Does REM sleep after slow-wave sleep balance local synaptic rescaling with global synaptic renormalization?

What each sleep stage seems to do to synapses

The synaptic homeostasis hypothesis holds that wakefulness strengthens many glutamatergic connections as the brain encodes experience, and sleep then scales them back down. In rats, sleep lowered levels of AMPA receptors — the receptor proteins that largely determine how strongly a glutamatergic synapse responds — in both the hypothalamus and the neocortex, with the strongest drops for receptors containing the GluA1 subunit [1]. When the researchers compared selective REM sleep with total sleep deprivation, slow-wave sleep emerged as the driving force behind this down-regulation, and REM sleep added no measurable contribution [1].

Cortical imaging in sleeping mice tells a more textured story. During SWS episodes, average calcium activity in excitatory pyramidal cells fell, but the spread of activity across the population widened: a subpopulation of cells actually became more active, and those cells were the ones most active during sleep spindles, the brief bursts of oscillatory activity linked to memory processing [3]. So SWS is not a uniform dimmer switch. It appears to combine a general lowering of excitability with a selective boost to circuits that were recently engaged.

Why REM sleep after SWS could be the balancing act

A review of sleep and memory consolidation proposes that replay of neuronal firing patterns during SWS, together with hippocampal ripples, thalamic spindles, neocortical slow oscillations and noradrenergic activity, helps convert hippocampus-dependent episodic memories into schema-like neocortical memories [2]. In that framework, REM sleep following SWS might balance the local synaptic rescaling that accompanies this memory transformation with a sleep-dependent homeostatic process of global synaptic renormalization [2]. The wording matters: this is offered as a possible division of labor, not a measured outcome.

The mouse imaging offers a compatible observation. REM episodes that followed SWS produced a general down-regulation of pyramidal cell calcium activity, including the spindle-active subpopulation that had been boosted during SWS, and this suppression persisted into later sleep and wakefulness [3]. Notably, parvalbumin-positive inhibitory interneurons increased activity during SWS but stayed unchanged during REM, which suggests the REM down-regulation is achieved by a different route than the SWS one [3]. That is consistent with REM acting as a broad, stage-specific reset, though calcium activity is a proxy for neural activity rather than a direct readout of synapse size.

Where the evidence for a handoff is thin

The strongest direct receptor measurement points to SWS as the main driver and finds no additional contribution from REM sleep [1]. That study compared REM sleep and total sleep deprivation but did not test an interaction between the two stages, so it cannot confirm or rule out a sequential SWS-then-REM balancing process [1]. The cortical imaging likewise describes REM as a general down-regulation, which fits the renormalization half of the proposal but does not show that it rescues or offsets specific local rescaling [3].

Other work adds context without settling the question. In patients with thalamic stroke, the normal overnight change in slow-wave slope — a marker thought to reflect synaptic renormalization — was reduced compared with patients whose strokes were outside the thalamus, and that reduction was related to greater daytime sleepiness [4]. This points to the thalamus as centrally involved in renormalization, but the study did not examine REM sleep or its interaction with SWS [4]. A review of sleep EEG oscillations and synaptic plasticity across NREM and REM sleep likewise treats the transition between slow-wave sleep and REM sleep as a distinct topic, describing global downscaling and synaptic refinement without providing new direct measurements [5].

So is the two-stage balance real?

The most defensible reading is that the two stages do different jobs. SWS is the better-supported driver of global synaptic renormalization, and it also permits a spindle-related, selective upregulation of recently active circuits [1][3]. REM sleep that follows is associated with a further, broad down-regulation of cortical activity through a mechanism that does not depend on the same inhibitory route as SWS [3], and it is proposed as the stage that reconciles local rescaling with global renormalization during memory transformation [2]. Whether REM actually performs that balancing function — rather than simply adding another round of down-regulation — remains an open question that the current evidence cannot answer.

Sources used in this answer

Related research questions