CRISPR-Cas9 genome editing has emerged as a powerful approach for treating genetic disorders such as Duchenne muscular dystrophy, a severe condition caused by mutations in the DMD gene that encodes the dystrophin protein. By repairing the genetic instructions needed to make dystrophin, genome editing offers the possibility of long-term therapeutic benefit. However, ensuring its safety remains a critical challenge because CRISPR permanently alters DNA, particularly due to unintended off-target mutations that could affect other genes and potentially lead to detrimental consequences.

One key factor influencing both efficiency and safety is the method used to deliver CRISPR components into cells. Viral vectors such as adeno-associated viruses (AAVs) have been widely used, but their DNA-based nature raises concerns about prolonged persistence and unintended genomic alterations. LNPs, which deliver CRISPR components as RNA, offer an alternative that may reduce these risks. Despite this promise, systematic evaluation of the mutational outcomes associated with LNP delivery has been limited.

To address this gap, the researchers developed a multi-layered approach to assess genome-editing outcomes with high sensitivity and specificity. They first analyzed on-target editing in a mouse model using a dual-guide RNA strategy targeting the DMD gene. Compared with AAV delivery, LNP-mediated editing resulted in fewer insertion events and showed no detectable integration of vector-derived sequences at the target site. Notably, LNP delivery maintained consistent editing efficiency even after repeated administration, suggesting lower immunogenicity and improved suitability for therapeutic use.

The research team then systematically evaluated 13 widely used computational tools for predicting off-target sites. While some tools successfully identified many potential sites, they also generated large numbers of false positives, highlighting a trade-off between sensitivity and precision. To refine predictions, they incorporated experimental cleavage data obtained from in vitro assays, improving confidence in candidate off-target locations.

To determine whether these candidate sites were actually mutated in cells, the researchers conducted high-depth whole-genome sequencing in human iPS cells, which provide a genetically stable and physiologically relevant model. They also introduced a novel "indel cluster" method to detect clusters of insertions and deletions characteristic of genome-editing activity, thus enabling these events to be distinguished from random background mutations. This approach enabled the identification of a small number of high-confidence off-target sites across the genome.

By integrating computational predictions, in vitro cleavage mapping, and in-cell mutation analysis, the research team identified only a limited set of candidate off-target events, most of which were associated with a single guide RNA and frequently located in repetitive or difficult-to-map genomic regions. Further analysis indicated that many apparent signals in such regions likely reflect technical artifacts rather than true editing events. Importantly, only a few sites overlapped with genes, and those showed minimal evidence of functional impact, supporting the overall safety profile of LNP-based delivery.

This study establishes a practical and scalable framework for evaluating genome editing safety, highlighting the importance of integrating complementary methods to achieve both sensitivity and specificity. By demonstrating the advantages of LNP delivery and providing a robust strategy for off-target assessment, this work lays the foundation for more reliable preclinical evaluation of genome-editing therapies and supports their continued development toward clinical application.

• doi: 10.1016/j.omtn.2026.102958