Proc.NatlAcad.Sci.USAhttp://dx.doi.org/10.1073/pnas.1321126111(2014)

Credit: NATIONAL ACADEMY OF SCIENCES

Preventing a response from the immune system is one of the main problems hindering the development of protein pharmaceuticals. Such a response is caused by cells in the immune system binding to parts of the protein (known as epitopes) and thereby recognizing that the protein is not naturally produced by the body. This can lead to the production of antibodies that bind to the protein, reducing or eliminating its therapeutic effect. Introducing mutations into epitopes is one method of preventing the immune system from recognizing the foreign nature of the protein and thus reducing the immune response. However, introducing enough mutations into the protein's amino acid sequence whilst retaining its overall structure, stability, biological function — and clinical effectiveness — is extremely difficult.

A team led by Chris King at the University of Washington, has now integrated a machine-learning approach with computational protein re-design techniques to enable the removal of epitopes from two proteins. The process involves identifying the epitopes contained within a protein's sequence using computational predictions that include data from binding experiments. The protein's sequence is then improved using a function that introduces mutations whilst retaining the overall structure and penalising the inclusion of sequences known, or calculated to be, epitopes. Mutations in the most stable structures are then combined in a stepwise manner to generate optimized variants. This approach was found to introduce mutations that were similar to previous attempts to de-immunize proteins that used alternative experimental techniques.

To demonstrate the potential of their approach the team attempted to remove three previously identified epitope regions from the toxin domain of a cancer therapeutic called HA22. This therapeutic has been shown to be effective for treating leukemia; however, it can generate an immune response that neutralizes its effectiveness. King and colleagues succeeded in developing several variants that retained similar activity to the original protein. The two most active variants were then tested in patients and produced a significantly reduced immune response.