The advent of CRISPR-mediated genome-editing has revolutionized the generation of genetically modified animal models. In the few short years since its introduction, CRISPR has played a pivotal role in numerous animal model studies. In the past, genetic studies in mice have primarily relied on gene knockouts and knockins made using ES cells. In contrast to the labor-intensive ES cell methods, CRISPR/Cas9 approaches can now produce knockouts, knockins or Large-fragment Knockin with less effort.
This new technology has led to a flood of activities by the research community in making CRISPR-based mouse models. Although CRISPR off-target effects have been a concern, researchers have tried to mitigate this concern by predicting potential CRISPR off-target sites based on homology to the target site, and then sequencing these potential off-target sites in CRISPR-treat mice to confirm their intactness1-3. Based on these analyses, it is generally assumed that CRISPR off-target effects are very infrequent, and therefore not a major confounding factor in the generation of mouse models.
However, a shocking new finding by Schaefer et al. has fundamentally challenged the utility of CRISPR-based animal models4. The group performed whole-genome deep sequencing on CRISPR-treated mice, and found a very large number of single nucleotide variants (SNVs), about ~1700 per mouse, caused by CRISPR. Surprisingly, most of these mutations were not indels and were not associated with sites homologous to the CRISPR gRNA. Thus, the use of in silico modeling was ineffective at predicting these mutations, many of which were in protein-coding and non-coding RNA genes. Interestingly, although most of the observed SNVs were not predicted, they did not occur randomly, as most of them were common between independently generated animals treated with the same CRISPR and gRNA.
One particularly troubling aspect of the Schaefer et al. finding is that the pattern of mutations they observed makes it very difficult to tell whether phenotypes found in a CRISPR-based mouse model are the result of the targeted mutation or off-target mutations. Traditionally, there are two approaches that researchers use to deal with potential off-target effects of genetic modifications. First, by back-crossing mutant mice to untreated, wild-type mice, unwanted mutations can be removed if they are physically far from the targeted mutation-of-interest or on different chromosomes. Second, by generating multiple independent mouse lines with the same targeted mutation, the effects of random off-target mutations can be eliminated, since they should not be present in multiple lines. Due to the large number and the non-random nature of the SNVs present in CRSIPR-treated mice, it is unlikely that either of these two strategies would be sufficient to eliminate the possibility that any observed mouse phenotype is due to an unpredicted off-target mutation. In the case of backcrossing, it cannot remove mutations that are close to the targeted mutation on the same chromosome, and these would not be rare if there are thousands of off-target mutations in each CRISPR-treated animal. Generating multiple independent mouse lines is also not helpful given that the same off-target mutations have a high probability of occurring in different lines.
The new results by Schaefer et al. seriously challenge the utility of CRISPR as a tool for generating animal models. It is vitally important that more work is done in the future to further test and improve the accuracy of CRISPR before it can be reliably used in animal model generation. In the meantime, new, rapid ES cell-based methods may be the best choice for constructing animal models. Homologous recombination methods in ES cells are highly accurate, and new proprietary approaches, such as our TurboKnockout method, but without the off-target concern.