Decoding CRISPR–Cas9 PAM recognition with UniDesign

Computational protein design (CPD) methods hold great promise in protein engineering. A critical first step in CRISPR–Cas mediated gene editing is recognizing a preferred protospacer adjacent motif (PAM) sequence on the target nucleotides by the protein’s PAM interacting amino acids (PIAAs). Such PAM requirement defines the editable sequence and the PAM specificity of each CRISPR–Cas protein. A deep understanding of the PAM recognition process through computational design methods is therefore useful in assisting CRISPR–Cas engineering to either relax or tighten the PAM requirement. Here we describe a universal CPD framework (UniDesign) suitable for modeling and designing the protein–nucleic acid interaction. As a proof of concept, we applied UniDesign to decode the PAM–PIAA interactions in eight CRISPR–Cas9 proteins. We first show that, given native PIAAs, the computationally determined PAM sequences by UniDesign are largely identical to those of known natural PAMs of all tested CRISPR–Cas9 proteins. We next subjected the natural PAM sequences of different CRISPR–Cas9 proteins to UniDesign and computationally determined the preferred PIAA residues for each, which again largely recapitulated the key residues (>70% by identity or >80% by similarity). These results demonstrate that UniDesign faithfully captures the mutual preference between the PAM and the PIAAs, thus suggest it as a useful tool for engineering CRISPR­–Cas and other proteins that interact with nucleic acids. UniDesign is open-sourced at https://github.com/tommyhuangthu/UniDesign.