Affinity maturation of a single-domain antibody via rational mutagenesis informed by artificial intelligence

Abstract

Antibodies can originate from various sources such as animal immunizations, patient material, DNA/RNA libraries and (AI-based) de novo design. Pre-made (e.g. synthetic) DNA libraries have substantial advantages over traditional immunization campaigns in terms of speed, cost, animal welfare, amount of required antigen and the possibility to select against otherwise problematic (e.g. toxic, unstable or expensive) antigens. The main downside is that selections from these non-immune libraries often result in binders with relatively low affinity due to the absence of in vivo affinity maturation. Now that in silico structure prediction tools are coming of age, the resulting models could conceivably be used to rationally design mutations, e.g. to enhance binding affinity. Here we used a synthetic single-domain antibody (sdAb) library to select new binders against interleukin-12 (IL-12). One of the selected candidates, 2F3, was predicted to bind to the p40 subunit of IL-12 in a manner that would compete with binding of the receptor protein IL-12R{beta}1. Intriguingly, this epitope is also targeted by the clinically approved antibodies Briakinumab and Ustekinumab, which are used to treat a variety of inflammation disorders. The predicted structure of the IL-12/sdAb complex was used to rationally design eleven point mutations, of which three showed to enhance the binding affinity. By combining these mutations, we could enhance the affinity by eightfold compared to the original lead clone. This shows that correctly (AI-)predicted protein structural models can be used for rational mutagenesis to enhance binding affinity to relevant drug targets.