PerspectiveGenome Editing

When genome editing goes off-target

See allHide authors and affiliations

Science  19 Apr 2019:
Vol. 364, Issue 6437, pp. 234-236
DOI: 10.1126/science.aax1827

Editing DNA in eukaryotic cells with CRISPR-based systems has revolutionized the genome engineering field. Cas (CRISPR-associated) endonucleases are directed to a particular location in the genome by a short guide RNA, providing an easily programmable strategy to target any section of DNA. As of now, two CRISPR-based approaches can introduce targeted, permanent edits. DNA cleavage with the Cas endonuclease facilitates small insertions or deletions of nucleotides that can disable the targeted gene (1). A second modified “base editor” system can generate precise single-base mutations in the targeted DNA (2). For both approaches, it is imperative that DNA modifications are made in the intended region (“on-target”) and not elsewhere in the genome (“off-target”). On pages 286, 289, and 292 of this issue, Wienert et al. (3), Zuo et al. (4), and Jin et al. (5), respectively, describe methods that identify off-target activities, which will be invaluable in therapeutic contexts as well as for stringent evaluation of future iterations of gene-editing tools.

The specificity of gene editing tools is critical to their utility, which is why off-target potential is a major concern. For therapeutic applications, unintended mutations introduced in a patient's DNA could permanently disrupt normal gene function and lead to unpredictable complications. CRISPR tools can also generate a variety of engineered cell lines and animal and plant models for research purposes. The data generated with these cellular and organismal model systems depend on the specificity of the DNA-editing tool because off-target mutations can confound experimental results. As a consequence, much research has gone into identifying and minimizing potential off-target sites of Cas activity.

Traditional CRISPR-based genome editing introduces double-strand breaks in DNA using a catalytically active Cas enzyme. This break can be corrected by an error-prone nonhomologous end-joining process in which DNA bases are randomly inserted or deleted at the target site. For base editing, a nucleotide deaminase is fused to a catalytically impaired Cas enzyme. This tool does not generate double-strand breaks in DNA but instead uses the deaminase to convert one nucleotide to another in the targeted region. In both cases, it is critical to have an appropriate method to identify off-target activity for the given tool.

Early work profiling off-targets of Cas nucleases used computational methods to predict genomic sites likely to be cleaved by a particular guide RNA based on sequence similarity (6). However, such in silico investigations are limited because they only experimentally validate selected regions for unintended mutations. An ideal off-target detection platform should be unbiased and examine the entire genome. Newer in vitro approaches look for sites of DNA cleavage upon incubating Cas9, guide RNA, and purified genomic DNA (7, 8). Although highly sensitive, these methods do not account for cellular properties that present potential obstacles to accessing DNA such as chromatin and nuclear architecture. By contrast, in vivo experimental methods deliver Cas9 and guide RNA to living cells to identify resulting off-target events in a particular cellular context (9, 10). However, these approaches can have lower sensitivity and need additional components, which can be difficult to deliver, limiting application to many samples.

Wienert et al. developed a new approach to identify off-target cleavage by Cas endonucleases in cells and tissues. The technique, called DISCOVER-Seq (discovery of in situ Cas off-targets and verification by sequencing) relies on endogenous DNA repair machinery that is naturally recruited to sites of double-strand breaks in the genome (see the figure). The authors determined that the protein MRE11 (meiotic recombination 11), a subunit of a complex that repairs DNA, was recruited to sites of Cas9-induced genome breaks. By isolating MRE11 and sequencing the bound DNA, Wienert et al. identified the locations of cleavage events for an RNA guide of interest. The method worked in induced pluripotent stem cells from a patient with Charcot-Marie-Tooth syndrome, as well as in mouse livers that were edited with virally delivered Cas9.

Previous in vivo methods required the introduction of additional components beyond the Cas enzyme and RNA guide, which can be a technical challenge for some cell types. Because DISCOVER-Seq utilizes endogenous DNA repair machinery to identify sites of double-strand breaks, no such additional factors are required. As a result, the approach of Wienert et al. can be applied to a variety of samples, including patient-derived primary cells. This opens a range of possibilities for stringently evaluating off-targets for therapeutic genome editing. By screening a panel of potential RNA guides in cultured patient-derived cells, it may be possible to identify off-target sites that might have otherwise been missed because of differences between a particular patient's genome and a standardized reference genome. Additionally, RNA guides can be designed and tested for patient-specific mutations for rare genetic disorders. Such a personalized approach to prevalidate RNA guides for individual patients ex vivo before treatment could provide an additional level of safety for CRISPR-Cas endonuclease therapeutics.

Detecting CRISPR off-targets

Two CRISPR-based approaches edit DNA, but neither system is perfect. Methods to detect the location of undesired edits provide information about editing accuracy.


The detection of off-target activity from fusion Cas9 base editors poses additional technical difficulties. The single-nucleotide variants produced by base editors are difficult to detect, especially in heterogeneous samples in which rare off-target mutations may be masked within the population. There are also potential unexpected effects that could arise from introducing the nucleotide deaminase. Zuo et al. developed an approach called GOTI (genome-wide off-target analysis by two-cell embryo injection) to identify potential off-targets of cytosine and adenine base editors in vivo in mouse embryos. A Cas base editor and RNA guide were injected into a single blastomere of a two-cell embryo (along with a molecule to fluorescently mark the edited cell). Progeny cells of the edited or nonedited blastomeres were sorted on the basis of fluorescence and then sequenced, providing an internal control from the same embryo for accurate determination of editing-induced single-nucleotide variants. Zuo et al. found that the cytosine base editor (BE3) generated around 20 times more single-nucleotide variants than the adenine base editor, Cas9, or control. Most off-target mutations were independent of the RNA guide, implying that off-targets did not arise from Cas9 itself but rather from random off-target activity of the fused deaminase.

Jin et al. reported similar genome-wide off-target findings for cytosine and adenine base editors in rice. The authors introduced the base editors into single cells and then evaluated genetic changes in the resulting rice plant, using whole genome sequencing. They found substantially more off-target single nucleotide variants in plants treated with the cytosine base editor than in plants treated with the adenine base editor or control. The results of Zuo et al. and Jin et al. are an important supplement to other work examining deaminase off-targets in vitro (11, 12). By editing at the single-cell stage, both groups could sequence a homogeneous population of cells and observe off-target changes scattered throughout the genome that were previously undetected.

The observed off-targets are not entirely surprising given the properties of the deaminase effectors. The BE3 base-editor system uses rat APOBEC (apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like), a cytidine deaminase that can bind to single-stranded DNA independently from Cas9. This could explain why Zuo et al. and Jin et al. observed most of the random mutations in actively transcribed genes, where DNA is unwound by transcriptional machinery and single-stranded DNA is available for binding and editing by APOBEC. This raises additional concerns about the potential effect of these random mutations, because they could disrupt highly transcribed protein-coding genes. By contrast, the adenine base editor uses a modified TadA (tRNA-specific adenosine deaminase) protein from bacteria as a deaminase (13). Unlike APOBEC, TadA lacks the ability to bind DNA on its own, and thus its activity is more likely restricted to sites of RNA guide–specified Cas9 binding rather than acting independently on random DNA sequences it encounters.

Substantial work has already been done to minimize off-target effects of Cas9 itself, including RNA guide–design strategies, ribonucleoprotein delivery, and protein engineering (14). Similar efforts should be made to improve the specificity of base editors by limiting deaminase activity outside of Cas9 binding. This could be done by utilizing different deaminase effectors or rationally engineering the deaminase to decrease its DNA binding ability. Overall, improved identification of off-targets provides an opportunity to optimize guide development as well as improve gene-editing tools themselves, advancing the capabilities of genome editing.

References and Notes

Acknowledgments: H.R.K. acknowledges the National Science Foundation Graduate Research Fellowship Program and Stanford Bio-X Fellowship Program. L.S.Q. acknowledges the Pew Scholar Foundation, Alfred P. Sloan Foundation, and Li Ka Shing Foundation.
View Abstract

Stay Connected to Science

Navigate This Article