Our brain demands a lot of energy compared to our other organs. However, it can also make do when energy supplies are sparse, flexibly performing information processing using what is available. How the brain resourcefully allocates this limited energy across internal states remains a key question in neuroscience.

Sleep provides a useful window to the answer of this question. Although sleep is associated with rest, the brain remains highly active. This is especially true during rapid eye movement (REM) sleep, the stage closely linked to dreaming and memory processing. REM sleep is sometimes called "paradoxical sleep" because the body is largely still while the brain shows wake-like activity. Researchers at Tohoku University have now uncovered another paradox within REM sleep: while energy supply to the dreaming brain appears to rise, the energy molecule directly used by neurons falls.

The findings were published in Communications Biology on July 27, 2026.

"Ever felt exhausted after a vivid dream?" asks Professor Ko Matsui of Tohoku University. "Sleep may appear peaceful, but the brain is highly active - especially when dreaming. We were intrigued by this paradox, and wanted to look into the scientific basis behind why dreaming is somehow tiring."

To get a better understanding, the team kept the mouse skull transparent using a UV-curable resin, allowing them to observe the brain during natural sleep. With wide-field fluorescence imaging, they monitored brain blood volume fluctuations as an indicator of "fuel" supply, neuronal ATP as the energy molecule that powers neurons, and astrocytic pyruvate as a key molecule connecting blood-derived glucose to brain energy metabolism.

Non-REM sleep is best known for strong neuronal activity in the delta-band frequency, but subtle theta-band fluctuations are also present. The researchers found that these theta-band fluctuations could predict brain blood volume changes several seconds later, suggesting that the sleeping brain adjusts vascular dynamics to ongoing neuronal activity and metabolic demand.

The transition from non-REM sleep into REM sleep showed a different pattern. About 50 seconds before the classically defined onset of REM sleep, brain blood volume began to rise. This increase started in the posterior cortex and spread forward, suggesting a large-scale metabolic preparation process. After REM sleep began, astrocytic pyruvate also increased, consistent with enhanced substrate availability or elevated astrocytic glycolytic activity. Paradoxically, however, neuronal ATP decreased.

Several mechanisms may explain the ATP decrease. Neurons may consume large amounts of ATP during REM sleep to support memory-related synaptic reorganization, hippocampal-cortical communication, or large-scale circuit transitions. Alternatively, metabolic transfer from astrocytes to neurons may be altered, or mitochondrial ATP production may shift.

The research also points toward a broader principle of biological computation. Unlike conventional computers, animal brains operate under strict metabolic constraints. Rather than distributing energy uniformly, the brain may reroute energy depending on behavioral state, memory demand, and internal needs.

"Understanding how the brain balances energy supply and consumption may help explain what makes biological intelligence so efficient," explains lead investigator Yusuke Takahashi. "REM sleep gives us a natural example of how the brain reorganizes its energy economy to support complex internal processing."

Sleep is a vital function that can help consolidate memories and keep us mentally sharp the next day. These research findings about energy and sleep are a crucial step forward to helping us understand the science behind sleep, and how a good night's rest is not just important for our bodies - but our minds too.