Specificity is a recurring theme in genome editing—the question is how to ensure high activity at the target site while not introducing modifications elsewhere in the genome. The strategies taken thus far for CRISPR (clustered, regularly interspaced, short palindromic repeats) editing optimize the selectivity of the guide RNA or use dimeric forms of Cas9 that require two simultaneous binding events for a double-strand cut. David Liu from Harvard University has explored a strategy that limits the time window during which the Cas9 nuclease can cleave DNA.

Addition of a ligand triggers intein self-cleavage and activates Cas9. Figure adapted from Davis et al., Nature Publishing Group.

Liu points out that genome editing is not a steady-state process. There are only two desired substrate molecules in each diploid cell; therefore, “once a genome-editing agent has catalyzed the two turnovers, the only things it can do are undesirable,” says Liu. “If you limit exposure of a cell to an active genome-editing agent to a time window just long enough to achieve desired levels of on-target modification, you should increase the overall specificity of genome modification.”

To exert temporal control over Cas9, Liu's graduate student Kevin Davis made use of an intein, developed in the Liu lab over a decade ago, that self-cleaves in response to hydroxytamoxifen (HT). Davis fused the intein to the Cas9 alpha-helical domain and showed that this fusion protein does not cleave DNA. Addition of HT triggers the splicing of the intein and leaves an activated Cas9. The team targeted endogenous loci in human cells and assessed the degree of nonhomologous end joining, a process used by the cell to repair cut DNA. They observed an increase in specificity of up to two orders of magnitude when Cas9 activity was restricted to 12 hours. The researchers are now looking to modify genes associated with human diseases using this strategy.

It is still being debated exactly how specific CRISPR reagents are and how much specificity is required for a given research or therapeutic application. Although several studies have reported minimal off-target cuts in cells and also in mice (see Correspondence on p479 in this issue), the recently published study presenting an edited human embryo highlighted—among the important ethical issues surrounding this work—the need to better understand the technology, as off-target genome modifications were extensive in the zygotes.

Liu emphasizes the need for “a test that relates, preferably in a quantitative way, the relationship between off-target genome modification and organismal consequences,” akin to an Ames test that determines the mutagenic potential of chemical compounds. “It would help enormously,” says Liu, “to know that if the ratio of on- to off-target modification falls below a certain level for a certain application, one should do secondary tests to look for negative biological consequences.”

But Liu sees reasons to be upbeat. Once the off-target mutations introduced by genome-editing tools are comparable to, or below, the natural rate of spontaneous mutation in healthy human cells, they are unlikely to increase toxicity to the cell or organism. How much improvement is needed to achieve this goal is at present unclear, but Liu is optimistic that a combination of approaches, such as engineered Cas9 variants and direct delivery of Cas9–guide RNA complexes, will be successful.