The development and application of genome editing tools has accelerated in recent years. However, their widespread application, especially in the medical field, is delayed for various issues with the safety of the tools being one of the top concerns. Much of the detail of how proteins find their target sites and how they cleave at the sites remains unclear. Despite much progress in lab environments, where the tolerance for imprecise cutting is relatively high, in vivo treatment remains difficult.

Most genome editing tools are derived from naturally occurring regulatory proteins. Better understanding of the mechanisms used by these natural proteins should facilitate the use of genome editing tools. In nature, gene regulation is usually sparked by a change in the cellular environment, such as viruses invading a bacterium. Restriction enzymes defend the host bacteria by recognising specific sites on the viral DNA and cutting invading viruses at those sites. We aim first to understand how these proteins translocate along the viral DNA molecules toward their recognition sites, and then to see how their accuracy of cleavage can be increased.

Protein translocation along DNA molecules has been studied for more than 40 years. Atomic force microscopy in fluid mode allows direct observation of protein/DNA interactions. This application is about twenty years old but most of the images taken, lack the spatial and temporal resolution for quantitative studies of protein translocation dynamics. Here we achieve second-level and nanometre-scale tracking of the translocation of EcoRV, a Type IIP restriction enzyme, using fast-scan atomic force microscopy (AFM) and DNA origami techniques. We find that EcoRV tends to jump toward its recognition site from afar and then switch to slow sliding mode when it is within about 20 base pairs of its recognition site.

Our methods demonstrate how BcgI, a type IIB restriction enzyme, brings together two recognition sites both in cis and in trans before cleavage, minimising mis-cleavage at a single recognition site. We show that the collision looping model is valid but not the sliding model. The two restriction enzymes were chosen as they represent different model systems of typical restriction enzymes.

Our methods will be useful in studies of other types of restriction enzymes and other proteins or protein complexes that interact with DNA. We expect that these methods will see broader applications in studies of protein-DNA interactions. We also hope that our studies will contribute to the safe application of the genome-editing tools in medical contexts.