A month ago, I started my second short project with OPIG, on antibody modelling. I had become interested in structural bioinformatics when I was working on the advanced programming project, where information was mined from a dataset to answer biological questions.
It all begins with protein folding.
Upon translation, a chain of amino acids folds into different shapes for its function. These proteins might not be at their lowest energy state – rather, they hold to a level of entropy for the metabolic reactions they are involved in (such that they can lose or gain energy to change the conformation of another protein and trigger the proceeding pathway of changes). It is therefore not reasonable to predict the conformation of protein simply by calculating their lowest entropic states.
Adding to this complexity is the antibody repertoire. Human only has one set of gene for antibody production. But how do we produce antibody that are specific to a wide range of antigens? Gene recombination of the variable-diversity-joining region initiates the diversity of the antibody repertoire. However, native antibodies produced from human germlines are found to be relatively generic as compared to the functioning antibodies. We produce a wide range of different antibodies from the germline – the one that is sufficiently “sticky” to the target antigen is recognised, then undergoes hypermutation such that the complementarity to the target antigen is improved.
Engineering proteins have shown an extent of success. We infect the production factory (desired cell line for the target protein) with carefully designed genes, express them in large quantities for medicinal use. Concerns with this method include the non-human protein folding machinery in animal or plant cells which could destroy the function of the designed protein, which renders rejection from the human body; although, as mentioned above, the lack of knowledge in protein folding means that we are unable to accurately predict whether the designed genes will be translated and folded as designed. To design new proteins, or in our case, humanised antibodies, we try to recover the decision process by which the protein folds in the observed way for its function.
We employ knowledge-based prediction of antibody conformation based on sequence information, with tools previously developed by our group, including the Structural Antibody Database, Antibody Numbering system and CDR canonical form clustering methods.
Antibodies are specific to their targets. This specificity is determined by the complementarity-determining regions (CDRs) on the heavy and light chains of the antibody. CDR or SDR (specificity-determining region) grafting techniques have been used to humanised antibodies produced in other species upon the exposure of a target antigen. CDR grafting involves replacing the CDR on a human antibody framework; whereas SDR grafting only minimally mutates residues which are contributing to the antigen-binding function. The latter method minimises the introduction of foreign materials to the human host, hence supposedly raises the compatibility. However, point mutations also incur a change in the chemical properties of the region, and owing to the size differences of amino acids, a different conformation or bending angle might be preferrable. I wondered, if the discrepancy between this point mutation in SDR grafting, and the in vivo antibody maturation which also involves point mutation, would suggest the preferrable condition in the biological space as oppose to the physical space: is physically minimalistic antibody (or nanobody) biologically stable?