Skip to Main Content

The fear that CRISPR-based genome repair for preventing or treating genetic diseases will be derailed by “editing gone wild” has begun to abate, scientists who are developing the technique say. Although there are still concerns that CRISPR might run amok inside patients and cause dangerous DNA changes, recent advances suggest that the risk is not as high as earlier research suggested and that clever molecular engineering can minimize it.

“Progress is being made at a pretty stunning rate,” said biochemist David Liu, of Harvard University and the Broad Institute, who has developed several versions of CRISPR. A parade of new discoveries, he said, “suggests that it’s possible to use these genome-editing tools and not make unintended edits.”

The basic reason for such “off-target effects” is that CRISPR’s guide molecule, which is usually 20 genetic letters long, isn’t as precise as often advertised. Although it’s supposed to lead the DNA-cutting enzyme (the other principle component of CRISPR) to regions of the genome with those exact 20 letters, in practice it’s happy to find regions with just 18 or even fewer — kind of like a bloodhound that’s primed to find a suspect based on the scent in his particular Nikes but that goes snuffling after everyone wearing Nikes.


A study last year involving just three mice raised fears that CRISPR-Cas9, the classic version of the genome editor, might be a big, stupid bloodhound. Scientists at Stanford University reported they found hundreds of off-target effects, which they said was “of concern”: Messing with a tumor-suppressor gene could cause cancer, while other edits “could be detrimental to key cellular processes,” they wrote.

The stock prices of several CRISPR companies swooned, but CRISPR experts immediately pounced. In addition to using so few mice, the study, they said, didn’t account for the fact that DNA differences between the two CRISPR’d mice and the control mouse might simply have reflected different ancestry. After all, except for identical twins, one individual’s DNA differs from another’s at millions of sites, no CRISPR’ing required.


New research, submitted to a journal, supports the critics. Scientists at the Wellcome Sanger Institute in England report that, using the same basic approach as the OMG Stanford study, they found no such problems: They used CRISPR-Cas9 to change a gene called Tyr, which is responsible for black fur, in 10 mouse embryos. Comparing the CRISPR’d genomes to those of nine control embryos — plus, crucially, all the parents — they found only targeted edits and the expected (few) spontaneous DNA changes.

“We don’t have any detectable off-target damage attributable to our CRISPR,” said Sanger’s Vivek Iyer. That, he and his colleagues wrote, “should support further efforts to develop CRISPR-Cas9 as a therapeutic tool.”

The apparent absence of off-target edits in mice — especially in a study of only 19 of them —  certainly doesn’t guarantee the same for patients who might one day undergo genome-editing. “For human therapeutics, the bar is much higher,” said biochemist Erik Sontheimer of the University of Massachusetts Medical School. “But it’s definitely a tractable problem.”

His own research supports that optimism: It seems that nature is full of CRISPR enzymes that are more accurate than the original Cas9, which comes from Streptococcus pyogenes bacteria. Sontheimer tested a Cas9 from the bacterium Neisseria meningitidis. In a head-to-head comparison in human embryonic kidney cells (a lab stalwart) growing in dishes, classic Cas9 hit the wrong target hundreds of times, while the NME version “exhibits a nearly complete absence of unintended targeting in human cells,” Sontheimer and his team wrote in a paper submitted to a journal. (It and the Sanger paper were posted on the bioRxiv website and have not yet been peer-reviewed.)

Although Sontheimer has applied for a patent on the use of NME Cas9 for genome-editing, he doesn’t think it’s necessarily the final answer. “The point is that there are [Cas9’s] out there for which you don’t have to do a lot of engineering to increase their accuracy,” he said. “They’re naturally super-accurate.”

Scientists around the world are nevertheless busily tweaking Cas9 and other elements of CRISPR, striving for the perfect bloodhound. Take Liu’s CRISPR invention, which Harvard biologist called “the most clever CRISPR gadget.” The “base editor” changes a single DNA nucleotide, or base (such as C or G), into another (A or T). That has the potential to correct genes where a single wrong letter causes a disease, such as the blood disorder beta thalassemia. Alternatively, a single-base change can produce a “stop” signal that makes a disease-causing gene halt in its tracks — no more disease-causing proteins to cause, say, hereditary deafness and the form of blindness called retinitis pigmentosa.

The base editor targets regions five letters long, however, opening the door to errant changes: If the editor is supposed to change a C to a T to repair a gene, but there’s a second C nearby, it might change that one, too, with who-knows-what consequences.

Scientists led by Dr. Keith Joung of Massachusetts General Hospital, however, have discovered a way to drastically minimize that bystander effect. According to an unpublished paper they posted last week, they tweaked the base editor to, in essence, change that C only if the letter before it is, say, a T. It’s akin to priming the bloodhound to find suspects wearing Nikes only if the aroma of Dockers is also around.

Joung (who declined to discuss the work until a journal publishes it) tested it on human cells growing in the lab. They picked 60 regions of the genome likely to tempt the original base editor into unintended editing because of their similarity to the target. Sure enough, the editor changed 36 of them in many of the cells.

But the tweaked editor, for which Joung has filed a patent application, left 21 of the 36 completely alone. It mistakenly edited 15 sites, but in fewer cells than the original editor. And when Joung (a co-founder of genome-editing company Editas Medicine) and his team tested how well their tweaked editor fixed the mutation that causes beta thalassemia, also in human cells growing in the lab, it beat the original hands down, causing “undetectable” levels of off-target editing.

Liu, who was not involved in the study, called it “a brilliant approach to minimizing bystander editing.”

There is no question that if scientists aren’t careful, CRISPR can induce substantial off-target mutations. In another study Joung’s lab submitted to a journal, they show that when “promiscuous” forms of CRISPR were slipped into mice’s livers, as some genome-editing companies hope to do for some human metabolic diseases, it edited hundreds of spots in the mouse genome that it wasn’t supposed to.

That should serve as a warning to do-it-yourself CRISPR fans: You can craft a genome-editor that cuts and cuts and cuts DNA throughout the genome.

For scientists who know what they’re doing, though, “my general sense is that off-targets are not a major issue, especially given the development of new engineered forms of Cas9,” said Neville Sanjana of the New York Genome Center. He suspects that, for any experimental CRISPR therapy, it will be necessary to “precisely quantify the locations and frequency of off-target modification,” just as drug developers measure toxicity and adverse effects.

A key question for genome editors is whether doing that will require sequencing the entire genome of patients in clinical trials. That could be expensive.

Another question is how to handle humans’ tremendous genetic diversity: We differ from each other by millions of DNA letters. That suggests that genome-editing therapies “would need to be tailored to each patient,” said Sanger’s Iyer: The region of the genome containing disease-causing DNA, and therefore CRISPR’s targets, might be spelled slightly differently in different patients.

Regulators will have to decide how much off-target CRISPR’ing is acceptable. Since people’s genomes experience constant natural mutations, due to cosmic rays and other forces, the level of acceptable off-target editing “should not be zero percent,” said Liu, “but editing that’s a tiny fraction of these natural changes” (and not in, say, tumor-suppressor genes).

He added, “I think we’ll accept more risk if the consequence of not treating patients [with genome-editing] is that they die.”