Yes. This is a very important paper, but the headline “40 percent of MRI signals do not correspond to actual brain activity” is a little too dramatic. The paper does not show that 40% of fMRI is meaningless, nor that 40% of all MRI signals are false. It shows something subtler—and, scientifically, more consequential:
In roughly 40% of cortical voxels that showed a significant task-related BOLD change, the direction of the BOLD signal was opposite to the direction of oxygen metabolism.
The paper is Samira Epp et al., “BOLD signal changes can oppose oxygen metabolism across the human cortex,” published in Nature Neuroscience in December 2025.
Why this matters
Conventional fMRI does not measure neurons firing. It measures the BOLD signal—blood-oxygen-level-dependent contrast, essentially changes in deoxygenated haemoglobin. Researchers then infer neuronal activity through a physiological chain:
neuronal activity → increased energy/oxygen demand → altered cerebral blood flow → altered blood oxygenation → BOLD signal
The crucial assumption has usually been that this chain behaves predictably enough that:
positive BOLD ≈ increased neural activity
negative BOLD ≈ decreased neural activity
That assumption is what this study attacks.
The investigators simultaneously estimated cerebral blood flow, oxygen extraction fraction (OEF), and cerebral metabolic rate of oxygen, CMRO₂, while 40 healthy subjects performed calculation and autobiographical-memory tasks. CMRO₂ is substantially closer to the energetic consequences of neuronal activity than ordinary BOLD.
And they found this:
about 40% of voxels with significant BOLD responses were “discordant.”
Sometimes:
BOLD ↑ while CMRO₂ ↓
and sometimes:
BOLD ↓ while CMRO₂ ↑.
That second case is particularly striking. A conventional activation map might tell you:
“This region became less active.”
Yet its oxygen consumption actually increased.
So the image is not necessarily wrong. Our interpretation of the image may be wrong.
The physiological discovery may be even more interesting than the 40%
The conventional model assumes that when neurons need more oxygen, the brain mostly responds by increasing blood flow.
Epp et al. found two physiological strategies.
In “concordant” voxels, the classical model works quite well:
↑ neuronal metabolism → ↑ CBF → BOLD change
About 87% of the explained metabolic response in these voxels was associated with changes in blood flow.
But the discordant voxels behaved differently. They frequently changed the oxygen extraction fraction instead:
same or modest blood flow → extract more oxygen from the blood → ↑ metabolism
In these voxels, OEF accounted for about 58% of the explained variance, versus 42% from CBF.
So there appear to be at least two ways the brain can pay its energetic bill:
bring more oxygen in, or
extract more oxygen from what is already there.
Ordinary BOLD is much better at seeing the first mechanism than the second.
That means the deeper finding is:
Neurovascular coupling is not one uniform transfer function across the cortex.
And therefore there cannot be a universally valid translation:
BOLD amplitude → neuronal activity.
The authors explicitly conclude that BOLD alone can produce misleading interpretations of underlying neuronal activity.
And here comes psychiatry
This is where I think the paper becomes especially consequential.
Imagine two groups:
patients with depression
versus
healthy controls.
Suppose an fMRI study finds:
↓ BOLD in dorsolateral prefrontal cortex.
For thirty years, the temptation has been to write:
“Depressed patients demonstrate hypoactivation of the DLPFC.”
But logically, the experiment has actually demonstrated:
reduced BOLD response in the DLPFC.
The rest is inference.
After Epp et al., at least three explanations become plausible:
A. neuronal/metabolic activity really decreased.
B. neuronal activity is unchanged, but vascular response differs.
C. neuronal metabolism actually increased, but oxygen extraction changed in such a way that BOLD moved in the opposite direction.
That distinction becomes especially dangerous in aging, Alzheimer’s disease, vascular disease, and perhaps many psychiatric populations because the patient and control groups may differ systematically in vascular physiology. TUM explicitly emphasizes this problem.
Consequently, some apparent findings of “hypoactivation” and “hyperactivation” in the psychiatric fMRI literature may need reinterpretation.
Not necessarily rejection.
Reinterpretation.
Default Mode Network: particularly interesting
There is another detail I suspect you will find fascinating.
The discordant voxels were especially prominent in the default mode network (DMN).
That matters because an enormous psychological literature has interpreted task-related negative BOLD in DMN regions as:
“deactivation.”
The usual story goes:
task begins → DMN activity goes down.
But the study shows that negative BOLD does not necessarily mean reduced oxygen metabolism. In some DMN voxels, metabolism can go up while BOLD goes down.
That potentially complicates an extraordinary number of stories we have told about:
self-referential processing,
mind wandering,
autobiographical memory,
depression and rumination,
schizophrenia,
Alzheimer’s disease,
consciousness.
The network itself doesn’t disappear.
But some of the psychological narratives built on the polarity of BOLD become less secure.
There is an epistemological lesson here
For years popular neuroscience has given us beautiful coloured maps labelled:
fear
reward
self
empathy
trauma
religious experience.
But the instrument never saw any of those things.
It saw:
changes in magnetic properties produced partly by changing concentrations of deoxygenated haemoglobin.
Then came several layers of inference:
BOLD → metabolism → neuronal activity → computation → psychological construct.
Epp et al. demonstrate that trouble can already arise at the first physiological inference.
Thus:
BOLD \not\equiv neuronal\ activity
and, even more importantly,
\operatorname{sign}(\Delta BOLD)
\not\equiv
\operatorname{sign}(\Delta neuronal\ metabolism).
That is the genuinely disruptive part of the paper.
So I would formulate its significance this way:
fMRI remains an extraordinarily useful instrument, but the coloured activation map should no longer be treated as a transparent window onto neuronal activity. It is a vascular-metabolic shadow whose relationship to neuronal activity varies across the brain.
And perhaps this is another wonderful case of your familiar distinction between the thing and our representation of the thing: neuroscience had gradually forgotten that BOLD was a proxy, and began talking as though it were the brain itself.
Epp et al. have put the proxy back between us and the brain.
