Structural biology and bioinformatics in drug design: opportunities and challenges for target identification and lead discovery

Abstract

Impressive progress in genome sequencing, protein expression and high-throughput crystallography and NMR has radically transformed the opportunities to use protein three-dimensional structures to accelerate drug discovery, but the quantity and complexity of the data have ensured a central place for informatics. Structural biology and bioinformatics have assisted in lead optimization and target identification where they have well established roles; they can now contribute to lead discovery, exploiting high-throughput methods of structure determination that provide powerful approaches to screening of fragment binding.

Figure 1

Drug discovery classically follows the path from target selection, through lead discovery to lead development. Although structural biology has historically had a role in the final stages during lead optimization, it is now having an effect at all stages. Homology recognition and structural genomics can aid target selection, while structure-based screening assists the lead discovery and development processes.

Figure 2

An example of structurally conserved clusters for an ensemble of superposed structures. Two members of the GTP-binding protein family are shown (Pdb codes: 1ftn; 1a4r). Regions with the same colours belong to structurally conserved clusters.

Figure 3

The formula for calculating the two sequence-based scoring systems (a) conservation score, (b) divergent score and the structure based score (c) structural conservation score are shown. (d) The empirically determined weights of the sequence based score and structural conservation score, which improves the functional site prediction, are shown.

Figure 4

The Pyramid system allows lead discovery through a fragment-based approach of molecular fragment matching and fitting. (a) High resolution target structure determination. (b) Generation of Astex drug fragment library. Virtual screening used to enrich the library for fragments likely to bind the target. (c) Drug fragment cocktails used for protein crystal soaks, 4–8 compounds per cocktail. (d) High throughput protein/ligand X-ray crystallography. Automated X-ray data collection and analysis. (e) Electron density analysed by AutoSolve in order to identify bound drug fragment. (f) Structure-based optimization of hits to leads.

Figure 5

Pyramid hits-to-leads generation for Cdk2. This figure shows the electron density of various fragments bound to Cdk2. From around 500 compounds screened, 11 cocktails showed hits. From these fragments AT381 was selected with around 1 mM activity. Subsequent steps, represented by green arrows, are the optimization of fragment AT381 to improve potency, selectivity and ADME properties. © Astex Technology Ltd. 2005.

Figure 6

(a) The fibroblast growth factor (FGF; green) and its receptor (blue) contain globular domains that form a complex with the co-factor heparin, without significant changes to their 3D structures. (b) A more detailed examination of the interaction between the FGF and domain 2 (top as shown) of the receptor shows that binding sites in both proteins consist of a discontinuous epitope on surfaces that are comparatively flat for protein structures. The binding sites broadly consist of a hydrophobic centre bordered by charged and polar patches.

Figure 7

The human non-homologous end joining protein Xrcc4 binds to a flexible linker between tandem BRCT domains of DNA ligase IV, imposing structure on the linker through the interaction. (a), (b) show the full Xrcc4 structure with ligase linker bound (shown in green); (c) shows a close-up of the interaction, highlighting the structure that is imposed on a previously unstructured peptide.

Figure 8

(a) Human recombinase Rad51 binds BRC repeats of BRCA2 in an interaction that is essential for function in recombination. Although this is usually essential for normal DNA repair, there is an advantage in disrupting recombination during radiotherapy and chemotherapy, which function through the introduction of DNA damage in cancerous cells. While Rad51 independently forms a stable globular structure, only upon interacting with Rad51 does the BRC peptide fold into a defined three dimension structure. Closer examination of the interaction (b, c) shows discrete regions of interaction that may be useful drug targets in disrupting the interaction and thus blocking recombination.

Figure 9

Protein–protein interactions can be described by three models. (a) Proteins of preformed globular structure that interact with no change to their structure through a discontinuous epitope. (b) Proteins of preformed globular structure that adapt upon interaction to form a complex of novel conformation. (c) A natively unstructured protein that folds upon interaction with another partner.