Have you ever awakened from a vivid dream feeling strangely tired?
Our study does not show that dreaming causes fatigue. Nevertheless, it revealed a real and unexpected metabolic phenomenon during rapid eye movement, or REM, sleep: brain blood volume increased, yet the ATP available inside neurons decreased.
REM sleep is already known as “paradoxical sleep.” The body is largely still, while the brain shows activity resembling wakefulness. In humans, this sleep stage is closely associated with vivid dreaming, and it is also thought to contribute to memory reorganization and emotional processing.
We have now found another paradox hidden within this already paradoxical state.
Following energy across blood vessels, astrocytes and neurons
The brain requires a continuous supply of energy. Neurons consume ATP to generate and transmit electrical signals, restore ion gradients and modify synaptic connections. Oxygen and glucose reach the brain through the vasculature, but the path from blood-borne substrates to usable neuronal ATP is not necessarily direct.
Astrocytes are positioned between blood vessels and neurons and are thought to contribute to the processing and allocation of metabolic substrates. We therefore wanted to follow three interconnected components of the brain’s energy network:
- brain blood volume, as a vascular signal associated with substrate delivery;
- pyruvate inside astrocytes, as a central metabolite linking glucose breakdown to energy metabolism; and
- ATP inside neurons, as the molecule that directly powers neuronal functions.
A methodological challenge was how to observe these signals without substantially disturbing the vascular and glial environment.
Instead of making a conventional cranial window, we preserved the optical transparency of the intact mouse skull with a transparent UV-curable resin. This allowed us to use wide-field fluorescence imaging to observe a large area of the cerebral cortex while the mice naturally moved among non-REM sleep, REM sleep and wakefulness.
The project was led by graduate student and first author Yusuke Takahashi, working together with Assistant Professor Yoko Ikoma and me in our laboratory at Tohoku University.
The sleeping brain continuously adjusts its vascular support
We first examined how neuronal activity and brain blood volume were related during non-REM sleep.
Non-REM sleep is best known for prominent slow delta activity. However, weaker theta-band fluctuations are also present. To our surprise, these theta-band fluctuations showed a particularly close relationship with subsequent brain blood-volume changes.
Changes in theta-band ECoG activity occurred approximately four to five seconds before the corresponding vascular changes. When we convolved the theta-band signal with a simple brain-blood-volume response function, much of the measured blood-volume fluctuation could be reproduced.
This result suggested that vascular activity during sleep is not simply a static background process. Even during non-REM sleep, the brain appears to continuously adjust vascular delivery according to ongoing neuronal activity.
We also found a spatial structure in these fluctuations. During non-REM sleep, relatively fast blood-volume waves repeatedly propagated from anterior to posterior cortical regions, crossing much of the observed cortex in approximately one second.
The brain begins preparing for REM sleep before REM is detected
The transition into REM sleep revealed a very different vascular pattern.
Approximately 50 seconds before REM sleep could be conventionally identified from the ECoG and muscle activity, brain blood volume had already begun to rise.
This increase did not occur simultaneously throughout the cortex. It started in posterior cortical regions and slowly propagated toward anterior regions over approximately 15 to 20 seconds.
REM sleep therefore appeared to involve more than an abrupt switch in neuronal activity. A large-scale vascular and metabolic transition was already developing before the classically defined beginning of REM sleep.
Our initial expectation was wrong
The increase in brain blood volume raised an obvious question: did intracellular energy availability increase as well?
When we artificially increased brain blood volume using the vasodilator sodium nitroprusside, the result followed the expected pattern. Astrocytic pyruvate increased, and neuronal ATP also increased. Greater vascular delivery was therefore capable of increasing intracellular metabolite levels.
We initially expected natural REM sleep to produce a similar response.
Astrocytic pyruvate did rise during REM sleep, particularly in posterior cortical regions. This was consistent with increased substrate availability, increased astrocytic glycolytic activity, or both.
Neuronal ATP, however, moved in the opposite direction.
During REM sleep, neuronal ATP decreased sharply despite the concurrent increase in brain blood volume. The reduction was especially prominent in posterior cortical areas, including the retrosplenial and visual cortices.
Thus, the three components of the metabolic network were not changing in parallel:
- brain blood volume increased;
- astrocytic pyruvate increased;
- neuronal ATP decreased.
Natural REM sleep had broken the relationship that we had observed during pharmacological vasodilation.
This was the result that transformed the study from an investigation of sleep-related vascular dynamics into a study of an energy paradox.
Video caption: Neuronal ATP falls while brain blood volume rises during REM sleep. Wide-field fluorescence imaging shows an increase in the brain blood-volume signal and a simultaneous decrease in neuronal ATP across the cerebral cortex, particularly in posterior cortical regions (modified from Supplementary Movie 5 in original article).
Where did the neuronal ATP go?
Our study does not yet establish why neuronal ATP decreases during REM sleep. We currently consider at least three broad possibilities.
First, neurons may consume unusually large amounts of ATP during REM-specific information processing. Synaptic reorganization, hippocampal–cortical communication and large-scale circuit transitions may all impose substantial energy demands.
Second, the transfer of metabolic substrates from astrocytes to neurons may become restricted. Changes in monocarboxylate transporter activity could alter the transfer of lactate or related substrates, allowing pyruvate to accumulate in astrocytes while neuronal ATP falls.
Third, mitochondrial ATP production may become less efficient during REM sleep. A temporary shift away from oxidative phosphorylation could contribute both to pyruvate accumulation and to reduced neuronal ATP.
These possibilities are not mutually exclusive. The ATP decrease may reflect a combination of increased consumption, altered substrate transfer and state-dependent changes in energy production.
Determining the relative contribution of these mechanisms will require direct manipulation of transporters, mitochondrial function, cellular pH and metabolic pathways during natural sleep.
What vascular imaging can—and cannot—tell us
The findings may also influence how we interpret hemodynamic brain imaging.
Functional magnetic resonance imaging, or fMRI, does not directly measure neuronal activity. It infers brain activity mainly from changes in blood flow, blood volume and blood oxygenation.
Our results do not reduce the value of fMRI. Hemodynamic signals remain powerful indicators of brain-state-dependent activity. However, the REM sleep paradox shows that increased vascular signals do not necessarily mean that more ATP is available inside neurons.
During REM sleep, brain blood volume increased while neuronal ATP decreased.
Blood vessels therefore represent one layer of energy regulation, but they do not reveal the entire intracellular energy state. Astrocytes and other cellular metabolic processes may provide an additional layer that determines how blood-derived substrates are processed and made available to neurons.
We propose a dual-layered adaptive energy-allocation strategy:
- vascular regulation adjusts the delivery of blood-borne substrates; and
- astrocyte-associated metabolic regulation influences how those substrates are allocated and transferred.
Together, these layers may allow the brain to adapt energy use to different modes of information processing.
The brain as an energy-constrained computer
Modern computers can consume large amounts of electricity and produce considerable heat. The biological brain, by contrast, performs perception, memory, decision-making and internally generated experiences under severe metabolic constraints.
The brain may achieve this efficiency not by distributing energy uniformly, but by continually reallocating limited resources according to its current computational state.
REM sleep provides a striking natural example. Vascular delivery increases, astrocytic metabolism changes and neuronal ATP simultaneously falls. What initially appears to be an energy failure may instead reflect a highly organized, state-dependent redistribution of metabolic resources.
Whether this transient neuronal “energy drain” contributes to the familiar feeling of tiredness after a vivid dream remains speculative. The cellular phenomenon itself, however, is clear: during REM sleep, more blood does not necessarily mean more neuronal energy.
This paradox may provide a route toward understanding not only sleep and memory, but also the broader principles that allow biological intelligence to operate with exceptional energy efficiency.
Read the paper
Yusuke Takahashi, Yoko Ikoma and Ko Matsui.
Energy paradox in REM sleep: balancing supply and consumption in brain metabolism.
Communications Biology 9, Article number: 979 (2026)
https://doi.org/10.1038/s42003-026-10646-6
Continue the conversation
Why might the brain allow neuronal ATP to fall during REM sleep despite increasing vascular delivery?
We welcome questions, alternative interpretations and ideas for future experiments.