Neural replay, the brain’s mechanism of briefly re-activating patterns of neural activity in the same sequence they occurred during a prior experience, has long been considered a hallmark of the hippocampus, tightly tied to spatial navigation and memory consolidation. A new technique developed by researchers at the Institute for Basic Science (IBS) in Daejeon, South Korea, and KAIST now shows that replay is far more widespread than previously appreciated, and that it can be detected in brain regions as different from the hippocampus as the visual cortex, even during passive movie viewing.
The method, described in Nature Communications on July 4, overcomes a fundamental limitation of existing replay detection approaches. Traditional methods rely on what are called “place cells”, hippocampal neurons that fire when an animal occupies a specific location, to construct a spatial template against which to compare later neural activity. If the same sequence of place cells fires during rest as during running, replay is inferred. But this approach excludes everything that isn’t a place cell, and everything that happens outside the hippocampus.
“Our method estimates the likelihood of spike sequences based on pairwise firing-order probabilities observed during active behavior, regardless of what the cells encode,” said senior author Min Whan Jung, a professor at KAIST and associate director of the IBS Center for Synaptic Brain Dysfunctions. “It works for any cell type and any brain region.”
How it works
The core innovation is statistical. During a behavior period, say, a rat running on a track, the method calculates for every pair of recorded neurons the probability that neuron A fires before neuron B. This creates a pairwise firing-order probability matrix. During a later period, rest, sleep, or passive viewing, the same calculations are performed. If the post-behavior pairwise probabilities match the behavior-period probabilities more often than chance, replay is occurring.
The approach is fundamentally different from template-matching methods, which require the experimenter to define what constitutes a “replay event” in advance and to set parameters accordingly. The likelihood-based method makes no such assumptions. It is parameter-free in the sense that it does not impose a predefined detection threshold on the pattern itself, it simply asks whether the observed spike-order statistics are more consistent with the behavioral template than with chance.
The researchers validated the method using three independent data types: simulated spike trains with known replay ground truth, single-unit recordings from rats running on linear tracks, and calcium imaging data from mice. In all three cases, the method showed strong agreement with conventional replay detection metrics, while extending detection to cell types and brain regions that previous methods could not reach.
Beyond the hippocampus
The most striking demonstration came from an application that would have been impossible with traditional methods. The team recorded neural activity from the hippocampus and primary visual cortex of head-fixed mice as they passively viewed naturalistic movie clips, a stimulus that has no spatial component and therefore no place cells to template-match against.
The method detected significant replay in both the hippocampus and the visual cortex during post-movie rest periods. Replay in the hippocampus during non-spatial tasks has been documented before, but only with specialized analytical approaches. Replay in the visual cortex, by contrast, is a finding that challenges the traditional view of replay as a primarily hippocampal phenomenon dedicated to spatial memory consolidation.
“This suggests that structured replay is a general property of neural circuits, not a specialization of the hippocampus,” said first author Namjung Huh, a postdoctoral researcher at IBS. “The visual cortex is replaying patterns of activity that were established during movie viewing. That implies replay may be a fundamental mechanism for consolidating any sequentially structured experience, not just spatial ones.”
What replay is and why it matters
Neural replay was first discovered in the hippocampus in the 1990s, when researchers observed that place cells active during a rat’s run on a track would re-activate in the same temporal order during subsequent sleep or quiet rest, but compressed in time, a phenomenon sometimes described as the brain “rehearsing” recent experiences. Decades of research have shown that disrupting replay impairs memory consolidation, and that replay strength correlates with learning.
The discovery that replay is not limited to the hippocampus nor to spatial tasks has broad implications. It suggests that sensory cortices, the brain regions that process what we see, hear, and feel, may consolidate their own experiences through replay, independently of the hippocampus. This could mean that memory consolidation is a more distributed process than the dominant hippocampal-centric model suggests.
It also opens new experimental possibilities. Because the method works with any cell type and any brain region, researchers can now ask whether replay occurs in the motor cortex after learning a new skill, in the auditory cortex after hearing a melody, or in the prefrontal cortex after a decision-making task. Each of these questions was previously difficult or impossible to address with existing tools.
Caveats and limitations
The method detects replay patterns but does not, on its own, establish whether those patterns are functionally important for memory consolidation. It is a detection tool, not a causal intervention. Disruption experiments, where identified replay events are selectively silenced, would be needed to link visual cortex replay to visual memory. The authors also note that the method requires sufficiently large spike counts to construct reliable pairwise probability matrices, and may be less sensitive for very sparse neural populations.
Still, for a field that has spent three decades tethered to hippocampal place cells, the ability to detect replay anywhere in the brain, without a spatial map, without cell-type restrictions, and without predefined pattern templates, represents a methodological step change.
Source: Huh N, Yun I, Lee JW, Jung MW. A likelihood-based method for identifying replay from spike sequences. Nature Communications (2026). DOI: 10.1038/s41467-026-74822-2