Summary: Researchers have resolved a central question in neuroscience: how the brain can remain adaptable enough to learn new information while preserving older memories. The study shows that the hippocampus reuses a dedicated core group of neurons to route different memories through distinct activity patterns, preventing new learning from overwriting established information.
Using simultaneous recordings from hundreds of individual neurons in freely moving mice, the team identified a hippocampal “memory switchboard” that separates incoming and outgoing signals by directing them through divergent firing patterns. This mechanism protects long-term memories as the brain continues to incorporate new experiences.
Key Facts
- The plasticity–stability challenge: Brains must balance flexibility for learning with stability for preserving prior knowledge. How neural circuits achieve this balance has remained an open question.
- CA1 core hub: About one in four neurons in the CA1 region of the hippocampus functions as a shared hub linking upstream inputs and downstream outputs.
- Divergent firing channels: Those CA1 hub neurons receive fast-varying input from CA3 but transmit to the retrosplenial cortex (RSC) using distinct firing patterns, effectively creating separate channels for incoming and outgoing information.
- Simultaneous multi-region recording: The research captured activity from DG, CA3, CA2, CA1 and RSC using high-density electrodes while mice ran a rewarded track, allowing behavioral alignment of spiking activity across regions.
- Nighttime replay: The same CA1 hub neurons active during waking behavior also replayed waking activity during sleep-associated sharp-wave ripples, supporting memory consolidation and keeping hippocampus-to-cortex pathways engaged.
- Clinical and AI relevance: The circuit blueprint offers insight into memory circuit failures such as Alzheimer’s disease and suggests a biological strategy to reduce “catastrophic forgetting” in artificial intelligence systems.
Source: NYU Langone
Overview
A new mouse study led by NYU Langone Health scientists shows that the brain can reuse a core set of hippocampal neurons to handle many different memories without mixing or erasing prior information. The findings, published online May 13 in Nature, explain how the hippocampus can transform variable inputs into stable outputs along the hippocampal–retrosplenial axis.
The investigators examined a functional chain that includes CA3 (a hippocampal subregion that supplies rapidly changing input), CA1 (a central hippocampal hub), and the retrosplenial cortex (a cortical area involved in navigation and scene representation). They found that a minority of CA1 neurons carried most incoming signals from CA3, but those same neurons fired in a different, nonoverlapping pattern when conveying information to RSC. That separation of firing patterns lets the same neurons participate in both input processing and output transmission without blending the channels—analogous to an electronic switchboard routing calls through different circuits.
This arrangement allows the retrosplenial cortex to maintain a stable representational map while upstream hippocampal regions continue to encode new experiences. The team observed that these CA1 hub neurons remained active in sleep, replaying waking sequences during sharp-wave ripples. Nighttime replay likely helps transfer and consolidate memory traces from hippocampus to neocortex, keeping the pathway functional across behavioral states.
Co-lead authors Joaquín Gonzalez, PhD, and Mihály Vöröslakos, MD, PhD, emphasized that the key advance was the ability to monitor hundreds of individual neurons across multiple deep-brain and cortical areas simultaneously in freely moving animals. That broad, simultaneous sampling made it possible to detect low-dimensional communication subspaces—distinct patterns of joint activity that map inputs to outputs—within CA1.
Co-senior authors Zhe S. Chen, PhD, and György Buzsáki, MD, PhD, note that identifying a switchboard-like motif deep in the hippocampus offers a testable framework for understanding memory circuit breakdown in dementia and for designing AI systems that can learn continuously without catastrophic forgetting.
For the experiments, six mice were trained to shuttle along a straight track for water rewards. High-density electrode arrays recorded spiking activity across dentate gyrus, CA3, CA2, CA1 and RSC while the animals moved and while they slept. The researchers matched individual spikes to precise locations and behaviors, then used linear dimensionality-reduction methods to identify distinct communication subspaces linking hippocampal inputs to cortical outputs. Reactivation during sleep revealed that CA1–CA3 subspace patterns correlated with replay, whereas CA1–RSC patterns showed different reactivation dynamics consistent with a balance between plasticity and stability.
While the results provide a clear circuit-level mechanism in mice under controlled conditions, the authors caution that further work is needed to determine how these motifs operate in more naturalistic settings and in the human brain.
Funding: Supported by NIH grants RF1DA056394, P50MH132642, R01MH122391, and U19NS107616.
Contributors include Joaquin Gonzalez, Mihály Vöröslakos, Deren Aykan, Nina Soto, Noam Nitzan, Rachel Swanson, Mursel Karadas, Zhe Sage Chen and György Buzsáki, among others at NYU Langone.
Key Questions Answered:
A: The neurons change their temporal firing patterns rather than switching which cells are active. The CA1 hub neurons use different activity rhythms to represent incoming signals from CA3 versus outgoing messages to the retrosplenial cortex, keeping the two channels separate even though they involve overlapping cells.
A: Sleep recordings reveal replay events—sharp-wave ripples—when waking activity is reactivated and consolidated. Observing the same hub neurons replay waking patterns at night shows how the hippocampus can maintain a pathway to cortex that supports long-term storage.
A: The study suggests a biological blueprint to mitigate catastrophic forgetting: overlapping neuronal pools can support distinct input–output mappings by using separable activity subspaces. AI architectures that mimic this separation may update continuously without erasing prior knowledge.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full by the editorial team.
- Additional context and clarifications were added by staff editors.
About this memory research news
Author: Shira Polan
Source: NYU Langone Health
Contact: Shira Polan – NYU Langone Health
Image: Image credited to Neuroscience News
Original Research: Closed access. “Subspace communication in the hippocampal–retrosplenial axis” by Joaquin Gonzalez, Mihály Vöröslakos, Deren Aykan, Nina Soto, Noam Nitzan, Rachel Swanson, Mursel Karadas, Zhe Sage Chen & György Buzsáki. Nature
DOI: 10.1038/s41586-026-10481-z
Abstract
Subspace communication in the hippocampal–retrosplenial axis
The ability of hippocampal circuits to convert upstream inputs into downstream outputs underlies navigation and memory, but the circuit mechanisms that allow flexible adaptation to experience are not fully understood. This study uses large-scale (up to 1,024 channel) recordings across the hippocampal–retrosplenial cortex circuit in behaving mice to access spiking activity in dentate gyrus, CA3, CA2, CA1 and RSC simultaneously. Applying a linear dimensionality-reduction technique (partial canonical correlation analysis), the authors identify low-dimensional communication subspaces between pairs of regions while accounting for third-area influences.
These subspaces reveal distinct input–output transformations within CA1, linking upstream hippocampal activity to downstream cortical targets. Intrinsic firing properties and anatomical position constrain subspace membership, and members map to deep sublayers of the CA3–CA1–RSC axis during spatial and nonspatial tasks. Overlapping neuronal pools can be recombined into different subspaces to support distinct interareal interactions across changing experiences and brain states. Reactivation of CA1–CA3 subspaces, but not CA1–RSC subspaces, during post-experience sleep correlates with replay, reflecting a balance between plasticity and stability in input–output transformations along the hippocampal–retrosplenial axis. The findings support a model in which hippocampal–neocortical communication flexibly reconfigures predetermined circuit motifs to encode experience.