[10] DNA TRIPLEXES AND QUADRUPLEXES 225

ms. In addition, constraints for hydrogen bonding were also incorporated. 51 The stem structure of the NOE-restrained energy-minimized DNA struc- ture calculated for the hairpin 3' d(TaxC4As)3' on the basis of NMR data (see Figs. 2 and 5) is essentially identical to the structure originally proposed by Pattabiraman. 26 The structure formed has reverse Watson-Crick base pairing and a backbone resembling that of B-DNA. The helix diameter of the NOE-refined model is about 20/~ with 9.4 residues/turn. 51 The most striking feature for this molecule is that the grooves have a similar width that make it distinct from B-DNA in terms of its interactions with drugs and proteins (Fig. 6). The average distance between the closest phosphorus atom across one groove is 16/~, whereas 13/~ is found for the other. 51 In B-DNA, the corresponding measurements are 17.4 and 11.5/~ for major and minor groove, respectively. The deoxyribose ring conformations, glycosidic torsion angles in the refined model, as well as the backbone conformation are fully consistent with the NMR analysis.

These data show that parallel-stranded duplex DNA can occur for natural DNA sequences under physiological conditions. In light of the radically different overall structure of parallel-stranded DNA, it is intrigu- ing that each of the strands still is, in essence, a B-type structure. This may be the major factor determining the relative ease with which parallel DNA in A-T sequences is formed.

Acknowledgments

This research was supported by the Alber ta Heri tage Foundat ion for Medical Research and the Medical Research Council of Canada. W e thank Dr. J. Aramini for his assistance in editing this manuscript .

[10] IH N M R S p e c t r o s c o p y o f D N A T r i p l e x e s a n d Q u a d r u p l e x e s

By JULI F E IGON, K A R L M. K O S H L A P , and FLINT W . SMITH

Introduction

An exciting development in the area of structural studies of DNA in the past few years is the rediscovery that DNA can adopt a variety of conformations in addition to B-DNA. Alternative hydrogen bonding ar- rangements of DNA nucleotides resulting in three- and four-stranded struc-

Copyright © 1995 by Academic Press, Inc. METHODS IN ENZYMOLOGY, VOL. 261 All rights of reproduction in any form reserved.

226 DNA AND RNA STRUCTURE [101

tures were proposed in the late 1950s and early 1960s. 1-3 The "rediscovery" followed from biochemical and molecular biological evidence that purine- rich sequences in a variety of biological control regions in the genome could adopt unusual structures under appropriate conditions. These purine- rich sequences were subsequently found to form structures that fall into two general classes: triplexes 4,5 and quadruplexes. 6 The recent advances in structure determination by multidimensional NMR methods have made it possible to determine these structures at high resolution.

The possible in vivo occurrence of a triplex form of DNA was postulated based on a variety of enzymatic digestion, chemical modification, and two- dimensional gel electrophoresis studies. 4,7 These studies provided evidence that homopurine :homopyrimidine duplexes could form an alternative structure called H-DNA, 8 in which half of the pyrimidines fold back to bind in the major groove via Hoogsteen pairing and the excess purines loop out as single strand. Various lines of evidence have indicated that these structures may have a functional role in vivo . 4 Triplex formation has also been extensively studied because of the potential for sequence-specific targeting of duplex DNA by third-strand binding. 5,9 Applications of this approach range from potential nucleic acid therapeutics 1°'11 to chromo- some mappingJ 2

Triplexes are formed from a Watson-Crick duplex composed of one homopurine and one homopyrimidine strand and a homopyrimidine or purine-rich third strand that binds in the major groove to form base trip- le ts . 9'13 The more well characterized is the pyrimidine, purine • pyrimidine (pyrimidine motif) triplex, in which the second pyrimidine strand binds parallel to the purine strand of the Watson-Crick duplex via Hoogsteen base pairing. The resulting triplexes are composed of T. A. T and C ÷. G. C

1 G. Felsenfeld, D. R. Davies, and A. Rich, J. Am. Chem. Soc. 79, 2023 (1957). 2 W. Guschlbauer, J.-F. Chantot, and D. Thiele, J. Biomol. Struct. Dyn. 8, 491 (1990). 3 R. D. Wells, D. A. Collier, J. C. Hanvey, M. Shimizu, and F. Wohlrab, FASEB J. 2,

2939 (1988). 4 S. M. Mirkin and M. D. Frank-Kamenetskii, Annu. Rev. Biophys. Biomol. Struct. 23, 541

(1994). 5 J.-S. Sun and C. Hrl~ne, Curr. Opin. Struct. Biol. 3, 345 (1993). 6 j. R. Williamson, Annu. Rev. Biophys. Biomol. Struct. 23, 703 (1994). 7 M. D. Frank-Kamenetskii, in "Methods in Enzymology" (D. M. J. Lilley and J. E. Dahlberg,

eds.), Vol. 211, p. 180. Academic Press, San Diego, 1992. 8 S. M. Mirkin, V. I. Lyamichev, K. N. Drushlyak, V. N. Dobrynin, S. A. Filippov, and M. D.

Frank-Kamenetskii, Nature 330, 495 (1987). 9 N. T. Thuong and C. Hrl~ne, Angew. Chem. Int. Ed. Engl. 32, 666 (1993).

10 K. M. Nagel, S. G. Holstad, and K. E. Isenberg, Pharmacotherapy 13, 177 (1993). 11 j. Prins, E. G. de Vries, and N. H. Mulder, Clin. Oncol. 5, 245 (1993). 12 S. A. Strobel, H. E. Moser, and P. B. Dervan, J. Am. Chem. Soc. 110, 7927 (1988). 13 I. Radhakrishnan and D. J. Patel, Biochemistry 33, 11405 (1994).

[101 D N A TRIPLEXES AND QUADRUPLEXES 227

triplets, formation of which requires protonation of the Hoogsteen-paired C at N3 (Fig. 1). More recently, purine-purine-pyrimidine (purine motif) triplexes have been characterized in which the additional purine-rich strand binds in an antiparallel orientation to the Watson-Crick purine strand. The third strand is G-rich but not necessarily all purines. In this motif, G bonds to Watson-Crick G. C base pairs to form G. G. C triplets, while either A or T may bond to A. T base pairs, forming A" A- T or T. A. T triplets, re- spectively.

Studies of quadruplexes in our laboratory were initiated as investigations into the structures formed by the G-rich strand of telomeres. 14'15 These short, G-rich repeat sequences were proposed to form quadruplex struc- tures. ~6'17 NMR and X-ray crystallographic studies confirmed the G-quartet structure (Fig. 2) and provided three-dimensional structures of telomeric quadruplexes. 18'19 G-quadruplexes have also been proposed to occur in immunoglobulin switch regions 2° and the dimerization domain of the HIV- 1 RNA genome. 21'22 G-quartets have appeared as structural motifs in a variety of both RNA and DNA aptamers, 23-2s which are oligonucleotides selected in vitro from random sequences for their ability to bind a specific target molecule. 29

Proton NMR studies of DNA structures have inherent assignment prob- lems that are not present in protein NMR studies. 3°'31 In proteins, intrapep-

a4 y. Oka and C. A~ Thomas, Jr., Nucleic Acids Res. 15, 8877 (1987). 15 E. Henderson, C. C. Hardin, S. K. Walk, I. Tinoco, Jr., and E. H. Blackburn, Cell 51,

899 (1987). 16 j. R. Williamson, M. K. Raghuraman, and T. R. Cech, Cell 59, 871 (1989). 17 W. I. Sundquist and A. Klug, Nature 342, 825 (1989). is F. W. Smith and J. Feigon, Nature 356, 164 (1992). 19 C. Kang, X. Zhang, R. Ratliff, R. Moyzis, and A. Rich, Nature 356, 126 (1992). 20 D. Sen and W. Gilbert, Nature 334, 364 (1988). 21 W. I. Sundquist and S. Heaphy, Proc. Natl. Acad. Sci. U.S.A. 90, 3393 (1993). 22 G. Awang and D. Sen, Biochemistry 32, 11453 (1993). 23 R. F. Macaya, P. Schultze, F. W. Smith, J. A. Roe, and J. Feigon, Proc. Natl. Aead. Sci.

U.S.A. 90, 3745 (1993). 24 p. Schultze, R. Macaya, and J. Feigon, J. MoL BioL 235, 1532 (1994). 25 L. C. Bock, L. C. Griffin, J. A. Latham, E. H. Vermaas, and J. J. Toole, Nature 355, 564 (1992). 26 D. E. Huizenga and J. W. Szostak, Biochemistry 34, 656 (1995). 27 K. Y. Wang, S. McCurdy, R. G. Shea, S. Swaminathan, and P. H. Bolton, Biochemistrv 32,

1899 (1993). 28 K. Padmanabhan, K. P. Padmanabhan, J. D. Ferrara, J. E. Sadler, and A. Tulinsky, J. Biol.

Chem. 268, 17651 (1993). 29 j. W. Szostak, TIBS 17, 89 (1992). 30 K. WtRhrich, "NMR of Proteins and Nucleic Acids." John Wiley & Sons, New York, 1986. 31 j. Feigon, V. Sklenar, E. Wang, D. E. Gilbert, R. F. Macaya, and P. Schultze, in "Methods

in Enzymology" (D. M. J. Lilley and J. E. Dahlberg, eds.), Vol. 211, p. 235. Academic Press, San Diego, 1992.

228 D N A AND R N A STRUCTURE [10]

A

C

H H

o.-~.~.-Lo B o3..~.~

R N-R---O R

I~1 N-R---O ..f'-'-~ Me

0 H

~ N

I I

. f ,, ,, " ,~ .¢"N ,- ~ o---H-N

R J ~ I - - H

A

D , ~ N ~R

N

/'~'-'~".~...o ~ = . H

Rr \ " / ,,, \ - / . - ( / ) \

M e , ~ J ' ~ N ~ R

E o¢A2-~ /~"~.__of..=

" ' ~ / o~£

[10] DNA TRIPLEXES AND OUADRUPLEXES 229

tide connectivities are through-bond, interpeptide sequential connectivities are through-space, and sequential connectivities between these can be made without structural assumptions. In contrast, for DNA, sequential connectivi- ties rely only on through space connectivities for both the intranucleotide and internucleotide steps. These connectivities are between base H8,H6 and sugar HI',H2',H2" and are sequential along a strand only in the case of a helical structure with all anti nucleotides. 32 Thus, the standard sequential connectivities for DNA are only valid for A- or B-DNA helices. In the case of loops, syn bases, or other alternative conformations, these methods can fail or give erroneous assignments. 31

Here we review the methods used in our laboratory for 1H NMR struc- tural studies of DNA triplexes and quadruplexes. In general, these methods start with those used for DNA duplexes, which have been well reviewed previously, 3°m,33-35 as well as Schmitz and James, in Chapter [1], and will not be discussed here.

NMR Sample Preparation

DNA oligonucleotides are synthesized chemically on a DNA synthesizer and purified by Sephadex size exclusion chromatography as previously described. 31 For G-rich samples that can form quadruplexes, certain precau- tions should be taken to ensure that the sample contains only the cation of interest, as these samples will tightly bind any K ÷ impurity that may be present. 36 The Sephadex column should be prewashed with 10-20 ml of a i M solution of the desired cation and then thoroughly washed with water before loading the sample to ensure that only the desired counterion is present. In both quadruplexes and triplexes, the structure formed will often depend on the cation(s) present. Counterions can be exchanged by passing

32 j. Feigon, A. H.-J. Wang, G. A. van der Marel, J. H. van Boom, and A. Rich, Nucleic Acids" Res. 12, 1243 (1984).

33 B. R. Reid, Q, Rev. Biophys. 20~ 1 (1987). 34 D. E. Wemmer and B. R. Reid, Annu. Rev. Phys. Chem. 36, 105 (1985). 35 D. J. Patel, L. Shapiro, and D. Hare, Q. Rev. Biophys. 20, 35 (1987). 36 F. W. Smith and J. Feigon, Biochemistry 32, 8682 (1993).

F16.1. The base pairing schemes of triplets found in pyrimidine • purine • pyrimidine tri- plexes (A and B) and purine.purine-pyrimidine triplexes (C-E). In pyrimidine motif tri- plexes, the third-strand pyrimidines bind to the purines by formation of Hoogsteen hydrogen bonds. In purine motif triplexes, the third-strand bases bind to the Watson-Crick purines by reverse Hoogsteen hydrogen bonding. Arrows are drawn between protons whose NOE cross- peaks define the base pairing scheme and/or are used in assignments. Dashed arrows indicate NOE conneetivities that are not always observable due to exchange.

230 DNA AND RNA STRUCTURE [10l

H

l "N uuluu)-1 ~ N F:I

H ~ N \ O -

\ H " H

H

FIG. 2. The base-pairing scheme in a G-quartet. An inosine replaces one of the guanines in this quartet to illustrate the short IH2-GH8 internucleotide distance.

the sample over a Dowex AG50 column equilibrated with the appropriate cation. The sample is then stored lyophilized until it is dissolved in the desired solution for the NMR experiment.

For triplex samples, we have found that concentrations above approxi- mately 2 mM generally lead to some sample aggregation and a concomitant decrease in signal to noise. In contrast, for the folded quadruplex structures, much higher concentrations can be studied with no deleterious effects. In the dimeric quadruplexes we routinely make samples over 5 mM in strand (>2.5 mM dimer). These samples give excellent signal-to-noise ratios with relatively few scans. This adds greatly to the ability to observe weak cross-peaks.

Formation and stability of both triplex and quadruplex structures are cation dependent, and in the case of pyrimidine motif triplexes there is a strong pH dependence as wel l . 37 Fig. 3 shows the imino resonances of a putative purine motif triplex in different cation conditions. 38 Very different

37 y. K. Cheng and B. M. Pettitt, Prog. Biophys. Mol. Biol. $8, 225 (1992). 38 R. M. Macaya, "NMR Studies of Intramolecular DNA Triplexes and Quadruplexes." Ph.D.

Thesis, University of California, Los Angeles (1993).

[101 D N A TRIPLEXES AND QUADRUPLEXES 231

i . . G T G G G

• r,." +);., ~ C C C C-~'

t 14 12 10 8 PPM

FIG. 3. The one-dimensional spectra of the imino, amino, and aromatic regions of a putative purine • purine • pyrimidine triplex under different ionic conditions at 1 °. The sample conditions were: (A) 100 mM NaCI, pH 6.3; (B) 100 mM LiC1, pH 6.0; (C), 100 mM LiC1, 20 mM ZnCI2, pH 5.9; (D) 20 mM ZnC12, pH 6.0; and (E) 50 mM MgClz, 5 mM LiC1, pH 6.7 (from Macaya38).

spectra are observed depend ing on the cations present . In addition, a l though it may not be evident f rom the one-d imens iona l spectra, under all of these condi t ions there is m o r e than one confo rma t ion present . This is the rule ra ther than the except ion with pur ine mot i f triplexes, making them difficult to character ize spectroscopically.

232 DNA AND RNA sx~ucxu~z 110]

For optimal observation of imino cross-peaks in both triplexes and quadruplexes, the pH of the sample should be pH 6 or lower. For pyrimidine motif triplexes, low pH is needed to drive the equilibrium toward the triplex conformation, since protonation of third-strand Cs is required. In this case, we have found the optimal pH to be 5.2. Below this pH, one needs to be concerned about protonation of other bases. Since divalent cations are often used to stabilize the triplex structures, phosphate buffer is not satisfac- tory. Since other buffers are protonated, we use no added buffer, rather we simply adjust the sample pH of the NMR sample by addition of aliquots of base (usually NaOH) or acid (usually HC1) as needed.

~H NMR Spectra of Triplexes

Sequence Design

In our 39'4° and other 41-44 early studies of triplexes, the pyrimidine motif triplex structure was generally formed by mixing an oligonucleotide com- posed of purines with the complementary pyrimidine strand in a 1 : 2 ratio and lowering the pH to stabilize the triplex conformation. These early studies confirmed the proposed T . A . T and C +. G. C triplet pairing schemes (Fig. 1) and provided the first evidence that the sugar conforma- tions in triplexes were not all C3'-endo 39 as proposed from fiber diffraction by Arnott and co-workers. 45 However, detailed structural characterization of such triplexes was hindered by the fact that in addition to triplexes, multiple other structures could be (and were) present in the samples. 39'46 These include, in addition to duplex and single strand, self-complexes of the purine and possibly the pyrimidine strand and also possibly a second triplex composed of two purine and one pyrimidine strand. 46 The self- complexes of the purine strand could include quadruplexes, depending on sequence. Because of the problem of multiple conformations, we turned to the design of unimolecular triplexes in which a single strand folds to form an intramolecular triplex with short loops connecting the Watson-Crick and

39 p. Rajagopal and J. Feigon, Biochemistry 28~ 7859 (1989). 40 p. Rajagopal and J. Feigon, Nature 339, 637 (1989). 41 C. de los Santos, M. Rosen, and D. Patel, Biochemistry 28, 7282 (1989). 42 D. S. Pilch, C. Levenson, and R. H. Shafer, Proc. Nat. Acad. Sci. U.S.A. 87~ 1942 (1990). 43 t . S. Kan, D, E. Callahan, T. L. Trapane, and P. S. Miller, J. BiomoL Struct. Dyn. 8, 911 (1991). 44 M. M. W. Mooren, D. E. Pulleyblank, S. S. Wijmenga, M. J. J. Blommers, and C. W. Hilbers,

Nucleic Acids Res. 18, 6523 (1990). 45 S. Arnott and E. Selsing, J. MoL Biol. 88~ 509 (1974). 46 j. Feigon, D. E. Gilbert, E. Wang, P. Rajagopal, G. A. van der Marel, and J. H. van Boom,

in "Proceedings of the Sixth Conversation in Biomolecular Stereodynamics" (M. H. Sarma and R. H. Sarma, eds.), Vol. 3, p. 207. Adenine Press, New York, 1990.

[ 10] DNA TRIPLEXES AND QUADRUPLEXES 233

, 3' 5 ' A G A G A G A A ~ T T C T C T C T T3 T C T C T C T T

5'

' rcTc c

÷ + ÷

T3 A G A G A G A A

Fie, 4. A schematic of the s e ~ e n c e and Riding pathway of a pyrimidine • purine • pyrimi- dine intramolecular t ~ l e x .

Watson-Hoogsteen strands (Fig. 4). 47 This design has been used in nearly all triplex structural studies done since our original report. It should be cautioned, however, that one still needs to be careful in choosing the sequence and conditions, since alternative or multiple structures may form depending on the sequence and buffer conditions (see Fig. 3) or if the molecule is not well designed. For example, designed triplexes with tracts of Gs may form quadruplexes instead of or in addition to triplexes. It is also possible that two molecules may dimerize to form an intermolecular triplex. The polarity of the Hoogsteen-paired strand relative to the Watson- Crick purine strand needs to be considered as well, since for pyrimidine motif triplexes the Hoogsteen-paired strand is parallel, whereas for purine motif triplexes the Hoogsteen strand is usually antiparallel to the Watson- Crick purine strand. Finally, the loop sequences should be carefully consid- ered, both from the standpoint of stability and possible alternative structure formation. We note that the intramolecular triplex forms a nice model system for the proposed H-DNA, 8 which may form in vivo.

Basic Features of One-Dimensional 1H Spectra of Triplexes in 11:0

Formation of triplexes can be most readily monitored by observation of the exchangeable imino and amino proton resonances. Fig. 5 shows the one-dimensional 1H spectra of the exchangeable resonances of a pyrimidine motif intramolecular triplex. 48 Each Watson-Crick and Hoogsteen base

47 g. Sklenar and J. Feigon, Nature 345, 836 (1990). 48 R. F. Macaya, E. Wang, P. Schultze, V. Sklenar, and J. Feigon, J. Mol. BioL 225, 755 (1992).

234 DNA AND RNA STROCTU~ [10]

Im inos G4,G6 /

T18 T16 T28T30 C27.1.IC29j~ I 11~ 4 J j G2kll T31 T'sln Amino8

c~5 ( , Ic,2r c25,

16 15 14 13 12 11 10 9 p.p.m.

FIG. 5. A portion of a one-dimensional spectrum of the imiao and C + amino resonances of the intramolecu]ar triplex shown in Fig. 3. The sample conditions were 100 rmA,/NaC], 5 mM MgC]2, 1.9 m~r DNA, 90% HzO/10% D20, 1 °. Sequence-specific assignments, derived from analysis of NOESY spectra in H20, are indicated (after Macaya et aL48).

pair gives rise to one hydrogen-bonded imino resonance in the 12-16 ppm range, and observation of the additional Hoogsteen pair iminos is good evidence for triplex formation. The protonated C iminos in pyrimidine motif triplexes generally resonate fairly low field, between about 14 and 16 ppm. The protonated C amino resonances also appear at lower field than unprotonated C aminos, between 9 and 10.5 ppm; observation of these resonances provides further strong evidence that a triplex has formed.

Assignment of Exchangeable Resonances of Triplexes

The usual starting point for the study of a new triplex is to first perform a one-dimensional NMR temperature study of the sample in 1-120 to assess triplex formation and stability. 48 Assignment of the imino proton resonances can usually be obtained in a straightforward way from a 2D NOESY spec- trum acquired using a water suupression pulse sequence for the observe pulse. In our laboratory the best results have been obtained with an li-echo pulse sequence. 49 The WATERGATE 5° pulse sequence can also be used, but some of the broader (faster exchanging) resonances may not be observed. The low pH used to promote formation of pyrimidine motif triplexes also functions to slow the exchange rate of the imino and amino resonances, thus facilitating observation of their NOE cross-peaks. These spectra need to be obtained at 10 ° or below in order to observe all the cross-peaks; best results are almost always obtained at 1 °.

The T. A. T and C +. G. C triplets originally proposed for the pyrimi- dine motif (Fig. 1A) were first confirmed by NMR on the basis of intratriplet NOEs (as discussed later). 39-41'47 Under optimal conditions, two sets of

49 V. Sklenar and A. Bax, J. Magn. Reson. 74, 469 (1987). 50 V. Sklenar, M. Pilotto, R. Leppik, and V. Saudek, J. Magn. Reson. Ser. A 102, 241 (1993).

[ 10] DNA TRIPLEXES AND QUADRUPLEXES 235

sequential imino-imino connectivities can be traced: one set between the Watson-Crick paired iminos and the other between the Hoogsteen-paired iminos (Fig. 6). The C ÷ iminos are usually readily identifiable by their cross-peaks to the intranucleotide pair of lowfield shifted C ÷ amino proton resonances and provide a useful starting point for the sequential assign- ments along the Hoogsteen-paired strands. However, the C ÷ iminos are particularly labile, especially at higher pHs.

Confirmation of the imino assignments and assignment of some of the aromatic and amino resonances can be obtained by analysis of their intra- triplet NOE cross-peaks (Fig. 1 and Fig. 7). 47,48 Watson-Crick A - T pairs are identified by strong intrabase pair T imino-AH2 NOEs; Hoogsteen- paired A. T are identified by strong intranucleotide T imino-AH8 NOEs; and the T. A. T triplet is identified by NOEs between an A amino pair and both a Watson-Crick and Hoogsteen imino. Watson-Crick G. C pairs are identified by strong G imino to C amino cross-peaks and a single broad G imino to G amino cross-peak, and Hoogsteen C +. G base pairs are identified by C + imino to C + amino cross-peaks. Cross-peaks between the C + aminos and the C aminos in a C + • G. C triplet are sometimes, but not always, identifiable. Pairs of sharp, resolved resonances are observed for

@ e N

I i

@.

h i~.2 14.4 ~ . 6 12'.a

0pm

FIG. 6. The imino-imino region of a two-dimensional NOESY spectrum of the same triplex and with sample conditions similar to those listed in the caption to Fig. 4. The imino-imino sequential connectivities for the Watson-Crick base pairs are indicated above the diagonal, and those for the Hoogsteen base pairs are indicated below the diagonal (after Macaya et al.48).

236 DNA AND RNA STRUCTURE [101

O

++ ,<

o~

. . . . A A~H8 ' ~ H 8 • A5N6~IC~IAsH]~"~-~-A'7HcB N4 V .~..~i(v~TH 2 ~ w ~ 15 ...

A3 N(L~mM~I G4H8 41, tr,l G ~ . , ] m ~ . , a

AsH2 A5N6 5 T M

~0 e+9~

• + ++\ l \ +

,<I 0/..,,,,., °.,,. / / c29N4

15'.2 t4.4 13'.6 ppm

~C19N4 ct7~ as

O

to n

tO

"o;

12'.e

FIG. 7. The imino-aromatic, amino region of the same NOESY spectrum as Fig. 5, with the intratriplet cross-peaks labeled (after Macaya et a/.48).

both the A and C amino protons, but the G amino protons are generally more rapidly exchanging, resulting in a single broad cross-peak for the pair that often cannot be assigned.

Particularly interesting and informative are some of the in ter tr iple t NOE cross-peaks that are observed. For example, pyrimidine iminos of the Hoogsteen base paired strand have cross-peaks to the 5'-neighboring purine H8 protons and to their H2', H2" protons. Such cross-peaks are not observed in duplex B-DNA but are characteristic of triplex DNA. NOE cross-peaks are seen not only between the T methyl protons of the Hoogsteen base paired strand and their own imino protons, but also to the 5'-neighboring Hoogsteen base paired imino protons. In addition to NOE cross-peaks to their own imino protons, C+ amino protons may also have cross-peaks to the imino protons of the Hoogsteen base pair in both the 5' and 3' directions, to the imino proton of the G of its own triplet, and to the imino proton of the Watson-Crick base pair in the 3' direction (i.e., 3' with respect to the purine strand).

Purine • purine • pyrimidine triplexes represent the second structural mo- tif of triple-stranded DNA. In this instance, the third purine-rich strand binds to the purine strand of the Watson-Crick duplex by the formation of reversed Hoogsteen-type hydrogen bonds and in an antiparallel orien- tation. The resultant triplex is composed of G. G. C and A- A. T or T- A- T triplets. As might be expected, both similarities and differences with the

[101 DNA TmPLEXES AND OUADRUPLEXES 237

pyrimidine • purine • pyrimidine motif are observed. 51-53 Sequential imino- imino connectivities will be interrupted in the third strand if there are A. A. T triplets, since the reverse Hoogsteen A. A pair has no imino proton. The NOE cross-peaks described previously for T. A. T triplets are also seen for T- A. T triplets found in purine • purine • pyrimidine triplexes. Imino protons of G- C Watson-Crick base pairs exhibit the expected fea- tures; there are NOE cross-peaks between the G imino protons to their own C amino protons, to any adjacent AH2 protons, and possibly to neigh- boring C or A amino protons. In contrast to the usual case for G amino proton resonances, clear NOE cross-peaks can be identified between the third-strand G amino protons and the G iminos on the same nucleotide, as well as intratriplet NOE cross-peaks to the C amino protons and to the triplet in the 3' direction (i.e., 3' with respect to the Watson-Crick purine strand). Although only one G amino resonance is observed for the pair of G amino protons, it is less broad and thus the cross-peaks are more easily detected than those from the Watson-Crick-paired G aminos. 53 Another characteristic NOE cross-peak is between a third-strand G imino proton and the H8 of the Watson-Crick strand G to which it is bound. NOE cross- peaks may also be observed between third-strand G imino protons and A amino protons of adjacent T. A. T triplets. As in the case of third-strand pyrimidine imino protons mentioned earlier, third-strand G imino protons in this motif may also have NOE cross-peaks to the H8, H2', and H2" protons of the Watson-Crick purine in the 5' direction (i.e., 5' with respect to the Watson-Crick purine strand). In A. A. T triplets, third-strand A amino protons have NOE cross-peaks to both the H8 and amino protons of the Watson-Crick A to which it is bound, as well as to the adjacent 5' imino proton. Third-strand AH2 protons have cross-peaks to the amino protons of the Watson-Crick A to which it is bound, to C amino protons of a 5' triplet (again 5' with respect to the Watson-Crick purine strand), and to the adjacent 3' imino proton. 54

Assignment of Nonexchangeable Resonances

Strategies for the assignment of triple-stranded DNA have evolved from techniques developed for canonical right-handed D N A . 3°'31'33-35 Since sequential assignment methods developed for DNA assumed a right- handed helix with all anti glycosidic torsion angles, in the early NMR studies of triplexes it was not obvious that these methods would lead to correct

5~ I. Radhakrishnan, C. de los Santos, and D. J. Patel, J. Mol. Biol. 221, 1403 (1991). 52 I. Radhakr ishnan and D. J. Patel, Structure 1, 135 (1993). 53 K. Dittrich, J. Gu, R. Tinder, M. Hogan, and X. Gao, Biochemistry 33, 4111 (1994). .s4 I. Radhakrishnan, C. de los Santos, and D. J. Patel, J. Mol. Biol. 234, 188 (1993).

238 DNA AND RNA STRUCTURE [10]

assignments. However, now that a right-handed helical structure for tri- plexes has been established, it is clear that sequential base-sugar assign- ments can usually be traced along each of the three strands, although some differences from standard B-DNA do exist and should be considered in the assignment process. Detailed description of the assignment process for pyrimidine motif triplexes can be found in Macaya et al. 48

Work on assigning the nonexchangeable resonances usually begins once the NOESY 55 spectrum of the sample in 1-120 has been at least partially assigned. This generally involves the data already obtained from the spec- trum of the exchangeable resonances and analysis of NOESY spectra of the sample in D20, as well as T O C S Y , 56,57 HOENOE, 58 and P . C O S Y 59'6°

spectra. For standard triplexes, sequential connectivities can usually be found along the entire length of the individual "strands. ''48 The easiest approach is usually to get sequential assignments in the base HI ' region using the more crowded base-H3' and base-H2',H2" regions for confirma- tion of assignments. For the purine "strand," use can be made of the H8 assignments already obtained from analysis of the NOESY spectrum of the exchangeable resonances. Although the NOESY spectra in 1-120 are usually obtained at a lower temperature than is used for the NOESY spectra in D20, which give optimal resolution at higher temperature, the H8 chemical shifts correlate fairly well between the two spectra.

Certain trends can be seen in the chemical shifts of the aromatic reso- nances in pyrimidine motif triplexes. In general, the purine resonances are shifted upfield and the Hoogsteen-paired pyrimidine resonances are shifted downfield from their usual chemical shifts in B-DNA. The AH8 especially are shifted to higher field and can appear as low as 7 ppm compared to around 8 in B-DNA (see Fig. 7). The chemical shifts of the base protons of the protonated C in the third strand are pH dependent; therefore, care should be taken in comparing one sample to another where small changes in pH may cause relatively large changes in chemical shift. This may also be a problem if Cs are used in the loops, since they may become protonated at the lower pHs usually used to study triplexes.

The TH6 and T methyl and the CH6 and CH5 resonances can be identified from the P.COSY and H O H A H A spectra. Analysis of the H O H A H A and P.COSY spectra should also facilitate identification of the individual deoxyribose spin systems. (Coupling constants for use in structure

55 A. Kumar, R. R. Ernst, and K. WiJthrich, Biochem. Biophys. Res. Commun. 95, 1 (1980). 56 A. Bax and D. G. Davis, J. Magn. Reson. 65~ 355 (1985). 57 L. Braunschweiler and R. R. Ernst, J. Magn. Reson. 53, 521 (1983). 58 V. Sklenar and J. Feigon, J. Am. Chem. Soc. 112, 5644 (1990). 59 D. Marion and A. Bax, J. Magn. Reson. 811, 528 (1988). 60 L. Mueller, Z Magn. Reson. 72, 191 (1987).

[ 10l DNA TRIPLEXES AND QUADRUPLEXES 239

calculations can also be obtained from P,COSY or P.E.COSY spectra.) Unambiguous identification of NOE cross-peaks arising from the CH6 resonances can be obtained from a H O E N O E spectrum. Simplification of the spectrum can also be accomplished by deuterating the purines at C8. This is useful both for resolving overlapping cross-peaks and for verification of the purine H8 assignments. Other methods for resolving overlapped regions include varying the temperature, changing the sample conditions, and the use of homonuclear 3D NOESY-TOCSY 6t'62 or NOESY-NOESY 63 spectra. 3D NOESY-TOCSY have been used by Patel and c o - w o r k e r s 64"6s

to trace the sequential connectivities in the base-H2',2" region of intramo- lecular triplexes. The CH5 planes in o3 were also used to edit effectively the spectra to only the CH6 NOE cross-peaks, 66 analogous to the 2D HOENOE experiment. 5s Although these experiments can in specific in- stances be helpful in resolving overlapping cross-peaks and provide redun- dant information that can be used as a check on assignments, we have found that for the most part cross-peaks that are unresolved in two-dimensional NOESY experiments, particularly the H4', H5', and H5" cross-peaks, re- main unresolved in the 3D experiments.

What has thus far been said concerning the assignment of nonexchange- able resonances also applies to the Watson-Crick duplex region of purine- purine-pyrimidine triplexes; however, interesting sequence-specific pertur- bations may arise in the antiparallel purine-rich third strand. 54 For example, in triplexes containing G - G . C and T. A. T triplets, TpG steps are un- derwound and somewhat extended so that sequential NOEs between these residues are either absent or weak. A partially compensating trend is ob- served at GpT steps. Distinctions are also present in the chemical shifts (i.e., the H2',H2" protons of T in a TpG step and the HI ' proton of the 5'G in a GpG step resonate upfield of their usual positions). In general, third-strand purine H8 protons resonate downfield from where they would typically be found in canonical duplex DNA.

Final mention should be made of a set of NOE cross-peaks that are characteristic of pyrimidine.purine, pyrimidine triplexes. These are the cross-peaks between purine H8 protons and the HI ' resonances of the 3'- neighboring pyrimidine of the Hoogsteen base-paired strand. These are

61 G. W. Vuister, R. Boelens, and R. Kaptein, J. Magn. Reson. 80, 176 (1988). 62 H. Oschkinat, C. Griesinger, P. J. Kraulis, O. W. Sorenson, R. R. Ernst, A. M. Gronenborn,

and G. M. Clore, Nature 332, 374 (1988). 63 R. Boelens, G. W. Vuister, T. M. G. Koning, and R. Kaptein, J. Am. Chem. Soc. 111,

8525 (1989). 64 I. Radhakrishnan, D. Patel, and X. Gao, J. Am. Chem. Soc. 113, 8542 (1991). 6s I. Radhakrishnan, D. J. Patel, and X. Gao, Biochemistry 31, 2514 (1992). 66 M. E. Piotto and D. G. Gorenstein, J. Am. Chem. Soc. 113, 1438 (1991).

240 DNA AND RNA STRUCTURE [101

not only useful in terms of verifying the assignments made, but also provide interstrand distance constraints that may be used in structure calculations. The analogous cross-peaks are not observed in purine.purine.pyrimi- dine triplexes.

IH NMR Spectra of Quadruplexes

Folding Topologies and Sequence Design

There are four possible variations in strand orientation in guanine quad- ruplexes (Fig. 8). Four tracts of guanine residues come together to form the quartets, but any two strands can have parallel or antiparallel orientation. Quadruplex structures with all possible strand orientations have now been solved. All parallel strand orientation is observed for tetrameric se- quences. 62-71 The other strand orientations have been observed for dimeric or unimolecular quadruplexes that contain loops joining the G t r a c t s J 8't9'23'27'72-76 Loops at the ends of quadruplexes can cross an edge or the diagonal of the end quartet. It is also possible for a loop to span from one end of the quadruplex to another along a g r o o v e . 73'77 Depending on the orientation of the loops, all adjacent strands can be antiparallel or there can be a mixture of parallel and antiparallel strands. If a strand's direction is reversed, the guanine bases must be rotated 180 ° about the glycosydic bond to maintain the quartet hydrogen bonding scheme. Thus, both syn and anti nucleosides are found in quartets made of tracts having varying polarities. The relationship between loop length and sequence and loop orientation are not known.

In design of a sequence, consideration should be given to the potential for an axis of symmetry. Note that a guanine quartet has two faces; there- fore, in the absence of all parallel strands, an odd number of quartets cannot be symmetrical. TM There are several advantages and disadvantages

67 F. Aboul-ela, A. I. H. Murchie, and D. M. J. Lilley, Nature 3611, 280 (1992). 68 G. Laughlan, A. I. H. Murchie, D. G. Norman, M. H. Moore, P. C. E. Moody, D. M. J.

Lilley, and B. Luisi, Science 265, 520 (1994). 69 C. Cheong and P. B. Moore, Biochemistry 31, 8406 (1992). 70 y. Wang and D. J. Patel, Biochemistry 31, 8112 (1992). 71 y. Wang and D. J. Patel, J. Mol. Biol. 234, 1171 (1993). 72 y. Wang and D. J. Patel, Structure 1, 263 (1993). 73 y. Wang and D. J. Patel, Structure 2, 1141 (1994). 74 F. W. Smith, F. W. Lau, and J. Feigon, Proc. Natl. Acad. Sci. U.S.A. 91, 10546 (1994). 7s p. V. Scaria, S. J. Shire, and R. H. Sharer, Proc. Natl. Acad. Sci. U.S.A. 89, 10336 (1992). 76 K. Y. Wang, S. Swaminathan, and P. H. Bolton, Biochemistry 33, 7517 (1994). 77 O. L. Acevedo, L. A. Dickinson, T. J. Macke, and C. A. Thomas, Jr., Nucleic Acids Res.

19, 3409 (1991).

[ 10] DNA TRIPLEXES AND QUADRUPLEXES 241

A B C

M

FIG. 8. The possible strand polarities in G-quadruplexes: (A) all strands parallel, (B) three parallel and one antiparallel strand, (C) two pairs of parallel and two pairs of antiparallel strands, and (D) all antiparallel strands. The G tracts may be independent or connected by loops. Groove widths are indicated as M (medium), W (wide), and N (narrow). Structures of quadruplexes with these strand orientations include: (A) d(TG4T) (solution and crystal), 67"68 d(T2AG3), d(T2G4), 7° d(TrGaT), 71 and r(UG4U); 69 (B) d(T2G4)4; 73 (C) d(G4T4G4), 18's8 d(G4T4G4T4G4T4G4), TM and d(AG3T2AG3TzAG3T2AG3); 72 and (D) d(GnT4G4) crystal struc- ture 19 and d(G2T2G2TGTG/T2G2). 23~4'27 These structures are NMR-derived solution struc- tures except where noted.

to having symmetric molecules. A two-fold symmetry more than halves the number of resonances and doubles the strand concentration at a given quadruplex concentration. Both effects greatly clarify NMR spectra at the expense of a potential ambiguity in NOEs between protons near the symme- try axis. The main advantages are that the spectra are more simple and have better signal-to-noise ratio. A problem with the unimolecular quad- ruplexes is that they may be nearly symmetric; therefore, many residues may be severely overlapped.

Basic Features of One-Dimensional 1H Spectra of Quadruplexes

Formation of quadruplexes can be most readily monitored by observa- tion of the exchangeable imino and amino proton resonances. Fig. 9 shows the one dimensional 1H spectra of the exchangeable and nonexchangeable resonances of a symmetric bimolecular quadruplex formed by d(G4T4G4) [Oxy-1.5].36 Each guanine quartet gives rise to four hydrogen-bonded imino resonances in the 10.5-12.5 ppm range, and observation of these iminos is good evidence for quadruplex formation. TM The number of imino resonances is the first indication of a symmetric molecule. If an inosine residue has been included, the inosine imino resonance should be observed near 14 ppm. An unusual feature in the DzO spectrum is resonance intensity from some of the exchangeable imino and amino protons. These protons can be unusually long lived in quadruplexes, especially in foldback quadruplexes, and depending on sequence and conditions some of these may be observable for days to months after dissolving the sample in D20. TM Guanine H8

242 DNA AND RNA STRUCTURE [10]

1.921210

I ,IA 3 .

12 10 8 6 4 2

I I

PPM

FIG. 9. One-dimensional ~H spectra of the symmetric dimeric quadruplex formed by d(G4T4G4): (A) sample in HzO and (B) sample in D20. Characteristic features include hydro- gen-bonded G imino resonances at 10.5-12.5 ppm and G amino resonances between 5 and 11 ppm (after Smith and Feigon36).

protons resonate between 6.5 and 8.5 ppm, with those arising from syn and anti residues generally at the lower and upper end of that range, respectively. Inosine H8 protons resonate at slightly higher ppm than a guanosine H8 in the same environment. For the most part, loop resonances are distributed throughout the quadruplex resonances.

Assignment of Nonexchangeable Resonances of DNA Quadruplexes

In general, the assignment strategies for quadruplex DNA structures are similar to those used for B-DNA or triplex DNA: resonance assignments are made using a combination of through-bond (P.COSY, TOCSY, aH-31P H S Q C 78) and through-space (NOESY) NMR experiments. The difference arises in that the NOEs used to find "standard" sequential assignments are often missing or reversed in quadruplex spectra.

In tetrameric quadruplexes, made up of four separate strands, the gua- nine tracts align with parallel orientation in all quadruplexes studied to date . 67-7a All guanosine residues are in the anti conformation, all the quar- tets are in one orientation, and B-DNA assignment strategies may be applied. Since all residues are in the anti conformation as in B-DNA and a right-handed helix is formed, the "standard" sequential HI ' -H8 and H2',2"-H8 sequential NOEs are observed along each strand. The sequential

78 G. Bodenhausen and D. Ruben, Chem. Phys. Lett. 69, 185 (1980).

[ 10] DNA TRIPLEXES AND QUADRUPLEXES 243

connectivities generally extend into the sequences flanking the G-tract. In all reported structures, the four strands have identical sequences and are symmetric. Only one strand is observed, although a second set of resonances corresponding to the uncomplexed single strand is also seen in most cases. The only NOESY evidence for the quartet structure that connects the four strands are the long-range Gimino-GH8 NOEs not otherwise expected.

In the case of unimolecular or bimolecular quadruplexes, strand reversal is required. All such foldback quadruplexes studied to date have some syn bases. 18'19'23'27'72-76'79 This is a consequence of the change in strand polarity; the geometry of the G-quartet cannot be maintained without flipping some nucleotides from anti to syn. Within a given quartet, this will depend on the loop orientation and cannot be predicted a priori. The first step in assigning foldback quadruplexes is to identify any syn bases. These are clearly evident in the aromatic HI ' region of the NOESY spectrum as very large NOE cross-peaks. Large pyrimidine H5-H6 cross-peaks appear in the same region, but they can easily be distinguished using the P.COSY spectrum. If there are no syn guanosine residues, the structure is probably not an antiparallel quadruplex. De- pending on the glycosydic conformation of adjacent residues, the NOEs available for sequential assignment varies, as shown in Fig. 10 and Table I and as discussed next.

1. 5' anti-3' anti: In this case, normal B-DNA NOE connectivities should be present, assuming a regular backbone. For instance, in the aromatic HI ' region, there should be GnHI' and GnH2',H2" to Gn÷IH8 NOEs in addition to intraresidue NOEs.

2. 5' syn-3'anti: If a syn residue is followed by an anti residue, there is a characteristic set of NOEs. Two such sequences are present in each GGGG tract in Oxy-l.5, which has syn-anti-syn-anti guanines along each "strand" of the quadruplex. In addition to the GnHI'-Gn+IH8 observed in B-DNA, there is a weak Gn+IHI'-GnH8 (reversed from B-DNA). A weak GnH8-G,+IH8 NOE is also observed. The Gn+IH8 has strong NOEs to both the Gn and Gn÷IH2',H2". The GnH8 has weak NOEs to the G,H2',H2" and none to the Gn÷IH2',H2". Similarly, both GnH3' and Gn÷IH3' have medium NOEs to the Gn+1H8, but have weak NOEs at most to the GnH8.

3. 5'anti-3syn: No sequential basen-sugarn-basen+~-sugarn+~ NOEs are

79 y. Wang, S. de los Santos, X. Gao, K. Greene, D. Live, and D. J. Patel, J. Mol. Biol, 222, 819 (1991).

244 D N A AND RNA STRUCTURE [10]

,

Sequential connectivity ~ 5'svn-svn ~' reversed _ (Ta|l-to-Flead)

Both s y n ~ 5 ' svn-anti 3 ' sequential connectivities (Tail-to-Tail)

Standard B-DNA ~ a n t i sequential connectivity ~ ~ .... 5'anti-anti3' (Head-to-Tail)

~ ~ 1 ~ Neither 5'anti-svn3' sequential (Head-to-Head) connectivity

syn 3 ' /

3' 3' 5'

H e a d - t o - H e a d ~ ~ ~ [ i ~

Head-to-Tail .)tg'd//~''~ J g ' d ~ h 4 ' d ~ - -

FIG. 10. (A) Schematic representation of NOEs to be expected between possible GpG steps observed in G-quadruplexes. (B) Illustration of intra- and interstrand NOEs containing the GpG steps in (A). Circles represent the sugar protons, primarily HI', H2', H2", and H3'. Bold circles indicate that the residue is in the syn conformation. Weak NOEs are observed between adjacent quartets, and very weak NOEs are observed around each quartet.

observed between 5'anti-3'syn steps. This is because the syn conformation of the guanine places the G H 8 > 5/~ away f rom the I - I I ' ,H2 ' ,H2" ,H3 ' protons on the preceding 5' sugar. The only N O E s observed between nonexchangeable resonances for these steps are directly f rom sugar to

[10] D N A TRIPLEXES AND QUADRUPLEXES 245

m <

0

0 z

%

oo

O Z

246 DNA ANI) RNA STRUCTURE [10]

sugar and are usually weak and/or overlapped. Although one group has reported GnH8-Gn+IH8 NOEs at this step, TM we have found this to be in error, s°

4. 5'syn-3'syn: With both bases in the syn conformation, the GnH8 is in position to make some weak NOEs to the Gn+l sugar. The most notable of these is the G~H8 to Gn+IHI'.

The spectra of quadruplexes studied in our lab and elsewhere have essentially been made up of these subunits. A good example is provided by the sequential H8-HI' NOEs observed in spectra of d(G3T4G3) dimeric quadruplex TM (Fig. 11), which has GGG tracts with two different conforma- tion sequences, 5'syn-anti-anti 3' and 5'syn-syn-anti 3'. At the anti-anti steps, normal B-DNA such as GnHI'-Gn+IH8 connectivities were observed. At the syn-anti steps, reversed G~H8-Gn+IHI' NOEs were also observed. At the syn-syn steps only the reversed GnH8-Gn÷IHI' NOEs were observed.

Connecting Isolated Clusters. Often, in the absence of 1H-3~P hetero- nuclear data, clusters of residues (i.e., runs of nucleotides along a strand) cannot be identified unambiguously on the basis of NOEs among the clus- ter) TM That is, a sequential run of two or more guanines may be identified, but its placement in the strand will not be known. This occurs largely because 5'syn-anti 3' groups are often almost completely isolated from the rest of the molecule in terms of nonexchangeable short-range NOEs. However, occasional NOEs to loop resonances can help orient the clusters even though it is not known if they are short-range or long-range NOEs. Identification of clusters near loops may limit the assignment possibilities of other clusters.

Unusual NOEs can be useful to bridge the gaps left in the assignments. Weak NOEs can sometimes be observed directly between sugars. One such NOE is the Hl'n to H3'n+x connectivity. In NOESY spectra at long mixing time, with a good signal-to-noise ratio, these sequential connectivities are visible as very weak cross-peaks. Fig. 12 shows sequential GnHI'-G,+IH3' connectivities along the purine tracts of two related dimeric and unimolecu- lar quadruplexes. Note that the interresidue NOEs are weak. Many of the cross-peaks are obscured by overlapping intraresidue peaks, but the few resolved cross-peaks are useful confirmation of assignments otherwise made using long-range H8-H8 NOEs. However, because these cross-peaks are often unresolved or near the noise floor of the spectrum, especially for larger quadruplexes, these NOEs should only be used to confirm assignments previously determined.

One of the most useful techniques for the identification of isolated

s0 F. W. Smith, "Solution Structures of Guanine Quadruplex DNA Oligonucleotides." Ph.D. Thesis, University of California, Los Angeles (1995).

[ 10] D N A TRIPLEXES AND QUADRUPLEXES 247

i . . . . ;

I I

20

101

d(G1 G2 G3 T4T5 T6 T7 G8 G9 G10) d(G11G12G13T14T15T16T17G18G19G20)

H6,8

. . . . . . <- . . . . 016 I

12#_ . . . . . . <___0_ r . . . . 0 5 1 o 4 1, ~ < 1 5 ~ < _ . ~ . . ~

/~ 'I i 14 II I e ,~ , * 0 - < - -

I I

Ii I I I Ii I I I

_ _ _ ~ . . . . :E'_ _= _ r l ~

13 ;017 ~ 018 ~- . . . . . - < . . . . . . ~ - - - - 4

r~

¢D 't6

v 0

"¢6

"¢6

810 716 712 (ppm)

FIG. 11. This portion of a NOESY spectrum of d(G3TnG3) illustrates the NOEs observed at 5'syn-anti3' (tail-to-tail) and 5'anti-anti3' and 5'syn-syn3' (head-to-tail) steps. This dimer is not symmetric; therefore, the two sets of connectivities are indicated by solid and dashed lines. Normal B-DNA-type connectivities (horizontal then vertical) are observed for the anti- anti steps, G12-G13 and G9-G10, as well as out into the loops G3-T4 and G13-T14. At the syn-syn steps, G1-G2 and G18-G19, the connectivities are in the reverse direction (vertical then horizontal). At the syn-anti steps, G2-G3, G8-G9, G11-G12, and G19-G20, both connectiv- ities are seen (vertical then horizontal and horizontal then vertical). There are no anti-syn Gn-Gn+l steps in this molecule, but if there were, there would not be any sequential cross- peaks in this region. There is no connectivity at the anti-syn T7-G8 or T17-G18 steps at the ends of the loops (after Smith et a/.74).

clusters is to assume a regular B - D N A - t y p e backbone . Wi th this assump- t ion, the sequence of syn and anti residues is directly re la ted to the quar te t or ien ta t ion . This allows pred ic t ion of the sequence for o ther tracts, based on the ones previous ly identified. As an example , if an ant i -an t i re la t ionship has b e e n ident if ied for two sequent ia l residues, there should be ano the r ant i -ant i and two s y n - s y n groups f rom the o ther three tracts in a fold-back

248 DNA AND RNA STRUCTURE [10]

A bO "r"

12

0

HI'

lO

lO-ll

C0

I I

e ' -~' 0 9 • °

18-19--~b.~ v20 2~-1__~ "" l u 2

616 614 6:2 610 518 ppm FIG. 12. Portion of NOESY spectra of (A) dimeric and (B) unimolecular quadruplex,

showing cross-peaks between HI' and H3'. Direct sugar-sugar NOEs are clearly seen in the well-resolved and high signal-to-noise spectra of the symmetrical dimer quadruplex d(G4T4G3I) in (A), but are mostly obscured in the unimolecular version d(G4TUTUG4T4G4UUTTG3I) in (B). The few peaks that are visible are useful because of the lack of other NOEs at 5'anti- syn3' steps (from SmithS°).

quadruplex. Finally, it is worth noting that symmetry arguments can be used to eliminate many possibilities. 18'36

Conformation-Independent Assignments. The difficulty in using the pre- viously described NOEs for the assignment of quadruplex D N A is twofold. First, unlike B-DNA or triplex DNA, it is not known what NOEs should be expected. Second, for some steps, sequential NOEs are weak or nonexis- tent. In principle, the best solution to this problem is to use a 1H-31p heteronuclear COSY or HSQC to get assignments through the phosphate groups. Ideally, since each phosphorus atom is coupled to the Xn÷~H3' and XnH4',H5',H5", these experiments give unambiguous sequential assign- ments without regard to the glycosidic bond conformation. Unfortunately, it is most often the case that the relatively narrow spectral dispersion

[1 0] DNA TRIPLEXES AND QUADRUPLEXES 249

in the 31p chemical shifts and the proton H4',H5',H5" region precludes unambiguous assignment by this method. In our small symmetric molecules (12 nonequivalent residues) we were able to obtain through bond assign- ments for all residues, 36 but for the larger (28 residue) unimolecular quad- ruplexes we could only confirm a few assignments using these experiments. 8° However, these 1H-31P correlation experiments can still be quite useful, since it is often the case that in regions where structural predictions (and thus NOE-based assignments) are most questionable (i.e. in loops), the phosphates have unusual conformations and resonate at unusual chemical shifts, allowing these peaks to be resolved and assigned.

Use of Inosine and Deoxyuracil Substitutions. A particularly useful ap- proach is to use derivative sequences in which one or more guanines or thymines is substituted by inosine or deoxyuracil, respectively. 23"36 Most often a single inosine or deoxyuracil substitution will have little effect on the structure, but care must be taken to verify this. In the case of inosine substitutions, we have obtained the best results with an inosine at the 3' end. The replacement of the exchangeable amino group protons with a nonexchangeable H2 aromatic proton greatly assists in the assignment of the exchangeable resonances by providing an unambiguous starting point, as discussed in the next section. The H8 of inosine resonates downfield of guanosine H8 by 0.2-0.5 ppm.

Assignment of Exchangeable Resonances of Quadruplexes

Identification of Quartet. As discussed previously, the one-dimensional 1H spectra of the exchangeable imino and amino resonances provide the initial evidence for quadruplex formation and an indication of its symmetry. The quartet hydrogen bonding scheme can be verified by observation of imino-amino-H8 NOEs between pairs of guanines. If both hydrogen- bonded iminos and aminos are present, this eliminates a G-G hairpin- type conformation. However, even if sequential assignments have been unambiguously obtained along each strand of the quadruplex, this does not give any information on the relative orientation of bases within quartets (i.e., which guanine is hydrogen bonded to another guanine). This is because there are no through bond connectivities between the exchangeable reso- nances that define the base pairs and the assigned nonexchangeable reso- nances. The exception to this is in molecules where there is an appropriate inosine substitution. In this case, strong internucleotide imino-IH2-GH8 NOEs are observed (discussed later).

Unlike the case for DNA duplexes and triplexes, imino-imino NOEs cannot be used to make sequential connectivities along each strand. This

250 DNA AND RNA STRUCXURE [ 10]

is because the imino connectivities form a complicated network both within a quartet and between quartets, 36 and these connectivities can only be followed after the iminos have been assigned.

Molecularity of Quadruplex. It is not always obvious if the quadruplex formed is unimolecular, bimolecular, or tetramolecular. A good example is the thrombin-binding aptamer, d(GGTTGGTGTGGTFGG), which forms a unimolecular quadruplex with two G-quartets. 23,24,27 The spectra we obtained on this molecule were consistent with two G-quartets, but a priori we could not distinguish between a dimeric symmetrical quadruplex with four G-quartets or a unimolecular quadruplex with two G-quartets. 23 In this case, a mixing experiment with two different inosine-substituted derivatives provided evidence that the structure was unimolecular, as the NOESY spectra of the mixture was the sum of the individual NOESY spectra. If mixing experiments are done, care must be taken to heat the samples high enough to denature them and allow mixing to take place. Since quadruplexes are exceptionally stable, especially with K ÷, this can be difficult.

Topology Dependence. The assignment of the amino and imino reso- nances is complicated by the lack of intraresidue NOEs to the sugar or H8 resonances. The only neighboring protons are those of the hydrogen- bonded neighbors and the adjacent quartets. In the case of symmetric, all anti, tetrameric quadruplexes, each guanine is hydrogen bonded to a symmetric copy of itself; therefore, the assignment is straightforward. In the case of less than four-fold symmetry, however, it is not initially known which residues are hydrogen bonded. Depending on the arrangement of the individual G-tracts (determined by loop orientation), each imino proton has several possible nonexchangeable neighbors. This problem can be han- dled by making tentative assignments using each of the possible topologies. Assuming that the nonexchangeable H8 resonances have been previously assigned, a model can be used to predict which residue is H bonded to a known residue and use H8-amino and H8-imino NOEs to assign these exchangeable resonances around the quartets. Then the NOEs between the imino resonances can be checked for agreement with the model used to derive the assignments. NOEs assigned using incorrect topology models do not give a self-consistent description of the structure. Typically, this shows up as imino-imino distances that are inconsistent with the model structure. All possible folding models must be considered independently using the appropriate model-derived assignments.

This method can be confusing because each resonance has multiple possible assignments due to the number of possible loop arrangements. In the analysis, each NOE must be assigned for each model and then checked for in the model. A convention that we have found very useful has been

[ 10] DNA TRIPLEXES AND QUADRUPLEXES 251

3' 3' 5'

Head.to-Head "~ f "

Head-to-Tail J (Tail-to-Head) . J "

Tail-to j / -Tail

5' 3'

Fic. 13. Schematic representation of NOEs to be expected between imino protons in quartets of varying orientation. Bold circles indicate that the residue is in the syn conformation. Parallel and antiparallel strands are illustrated.

to keep the assignment of the exchangeable resonances tied to the known neighbor H8 (to which they show NOEs) assignment. For example, in a quartet made up of G1, G12, G17, and G28, an imino with an NOE to G17 H8 could be from G1, G12, or G28 depending on the model; our convention is to temporarily assign it as "G17-associated imino," reflecting the observed NOE to G17. Each model is constructed with the same imino- naming convention. It is then a simple matter to compare the NOEs, which now have a single assignment for all models, to each of the models in turn. "Associated" assignments are strictly a tool to simplify quadruplex assignment; it is important to use the actual assignments when calculating structures or reporting results.

In the structures we have studied, interquartet imino-H8 NOEs are rarely observed; however, Wang and Pate172 report both interquartet and intraquartet imino-H8 NOEs, with the former being stronger, in the unimo- lecular quadruplex formed from four repeats of the human telomere se- quence.

To follow the complex network of imino-imino connectivities in quad- ruplexes, we have made schematics of the imino protons to allow charting the observed NOEs. Fig. 13 is an idealized map showing the types of imino-imino NOEs we have observed. In an actual spectrum, many connec- tivities may be missing due to spectral overlap, noise, proximity to the diagonal, and symmetry effects. Typically, imino-imino NOEs are weak or very weak. Very weak NOEs are observed around each quartet, whereas the stronger NOEs are between imino protons on adjacent quartets. In our studies of quadruplexes, we have seen three different interquartet NOE

252 DNA AND RNA STRUCTURE []0]

patterns. In a 5'anti-syn 3' step, the imino protons are positioned directly above one another, parallel to the helix axis. Each imino proton has a single NOE to the adjacent sequential imino proton and no others to the adjacent quartet. In a 5'syn-3'anti step, a single NOE is observed, but it is between imino protons on adjacent strands. The only NOEs observed at this step are long-range NOEs. At an anti-anti or syn-syn step the twist of the helix places the imino protons in an offset position. Thus both sequential and long-range NOEs are observed.

Model-Independent Quartet Assignments. As discussed previously, mod- els can effectively be used to find a set of assignments that give an internally consistent set of NOEs. The model is used when interpreting through-space connectivities to find assignments. Prior knowledge of DNA structure is used to predict which protons should be close to each other. Then the NOEs so assigned are used to calculate another model. The danger exists that there may be more than one possible solution. This danger is especially great in quadruplex structures due to the many breaks in sequential connec- tivities. If all the cross-peak data are included, this method should not fail since the final refined structure will have to satisfy all of the NOE restraints. Mistakes can easily be made, however, if only some of the cross-peaks are assigned and used. For example, in the case of Oxy-l.5, our initial (unpublished) assignments were obtained using an edge-looped model. Nearly all nonexchangeable NOEs were satisfied, but the imino-imino NOEs gave completely inconsistent results. Once the correct model was used, all the NOEs made sense. Because these are such complicated systems, it is most desirable to have some independent model-free confirmation of assignments. In favorable cases, these can be obtained using inosine- substituted molecules. The inosine imino is unambiguously identified by its strong I imino-IH2 NOE, and the intraquartet IH2-GH8 NOE unambigu- ously identifies which G is the hydrogen-bonded neighbor (Fig. 2). Once two assigned nucleotides are unambiguously located next to each other in a quartet, the number of possible topologies that fit these assignments is greatly reduced.

Structure Determination

Qualitative Analysis of Triplex and Quadruplex Structure

The analysis of the NMR spectra discussed previously also provides qualitative information on the folded structures formed from purine-rich oligonucleotides. As discussed previously, the imino proton spectra provide evidence for hydrogen-bonded structures, and the NOEs from the ex-

[10] DNA TRIPLEXES AND QUADRUPLEXES 253

changeable imino and amino resonances usually provide the information needed to determine the hydrogen bonding of a given base pair. These methods have been used to determine the structures of the base triplets in both the pyrimidine 39-41'47 and purine 52-54 motif triplexes, as well as the structure of alternative base triplets incorporated into triplexes 81-83 (e.g., the G-T-A triplet65's4). Syn versus anti base orientations are also usually readily determined from the intensity of the base H6,H8-HI ' NOEs. Both NOE intensities and coupling constants can be used to determine the deoxyribose conformation. The latter method was used to show unambiguously that triplexes did not have all C3'-endo sugar puckers as previously believed, based on the early fiber diffraction work. as're's5 J values for the deoxyribose protons can be obtained from simulation of the phase sensitive COSY cross-peak patterns; this can be done either manually, visually matching the simulated spectra to the experimental spectra, or by an automated iterative method. Once J values are obtained, they can be used with the program P S E U R O T 86 to determine the sugar conformation.

Once the base-pairing schemes, glycosidic torsion angles, sugar puckers, and general helix type are determined, a model structure can usually readily be constructed and refined using the distance and dihedral angle restraints derived from N O E and J coupling data, respectively.

Approaches to Determining Complete Three-Dimensional Structures

Structures of triplexes and quadruplexes can potentially be determined to higher resolution than D N A duplexes because of the additional con- straints provided by the greater number of intrastrand N O E connectivities and, in the case of quadruplexes, by the very slowly exchanging imino and amino resonances that thus give rise to NOEs that can be quantitated much more accurately than is the usual case. The determination of the molecular structures of DNAs with the precision of atomic resolution has some unique problems not encountered in the determination of protein structure. Experi- mental NOEs alone, if sufficient in number, are enough to define all the atomic positions in a given protein, whereas the number of experimental

81 R. F. Macaya, D. E. Gilbert, S. Malek, J. Sinsheimer, and J. Feigon, Science 254, 270 (1991).

82 K. M. Koshlap, P. Gillespie, P. B. Dervan, and J. Feigon, J. Am. Chem. Soc. 115, 7980 (1993).

83 I. Radhakrishnan and D. J. Patel, J. MoL BioL 241, 600 (1994). 84 E. Wang, S. Malek, and J. Feigon, Biochemistry 31, 4838 (1992). 85 R. F. Macaya, P. Schultze, and J. Feigon, J. Am. Chem. Soc. 114, 781 (1992). 86 F. A. A. M. de Leeuw and C. Altona, J. Comp. Chem. 4, 428 (1983).

254 DNA AND RNA STRUCTURE [10]

1H NOEs alone, no matter how abundant, can define the atomic positions of the base and deoxyribose portions of any given oligonucleotide with only marginal precision and cannot define the atomic positions of the phos- phodiester backbone. This is because of the directional distribution of NOE contacts. In the core of a protein, many 1H-1H contacts are observed in all directions, not necessarily to sequentially neighboring residues. In DNA, however, typically the great majority of cross-peaks are either intraresidue or between immediately neighboring residues along the strand. Early attempts to calculate three-dimensional structures of B- DNA oligonucleotides were not very successful; the structures were often underwound and had unusual groove widths and bends that have not borne the test of time. 87 In general, the NOE constraints used were insufficient to give a structure that was any better than a model structure based on a B-DNA helix, and deviations from this could easily be overinterpreted to be meaningful.

However, the situation has improved considerably in the last few years, especially for folded structures such as intramolecular triplexes and quad- ruplexes, which provide additional restraints for the structure calculations. Improvements in the analysis of NOE data, addition of dihedral bond angle constraints of the deoxyribose rings derived from analysis of 3J(H,H) coupling constants, and improvements in computational analysis of the data have all contributed to this progress. Furthermore, experimental informa- tion provided by 3J(H,P) couplings, which can be measured from (1H, 31p)

heteroCOSY, is beginning to be used in determining the torsional angles of the phosphodiester backbone. However, one must still be cautious when evaluating reported structures; as with crystallography, their accuracy and precision depends on the quality of the input data, the quality of the data analysis, and the method of structure calculation.

In general, structures are refined by simulated annealing and energy minimization of initial starting structures) 8 Although initial starting struc- tures can be models, such as a triplex generated from A- and/or B-DNA helix coordinates derived from fiber diffraction, the preferred method is to use metric matrix distance geometry to avoid introducing any possible bias into the final structures from the initial standard conformations. Re- finement should of course include as many restraints as possible, including distance restraints from both exchangeable and nonexchangeable protons and dihedral bond angle constraints from analysis of coupling constants.

87 D. J. Patel, L. Shapiro, and D. Hare, Annu. Rev. Biophys. Biophys. Chem. 16, 423 (1987).

8s A. T. Brtinger, "X-PLOR (Version 3.1) Manual." Yale University Press, New Haven and London, 1992.

[ 10] DNA TRIPLEXES AND OUADRUPLEXES 255

In practice, it turns out to be necessary to include explicitly standard hydro- gen bond constraints for the base pairs in addition to the directly observed H-H contacts. To define sugar puckers, the coupling constants between sugars can be analyzed using a two-state ring flip model with the program PSEUROT, 86 which fits the experimental coupling constants with two pseu- dorotation parameters and the population ratio between them. The in- dividual values for the dihedral angles corresponding to the major con- former are used as restraints in our refinements. Two out of the five defined torsion angles are sufficient to define the complete five carbon ring con- formation.

In the case of symmetric dimers or tetramers, the special problem arises that intramonomer NOE contacts cannot be distinguished from intermono- mer contacts. 89 Such ambiguities can be resolved by testing constraint viola- tions on a raw starting model in standard conformation under the initial assumption of all intramonomer contacts. Significantly violated constraints are then switched to their symmetry related intermonomer equivalents to see if they are better fulfilled this way. For the actual refinement, the complete set of contacts is duplicated, substituting the symmetry equivalent residue numbers in the second set.

Once the energy-refined structures are obtained, a final NOE-based refinement step can be done. This involves calculation of the full relaxation matrix or direct NOE refinement, using data from at least two different NOESY mixing times. This last step usually does not result in large changes in the atom positions, but it does significantly improve the agreement between observed and calculated NOE intensities.

Acknowledgments

This work was supported by grants from NIH (R01 GM37254 and R01 GM48123) and NSF (DMB 89-58280) with matching funds from AmGen Inc., Monsanto Co., and Sterling Winthrop Drug Inc. to J. Feigon. The authors thank Peter Schultze for helpful comments on the manuscript.

89 p. Schultze, F. W. Smith, and J. Feigon, Structure 2, 221 (1994).