EpiSweep: Computationally Driven Reengineering of Therapeutic Proteins to Reduce Immunogenicity While Maintaining Function

Abstract

Therapeutic proteins are yielding ever more advanced and efficacious new drugs, but the biological origins of these highly effective therapeutics renders them subject to immune surveillance within the patient’s body. When recognized by the immune system as a foreign agent, protein drugs elicit a coordinated response that can manifest a range of clinical complications including rapid drug clearance, loss of functionality and efficacy, delayed infusion-like allergic reactions, more serious anaphylactic shock, and even induced auto-immunity. It is thus often necessary to deimmunize an exogenous protein in order to enable its clinical application; critically, the deimmunization process must also maintain the desired therapeutic activity. To meet the growing need for effective, efficient, and broadly applicable protein deimmunization technologies, we have developed the EpiSweep suite of protein design algorithms. EpiSweep seamlessly integrates computational prediction of immunogenic T cell epitopes with sequence- or structure- based assessment of the impacts of mutations on protein stability and function, in order to select combinations of mutations that make Pareto optimal trade-offs between the competing goals of low immunogenicity and high-level function. The methods are applicable both to the design of individual functionally deimmunized variants as well as the design of combinatorial libraries enriched in functionally deimmunized variants. After validating EpiSweep in a series of retrospective case studies providing comparisons to conventional approaches to T cell epitope deletion, we have experimentally demonstrated it to be highly effective in prospective application to deimmunization of a number of different therapeutic candidates. We conclude that our broadly applicable computational protein design algorithms guide the engineer towards the most promising deimmunized therapeutic candidates, and thereby have the potential to accelerate development of new protein drugs by shortening time frames and improving hit rates.

Computationally-driven identification of antibody epitopes

“…Ab–Ag docking models were generated by the ClusPro server ( Brenke et al, 2012 ; Comeau et al, 2004a ), which in a recent Ab–Ag docking benchmark ( Brenke et al, 2012 ) demonstrated a near-native docking model among its top 30 predictions in 95% of test cases. For each model, site-directed mutagenesis based Ag variants were computationally designed ( Choi et al, 2017 ; Parker et al, 2013 ) to disrupt Ab binding, as evaluated by a sequence potential ( Pons et al, 2011 ), while maintaining Ag stability, as evaluated by molecular mechanics modeling ( Gainza et al, 2013 ; Pearlman et al, 1995 ). These two properties were balanced in a Pareto optimal fashion ( He et al, 2015 ), with the goal of ensuring that variants still express and fold similarly to the wild type protein, thereby enabling confident interpretation of Ab binding results.…”

Engineering Stability, Viscosity, and Immunogenicity of Antibodies by Computational Design

“…[231][232][233][234][235] The integrated methods developed by Bailey-Kellogg, Griswold, and coworkers are available upon request through the corresponding authors as the EpiSweep package. 199…”

Development of a strategy and computational application to select candidate protein analogues with reduced HLA binding and immunogenicity

“…Previous studies have reported different strategies to generate amino acid substitutions for the purpose of reducing immunogenicity (for recent studies see refs ). In several studies, alanine replacements have been used (see refs ).…”

“…Other de‐immunization studies also based on introducing single amino acid substitutions into known epitopes have been reported in the literature (see, for example, refs , as cited above). In those studies, mild and conservative substitution strategies, such as ‘alanine scans’, were shown to be effective.…”