The authors behind a contentious 2022 Science paper that purported to measure neuronal activity using functional MRI (fMRI) retracted the work today.
The retraction marks the end of the road for the method, called “direct imaging of neuronal activity,” or DIANA, says Noam Shemesh, principal investigator at the Champalimaud Centre for the Unknown, who was not involved in the now-retracted work. But many neuroimaging researchers still hope to one day use fMRI to capture neuronal activity. “MRI is such a rich modality. It has such rich physics, and not all of it has been exploited in the functional sense,” Shemesh says.
DIANA collected fMRI data in a way that enabled the researchers to measure signal changes on the order of tens of milliseconds. The team, led by Jang-Yeon Park at Sungkyunkwan University, captured a signal peak in the somatosensory cortex of mice 25 milliseconds after shocking their whisker pads.
Despite an initial flurry of excitement from the field, other labs could not replicate the results. As a result, the paper received an editorial expression of concern in August 2023 because “the methods described in the paper are inadequate to allow reproduction of the results and … the results may have been biased by subjective data selection,” the notice states.
Following the editorial expression of concern, “we reanalyzed the data. Unfortunately, the additional results revealed unexpected MR signal characteristics and did not robustly support the original conclusions. We are therefore retracting the paper,” the retraction notice states. Science did not have any additional comment beyond what is outlined in the expression of concern and retraction notice.
“They’re doing the right thing by stepping back from it,” says Ravi Menon, professor of medical biophysics, medical imaging, neuroscience and psychiatry at the University of Western Ontario.
D
irectly capturing neuronal activity using fMRI, rather than measuring it indirectly through changes in blood flow and oxygenation, has long been an unrealized dream of the fMRI field. So when the DIANA paper came out, many researchers, including Shemesh, rushed to try out the method on their own scanners.
“I took my best student and ran to the basement, and we basically coded the sequence on the spot and did the experiment for the first time,” Shemesh says. He saw the signal peak but soon realized it was an artifact.
Two other groups came to the same conclusion and published their failed replication attempts in March 2024. One team observed the signal peak described by Park’s group but uncovered that it was an artifact created by data-collection timing: An electrical pulse meant to synchronize the scanner and animal stimulation inadvertently shifted data collection by 10 microseconds.
The other group only saw the peak when they excluded data that did not resemble the signal they were looking for.
At the time, Park stood by DIANA. “I firmly believe, despite the controversy, the DIANA signal exists,” he told The Transmitter in March 2024. “Further studies from other groups—including me and my collaborators—will clearly reveal the truth. I think it’s just a matter of time.”
But earlier this month, prior to the retraction, Park and his colleagues outlined artifacts that contribute to the DIANA signal in a preprint posted on bioRxiv. The introduction and discussion sections describe why the authors decided to retract the paper, Park told The Transmitter in an email. The preprint confirms the timing delay but does not address the data-selection issues described in one of the replication attempts.
Park’s team aimed to hold the protons in the brain at a steady state of magnetization; that way, they could attribute any changes in the magnetization to the electrical current produced by an action potential.
But many factors can cause small deviations from that steady state, says Shella Keilholz, professor of biomedical engineering at the Georgia Institute of Technology and Emory University. These include the signal spoiling mechanisms used to maintain the steady state, and the timing delay.
“These are subtle signal inconsistencies,” says Menon, who was a co-investigator on one of the replication efforts. “Normally, nobody cares about them because they’re so tiny.”
But those fluctuations start to drown out any meaningful measurements once you search for something as small as the DIANA signal, Keilholz says, which is marked by a roughly 0.2 percent change in signal strength.
Given these artifacts, it is “premature” to conclude that the signal “is primarily attributed to neuronal activity, as further investigations are required,” Park and his team write in the preprint.
B
ut the artifacts do not explain why the team observed changes in activity that peaked in the thalamus and then in the somatosensory cortex on a timescale aligning with data collected in electrophysiology experiments, Park and his colleagues argue—and the consistency of this result is “difficult to ascribe to coincidence alone.”
Park reported that timing gap in the original 2022 paper. “That was so convincing” that the signal was capturing neuronal activity, Keilholz says. “I was so excited to see that when it first came out.” But now the data in the preprint show signal changes across the whole brain and at various times. “That should be an alarm sign, right? That’s not something we expect to see. When you have a small sensory stimulus, it shouldn’t light up your whole brain,” she says.
The authors do not write off the entire signal as artifact, however. “We anticipate that sufficient time and comprehensive support from the academic community would be required to fully elucidate the relationship between DIANA signals and brain activation, including neuronal activity,” the preprint states.
“I think it’s a sad fact about science that we make mistakes, and when the mistakes lead us in interesting directions, sometimes it’s difficult to completely let go,” says Alan Jasanoff, professor of biological engineering at the Massachusetts Institute of Technology and the McGovern Institute for Brain Research, who led one of the replications. But at this point, he says, “I feel like we have a sufficient explanation of the signal in terms of non-neuronal sources that I wouldn’t try to do neuroscience based on it.”