Study of macromolecular interactions using computational solvent mapping
The term "binding hot spots" refers to regions of a protein surface with large contributions to the binding free energy. Computational solvent mapping serves as an analog to the major experimental techniques developed for the identification of such hot spots using X-ray and nuclear magnetic resonance (NMR) methods. Applications of the fast Fourier-transform-based mapping algorithm FTMap show that similar binding hot spots also occur in DNA molecules and interact with small molecules that bind to DNA with high affinity. Solvent mapping results on B-DNA, with or without Hoogsteen (HG) base pairing, have revealed the significance of "HG breathing" on the reactivity of DNA with formaldehyde. Extending the method to RNA molecules, I applied the FTMap algorithm to flexible structures of HIV-1 transactivation response element (TAR) RNA and Tau exon 10 RNA. Results show that despite the extremely flexible nature of these small RNA molecules, nucleic acid bases that interact with ligands consistently have high hit rates, and thus binding sites can be successfully identified. Based on this experience as well as the prior work on DNA, I extended the FTMap algorithm to mapping nucleic acids and implemented it in an automated online server available to the research community. FTSite, a related server for finding binding sites of proteins, was also extended to develop PeptiMap, an accurate and robust protocol that can determine peptide binding sites on proteins. Analyses of structural ensembles of ligand-free proteins using solvent mapping have shown that such ensembles contain pre-existing binding hot spots, and that such hot spots can be identified without any a priori knowledge of the ligand-bound structure. Furthermore, the structures in the ensemble having the highest binding-site hit rate are closest to the ligand-bound structure, and a higher hit rate implies improved structural similarity between the unbound protein and its bound state, resulting in high correlation coefficient between the two measures. These advances should greatly enhance researchers' ability to identify functionally important interactions among biomolecules in silico.