A highly influential paper that claimed to have identified specific microbial signatures associated with different human cancers has been retracted after journal editors found the conclusions to be unsupported.
The research used machine-learning algorithms to detect associations between specific, resident microorganisms in the human body and the presence and type of cancer, sometimes with accuracy approaching 100%. Published in Nature in 2020, the work spawned multiple follow-up studies by other groups, as well as a company to turn the results into tools for early cancer diagnosis. But the findings came under scrutiny after critics alleged errors in the paper’s methods, leading the journal first to issue an editorial note in February and ultimately to pull the paper today.
“Expert post-publication peer review of the issues raised and the authors’ responses has confirmed that some of the findings of the article are affected and the corresponding conclusions are no longer supported,” states Nature’s retraction notice. All the authors agreed with the decision, it adds.
Author Rob Knight, a microbiologist at the University of California San Diego and a leader in microbiome studies, tells ScienceInsider in an email, however: “Our research team believes that the major conclusions of the original 2020 Nature manuscript regarding the ability of the microbiome to distinguish tumor types are true. We look forward to further illustrating this view in forthcoming research with improved methodology and more comprehensive data sets.”
The retraction is already having cascade effects on other research. National Cancer Institute researcher Eytan Ruppin, who used the 2020 paper’s data set for a 2022 study in Nature Communications, said last year he would wait to see how the allegations unfolded. He now tells ScienceInsider it is the “appropriate time” for his team to act. “I aim to contact the journal and discuss with them two basic options—to redo [our] analysis with the corrected data, or to retract our work.”
In the 2020 Nature study, Knight and colleagues compared a database of DNA in blood and tissue samples of people with cancer, as well as in samples of some without, with a database of most known microbial DNA. Using machine-learning algorithms, the researchers developed computer models that could in theory predict the presence of cancer and its type from a blood sample.
In 2019, Knight and the paper’s first author, Greg Sepich-Poore, co-founded a company, Micronoma, to commercialize the research. Micronoma raised millions in funding, and developed an experimental lung cancer blood test it hoped to move into clinical studies—the test received a “breakthrough device” designation from the U.S. Food and Drug Administration last year, which gave it priority for review.
However, researchers in the United Kingdom noticed oddities among the Nature paper’s findings. Knight’s team had listed seaweed bacteria among the microbes indicative of bladder cancers, for example—a sign of possible problems with the analysis. In a preprint posted last year (and now published in mBio), the U.K. researchers and Steven Salzberg, a computational biologist at Johns Hopkins University, went further, detailing “major data analysis errors.”
According to them, Knight’s group failed to properly filter out human DNA from their cancer tissue data, leading millions of human sequences to be incorrectly identified as microbial. A separate computational error in the team’s analysis inadvertently generated phantom patterns in the data—resulting in apparent associations between cancer and microbes when there were none, Salzberg and colleagues wrote.
At the time, Knight and colleagues rejected the criticisms, which they argued didn’t fundamentally alter the paper’s conclusions. But in February, Nature added a note to the article stating that “concerns have been raised about the data and conclusions presented in this article.” The decision to pull the paper comes after the journal considered additional responses from the authors—including a separate journal article that reanalyzed some of the 2020 paper’s data.
Salzberg says he is pleased with the decision. “Once I realized the problems in the paper, I felt really the only proper response would be to retract.” Other researchers have also reacted positively. “It’s reassuring to see that concerns were taken on board, and the authors have agreed to the retraction,” says bioinformatician and microbiologist Lesley Hoyles of Nottingham Trent University, who was not involved in the work.
The fate of technology built on the findings is uncertain. In 2023, Sandrine Miller-Montgomery, president and CEO of Micronoma at the time, told ScienceInsider the company’s products weren’t affected, and that Micronoma had generated “an independent and proprietary microbial database.” However, the company has not posted news to its website since last year, and Miller-Montgomery stepped down last fall. Her successor, Eddie Adams, posted on social media last month that Micronoma was “selling off a large quantity of scientific equipment” as part of a “laboratory equipment liquidation.” Neither Adams nor the company responded to a request for comment.
Researchers emphasized that the retraction does not invalidate research on cancer microbiomes as a whole. “The outcome should entrench neither the naysayers nor the proponents of tumor microbiomes,” says Ivan Vujkovic-Cvijin, a microbiome scientist at Cedars-Sinai Medical Center.
Still, it’s a cautionary tale about being too uncritical of research using these complex methods, Hoyles suggests. “The microbiome community needs to be open to reassessing research outputs as our understanding of the limitations of different bioinformatics tools and machine-learning approaches evolves,” she says.
More: https://www.science.org/content/article/journal-retracts-influential-cancer-microbiome-paper
