Sequence Display generates large-scale sequence–activity datasets for reprogramming SlugCas9 toward difficult-to-access PAMs

Protospacer adjacent motif (PAM) requirements constrain the targeting scope of CRISPR-Cas9 systems and limit genome editing applications that require precise nuclease positioning. Although broad-PAM and near-PAMless Cas9 variants have expanded the editable genome, robust targeting across certain PAM families remains difficult, particularly at NNAN PAMs. Here, we apply Sequence Display to generate multiplexed, large-scale sequence–activity datasets for Staphylococcus lugdunensis Cas9 (SlugCas9) across the NNAN PAM family, including NNAA, NNAT, NNAC and NNAG. These datasets quantitatively resolve how individual and combinatorial mutations reshape nuclease activity across distinct PAM contexts. By integrating Sequence Display-derived datasets with protein language models, we construct fine-grained activity landscapes and identify SlugCas9 variants with enhanced and broadened activity across NNAN PAMs. The resulting variants reprogram PAM recognition beyond the native G-compatible landscape of SlugCas9 and enable improved base editing at selected endogenous NNAN-constrained sites in cultured mammalian cells, including disease-relevant sites in APOC3 and PCSK9. Together, our results establish a strategy for reprogramming PAM preference in compact Cas9 nucleases and highlight Sequence Display as a general platform for evolving genome editors toward specific and difficult-to-access PAM spaces. Compact CRISPR-Cas9 enzymes are easier to deliver but recognize few targets. Here, authors use Sequence Display to generate large-scale SlugCas9 sequence–activity datasets and apply protein language models to engineer variants that recognize difficult NNAN PAMs and edit selected human genomic sites.