Background: Delivery vehicle choice and off-target variant assessment

In mouse muscle, CRISPR-Cas9 delivered by lipid nanoparticles (LNPs) produced a lower proportion of target-site insertions than delivery by adeno-associated virus (AAV). In human cells, however, a high frequency of off-target variants associated with a particular guide RNA (gRNA) was observed. The researchers presented a framework combining computational predictions, DNA cleavage assays, and cellular variant analysis to assess these risks.

The study focused on exon 45 of the dystrophin gene in Duchenne muscular dystrophy (DMD). The strategy uses two gRNAs to induce exon skipping and restore the protein reading frame. The LNPs encapsulated Cas9 messenger RNA (mRNA) and both gRNAs. DNA-free delivery is expected to reduce the risk of vector DNA fragment insertion. The risk of Cas9 cutting sites other than the intended target must still be assessed separately.

Key findings: Insertion analysis in mice and 11 candidates in human cells

The mouse experiment used a reporter model, with four animals per group. A dose was administered to the left gastrocnemius, followed by the same dose to the right gastrocnemius 28 days later. Both tissues were analyzed 28 days after the second administration. The proportions of reads containing insertions or deletions (indels) were 3.60% and 3.55% after the first and second LNP administrations, respectively, compared with 10.29% and 0.68% after the corresponding AAV administrations. Because repeat administration involved the contralateral muscle, these results cannot be extrapolated to repeat dosing in the same muscle or to treatment effects in patients.

The proportion of reads containing insertions was 0.51% after each LNP administration and 5.69% after the first AAV administration. Among insertions longer than 10 base pairs whose origins were traced, more than half of those observed with LNPs matched the human DMD exon 45 sequence included in the reporter. Sequences derived from Cas9 mRNA or gRNAs were reported to be barely detectable; this does not establish the absolute absence of foreign sequences.

For off-target analysis, the researchers compared 13 prediction tools. They then integrated Cas-OFFinder sequence-based predictions, CIRCLE-seq, which identifies DNA cleavage sites in vitro, and whole-genome sequencing (WGS) of two human induced pluripotent stem cell (iPSC) lines with normal karyotypes. Both lines came from the same healthy individual, and high-dose LNP treatment was used to increase detection sensitivity. Regions with three or more overlapping distinct indel patterns were defined as “indel clusters,” in an attempt to distinguish candidate editing-related variants from variants arising spontaneously during culture.

The integrated analysis identified 11 off-target candidates, 10 of which were associated with gRNA #23. Targeted amplicon sequencing found an indel frequency of 58–85% at one candidate site on chromosome 9. This is the combined frequency of insertions and deletions measured at a specific site in iPSCs treated with high-dose LNPs. It does not represent insertion frequency alone or the rate of adverse events in patients.

Implications and outlook: Preclinical safety assessment requires cross-validation

The prediction tools did not simultaneously improve sensitivity and precision. This highlights the need to distinguish and cross-check computational candidates, in vitro cleavage, and variants occurring in cells. Off-target variant patterns also differed between iPSCs and human muscle-lineage cells, and detection reproducibility across conditions was limited.

Even WGS at an average depth of about 60–80× cannot capture all low-frequency variants. In repetitive regions, sequence-alignment errors can also affect candidate calls. The LNP and AAV doses were not rigorously matched on a molar basis, so caution is required when generalizing differences between delivery vehicles.

This study provides a basis for considering delivery vehicle choice alongside gRNA-specific risk assessment. Efficacy in patients, long-term safety, and whether regulatory approval requirements are met remain subjects for further validation. The study design and findings were published in a Molecular Therapy Nucleic Acids paper.