The waggle dance of the honey bee is one of the marvels of nature. A bee that has returned to the hive from foraging shakes its body in a figure-eight movement that tells hive mates where to find a food source. Discovered decades ago by Austrian ethologist Karl von Frisch, who won a Nobel Prize in 1973 for his work, the dance grew more wondrous starting around 2000, with evidence that bees have an internal “odometer,” relying on visual cues to measure the distances they convey in the dance.
Now, questions are swirling around those claims, after two scientists detailed concerns about possible miscalculations, reuse of images, and manipulation of data in 10 of the key papers. “If you look at any one of these issues alone, you could chalk it up to human error,” says biologist Laura Luebbert, a postdoctoral researcher at the California Institute of Technology (Caltech) who co-authored the critique, posted on the arXiv preprint server and in a blog post this week. But, she says, “This happened over and over again, across many different papers over a long period of time.”
The accusations have put an uncomfortable spotlight on the one author shared by all of the papers: bioengineer and neuroscientist Mandyam Veerambudi “Srini” Srinivasan, now an emeritus professor at the University of Queensland. Srinivasan, whose bee research has won him numerous awards including the prestigious national honor known as the Order of Australia, says he was unaware of the critical preprint until contacted for comment for this article.
Srinivasan told Science his conclusions “remain firm and sound, and have been replicated independently in many subsequent studies.” He said he is “surprised and disappointed” by the “unprofessional manner” in which the Caltech researchers went about their investigation. “There was definitely no falsification or manipulation of data,” says Srinivasan, who calls the allegations “totally bizarre.”
Some in the field are backing him. Jürgen Tautz, a bee expert from the Julius Maximilians University of Würzburg who collaborated with Srinivasan on several of these projects, says he is “highly convinced that the accusations lack any basis.”
Other bee researchers are concerned, however. The “precise claims” in the work “clearly have to be revisited,” according to Fred Dyer at Michigan State University, who studies bee behavior. But the key conclusion that honey bees use visual cues to navigate to their foraging grounds “still stands,” he says.
When Luebbert first read about the bee odometer for a university journal club in 2020, she felt something was amiss. Senior faculty discouraged her from pursuing her doubts, she says, but she persevered. “I just felt like, why bother if it’s not true, if it’s not real, if I can’t believe in it.” She contacted science integrity consultant Elisabeth Bik, who agreed the papers needed scrutiny and encouraged her to post her doubts to the anonymous PubPeer server. “I am definitely concerned,” Bik now says, about the integrity of Srinivasan’s data.
With the help of her Caltech research supervisor Lior Pachter, a computational biologist, Luebbert has now reanalyzed the data in the 10 of Srinivasan’s studies, from between 1996 and 2010. In one, published in 2000 in Science, Srinivasan had honey bees fly within tunnels lined with black and white barcodelike patches to serve as landmarks, then filmed their waggle dances. He concluded that each millisecond of a waggle dance encodes 17.7° of apparent motion of a landmark. But when Luebbert redid the math with the data from the original study, she found that each millisecond should encode 13.86° of image motion.
Although these results are odd, they are “not impossible” given the variation that can occur in these data sets, says Roger Schürch, a behavioral ecologist at Virginia Polytechnic Institute and State University who has created a universal formula for understanding bee waggle dance.
What’s more telling, according to Schürch, is that the 2000 paper states the bees were filmed at 25 frames per second, but the results are not consistent with that rate. Schürch does not think the team “made up values,” but says Srinivasan and his colleagues might have changed something in their methods without reporting it. He thinks the concerns listed for the affected papers merit further investigation and potential retraction down the line if necessary. A spokesperson for Science says the journal is evaluating these concerns.
Luebbert and Pachter also allege that Srinivasan reutilized data from a 1996 study in the Journal of Experimental Biology as the results for different bee experiments, under different conditions, reported in a 1997 JEB paper. The 1996 paper studied 88 bees in an 11-centimeter-wide tunnel, whereas the 1997 study described 56 bees flying in a 7-centimeter tunnel. The same pattern of reporting identical data for different experiments with different conditions also appears within the 1997 study, according to their arXiv report.
JEB says that after Luebbert and Pachter alerted it to the concerns, it investigated and published a note “to alert readers to issues with the data.” Srinivasan says he told the journal that because the original data for these studies are not available, it’s not possible to determine where the errors lie, but that he takes full responsibility for the mistakes mentioned.
Luebbert and Pachter say a 2000 paper in Biological Cybernetics suffers similar problems. They contend the paper includes data duplicated from the 1996 JEB paper and reports “highly similar” data and some identical data points for different experiments throughout the paper. In two of the graphs, “not only do the points align, but the fitted exponential curve has exactly the same exponent, except it is flipped,” Pachter says.
Javaan Singh Chahl, a neuroscientist and roboticist from the University of South Australia who worked on the 2000 research but didn’t take part in the data gathering or analysis, notes that limits in the technology from the time could explain apparent echoes. “If you can only measure between zero and six and all the measurements are in the middle, they’ll tend to be three and four,” he says. Still, he thinks Srinivasan may have presented only data that suited the studies’ graphing format, which Chahl calls “selection bias.”
Luebbert and Pachter also cite another anomaly: Six of the 10 papers have R-squared values—a value between zero and one that shows how well data fit a predicted trend—of about 0.99. These are “ridiculously high” correlations, Pachter says, and unlikely given the data were collected across a wide range of conditions.
Luebbert and Pachter’s calculations suggest that in a 2005 paper published in the Proceedings of the National Academy of Sciences (PNAS), the R-squared value should be 0.918 instead of 0.99. (PNAS says it is looking into the matter.) Similarly, the 2000 Science paper reports an R-squared value of 0.997, but when Luebbert put the same values and their variability through a computer simulation, only 0.006 of the results would have led to an R-squared value of 0.997.
When asked about the R-squared issues, Srinivasan shared his own recalculations of the values and says the results “are in very good agreement with the published values.” But Pachter says this analysis “deflects and is disingenuous,” and only covers two of the six investigated papers.
Overall, “There’s a real cost and a real harm when something like this happens,” Pachter says. “A lot of resources get distributed to scientists. The scientist is highly decorated, presumably got a lot of funding, probably sucked away resources from others.”
Aung Si, a former Ph.D. student of Srinivasan who now does unrelated research at the University of Cologne, says that while working in the Srinivasan lab he “never felt pressured to produce a particular favorable result, or to suppress a result that contradicted a previous finding.” But given the allegations, Si agrees that certain early experiments “should be replicated to set the record straight.”
