Structure-Based Classification of CRISPR/Cas9 Proteins: A Machine Learning Approach to Elucidating Cas9 Allostery

Area of Science:

• Biochemistry
• Molecular Biology
• Bioinformatics

Background:

• CRISPR/Cas9 gene editing relies on allosteric regulation for specificity and stability.
• Understanding these mechanisms is crucial for developing precise Cas9 variants with fewer off-target effects.

Purpose of the Study:

• To introduce a novel structure-based machine learning (ML) approach for identifying long-range allosteric networks in Cas9.
• To systematically map and refine these networks in Streptococcus pyogenes Cas9 (SpCas9) to understand its stability and specificity.

Main Methods:

• Trained an ML model on all available Cas9 structures to analyze the Cas9 structural landscape.
• Applied a two-stage SHAP feature selection strategy using Cα-Cα inter-residue distances to identify critical Lysine-Arginine (Lys-Arg) residue pairs.
• Utilized molecular dynamics (MD) simulations and mutational analysis to investigate the identified allosteric networks and an electrostatic valley.

Main Results:

• Identified 28 critical Lys-Arg residue pairs mediating SpCas9 interdomain communication, stability, and specificity.
• Discovered a hierarchical allosteric network with distinct stabilization behaviors among Lys-Arg pairs.
• Revealed an electrostatic valley crucial for SpCas9 structural integrity, with mutations destabilizing its DNA-bound conformation and impacting specificity.

Conclusions:

• Developed a novel ML framework integrating structural analysis and MD simulations to understand Cas9 allostery.
• Identified key allosteric residues and the electrostatic valley, providing a rational strategy for engineering high-fidelity Cas9 variants.
• The approach is applicable to other enzymes for understanding allosteric regulation and protein design.