Sleep 301

Networks that regulate REM. REM is actively generated by interacting brainstem circuits. GABA is a chemical messenger that usually inhibits its target neurons. In mice, activating a GABA-releasing pathway from the medulla, part of the brainstem, can trigger REM from non-REM (NREM) sleep, in part by inhibiting REM-suppressing neurons. Inhibiting an inhibitor can therefore promote a state. This is experimental animal evidence, not a complete map of human REM control. 1

Motor inhibition and dreaming. Atonia means a marked reduction in muscle tone. In REM, most skeletal muscles have greatly reduced activity while breathing and eye movements continue. Rat experiments implicate inhibitory GABA and glycine signaling at motor neurons, the nerve cells that drive muscles. These motor mechanisms do not explain all dream experience: human studies also find dreaming during NREM. 237

Signals and scale. Scalp electroencephalography (EEG) records electrical brain activity through scalp electrodes. Electrooculography (EOG) tracks eye movements; electromyography (EMG) records muscle electrical activity. Together, these signals help classify stages. A hypnogram plots the sequence of scored stages; microarchitecture describes briefer events within them. A stage label is therefore a useful summary, not a complete description of the sleeping brain. 459

Slow oscillations and spindles. Slow oscillations alternate between relatively active and quiet neuronal states, typically below about one cycle per second. Spindles are brief bursts, commonly around 12-16 cycles per second, generated through interactions between the thalamus and cortex. Frequency is measured in hertz (Hz), or cycles per second. These rhythms can become coupled, meaning their timing is coordinated. 5

Reactivation and consolidation. Memory consolidation is the stabilization and reorganization of information after learning. One proposed mechanism involves reactivation of learned patterns during sleep and communication between the hippocampus, important for forming memories, and cortical networks. Slow oscillations, spindles and ripples, brief high-frequency bursts, can align in time during NREM. Human recordings link this coordination with neuronal communication and memory reactivation. 56

What the evidence supports. In one human study, researchers replayed learning-related sounds during sleep and detected associated memory patterns. Deep recordings came from people undergoing epilepsy monitoring, while scalp recordings came from healthy participants. This supports a mechanism under specific conditions; it does not prove that one stage stores all memories, or that increasing a wearable's deep-sleep estimate improves learning. 6

Autonomic regulation. The autonomic nervous system adjusts functions such as heart rate and blood pressure. These generally decrease as NREM deepens, but brief increases can accompany arousals. REM is more variable, and breathing can become less regular. These are patterns across states, not rules that every heartbeat or breath follows. 7

Sleep state and biological time. Growth hormone release often accompanies early slow-wave sleep, but not every overnight hormone change is caused by a stage. In a human experiment that shifted sleep relative to the internal clock, the cortisol rhythm remained largely unchanged. Sleep-related effects and circadian effects must therefore be distinguished when interpreting overnight physiology. 78

Recorded signals and inferred states. Polysomnography (PSG) combines brain, eye and muscle recordings with measures such as breathing and blood oxygen. Many wrist and ring trackers instead use movement data and photoplethysmography (PPG), an optical pulse measurement, to estimate sleep through algorithms. Their stage labels and sleep scores are not direct measurements of memory processing or every biological event within a stage. 49

Validation and uncertainty. Against a PSG reference, sleep sensitivity is the proportion of PSG-scored sleep correctly identified; wake specificity is the proportion of PSG-scored wake correctly identified. A device can perform well on one and poorly on the other. Validation must match the device, software, metric and population. Correlation also differs from causation: two measures changing together does not establish that one caused the other. A consumer sleep score alone is not a diagnosis. 910