To the Editor:

The clustered, regularly interspaced, short palindromic repeats (CRISPR)-Cas system is a highly efficient technology for genome editing of mouse zygotes1. Previously, we reported cotransmission of a Cas9-induced mutation in the X-linked Ar gene and an off-target mutation to offspring of founder animals from pronuclear injection of Cas9 mRNA and two single guide RNAs (sgRNAs)2. No off-target damage is observed in offspring derived from founder animals injected with Cas9 nickase mRNA and a pair of sgRNAs, showing that CRISPR-Cas specificity is improved by employing nickases2,3. Here, we used whole-genome sequencing to more thoroughly assess any damage induced by injection of Cas9 versus Cas9 nickase.

Two female founder animals (C57BL/6J × CBA) carrying biallelic deletions in Ar induced with Cas9 and Cas9 D10A nickase (animals F18 and F25, respectively) were mated to C57BL/6J males, and heterozygous Ar mutant offspring were sequenced. To control for strain-specific variants, we also sequenced a C57BL/6J and a CBA animal from our breeding colonies. Whole-genome sequencing was performed at a sufficient depth (20–25×) to detect more than 95% of heterozygous variants (Supplementary Methods).

We applied two standard computational methods to detect small insertions and deletions (indels), the hallmark of endonuclease-induced damage. We readily detected the Ar mutant alleles transmitted to each of the F1 offspring (Supplementary Fig. 1). We then looked for damage to related sites in the genome with up to five mismatches. Of 8,441 possible off-target sites (Table 1), damage was observed only in offspring of the Cas9-treated founder animal at OTS3, a bona fide off-target site for one of the two Ar sgRNAs2 (Supplementary Fig. 2).

Table 1 Summary of variants detected by whole-genome sequencing

We extended our analysis to the remainder of the genome on the basis of the following assumptions. We reasoned that the variants would not be shared with the control inbred strains (C57BL/6J or CBA) and would not be shared in offspring from individual founder animals. Variants associated with repetitive DNA sequence were discarded as these cannot be distinguished from the large number of naturally occurring variants in repetitive DNA sequence in individual mice. These filtering steps (Supplementary Tables 1 and 2) reduced the number of indels to a small set of 120 high-quality variants (Supplementary Data). Twenty-four variants were selected at random, of which 22 were confirmed by experimental validation.

On the basis of the ungapped and gapped alignments of the two Ar sgRNAs to genomic sequence at the variant sites, none of the 120 variants appeared to be a true off-target site (Supplementary Fig. 3). We conclude that these variants arose spontaneously and were not the result of unconstrained Cas9 endonuclease activity. Our study is consistent with three recent reports showing negligible genome-wide damage in Cas9-engineered human induced pluripotent stem cells4,5,6.

In contrast to embryonic stem cell technology, where extensive genetic variation arises in culture, undesired mutations induced by the Cas9 endonuclease will be rare in zygotes. Because unlinked mutations will segregate away through breeding, phenotyping of two independent founder animals would be sufficient to establish causality. Alternatively, mutant founder animals generated with two unrelated guide RNAs would rigorously control for any confounding alleles.

Author contributions

V.I. and B.S. contributed equally to this work. V.I., A.H. and T.K. performed the computational analysis of whole genome sequence. B.S. and W.Z. experimentally validated the variants in mice. X.H. and W.C.S. wrote the manuscript.