Optimization of C-to-G base editors with sequence context preference predictable by machine learning methods.

Area of Science:

• Molecular Biology
• Gene Editing Technologies
• Bioinformatics

Background:

• Efficient and precise C-to-G base editing is crucial for genetic research and therapeutic applications.
• The sequence context influencing C-to-G base editor (CGBE) efficiency and specificity remains poorly understood.
• Existing CGBEs often lack predictable editing outcomes, limiting their broader application.

Purpose of the Study:

• To engineer novel C-to-G base editors (CGBEs) with enhanced efficiency and fidelity.
• To elucidate the sequence context governing CGBE editing outcomes.
• To develop predictive models for CGBE activity using machine learning.

Main Methods:

• Engineering of C-to-G base editors (OPTI-CGBEs) by modifying uracil-DNA glycosylase and deaminase components and codon optimization.
• Determination of motif preference for OPTI-CGBEs across 100 endogenous sites in HEK293T cells.
• Development of a deep-learning model using a large sgRNA library (41,388 sequences) to predict editing outcomes based on sequence context.

Main Results:

• Engineered OPTI-CGBEs demonstrate high efficiency and fidelity for C-to-G transversion.
• A deep-learning model accurately predicts OPTI-CGBE editing outcomes for targeted sites based on sequence context.
• OPTI-CGBEs successfully perform efficient base editing in mouse embryos, generating edited offspring.

Conclusions:

• The developed OPTI-CGBEs offer a powerful tool for efficient and precise C-to-G base editing.
• Predictive machine-learning models enhance the reliability and applicability of CGBEs by accounting for sequence context.
• These engineered CGBEs hold significant potential for genetic applications, including in vivo editing and therapeutic development.