A continuous transition from A-DNA to B-DNA in the 1:1 complex between nogalamycin and the hexamer dCCCGGG.

The antibiotic nogalamycin, a drug with high specificity for TG and CG steps in double-stranded DNA, has been crystallized as a 1:1 complex with the hexamer d(CCCGGG). The antibiotic is inserted at the central CG step of the duplex, with the two sugars oriented in the same direction and with strong interactions with the DNA within the grooves. The amino-glucose residue makes an integral part of a well defined major groove hydration network with van der Waals contacts and several strong hydrogen bonds to the duplex. The nogalose residue resides in the minor groove, making primarily van der Waals contacts. The single site allows an accurate molecular description of the intercalation, without perturbations from end effects observed previously. The local unwinding induced by nogalamycin is completely relaxed 2 base pairs away from the intercalation site. The two strands of the DNA show a continuous deformation from the A to the B form: 1) the cytosines toward the 5' end of the nogalomycin site in each strand have c3'-endo conformations while 5 guanosines toward the 3' ends have c2'-endo conformations; 2) within each strand, the phosphate-phosphate distances increase in a continuous manner from 5.7 A (A-form) to 7.1 A (B-form).

The anthracycline antibiotic nogalamycin ( Fig. 1) is biologically important for its anti-tumor activities and its ability to inhibit DNA-directed RNA synthesis in vivo (1). X-ray studies have shown how it binds to double-stranded DNA by intercalation at TpG, CpG, and CpA steps (2,3). The anthracycline ring is parallel to the base pairs with its long axis aligned perpendicular to the base pairs and with the hydrophobic nogalose and amino-glucose sugar moieties lying in the minor and major grooves, respectively. DNA footprinting studies have indicated preferential binding at d(CA)⅐d(TG) steps (4). NMR studies using nuclear Overhauser effect data have confirmed that sequences containing more than one potential site can form two different intercalation complexes with roughly equal proportions (5). The x-ray structures of the DNA complexes with nogalamycin and with daunomycin and adriamycin, other anthracycline antibiotics with anti-tumor activities and similar sequence preferences, have utilized modifications to the hexameric sequence d(CGTACG). The initial structure of this se-ries used in the high resolution x-ray structure of a complex with daunomycin (6,7). Intercalated at steps 1 and 5. This structure and the related structures with d(CGATCG) or d(T-GATCA) utilized the same 2:1 drug:DNA stoichiometry, crystallizing as an isostructural series in space group P4 3 2 1 2 with 0.5 complex in the asymmetric unit. Later structures with nogalamycin in other space groups with either 0.5 or 1 complex per asymmetric unit have utilized *CGT s A*CG (6,7) and TGATCA (8,9), where *C indicates C or 5-MeC and where the phosphate group of the central step was replaced by a (R)-p-phosphorothioate to improve the crystallization characteristics.
The intercalation of nogalamycin into DNA presents a mechanistic problem, since the dumbbell shape of the intercalator with its bulky sugar residues presumably requires substantial deformation of the DNA to effect entry. These deformations may persist in the resultant complex structure where the highly hydrophobic nogalose sugar lying in the minor groove will displace water molecules from the DNA in order to utilize maximum Van der Waals interactions. In the structural studies where there are two drug molecules per hexamer at steps 1 and 5, the deformation required to effect the intercalation process may be reduced, because there is no longer a requirement to propagate deformation beyond the terminal steps of the DNA. Another consideration in all previous crystallographic work is that the orientation of the two drug molecules is 2-fold symmetric about the central step in the DNA with 4 base pairs separating the two intercalation steps.
We were interested in determining the structure of a nogalamycin:DNA complex containing a single drug molecule. Early work by Viswamitra and Salisbury 1 established that crystalline complexes could be obtained with nogalamycin and the synthetic hexamer d(CCCGGG). However there were severe problems with the reproducibility of crystallization and the poor diffraction and thermal characteristics of the crystals, which optimally are unstable above 10°C. However we were particularly interested in this sequence, which has the advantage of a single, centrally located, putative site and therefore the potential for observing a stepwise effect of intercalation on the DNA conformation.

MATERIALS AND METHODS
Preparation and Crystallization-The hexanucleotide was synthesized by automated methods with an Applied Biosystem 391 synthesizer and phosphoroamidite monomers obtained from Millipore with 4 ϫ 1 mol G-loaded CpG columns. The product was hydrolyzed with 3 M aqueous ammonia, washed with ether, and precipitated with ethanol.
After lyophilization, the unpurified product was tested to determine the optimum stoichiometry for crystallization. Since nogalamycin is almost insoluble in water at pH 7, the complex was formed by dissolving the nogalamycin in chloroform and extracting with a 2 mM solution of the hexamer at 30°C. Complexation was accompanied by a progressive reddening of the aqueous phase. This process was hindered or prevented by increasing concentrations of salt in the hexamer solution, whether added as buffer (sodium cacodylate) or as MgCl 2 .
Crystallization was done by the sitting drop method in nine-well Dow Corning plates in groups of three wells with the plate ground flat to permit sealing with standard microscope slides and with the central well in each group connected to the two flanking wells by a notch cut in the glass. Each three-well group thus forms a microdiffusion chamber suitable for reverse vapor diffusion by placing the absorbent solution (here 50% methylpentanediol (MPD)) 2 in the central well and the droplets containing the complex in the flanking wells. Red platelets were obtained at 4°C in the presence of spermine at nogalamycin:hexamer: spermine ratios of 1:2:1 and upwards, but optimally near a 2:2:1 ratio where the excess nogalamycin was solubilized after the addition of MPD. The crystal habit was pinacoids {001} competing with tetragonal bipyramids {101}. Rapid cooling from ambient temperatures favored the pinacoid form. Increasing the starting concentrations of MPD above 12% favored the bipyramidal form. Although the crystallization was successful with unpurified product, the crystals were prone to defects detectable by anomalies in the birefringence edges. The crystals melted rapidly at temperatures above 10°C, decomposing slowly, with loss of birefringence, above 4°C. The crystallization was reversible, indicating a stable complex. Further purification of the hexanucleotide by reverse phase HPLC reduced the occurrence of crystal defects, but only after desalting and replacement of ammonium cations with sodium.
The experimental conditions for obtaining crystals for intensity measurements were as follows. The nogalamycin was solubilized in the DNA as described above. The initial droplet composition was: 0.3 mM hexanucleotide (sodium form), 0.6 mM nogalamycin in a droplet volume of 100 -120 l containing 2 mM MgCl 2 and 12% MPD and 0.3 mM spermine. The central reservoir in each group contained 200 l of 50% aqueous MPD. The plate was sealed, placed in an oven at 38°C for 6 h, and then transferred to a double-sealed insulated box containing water in two compartments, one inside the other. The complete system was set inside a cold room at 3-4°C and the plate removed from the bath after 24 h. Crystals appeared within 2 weeks growing to an maximum size of 0.25 ϫ 0.25 ϫ 0.15 mm over a period of a month with both {001} and {101} forms present. These crystals were stable for a period of about 4 months at 4°C.
Data Collections-Diffraction data were recorded at the LURE synchrotron facility, in Orsay, at a wavelength of 0.9 Å using a MAR Research Image plate system connected to the monochromated W32 beam line (10). Two crystals were randomly oriented in capillaries without alignment of a crystallographic axis with the spindle. The temperature was maintained throughout the recordings below 4°C to limit decomposition. The crystal-to-film distance was set according to the maximum observed diffraction limit. Two different crystals were used each covering a large domain of rotation. Each intensity data set was processed separately using the MOSFLM package (11) interfaced with the CCP4 suite of programs (12). Table I summarizes the results of the processing. Afterwards, the two different data sets were merged together with a R eq ϭ 5.2% at the maximum resolution of 2.25 Å, and a   1. Chemical formulae of daunomycin and nogalamycin with the numbering adopted in this study. A polar amino-sugar is linked at the right side of the aglycone (C7) in daunomycin and at the left side (C1) in nogalamycin. In both drugs, the opposite side of the aglycone bears a hydrophobic residue: a small methoxy group for daunomycin and a bulky nogalose residue for nogalamycin. In spite of these differences, the two drugs have their aglycone inserted the same way in DNA.
Structure Determination-The structure was solved using the molecular replacement method. A model was built using spare parts from different sources. The first part was the nogalamycin intercalated between the C-G duplex, obtained from previous analyses on the 2:1 nogalamycin complexes (6,7). To this central piece (model B) was added two dinucleotide sequences C2/G2 and G2/C2, constructed in the B form with an helix-generating program (models A and C). Further models using the co-ordinates of nogalamycin from high resolution work (13) were also built as alternative starting points.
The multibody technique developed in the AMoRE program (14) was employed. Model B (nogalamycin plus the 2 flanking base pairs) was first oriented and translated according to the best solution (correlation factor C ϭ 44 and R factor ϭ 51%). To this fixed solution was added, in the translation search, the best rotation solution for the second, then the third spare part (models A and C). Only one solution with a correlation factor of 61 and an R factor of 47% emerged. The second peak was 45% in correlation. A graphic inspection showed that the three pieces were on top of each other, with correct orientation to built the missing phosphodiester linkages between them. The model was then submitted to the fast rigid body refinement included in AMoRe, the FITING step (15). The final correlation coefficient was C ϭ 64.9 and the R factor ϭ 44.1% (resolution: 15-3 Å) in P4 3 2 1 2, with a good packing while the best solution in P4 1 2 1 2 gave C ϭ 59.6 and R ϭ 49.3 with important overlaps in the packing. Once the three parts were oriented and translated according to the best correlation factor, the model was further divided in seven parts (the six CG or GC base pairs plus the nogalamycin) and refined again with the FITING program. The whole model was reconstructed on a graphic display using 2F o Ϫ F c maps. The R factor was 33% at 3.1 Å resolution.
Refinements-The rigid body model was submitted to the annealing procedure described in XPLOR (16). The starting resolution of 3.1 Å was gradually increased to the maximum of 2.4 Å. A combination of crystallographic and molecular dynamic refinements were performed during which solvent atoms with good hydrogen bonding geometry were gradually added to the model following visual inspection of 2F o Ϫ F c and F o Ϫ F c maps on a graphic system. The maximum temperature of the molecular dynamics was gradually decreased from 900 to 300 K as the refinement stabilized following inclusion of well defined solvent atoms. During this process, the conformation of the sugar residues changed toward their final values without manual intervention. Difference maps indicated that one of the methoxy groups of the nogalose residue had changed and needed manual repositioning. The co-ordination geometries around prominent solvent peaks were periodically examined to determine whether they would be better modeled as hydrated Na ϩ or Mg 2ϩ ions. Solvent atoms were also discarded if during B factor evolution, individual values became greater than B ϭ 60 Å 2 . The final stages of the refinement was performed with NUCLSQ (17) essentially to flatten the rings, but otherwise no constraints were placed on torsion angles or sugar conformations. The final geometric parameters and deviations from ideality are given in Table II.

Overall Molecular Structure
The refined structure is shown in Fig. 2 (stereo view) which gives a view of the unsolvated complex perpendicular to the long axis of the drug and the DNA helix axis. The residues are numbered from the 5Ј end C1 to G6 in chain 1 and C7 to G12 in chain 2. The DNA is oriented with the C1:G12 base pair at the top and the C7:G6 base pair at the bottom; the nogalamycin is intercalated between the C3:G10 and G4:C9 base pairs with the long axis of the chromophore perpendicular to the mean direction of the 2 flanking base pairs and with the nogalose and amino-glucose residues pointing toward the G6:C7 base pair at the bottom. The hydrophobic nogalose sugar residue is situated in the minor groove on the right-hand side of Fig. 2; it is associated with the duplex mainly by van der Waals interactions. The hydrophilic amino-glucose lies in the heavily solvated major groove on the left-hand side of the Fig. 2, it has several strong hydrogen bonds strengthening its association with the duplex. The overall shape of the DNA in the complex is intermediate between the A-and B-conformation. The sugar The aglycone inserted at the central step is slightly bent. This view shows how the major groove is widened in the region where the amino-glucose interacts with the 2 base pairs below the intercalation step. Also visible in the minor groove is the carbomethoxy group at C7 which is hydrogen-bonded to the guanine one step above. The symmetry related duplexes, tightly stacked on top and bottom of the complex, are depicted in red.
groups of the cytosine residues have conformations belonging to the c3Ј-endo family. The sugar groups of the guanosine residues have conformations belonging to the c2Ј-endo family with the exception of G10, which is c3Ј-endo. Thus 5 of the base pairs are composed of a cytosine in the A-conformation and a guanosine in the B-conformation.

Crystal Packing
The complex is oriented in the cell so that the helix axis of the duplex is parallel to the x and y axes (Fig. 3). There is a tight contact between the base planes of the C1:G12 base pair and those of a symmetry related G6*:C7* base pair (see Figs. 2, 3 and 9A). This leads to the formation of two sets of stacked bases, extended along the x and y axes, and crossing each other at right angles. They associate in the packing (by solvent-mediated well defined hydrogen bonds) where the major grooves face and interact with each other. The interactions in this region involve the phosphate backbone (P2-P4) of one duplex and the major groove of the crossing complex and a small number of critical solvent molecules. The protruding amino-glucose groups of both complexes form an integral part of the association. On the opposite side of the duplex, where the hydrophobic nogalose is situated in the minor groove, the packing arrangement delimits an infinite channel parallel to the z axis and about 20 Å in length and 10 Å in width. This region of the structure is characterized by higher thermal parameters for both the nogalose (B ϭ 28 -31 Å 2 ) and the duplex (B ϭ 22-29 Å 2 ) and is apparently void of well ordered solvent atoms.

The Nogalamycin in the Complex
In the Minor Groove (Fig. 4)-The nogalose residue sits in the minor groove of the duplex reaching almost 2 bases after the C-G site and oriented toward the same end of the duplex as does the amino-glucose. The minor groove is shallower and wider than the prototype B-DNA structure observed (18) for AT-rich structures and is closer toward a B-DNA structures containing CG-rich stretches, where there is enough width to allow a double chain or zig-zag pattern of water molecules (19 -21). This characterization is appropriate in that the nogalose is in van der Waals contact with chain 1 (G4 to G6), but less so with chain 2 (Fig. 4). There is an apparently empty channel between the nogalose residue and the other bank of the minor groove. (residues C7, C8, and C9).
In the Major Groove-The amino-glucose residue resides in the major groove extending from the central C3-G4 site toward the 3Ј end of chain 1. The amino-glucose spans the major groove edge of the G:C pair G4:C9 but with only one direct hydrogen bond between the amino-glucose and the duplex. This is between O2G and N7 of G4 (2.84 Å). On the other side of the edge, the interaction between O4G and N4 of C9 is mediated by a bridging water molecule W2 (O4G-(2.79 Å)-W2-(2.89 Å)-C9N4). In this respect the interactions between the amino-glucose and the duplex are close to previous structures. There is an interaction between the amino-glucose and the duplex 2 base pairs after the intercalation site. This involves the dimethylamino group and the amino group C8N4 of the G5:C8 base pair and a bridging solvent molecule W28. This hydrogen bonding interaction is strong as indicated by the hydrogen bonding distances: (N3G-(2.60 Å)-W28-(2.81 Å)-C8N4).
The orientation of the dimethylamino group is probably influenced to some extent by solvation effects where the aminoglucose partakes in the intricate network of hydrogen bonds corresponding to the "crossover" contact. There is a direct hydrogen bond (3.08 Å) between O4G of the aglycone and a phosphate oxygen (C2*O1P). The solvent molecules W2 and W28 are also hydrogen-bonded to C2*O2P (2.87 and 2.79 Å, respectively). The dimethylamino group is probably influenced by W28, which participates in the strong intermolecular interaction (N3G-(2.60 Å)-W28-(2.79 Å)-C2*O2P). The solvent molecule W28 is tetrahedrally co-ordinated with a very low thermal parameter (B ϭ 13 Å 2 ), see Fig. 8. However, there are some differences in the sugar conformations when compared to nogalamycin in the 2:1 complexes (Fig. 5). The free and bound state comparisons (least squares fit) of the nogalamycin alone clearly show that either the bending and conformations of the sugars or dimethyl-amino group orientation are different.
The aglycone is inserted at the C-G central step, with specificity determined by the flanking sugar interactions, one step behind and two steps ahead as indicated previously and by interactions between the O1 and O12 atoms and the bridging phosphate group between C3 and G4. The orientation is similar as to the other 2:1 complexes and induces equivalent unwinding (Table III). A new feature in this structure is a bridging water molecule W17 which connects the atoms O1 and O12 of the anthracycline aglycone to the phosphate group which spans the intercalation site on the C3-G4 step (O12-(2.97 Å)-W17 (3.09 Å)-G4O1P) (Figs. 6 and 8).
The Axial Carbomethoxy Group-This group is situated within the minor groove of the duplex interacting one step  x (and y). This illustrates the tight interactions between crossing helices on one side of the major groove. On the opposite side, the nogalose sugar and the P10 -P13 phosphate backbone point toward the channels, with essentially no restriction. This kind of packing has been observed recently in RNA crystals (35). before the C-G insertion site. There is a hydrogen bond between the carbonyl oxygen O14 of the aglycone and G10N2 (3.15 Å). There are two additional polar contacts through water molecules toward the furanose oxygen atoms of residues C1 and C2. The geometry of these interactions suggests that alteration to this carbomethoxy group (i.e. to a free carboxylate) could affect both the solubility and the binding affinity of the drug by strengthening the hydrogen bonding in this region.

Distribution of Solvent
The hydration on the major groove side of the duplex is ordered with readily visible water molecules and an abundance of significant electron density. This contains the more polar of the two sugar residues, the amino glucose. On the minor groove side there are very few ordered water molecules. The way in which the nogalose sits in the minor groove leaves a space which should be filled with solvent/MPD (Fig. 3); however, the electron density is too low to say what the fill is. One groove is more hydrated than the other, and low B values are observed.
About 47 water molecules were localized with temperature B factors ranging from 12 to 60 Å 2 . They form in the major groove a complicated hydrogen bonded network with many tetra-coordinated molecules. The most important, already described above, are W28 which connect the O4G of the amino-glucose to O2P of C2*, the N4 of C8, and W20 (2.9 Å, this water molecule is one having the lowest B temperature factors, B ϭ 12 Å 2 ). W20 also connects O2P of C3 (2.65 Å, symmetry-related) and two other water molecules: W15 (2.80 Å) and W30 (2.79 Å). Finally, W30 is connected to W9 (3.03 Å), N4 of C7 (2.81 Å), and W34 (2.99 Å). The water molecules in this region represent a remarkable arrangement of five tetra-co-ordinated water molecules, that builds a strong zig-zag spine, not analogous to the one observed in the minor groove of the B form of DNA (19,21), but equivalent in characteristics.
In the same region a hydration pentagon is observed in the major groove, probably related (22) to the A-like conformations of the cytosines, connecting the O1P atom of C2 and the nogalamycin O5G atom (Fig. 7). These water molecules have low B values, in the range 13-20 Å 2 .
Some phosphate groups are bridged by water molecules like W11 or W10. These molecules are located on the first strand C1-G6. The second strand is less hydrated. The minor groove contains six water molecules (with ϽBϾ Ն 45 Å 2 ), while more than 25 water molecules are present in the major groove (Fig. 8).

Helical Properties
The helical parameters of the duplex and the torsion angles of the sugar-phosphate backbone were calculated with the program NEWHEL (23) and are given in Tables III and IV. In order to make enough room for the nogalamycin to enter between the central C3:G10 and G4:C9 base pairs (step 3), the helix has been unwound, allowing a gap. This separation is accompanied by some changes in the torsion angles along the sugar-phosphate backbone, including changeover in sugar conformation from c3Ј-endo to c2Ј-endo at the intercalation site for chain 1 and one step further along from the intercalation site the 3Ј end of chain 2.
In this structure, it is possible to see how the deformation is propagated in a diminished fashion two steps above and below the intercalation site. At the central step the helical twist angle is reduced from 36°(for B-DNA) to 20.0°corresponding to an unwinding of 16.0°. The rise parameter h, corresponding to the vertical displacement of the C1Ј-C1Ј vectors parallel to the helix axis, is 5.9 Å compared with a value of 3.4 Å for B-DNA. The cavity in which the nogalamycin aglycone sits is not flattened across top and bottom but is buckled upwards from the centre by Ϫ14°for the C3:G10 base pair and downwards by ϩ11°for the G4:C9 base pair. The propellor twist angles for the C3:G10 and G4:C9 base pairs are reduced to values of Ϫ3°and ϩ3°, respectively. The diamond-shaped cavity is similar to that observed in previous structures containing nogalamycin, where  Table III shows how the propellor twist and buckle angles change from special values at the intercalation site to more standard for the terminal base pairs. The only base pair away from the cavity that shows an appreciable buckling angle is the intermediate base pair C2:G11, which has a value of ϩ7.0°. This corresponds to a change of ϩ21°from the C3:G10 base pair. It is probably correlated with the observation that the sugar-pucker changes from c3Ј-endo for G10 to c2Ј-endo for G11.
The manner in which the individual base pairs stack on top of each other can be seen in Fig. 9. There is an overall similarity to the stacking observed in d(G4C4) (24) and other C-and G-rich structures (25) that crystallize in the A-conformation. There is partial overlap between guanine residues of successive base pairs but the overlap between cytosine bases is poor. The large values of Ϫ17.0°and Ϫ13.7°for the propellor twist angles of base pairs C1:G12 and G6:C7 are probably indicative of structural adjustments associated with optimal stacking of adjacent complexes along the x and y axes.
The adjustments along the sugar-phosphate backbone are both complex and regular with individual backbone torsion angles falling near to the average values for A-DNA or B-DNA depending on the sugar conformation of the particular residue. The sugar conformations belong to the c3Ј-endo family for each cytosine residue and to the c2Ј-endo family for each guanosine residue except for G10 which is c2Ј-endo. In terms of pseudorotation parameters (26,27), depicted in Fig. 10, the cytosine   "Propeller twist" "Buckle" residues and G10 are clustered in the region 3 E (Northern) and the remaining guanosine residues are spread over region 3 E to 2 E, situated on the other side (Southern) in the diagram. For chain 1 the conformation is A-like for the first three nucleotides changing at the intercalation step to a B-like conformation for the last three residues. For chain 2 the conformation is A-like for the first four residues, changing to B-like for the last two residues. The transition from A to B shows up in the P i Ϫ P iϩ1 distances along each chain. One interesting feature of these values is that for the intercalation step, the distances, 6.36 Å from P3 to P4 in chain 1 and 6.14 Å from P9 to P10 in chain 2, are intermediate between the standard values of 5.9 Å for A-DNA and 7.0 Å for B-DNA (Table V).
The backbone torsion angles in Table IV are in general close to values associated with A-DNA and B-DNA. The only unusual changes from normal A-DNA and B-DNA values are ␤ ϭ 202°, ϭ 212°, and (⑀ Ϫ ) ϭ Ϫ1°for residue G4. Changing ␤ from a "normal" value around 162-202°for residue G4 allows the bases C3 and G4 to open. On the other chain, where the sugar conformations of residues C9 and G10 are c3Ј-endo, the value ␤ ϭ 174°for G10 is normal, but a small change to higher values would have a substantial effect on the opening of bases C9 and G10. The other unusual value for and ⑀ Ϫ for residue G4 is associated with a swinging of the phosphate group of P4 toward the major groove. In this structure, the phosphate oxygen atom O1P4 is now able to form a hydrogen-bonded interaction via the bridging water molecule W17 to O12 and O1 of the aglycone. DISCUSSION Intercalation is one of the most important way the drugs interact with DNA. This normally leads to perturbation of the double helix, altering the cell metabolism and very often leading to the end of the cell growth (28). A large number of such chemicals are known, some are synthetic (9-aminoacridine, acridine orange, ethidium bromide) some are naturally produced by plants (ellipticine) or fungi or bacteria (daunomycin, actinomycin, nogalamycin . . . ). They all have strong biochemical activities and are evaluated in medicine as antibiotics, antimitotics, carcinogenics . . . The mechanism of drug intercalation has been extensively investigated (29). Thermodynamic and kinetic studies by spectroscopic techniques are in favor of a two-step mechanism (30,31), while a close description of the intercalation geometry has been determined by x-ray diffraction on a number of short dinucleotide complexes as well as some longer sequences (hexamers). But up to now, all the crystallized complexes have the drug inserted in the first/last step of the DNA which limits the description because of termination effects.
Why Did We Use This Hexameric Sequence?-The original purpose of this work was to get a more detailed description for the intercalation process for a single site and in particular determine the extent to which the helical parameters are changed without the complication of termination effects. The sequence d(C 3 G 3 ) was chosen it was a self-complimentary sequence, containing a single CpG site. This avoided potential difficulties arising in the case of a hexamer with a central TpG site where the crystallization process would require some kind of discrimination between two different chains. There is an additional problem with nogalamycin as the intercalant. The DNA must at least partially melt at the CpG site to allow nogalamycin with its bulky nogalose and amino-glucose side groups to enter. In addition, there may be some conformational changes needed such as reorientation of the sugar residues with respect to the aglycone to permit entry. Also there may a preliminary transition intermediate defining specificity in the direction of approach.
Our crystallization problems seem to concur with these requirements. First we observed that the crude hexamer readily formed a crystalline complex but that addition of salt, whether intentionally or as an artifact of HPLC purification, hinders solubilization of the drug in the nucleotide solution. High salt conditions tend to stabilize the duplex. Second, we found that the ability to form well diffracting crystals involved annealing at 38°and slow cooling to 4°C. The instability of the crystals above 4°C is probably directly related to the weak interactions between the extended solvent channels onto which the minor     groove of the complex and the nogalose face. Re-annealing followed by slow-cooling probably helps purify the complex in terms of improving conformational homogeneity of the sugar residues thus reducing the tendency for crystal defects. The refinement of the structure from the initial model is interesting because the oligonucleotide pieces A and C were initially built as B-DNA models. The conversion of the sugar conformations to c3Ј-endo for cytosine occurred after the addition of the prominent solvent atoms and subsequent annealing at 900 K with XPLOR. We finished the refinement using NULCSQ, because we suspected that the nonplanarity of the aromatic core of the aglycone was an artifact coming from the dictionary. In our final model, the aromatic core of the anthracycline ring system was still slightly nonplanar and similar to the previous nogalamycin complexes (6 -9). In NUCLSQ, the option of constraining the sugar puckers was not used and the pseudo-rotation parameters were not restricted to specific A-or B-DNA target values but left to refine freely.
How Does the Drug Enter?-In the case of nogalamycin with its two bulky side groups, there is a problem with complex formation, which requires some kind of melting or opening up of the base pairs at the intercalation step. This can be achieved either by melting the duplex and/or by local A-to B-DNA interconversion. The A/B-DNA interconversion is a known process that occurs in solution, it may even be trapped during the crystallization process (32).
From a thermodynamical viewpoint, changing the conformation of the sugar residues from c2Ј-endo and c3Ј-endo or vice versa is not a problem at 4°C, since both conformations have been observed together in the same crystal. However the regularity of the sugar conformations in these two domains seems inconsistent with the low melting temperature for the crystals. This suggests some distinct stabilizing property associated either with the sugar conformations and/or the crystal packing and hydration. There are a variety of sugar conformations around intercalation site in earlier 2:1 complexes (8,9), whereas here the variation is well defined. Although there is good agreement with crystalline complexes of dinucleoside monophosphates where opening up is observed by variation of ␤ and ␥ torsion angles at the intercalation step (33,34), we have now a better description because our structure is extended 2 base pairs on each side of the intercalation step.
Based on This Structure, Is There a Sequence Specificity for Nogalamycin?-Nogalamycin has been described as having a high affinity for TpA or CpG or TpG steps. Comparison with daunomycin is interesting because the orientation of the polar and hydrophobic groups in the two molecules are reversed (Fig.  1). First, the intercalation of daunomycin would require less energy than nogalamycin. Second the insertion of the aglycone in the DNA follows the same orientation in the two molecules: either the hydrophobic nogalose, in the case of nogalamycin, or the polar amino-sugar, in the case of daunomycin, is located in the minor groove (6, 7), a strong indication that the hydrophobicity/hydrophilicity parameter is not the principal driving force for the insertion. On the other side of the duplex, nogalamycin fills the major groove with its bulky amino-glucose, while daunomycin has only a small methoxy group. The only group that shares similar orientation and interactions between the two drugs is the carbonyl group (methylketone in daunomycin, methyl ester in nogalamycin) at C9. In all of the structures, it interacts with the G11 residue one step before either directly or mediated by a water molecule, despite a different configuration (axial in nogalamycin, equatorial in daunomycin).
Surprisingly, nogalamycin does not interact strongly by direct bonding to the CG step where it sits. In our structure only one direct interaction is observed in the major groove, O2G 3 N7 of G4, d ϭ 2.84 Å. In the 2:1 complexes already described it is the O4G of the drug which interacts directly to N4 of C9: d ϭ 3.10 Å (in the present work the corresponding distance is 3.56 Å, mediated by the bridging water W2).
In contrast nogalamycin forms a number of important hydrogen bonds between the three donor/acceptor groups (O2G, N3G, and O4G) of the amino-glucose residue and the G4:C9 base pair which is spanned by the amino-glucose. This must be compared with previous structures and is certainly at the origin of the high specificity of nogalamycin. However the comparison with other structures is not straightforward as they all contain m5C-G steps instead of pure C-G steps (The present structure is the first of this kind).
As in previous 2:1 complexes the dimethyl-amino group N3G is strongly involved in the hydrogen bond network, but here, due to a rotation, there is a different hydration. It interacts through the tetra co-ordinated solvent residue W28 to N4 of residue C8. W28 is part of a strongly defined network, including a symmetry-related phosphate backbone (Fig. 8). In this respect, nogalamycin displays a completely different type of interaction network from that observed in daunomycin complexes which may well explain why its specificity is increased.
Finally, the parameters of the duplex at the two ends go back to "normal" values, indicating that the DNA deformation is rapidly compensated, at least Ϯ two steps before and after the intercalation site.