Postdoc’s grad-school sleuthing raises questions about bee waggle-dance data

Four years ago, Laura Luebbert noticed something unusual in two classic papers on bee navigation she was assigned to review for a graduate student journal club. Unfamiliar with the topic, she then read two others by the same authors to learn more.

“I had the feeling that I was looking at the same data over and over again,” wrote Luebbert, who is now a postdoctoral researcher at the California Institute of Technology, in a blog post earlier this week.

She had already detailed her questions about the data in a 2020 post on PubPeer, but it flew mostly under the radar, she says. Later, she and her adviser, Lior Pachter, professor of computational biology and computing and mathematical sciences at Caltech, looked into more papers co-authored by the same group and found additional issues.

By the time they were done, they had found what they say are data integrity issues in 10 studies of honeybee navigation published between 1996 and 2010. The potential issues include data duplication, manipulation and unusually high r-squared values, Pachter and Luebbert wrote in a May preprint. Collectively, the 10 papers in question have been cited more than 1,000 times, according to Clarivate’s Web of Science.

“If you look at any one paper and the issues in any one paper that we found, you can attribute them to human error,” Luebbert says. “It’s the entirety of it that really makes it something much bigger than that, and much more concerning.”

On 25 June, the Journal of Experimental Biology issued an expression of concern about two of the papers published in that journal in 1996 and 1997.

The single common author on all 10 papers is Mandyam V. Srinivasan, now an emeritus professor at the Queensland Brain Institute at the University of Queensland, who is known for his work on bee vision and navigation. The papers describe how honeybees can use vision to accurately fly to a food source and communicate this information to other bees through the waggle dance. Pachter says his investigation with Luebbert does not call into question Karl von Frisch’s Nobel Prize-winning discovery of the honeybee waggle dance, but rather the derivative work from Srinivasan.

In an email to The Transmitter, Srinivasan says the Journal of Experimental Biology contacted him about a year ago with questions about the two papers it had published, and his responses to them appeared in the recent expressions of concern, but he was unaware of the preprint until now.

The University of Queensland has policies and procedures in place for research conduct, but refers concerns about staff members’ work at previous institutions to those institutions, in line with Australian government guidelines, a university spokesperson told The Transmitter. The University further noted that it had not received The Transmitter’s request for comment, which was sent to the Queensland Brain Institute, before this story’s publication.

L

uebbert and Pachter submitted their preprint to the Journal of Experimental Biology in March. But, they told The Transmitter, the journal did not want to publish a critique that also mentioned papers from other journals. Eventually, they posted the preprint on arXiv in May. (The Transmitter is an editorially independent publication supported by the Simons Foundation, where Transmitter editor-in-chief Ivan Oransky serves as program officer for a gift to Cornell Tech to support arXiv.)

Luebbert discovered possibly duplicated figures both within and between the 1996 and 1997 Journal of Experimental Biology papers, as she mentioned in her 2020 PubPeer post. And in the 1997 paper, identical data appear in multiple figures, Luebbert alleges, but the figures report different experimental conditions.

In the expressions of concern about these papers, Srinivasan offered his explanations for the discrepancies and potential errors. He also stated that he takes “sole responsibility” for any errors, but that “the issues do not alter the overall results and conclusions of the paper.”

When asked if there will be further investigation into these papers, the managing editor of the Journal of Experimental Biology, Michaela Handel, wrote in an email to The Transmitter that “if additional concerns are raised about these or other papers published in Journal of Experimental Biology, we will investigate.” The link to the 1996 paper with an expression of concern previously went to an error page on the journal’s website, but it was fixed after an inquiry from The Transmitter.

Among the other issues Luebbert and Pachter describe is a potential miscalculation of the honeybee odometer in a 2000 Science paper. That paper reports that each millisecond of the waggle dance encodes 17.7 degrees of image motion, but when Luebbert recalculated the value from the data presented, it was 13.86 degrees, according to the preprint.

The issue with this calculation “seems not so grave if we just look at the data,” wrote Roger Schürch, assistant professor of entomology at Virginia Tech who has performed similar experiments, in an email to The Transmitter. “It might be in the realm of possible calibrations,” he added.

The issue, though, as Luebbert and Pachter point out in their preprint, is that the frame rate of the camera described in the methods is not concordant with the data presented in the paper. “It would be great if it could be clarified if some details of the methodology were omitted due to the brevity of the Science format,” Schürch wrote.

Of the 10 papers examined in the preprint, 6 have unexpectedly high r-squared values—close to or above 0.99—for data collected in variable conditions, Pachter says. For example, one of the graphs in a 2005 Proceedings of the National Academy of Sciences paper denotes an r-squared value of 0.999, but Pachter says his recalculation shows it was 0.918.

The Journal of Experimental Biology made Srinivasan aware of questions related to his publications in other journals earlier this year, Srinivasan wrote in his email to The Transmitter, and he wrote a reply to the journal that was not published. In this unpublished reply, which was reviewed by The Transmitter, Srinivasan recalculated a subset of r-squared values in four papers, writing, “To the best of my knowledge, there are no errors in the calculation of the regression coefficients.” He says that the method he used to calculate the regression coefficients, by using mean values, explains why they are “rather high.”

There are currently no expressions of concern listed for eight papers mentioned in the preprint that were published in the Journal of Experimental Biology, Science, Biological Cybernetics, The International Journal of Robotics Research, Annual Review of Neuroscience, PLOS Biology and Proceedings of the National Academy of Sciences.