Why do we toss and turn after a stressful day, yet sleep soundly after a joyful one? How exactly does the brain tailor our sleep based on waking experiences?
While it is widely accepted that sleep is crucial for memory consolidation—spontaneously reactivating daytime memories to cement them in the brain—a compelling question remains: Could this nighttime memory replay actually be reverse-engineering our sleep? At the cellular level, does memory activity actively shape sleep architecture according to our waking experiences?
In a new study published online in Science on June 4, 2026, researchers led by Dr. Bo Lei at the Beijing Academy of Artificial Intelligence (BAAI) and Dr. Yi Zhong at 蜜桃app reveal that memory reactivation is far more than a mere byproduct of consolidation. Instead, it serves as an active regulator that adaptively sculpts sleep. The team demonstrated that the reactivation of negative memories triggers transitions from non-rapid eye movement (NREM) sleep to wakefulness, to fragmented sleep. Conversely, replaying positive memories enhances NREM sleep continuity and bolsters resistance to environmental disturbances.
1. Memory Reactivation as an Active Sleep Regulator
To unravel how daily experiences dictate sleep patterns, the researchers first monitored mice after a fear-conditioning task. Following a negative experience, the animals exhibited highly fragmented sleep, characterized by shorter NREM episodes and a surge in transitions from NREM sleep to wakefulness. Similar arousals could be artificially triggered by playing auditory cues during sleep to reactivate these specific fear memories. Crucially, silencing these memory engrams abolished the sleep fragmentation, proving that the structural changes in sleep were entirely memory-dependent. To precisely pinpoint these reactivation events, the team developed an AI-driven EEG encoder that accurately distinguishes “memory reactivation sleep” from baseline sleep.
By combining two-photon in vivo imaging with EEG/EMG recordings, the researchers tracked memory-responsive cells in the basolateral amygdala (BLA) at single-cell resolution. They observed that following a fearful experience, the specific cellular ensembles encoding that memory spontaneously fired during NREM sleep, directly correlating with subsequent arousals. Using an AI-based calcium signal decoder, the team could even predict impending awakenings purely based on the reactivation of these memory cells.
Figure 1. Two-photon imaging combined with EEG/EMG recording revealed the correlation between memory cell activity and arousal
Subsequent optogenetic and chemogenetic manipulations confirmed this causal link: inhibiting these negative memory engrams blocked cue-induced arousals, while directly activating them during NREM sleep reliably induced wakefulness. These findings establish BLA memory engrams as the neural drivers of memory-dependent sleep regulation.
2. A Hierarchical Hippocampus-Amygdala Circuit
Given the central role of the hippocampus and BLA in emotional memory, the team investigated how these two regions coordinate during sleep. They found that memory engram cells in the dorsal dentate gyrus (dDG) of the hippocampus are both necessary and sufficient for experience-dependent sleep regulation as those in the BLA.
Through concurrent dual-region manipulations—optogenetically activating dDG engrams while chemogenetically inhibiting BLA engrams—the researchers elucidated the hierarchical organization of this circuit. The findings reveal that the hippocampus serves as an upstream prerequisite, coordinating the reactivation of BLA engram cells to translate memory reactivation into distinct sleep-wake transitions.
3. Positive Memories Promote Sleep Maintenance
In contrast to the arousal-promoting effects of negative memories, reactivating positive memories stabilized and promoted sleep. When the researchers activated BLA engrams associated with a rewarding social experience, the mice displayed significantly enhanced NREM sleep continuity, with fewer awakenings and longer continuous sleep bouts. Furthermore, when subjected to random, disruptive background noise, activating these positive engrams provided a profound "sleep-protective" effect, allowing the mice to sleep soundly through the disturbances.
Whole-brain projection mapping and activity analyses revealed the potential circuit mechanisms: positive and negative memory engram cells exhibit stronger functional connectivity with NREM sleep-promoting circuits and arousal-promoting circuits, respectively.
Figure 2. The sleep-promoting effect of positive memory reactivation and the underlying circuit mechanisms
4. Sleep fragmentation in the chronic stress model is induced by the reactivation of stressful memory
Finally, the research team extended this discovery to decipher the mechanisms of pathological sleep disturbances caused by chronic stress. In a depression-like mouse model induced by chronic immobilization stress (CIS), mice exhibited typical sleep fragmentation. In vivo calcium activity recordings revealed that during NREM sleep in CIS mice, the spontaneous reactivation of engram cells associated with the restraint memory was highly correlated with frequent arousals.
Remarkably, targeted chemogenetic inhibition of these specific negative memory engrams during sleep completely reversed sleep fragmentation, restoring normal sleep continuity and correcting aberrant activity across the brain's sleep circuits. These results indicate that the reactivation of negative memories during sleep is the critical mechanism underlying sleep disturbances caused by chronic stress.
Figure 3. The model of memory reactivation regulating sleep
This study reshapes our understanding of the sleep-memory interplay. Rather than acting as a passive incubator for memory consolidation, sleep is a highly adaptive state continuously sculpted by daytime experiences. By reactivating memories, the brain primes animals to adopt a vigilant sleep strategy after facing danger or allows them to enjoy deep, restorative rest after safe, rewarding experiences. These findings offer a novel framework for understanding sleep disturbances caused by chronic stress and provide new therapeutic targets.
The study was led by Dr. Bo Lei from the Beijing Academy of Artificial Intelligence (BAAI), and Professor Yi Zhong from the School of Life Sciences, 蜜桃app. Ph.D. student Menghan Yu was the first author of the paper. Junjie Wang, Zihan Zhai, and Ruoyi Huang were co-second authors, with significant contributions from Guofan Fan, Xiaoya Su, Yijun Niu, Haochen Zhu, Jiayi Chen, Grace Jiang, and Tian Zhang. The work was supported by the National Natural Science Foundation of China, the Sci-Tech Innovation 2030 "Brain Science and Brain-Like Intelligence Technology" Major Project, and the 蜜桃app-Peking Joint Center for Life Sciences.
Link to paper:
Editors: Li Han, John Paul Grima
