeakbetween protons H2B and H5A, which is notobserved. Likewise, this alternative orientationwould cause any NOE crosspeaks between H5Aand both H6B and H5B to be either very weak ornon-existent. Thus, the lack of any NOE crosspeaksother than that of T4Me to H2B, as well as strongcrosspeaks between both the central and outerprotons of one of the spin systems mentioned aboveand the resonance identified as either H5A or H2Adistinguishes ring B from C, as well as protons H6Bfrom H4B, and H5A from H2A.

The remaining three spin system must beassigned to the C ring. Although the C ring is afive-spin system, resonances for H2C and H6C andfor H3C and H5C are degenerate. This is apparentlydue to motional averaging (ring flips). H3/5C wasassigned from the P.COSY spectrum as the centralspin in the three-spin system. H4C and H2/6C aredistinguished on the basis of their NOESY cross-peaks, i.e. crosspeaks between H4C to both A17H8and T16H1', and H2/6C to A17H2 (Figure 3, peakst, m and s).

Solution Structure of a D3-containing DNA Triplex 1055

Table 1. DTA triplex chemical shiftsResidue Imino Aminos H6,H8 H2,H5,Me H1' H2' H2" H3'

A1 —a 6.99, 7.73 7.73G2 12.69 —A3 — 7.63, 7.71 7.43 7.56 5.92 1.89 2.51 4.76T4 13.34 — 6.32 0.34 5.61 1.81 2.15 4.70A5 — 7.08, 7.47 7.67 7.14 6.05 2.59 2.90 4.86G6 12.46 7.20 — 5.76 2.38 2.87 4.85A7 — 7.52, 7.71 7.37 7.35 5.84 2.44, 2.72 4.88A8 — 7.52, 7.72 7.77 7.94 6.04 2.44, 2.57 4.77C9 — 7.41 5.50 5.75 1.97, 2.32 4.66C10 — 7.73 5.97 6.04 2.01, 2.36 4.50C11 — 7.55 5.59 5.73 1.88, 2.34 4.66C12 — 7.94 6.12 6.21 2.41, 2.67 4.73T13 14.89 — 7.66 1.79 6.09 2.26, 2.71 4.85T14 14.09 — 7.53 1.65 6.07 2.28, 2.61 4.90C15 — 7.12, 8.18 7.55 5.53 5.92 2.02, 2.45 4.74T16 13.12 — 7.37 1.63 6.02 2.36, 2.51 4.98A17 — 6.45, 6.99 8.30 7.55 5.99 2.65, 2.80 4.90T18 13.73 — 7.22 1.29 5.92 2.18, 2.50 4.78C19 — 6.89, 8.17 7.49 5.45 5.91 2.01, 2.46 4.60T20 14.45 — 1.53T21 — 1.71A22 —T23 —A24 —T25 13.98 —C+26 15.56 9.17, 10.04 7.92 5.91 6.09 2.18, 2.66 4.57T27 12.81 — 7.65 1.74 6.34 2.33, 2.53 5.04D328 — — — — 6.04 2.55, 3.05 4.88T29 13.07 — 7.31 1.64 6.16 2.26, 2.59 4.92C+30 15.37 9.38, 9.76 7.86 5.71 6.12 2.20, 2.67 4.73T31 13.34 — 7.70 1.70 6.18 2.27, 2.69 4.87T32 12.51 — 7.53 1.59 6.33 2.25, 2.30 4.55

D328 aromatic resonancesH2A = 8.22 H2B = 7.17 H2/6C = 7.08H5A = 7.45 H4B = 6.22 H3/5C = 6.46

H5B = 6.55 H4C = 6.16H6B = 6.88

Chemical shifts for the exchangeable and non-exchangeable resonances are at 1°C and 25°C,respectively. Not all resonances in the loops and in the two 5'-terminal triplets could be assigned dueto chemical exchange broadening.

a Not applicable.

Evidence for D 3 intercalation

Numerous NOE crosspeaks are observed be-tween protons from all three aromatic rings ofthe D3 base and other nucleotides in the DTA triplex.In all cases, the observed NOE crosspeaks are toprotons of either the T29·A5·T16 triplet or theT4·A17 base-pair. No crosspeaks are seen betweenthe D3 base protons and those of the T27·A3·T18triplet. Figure 3 shows the H1', H5 aromaticregion of a 2H2O NOESY spectrum of the DTAtriplex at 25°C. The lettered crosspeaks are be-tween protons of the D3 ring systems and thoseof either the targeted T4·A17 base-pair or theadjacent T29·A5·T16 triplet. Specifically, crosspeaksa through c involve the A ring of D328, cross-peaks d through k involve the B ring, andcrosspeaks l through u involve the C ring.Crosspeak v is the large T27H3'-T27H6 crosspeak,indicating that the T27 sugar has an N-typepucker. In addition, each ring primarily has NOEsto a different strand of the triplex. All crosspeaksfrom the A ring are to the third strand base, T29.

All crosspeaks from the B ring are to theWatson–Crick purine strand (T4 and A5), exceptcrosspeak d, which is to T29. All crosspeaks fromthe C ring are to the Watson–Crick pyrimidinestrand (T16 and A17) except crosspeak l, which isto A5.

Figure 4 shows the H2',2",Me-aromatic, H1'region of the 2H2O NOESY spectrum. Crosspeaks athrough c involve the A ring of D328, crosspeaks dthrough h involve the B ring and crosspeaks ithrough m involve the C ring. Just as in the H1',H5-aromatic region, each ring primarily has NOEsto a different strand of the triplex. The A ring hascrosspeaks to only T29 (third strand), the B ring hascrosspeaks to only T4 (Watson–Crick purine strand)and the C ring has crosspeaks to only T16(Watson–Crick pyrimidine strand). The one excep-tion is the T4Me, which has crosspeaks to bothA and B rings. Both the T4Me and the T4H6resonances have large upfield shifts compared tothe usual triplex TMe and TH6 chemical shifts,as is illustrated by the position of their crosspeak(Figure 4, peak n); for comparison, the crosspeak

Solution Structure of a D3-containing DNA Triplex1056

Figure 3. H1', H5 aromatic regionof the 2D NOESY spectrum ofthe DTA triplex in 2H2O at 25°Cand tm = 200 ms. The sequentialassignments are indicated withcontinuous lines. Disruptions inthe sequential connectivities be-tween T4 and A5, T16 and A17, andT27 and D328 are connected withbroken lines, and the intra-residue(aromatic-H1') crosspeak is labeledwith the residue number. Themissing crosspeaks are indicatedwith boxes. Inter-residue crosspeaksinvolving the D3 aromatic ringsare: a, T29H6-H2A; b, T29H6-H5A; c, T29H1'-H5A; d, T29H1'-H6B; e, T4H1'-H5B; f, T4H1'-H4B;g, T4H6-H5B; h, T4H6-H6B; i, H6B-A5H8; j, H5B-A5H8; k, H4B-A5H8;l, H3/5C-A5H2; m, T16H1'-H4C;n, T16H1'-H3/5C; o, H4C-T16H6;p, H3/5C-T16H6; q, A17H1'-H3/5C;r, H3/5C-A17H2; s, H2/6C-A17H2;t, H4C-A17H8; and u, H3/5C-A17H8. Crosspeak v is the large,intra-residue T27H3'-H6 NOE.Intra- and inter-ring crosspeaks ofthe D3 base are marked with a *.The spectrum was acquired with aspectral width of 5000 Hz in bothdimensions; 1024 and 400 complexpoints were collected in t2 and t1,respectively, with 32 scans per t1

block. In both dimensions 400 pointswere apodized using a 45°-phase-shifted squared sinebell; both di-mensions were zero-filled to 2048complex points.

between T18Me and T18H6 is also labeled (Fig-ure 4, peak o).

Just as in the imino-imino region of the H2ONOESY spectrum, some disruptions in the sequen-tial connectivities are observed. No base-H1' orbase-H2',H2" sequential connectivities are observedbetween D328 and T27, between T4 and A5, orbetween A17 and T16 (Figure 3).

NOE restraints

A total of 728 distance restraints were used for therefinement steps prior to relaxation matrix refine-ment. Of these, 62 are hydrogen-bond restraints, 408are derived from 2H2O NOESY spectra, 132 arederived from the H2O NOESY spectrum, and 126are repulsive restraints derived from predicted2H2O NOESY crosspeak volumes. The distributionof distance restraints by residue is shown in Fig-ure 5(a). In Figure 5(b), the distance restraints areclassified as intra-residue, sequential and inter-residue (involving residues other than neighboringresidues) restraints. For relaxation matrix refine-ment, the 2H2O and repulsive restraints were

converted to volume restraints at all five NOESYmixing times yielding 2670 volume restraints. Theremaining 194 water and hydrogen-bond distancerestraints were left as distance restraints.

Torsion angle restraints

Torsion angle restraints for the sugar puckers andd backbone angles were calculated from H1' toH2',2" crosspeaks in a 2H2O P.COSY spectrum (datanot shown). The residues that had sufficientlyresolved crosspeaks to obtain reliable couplingconstants using CHEOPS are indicated in Figure 5.The peak fits had correlation coefficients thatranged from 90 to 96%.

To restrain the e backbone angle, a 31P-1H 2H2OheteroCOSY was collected (data not shown) fromwhich the J3

P,H3' couplings were determined to be nogreater than 14 Hz, indicating that the e dihedralangles were not in the +gauche conformation(Blommers et al., 1991; Roongta et al., 1990). Allresidues were restrained to exclude the +gaucheconformation for the e angle.

We are aware that there has been some recent

Solution Structure of a D3-containing DNA Triplex 1057

Figure 4. The H2',2",Me-aromatic,H1' region of the same 2D NOESYspectrum as in Figure 3. Inter-residue crosspeaks involving the D3

aromatic protons are: a, T29Me-H2A; b, T4Me-H5A; c, T4Me-H2A;d, T4Me-H6B (partially overlapp-ing T4Me-A3H8); e, T4Me-H2B;f, T4H2',2"-H4B; g, T4H2',2"-H5B;h, T4H2',2"-H6B; i, T16Me-H4C (par-tially overlapping T29Me-T29H1');j, T16Me-H3/5C; k, T16Me-H2/6C;l, T16H2',2"-H4C; and m, T16H2',2"-H3/5C. Crosspeak n is the upfieldshifted T4Me-T4H6 NOE andcrosspeak o, is a typical TH6-TMeNOE (T18Me-T18H6).

debate concerning the validity of fitting couplingconstants to a two-state sugar pucker model(Harbison, 1993; Zhu et al., 1994). However, most ofthe measurable sugar puckers were almost exclu-sively S-type, with the exception of T27, which ispredominantly N-type.

DTA triplex structure

Figure 6 shows several views of the refined DTAtriplex structures. The seven lowest energy struc-tures are shown superimposed in Figure 6(a). Theaverage RMS deviations from ideal geometry for thebonds, angles and impropers are 0.008 A, 1.592° and0.276°, respectively, indicating that these structureshave very good covalent geometry. The loops havefew restraints (Figure 5) and adopt many differentconformations (not shown). The terminal triplets arestrongly perturbed by the conformation of theloops, which artificially increases the variability ofthese triplets. In addition, the two triplets at the 5'end of the triplex (with respect to the purine strand)are only partially assigned (Table 1) leading to manyfewer restraints and greater conformational free-dom. For these reasons, only the atoms in this‘‘core’’ of the triplex (triplets 3 to 7) were used tocalculate the superposition of structures. Theseeffects are also apparent in the RMSD of theindividual residues (Figure 7). The loops and theterminal triplets have a larger RMSD compared to

the core of the triplex. The average pairwise RMSDand the average R1/6-factor (Brunger, 1992) for thecore of the DTA triplex are 1.3 (20.19) A and3.9 (20.1)%, respectively. The number of NOEdistance violations greater than 0.2 A varied fromone to three per structure and the number ofdihedral angle violations greater than 5° variedfrom four to seven per structure.

The backbone angles, glycosidic angles, sugarpuckers and angles specific to the D3 base are shownin Figure 8 for all seven lowest energy structures.Most torsion angles fall into the standard range ofconformations (Saenger, 1984). However, severalangles consistently adopt unusual conformations inmany of the refined structures: the glycosidic angle,x, in residues A5, A17 and T27, prefers the high-anticonformation; C19, C+26 and T27 all have N-typesugar puckers; and the T27 b angle prefers the+gauche conformation.

The helix parameters are shown in Figure 9 as afunction of residue or base step. The X-displacementis very uniform with good precision, although theprecision decreases towards the 5' end (with respectto the purine strand) of the triplex due to sub-stantially fewer restraints on the two 5'-terminaltriplets. The axial rise is also fairly uniform exceptfor the T4-A5 base step, where the rise more thandoubles as a result of the D3 intercalation. Thehelical twist exhibits the greatest stepwise variationof all the helix parameters and also has the lowest

Solution Structure of a D3-containing DNA Triplex1058

Figure 5. Number of NOE dis-tance restraints per residue. (a) Thedistance restraints are classified bytype as: quantitative, non-exchange-able NOEs from 2H2O NOESYspectra; qualitative, exchangeableNOEs from the H2O NOESY spec-trum; hydrogen bonds; and repul-sive. (b) The distance restraints areclassified by their range: intra-residue, sequential and inter-residue(other than sequential). The lightgray regions correspond to theloop residues of the DTA triplex.Residues with restraints on the dtorsion angle are marked with a *.

precision, indicating that it is the least well-definedhelical parameter. Again, the 5' end of the triplexdisplays the most variation.

The conformation of the D3 base is best illustratedby the superposition of the D3 base from the sevenlowest energy structures (Figure 6(c)). The C ringand peptide group are clearly not as restricted intheir conformation as the rest of the D3 base. Themajor conformation is depicted schematically inFigure 1(a).

The stacking of the D328 base on the neighboringT16·A5·T29 triplet and T4·A17 base-pair is illus-trated in Figure 10. The sequence specificity of D3

intercalation is also addressed, where the T4·A17base-pair has been isosterically replaced with thethree other possible Watson–Crick base-pairs, andthe T16·A5·T29 triplet has been isosterically re-placed with a C+·G·C triplet. The D3 base stacks withgreater overlap and the peptide group is morecentered over the polar groups between the bases,when the target base-pair is a T·A or C·G base-pairinstead of a A·T or G·C base-pair. The D3 base

stacking is illustrated from a different view inFigure 6(b), highlighting its similarity to a triplet.

Discussion

Intramolecular triplex formation

The data presented here demonstrate that theDTA sequence forms a stable intramolecular triplex.The NOESY spectrum of the exchangeable reson-ances (Figure 2) strongly resembles those of othertriplexes (Sklenar & Feigon, 1990). Fifteen sharp,well-defined resonances are observed in the spectralregion where hydrogen-bonded imino protons arefound. These have been assigned to the iminoprotons of the eight Watson–Crick and the sevenHoogsteen base-pairs. Further evidence for triplexformation is indicated by the presence of C+ aminoresonances, which are characteristically shifteddownfield. With the exceptions discussed above,sequential imino-imino NOE connectivities may betraced, as is seen in Figure 2(a).

Figure 6. Solution structures of the DTA triplex. The Watson–Crick purine strand is yellow, the Watson–Crickpyrimidine strand is red and the third strand is blue. (a) Stereo view of the seven lowest energy structures lookinginto the groove formed by the Watson–Crick pyrimidine strand and the third strand. The D3 base is green. Loop residuesare not shown. (b) Similar view as in (a) but of a single structure to illustrate how the D3 base mimics a triplet. RingsA, B and C of the D3 base are blue, yellow and red, respectively, and the base of the D3 base is shown as filled sticks.(c) View of the D3 base of the seven lowest energy structures looking into the major groove of the duplex portion ofthe triplex. The rings are colored as in (b). The superpositions involved all heavy atoms shown in each Figure exceptfor the terminal triplets, the C+26·G2·C19 triplet and the D328 peptide group and C-ring.

(a)

(b)

(c)

Solution Structure of a D3-containing DNA Triplex1060

Figure 7. Root-mean-squared de-viation (RMSD) of the seven lowestenergy structures on a per residuebasis. The RMSD is shown for allatoms, only the backbone atoms,and only the base atoms. The shadedregions mark the loop residues ofthe triplex. The best fit of thestructures was determined using allatoms except the loops, the terminaltriplets, the C+26·G2·C19 triplet andthe D328 peptide group and C-ring.

Helix morphology

The DTA triplex forms a regular helix that, inmany ways, is more similar to B-DNA than A-DNA.This can be seen in the axial rise, which shows littlevariation except at the T4·A17-A5·T16 base step,where the rise more than doubles to accommodateD3 intercalation (Figure 9(a)). If the intercalationstep is ignored, the average rise is 3.2 A, which issignificantly closer to the rise in B-DNA (3.4 A) thanin A-DNA (2.6 A).

In contrast to the axial rise, the helical twist ofthe DTA triplex is less regular (Figure 9(b)) andresembles A-DNA more than B-DNA. The helicaltwist displays greater variation because only the eand d backbone angles are restrained. In addition,the initial base step (with respect to the purinestrand) is especially variable for several reasons:

(1) it is the terminal triplet that tends to be lessstable; (2) the terminal triplet has few resonanceassignments (Table 1) and, thus, few restraints;and (3) even though T20 and T25 are connectedby a loop, the loop is unassignable and thereforeunrestrained during refinement, so it adopts a widerange of conformations, which strongly perturbs theconformation of the initial triplet. A similar effect isseen for the second base step, but to a lesser extent.The intercalation site also displays substantialvariability, which is related to the multiple con-formations of the C ring of the D3 base (dis-cussed below). If we ignore the A1·T20-G2·C19terminal step and the T4·A17-A5·T16 intercalationstep, then the average twist is 31°, which, unlike theaxial rise, is more similar to the twist in A-DNA(33°) than in B-DNA (36°).

The X-displacement of the Watson–Crick base-

Figure 8. Torsion angles and sugar puckers of all seven lowest energy structures, listed by residue. (P) is thepseudo-rotation angle of the sugars with the N-type and S-type pucker ranges indicated. D328 has three additionalangles, u1, u2 and u3, as defined in the legend to Figure 1(a).

Solution Structure of a D3-containing DNA Triplex 1061

Figure 9. Helical parameters as a function of base stepor residue, averaged over the seven lowest energystructures. Error bars indicate the standard deviation. Theaverage value for the DTA triplex along with the valuesfor A- and B-DNA are indicated. (a) Axial rise. Theaverage value was calculated omitting the T4·A17-A5·T16base step. (b) Helical twist. The average value wascalculated omitting the A1·T20-G2·C19 and T4·A17-A5·T16 base steps. (c) X-displacement.

The sugar puckers of the DTA triplex are moresimilar to those in B-DNA than in A-DNA. With theexception of three residues in the third strand, thesugars in the core of the DTA triplex prefer anS-type pucker (Figure 8).

Comparison with other triplex structures

Taken together, the helix parameters indicate thatthe DTA triplex has a morphology more like B-DNAthan A-DNA. This is in contrast to early fiberdiffraction studies on poly(dT)·poly(dA)·poly(dT),which concluded that the helical morphology wassimilar to that of A-form DNA (Arnott & Selsing,1974). More recently, fiber diffraction studies ontwo different sequences have suggested that thediffraction patterns are more characteristic ofB-form DNA than A-form DNA (Liu et al., 1994). Inaddition, NMR-derived structures of intramolecularpyrimidine·purine·pyrimidine triplexes containinga non-canonical G·T·A triplet at a single site (GTAtriplex) (Radhakrishnan & Patel, 1994b; Wang et al.,1992) and a non-canonical T·C·G triplet at a singlesite (TCG triplex) (Radhakrishnan & Patel, 1994a)have been solved. A comparison of the helixparameters is given in Table 2. In general, thetriplexes have similar helix parameters. The greatestdifference is found in the X-displacement; theArnott fiber diffraction structure has a largeX-displacement, while the DTA, GTA and TCGtriplexes all have significantly smaller X-displace-ments. As a result, the Arnott fiber diffractionstructure has a significantly deeper major groovethan the solution structures. Similarly, the DTAtriplex has the shallowest major groove because ithas the smallest X-displacement of all the triplexstructures. An earlier NOE-derived model structureof a triplex, similar in sequence to the DTA triplexbut containing only canonical T·A·T and C+·G·Ctriplets, also has a similar rise (3.1 A), twist (32°)and X-displacement (−2.2 A) (Macaya et al., 1992b).

It is worth noting that the NMR-derived triplexstructures were refined by different methodologies.The GTA and TCG structures were both refinedfrom A- and B-DNA starting models. In contrast,the DTA structure presented here was refined usingdistance geometry, which samples a larger area ofconformational space (and results in a larger RMSDbetween structures). Although these structures allhave similar helical parameters, they do havesignificant differences. It is unclear to what extentthe refinement method contributes to these differ-ences.

The sugars in triplexes were originally thoughtto be C3'-endo because Arnott’s fiber diffractionstructure was very similar to A-DNA (Arnott& Selsing, 1974). However, NMR studies ofintramolecular triplexes have shown that thesugars are predominantly S-type (Macaya et al.,1991, 1992a,b; Radhakrishnan & Patel, 1992). TheDTA triplex and the GTA and TCG triplexes(Radhakrishnan & Patel, 1994a,b) also have pre-dominantly S-type sugar puckers. C2'-endo sugar

pairs from the global axis has an average value of−1.4 A for the DTA triplex (Figure 9(c)), and isintermediate between A-DNA (−5.4) and B-DNA(−0.7) but is more similar to B-DNA. The negativeX-displacement means that the DTA triplex has adeep major groove, induced by the binding of thethird strand to the Watson–Crick purine strand, butnot as deep as in A-DNA. The Y-displacement (notshown) is very small, which is in agreement withboth A- and B-DNA.

Solution Structure of a D3-containing DNA Triplex1062

Figure 10. Stacking of the D3 baseon the target T4·A17 base-pair orthe neighboring T29·A5·T26 triplet,and comparison to the stacking onalternate base-pairs or triplets. TheD3 base is blue and the other basesare orange. Atoms that could poten-tially act as hydrogen bond donorsand acceptors are purple (nitrogen),red (oxygen) and white (hydrogens).(a) Stacking of the D3 base on thenative T·A base-pair. (b) Same as (a)except the native T·A base-pair hasbeen isosterically replaced by a C·Gbase-pair. (c) The native T·A base-pair has been replaced by an A·Tbase-pair. (d) The native T·A base-pair has been replaced by a G·Cbase-pair. (e) Stacking of the D3 baseon the neighboring T·A·T triplet. (f)The neighboring T·A·T triplet hasbeen isosterically replaced with aC+·G·C triplet.

puckers are also claimed in the more recent fiberdiffraction results (Liu et al., 1994).

In summary, the general morphology of thetriplexes falls into a distinct family with thefollowing characteristics: (1) an axial rise verysimilar to that in B-DNA; (2) a low helical twist thatis similar to that in A-DNA; (3) an X-displacementthat is intermediate between that in A and B-DNA,but generally closer to B-DNA; and (4) apredominance of S-type sugar puckers as inB-DNA. These characteristics suggest that triplexesresemble B-DNA more than A-DNA.

D3 intercalation

As previously reported, D3 intercalates on the 3'side (relative to the purine strand) of its targetbase-pair rather than forming sequence-specifichydrogen-bond interactions (Koshlap et al., 1993).Disruptions in the sequential connectivities in the

imino region (Figure 2(a)) and the aromatic to H1'region (Figure 3) of NOESY spectra indicatedintercalation (Koshlap et al., 1993). The intercalationexplains the strong sequence dependence on the 3'neighboring triplet and the relative insensitivity tothe 5' neighboring triplet observed by Kiesslinget al., (1992).

The D3 base is readily accommodated into thetriplex, although the intercalation perturbs the helixin several ways. In comparison to our previousstudies on a similar triplex that does not contain theD3 base (Macaya et al., 1992b; Sklenar & Feigon,1990), the resonances for the T25·A1·T20 and theC+26·G2·C19 triplets and the TATA loop residuesare much broader, even to the point of beingunobservable, indicating that intercalation of the D3

base destabilizes the 5' end of the triplex. Anothereffect of D3 intercalation is an increase in the axialrise with a concomitant unwinding of the helix atthe intercalation site (Figure 9(b)). In fact, the low

Table 2. Helix parametersStructurea Rise (A) Twist (°) X-disp (A) Sugar pucker

DTA 3.2b 31c −1.4 Generally S-typeGTA 3.4 31 −1.9 Generally S-typeTCG 3.4 32 −2.1 Generally S-typeLiu (TAT) 3.3 28 N/Ad S-typee

Liu (hetero) 3.2 30 N/Ad S-typee

Arnott 3.3 30 −3.6 Assumed N-typeB-DNA 3.4 36 −0.7 S-typeA-DNA 2.6 33 −5.4 N-type

a DTA refers to the DTA triplex structure. GTA and TCG refer to the two NMR-derived triplexstructures containing a non-canonical triplet (Radhakrishnan & Patel, 1994a,b). Liu (TAT) andLiu (hetero) refer to the two DNA sequences used in fiber diffraction studies (Liu et al., 1994).Arnott refers to the original fiber diffraction structure (Arnott & Selsing, 1974). B-DNA andA-DNA refer to the standard DNA duplex conformations (Saenger, 1984).

b The axial rise for the DTA triplex is calculated omitting the T4 A17-G2 T16 base step.c The helical twist for the DTA triplex is calculated omitting the A1 T20-G2 C19 and T4 A17-A5

T16 base steps.d N/A, not available.e The sugar conformation is claimed by the authors, but no evidence is given (Liu et al., 1994).

Solution Structure of a D3-containing DNA Triplex 1063

twist observed at the A3-T4 step is even smaller ifone considers that the D3 base acts like another base,so that the A3-T4 base step is really a dualA3-D328-T4 base step. Similar effects are seen inintercalating drug–DNA complexes (Addess et al.,1992; Gao & Patel, 1988; Gilbert et al., 1989).

Other effects of D3 intercalation are manifestedin unusual torsion angles and sugar puckers (Fig-ure 8). Backbone restraints were only applied tosome of the d angles and loosely applied to the eangles. As a result, most backbone angles are onlyindirectly determined and the precision of thebackbone angles varies from well defined to almostrandom (Figure 8). Nevertheless, most angles arepredominantly within normal ranges (Saenger,1984). In several of the structures, the T27 b angleis in the +gauche range instead of the usual transconformation. The +gauche conformation places T27in a more lateral position (with respect to the globalhelical axis) from C+26. Also, the T27 x angle is veryclose to the high-anti conformation, which effec-tively rotates the T27 sugar and phosphate closer toC+26. The sugar pucker of T27 is also unusual inthat it adopts an N-type conformation while mostother residues are S-type. This N-type sugar puckercreates a large T27H6 to T27H3' crosspeak inthe NOESY spectrum (Figure 3, peak v) and a‘‘reversed’’ coupling pattern in the T27H1' toT27H2',2" crosspeaks in the P.COSY spectrum (notshown). These structural changes are a consequenceof the D328 base skipping over the T4·A17 base-pairto intercalate, thereby pulling T27 out of plane withA3 and T18. The N-type sugar conformation of T27also results in a decrease in the twist of D328,positioning the D328 base so that it more closelymimics a triplet.

The x angles of A5 and A17, which are both 3' tothe intercalation site relative to their respectivestrands, adopt a high-anti conformation. Thisenables the backbone to stretch further than usual,allowing a greater axial rise at the intercalation site.The base in the equivalent position in the thirdstrand is D328, which does not have a high-anticonformation, but instead pulls on and induces ahigh-anti conformation in T27.

In the refined structures, the C19 and C+26 sugarshave predominantly N-type puckers. However, nodihedral angle restraints were applied to thesesugars and there is no direct NMR evidence tosupport this result. The P.COSY H1'-H2',2" couplingpatterns could not be used to confirm these resultsbecause they are poorly resolved (data not shown).In the aromatic H3' region of the NOESY spectrum(Figure 3), both the C19 and C+26 intra-residuecrosspeaks are weak, indicating that these sugarsare not N-type. However, both residues generallyhave weaker crosspeaks than other residues andC+26 has especially broad crosspeaks and no iminoNOE crosspeaks (Figure 2(a)), indicating that theC+26·G2·C19 triplet is in intermediate exchangewith an alternate (probably non-hydrogen bonded)state. A protonated cytosine in the same positionhas been previously found to prefer N-type sugar

puckers (Macaya et al., 1992a). It is unclear whetherthe N-type sugar puckers are a result of the D3 baseintercalation or a sequence effect.

There are other unusual torsion angles in the loopresidues and the terminal triplets, but they areprobably not due to D3 intercalation. The loopresidues do not necessarily conform to standard A-or B-DNA-like angles, and the terminal triplets areonly partially assigned (Table 1), have many fewerrestraints (Figure 5), and are strongly perturbed bythe conformation of the loops.

D3 base conformation

The conformation of the D3 base is important indetermining the interactions between the D3 baseand its neighboring bases. Figure 6(c) shows thesuperposition of the D3 base from the seven lowestenergy structures. The orientation of rings A and Bare well defined by their numerous NOE inter-actions with the T4, A5, T16, A17 and T29 sugarsand with each other (Figures 3 and 4). The x angleadopts an anti conformation (Figure 8; wherethe N1A, C2A and H2A atoms are analogous to theN9, C8 and H8 atoms in a purine) that fixes theorientation of the A ring parallel to its neighboringbases and points the B ring into the major groove.In this configuration, the position of rings A and Bare similar to those of a third strand pyrimidine andits Hoogsteen base-paired residue in the Watson–Crick purine strand, respectively.

The u1 torsion angle (Figure 1(a)), which connectsrings A and B, determines the rotational position ofthe B ring and, thus, the position of the peptidegroup and the C ring. In all seven structures, the u1

torsion angle positions the B ring to be co-planarwith the A ring and points the peptide group andC ring towards the Watson–Crick pyrimidine strand(Figures 6(c) and 8). Neither the rings nor thepeptide group were restrained to be co-planar. Thisconfiguration places rings B and C in a similarposition to a Watson–Crick base-pair.

The u2 torsion angle adopts several orientations(Figures 6(c) and 8) but predominantly favors theorientation with the carbonyl group pointing out ofthe major groove. The u2 torsion angle affects theposition of the C ring, but only by 02 A at most.In theory, the conformation of the peptide groupcould be determined by NOE interactions involvingthe amide proton, but the amide resonance was notobserved due to rapid chemical exchange withwater. This also suggests that the amide proton isnot involved in a hydrogen bond. Thus, theorientation of the peptide group is only indirectlydetermined by restraints to rings B and C.

The u3 torsion angle is the most variable and leastwell defined in the D3 base (Figure 8). The chemicalshift degeneracy of the C ring protons indicates thatthe chemical shifts are motionally averaged and thatthe C ring is rotating fast on the NMR time scale.Interestingly, significant conformational variation ofthe C ring is also observed in the refined structures(Figures 6(c) and 8). However, since the NOE

Solution Structure of a D3-containing DNA Triplex1064

restraints used during the refinement are alsodegenerate, it is unclear whether the variability ofthe C ring seen in the calculated structures reflectsactual rotation of the C ring or if it simply reflectsthe degeneracy of the restraints. The large standarddeviation in the axial rise at the T4-A5 base step(Figure 9(a)) is a direct consequence of the con-formational variability of the C ring. Even if therise is halved (to account for the D3 base), the riseis still large (04.3 A) due to those structures inwhich the C ring is in a perpendicular (unstacked)conformation.

Stacking of the D 3 base

As previously reported, the stacking of the D3

base causes upfield ring current shifts of 01 ppmfor the T4H6 and T4Me relative to other TH6-TMecrosspeaks (Figure 4 and Table 1) (Koshlap et al.,1993). This effect can be rationalized by examiningFigure 10(a), where the stacking of the D3 base inrelationship to the neighboring T4·A17 base-pairand T29·A5·T16 triplet is shown. The B ring (andpartially the A ring) stacks directly over the T4Meand T4H6, which induces an upfield shift.

The most prominent feature of the stacking is therelative disorder of the D3 base position. Comparedto the neighboring base-pair and triplet, the positionof the D3 base is less well defined. In fact, the D3

base has the largest RMSD of any residue in the coreof the triplex and is also the only residue (other thanthe loops) that has a greater RMSD of its base thanits backbone (Figure 7). The major contribution tothe RMSD comes from the peptide group and theC ring, as already discussed. The D3 base generallyhas some overlap with the T29·A5·T16 triplet andsubstantial overlap with the T4·A17 base-pair(Figure 10(a) and (e)). This may explain why thestability of D3-containing triplexes is stronglyaffected by the sequence of the adjacent base-pair,but are less affected by the sequence of the adjacenttriplet (Kiessling et al., 1992).

An interesting result is that the D3 base mimics atriplet with each ring assuming the position of abase in each strand (Figure 6(b)). From the 3' end ofthe triplex (with respect to the purine strand), thefirst four triplets form a regular stacking patternwith the D3 base mimicking a fifth triplet. Thenthere is large unwinding of the helix between theD3 base and the adjacent T4·A17 base-pair (Fig-ure 9(b)), so the mimicry is not perfect. After theT4·A17 base-pair, the helix continues with a regulartwist again (excluding the A1-G2 step, which is illdefined). As discussed above, the unwinding at theT4-A5 step is more severe than is indicated inFigure 9(b), since the step is really two steps ofT4-D328 and D328-A5. In fact, rings B and C (whichmimic a Watson–Crick base-pair) have almost 0°twist relative to the T4·A17 base-pair. The un-winding is expected because the helical rise mustincrease to accommodate the D3 intercalation.

The D3 base superimposes on a triplet best whenthe angles u1 and u2 are both in the conformation

shown in Figure 1(a), which is also the majorconformation of the seven lowest energy structures.In such a superposition, the glycosidic bond of theD3 base is rotated away from the purine strandrelative to a canonical triplet. This is precisely theeffect that the N-type sugar pucker of T27 has onD328, as discussed above. These observationssuggest that it may be possible to design a base thatconfers greater stability to the triplex by designingit to mimic a triplet more closely.

Sequence specificity

The sequence specificity of D3 intercalationappears to be derived primarily from stackinginteractions between the D3 base and its neighbor-ing bases. Although stacking interactions aredifficult to determine, the D3 base does have slightlymore overlap with T·A and C·G base-pairs (goodtargets for the D3 base) than it does with A·T andG·C base-pairs (poor targets for the D3 base; Fig-ure 10). Sequence-specific hydrogen bond formationcannot be unambiguously ruled out, although thethree chemical groups of the D3 base that canpotentially form hydrogen bonds (the amide proton,the carbonyl oxygen and the N3A imidazolenitrogen) do not show evidence of hydrogenbonding. The amide proton is exchanging rapidly,which is inconsistent with hydrogen bond for-mation. The imidazole N3A nitrogen faces outwardinto the major groove with no nearby hydrogenbond acceptors (Figure 10). In addition, the onlyavailable acceptors for the peptide amide proton arethe A5 and A17N1, the A5N7 and the T4O2, all ofwhich are out of plane with the peptide group andcould form only weak hydrogen bonds. Similarly,the only hydrogen bond donors near the peptidecarbonyl group are the A5 and A17 amino groupsand the T4 and T16 imino protons, which would allyield weak hydrogen bonds.

Electrostatic interactions may also be important,although the exact nature of the interaction is notclear. For example, in the major groove, the peptidepolar groups are near several polar groups of theT·A base-pair (Figure 10(a)). However, in the majorgroove of a G·C base-pair (a poor target for the D3

base) the same polar groups are present in a similararrangement (Figure 10(d)). Similarly, the majorgroove of a C·G and on A·T base-pair resemble oneanother, although a C·G base-pair is a good targetfor the D3 base and an A·T base-pair is a poor target.In an effort to elucidate further the determinants ofsequence specificity, we are currently solving thesolution structure of the D3 base targeted to a C·Gbase-pair.

D3 intercalation strongly prefers a neighboringT·A·T triplet over a C+·G·C triplet. Comparison ofthe D3 base stacked over a T·A·T and a C+·G·C tripletdoes not present any obvious reason for thispreference (Figure 10(e) and (f)). However, thepreference for a neighboring T·A·T triplet may bemore general than just for the D3 base, since thecanonical triplets also prefer neighboring T·A·T

Solution Structure of a D3-containing DNA Triplex 1065

triplets over C+·G·C triplets (Kiessling et al., 1992).This preference may be an effect of the positivecharge in a C+·G·C triplet.

Intercalation occurs on the 3' side of the targetbase-pair instead of the 5' side. This is due to theright-handedness of the triplex. If one observes thehelix in Figure 6(b), the third strand proceeds 5' to 3'from top to bottom of the figure. In this orientation,D328 stacks below and to the left of T27, betweenthe T27 and the A3 sugars. D328 also stacks aboveand to the right of the T29 sugar. If D328 were tointercalate on the 5' side of (above) T4, the base stepbetween T27 and D328 would be compressed andoverwound, which would force D328 towards thepurine strand. Since D328 stacks between the A3and T27 sugars, the compression in the rise and theoverwinding of the twist would force D328 into theA3 sugar causing interference with the intercalation.However, if D328 intercalates on the 3' side of(below) T4 (as is observed), the base step betweenD328 and T29 would be compressed and over-wound, which would force D328 away from thepurine strand. Since D328 does not stack betweentwo sugars on the 3' side and because theoverwinding rotates D328 further from the purinestrand, there is no obstruction from other residues.

Conclusions

In summary, a pyrimidine·purine·pyrimidinetriplex containing a D3 base forms a stable structureresembling other triplex structures, despite havingto accommodate D3 intercalation. The stability ispartially due to the ability of the D3 base to imitatea triplet, although imperfectly. This suggests thatgreater stability may be achieved by designing abase to more closely mimic a triplet. Pyrim-idine·purine·pyrimidine triplexes seem to haveseveral features in common, including an axial riseof 03.2 to 3.4 A, a low helical twist of 030 to32°, an X-displacement of 0−1 to −3 A, and apredominance of S-type sugar puckers. Sincetriplexes conform to neither A- nor B-DNAexclusively, they might best be considered a newform of DNA, T-form DNA, a separate, distinctfamily of structures.

Materials and Methods

Sample preparation

The phosphoramidite derivative of the artificialnucleoside, D3, was prepared as described (Griffin et al.,1992). The DNA oligonucleotides were synthesized on anABI 380B DNA synthesizer. The oligonucleotides weredeprotected with concentrated ammonia at 55°C for24 hours, and subsequently desalted on a Sephadex G10column. FPLC purification (DMT-on) was performedusing a gradient of 0 to 40% acetonitrile/100 mM tri-ethylammonium acetate over 40 minutes. Detritylationwas accomplished with 80% acetic acid for 30 minutes.Final FPLC purification (DMT-off) used a gradient of 0 to27% acetonitrile/100 mM triethylammonium acetate over30 minutes. To remove the triethylammonium acetate, the

sample was run over a BioRad AG 50W-X8 ion-exchangecolumn, followed by chromatography on a Sephadex G15column. The appropriate fractions were then pooled anddried by lyophilization. The final conditions of the NMRsamples were 100 mM NaCl, 5 mM MgCl2, and 02 mMDNA in 400 ml of either 99.996% 2H2O or 90% H2O/10%2H2O; the pH was adjusted to 5.2 with NaOH/HCl.

NMR spectroscopy

All NMR experiments were performed on a GeneralElectric GN500 spectrometer. Phase-sensitive nuclearOverhauser effect (NOESY) spectra (Kumar et al., 1980;Macura & Ernst, 1980) were obtained employing themethod of States et al. (1982). For spectra in 2H2O, theresidual 2H2O peak was suppressed by irradiation duringthe recycle delay. Water suppression for samples in H2Owas achieved by using a 11�-spin echo read pulse(11ECHO; Sklenar & Bax, 1987). The carrier was centeredat the water resonance, and the imino and aminoresonances were maximized by setting t = 50 ms. Furtherelimination of the residual water was accomplishedduring processing using a time-domain convolutionroutine (Marion et al., 1989).

For NOE restraints involving non-exchangeable reson-ances, a set of 2H2O NOESY spectra was collected at 25°Cwith mixing times of 40, 80, 120, 160 and 200 ms withoutremoving the sample from the magnet. All NOESYspectra in the series were collected and processedidentically except for the mixing time. The acquisitionparameters were a sweep width of 5000 Hz in both t2 andt1, 1024 points in t2, 350 points in t1, 32 scans per t1 block,and a recycle delay of two seconds. The t2 apodizationwas a 450 point, 45° phase shifted, squared sinebell witha first-order polynomial baseline flattening and the t1

apodization was a 350 point, 45° phase shifted, squaredsinebell with a second-order polynomial baseline flatten-ing. The resulting datasets were zero filled to 4096 by 4096complex points for ample digital resolution duringintegration of the crosspeak volumes.

For NOE restraints involving exchangeable resonances,an H2O NOESY spectrum was collected at 1°C with amixing time of 100 ms. The excitation maximum for the11ECHO water suppression was set to 13.9 ppm. Allprocessing, peak picking and integration were performedusing FELIX 2.3 (Biosym Technologies/MSI) running onSilicon Graphics workstations. Further details of theacquisition and processing parameters are given in theFigure captions.

For proton coupling data, a P.COSY spectrum (Marion& Bax, 1988; Mueller, 1987) was collected at 25°C with a90° mixing pulse. The acquisition parameters were asweep width of 5000 Hz in both t2 and t1, 1024 points int2, 550 points in t1, 96 scans per t1 block, and a recycledelay of 1.8 seconds. Using the technique of Marion &Bax (1988), a one-dimensional reference spectrum wascollected to subtract the dispersive component of thediagonal. The same parameters as the P.COSY were usedexcept that 1536 scans were collected and the acquisitiontime was doubled. To insure that the total relaxation delayand H2O pre-saturation between scans was the same inboth the reference and P.COSY spectrum, an additionaldelay equal to the acquisition time was added to the endof the P.COSY pulse sequence. The t2 apodization was a700 point, 35° phase shifted, squared sinebell; and the t1

apodization was a 550 point, 45° phase shifted, squaredsinebell. Selective regions of the spectrum were processedand zero filled to a resolution of 0.82 point/Hz forcrosspeak pattern fitting.

Solution Structure of a D3-containing DNA Triplex1066

NOE restraints

Crosspeak volumes from the 2H2O NOESY mixing timeseries were integrated for each mixing time. Thesevolumes were either used for relaxation matrix refine-ment directly or converted to distances for restraineddistance refinement. Short mixing time data can be usedto estimate the initial build-up rates of these crosspeakvolumes, which are inversely correlated to their distances.Only the 40 ms mixing time data were used to calculatethe distances because the data from longer mixing timesindicated significant spin diffusion. The 40 ms volumeswere converted into distances using a set of known, fixeddistances for calibration and the relationship Volij A 1/r6

ij ,where Volij is the volume of the crosspeak betweenprotons i and j, and rij is the distance between protons iand j. The crosspeaks used for calibration were C9 H5-H6,C19 H5-H6, C30 H5-H6 and D328 H5B-H6B, chosenbecause they were isolated and could be cleanlyintegrated.

The exchangeable crosspeaks were qualitatively catego-rized as strong, medium or weak corresponding to 3.0, 4.0and 5.0 A, respectively. The crosspeak intensities forexchangeable resonances are not accurately quantifiablebecause they exchange with water at differing rates andbecause correction for the 11ECHO sin3 excitation profileis only accurate near the maximum.

Crosspeaks that overlapped each other and that couldnot be integrated separately were, nevertheless, utilizedby integrating the overlapping crosspeaks as a singlecrosspeak and converting the total volume into a singledistance restraint. The resulting distance restraint wasentered into XPLOR by using a ‘‘sum’’ averaging, whichallows multiple assignments to be defined and whichcorrectly handles degenerate crosspeaks. Several reson-ances in the C ring of the D3 base are degenerate and werealso averaged using the sum averaging method. Anycrosspeaks involving methyl groups had a 1/r3 averagingin XPLOR (James et al., 1991). All other distance restraintswere calculated using ‘‘center’’ averaging. In all casesexcept overlapping crosspeaks, the lower bound was thevan der Waals’ (VDW) radii and the upper bound was+10% of the distance. For overlapping crosspeaks thelower bound could potentially be less than the VDWradii when using the sum averaging and requires that thelower bound be determined by the formula: lowerbound = VDW/(n1/6), where n is the number of over-lapping crosspeaks integrated as a single crosspeak. Theupper bound was +20% of the distance.

Background noise was integrated at various regions ofthe spectrum and the average value was subtracted fromeach crosspeak before converting them to distances. Thestandard deviation of the background noise level alsogave an estimate of the smallest measurable volume and,thus, the longest measurable distance. Crosspeaks thatwere below the noise level in the 40 ms NOESY spectrumbut were visible at longer mixing times were given anupper bound restraint of 6 A.

Repulsive restraints, which were obtained in aniterative approach (see Structure determination, below)were also utilized. When used as distance restraints, thelower bound was 4 A and the upper bound was 99 A.When used as volume restraints, the target volume wasset to the noise level.

Torsion angle restraints

To determine the coupling constants in the deoxyri-boses, the H1'-H2' and H1'-H2" crosspeaks of a P.COSY

spectrum were fit using CHEOPS (Macaya et al., 1992a;Schultze & Feigon, unpublished program). The dihedralangles n1 and n2 were determined from these couplingconstants by fitting them to a two-state model for thesugar pucker with the program PSEUROT (de Leeuw &Altona, 1983) and using the dihedrals from the majorconformation. The dihedrals were restrained with anerror of 25°. These two dihedral restraints fix thebackbone torsion angle d. Twelve residues were resolvedenough to determine coupling constants, leading to 24dihedral angle restraints. The e dihedrals were derivedfrom a 1H-31P heteroCOSY spectrum from which it wasdetermined that no 3JP,H3' couplings were greater than14 Hz, which excludes the +gauche conformation for the edihedral angle. Therefore, the e angle was restrained to−120 (2120)° for all 31 phosphate groups. Twenty threeresidues from the core of the triplex and the CCCCloop had their x angle loosely restrained to the anticonformation (−120(290)°). The x angles for G2, theT25·A1·T20 triplet, and the TATA loop were notrestrained, since the conformation of their glycosidicbond could not be determined from their base to H1'crosspeaks. The x angle for D328 was also not restrainedso as not to bias the conformation of the D3 base. Tomaintain proper chirality, an additional dihedral anglerestraint was applied to each chiral center, C1', C3' andC4', and to each prochiral center, C2', C5' and P.

Structure determination

Structure calculations were performed on DigitalEquipment Corporation Alpha workstations usingXPLOR V3.1 (Brunger, 1992). The structure determinationis composed of an initial refinement, which essentiallyfollows the protocols outlined in the XPLOR V3.1 manual,with only minor changes, followed by an iterative, fullrelaxation matrix refinement. Iterative relaxation matrixrefinement involves back-calculating the NOE volumesfrom the structures produced by the initial XPLORprotocols and refining these back calculated volumesagainst the observed volumes (Yip & Case, 1989). Afterone round of relaxation matrix refinement, thosecrosspeaks that are observed in the back-calculatedspectra but are clearly not observed in the NOESY spectraare entered as repulsive restraints and the structures arere-calculated. Several rounds of this procedure are doneuntil no further repulsive terms are found.

The XPLOR protocols essentially consist of distancegeometry embedding of the structure consistent with theNOE distance and dihedral angle restraints, regulariz-ation of the structure, further refinement using simulatedannealing (SA) with a simplified VDW energy termand a final refinement with a long SA protocol and a fullVDW energy term. In addition, we supplementedthese protocols with a final relaxation matrix refinement.The following changes to the protocols were made: (1)distance geometry sub-embedding (dg sub embed.inp),which used atoms N1A, C1B, N and C1C of the D3

base for sub-embedding (Figure 1(a)); 100 structureswere calculated; (2) regularization (dgsa.inp), which de-creased the time step to 2.5 fs to prevent wildlyoscillating structures during dynamics, but increased thenumber of steps to 2000 to maintain the same totalsimulation time; 100 structures were calculated; (3)refinement (refine.inp), which decreased the time stepto 2.0 fs and increased the number of steps to 5000during dynamics; 100 structures were calculated; (4) fullVDW refinement (refine gentle.inp), which increased thetemperature during dynamics to 500 K. Included hydro-

Solution Structure of a D3-containing DNA Triplex 1067

gen bond energy term and used the planarity restraint tokeep the individual bases and the rings of the D3 base flat;the 40 lowest energy structures from the previous stageof refinement were refined; and (5) relaxation matrixrefinement, in which volumes from all five NOESYmixing times were used. The non-exchangeable andrepulsive NOE distance restraints were replaced with thecorresponding volume integrals. The exchangeable (andhydrogen bond) NOEs were left as distance restraints.The NOE energy scale was increased to 100. Isotropictumbling was assumed with a correlation time of 5 ns.The NOE integrals were scaled to the same fourcrosspeaks as in the distance refinement. The relaxationenergy parameters were set to cutoff = all, value = 6.0,potential = well, iexp = 1/6, eexp = 2, expo = 2, weight= 0.5. This is essentially a Vol1/6 weighting of the integrals(Brunger, 1992; James et al., 1991). All other parametersare the same as in the full VDW refinement. The overallrelaxation matrix refinement protocol is: (1) 100 steps ofenergy minimization with relaxation energy calibrationturned off; (2) 1000 steps of dynamics with a time step of1.0 fs at 300 K with temperature bath coupling and autocalibration turned on for the relaxation energy term;(3) the dynamics trajectory is averaged over the lasthalf of the dynamics run taken every 10 steps to obtainan average structure; and (4) 200 steps of energyminimization done in two 100-step phases with relaxationmatrix calibration done after each phase. The 20 lowestenergy structures from the previous stage of refinementwere refined.

Structure analysis

The program CURVES (Lavery & Sklenar, 1988) wasused to calculate the helical parameters in all thestructures. Only the Watson–Crick duplex portion of thetriplex was used in these calculations. XPLOR V3.1 wasused to measure all the backbone angles and sugarpuckers.

AcknowledgementsThis work was supported by NIH grant R01 GM37254

to J.F., and an Office of Naval Research grant to P.B.D. Thecoordinates for the structures have been deposited withthe Brookhaven Protein Data Base (accession no. 1WAN).

ReferencesAddess, K. J., Gilbert, D. E., Olsen, R. K. & Feigon, J.

(1992). Proton NMR studies of [N-MeCys3, N-MeCys7]TANDEM binding to DNA oligonucleotides:sequence specific binding at the TpA site. Biochem-istry, 31, 339–350.

Arnott, S. & Selsing, E. (1974). Structures for thepolynucleotide complexes poly(dA)·poly(dT) andpoly(dT)·poly(dA)·poly(dT). J. Mol. Biol. 88, 509–521.

Bax, A., Griffey, R. H. & Hawkins, B. L. (1983). Correlationof proton and nitrogen-15 chemical shifts by multiplequantum NMR. J. Magn. Reson. 55(55), 301–315.

Beal, P. A. & Dervan, P. B. (1991). Second structural motiffor recognition of DNA by oligonucleotide-directedtriple-helix formation. Science, 251, 1360–1363.

Belotserkovskii, B. P., Veselkov, A. G., Filippov, S. A.,Dobrynin, V. N., Mirkin, S. M. & Frank-Kamenetskii,M. D. (1990). Formation of intramolecular triplex in

homopurine-homopyrimidine mirror repeats withpoint substitutions. Nucl. Acids Res. 18, 6621–6624.

Blommers, M. J. J., van de Ven, F. J. M., van der Marel,G. A., van Boom, J. H. & Hilbers, C. W. (1991). Thethree-dimensional structure of a DNA hairpin insolution. Two-dimensional NMR studies and struc-tural analysis of d(ATCCTATTTATAGGAT). Eur. J.Biochem. 181, 33–51.

Broitman, S. L., Im, D. D. & Fresco, J. R. (1987). Form-ation of the triple-stranded polynucleotide helix,poly(A·A·U). Proc. Natl Acad. Sci. USA, 84, 5120–5124.

Brunger, A. T. (1992). X-PLOR (Version 3.1) Manual, YaleUniversity Press, New Haven and London.

Cooney, M., Czernuszewicz, G., Postel, E. H., Flint, S. J.& Hogan, M. E. (1988). Site-specific oligonucleotidebinding represses transcription of the human c-mycgene in vitro. Science, 241, 456–459.

de Leeuw, F. A. A. M. & Altona, C. J. (1983).Computer-assisted pseudorotation analysis of five-membered rings by means of proton spin-spincoupling constants: Program PSEUROT. J. Comp.Chem. 4, 428–437.

de los Santos, C., Rosen, M. & Patel, D. (1989). NMRstudies of DNA (R+)n·(Y−)n·(Y+)n triple helices insolution: Imino and amino proton markers of T·A·Tand C·G·C+ base-triple formation. Biochemistry, 28,7282–7289.

Degols, G., Clarenc, J. P., Lebleu, B. & Leonetti, J. P. (1994).Reversible inhibition of gene expression by apsoralen functionalized triple helix forming oligonu-cleotide in intact cells. J. Biol. Chem. 269(24),16933–16937.

Felsenfeld, G., Davies, D. R. & Rich, A. (1957). Formationof a three-stranded polynucleotide molecule. J. Am.Chem. Soc. 79, 2023–2024.

Gao, X. & Patel, D. J. (1988). NMR studies of echinomycinbisintercalation complexes with d(A1-C2-G3-T4) andd(T1-C2-G3-A4) duplexes in aqueous solution:sequence-dependent formation of Hoogsteen A1·T4and Watson–Crick T1·A4 base-pairs flanking thebisintercalation site. Biochemistry, 27, 1744–1751.

Gilbert, D. E., van der Marel, G. A., van Boom, J. H. &Feigon, J. (1989). Unstable Hoogsteen base-pairsadjacent to echinomycin binding sites within a DNAduplex. Proc. Natl Acad. Sci. USA, 86, 3006–3010.

Greenberg, W. A. & Dervan, P. B. (1995). Energetics offormation of sixteen triple helical complexes whichvary at a single position within a purine motif. J. Am.Chem. Soc. 117, 5016–5022.

Griffin, L. C. & Dervan, P. B. (1989). Recognition ofthymine·adenine base-pairs by guanine in a pyrim-idine triple helix motif. Science, 245, 967–971.

Griffin, L. C., Kiessling, L. L., Beal, P. A., Gillespie, P. &Dervan, P. B. (1992). Recognition of all four base-pairsof double-helical DNA by triple-helix formation:design of nonnatural deoxyribonucleosides forpyrimidine·purine base-pair binding. J. Am. Chem.Soc. 114, 7976–7982.

Grigoriev, M., Praseuth, D., Robin, P., Hemar, A.,Saison-Behmoaras, T., Dautry-Varsat, A., Thuong,N. T., Helene, C. & Harel-Bellant, A. (1992). A triplehelix-forming oligonucleotide-intercalator conjugateacts as a transcriptional repressor via inhibition of NFkB binding to interleukin-2 receptor a-regulatorysequence. J. Biol. Chem. 267(5), 3389–3395.

Harbison, G. S. (1993). Interference between J-couplingsand cross-relaxation in solution NMR spec-troscopy—consequences for macromolecular struc-ture determination. J. Am. Chem. Soc. 115, 3026–3027.

Solution Structure of a D3-containing DNA Triplex1068

Havre, P. A., Gunther, E. J., Gasparro, F. P. & Glazer, P. M.(1993). Targeted mutagenesis of DNA using triplehelix-forming oligonucleotides linked to psoralen.Proc. Natl Acad. Sci. USA, 90(16), 7879–7883.

James, T. L., Gochin, M., Kerwood, D. J., Schmitz, U. &Thomas, P. D. (1991). Computational Aspects of theStudy of Biological Macromolecules by NMR (Hoch,J. C., ed.), Plenum Press, New York.

Ji, H. M., Smith, L. M. & Guilfoyle, R. A. (1994). Rapidisolation of cosmid insert DNA by triple-helix-medi-ated affinity capture. Genet. Anal. Techn. Appl. 11,43–47.

Kiessling, L. L., Griffin, L. C. & Dervan, P. B. (1992).Flanking sequence effects within the pyrimidinetriple-helix motif characterized by affinity cleaving.Biochemistry, 31, 2829–2834.

Koshlap, K. M., Gillespie, P., Dervan, P. B. & Feigon, J.(1993). Nonnatural deoxyribonucleoside D3 incorpor-ated in an intramolecular DNA triplex bindssequence-specifically by intercalation. J. Am. Chem.Soc. 115, 7908–7909.

Kumar, A., Ernst, R. R. & Wuthrich, K. (1980). Atwo-dimensional nuclear Overhauser enhancement(2D NOE) experiment for the elucidation of completeproton-proton cross-relaxation networks in biologicalmacromolecules. Biochem. Biophys. Res. Commun. 95,1–6.

Lavery, R. & Sklenar, H. (1988). The definition ofgeneralized helicoidal parameters and of axiscurvature for irregular nucleic acids. J. Biomol. Struct.Dynam. 6(1), 63–91.

Le Doan, T., Perrouault, L., Praseuth, D., Habhoub, N.,Decout, J.-L., Thuong, N. T., Lhomme, J. & Helene, C.(1987). Sequence-specific recognition, photocross-linking and cleavage of the DNA double helix by anoligo-[a]-thymidylate covalently linked to an azido-proflavine derivative. Nucl. Acids Res. 15, 7749–7760.

Lipsett, M. N. (1964). Complex formation betweenpolycytidylic acid and guanine oligonucleotides.J. Biol. Chem. 239, 1256–1260.

Liu, K., Miles, H. T., Parris, K. D. & Sasisekharan, V.(1994). Fibre-type X-Ray diffraction patterns fromsingle crystals of triple helical DNA. Nature Struct.Biol. 1, 11–12.

Luebke, K. J. & Dervan, P. B. (1991). Nonenzymaticsequence-specific ligation of double-helical DNA.J. Am. Chem. Soc. 113, 7447–7448.

Macaya, R. F., Gilbert, D. E., Malek, S., Sinsheimer, J. &Feigon, J. (1991). Structure and stability of X·G·Cmismatches in the third strand of intramoleculartriplexes. Science, 254, 270–274.

Macaya, R. F., Schultze, P. & Feigon, J. (1992a). Sugarconformations in intramolecular DNA triplexesdetermined by coupling constants obtained byautomated simulation of P.COSY crosspeaks. J. Am.Chem. Soc. 114, 781–783.

Macaya, R. F., Wang, E., Schultze, P., Sklenar, V. & Feigon,J. (1992b). Proton NMR assignments and structuralcharacterization of an intramolecular DNA triplex.J. Mol. Biol. 225, 755–773.

Macura, S. & Ernst, R. R. (1980). Elucidation of crossrelaxation in liquids by two-dimensional N.M.R.spectroscopy. Mol. Phys. 41, 95–117.

Maher, L. J. (1992). DNA triple-helix formation—anapproach to artificial gene repressors. BioEssays14(12), 807–815.

Marion, D. & Bax, A. (1988). P.COSY, a sensitivealternative for double-quantum-filtered COSY.J. Magn. Reson. 80, 528–533.

Marion, D., Ikura, M. & Bax, A. (1989). Improved solventsuppression in one- and two-dimensional NMRspectra by convolution of time-domain data. J. Magn.Res. 84, 425–430.

Mergny, J.-L., Sun, J.-S., Rougee, M., Montenay-Garestier,T., Barcelo, F., Chomilier, J. & Helene, C. (1991).Sequence specificity in triple-helix formation: exper-imental and theoretical studies of the effect ofmismatches on triplex stability. Biochemistry, 30,9791–9798.

Moser, H. E. & Dervan, P. B. (1987). Sequence-specificcleavage of double helical DNA by triple helixformation. Science, 238, 645–650.

Mueller, L. (1987). P.E.COSY, a simple alternative toE.COSY. J. Magn. Reson. 72, 191–196.

Olivas, W. M. & Maher, L. J. (1994). Analysis of duplexDNA by triple helix formation—application todetection of a P53 microdeletion. Biotechniques, 16(1),128–132.

Radhakrishnan, I. & Patel, D. J. (1992). Solution con-formation of a G·TA triple in an intramolecularpyrimidine·purine·pyrimidine DNA triplex. J. Am.Chem. Soc. 114(17), 6913–6915.

Radhakrishnan, I. & Patel, D. J. (1994a). Solutionstructure and hydration patterns of a pyrim-idine·purine·pyrimidine DNA triplex containing anovel T·CG base-triple. J. Mol. Biol. 241, 600–619.

Radhakrishnan, I. & Patel, D. J. (1994b). Solution structureof a pyrimidine·purine·pyrimidine DNA triplexcontaining T·AT, C+·GC and G·TA triples. Structure, 2,17–32.

Radhakrishnan, I., de los Santos, C. & Patel, D. J. (1991).Nuclear magnetic resonance structural studies ofintramolecular purine·purine·pyrimidine DNAtriplexes in solution. Base triple pairing alignmentsand strand direction. J. Mol. Biol. 221, 1403–1418.

Rajagopal, P. & Feigon, J. (1989a). NMR studies oftriple-strand formation from the homopurine-homopyrimidine deoxyribonucleotides d(GA)4 andd(TC)4. Biochemistry, 28, 7859–7870.

Rajagopal, P. & Feigon, J. (1989b). Triple strand formationin the homopurine:homopyrimidine DNA oligonu-cleotides d(GA)4 and d(TC)4. Nature, 339, 637–640.

Roberts, R. W. & Crothers, D. M. (1991). Specificity andstringency in DNA triplex formation. Proc. Natl Acad.Sci. USA, 88, 9397–9401.

Roongta, V. A., Jones, C. R. & Gorenstein, D. G. (1990).Effect of distortions in the deoxyribose phosphatebackbone conformation of duplex oligodeoxyribonu-cleotide dodecamers containing GT, GG, GA, AC, andGU base-pair mismatches on 31P NMR spectra.Biochemistry, 29, 5245–5258.

Saenger, W. (1984). Principles of Nucleic Acid Structure,Springer-Verlag, New York.

Sklenar, V. & Bax, A. (1987). Spin-echo water suppressionfor the generation of pure-phase two-dimensionalNMR spectra. J. Magn. Reson. 74, 469–479.

Sklenar, V. & Feigon, J. (1990). Formation of a stabletriplex from a single DNA strand. Nature, 345,836–838.

States, D. J., Haberkorn, R. A. & Ruben, D. J. (1982). Atwo-dimensional nuclear Overhauser experimentwith pure absorption phase in four quadrants.J. Magn. Res. 48, 286–292.

Strobel, S. A. & Dervan, P. B. (1991). Single-site enzymaticcleavage of yeast genomic DNA mediated by triplehelix formation. Nature, 350, 172–174.

Strobel, S. A., Doucette-Stamm, L. A., Riba, L., Housman,D. E. & Dervan, P. B. (1991). Site-specific cleavage of

Solution Structure of a D3-containing DNA Triplex 1069

human chromosome 4 mediated by triple-helixformation. Science, 254, 1639–1642.

Sun, J.-S. & Helene, C. (1993). Oligonucleotide-directedtriple-helix formation. Curr. Opin. Struct. Biol. 3,345–356.

Wang, E., Malek, S. & Feigon, J. (1992). Structure of aG·T·A triplet in an intramolecular DNA triplex.Biochemistry, 31, 4838–4846.

Wells, R. D., Collier, D. A., Hanvey, J. C., Shimizu, M. &Wohlrab, F. (1988). The chemistry and biology ofunusual DNA structures adopted by oligopurineand oligopyrimidine sequences. FASEB J. 2, 2939–2949.

Yip, P. & Case, D. A. (1989). A new method for refinementof macromolecular structures based on nuclearOverhauser effect spectra. J. Magn. Reson. 83,643–648.

Yoon, K., Hobbs, C. A., Koch, J., Sardaro, M., Kutny, R.& Weis, A. L. (1992). Elucidation of the sequence-specific third strand recognition of four Watson–Crick base-pairs in a pyrimidine triple-helix motif:T·AT, C·GC, T·CG and G·TA. Proc. Natl Acad. Sci.USA, 89, 3840–3844.

Zhu, L., Reid, B. R., Kennedy, M. & Drobny, G. P. (1994).Modulation of J couplings by cross relaxation in DNAsugars. J. Magn. Res. ser. A, 111, 195–202.

Edited by P. E. Wright

(Received 21 November 1995; received in revised form 17 January 1996; accepted 22 January 1996)