ASHINGTON — At scientific meetings on genome-editing, you’d expect researchers to show pretty slides of the ribbony 3-D structure of the CRISPR-Cas9 molecules neatly snipping out disease-causing genes in order to, everyone hopes, cure illnesses from cancer to muscular dystrophy. Less expected: slides of someone kneeling on a beach with his head in the sand.
Yet that is what Dr. J. Keith Joung of Massachusetts General Hospital showed at the American Society of Hematology’s workshop on genome-editing last week in Washington. While the 150 experts from industry, academia, the National Institutes of Health, and the Food and Drug Administration were upbeat about the possibility of using genome-editing to treat and even cure sickle cell disease, leukemia, HIV/AIDS, and other blood disorders, there was a skunk at the picnic: an emerging concern that some enthusiastic CRISPR-ers are ignoring growing evidence that CRISPR might inadvertently alter regions of the genome other than the intended ones.
“In the early days of this field, algorithms were generated to predict off-target effects and [made] available on the web,” Joung said. Further research has shown, however, that such algorithms, including one from MIT and one called E-CRISP, “miss a fair number” of off-target effects. “These tools are used in a lot of papers, but they really aren’t very good at predicting where there will be off-target effects,” he said. “We think we can get off-target effects to less than 1 percent, but we need to do better,” especially if genome-editing is to be safely used to treat patients.
Off-target effects occur because of how CRISPR works. It has two parts. RNA makes a beeline for the site in a genome specified by the RNA’s string of nucleotides, and an enzyme cuts the genome there. Trouble is, more than one site in a genome can have the same string of nucleotides. Scientists might address CRISPR to the genome version of 123 Main Street, aiming for 123 Main on chromosome 9, only to find CRISPR has instead gone to 123 Main on chromosome 14.
In one example Joung showed, CRISPR is supposed to edit a gene called VEGFA (which stimulates production of blood vessels, including those used by cancerous tumors) on chromosome 6. But, studies show, this CRISPR can also hit genes on virtually every one of the other 22 human chromosomes. The same is true for CRISPRs aimed at other genes. Although each CRISPR has zero to a dozen or so “known” off-target sites (where “known” means predicted by those web-based algorithms), Joung said, there can be as many as 150 “novel” off-target sites, meaning scientists had no idea those errors were possible.
One reason for concern about off-target effects is that genome-editing might disable a tumor-suppressor gene or activate a cancer-causing one. It might also allow pieces of two different chromosomes to get together, a phenomenon called translocation, which is the cause of chronic myeloid leukemia, among other problems.
Many researchers, including those planning clinical trials, are using web-based algorithms to predict which regions of the genome might get accidentally CRISPR’d. They include the scientists whose proposal to use CRISPR in patients was the first to be approved by an NIH committee. When scientists assure regulators that they looked for off-target effects in CRISPR’d cells growing in lab dishes, what they usually mean is that they looked for CRISPR’ing of genes that the algorithms flagged.
As a result, off-target effects might be occurring but, because scientists are doing the equivalent of the drunk searching for their lost keys only under the lamppost, they’re not being found.
One little-appreciated feature of CRISPR’s DNA-cutting enzyme is that it doesn’t stop at one. Even if the enzyme cuts its intended target, the risk of off-target cutting remains. The enzyme “still has energy to bind with off-target sites,” Joung said, so “it can still cleave those sites.”
Scientists from some of the leading genome-editing companies said they are confident they will be able to minimize off-target CRISPR’ing, by picking “high-quality” guide RNAs, among other methods. While “bad” RNAs hit as many as 152 wrong targets, studies show, good ones hit only one, and the algorithms “capture most of” the potential off-target effects, said Dr. Bill Lundberg, chief scientific officer of CRISPR Therapeutics. Still, he conceded, “At the end of the day we’re taking a cell where we can’t predict a priori where the edit has happened.”
Scientists have recently recognized another reason to worry about off-target effects: No two people’s genomes are identical. Off-target-identifying methods, which are based on a composite or “reference” human genome, might indicate that there are no stretches of DNA that CRISPR can mistakenly snip. But because of random mutations and genetic variations, some patients might have additional “123 Main Street”s, attracting CRISPR and its DNA-cutting enzyme where they’re not supposed to go.
“There are a significant percent of sites, more than I would have thought,” where that might happen, said Joung, “and it varies by ethnic group.”
Said Andrew May, chief scientific officer of Caribou: “There is going to have to be some consideration of that” as genome-editors try to bring CRISPR to patients.