Table of Contents:
In this study the similarity analysis of protein binding pockets is used to annotate protein structures, detect structurally related proteins, rationalize and predict cross-reactivities of known drugs to different proteins, and to classify protein families according to the similarities in their active sites.
Cavbase is a method to describe and compare protein binding pockets according to the physicochemical properties of their active sites. Binding pockets are automatically extracted in terms of clefts on the protein surface. A cavity is represented by pseudocenters mapping physicochemical properties of the flanking amino acids together with their surface exposure. Such classified cavities are stored in Cavbase, which is integrated into the protein ligand database Relibase. The property description in Cavbase was validated and adapted by comparing the cavity surface attributed with the physicochemical properties of the pseudocenters with Superstar and Drugscore hotspots of comparable atom probes. This allows for a better representation of the physicochemical properties of the active sites.
For the comparison of different binding sites, a clique algorithm is applied to detect common substructures. A significant speed-up of the clique algorithm was achieved using heuristic filters and optimizations. It is now possible to compare very large datasets of protein cavities. Additionally, a fast similarity measure comparable to bitstring patterns used in the pharmacophore matching of small molecules was developed. Using the fast bitstring matches as filter to reduce the number of cases that have to be considered in the more elaborate clique detection algorithm, the time consumption is significantly reduced.
By optimising binding to a selected target protein, modern drug research strives to develop safe and efficacious agents for the treatment of disease. The unexpected nanomolar affinity of the COX-2 specific inhibitors Celecoxib and Valdecoxib for isoenzymes of the totally unrelated carbonic anhydrase family is rationalized using Cavbase. The similarity analysis detects similar areas in the active sites of both proteins and helps in the explanation of this cross-binding.
The functional classification of protein families is presented using two pharmaceutically relevant protein families: the alpha-carbonic anhydrases and the protein kinases.
The family of the alpha-carbonic anhydrases contains presently 14 members, all involved in the interconversion of carbon dioxide and bicarbonate. For six carbonic anhydrase isozymes the crystal structure are known. A dataset of 24 cavities is used to classify the different isozymes. Cavbase successfully distinguishes between the different members of the carbonic anhydrase subfamilies using only information about the active sites. Furthermore, structural differences in a carbonic anhydrase II important for the catalytic mechanism are detected.
The second protein family contains information for all structural known kinases (263 kinase binding pockets). Kinases are a ubiquitous group of enzymes involved in a large variety of cellular signalling pathways thus providing an interesting panel of pharmaceutically relevant targets. The elucidation of selectivity discriminating features exhibited by the binding site of protein kinases is of fundamental importance to classify kinases and provides the basis for the design of selective inhibitors. We present the results for the classification and clustering of a set of protein kinase structures with respect to their binding sites as stored in Cavbase. We are able to generate a reasonable clustering of different subtypes of the kinase family. Cavbase also distinguishes between cavities of different activation states from one subfamily. Furthermore, a number of unexpected similarities of protein kinases not related in sequence space are found.