A Synthetic Nuclease Boosts CRISPR Editing Efficiency

Advancements in CRISPR-Cas9 systems have allowed scientists to make precise edits in genes, revolutionizing genetic engineering. However, current enzymes still produce off-target effects, so researchers have been exploring ways to extend the capabilities of these systems for more efficient gene editing approaches.¹

One of the ways scientists can do this is by redesigning nucleases—the enzymes that act as “molecular scissors” to nick DNA for gene editing—to be more specific. Nucleases rely on RNAs binding to specific sites on the target DNA that guide the enzymes on where to introduce cuts. However, the multidomain nature of nucleases makes it challenging for scientists to tweak them without disrupting their activity.

Now, researchers led by University of California, Berkeley biochemist Jennifer Doudna, who won the 2020 Nobel Prize for her CRISPR-related work, used an AI-based model to design a synthetic nuclease with less off-target activity. The enzyme showed activity comparable to or better than its natural counterpart. Their framework, published in Science, establishes an approach to create non-natural nucleases, expanding the scope for protein engineering to improve existing CRISPR-Cas9 tools.²

To design a synthetic nuclease, Doudna and her colleagues turned to an inverse protein-folding model that uses AI to reverse engineer protein sequences based on the desired three-dimensional structure. Using this approach in silico, the researchers redesigned TnpB, a family of CRISPR-Cas12a-like nucleases. Phylogenetic analyses revealed that the candidate redesigned TnpB sequences differed significantly from wild type (WT) TnpB sequence. However, the pipeline also introduced some mutations that would affect how the enzyme interacts with its guide RNAs and target DNA.

To prevent this, the researchers coupled the AI model with constraints on amino acid residue positions based on evolutionary information. The resulting sequences carried fewer mutations in crucial binding areas, indicating improved compatibility and potentially higher editing activity.

Doudna and her colleagues then generated full-length candidate TnpB variant proteins and screened their activity in bacterial cells. They shortlisted a few of these variants with the highest editing function to test them further in plant and animal cells.

Gene editing assays in animal and plant cells helped the researchers identify the synthetic TnpB variant (SynTnpB) with the highest editing efficiency and lowest off-target effects. To understand the mechanism behind this improved activity, Doudna and her team used cryo-electron microscopy.

Comparing the structures of WT TnpB and SynTnpB revealed that the engineered protein introduced electrostatic- and hydrogen-bonds that formed new interactions at the guide RNA-DNA interface. These findings suggest that an AI model-based and evolution-informed approach could help design better RNA-guided nucleases.