He Jiankui speaks at a gene-editing meeting last fall.Kin Cheung / AP

In the 1990s, virologists in New York learned of a genetic mutation that would become one of the most famous ever discovered. They found it in a man who could not be infected with HIV. He turned out to be missing just 32 letters in a gene called CCR5, and remarkably, it was enough to make him resistant to the virus killing so many others. About 1 percent of people of European descent carry two copies of this mutation, now known as CCR5-Δ32.

In 2018, a Chinese scientist named He Jiankui made the mutation infamous when he attempted to use CRISPR to edit CCR5-Δ32 (pronounced “CCR5-delta-32”) into human embryos. He chose this mutation, he said, because the babies’ father was HIV-positive, and he wanted to make the resulting twin girls resistant to the virus. CCR5-Δ32 is also, after all, one of the most studied mutations.

He’s work immediately provoked outrage among scientists, who knew enough to know how much they did not know about the risks of altering CCR5. And now a new study suggests that CCR5-Δ32 is indeed harmful overall.

The girls’ CCR5 genes were altered, according to data He presented, but they do not exactly match the 32-letter deletion; it’s unclear whether either of them is actually resistant to HIV. Even if they were unable to get HIV, a body of research already suggested that CCR5-Δ32 made people more vulnerable to the flu and West Nile virus. A “good” mutation in the context of HIV can be “bad” in another context. No one knew, exactly, the net effect of a CCR5-Δ32 mutation.

However, the new study, by Rasmus Nielsen and Xinzhu “April” Wei of UC Berkeley, shows that people with two copies of the mutation are 21 percent more likely to die at the age of 76, with a mortality rate of 16.5 percent, compared with 13.6 percent for those who have only one or zero copies. Only recently, Nielsen told me, have genetic databases even become big enough for these effects on mortality to be apparent.

The effect of CCR5-Δ32 on mortality is ultimately subtle, but it follows from what’s already known about this gene. CCR5 usually codes for a receptor on the surface of white blood cells, and it plays a role in normal immune responses. HIV co-opts CCR5 as a way to get into white blood cells. So to block HIV is, ironically, also to eliminate a small piece of the normal immune system.

“If you think about what these people are with Δ32, they’re like human knockouts for a fairly important gene in immune response,” says Bill Paxton, a microbiologist at the University of Liverpool who helped discover the role of CCR5 in HIV. “It’s not wholly surprising you [would] read a paper like this, and the finding is there.”

After HIV researchers made CCR5-Δ32 famous, scientists in other fields got interested in the mutation, too. Flu researchers who studied it found that it predisposes people to fatal outcomes with flu. West Nile virus researchers found the same with that disease. Neurobiologists have found evidence that CCR5-Δ32 actually enhances recovery from stroke. But this process of understanding the full scope of CCR5 has been piecemeal, essentially limited by what scientists think to look for.

Geneticists have proposed more systematic ways to understand all the effects of a single gene. Instead of picking a disease and looking for associated genes among a large group of people, geneticists can pick a gene of interest and look for associated traits. This is called PheWAS, or phenome-wide association study, where phenome refers to the set of observable traits. The idea is to look for links “that we just never knew to look for before,” says Marylyn Ritchie, a geneticist at the University of Pennsylvania who uses PheWAS in her research. Crucially, PheWAS requires not just DNA from volunteers but rich and detailed health data from those same volunteers—everything that could be conceivably linked to a gene, from height to brain volume to white-blood-cell count. PheWAS studies are necessarily limited by what health data have been collected.

Nielsen and Wei told me that they also tried to see whether CCR5-Δ32 was linked to other traits in the U.K. Biobank, and they found a few additional expected associations, such as white-blood-cell count. But their results are restricted by the health data scientists thought to begin collecting back in 2006, when the U.K. Biobank project began.

And despite its size, the U.K. Biobank is not representative of all people in all situations. In a population with more non-British ancestry and in a part of the world where certain viruses are more prevalent than others, CCR5-Δ32 could be more harmful or less beneficial. It depends on the environment, and the environment could also change in the future. A new epidemic could emerge, and so could new treatments.

That is ultimately what makes the proposition of editing genes like CCR5 so tricky. It can be hard to predict what the net effect will be, in a future we do not yet know, and harder still when all of the trade-offs today have not even been fully studied. In May, scientists launched an international commission on gene editing that will discuss these concerns, including how to balance the benefits and harm to not just a gene-edited child but also “subsequent generations.”

We want to hear what you think about this article. Submit a letter to the editor or write to letters@theatlantic.com.