Solution structures of DNA·RNA hybrids with purine-rich and pyrimidine- rich strands: Comparison with the homologous DNA and RNA duplexes

The structures of d(GAAGAGAAGC)·d(GCTTCTCTTC), d(GAAGAGAAGC)·r(GCUUCUCUUC), r(GAAGAGAAGC)·d(GCTTCTCTTC), and r(GAAGAGAAGC)·r(GCUUCUCUUC) have been determined in solution from NMR data. Globally, the pure DNA and RNA duplexes were in the B and A forms, respectively. The two DNA·RNA hybrids were neither A nor B, but closer globally to the A than the B form. However, the thermodynamically less stable d(GAAGAGAAGC)·r(GCUUCUCUUC) duplex has a significantly different conformation from r(GAAGAGAAGC)·d(GCTTCTCTTC). Structures were calculated based on the NMR data, using restrained molecular dynamics. A new approach to the treatment of conformational averaging based on a priori probabilities has been used. The nucleotides were treated by fitting the scalar coupling data and NOE time courses to a two-state model comprising N and S sugar puckers each with a different glycosidic torsion angle, and the mole fraction of the S state. Restraint sets for different distributions of N and S states within molecules were constructed, such that each nucleotide was weighted in the ensemble according to the mole fractions (or a priori probabilities). The individual nucleotide conformations were strongly restrained, whereas the internucleotide restraints were set relatively loosely. Ensembles of conformations were generated and assessed by comparison of the NOEs calculated from ensemble-averaged relaxation matrices with the experimental NOEs. The ensemble averages accounted for the experimental data much better than any individual member, or for structures calculated assuming a single unique conformation. The two hybrids populated different degrees of conformational space. There was a general trend in minor and major groove widths in the order d(GAAGAGAAGC)·d(GCTTCTCTTC), d(GAAGAGAAGC)·r(GCUUCUCUUC), r(GAAGAGAAGC)·d(GCTTCTCTTC), r(GAAGAGAAGC)·r(GCUUCUCUUC) and a similar progression in global character from B-like to A-like structures. Furthermore, r(GAAGAGAAGC)·d(GCTTCTCTTC) showed a greater dispersion of conformations in the ensemble than d(GAAGAGAAGC)·r(GCUUCUCUUC), reflecting the greater flexibility of the sugars. If conformational averaging of the nucleotides is ignored, incorrect virtual structures are produced that nevertheless are able to satisfy a substantial fraction of the experimental data.