Custom CRISPR-Cas9 PAM variants via scalable engineering and machine learning.

Area of Science:

• Biotechnology
• Molecular Biology
• Genomics

Background:

• CRISPR-Cas9 technology enables diverse genome editing but generalist enzymes risk off-target edits.
• Engineering and characterizing proteins for specific genome editing applications is time-consuming.
• Developing safer and more efficient genome editing tools is crucial.

Purpose of the Study:

• To develop a scalable method for reprogramming Cas9 enzymes using machine learning and high-throughput protein engineering.
• To create bespoke Cas9 editors tailored for specific genomic targets, improving safety and efficacy.
• To identify novel Cas9 variants with distinct protospacer-adjacent motif (PAM) requirements.

Main Methods:

• Combined high-throughput protein engineering with machine learning to derive custom Cas9 editors.
• Utilized structure-function-informed saturation mutagenesis and bacterial selections to generate engineered SpCas9 enzymes.
• Trained a neural network (PAM machine learning algorithm - PAMmla) to predict PAM specificity based on amino acid sequence.

Main Results:

• Generated nearly 1,000 engineered SpCas9 enzymes and characterized their PAM requirements.
• Identified efficacious and specific Cas9 enzymes that outperform existing variants in human cells.
• Demonstrated reduced off-target edits and enabled allele-selective targeting using in silico-directed evolution.

Conclusions:

• PAMmla integrates machine learning and protein engineering to create a diverse catalogue of SpCas9 enzymes with unique PAM specificities.
• This approach facilitates a move from generalist to bespoke Cas9 variants for safer and more efficient genome editing.
• The developed method enables user-directed Cas9 enzyme design for precise therapeutic applications.