Engineering affinity of humanized ScFv targeting CD147 antibody: A combined approach of mCSM-AB2 and molecular dynamics simulations

Abstract

This study aims to assess the effectiveness of mCSM-AB2, a graph-based signature machine learning method, for affinity engineering of the humanized single-chain Fv anti-CD147 (HuScFvM6-1B9). In parallel, molecular dynamics (MD) simulations were used to gain valuable insights into the dynamics and affinity of the HuScFvM6-1B9-CD147 complex. The result analysis involved integrating free energy changes calculated from the mCSM-AB2 with binding free energy predictions from MD simulations. The simulated structures of the modified HuScFvM6-1B9-CD147 domain 1 complex from MD simulations were used to highlight critical residues participating in the binding surface. Interestingly, alterations in the pattern of amino acids of HuScFvM6-1B9 at the complementarity determining regions interacting with the 31EDLGS35 epitope were observed, particularly in mutants that lost binding activity. The predicted mutants of HuScFvM6-1B9 were subsequently engineered and expressed in E. coli for subsequent binding property validation. Compared to WT HuScFvM6-1B9, the mutant HuScFvM6-1B9^L1:N32Y exhibited a 1.66-fold increase in binding affinity, with a K_D of 1.75 × 10⁻⁸ M. While mCSM-AB2 demonstrates insignificant improvement in predicting binding affinity enhancements, it excels at predicting negative effects, aligning well with experimental validation. In addition to binding free energies, total entropy was considered to explain the discrepancy between mCSM-AB2 predictions and experimental results. This study provides guidelines and identifies the limitations of mCSM-AB2 and MD simulations in antibody engineering.