Communication Subspaces in the Hippocampal-Prefrontal Circuit
How do brain regions coordinate memory-guided behavior? Behavior lines up with the activity patterns hippocampus and prefrontal cortex share, not with either region alone.
Published in: eNeuro, September 2025, 12(9) · Authors: Ryan A. Young, Justin D. Shin, Ziyi Guo, Shantanu P. Jadhav (Brandeis University)
📄 Read Paper · 💻 Code · 📊 Data (DANDI 000978)
🎯 TL;DR
Task behavior in rats lined up with low-dimensional communication subspaces, the activity patterns shared between hippocampal CA1 and prefrontal cortex (PFC), rather than with local activity in either region. Brain rhythms and running speed reshaped that shared channel.
❓ The Problem
Memory-guided decisions require coordination between the hippocampus (memory) and prefrontal cortex (decisions). Rhythmic network states have been theorized to help brain regions communicate, but it was unclear how these oscillations shape the communication subspaces that carry activity between regions, and how those subspaces in turn relate to behavior.
🔬 Key Finding
- Shared activity predicts behavior; local activity doesn’t. Activity in the CA1–PFC shared subspace predicted task behaviors, while activity confined to either region alone did not.
- Rhythms sculpt the channel. Network oscillations (theta rhythms and sharp-wave ripples) and running speed modulated the structure and performance of the communication subspace. Theta power mattered more than CA1–PFC theta coherence: coherence gave no measurable gain in prediction once theta power was accounted for.
- The shared activity forms rings. Visualized with manifold techniques, shared CA1–PFC activity organizes into ring-like structures that track trajectories through the maze.
🔧 Our Approach
- Simultaneously recorded 394 CA1 and 254 PFC neurons with multitetrode drives in 7 rats learning a continuous-alternation spatial memory task on a W-maze, across eight training sessions in a single day.
- Found communication subspaces with reduced-rank regression and canonical correlation analysis (CCA), splitting activity into what is shared between regions and what stays private to one.
- Predicted behavior from the shared and local activity patterns with Bayesian linear models, and compared high- versus low-rhythm network states.
Techniques: multitetrode recording · reduced-rank regression · canonical correlation analysis · Bayesian GLMs (TuringGLM.jl) · manifold visualization
💡 What It Means
This work bridges two ideas about how brain regions communicate: rhythmic coordination, where oscillations synchronize regions, and communication subspaces, where only certain activity patterns pass between regions. Here the rhythms reshape the subspace. Memory-guided behavior appears to be driven by shared CA1–PFC interactions that are dynamically modulated by oscillatory states, not by either region alone.
Open questions:
- Do these communication patterns change during sleep, when memories are consolidated?
- How do the rings form during learning: are they present from the start, or do they emerge?
- Can manipulating the communication subspace directly influence behavior?
📚 More Information
- Paper on eNeuro (DOI 10.1523/ENEURO.0336-24.2025)
- Analysis code: JadhavLab/CommSub on GitHub
- Dataset: DANDI Archive 000978
Citation
@article{young2025communication,
title = {Hippocampal--Prefrontal Communication Subspaces Align with Behavioral
and Network Patterns in a Spatial Memory Task},
author = {Young, Ryan A. and Shin, Justin D. and Guo, Ziyi and Jadhav, Shantanu P.},
journal = {eNeuro},
volume = {12},
number = {9},
year = {2025},
doi = {10.1523/ENEURO.0336-24.2025}
}