Energy paradox in REM sleep: balancing supply and consumption in brain metabolism

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Communications Biology volume 9 , Article number: 979 ( 2026 ) Cite this article

The brain’s capacity for information processing depends on precisely regulated energy dynamics. Yet how metabolic supply adapts to shifting computational demands across brain states remains unclear. Using wide-field fluorescence imaging through the intact skull of live mice, we simultaneously monitored brain blood volume (BBV), astrocytic pyruvate, and neuronal ATP levels during natural sleep. We found that large-scale metabolic dynamics are coupled to neuronal activity but reorganize in a state-dependent manner. During non-rapid eye movement (NREM) sleep, theta-band electrocorticogram (ECoG) activity predicted subsequent blood volume changes, accompanied by rapid anterior-to-posterior vascular waves. In contrast, REM sleep was marked by a pronounced increase in BBV, originating in the posterior cortex and slowly propagating across the brain. This was accompanied by elevated astrocytic pyruvate; paradoxically, however, neuronal ATP levels declined sharply. These findings reveal a dynamic interplay among neurons, astrocytes, and the vasculature, suggesting that distinct energy-allocation strategies underlie the brain’s computational flexibility.

Energy is essential for driving information processing 1 . In the brain, interwoven information-processing and metabolic networks finely regulate energy supply and consumption to meet computational demands.

In this study, we investigated energy dynamics in live mice through the intact skull using wide-field fluorescence imaging 2 to monitor local brain blood volume (BBV), as well as cytosolic pyruvate and ATP levels in astrocytes and neurons 3 . We observed a tight coupling between energy dynamics and brain activity, with this association shifting to distinct functional relationships depending on brain state (e.g., non-rapid eye movement (NREM) sleep, REM sleep, and wakefulness) 4 . Given the limited energy availability in the brain, such constraints and competition for energy have likely contributed to the evolution of distinct energy-information coupling strategies. Elucidating energy dynamics in the biological brain represents a key step toward understanding the mechanisms underlying the brain’s unique information-processing capabilities and its unparalleled energy efficiency compared to modern computers 5 .

The firing of an action potential disturbs the ionic environment, requiring energy to restore homeostasis 6 . During an action potential, extracellular Na⁺ enters the neuronal cytosol, followed by the efflux of intracellular K⁺. To restore the ionic balance to the basal state, the Na⁺/K⁺-ATPase actively pumps Na⁺ out of the cell and K⁺ back in, working against their respective concentration gradients using ATP as an energy source. Through the continuous activity of Na⁺/K⁺-ATPase, neurons are able to represent information by firing complex patterns of multiple action potentials. In addition to neurons, astrocytes also express Na⁺/K⁺-ATPase and play a major role in the reuptake of excess extracellular K⁺ released during neuronal activity 7 . It is widely assumed that a substantial portion of the brain’s energy consumption is devoted to fueling Na⁺/K⁺-ATPase activity in both neurons and astrocytes, highlighting the significant energetic cost of maintaining ionic homeostasis across cell types 8 , 9 .

In addition to action potential generation, synaptic plasticity, which alters signal flow within neuronal networks, also requires significant energy. Cytosolic ATP is essential for synthesizing neurotransmitters, packaging them into synaptic vesicles, and driving synaptic vesicle cycling. Moreover, the synthesis, trafficking, and insertion of receptors, transporters, and ion channels also demand energy. Thus, substantial energy is required to support synaptic plasticity and memory formation.

The ultimate source of brain energy substrates is glucose delivered via the vasculature; however, many pyramidal neurons are not in direct contact with blood vessels. Thus, glucose is often assumed to be primarily taken up by astrocytes. Nevertheless, as extracellular glucose concentrations can be substantial, neurons are also capable of directly taking up glucose 10 . In fact, neurons have been shown to directly take up more than half of blood-borne glucose 11 .

Glucose taken up by neurons is converted into pyruvate and subsequently used for ATP production through mitochondrial metabolism. In contrast, glucose taken up by astrocytes is converted to pyruvate, which can either enter astrocytic mitochondria for oxidative metabolism or be further converted into lactate. Lactate produced in astrocytes can be released into the extracellular space via monocarboxylate transporters (MCTs). This lactate is subsequently taken up by neurons through neuronal MCTs, a process known as the astrocyte, neuron lactate shuttle (ANLS) 12 , 13 . Lactate taken up by neurons is reconverted into pyruvate and ultimately utilized by mitochondria to generate ATP. The relative contributions of direct neuronal glucose uptake and astrocyte-mediated lactate shuttling to neuronal ATP production likely vary depending on brain state and metabolic demand.

Delivery of energy substrates to neurons can be modulated by several mechanisms. For example, vascular dilation and constriction regulate the initial step of glucose and oxygen delivery 2 , 14 . This process can be adjusted on demand either through modulation of neurovascular coupling 15 , 16 , 17 or via neuronal signaling to astrocytes 18 . In addition to receiving glucose from the vasculature, astrocytes can store energy by converting excess glucose into glycogen. In particular, in association with K⁺ uptake during periods of elevated neuronal activity, stored glycogen can be mobilized to generate glucose and subsequently pyruvate, which can be used either for oxidative metabolism within astrocytes or converted into lactate for export.

Under conditions in which astrocyte-mediated substrate transfer plays a dominant role, increased metabolic utilization of pyruvate for ATP production within astrocytes could alter metabolite availability within astrocyte-mediated transfer pathways, potentially creating a trade-off between astrocytic and neuronal energy states. Thus, astrocytes and neurons may exhibit a context-dependent, partially reciprocal relationship in their energy states, depending on the relative contribution of different metabolic pathways. Another level of regulation in astrocyte-mediated energy delivery is the efficiency of MCTs in both astrocytes and neurons. However, neuronal energy supply is supported by multiple parallel routes, including direct glucose uptake. Therefore, the energy states of astrocytes and neurons are likely governed by a flexible balance between these pathways, which may vary depending on metabolic demand associated with different modes of information processing. Conversely, metabolic state may, in turn, influence the properties of neuronal computation 19 , suggesting a bidirectional interaction between energy metabolism and information processing.

ATP is essential for cell survival; therefore, multiple mechanisms likely act to balance ATP consumption and production, ensuring stable cytosolic ATP levels under physiological conditions. Indeed, in vitro studies have shown that neuronal ATP levels decrease only under hyperexcitation induced by strong electrical stimulation or exogenous glutamate exposure, or following pharmacological inhibition of mitochondrial ATP production pathways 20 , 21 , 22 . Consistent with these findings, in vivo studies have demonstrated that neuronal ATP levels remain stable in response to mild sensory stimuli in mice 22 , while a marked reduction in ATP levels is observed during epileptic seizure activity 3 . These observations suggest that brain energy substrates exist in a dynamic equilibrium, with compensatory mechanisms largely maintaining stable energy levels in healthy cells. It is likely that the delicate balance between energy consumption and supply in the brain is continuously adjusted depending on the computational state of the neuronal circuit.

Although sleep is often associated with “rest,” the brain remains highly active, with neuronal activity persisting but exhibiting frequency patterns distinct from those observed during wakefulness 23 , 24 . During NREM sleep, delta-frequency oscillations in the electroencephalogram (EEG) or electrocorticograms (ECoG) become prominent, although other frequency bands, such as the theta band, are also present, albeit at lower power 25 . In this study, we unexpectedly found that fluctuations in theta-band power, rather than those in the dominant delta band, closely resemble BBV dynamics. Subsequent detailed analysis revealed theta-band ECoG activity during NREM sleep precedes BBV dynamics by ~ 4, 5 s and can predict their temporal evolution. This finding suggests a tight coupling between neuronal activity and energy dynamics, which may contribute to maintaining stable levels of energy substrates.

Interestingly, we found that the delicate balance between energy consumption and supply breaks down during the transition to REM sleep. REM sleep is characterized by the emergence of strong theta-band EEG activity and the absence of most other frequency bands 26 and is associated with processes of memory selection and storage 27 , which are proposed to be reflected as dreams. During this stage, synchronous activity between the hippocampus and cortex is enhanced 28 . These neuronal activities and the formation of synaptic plasticity are thought to require large amounts of metabolic energy. Consistent with this assumption, a marked increase in BBV was observed 4 , 29 , 30 , which is generally associated with increased delivery of energy substrates, although this relationship is not directly measured in the present study. In parallel with the BBV surge, cytosolic pyruvate levels measured in astrocytes increased.

However, unexpectedly, a prominent decrease in neuronal cytosolic ATP was observed 31 . This decrease could result from (1) a reduction in the supply of energy substrates (e.g., lactate) to neurons, (2) a decrease in energy production within the neuronal cytosol, or (3) a substantial increase in ATP consumption driven by information processing specific to REM sleep. Although mechanisms (1) and (2) appear unlikely, as they would not support the increased energy demands, if such reductions in neuronal energy supply or production were to occur, this metabolic shift during REM sleep may serve an as-yet-unrecognized adaptive function. This ATP drop may reflect transient local energy redistribution that prioritizes synaptic reorganization 32 over immediate energy homeostasis.

In this study, we aimed to uncover how brain-state-dependent energy dynamics arise from coordinated interactions among neurons, astrocytes, and vasculature, and how these dynamic metabolic processes underpin distinct modes of information processing across sleep-wake cycles. We propose that the dynamic interaction between information processing and metabolism constitutes a key mechanism underlying the unique characteristics of the biological brain 33 .

Our goal was to elucidate how cerebral energy dynamics couple to information processing. Because craniotomy can markedly alter vascular and astrocytic activity by changing intracranial pressure and possibly provoking inflammation, we imaged brain activity through the native, unthinned skull of live mice. Immediately after skin incision and skull exposure, we coated the bone with a transparen