A kinetic model predicts SpCas9 activity, improves off-target classification, and reveals the physical basis of targeting fidelity

Cas9 off-target sites can be predicted by many bioinformatics tools. Here the authors present low complexity mechanistic model that characterizes SpCas9 kinetics in free-energy terms, allowing quantitative prediction of off-target activity in bulk-biochemistry, single molecule, and whole-genome profiling experiments. The S. pyogenes (Sp) Cas9 endonuclease is an important gene-editing tool. SpCas9 is directed to target sites based on complementarity to a complexed single-guide RNA (sgRNA). However, SpCas9-sgRNA also binds and cleaves genomic off-targets with only partial complementarity. To date, we lack the ability to predict cleavage and binding activity quantitatively, and rely on binary classification schemes to identify strong off-targets. We report a quantitative kinetic model that captures the SpCas9-mediated strand-replacement reaction in free-energy terms. The model predicts binding and cleavage activity as a function of time, target, and experimental conditions. Trained and validated on high-throughput bulk-biochemical data, our model predicts the intermediate R-loop state recently observed in single-molecule experiments, as well as the associated conversion rates. Finally, we show that our quantitative activity predictor can be reduced to a binary off-target classifier that outperforms the established state-of-the-art. Our approach is extensible, and can characterize any CRISPR-Cas nuclease - benchmarking natural and future high-fidelity variants against SpCas9; elucidating determinants of CRISPR fidelity; and revealing pathways to increased specificity and efficiency in engineered systems.

Automated summary

The authors present a quantitative kinetic model that characterizes the kinetics of SpCas9 in terms of free energy, allowing for the prediction of off-target activity in various experiments. The model is trained and validated on high-throughput bulk-biochemical data and can predict binding and cleavage activity as a function of time, target, and experimental conditions. It outperforms existing classification schemes in predicting off-target activity.