Precision medicine for protein binding sites

Proteins interact with other molecules through their protein binding sites, which are functionally important regions on the protein surface. Each binding site usually binds one or a few specific molecules, the ligands (Figure 1). A ligand can be an ion, another protein, a nucleic acid, a small molecule, or a water molecule. By predicting the binding affinity of protein-ligand complexes with simulations, the efficiency of drug candidates can be assessed in silico, which is a topic that receives continued interest in drug discovery research.

Genetic variations such as sequence variants that occur in coding regions of genes may alter proteins’ amino acids and presumably affect protein function. It was found that disease-causing sequence variants are preferentially located at protein-protein interfaces rather than in noninterface regions of protein surfaces. Somatic sequence variants are known to cluster in oncogenic proteins, and in some oncoprotein-oncotarget pairs, significant enrichments of such mutations were found within protein-protein, protein-nucleic acid and protein-metal ion binding sites. As such, binding site sequence variants are of great interest to drug development chemists and clinicians who seek to predict an individual’s response to a drug, which is known as precision medicine.

Scoring the binding efficiency

To evaluate the binding affinity of a protein-ligand complex, a scoring function needs to be designed. In first instance, the ligand is rotated and translated as a (rigid) unit with respect to the (rigid) protein binding site and the interaction is measured. While this static approach gives a first guess of binding affinity, it cannot predict loss or gain of interactions reliably and accurately as the protein and ligand are known to flexibly alter their shapes when interacting. Therefore, new methods are developed to include both binding site and ligand flexibility, which allows the prediction of the structural response upon binding. The resulting complexes are more accurate, and the scoring function that measures the binding efficiency lead to more accurate protein-ligand predictions.

A set of essential coordinates is extracted that includes not only the large-scale motions associated with conformational changes of the protein, but also the local vibrations associated with local deformations of the binding site. The mobile block Hessian method, where some atoms move coherently in normal mode analysis, is used for the fast computation of these normal modes. It allows selecting rigid regions in the protein, hence reducing the computational cost. Simple elastic network models are used to compute the large-scale motions, while locally an all-atom description is required of the binding site. The entropy loss incurred by the constraints on the atoms is assessed by incorporating the vibrational free energy associated to the constraints into a thermodynamic cycle.

Figure 1:Prediction of the protein binding site, the ligand, the sequence variant, and their binding dynamics. Depicted is a small molecule ligand (cancer drug sunitinib) (grey-blue, sticks) binding to the predicted binding site (orange, balls and sticks) in human sterile20-like kinase protein (cyan, cartoon).

Precision medicine

Sequence variants can affect the binding affinity of small molecules to proteins and thus influence the individual's response to drug molecules. The ProBiS web server was developed by Konc et al. [1] which enables the detection of ligands for proteins, based on graph-theoretical matching criteria. The ProBiS algorithm was interfaced with molecular dynamics software, and extended to map sequence variants to proteins’ binding sites (GenProBiS). This will now be extended by systematic analysis of their influence on binding affinity and binding kinetics, using molecular dynamics simulations equipped with a Bayesian analysis methodology to fit trajectory data to the Smoluchowski diffusion model, using transition interface sampling and other techniques.

The developed approach will be validated on anti-cancer drugs, where the drugs sunitinib, imantinib, antibody cetuximab, and of small molecule drug warfarin will serve as test cases. Sunitinib is a cancer drug whose binding and selectivity towards protein targets is mediated by conserved waters and which has known issues in patients caused by its various modes of action due to the variability of human genome sequences: the presence and location of conserved structural water molecules in ligand binding sites will be registered and their importance for ligand binding will be assessed by determining their interactions with ligands.

The result will be novel computational solutions that will enable the prediction of pharmacodynamic and pharmacokinetic behaviour of active drug substances as a consequence of genetic variability.

IBiTech researchers currently active on the project

• Bhawna Dixit (contact)
• Wouter Vervust (contact)
• An Ghysels (contact)
• Sina Safaei (contact)

Funding sources

• Fund for Scientific Research – Flanders (FWO-Vlaanderen)

References

1. J. Konc and D. Janežič, “ProBiS-ligands: A web server for prediction of ligands by examination of protein binding sites,” Nucleic Acids Res., vol. 42, no. W1, pp. 215–220, 2014