The ability of the PM3 semiempirical quantum mechanical method to reproduce hydrogen bonding in nucleotide base pairs was assessed. Results Of PM3 calculations on the nucleotides 2'-deoxyadenosine 5'-monophosphate (pdA), 2'-deoxyguanosine 5'-monophosphate (pdG), 2'-deoxycytidine 5'-monophosphate ( pdC), and 2'-deoxythymidine 5'-monophosphate (pdT) and the base pairs pdA-pdT, pdGpdC, and pdG(syn)-pdC are presented and discussed. The PM3 method is the first of the parameterized NDDO quantum mechanical models with any ability to reproduce hydrogen bonding between nucleotide base pairs. Intermolecular hydrogen bond lengths between nucleotides displaying Watson-Crick base pairing are 0.1-0.2 , & less than experimental results. Nucleotide bond distances, bond angles, and torsion angles about the glycosyl bond (X), the C4,-Cs, bond (y), and the C5,-05, bond ( p ) agree with experimental results. There are many possible conformations of nucleotides. PM3 calculations reveal that many of the most stable conformations are stabilized by intramolecular C-H---0 hydrogen bonds. These interactions disrupt the usual sugar puckering. The stacking interactions of a dT-pdA duplex are examined at different levels of gradient optimization. The intramolecular hydrogen bonds found in the nucleotide base pairs disappear in the duplex, as a result of the additional constraints on the phosphate group when part of a DNA backbone. Sugar puckering is reproduced by the PM3 method for the four bases in the dT-pdA duplex. PM3 underestimates the attractive stacking interactions of base pairs in a B-DNA helical conformation. The performance of the PM3 method implemented in SPARTAN is contrasted with that implemented in MOPAC. At present, accurate ab inilio calculations are too timeconsuming to be of practical use, and molecular mechanics methods cannot be used to determine quantum mechanical properties such as reaction-path calculations, transition-state structures, and activation energies. The PM3 method should be used with extreme caution for examination of small DNA systems. Future parameterizations of semiempirical methods should incorporate base stacking interactions into the parameterization data set to enhance the ability of these methods.
