Summary: A new theory proposes that the universal sequence of sleep stages—non-REM (NREM) followed by REM—is critical for strengthening and organizing memories. Mouse experiments combined with circuit-based computational models indicate that NREM sleep amplifies and stabilizes newly formed memory traces, while subsequent REM sleep selectively prunes overlapping or irrelevant elements, keeping separate memories distinct. Reversing this sequence impairs memory, offering a potential explanation for why this sleep architecture is so conserved across species.
Researchers emphasize that further work is required, but these results present a coherent account of how the ordered pattern of sleep stages supports learning and memory consolidation.
Key Facts:
- Memory gardening: NREM promotes growth and stabilization of memory engrams; REM trims competing or overlapping components to maintain clarity.
- Order matters: Running REM before NREM degrades memories instead of refining them, showing the sequence is functionally important.
- Model-driven insight: Integrating mouse recordings with a biophysical circuit model highlighted how fluctuations in acetylcholine shape network dynamics across sleep states.
Source: University of Michigan
Background
Sleep remains essential for health but is still scientifically mysterious. One of its most striking features is the repeating cycle from quiet, synchronized brain activity (NREM) into a high-activity state often accompanied by vivid dreaming (REM). This NREM→REM pattern appears across vertebrate species and is rarely observed in reverse, suggesting a deep evolutionary purpose.
Sara Aton, a professor at the University of Michigan, and colleagues sought a mechanistic explanation for why sleep cycles consistently progress from NREM to REM. Their combined experimental and computational approach produced a parsimonious hypothesis: during NREM, memory traces expand and recruit neurons, while REM later trims back those traces to prevent interference between similar memories.
In plain terms, if memories were shrubs, NREM acts as the gardener that lets them grow and root, and REM is the pruning shears that remove tangled or excessive branches so each plant remains distinct.
According to Aton: “The sequence matters. If REM comes first, pruning happens before memories are reinforced, and the result is loss rather than refinement. When NREM precedes REM, critical connections are strengthened and REM can selectively remove overlap.”
Mouse experiments and modeling
The team monitored hippocampal activity in mice after a simple conditioning task: mice explored a new environment and then, in the experimental group, received a brief foot shock. A control group explored the environment without the aversive stimulus. By comparing sleep-period activity after these sessions, the researchers examined how different sleep stages affected memory-related neural patterns.
Because current recording methods cannot trace every single neuron that encodes a memory, the researchers developed a complementary circuit model. Michal Zochowski’s group created a simplified biophysical network that represents learning as changes in neuronal activity within an excitatory–inhibitory network influenced by acetylcholine (ACh).
The model captures how low ACh levels during NREM reduce inhibition and produce waves of synchronized activity led by engram neurons, recruiting additional excitatory cells into the memory ensemble. By contrast, high ACh during REM increases inhibitory tone so only the most strongly recruited excitatory neurons remain active, effectively pruning less robust or overlapping connections.
This complementary action—growth during low-ACh NREM followed by selective pruning during high-ACh REM—explains how the ordered sleep cycle can simultaneously stabilize memories and prevent interference between similar experiences.
Everyday implications
The findings align with everyday observations about sleep and memory. Zochowski notes that after a day with multiple similar events—such as several meetings—NREM appears to strengthen the individual memories, while REM helps preserve distinguishing details like who said what and when. Proper sequencing of these stages may therefore be important for retaining discrete, detailed memories.
The study appears in PLoS Computational Biology and was supported by the National Science Foundation, the Chan Zuckerberg Initiative, and the National Institutes of Health. The research builds on advances from the NIH BRAIN Initiative over the past decade.
The authors caution that their network model is necessarily simplified and that the experiments addressed relatively straightforward learning tasks. The hypothesis is testable and may evolve as researchers apply these ideas to more complex memory scenarios and new datasets.
About this sleep and memory research news
Author: Matt Davenport
Source: University of Michigan
Contact: Matt Davenport – University of Michigan
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Cholinergic modulation of neural networks supports sequential and complementary roles for NREM and REM states in memory consolidation” by Sara Aton et al., PLOS Computational Biology
Abstract
Cholinergic modulation of neural networks supports sequential and complementary roles for NREM and REM states in memory consolidation
Across vertebrates, sleep commonly cycles from NREM to REM, but the specific functions and the reason for this stereotyped sequence have been unclear. Using a simplified biophysical network model, the authors explored how cholinergic modulation through muscarinic receptors could influence network dynamics and memory consolidation.
They show that low and high acetylcholine levels, associated with NREM and REM respectively, may perform distinct and complementary roles. Low ACh during NREM reduces inhibitory neuron activation, causing network disinhibition and synchronized bursts led by engram neurons that recruit additional excitatory cells into the memory trace. High ACh during REM raises inhibition, suppressing weaker activity and pruning the expanded engram back to its most robust components. Together, these dynamics offer a testable hypothesis for how state-specific cholinergic modulation supports memory consolidation during natural sleep cycles.