Neuronal ensembles - small populations of co-active neurons within a brain region - form the building blocks of memory. Yet these ensembles are not static. Their composition shifts with new experiences, raising an important question: what determines which ensembles persist and which are replaced? A new open-access study by Hiroyuki Miyawaki and Kenji Mizuseki provides a detailed systems-level view of this process, showing that ensemble stability depends not only on local dynamics but also on how strongly an ensemble participates in coordinated activity across distant brain regions during sleep.
The researchers recorded neuronal activity from the prelimbic cortex layer 5 (PL5), basolateral amygdala (BLA), and ventral hippocampus CA1 (vCA1) in male rats undergoing fear conditioning, extinction training, and retention-of-extinction sessions. Across these stages, animals experienced recurring NREM sleep epochs, allowing the team to examine how ensembles reacted both during waking behavior and in sleep-dependent off-line periods thought to support memory consolidation.
PL5 ensembles were intrinsically more stable than those in the BLA or vCA1, suggesting a cortical tendency to preserve ensemble identity across days. However, this baseline stability did not fully predict whether specific extinction-related ensembles would persist into later retention sessions. Instead, the key predictor involved interactions occurring during the intervening sleep.
The researchers analyzed patterns of ensemble reactivation during post-extinction NREM sleep. Surprisingly, the simple amount of reactivation did not explain which ensembles survived. Instead, the stability advantage emerged from a more sophisticated phenomenon: inter-regional coactivation. Ensembles that reactivated in coordination with ensembles in other regions - particularly when their activation patterns aligned during sleep - were far more likely to be preserved into the retention-of-extinction session.
Importantly, extinction training did not change the overall number of ensemble pairs that coactivated across the three regions. Rather, extinction reorganized the fine-scale structure of these inter-regional interactions. The composition of coactive pairs shifted, altering which ensembles from each region became linked during sleep. This restructuring indicates that extinction modifies the network-level architecture of memory, not only the local encoding of new associations.
A further layer of insight came from examining the temporal structure of sleep activity. Preserved ensembles that participated in inter-regional coactivation were disproportionately active during periods of fast network oscillations within NREM sleep. These fast oscillations represent windows of heightened synchrony in which information transfer and network plasticity are believed to be particularly effective. The study shows that ensembles embedded in these high-frequency coordination events gain a competitive advantage, becoming the long-term carriers of extinction memory.
The results point to a powerful systems-level mechanism: local ensemble stability is enhanced when an ensemble contributes to broader network coordination across regions. Memory retention is thus not just a matter of replaying local patterns but depends on how well those patterns integrate into a multi-region architecture during sleep.
This understanding has broad implications. For example, the prelimbic cortex's relative stability may allow it to serve as an anchor region, providing a persistent framework into which more plastic subcortical ensembles - such as those in the amygdala or hippocampus - can embed their associations. Extinction learning, which involves suppressing fear responses, appears to rely on this cross-region communication. Ensembles that fail to join the inter-regional synchrony of sleep may be more easily overwritten, potentially explaining why extinction learning is sometimes fragile and context-dependent.
From the perspective of Seven Reflections' Dimensional Systems Architecture (DSA), these findings illustrate the principle that stability emerges not from isolated units but from coherence across interconnected fields. In DSA terms, neuronal ensembles act as local structures within a larger dynamic field. Their long-term persistence depends on resonance: the extent to which an ensemble aligns with and participates in broader system-wide oscillatory patterns. Sleep functions as a synchronization window where low-entropy, high-coherence field interactions determine which structures become stable carriers of information. Ensembles that enter these coordinated modes achieve structural integration, while others dissipate as transient patterns. Memory consolidation, from this view, reflects the selective stabilization of field-coherent structures rather than the simple strengthening of local nodes.
This research reinforces a growing appreciation that memory is not stored in isolated circuits but emerges from the dynamic organization of multi-region activity - particularly during sleep. By identifying how inter-regional coactivation sculpts which ensembles endure, the study provides a mechanistic link between local plasticity, global network structure, and long-term behavioral change. It also suggests that interventions targeting sleep oscillations may one day enhance the stabilization of adaptive memories, including those supporting fear extinction.
