(270d) Computationally Driven Deimmunization of Therapeutic Proteins

Novel biotherapeutics have reshaped drug discovery and promise a continued revolution in disease therapy, yet therapeutic proteins present unique design challenges for the biomolecular engineer. One distinguishing risk factor is the prospect of eliciting an immune response upon administration to human patients. An anti-drug immune response can compromise therapeutic efficacy and even threaten patient safety. In the case of therapeutic antibodies, codified humanization technologies can mitigate much of the risk associated with immune reactions. These grafting-based deimmunization strategies are not, however, broadly applicable to other protein classes, such as therapeutic enzymes.  Our capacity to effectively tap the full diversity of non-immunoglobulin therapeutic candidates will require more broadly applicable protein deimmunization technologies. To help meet this challenge, we have developed optimization algorithms that seamlessly integrate computational prediction of T cell epitopes and bioinformatics-based assessment of the structural and functional consequences of epitope-deleting mutations. Our methods address the fact that biotherapeutic deimmunization represents a dual-objective protein design space with tradeoffs between immunogenicity and stability or activity.

Here, we describe the experimental validation of our algorithms using the Enterobacter cloacae P99 β-lactamase, a component of Antibody Directed Enzyme Prodrug Therapies. We describe a high-throughput immunoassay for peptide-MHC II interactions, and we analyze the robustness of our algorithm in two complementary studies.  In aggregate, we have constructed a large panel of engineered enzymes and analyzed their thermal stability, catalytic activity, constituent peptide affinity towards human MHC-II immune proteins, and, in one case, relative immunogenicity in humanized mice.  The results demonstrate our algorithm’s predictive power and in particular a robust capacity to generate highly active yet immunoevasive therapeutic candidates.  These novel computational methods can be applied to virtually any protein target, and we anticipate that they could speed biotherapeutic development and mitigate the risk of immunogenicity-related complications.