Among the many attractive features of the CRISPR–Cas9 system is its ability to replace a target sequence with a donor of choice. The efficiency of this homology-directed repair (HDR) varies widely depending on the cell type and whether long double-stranded DNA constructs bearing selection markers or short single-stranded oligonucleotides (ssDNA) are used to replace the target sequence. Jacob Corn at the Innovative Genomics Initiative and the University of California, Berkeley, is interested in the ssDNA approach because it allows scarless editing, but he notes that the pathways enabling such editing are still being worked out. “We wanted to understand how we can make this work best given a mechanistic understanding of the complex,” he explains.

In work led by postdoctoral fellow Christopher Richardson, the team first determined the kinetics of the Cas9 nuclease in vitro. Cas9 stayed bound to a single target site for 5.5 hours, even after cutting the DNA. This long residency time surprised the researchers, but they found supporting evidence in the work of Jin-Soo Kim, who independently explored the time course of CRISPR editing in cells and found similar repair-time lengths. Thus the long residency time of Cas9 seen in vitro could determine in vivo rates of editing.

“We knew that Cas9 holds on to the duplex very tightly,” says Corn, “but at first we thought it treated all strands equally.” This turned out not to be the case. By labeling the ends of the four fragments that are created by Cas9 cleavage—proximal and distal to the PAM site on both the nontarget and the target (bound by the single guide RNA) strand—they saw that only the distal site on the nontarget strand was accessible to ssDNA binding. “It flaps out,” as Corn graphically puts it. Even catalytically inactive Cas9 (dCas9) and a Cas9 mutant that introduces single-strand breaks make the nontarget strand bulge; “even if it is not cut, it is accessible to annealing with oligos,” says Corn. The team envisions that a 30-nucleotide stretch on the nontarget strand is looped out from the complex and that this loop is free to bind to ssDNA.

With this mechanistic insight, the researchers designed asymmetric ssDNA oligos that were complementary to the nontarget strand and spanned the cut site, overlapping the PAM-proximal site by 90 nucleotides but the PAM-distal site by only 30. They saw HDR frequencies of nearly 60%.

Even with dCas9, they were able to achieve HDR rates of up to 1% using ssDNA that annealed to the nontarget strand released by Cas9. With a strong selectable phenotype, this efficiency could be adequate to obtain edited cells. The advantage of dCas9 is that one does not have to worry about off-target cutting and editing. Although it is generally believed that nickases—enzymes that cleave only a single (target or nontarget) strand—do not introduce mutations at the nick site, the researchers did observe mutations that led to silencing of a reporter. They assessed editing only at the target site, but the observation that a single nick can be mutagenic should caution that the genome-wide off-targeting potential of nickases may need to be revisited.

One focus in the Corn lab is on introducing single-nucleotide polymorphisms at therapeutically relevant sites in primary human cells, and they recently succeeded in doing this using their asymmetric ssDNAs. Corn thinks that the discovery of Cas9's asymmetry and long residence time has implications beyond genome editing and could affect the use of dCas9 as a transcriptional regulator.