TL;DR
This paper introduces a high-throughput "pairwise library screen" to systematically assess the specificity of Staphylococcus aureus Cas9 (SaCas9) in human cells. By creating a massive library of SaCas9 guide RNAs paired with a diverse set of on-target and off-target DNA sequences, they developed a robust model to predict SaCas9 activity and identified that 21-nucleotide spacers offer the best balance of efficiency and specificity.
What Problem Does This Paper Address?
CRISPR-Cas9 systems are powerful tools for genome editing, but their off-target activity – unintended edits at sites similar to the target – remains a significant concern, especially for therapeutic applications. While Streptococcus pyogenes Cas9 (SpCas9) has been extensively studied, the specificity of other Cas9 orthologs, like the smaller Staphylococcus aureus Cas9 (SaCas9), is less understood. SaCas9 is attractive for delivery via adeno-associated viruses (AAVs), but its precise editing profile, particularly how variations in guide RNA (sgRNA) spacer length and target DNA mismatches affect its specificity, needs thorough characterization. Existing methods for profiling off-target effects can be limited in scope or introduce their own biases.
Methods
The core innovation here is a "pairwise library screen" approach. They designed a synthetic library containing 88,692 unique sgRNA-target pairs. Each pair consists of an sgRNA and a target DNA sequence that is either perfectly matched or contains single/double mismatches, insertions, or deletions. To ensure accuracy and overcome limitations of oligonucleotide synthesis, they incorporated two types of barcodes:
- Hamming barcode: Unique to each sgRNA-target pair, allowing for identification of the original pair after editing.
- Randomized barcode (rBC): Unique to each individual molecule, enabling a "whitelisting" process. This step filters out molecules with synthesis errors present at the start (Day 0), ensuring that observed edits are due to SaCas9 activity and not pre-existing mutations.
After delivering SaCas9 and the library into HEK 293FT cells, they sequenced the DNA at different time points to measure indel formation (edits) at each target site. This allowed them to quantify the editing efficiency for each sgRNA-target pair and build a specificity model.
Datasets
The paper introduces a novel, large-scale synthetic dataset comprising 88,692 sgRNA-target pairs. This dataset was generated in silico and then synthesized, covering a wide range of potential on-target and off-target interactions. This synthetic library serves as both the training and validation dataset for their SaCas9 specificity model.
Key Insights and Ideas
- Spacer length is a critical determinant of SaCas9 specificity: The study reveals that 21-nucleotide (nt) sgRNA spacers are generally more efficient and specific, particularly against targets with single and double mismatches, compared to shorter or longer spacers.
- SaCas9 exhibits a "seed region" of low tolerance: Similar to other Cas9 enzymes, SaCas9 is less tolerant of mismatches in the PAM-proximal "seed" region (approximately the first 9 nucleotides of the spacer), with tolerance increasing towards the PAM-distal end.
- Mismatches are not equally tolerated: The identity and position of a mismatch significantly impact SaCas9 activity. 21 nt spacers show greater tolerance to double mismatches than other lengths.
- Barcoding is essential for high-fidelity screening: The dual-barcoding strategy (Hamming and rBC) is crucial for accurately measuring editing efficiency by filtering out synthesis errors and ensuring that only correctly integrated and error-free molecules are analyzed.
- A predictive model for SaCas9 specificity is achievable: The developed model, trained on the extensive pairwise library data, can accurately predict SaCas9 off-target activity and rank potential off-target sites.
Key Results
- Optimal spacer length: 21 nt spacers were found to be most efficient and specific, with 22 nt spacers offering a good balance. 23 nt spacers showed significantly lower efficiency.
- Mismatch tolerance: SaCas9 edited perfectly matched targets most efficiently. Single mismatches reduced editing by 3.4-fold on average, and double mismatches or bulges reduced it even further.
- Specificity model performance: The developed SaCas9 specificity model showed good correlation with an independent validation set (Spearman correlation of 0.75) and with previously reported GUIDE-seq data (Spearman correlation of 0.82).
- Library recovery: After filtering for error-free cassettes (whitelisting), 83% of the library remained for analysis, demonstrating the effectiveness of the barcoding strategy.
Limitations and Future Work
- Synthetic vs. Genomic: The study uses a synthetic cassette for target integration. While this allows for comprehensive testing of mismatches, it doesn't fully replicate the complexity of endogenous genomic loci, which have varying chromatin structures and regulatory elements.
- In vitro vs. In vivo: While SaCas9 is known for its in vivo applications, this study was conducted in cultured human cells. Future work could extend this approach to in vivo models.
- Model limitations: The specificity model performed less well for guides with very low on-target activity, potentially due to a low signal-to-noise ratio for these ineffective guides.
- Future Directions: The authors suggest that their barcoded pairwise library screening approach provides a general framework for understanding nucleic acid interactions and could be extended to characterize other nuclease-centric properties, such as DNA repair outcomes, and to design oligonucleotide probes or switches. The code for their specificity model will be made publicly available.
Summary updated