LONDON – A paper that raised concerns for the future health of Lulu and Nana, the world's first gene edited babies, has been fully retracted at the request of the authors, after they failed to identify a problem in data from the U.K. Biobank on which their analysis was based.

The study published in Nature Medicine in June, concluded the edits the twin girls are purported to have in their CCR5 chemokine receptor gene is associated with a 21% increase in mortality in middle and old age.

The researchers drew on genomic data and death register information of 409,693 volunteers in the U.K. Biobank. They concluded people who are homozygous for the delta32 mutation said to have been introduced into Lulu and Nana's germline, have a significantly higher death rate between the ages of 42 and 78, than people with no, or one copy. (See BioWorld, June 4, 2019.)

He Jiankui, the Chinese scientist who carried out the controversial germline editing, justified it on the grounds that being homozygous for delta32 confers resistance to HIV infection. The girls' father was said to be infected with the virus.

Now, Nature Medicine has announced the paper is being retracted in full, after the authors, Xinzhu Wei and Rasmus Nielsen, both of UC Berkeley, became aware of a genotyping calling bias in the underlying U.K. Biobank data.

"Further analysis confirmed that the central finding of the study – that homozygous CCR5-delta32 mutation is associated with increased mortality in the U.K. Biobank – is a result of this technical artefact," the retraction says.

An accompanying commentary, titled "The hidden cost of genetic resistance to HIV-1," also has been withdrawn.

"The one thing that all scientists fear the most is to find out that a major result they have published was based on erroneous data," Nielsen wrote on Twitter in advance of the official retraction.

"Our strongest results seem to have been caused by biased genotype calling in the genotype data we used from the U.K. Biobank," Nielsen told BioWorld. "This is a problem that we failed to identify in our initial analyses, and it seems to be able to explain the main result of the paper," he said.

A large number of scientists around the world carry out research using U.K. Biobank. Nielsen said he wanted to emphasize that the genotype calling bias observed in his work "is likely unique" to this variant and "not a general sign of poor data quality" in the U.K. Biobank.

The shortcoming of the analysis carried out by Nielsen was confirmed by David Reich and colleagues at Harvard University.

Early doubts turn out to be correct

However, it was Sean Harrison, senior research associate in epidemiology at Bristol University Medical School, who first took issue with the research. On June 20, he posted a lengthy blog, relating failed attempts to repeat the analysis and pointing to various shortcomings in Nielsen's paper.

Harrison, who is not a geneticist, but an epidemiologist/statistician who works a lot with U.K. Biobank data, noted it is not appropriate to lift the data wholesale. For the CCR5-delta32 analysis, measures were required such as removing people who are related to each other from the sample, and controlling for population stratification, which is a confounder of the association between single nucleotide polymorphisms (SNPs) and outcomes.

When he attempted to repeat Nielsen's analysis, using the same data, but accounting for what he saw as missteps in the way the data were prepared, Harrison found nothing. The same was true for two other SNPs that are said to be good proxies.

"I don't believe there is enough evidence to say that the delta32 deletion in CCR5 affects mortality in people of white British ancestry, let alone people of other ancestry," Harrison concluded in his blog.

Harrison and Nielsen had what Harrison called "a nice collaborative exchange," leading to the confirmation by Reich that the CCR5-delta32 data were faulty, and the prompt retraction of the paper.

After Harrison's intervention, Nielsen at first thought he had confirmed his findings using another database called gnomAD. He subsequently found there were errors in that database, too. (Those are unrelated to the issue with U.K. Biobank data, and are being corrected.)

But Nielsen also was shown evidence that the increase in mortality cannot be replicated in several other proprietary databases he did not have access to. "So if there is an effect on mortality of delta32/delta32 it is certainly not as strong as we previously reported," Nielsen said.

Reich, Nielsen, Wei and others are joint authors of a non-peer-reviewed paper published a week ago on Cold Spring Harbor Laboratory's Biorxiv preprint server, titled "No statistical evidence for an effect of CCR5-delta32 on lifespan in the U.K. Biobank cohort."

"Our analysis in the U.K. Biobank of more markers tagging delta32/delta32 individuals does not provide statistical confirmation that [these] individuals have shorter lifespans than other people. Additionally, in testing associations with a wide range of phenotypes we do find weak effects on a handful of traits, but none that suggest a 20% increased mortality rate," the new paper concludes.

All the protagonists are keen to stress that invalidating Nielsen's analysis does not let He off the hook for his controversial human gene editing research.

"Artificially knocking out a gene in human embryos without fully understanding its function is dangerous, especially as genes that serve no function are unlikely to survive evolutionary pressures. It seems very plausible therefore that there are negative consequences of having no functional copy of the CCR5 gene," Reich et al say in the Biorxiv paper.

As to where the research goes from here, Nielsen said, "I am not sure the world wants to hear more from us on this. We will leave further analyses up to other researchers. But current analyses suggests that if there is an effect on mortality, it is almost certainly not as strong as we initially reported," he told BioWorld.

No Comments