The promise of machine learning: using Seismic’s IMPACT platform to design IgG cleaving enzymes for chronic treatment of autoimmune diseases

Abstract

Abstract Proteases derived from human pathogens can specifically cleave IgG into F(ab′)2 and Fc fragments. This unique trait suggests a novel opportunity to use these molecules to treat auto antibody mediated disease. IdeS, an IgG cleaving enzyme derived from Streptococcus pyogenes has shown clinical proof of concept and is approved for use before kidney transplant. Due to the immunogenic nature of these proteases, the dosing regimen is impacted by pre-existing antibodies and the induction of anti-drug antibodies after dosing. To mitigate the impact of the immune system on our novel enzyme, we employed our IMPACT platform leveraging machine learning to reduce T and B cell epitopes and to ensure that our molecule exhibits desirable drug like properties while maintaining enzymatic activity. To extend the pharmacokinetics (PK) of our molecule, we fused it with a Fc domain. To evaluate IgG protease PK and pharmacodynamics (PD), intravenous immunoglobulin (IVIg) was injected at different time points after protease treatment and IgG levels were quantified by MSD. As expected, the addition of the Fc increased the molecule’s half-life that resulted in a PD effect at later time points than observed with a control enzyme without a Fc. Taken together, our IMPACT platform leveraging machine learning demonstrates that we can optimize drug-like properties and reduce the immunogenicity of a molecule while maintaining function, resulting in a potential novel treatment for chronic autoimmune diseases.