High-fidelity KKH variant of Staphylococcus aureus Cas9 nucleases with improved base mismatch discrimination.

The Cas9 nuclease from Staphylococcus aureus (SaCas9) holds great potential for use in gene therapy, and variants with increased fidelity have been engineered. However, we find that existing variants have not reached the greatest accuracy to discriminate base mismatches and exhibited much reduced activity when their mutations were grafted onto the KKH mutant of SaCas9 for editing an expanded set of DNA targets. We performed structure-guided combinatorial mutagenesis to re-engineer KKH-SaCas9 with enhanced accuracy. We uncover that introducing a Y239H mutation on KKH-SaCas9's REC domain substantially reduces off-target edits while retaining high on-target activity when added to a set of mutations on REC and RuvC domains that lessen its interactions with the target DNA strand. The Y239H mutation is modelled to have removed an interaction from the REC domain with the guide RNA backbone in the guide RNA-DNA heteroduplex structure. We further confirmed the greatly improved genome-wide editing accuracy and single-base mismatch discrimination of our engineered variants, named KKH-SaCas9-SAV1 and SAV2, in human cells. In addition to generating broadly useful KKH-SaCas9 variants with unprecedented accuracy, our findings demonstrate the feasibility for multi-domain combinatorial mutagenesis on SaCas9's DNA- and guide RNA- interacting residues to optimize its editing fidelity.

Accurate top protein variant discovery via low-N pick-and-validate machine learning.

A strategy to obtain the greatest number of best-performing variants with least amount of experimental effort over the vast combinatorial mutational landscape would have enormous utility in boosting resource producibility for protein engineering. Toward this goal, we present a simple and effective machine learning-based strategy that outperforms other state-of-the-art methods. Our strategy integrates zero-shot prediction and multi-round sampling to direct active learning via experimenting with only a few predicted top variants. We find that four rounds of low-N pick-and-validate sampling of 12 variants for machine learning yielded the best accuracy of up to 92.6% in selecting the true top 1% variants in combinatorial mutant libraries, whereas two rounds of 24 variants can also be used. We demonstrate our strategy in successfully discovering high-performance protein variants from diverse families including the CRISPR-based genome editors, supporting its generalizable application for solving protein engineering tasks. A record of this paper's transparent peer review process is included in the supplemental information.

Machine learning-coupled combinatorial mutagenesis enables resource-efficient engineering of CRISPR-Cas9 genome editor activities.

The genome-editing Cas9 protein uses multiple amino-acid residues to bind the target DNA. Considering only the residues in proximity to the target DNA as potential sites to optimise Cas9's activity, the number of combinatorial variants to screen through is too massive for a wet-lab experiment. Here we generate and cross-validate ten in silico and experimental datasets of multi-domain combinatorial mutagenesis libraries for Cas9 engineering, and demonstrate that a machine learning-coupled engineering approach reduces the experimental screening burden by as high as 95% while enriching top-performing variants by ∼7.5-fold in comparison to the null model. Using this approach and followed by structure-guided engineering, we identify the N888R/A889Q variant conferring increased editing activity on the protospacer adjacent motif-relaxed KKH variant of Cas9 nuclease from Staphylococcus aureus (KKH-SaCas9) and its derived base editor in human cells. Our work validates a readily applicable workflow to enable resource-efficient high-throughput engineering of genome editor's activity.