Antibodies are invaluable tools not just in fighting disease but in exploring fundamental biological pathways and mechanisms. As is well known, they owe their useful properties to their very high specificities. Unfortunately these specificities can also hinder the application of an antibody from one species in the study of antigens in another species. For instance, an antibody designed for a human protein could bind much more unfavorably to a mouse ortholog of the same protein because of subtle differences in residue identity and placement.
Emerging computational design of protein-protein interfaces has the potential to make contributions in redesigning antibody surfaces to bind to orthologs. To this end, Craik, Jacobson and others from UCSF describe energetic analyses of combinations of mutations that allow an antibody for a human protein implicated in cancer to bind with high affinity to its mouse ortholog. The wild-type antibody binds about 300 times less strongly to the mouse ortholog. Simple biochemical and structural studies fail to pinpoint the reason for this reduced binding affinity, since only three residues differ between the human and mouse proteins and these don't seem to be directly involved in binding.
The authors decided to predict mutations at the antibody-protein interface that could possibly improve binding affinity for the mouse protein. They identified six protein residues in the antibody that were in proximity to the three differing residues. They then built a homology model of the mouse protein based on the crystal structure of the antibody-human protein complex and placed the mouse protein homology at the same location as the human protein in the complex.
The calculations were done with an implicit solvent model and involved optimizing side chain conformations and then minimizing the energy of the complex using molecular mechanics. Admittedly this is a relatively crude, approximate process and the authors admit that factors like changes in protein conformational entropy which are very difficult to calculate have been neglected. In order to gauge the effect of mutations, each one of the six residues was mutated to every one of the 18 other amino acid residues. Many mutations failed to show an improvement in binding affinity. However eight mutations indicated possibly improved binding to the mouse protein.
To validate these calculations, antibodies with the predicted mutations were engineered using recombinant techniques and their binding affinity for the mouse protein were measured. One mutation from a proline to a histidine was predicted to lead to a positive change in affinity. Experiments however indicated decreased binding. This was an indication of increased flexibility (and probably some entropy loss) imparted by the more flexible histidine, a factor that was not included in the calculations. However, another calculated increase in affinity was validated by experiment; this involved a threonine to arginine mutation. The arginine was predicted to form a hydrogen bond with a backbone carbonyl and a glutamate side chain, contributing about 1.6 kcal/mol of binding energy. The experiments demonstrated a 30 fold increase in binding affinity thus confirming the prediction. The structural prediction also explained why the human ortholog did not bind as well with the mutant; the glutamate there was replaced with an aspartate, whose side chain was not long enough to interact with the arginine.
The real value of any model is in prediction. This is a goal that climate modelers, financial modelers and computational chemists struggle with. The various factors that are at play are usually too complex to model and approximations are essential. It's nice when, once in a while, one of these approximations actually works. Such results provide hope for future design of protein-protein interfaces, one of the most challenging and potentially immensely useful aspects of biomedical research.
Farady, C., Sellers, B., Jacobson, M., & Craik, C. (2009). Improving the species cross-reactivity of an antibody using computational design Bioorganic & Medicinal Chemistry Letters, 19 (14), 3744-3747 DOI: 10.1016/j.bmcl.2009.05.005