Aligning and modeling protein-DNA interfaces: Towards an understanding of protein-DNA binding specificity

Years of structural work characterizing protein-DNA complexes has provided tremendous insight into the mechanisms of binding specificity. Currently, modeling approaches capitalizing on this wealth of structural data are being use to further understand specificity. Protein-DNA complexes are used as templates on which to model alternate protein and DNA sequences, and binding energies are calculated for the modeled complexes. However, the relationship between the protein-DNA interface geometry (docking arrangement) and binding is still poorly understood. The focus of this dissertation is to analyze the geometric properties of protein-DNA interfaces and their effect on atomic-level modeling of binding specificity.; A new method is introduced to structurally align interfacial amino acids observed in protein-DNA complexes. The spatial relationships of individual amino acid-nucleotide pairs are quantified and a dynamic-programming algorithm optimally aligns residues from different complexes using these relationships. An interface alignment score, IAS, measures the alignment quality and provides a quantitative measure of the similarity in the docking geometry between two protein-DNA complexes. A large set of protein-DNA complexes are aligned and clustered based on their IAS values. Proteins within a single family form identifiable clusters; however, subgroup clustering is often observed within families. Although proteins with similar folds tend to dock in similar ways, important differences are observed even for structural motifs that almost perfectly align. Relationships are observed between the interfaces formed in cognate and non-cognate complexes involving the same proteins indicating a strong driving force to maintain certain contacts, even if this requires a distortion of the DNA.; A novel interface modeling approach is described that uses rotamer-based descriptions of protein sidechains and DNA bases, a Monte Carlo search algorithm and a molecular-mechanics energy function. Modeling and binding-energy calculations for the Zif268 zinc finger demonstrate excellent agreement with experimental binding data when the wild-type complex is used as the template. However, predictions using templates with decreasing interface similarity to the wild-type complex demonstrate decreasing accuracy. Sidechain placement, a critical feature in modeling protein-DNA binding, shows similar correlations. The prediction accuracy of key interacting sidechain-base pairs depends strongly on the interface properties of the templates complexes used.