After four weeks of Mathematical Biology, I decided to switch to a chemistry course to explore something drastically different. In Structural Biology, modelling takes a different approach. In the previous module of Structural Biology, we had been exposed to the experimental methods and a tiny bit of physics behind solving for the crystal structures of proteins using X-ray crystallography, and how the structures are realised computationally. This module then provides the subsequent development in terms of using these crystal structures (or using the pharmacophore structures) to aid drug design. The overall idea is to predict the way proteins interact.

On the first day, we had a lecture on protein-ligand binding with a practical on virtual screening of small molecules based on their crystal structures docking onto a target protein. We evaluated the results of docking by its ligand efficiency – the ratio of the binding energy to the count of heavy atoms in the ligand. The docking procedures were run on AutoDock, a software to which our course lecturer is closely affiliated. We also looked at the relaxed complex scheme – which takes into account the flexibility of protein during docking. Instead of docking one ligand into one protein, we dock the ligand into many structures of protein from molecular dynamic simulation. We then compute the average of the binding energy spectrum for the ligand, to assess for the lowest binding energy conformation, that should suggest the most likely binding conformation.

The second day was on antigen-antibody docking, using ZDOCK. On an antibody, there are 6 Complementarity Determining Regions (CDRs) – 3 on each of the heavy and light chains. These loops define the specificity of the antibody. When a paratope (antigen-binding area of the antibody) is found against an antigen of interest, the antibody is usually further engineered to improve bioavailability and binding affinity to the antigen. Our task is to assess whether certain mutations on the antibody have improved/deteriorated the binding to antigen. The rationale behind this is the computational modelling takes significantly less time than synthesising and expressing the mutated antibody. Therefore we used the ABodyBuilder (a collation of some hard work done by the Oxford Protein Informatics Group) to construct the predicted structure of the mutated antibody.

 

Docking then takes place to assess the impact of the flexibility of protein chains on the binding affinity changes upon mutation. We did not have the structure of the mutant antibody, so we could have possibly discounted some structural features which facilitate/hinder binding. Curating decoys with the lowest binding energy (ie. highest likelihood of true binding) is expected to account for that. In this case we curated decoys and found that their binding modes did not really resemble the true binding mode (only 2% of contact surface overlap!). In fact this has been a question which has lingered around for awhile – motivating CAPRI, a “competition” to find the best protein-protein docking methodology.

On Thursday we had a visit to the Diamond Light Source just outside of Oxford. The scientific principles of the synchrotron at the Diamond are the same as the one at CERN, but this synchrotron is used in a different way: while CERN tries to get rid of the energy released from electrons travelling at a high speed, the DLS deliberately use this light emission to “fire” at structures, ranging from a Rolls-Royce aircraft blade, to the a small protein crystal, in order to obtain their structures. It was a really cool trip, especially for the fact that the synchrotron was shut down for maintenance and that we could access a lot of the “unsafe” areas of the inner ring of the synchrotron!

Finally, on Friday we had a presentation day on the methodology and success/failure of structure-based drug discovery. It was really fun looking at the chemistry (of small molecules/ligand) and how they fit into the pocket on the target protein in 3D. Admittedly my organic chemistry knowledge was constrained to slightly beyond A-level – yes I do know about curly arrows and stereochemistry etc. Yet visualising the interactions between the organic molecules and the macromolecules (which usually arrange in a way such that there are various gating mechanisms for the activation and deactivation of the protein functions) on PyMOL and learning how creative people can be after they have found the pharmacophore structure and decide to adapt the structure to cater for the bioavailability have been truly amazing to know. The 2D chemistry feels more realistic now with the drug design processes and how drugs actually affect the conformation and thus function of the target (and sometimes non-target) proteins. It feels surreal that the biological system has been tightly assembled to maintain the function of the living organism (whoops, still a little bit of me from a physiological perspective).

All in all, it was tough to begin with, but the practicals and assignments had been thorough enough such that I acquired the knowledge I need to proceed with modelling antibody in the future (which could read, the minimum amount of chemistry I would need before I can set up and interpret the results from running several scripts in the black box of docking softwares…)

Article of interest: Zheng, H., Hou, J., Zimmerman, M.D., Wlodawer, A. and Minor, W., 2014. The future of crystallography in drug discovery. Expert opinion on drug discovery, 9(2), pp.125-137.