ABSTRACT
Gel migration and uranyl photoprobing have been used to study the effects of inosine and 2,6-diaminopurine (2,6-DAP) substitution on adenine-tract (A-tract) induced DNA curvature. Using a 10mer repeated sequence including five inosines we show by uranyl photoprobing that a narrow minor groove varying in phase with the helix repeat is not the cause of DNA curvature. Further, we have systematically studied by gel migration the effects on A-tract induced curvature of either single or full substitution with inosine and/or 2,6-DAP in a 5'-AAAAAGCCGC-3' sequence. DNA curvature is shown to increase when inosines are substituted for the guanosines in the sequence between the A-tracts. By comparing the effects of each monosubstitution it can be seen that when the G closest to the 3'-end of the A-tract is substituted the effect on DNA curvature is much stronger than when substitution is made at any other position. By contrast, curvature is abolished when 2,6-DAP residues are substituted for all adenines, and monosubstitution reveals that the effect of substituting a single adenine is strongest at the 3'-end of the A-tract. These results favor a model in which the curvature induced by an A-tract in DNA molecules is primarily located at the junction with the 3'-end of the A-tract, and this peculiar junction is created because the A-tract has a preference to form a non-B-DNA structure which builds up from the 5'-end.
Although macroscopic curvature of DNA was discovered more than a decade ago (1 ,2 ), the structural basis for this phenomenon remains a subject for discussion. Agreement exists that a distinct curvature is associated with the repetition of at least four adenines in phase with the helix repeat (2 -4 ), and different models have been presented to explain the structural importance of the adenine-tracts (A-tracts) in bent DNA. Two models, not mutually exclusive, suggest that the structure of an A-tract itself is the cause of curvature. The wedge model is based on evidence that 2 bp (ApA) form a wedge, and as a consequence the local curvature associated with an A-tract is produced by the additive effects of adjacent ApA steps within it (5 -8 ). Alternatively, the junction model assumes that A-tracts are locally straight, but curvature of DNA arises as a result of stacking interactions at the junctions between an unusual A-tract conformation and normal B-DNA, which cause the two different helices to lie non-parallel. DNA curvature in this model can occur at both the 3' and the 5' junctions between A-tracts and mixed DNA, but the two junctions are not equivalent, which implies a lack of 2-fold symmetry in the A-tract structure (2 ,3 ,9 ,10 ). In both of these models macroscopic curvature is produced when the A-tracts are repeated in phase with the helix screw. In agreement with Haran et al. (11 ) we wish to emphasize that the junction and the wedge models relate to different aspects of the DNA helix, and that the same DNA structure could, in principle, be adequately and simultaneously described by both models.
The third model is based mainly on X-ray crystallographic data. This model, recently discussed by Goodsell et al. (12 ), suggests that the A-tract is actually straight and it is the regions between the A-tracts that are bent (13 -15 ).
Electrophoretic behaviour in gels has furnished most information about the sequence dependence of DNA curvature, thanks to the anomalously slow migration of curved molecules in polyacrylamide gels (3 ,15 -19 ). These experiments have shown for instance that base substitutions within the A-tract have a strong negative effect on curvature, and inosine (I) is the only nucleoside that can be tolerated in place of adenosine in an A-tract without affecting curvature drastically (18 ,19 ). However, an I-tract does not of itself produce a bent DNA structure, mainly attributed to weaker stacking interactions of I-C base pairs (18 ). Furthermore, gel studies have shown that curvature is not significantly modulated by sequence changes outside the A-tracts except for a possible effect of a 5'-CA-3' step that may enhance curvature. In these studies only the four normal nucleotides were used as substitutions (9 ,20 ).
We have recently shown that the structure of DNA is affected by incorporation of 2,6-diaminopurine (2,6-DAP) and inosine nucleotides. Specifically we found that I-C-tracts behave like A-T-tracts and DAP-T-tracts like G-C-tracts as regards minor groove width (21 ,22 ).
In the present study, inosine and 2,6-DAP substitutions within the sequence 5'-GGGGGATTAT have been made in order to examine the relation, if any, between minor groove width as assayed by uranyl photoprobing and DNA curvature. Furthermore, the effects on curvature of systematic inosine or 2,6-DAP substitution in a 5'-AAAAAGCCGC sequence have been studied by gel migration.
Construction of two plasmids containing A-tracts or G-tracts was obtained by cloning of nine repeats of a 5'-AAAAAGCCGC sequence resulting in pA9 or five repeats of a 5'-GGGGGATTAT sequence resulting in pG5 in the SmaI site of the polylinker of pUC19. The plasmids were cut with EcoRI and PvuII and the large EcoRI-PvuII fragments (190 bp fragment for pA9 and 150 bp fragment for pG5) were isolated on a 5% native polyacrylamide gel and eluted from the gel slice by diffusion in 90 mM Tris-borate and 1 mM EDTA pH 7.5. These fragments were used as PCR templates.
PCR products were obtained by using the `reverse sequencing' and the `forward sequencing' primersjust outside the polylinker region. This resulted in fragments of 190 bp containing the A-tract sequence (`A-tract') and fragments of 150 bp containing the G-tract sequence (`G-tract'). In addition, two classically `mixed' sequences were used for the experiment illustrated in Figure 2 : PCR products of the crp DNA were prepared from a template containing two cyclic AMP receptor binding sites cloned in the BamHI site of the polylinker of pUC19 resulting in a DNA fragment of 178 bp (the `reverse', AACAGCTATGACCATG, and the `forward', GTAAAACGACGGCCAGT, sequencing primers were used for PCR); similarly, tyrT fragments of 158 bp were prepared as described previously (21 ). PCR reactions and labeling were performed as described in Bailly et al. (21 ). After electrophoresis for ~1 h, a thin section of the gel was stained with ethidium bromide so as to locate the band of DNA under UV light. The same band of DNA free of ethidium was excised, crushed and soaked in elution buffer (500 mM ammonium acetate, 10 mM magnesium acetate) overnight at 37oC. This suspension was filtered through a Millipore 0.22 micron filter and DNA was precipitated with ethanol.
The purified PCR products were subject to uranyl photocleavage (23 ,24 ) in a volume of 100 [mu]l containing 50 mM NaAc pH 6.2 and 1 mM uranyl nitrate. The samples were irradiated in open tubes placed just below a 40 W/03 Phillips fluorescent light tube with maximum emission at 420 nm. After 20 min irradiation, NaAc pH 4.5 was added to a final concentration of 0.2 mM with 2.5 vol ethanol. The tubes were placed on ice for 15 min and thereafter centrifuged for 15 min. The dried pellet was dissolved in 6 [mu]l formamide, 90 mM Tris-borate and 1 mM EDTA, pH 8.3, containing xylene cyanol and bromophenol blue, and the samples were heated at 90oC before loading 2 [mu]l onto a 10% denaturing polyacrylamide gel (19:1 acrylamide:methylenebisacrylamide). The autoradiograms were obtained by overnight exposure using intensifying screens. Alternatively, the dried gels were examined with a Molecular Dynamics 425E Phosphorimager using Kodak phosphor storage screens. Base line-corrected scans were analysed by integrating the densities between two selected boundaries using Image Quant software. The area of each band was transferred for analysis in Microsoft Excel 5.0.
2'-deoxy-inosine phosphoramidite was purchased from CruaChem and protected 2-amino-2'-deoxyadenosine phosphoramidite was prepared according to Gryaznov and Schultz (25 ). The synthesis of the oligonucleotides was performed by standard procedures.
The complementary DNA monomers were phosphorylated with [[gamma]-32P]ATP (Amersham) and polynucleotide kinase (Gibco, BRL) in a volume of 10 [mu]l. After 10 min of incubation at 37oC ligation buffer was added together with 1 U of DNA ligase (Gibco, BRL) in a final volume of 50 [mu]l. Ligation reactions were incubated overnight at 4oC. After precipitation of the DNA, 10 [mu]l 90 mM Tris-borate and 1 mM EDTA containing bromophenol blue were added to the dried pellet and a 2 [mu]l aliquot was run on an 8% non-denaturing polyacrylamide gel (30:1 acrylamide:methylenebisacrylamide).
The 190 bp EcoRI-PvuII fragment containing nine repeats of the 5'-AAAAAGCCGC sequence from pA9, was used as a template to make PCR products containing inosine and/or 2,6-DAP nucleotides. In the modified DNA molecules, thymines are base paired with 2,6-DAP and cytosines are base paired with inosines, instead of adenines and guanines respectively.
The four 190 bp fragments, each containing one of the top four repetitive units listed in Figure 1 B, differ basically in the position of the purine 2-amino group. However, substantial differences in gel migration between the normal DNA and the molecules containing inosine and/or 2,6-DAP can be seen (Fig. 1 A, lanes 1-4).
A significantly narrowed minor groove is one of the characteristics of an A-tract (26 ,27 ) which, according to the wedge model, could result in a macroscopic curvature in a DNA molecule when helically phased.
Uranyl photoprobing of protein-DNA complexes has been used to study protein-DNA backbone contacts (23 ,28 ,29 ) and, when performed at slightly acidic pH, the uranyl mediated photocleavage of naked DNA exhibits considerable sequence modulation(24 ,30 ). A/T-tracts are cleaved much more efficiently than G/C-tracts, and in general a correlation between a putatively narrowed minor groove of the DNA helix and increased susceptibility to uranyl (binding and) cleavage is apparent (24 ,31 ). There are no indications that this slightly acidic pH (6.0) causes any change in overall DNA conformation/dynamics (32 ) or in DNA bending in particular (H.-S.Koo, personal communication).
Using uranyl photocleavage as well as DNase I probing we recently found that the difference between A-T and G-C base pairs in terms of DNA structure can be traced to the presence of the 2-amino group (21 ). Thus we observed that G -> I and A -> DAP substituted DNA when probed with DNase I or with uranyl photoinduced DNA cleavage produced cleavage patterns that are compatible with a DNA structure characterised by a narrowed minor groove at I-C-tracts and a widened minor groove at DAP-T-tracts (21 ). Therefore, by inference we would expect an I5-tract like the one present in the sequence IIIIIDTTDT to exhibit a relatively narrowed minor groove and thus show increased susceptibility to uranyl cleavage as compared to the parent G5-tract sequence (150 bp EcoRI-PvuII fragment from pG5 was used as template for PCR). The results of cleavage of the C-rich strand presented in Figure 2 show that this is indeed the case. With the normal sequence there is slight hyperreactivity leading to cleavage in the 3'-end of the 3'-TAATA-5' sequence, whereas the C5-region is less efficiently cleaved. However, inosine substitution changes the cleavage hyperreactivity to the five Cs and with an additional 2,6-DAP substitution the uranyl cleavage exhibits a pattern typical for a narrowed minor groove at the C-tract with an increased cleavage towards the 3'-end. Thus the double inosine/2,6-DAP substituted sequence displays a cleavage pattern similar to an A/T-tract sequence, which indicates that an inosine tract corresponds to an A-tract in terms of minor groove width. Interestingly, however, a similar effect is obtained by merely substituting adenine with DAP in the intervening A/T-tract. This of course would convert the A/T-tract to a DAP/T sequence (and hence in structural terms to something more like a G/C-tract that would be poorly cleaved by uranyl), but the results indicate an additional structural transition towards a narrowed minor groove in the G5-tract. However, none of the 150 bp fragments containing these I or DAP modified G-tract sequences exhibited the anomalous migration that would indicate major macrosopic DNA curvature (Fig. 1 , lanes 5-8).
A most intriguing observation arising from the gel migration experiments was the additional retardation, and thus possible enhancement of curvature, upon substitution of inosine into the intervening sequences between A-tracts (Fig. 1 , lane 2). This effect was examined in greater detail by systematically replacing each guanine between the A-tracts one by one.
The 5'-AAAAAGCCGC sequence was used. Five oligonucleotides were prepared, each containing one inosine at a specific position, which were subsequently hybridized to the complementary sequence containing the four normal bases and multimers of the repeating decanucleotide unit were obtained by ligation. The gel analysis of the ligation products is shown inFigure 3 . From this, a quantitative measure of anomalous gel mobility was obtained by plotting the RL values versus the actual size of the DNA fragment. (RL is the apparent chain length of a certain DNA fragment as judged by its migration compared to a standard DNA fragment of normal mobility divided by the actual chain length of the fragment). Figure 4 shows such a plot where the mobility of each inosine monosubstituted molecule is compared with the mobility of the normal 5'-AAAAAGCCGC sequence.
Strikingly, while all inosine monosubstituted fragments migrate slower than the normal fragment, substitution of the guanine closest to the 3'-end of the A-tract has the strongest effect on the gel mobility. This indicates a difference between the two ends of the A-tract which points to an important structural parameter involved in DNA curvature at the 3'-end. However, since the inosines are not all positioned on the same strand in the five fragments,the conclusion is not clear-cut.
All single substitutions of the adenines with DAP have a detrimental effect on curvature, as judged from the relative gel mobility (Fig. 5 ). Three of the DAP-monosubstitutions (5'-ADAAAGCCGC, 5'-AADAAGCCGC and 5'-AAADAGCCGC) result in A-tracts having less than four contiguous adenines, which is a limiting number for DNA curvature by A-tracts. These fragments all migrate faster than the normal A-tract sequence and, not unexpectedly, substitution of the middle adenine (leaving only two pairs of adjacent adenines) has the strongest neutralizing effect on the anomalous gel mobility. Substitution of adenine numbers 2 and 4 from the 5'-end in each case leaves three contiguous adenines, but substitition of the adenine number 4, which is closer to the 3'-end, has the stronger neutralizing effect on the migration anomaly. A fortiori substitution of the most 5' (5'-DAAAAGCCGC) and most 3' (5'-AAAADGCCGC) adenines also affects curvature quite differently. Remarkably, substitution of the most 5' adenine has only a moderate effect on mobility, whereas substitution of the adenine at the 3'-end has a very strong influence on the migration. The results show consistently that substitution with 2,6-DAP has more pronounced consequences the closer it is to the 3'-end. Finally, as would be expected the sequence (D5G5)n did not show any sign of curvature and neither did the (D5I5)n sequence (Fig. 6 ).
The present results bear upon the question as to how DNA curvature is affected by both the structure of the A-tract per se as well as by the junctions between an A-tract and the intervening DNA sequence. It has previously been shown that A-tracts have a significantly narrowed minor groove (26 ,27 ,33 ). However, it is still uncertain whether this is the cause of DNA curvature or not. Our results with the phased I5-tracts, which according to uranyl photoprobing do have a narrowed minor groove much resembling that of the A5-tracts (Fig. 2 ) but which do not cause DNA curvature, strongly suggest that it is not the narrow minor groove per se that is responsible for the curvature. Furthermore, the non-bent structure of the A5I5 sequence likewise shows that, since the I5-tract is also not bent, the curvature cannot be a structural feature of the A-tract itself.
Therefore, our results favour a junction type model in which the major contribution to the curvature originates at one or both of the junctions between an A-tract and the proximal DNA helix, most likely predominantly confined to the actual junction base pair step(s). In this model the curvature can be attributed to the peculiar `high propeller twist conformation' of A-tracts (26 ) which, in order to maximize base stacking interactions with the adjacent B-DNA helix, creates a `kink' at the junction. Our results are fully consistent with such a model. Accepting that an I-C base pair structurally resembles an A-T base pair and in particular, having only two hydrogen bonds, is able to adopt or may even prefer the A-T high propeller twist conformation (34 ,35 ), I-C base pairs adjacent to an (A-T)n-tract will essentially extend the A-tract structure. Using this concept the (A5I5)n oligomer should, as discussed, be straight since it probably adopts a continuous A-tract conformation without junctions. Analogously, placing just a single I-C base pair proximal to an A5-tract will effectively extend it into an A6-like-tract where the junction now lies between the I-C base pair and the adjacent B-DNA. Since an A6-tract bends DNA more efficiently than an A5-tract (18 ), an increase in DNA curvature will result. The crystal structures of several DNA duplexes containing I-C base pairs within A-tracts show that an I-C base pair can indeed adopt a high propeller twist conformation (34 ,35 ). Substitution of guanosine with inosine or adenine with diaminopurine most certainly would also change the wedge components, and could thereby account for the change in curvature. However, it cannot be questioned that a junction between an A-tract structure and a B-DNA structure is needed since the A5I5 is not curved. If this lack of curvature were to be explained solely by base pair wedges then the A-I or the I-I wedges would have to be very strong and comparable to A-A wedges. This is highly unlikely since inosine tracts repeated in phase with the helix screw do not produce any DNA curvature.
Our results also clearly show that the two junctionsof an A-tract do not contribute equally to the overall curvature of DNA. The DNA curvature induced by A-tracts is diminished by introducing a 2,6-DAP residue at any point within the five adenines in the 5'-AAAAAGCCGC sequence. However, there is a position dependence which modulates the influence of the extra 2-amino group. Substitution of the fourth adenine from the 5'-end (5'-AAADAGCCGC) has a larger negative influence on curvature than substitution of the second adenine (5'-ADAAAGCCGC), which may be difficult to explain by simple dinucleotide wedges, because the sum of all wedges does not change although the relative phase has been shifted slightly (naturally nearest neighbour effects could also come into play). This pattern becomes more pronounced when the effect of substituting the adenine closest to the 3'-end (5'-DAAAAGCCGC) is compared to modifying the most 5' adenine (5'-AAAADGCCGC), though we cannot exclude the possibility that this difference could be explained with a change of base pair wedges.
From the standpoint of the junction model this perceived difference between the 3'- and 5'-ends of the A-tracts identifies the junction at the 3'-end as being the more important, in accordance with the findings of Koo et al. (3 ) that curvature seems to be concentrated primarily at the junction between the 3'-end of an A-tract and adjacent B-DNA. This conclusion is further substantiated by the observation that an extra inosine added at the 3'-end of an A5-tract (5'-AAAAAIGGGG) results in larger curvature than adding the inosine at the 5'-end (5'-AAAAAGGGGI) (Fig. 4 ).
It is likewise clear that the nature of the base pair at the non-A-tract side of the junction critically affects the degree of curvature. For instance, the 5'-AAAAAICCGC sequence is significantly more bent than the 5'-AAAAAIGGGG sequence, which can hardly be due to the difference in the intervening sequence since AAAAAGCCGC and AAAAAGGGGG exhibit identical curvature. Consequently, the non A-tract model for DNA curvature suggesting that it is the G/C sequences between A-tracts that possess a curved structure is not really compatible with these or previous data (11 ). Also arguing against this model is the finding that substitution of G with I at any position between the A-tracts has little influence on the curvature. In contrast 2,6-DAP monosubstitution within the A-tracts always diminishes bending. Therefore, it seems unlikely that DNA curvature could be explained by any distinct G/C sequence conformation.
The differential influence of the 5' versus the 3' junction could be due to a gradual build-up of the special structure through the A-tract instead of an all or none switch as also suggested by base pair lifetime measurement on A-tracts (10 ). This would explain why A-tracts of a certain length ( >= 4 bp) are required and why the 5'-A5I-tract results in greater curvature than the 5'-IA5-tract, assuming that the I-C base pair mimics an A-T base pair well enough to extend an already existing A-tract conformation (the 5'-A5I situation) whereas it cannot initiate the process which induces this conformation from the 5'-end (the 5'-IA5 situation).
Thus, in summary, our results are entirely consistent with a model in which the curvature induced by an A-tract in DNA molecules can be traced primarily to the junction with the 3'-end of the A-tract, and this peculiar junction is created because the A-tract has a preference to form a non-B/DNA structure which builds up progressively from the 5'-end. By this we do not seek to discredit the wedge model which undoubtedly provides an adequate explanation for the intrinsic curvature of non-A-tract DNA. We do however conclude that our data strongly support the idea that A-tract induced DNA curvature involves an additional strong specific effect located at the 3'-end junction of the A-tract which is not easily explained by simple first order base pair wedges.
We would like to thank Otto Dahl, Department of Chemistry, Copenhagen University for synthesizing the oligonucleotides and Neel Louv-Jansen for expert technical assistance. This work was supported by The Danish National Research Foundation and by grants from CRC, AICR, the Wellcome Trust and the Sir Halley Stewart Trust.
*To whom correspondence should be addressed. Tel: +45 353 27762; Fax: +45 313 96042; Email: pen@biokemi.IMBG.ku.dk
REFERENCES