doi:10.1006/jmbi.2000.3907 available online at http://www.idealibrary.com on J. Mol. Biol. (2000) 300, 551±561

A New Crystal Form for the DodecamerC-G-C-G-A-A-T-T-C-G-C-G: Symmetry Effects onSequence-dependent DNA Structure

Eric Johansson, Gary Parkinson and Stephen Neidle*

CRC Biomolecular StructureUnit, Chester BeattyLaboratories, Institute of CancerResearch, Fulham RoadLondon, SW3 6JB, UK

E-mail address of the corresponds.neidle@icr.ac.uk

0022-2836/00/030551±11 $35.00/0

The dodecanucleotide d(CGCGAATTCGCG) has been crystallised in thespace group P3212, representing a new crystal form for this sequence.The structure has been solved by molecular replacement and re®ned at1.8 AÊ resolution. The present structure contrasts with previous ones forthis sequence since it is situated on a crystallographic 2-fold axis, and thecrystal symmetry re¯ects the palindromic nature of this sequence. Somefeatures accord with previous observations, notably that the minorgroove is hydrated with a continuous spine of solvent. There is no evi-dence of alkali metal ions within this spine. The minor groove retains itsnarrow width, although it is now symmetric and extends over the A/Ttract. Various base and base-pair morphological parameters have beenexamined. Their values do not show signi®cant correlations with earlierreports, suggesting that crystal packing effects on them are more domi-nant than has been hitherto realised.

# 2000 Academic Press

Keywords: crystal packing; dodecanucleotide; structure; symmetry;base morphology

*Corresponding author

Introduction

The DNA double helix is inherently symmetric.The structural models of DNA derived from®bre diffraction re¯ect this symmetry, with pre-cisely regular helices, and crystallographicallyidentical conformations for each nucleotide(Chandrasekaran & Arnott, 1996). So for a localstretch of self-complementary sequence, an exact2-fold relationship (with a local 2-fold axis in theplane of and mid-way between the central base-pairs) would be expected between the two anti-parallel strands. By contrast, crystal structures ofnative oligonucleotides, even with self-complemen-tary sequences, often do not demonstrate suchsymmetry.

The determination almost 20 years ago, of thecrystal structure of the dodecamer sequenced(CGCGAATTCGCG) (Drew & Dickerson, 1981),the so-called Dickerson-Drew dodecamer, was alandmark in the ®eld of DNA structural studies.The structure analysis unequivocally provided anab initio demonstration of the overall correctness ofthe original double helical model of B-DNA. It alsoprovided evidence for a high degree of local varia-

ing author:

bility throughout the sequence (Berman, 1997;Chiu et al., 1999; Grzeskowiak, 1996). This struc-ture continues to be the subject of intense exper-imental and theoretical study, not least as aparadigm for evaluating molecular dynamics simu-lation methodologies (see, for example, Duan et al.,1997; Feig & Pettitt, 1999; Kosztin et al., 1999;Young et al., 1997). In particular, the nature andstructure of the solvent in the minor groove hascontinued to excite considerable interest. The orig-inal, medium-resolution observations (Drew &Dickerson, 1981) of a continuous spine ofhydration in the A/T narrow minor groove are inaccord with recent high-resolution studies(SolerLopez et al., 1999; Tereshko et al., 1999a,b).However, re-assignments of at least some watermolecules in the spine as the monovalent alkalimetal ions sodium or potassium (Shui et al.,1998a,b) have been controversial, and have notbeen reported in a cross-linked structure ofd(CGCGAATTCGCG) at 1.3 AÊ resolution (Chiuet al., 1999). An extended hydration spine ofpure water molecules has also been reported inthe 1.15 AÊ structure of the sequenced(GGCCAATTGG) (Vlieghe et al., 1999). NMR stu-dies suggest (Hud et al., 1999) that ammonium ionsmay occupy the same positions as monovalentmetal ions, and are consistent with the reports of

# 2000 Academic Press

Figure 1 (legend opposite)

552 Structure of C-G-C-G-A-A-T-T-C-G-C-G

partially occupied sodium ions (Shui et al.,1998a,b).

The Dickerson-Drew sequence was originallycrystallised in the orthorhombic space groupP212121 with the asymmetric unit being bothstrands of the duplex, and with no exact sym-metry, crystallographic or otherwise, relating thetwo strands. There is a surprising degree of asym-metry in both major and minor groove widthsalong the sequence, as well as in the base and heli-cal morphology parameters. A continuing questionis whether this lack of symmetry is a consequenceof crystal packing, the so-called ``tyranny of the lat-tice'' (Dickerson et al., 1994). In turn this raises thequestion of whether and to what extent thesequence-dependent features observed in this andother DNA structures are real or artifacts resultingfrom crystal packing. The recent observation (Liuet al., 1998; Liu & Subirana, 1999) of a calcium saltof the Dickerson-Drew dodecamer in the rhombo-hedral space group R3, showed a packing motifdistinct from that in the P212121 lattice, with thecentral ten base-pairs stacked in columns in thecrystal, and the end nucleotides in extra-helicalconformations. The asymmetric unit in this struc-ture is again the two strands, although here an

increased degree of symmetry was observed for anumber of the morphological descriptors of thestructure.

Here, we report a new crystal form for theDickerson-Drew dodecamer d(CGCGAATTCGCG),in another rhombohedral space group, but onewhich incorporates exact crystallographic sym-metry for the DNA. Unexpectedly, the impositionof symmetry has a signi®cant effect on geometricparameters. These are discussed in comparisonwith those observed in the Dickerson-Drew dode-camer and other crystal structures containing thissequence. The nature of the structured minor-groove solvent is also discussed.

Results and Discussion

Overall features

The single strand of the asymmetric unit sitsastride a crystallographic 2-fold axis, which passesthrough the centre of the two central A �T base-pairs (Figure 1(a)). The central ten base-pairednucleotides form a standard B-form anti-parallelduplex, with an average helical twist between suc-cessive base-pairs of 33.9 �, as determined by

Figure 1. (a) The structure of the duplex with the two symmetry-equivalent strands, showing the nucleotide label-ling used. (b) A view of the packing in the crystal lattice, looking along the diagonal between a and b axes. (c) Look-ing down the c axis.

Structure of C-G-C-G-A-A-T-T-C-G-C-G 553

CURVES (Lavery & Sklenar, 1989). The helix thushas 10.6 residues per turn. Successive duplexesstack on one another along the crystallographicc-axis (Figure 1(b)). The terminal 50-cytosine(Figure 2(a)) and 30-guanine nucleotides are both in

extra-helical conformations. The former lies in thechannels between adjacent helices (Figure 1(b)),with hydrogen bonding involving N3 and N4 tophosphate oxygen atoms of the adjacent helix(Figure 2(b)). The adjacent crystallographic 2-fold

Figure 2 (legend opposite)

554 Structure of C-G-C-G-A-A-T-T-C-G-C-G

relating one helix to the next stacked one, places asecond symmetry-related cytosine in proximity sothat it also hydrogen bonds to the same pair ofphosphate groups. The unpaired guanosine lies inthe minor groove of an adjacent duplex in thecrystal lattice (Figure 2(c)), and hydrogen bonds to

the terminal base-paired guanosine via a pair ofN2 � � �N3 hydrogen bonds. All four of the A �Tbase-pairs in the duplex have complete base-pairhydrogen bonding; the N1 � � �N3 distances areboth 2.9 AÊ and the two N6 � � �O4 distances areboth 3.1 AÊ .

Figure 2. Intermolecular interactions of the extra-helical bases (a) showing the electron density from a 2Fo ÿ Fc

map, (b) the intermolecular hydrogen bonds involving the two symmetrically-related 50-end cytosine nucleotides,(c) the 30-end guanosine.

Structure of C-G-C-G-A-A-T-T-C-G-C-G 555

Groove width

The variation in minor-groove width (Figure 3)shows narrowing in the A/T tract, in overallaccord with all other structures of this sequence.However the width pro®le is distinct from thatobserved in other dodecanucleotide crystal struc-tures. In these, the characteristic sequence-depen-dent narrowing in the A/T region is markedlyasymmetric, with the valley de®ning the narrowregion at its lowest point around base-pair A5 �T20or A6 �T19. By contrast, in the present structure thevalley is much broader and symmetric, re¯ectingthe exact symmetry of the duplex. The very deepvalley ¯oor now extends from base-pair A5 �T20 toT8 �A17, i.e. covers all four A �T base-pairs. The

Figure 3. Plot of minor groove widths, calculated with CUcrystal structures of the sequence d(CGCGAATTCGCG). Can

minor groove narrowing at both the 30 and 50-endsof the A/T tract contrasts with the asymmetricgroove in the P212121 dodecamers, so that at itsextreme there are differences of >1 AÊ in widths.There is also a small hump in the ¯oor of thegroove, of ca 0.3 AÊ , centred between the centralA6 �T19 and T7 �A18 base-pairs, which is unique tothis structure.

Base-pair morphologies

Figure 4(a)-(c) compares values for several heli-cal parameters (calculated with CURVES) in thepresent structure with those from four otherd(CGCGAATTCGCG) crystal structures. All show

RVES for C40 � � �C40 inter-strand separations, for severalonical B-DNA has a calculated groove width of 4.0 AÊ .

Figure 4. Plots of (a) propellertwists, (b) helical twists and (c) rollin several crystal structures ofd(CGCGAATTCGCG), calculatedwith CURVES.

556 Structure of C-G-C-G-A-A-T-T-C-G-C-G

variations with sequence, although the extent (andoccasionally the nature) of the variations frequentlydiffer in detail from one structure to another. Careis needed when making comparisons in view ofdifferences in accuracy and precision between thestructures, as well as in re®nement protocols andparameters, which cannot be readily quanti®ed.However some likely signi®cant differences areapparent: (i) propeller twists for the central sixbase-pairs vary between 13 and 19 �, consistentlyless than in the original Dickerson-Drew structure,and close to those in the potassium-bound form(Shui et al., 1998b). Overall, there appears to be a

small, though consistent difference between thetwist of the four central A �T base-pairs comparedto the neighbouring G �C base-pairs.

(ii) There appears to be a consistent pattern ofhelical twist values, at least for the central six base-pair steps. The values systematically decreasetowards the centre of the duplex, with a minimumof 31 � at the central ApT step. There is no clearrelationship between twist and groove width. Thedifferences between the present symmetricstructure and the other asymmetric ones are moststrikingly seen when comparing the two CpGsteps C3/G4 and C9/G10 (Figure 4(b)), with a

Structure of C-G-C-G-A-A-T-T-C-G-C-G 557

signi®cant asymmetry in values in the asymmetricstructures. Since there are differences between thevarious structures (with the symmetric structurehaving the lowest value of all), it is not possible tojudge what is the ``correct'' value for this step fromthe asymmetric structures alone.

(iii) Base-pair roll values (Figure 4(c)) for theasymmetric dodecamer structures show consist-ently negative values (i.e. with steps openedtowards the major groove). By contrast, the sym-metric structure has roll values close to zero fortwo out of the ®ve central base-pair steps, so thatthe central segment of the structure is slightly benttowards the minor groove. The positive roll at thesteps immediately 50 and 30 of this segment alsoopen towards the minor groove. The overall effectis that the whole decamer sequence is bent, by ca7 �. It is noteworthy that the central base-pair stephas a small positive roll by contrast with the nega-tive roll of its neighbour on either side. It is tempt-ing to speculate that this is related to the increasein minor groove width at this point (see above).

The rmsd between the central ten base-pairs ofthis structure and those of the Dickerson-Drewdodecamer is 1.1 AÊ (Figure 5). This only reducesby 0.2 AÊ , to a rmsd of 0.9 AÊ when comparing thecentral six base-pairs, suggesting that the differ-

ences between the structures are not con®ned tothe extremities (which are involved in intermolecu-lar hydrogen bonding), but are distributed ratherevenly throughout the structure.

Hydration and counter ions

A well-organised array of a total of nine watermolecules have been located in the A/T tract ofthe minor groove of this structure (Figure 6(a) and(b)). The arrangement is 2-fold symmetric aboutthe water molecule bridging between the two cen-tral thymine bases. There is thus a complete spineof hydration in the A/T region, which is made upof two layers of solvent molecules. The four sol-vent molecules in the ®rst shell are hydrogenbonded to purine N3 and pyrimidine O2 atomsand are bridged by ®ve second shell water mol-ecules. As the groove widens toward the C �Gbase-pairs, the groove is able to accommodatemore water molecules and the characteristic spineorganisation is lost, though a linear network ofwater molecules can still be observed, the last onehydrogen-bonding to the phosphate oxygen atomof cyt23. Some extra ®rst-shell water molecules areseen (W29), which bridge to O30 backbone atoms.All three ®rst-shell solvent molecules in the A/T

Figure 5. Plot of the presentstructure, overlaid onto the Dicker-son-Drew structure.

Figure 6 (legend opposite)

558 Structure of C-G-C-G-A-A-T-T-C-G-C-G

tract have weak hydrogen-bonds to two O40atoms, one on each strand. The ®rst-shell solventmolecules at the end of the spine are in G/Cregions, where the groove is wider. Thus they canonly interact with the O40 on one strand (albeitalso weakly), and cannot quite reach the one onthe other strand. The A/T tract continuous minorgroove hydration spine has all other hydrogenbond distances within standard accepted distances,given the likely margin of error. Recently reported(SolerLopez et al., 1999; Tereshko et al., 1999a,b)third and fourth shells of hydration were notobserved in this structure, probably on account ofits lower resolution.

The nature of the spine of hydration has recentlybeen the focus of considerable debate, and in par-ticular the extent to which the water molecules aredisordered or partially occupancy monovalentpotassium or sodium ions continues to be contro-versial (Chiu et al., 1999). Molecular dynamicssimulations have previously indicated that sodiumions can ``invade'' the minor groove (Young et al.,1997) and NMR experiments have shown thatanother monovalent ion, ammonium, can also dothis (Hud et al., 1999). There are considerable prac-tical problems when attempting to differentiatebetween water molecules and putative counter-ions from X-ray data, even from high-resolutionstructures. This is especially challenging with ionshaving the same or similar number of electrons asa water oxygen atom. Partial occupancies, wherethe site is occupied fractionally by water and by

the ion, are particularly hard to differentiate. Onthe other hand, objective crystallographic re®ne-ment techniques on a structure with data to betterthan 2 AÊ resolution, should be able to differentiatebetween eight electrons of a water oxygen mol-ecule, and 19 of potassium, although not betweenoxygen and sodium (11 electrons).

In this structure, we are not able to assign anyatoms as monovalent ions purely on the basis oftheir electron density. All of the ®rst-shell solventmolecules in the spine are hydrogen-bond tetra-coordinated. This geometry is that exhibited bywater molecules. The weak interactions with O40atoms have also been observed previously (Chiuet al., 1999). However, since the distances involvedare close to van der Waals separations we suggestthat they are more a consequence of the high den-sity of atoms in the minor groove than true hydro-gen bonds. The evidence from this structure atleast, is not indicative of a low level of occupancyby monovalent ions (sodium or potassium). Wetherefore suggest that the minor groove structurefor the solvent spine is most consistent with the®rst and second shell of the spine consisting ofwater molecules (although we cannot discountthem being ammonium ions). The question of whyhexa-coordinated solvent molecules are observedin some crystal structures and not others may berelated to differences in groove width, which canreadily occur within the overall constraints of thecrystal packing in these structures. In the case ofthe present structure, the extended and symmetri-

Figure 6. The minor groovehydration, (a) viewed with solventmolecules shown as red spheres,and (b) in schematic form, high-lighting the hydrogen-bondingdistances involving water mol-ecules.

Structure of C-G-C-G-A-A-T-T-C-G-C-G 559

cal minor groove results in a solvent spine whicheffectively ®lls the central six base-pairs of thesequence by maximising hydrogen bond inter-actions with DNA.

Conclusions

The existence of a new crystal form of the DNAsequence d(CGCGAATTCGCG), with exact 2-foldsymmetry, has enabled the crystallographic anal-ysis of this well-studied sequence to escape fromthe constraints of the previous orthorhombic crys-tal lattice. We now ®nd that features such as vari-ations in minor groove width, are necessarilyprecisely symmetric in the present structure, andhighlight their marked asymmetry in the earlierstructures. Since the sequence is palindromic, wesuggest that packing forces do play a signi®cantrole in deforming DNA in the crystal. In a similarway the individual values of many helicoidal par-

ameters are distinct from those in the asymmetricstructures.

We suggest that the geometry of the presentstructure may be a more faithful representation ofthe inherent symmetry in the central eight to tenbase-pairs of the self-complementary sequenced(CGCGAATTCGCG), and thus of its sequence-dependent features. As such, it represents morethan yet another set of local minima in the multi-dimensional energy surface of DNA, immobilisedin a crystal structure. The structural features foundhere do demonstrate the plasticity of DNA, andsuggest to us that we can only be con®dent, on thebasis of current data, about some sequence-depen-dent features of DNA structure (the narrow minorgroove and, in particular, the remarkably constantspine of hydration in A/T regions). We are veryfar from being able to develop a more extensiveexperimentally based picture of DNA sequence-structure relationships, in view of the paucity ofdiverse sequences in the crystal-structure database,

560 Structure of C-G-C-G-A-A-T-T-C-G-C-G

and just as important, the lack of structures withmeaningful estimates of uncertainty in geometricparameters.

Materials and Methods

Crystallisation and data collection

The DNA sequence d(CGCGAATTCGCG), puri®edby HPLC, was purchased from Oswel. The crystals weregrown using the hanging drop method at 284 K, as col-ourless, cubic crystals. The crystal used for data collec-tion was obtained from a drop containing 1 ml of 4 mMsingle-stranded DNA dissolved in sodium cacodylatebuffer (pH 6.8), 1 ml of 2 mM of a minor-groove bindingdrug (in the same buffer) and 2 ml of Hampton Screeningkit reagent number 12, consisting of 10 % 2-methyl-pen-tane-2,4-diol (MPD), 12 mM spermine tetrahydro-chloride, 80 mM KCl and 20 mM BaCl2. This drop wasequilibrated against 700 ml of 30 % MPD solution and thecrystal grew in a period of three weeks. The crystal was¯ash-frozen in liquid nitrogen.

Data was collected at the Daresbury Synchrotronsource, station 9.5, using a MAR 30 cm area detectorwith a double silica monochromator, at a wavelength of1.0 AÊ . The crystal-to-detector distance was 232.47 mmand the oscillation range was 2.5 �, leading to data collec-tion at a maximum resolution of 1.75 AÊ . Forty frameswere collected and the crystal did not appear to decayduring collection. Data reduction and scaling were car-ried out using DENZO and SCALEPACK (Otwinowski& Minor, 1993). Statistics for the data collection and pro-cessing are given in Table 1.

Table 1. Data collection and re®nement statistics

DNA sequenced(CGCGAATTCGCG) 2

DNA in asymmetric unit CGCGAATTCGCGSpace group P3212Number of reflections 3343Number of unique reflections 3268Number of reflections used in refinement 2997Number of parameters refined 1163Restraints 1359Number of DNA atoms 243Number of full occupancy waters 47Maximum resolution of observedreflections (AÊ ) 1.7Maximum resolution of refinement (AÊ ) 1.8R-factor (%) 22.4Rfree (%) 30.1

Resolution range (AÊ ) Completeness (%)30.00-3.77 94.63.77-2.99 96.42.99-2.61 95.42.61-2.37 92.92.37-2.20 90.12.20-2.07 86.02.07-1.97 81.71.97-1.89 71.01.89-1.81 64.91.81-1.75 10.1All reflections 79.1

Structure solution and refinement

Unit cell dimensions were determined to be:a � b � 26.39, c � 98.78 AÊ , with successful structure anal-ysis in the space-group P3212. The 2-fold symmetryelements of the space-group, and the fact that the centreof the single strand must sit astride one of these, wereexploited in the structure solution.

The calculated volume of the unit cell suggested thereare six duplexes in the cell, which contains 12 asym-metric units. Therefore each duplex must be astride acrystallographic 2-fold axis. This axis was presumed tobe coincident with the inherent 2-fold symmetry of theDNA, thus only necessitating rotation searches in onedimension. The initial model used was one strand fromthe Dickerson-Drew structure. The model was manuallypositioned onto the 2-fold axis. A rotation search wascarried out, using XPLOR version 3.1 (BruÈ nger et al.,1987) in the c-axis direction, which gave a single sol-ution, with a correlation coef®cient of 2.0.

Translation searches were then attempted. Repeatedefforts on several 2-fold axes in different trigonal space-groups proved unsuccessful when the search model wasthe dodecamer single-strand. Truncation of the modelwith deletion of a nucleotide at each end, creating a dec-amer and then searching the 2-fold at a height of c � 1/6in space group P3212, resulted in a solution with anR-value of 42 %, suggestive of a correct solution. Examin-ation of a 2Fo ÿ Fc electron-density map, revealed thatthe DNA ®tted well into the density, and that the term-inal 30-end guanine seemed to be folded back into theminor groove of the adjacent DNA. However, no densitycould be seen for the missing 50-end cytosine. Severalcycles of rigid-body re®nement were carried out at aresolution range of 8-3.0 AÊ , bringing the R-value downto 38.9 %. Density maps were generated by XPLOR andviewed using the CHAIN package.

XPLOR (version 3.1) (BruÈ nger et al., 1987) was used tocarry out positional re®nement in different groups andby increasing the resolution, until the ®nal individualpositional re®nement included all data down to 1.8 AÊ

(3346 re¯ections). The R-value at this point was 37.2 %.Temperature factor re®nement and positional re®nementthen brought the R-value down further to 34.4 %. Rfree

was 35.4 %.At this point, re®nement was continued with the

SHELX-97 package (Sheldrick, 1997). Solvent moleculeswere located in Fourier density maps, and accepted onthe basis of standard distance and B-factor criteria. Noordered barium ions were found in the structure. Notrace of the drug molecule used in the crystallisation wasfound in the maps.

The terminal 50-end cytosine was located at this pointand was also found to have adopted an extra-helical con-formation, reaching into the channels between adjacentduplex strands. The ®nal R and Rfree values were 22.4 %and 30.1 %, respectively. The symmetry equivalentstrand was generated by applying the ÿx � y, y, z � 2/3symmetry operator, together with a �1 translation in thex-direction. DNA helicoidal parameters were generatedwith the program CURVES (Lavery & Sklenar, 1989).

Co-ordinates accession number

Final co-ordinates and structure factors have beendeposited with the Nucleic Acid Database as entry No.BD0032.

Structure of C-G-C-G-A-A-T-T-C-G-C-G 561

Acknowledgements

We thank the Cancer Research Campaign for support(Programme Grant SP1384 to S.N.), and The Institute ofCancer Research for a Studentship (to E. Johansson).

References

Berman, H. M. (1997). Crystal studies of B-DNA: theanswers and the questions. Biopolymers, 44, 23-44.

BruÈ nger, A. T., Kuriyan, J. & Karplus, M. (1987). Crys-tallographic R-factor re®nement by moleculardynamics. Science, 235, 458-460.

Chandrasekaran, R. & Arnott, S. (1996). The structure ofB-DNA in oriented ®bers. J. Biomol. Struct. Dynam.13, 1015-1027.

Chiu, T., Grzeskowiak, K. & Dickerson, R. (1999).Absence of minor groove monovalent cations in thecrosslinked dodecamer C-G-C-G-A-A-T-T-C-G-C-G.J. Mol. Biol. 292, 589-608.

Dickerson, R., Goodsell, D. & Neidle, S. (1994). Thetyranny of the lattice. Proc. Natl Acad. Sci. USA,91, 3579-3583.

Drew, H. & Dickerson, R. E. (1981). Structure of aB-DNA dodecamer. 3. Geometry of hydration.J. Mol. Biol. 151, 535-556.

Duan, Y., Wilkosz, P., Crowley, M. & Rosenberg, J. M.(1997). Molecular dynamics simulation study ofDNA dodecamer d(CGCGAATTCGCG) in solution:conformation and hydration. J. Mol. Biol. 272, 553-572.

Feig, M. & Pettitt, B. M. (1999). Modeling high-resol-ution hydration patterns in correlation with DNAsequence and conformation. J. Mol. Biol. 286, 1075-1095.

Grzeskowiak, K. (1996). Sequence-dependent structuralvariation in B-DNA. Chem. Biol. 3, 785-790.

Hud, N. V., Sklenar, V. & Feigon, J. (1999). Localizationof ammonium lone in the minor groove of DNAduplexes in solution and the origin of DNA A-tractbending. J. Mol. Biol. 286, 651-660.

Kosztin, D., Gumport, R. I. & Schulten, K. (1999). Prob-ing the role of structural water in a duplex oligo-deoxyribonucleotide containing a water-mimickingbase analog. Nucl. Acids Res. 27, 3550-3556.

Lavery, R. & Sklenar, H. (1989). De®ning the structureof irregular nucleic acids-conventions and prin-ciples. J. Biomol. Struct. Dynam. 6, 655-667.

Liu, J. & Subirana, J. A. (1999). Structure ofd(CGCGAATTCGCG) in the presence of Ca2� ions.J. Biol. Chem. 274, 24749-24752.

Liu, J., Malinina, L., HuynhDinh, T. & Subirana, J. A.(1998). The structure of the most studied DNA frag-ment changes under the in¯uence of ions: a newpacking of d(CGCGAATTCGCG). FEBS Letters, 438,211-214.

Otwinowski, Z. & Minor, W. (1993). In Data Collectionand Processing (Sawyer, L., Isaacs, N. W. & Bailey,S., eds), SERC Daresbury Laboratory, Warrington,UK.

Sheldrick, G. M. (1997). SHELX-97, a crystallographicre®nement program, The University of GoÈttingen,GoÈ ttingen, Germany.

Shui, X. Q., McFailIsom, L., Hu, G. G. & Williams, L. D.(1998a). The B-DNA dodecamer at high resolutionreveals a spine of water on sodium. Biochemistry,37, 8341-8355.

Shui, X. Q., Sines, C. C., McFailIsom, L., VanDerveer, D.& Williams, L. D. (1998b). Structure of the potass-ium form of CGCGAATTCGCG: DNA deformationby electrostatic collapse around inorganic cations.Biochemistry, 37, 16877-16887.

SolerLopez, M., Malinina, L., Liu, J., HuynhDinh, T. &Subirana, J. A. (1999). Water and ions in a highresolution structure of B-DNA. J. Biol. Chem. 274,23683-23686.

Tereshko, V., Minasov, G. & Egli, M. (1999a). The Dick-erson-Drew B-DNA dodecamer revisited at atomicresolution. J. Am. Chem. Soc. 121, 6970.

Tereshko, V., Minasov, G. & Egli, M. (1999b). A``hydrat-ion'' spine in a B-DNA minor groove. J. Am.Chem. Soc. 121, 6970.

Vlieghe, D., Turkenburg, J. P. & VanMeervelt, L. (1999).B-DNA at atomic resolution reveals extendedhydration patterns. Acta Crystallog. sect. D, 55, 1495-1502.

Young, M. A., Jayaram, B. & Beveridge, D. L. (1997).Intrusion of counterions into the spine of hydrationin the minor groove of B-DNA: fractional occu-pancy of electronegative pockets. J. Am. Chem. Soc.119, 59-69.

Edited by A. Klug

(Received 6 April 2000; received in revised form 25 May 2000; accepted 29 May 2000)