The version of CRISPR whose selling point has been its precision suffers, ironically, from the same shortcoming that has dogged other forms of the genome editor — that it makes a lot of unintended, off-target DNA changes. In two studies published on Thursday, one in mice embryos and one in rice plants, scientists find that this “base editing,” a form of CRISPR invented in 2016, can cause hundreds of unintended mutations, potentially making its clinical use a genetic crapshoot.
Fears about off-target mutations from CRISPR have waxed and waned even as clinical development of CRISPR, for severe genetic diseases ranging from sickle cell and inherited blindness to progeria and genetic epilepsy, marches on. In one episode, a 2017 paper reporting sky-high off-target edits in mice briefly tanked the stocks of three CRISPR companies. The paper was retracted last year after the authors conceded that they’d probably made a rookie mistake, but concerns about unintended edits haven’t disappeared.
Partly because of that history, the latest research is being received cautiously. “I really like these studies and think they advance the field,” said biochemist David Liu of Harvard University, who invented base editing three years ago. They show that there are reliable ways to detect off-target editing, he said, and should light a fire under scientists to figure out ways to minimize it.
But details of the mouse study, in particular, “make me suspect that this is a worst-case scenario,” Liu said. He welcomed that, however, saying it suggests the outer limits of base editors’ flaws.
The mouse study tested two base editors, which differ from classic CRISPR in key ways. In classic CRISPR, a guide RNA tracks down a specified sequence of nucleotides in DNA and an accompanying enzyme cuts the double strands of DNA there, slicing out a disease-causing segment or inserting healthy DNA. Base editors, too, use guide RNA, but instead of making a double-stranded break they cut only one DNA strand, and then chemically convert one target nucleotide into another: C to T, G to A, or A to C, and so on. Thousands of inherited diseases arise from a single wrong nucleotide.
Because single-letter misspellings are scattered all over the genome, it’s been tough to measure whether base editors are changing only what they’re supposed to: Finding an A where a G is supposed to be might be because the base editor changed a G that it shouldn’t have, or because nature did.
Scientists in China, led by Hui Yang of the Shanghai Institutes for Biological Sciences, figured out a way around that problem: use base editing in one cell of a two-cell mouse embryo and leave the other one untouched. The two cells are genetically identical, so by comparing them and their descendants (the scientists waited two weeks to have enough cells, edited and unedited, to genome-sequence), it’s possible to count off-target edits without being misled by naturally occurring genetic differences.
Bottom line: An adenine base editor, which turns AT pairs into CG pairs, caused almost no off-target edits, the team of Chinese, U.S., and European researchers reported in the journal Science. But the cytosine base editor, which turns CG pairs into TAs, mistakenly changed about one nucleotide in 20 million. The mouse genome is 6 billion nucleotides long (about the same as humans’), so that rate translates to 150 off-target edits. (The study of base editing in rice similarly found that the cytosine but not the adenine form introduced significant, genome-wide off-target mutations.)
“It is an interesting and important report,” Jin-Soo Kim, director of the Center for Genome Engineering at Seoul National University, said of the mouse study. It isn’t clear which part of the editor — the DNA-cutting enzyme or the one that changes one nucleotide into another — is responsible for the hundreds of collateral edits, added Kim, who has developed cutting-edge techniques to measure off-target edits from base editors, so “it is now important to determine [that] and how to reduce or avoid them. A new problem always invites new opportunities.”
One question is how dangerous a rate of 1 in 20 million nucleotides might be. DNA is constantly mutating, due to hits from cosmic rays and random glitches, at a rate as high in 1 in 1 million. “One in 20 million is in that range,” said Liu, who co-founded Beam Therapeutics, a Cambridge, Mass., company that hopes to turn base editing into therapies for severe inherited genetic diseases. It therefore might “have little or no impact” on patients who one day undergo base editing therapy.
The mouse study used extremely high levels of the base editor, beyond what’s likely to be needed for therapy: It got on-target (that is, intended) editing rates of 70 percent and higher, which is considered far above what would be needed to treat and even cure most patients. Introducing a lower concentration of base editor into the mouse cells produced fewer off-target edits, said co-author Lars Steinmetz of Stanford University, who led the genomic analysis, “but they didn’t disappear.”
Liu, too, has warned that base editors can cause unintended mutations. When DNA replicates for cell division and when genes get transcribed, a temporary “bubble” of single-stranded DNA can form, he explained. “Random encounters” between base editors’ nucleotide-changing enzymes (called deaminases) and these transient bubbles can produce random, off-target editing, especially by the enzyme in the cytosine base editors. It’s attracted to single-stranded DNA more than 1,000 times more strongly than the enzyme in adenine base editors.
That might explain why the mouse study found more unintended edits with the cytosine editor, but it also offers an escape route: tinker with this enzyme. Unpublished data, Liu said, show that new base editors, with different enzymes, show greatly reduced and possibly zero hits on unintended targets.