Abstract

Despite the power of CRISPR in genome editing, its clinical application is limited by off-target effects; these effects are currently difficult to evaluate at the genome level but are likely to involve chromatin context. Here, we developed the Endogenous Genome-wide Off-target Library Detection (EGOLD) method for high-throughput detection of off-target effects and identification of chromatin context bias in gene editor evaluation. Applying EGOLD to define the off-target characteristics of 17 base-editing tools revealed 2,145,592 total off-targets, with 1236–618,774 events detected per tool. The frequency of off-targets of CRISPR/Cas9 and derivative base editors ranged from 40% to 80% and were strongly influenced by the chromatin context. Using a large-scale endogenous off-target dataset with strict target site conditions to exclude the influence of sequence context, we found that off-target effects occurred in open chromatin genomic regions at a significantly greater frequency than in closed chromatin regions. The incorporation of EGOLD-Seq off-target chromatin context data to train machine learning-based models of gene editor activity substantially improved off-target prediction accuracy. These findings and the accompanying toolkit can guide mechanistic research and the development of safe and precise CRISPR-based tools.