Search

Approximately one third of proteins are thought to bind metal ions. These ions often improve the protein’s structural stability or impart functions such as conformational changes or catalytic activity. Due to their prevalence and functional roles, metal binding sites are attractive targets for protein design. Zinc binding sites are among the most common design targets due to their well-defined geometry and wide range of functions. Previous studies have successfully designed zinc binding sites by mutating residues in native proteins or by designing symmetric de novo scaffolds with zinc-coordinating residues; however, achieving atomic-level accuracy in modelling these sites remains a challenge. Here, we explore possible reasons for this continued challenge and introduce novel approaches for metalloprotein design. We first examine two crystal structures of metalloproteins designed by introducing coordinating residues into native scaffolds. We find that one of these sites shows significant changes in backbone conformation versus the predicted structure to accommodate the formation of second-shell hydrogen bonds, and the second binding site fails to form in the absence of such interactions. When these interactions are explicitly modelled, we find evidence of zinc binding at the modeled site. We next describe the development of a novel method for metalloprotein design based on the SEWING method for de novo backbone formation. This method, which we call Ligand SEWING, begins with a partially-coordinated metal binding site from a native protein and appends structural motifs to complete metal coordination and extend the protein backbone. Using this method, we design four proteins that are monomeric and bind zinc with high nanomolar to low micromolar affinity. During the design process, we note that designs are more successful when they do not model the charge of the zinc ion, indicating deficits in modelling binding site electrostatics (most likely due to not distributing the charge over coordinating residues). While accurate modelling of these interactions will require scoring modifications and possibly improved computational power, they may be partially circumvented by ignoring the ion charge and manually modelling second-shell interactions. Alternatively, high-throughput design and screening methods may allow functional design without atomic-level accuracy in modeling.