Summary: A new preclinical study shows the hippocampus does more than store memories: it actively reorganizes them to predict future rewards. By tracking hippocampal activity in mice over several weeks, researchers observed neurons shift their firing from the moment of reward to the moments that lead up to it, effectively building a predictive model of upcoming outcomes. These findings offer a fresh framework for understanding why learning and decision-making commonly decline early in Alzheimer’s disease.
Source: McGill University
Key Facts:
- Predictive mapping: The hippocampus continually updates an internal model of the environment, moving neural responses from the time of reward toward the events that precede reward.
- Advanced imaging: Researchers used longitudinal calcium imaging to mark active neurons and follow the same cells across weeks, revealing slow learning processes that standard electrode recordings often miss.
- Beyond Pavlovian conditioning: While simple reward associations involve ancient brain circuits, this study demonstrates the hippocampus contributes a context-rich, memory-driven form of anticipation.
- Alzheimer’s implications: Disruption of this predictive signaling may underlie early deficits in learning and decision-making seen in Alzheimer’s disease.
A preclinical study published in Nature provides direct evidence that the hippocampus reorganizes memory representations to anticipate future outcomes. The research, led by the Brandon Lab at McGill University in collaboration with colleagues at Harvard University, tracked hippocampal activity as mice learned a cognitively demanding, reward-based task over several weeks.
“The hippocampus is often described as the brain’s internal model of the world,” said senior author Mark Brandon, Associate Professor in McGill’s Department of Psychiatry and researcher at the Douglas Research Centre. “Our data show that this model is not fixed. It updates day by day as the brain learns from prediction errors. As outcomes become expected, hippocampal neurons begin to fire earlier, reflecting what will happen next.”
A new view of learning in action
The hippocampus is well known for building cognitive maps of physical space and storing past experiences. Scientists have also observed that these maps change over time, but those changes were often assumed to be random or noise. This study demonstrates that the reorganization is structured and meaningful: neural representations shift to encode predictive information about forthcoming rewards.
Using calcium imaging—techniques that make active neurons glow—the team could identify and follow the same hippocampal cells across multiple weeks of training. Unlike traditional electrode methods, which generally track neurons only for short sessions, these imaging approaches capture slow, progressive changes in neural tuning as learning unfolds. The researchers found that activity which initially peaked at the reward gradually migrated backward in time, ultimately appearing before mice reached the reward location.
This backward shift in activity suggests the hippocampus actively transforms memory traces into anticipatory signals. Rather than merely registering that a reward occurred, hippocampal networks learn to predict the reward based on preceding cues and context, improving the animal’s ability to anticipate and act upon upcoming outcomes.
Insights into learning and Alzheimer’s disease
Classic studies of reward learning, such as Pavlov’s conditioning experiments, illustrated how simple cues become associated with outcomes via primitive circuits. The current findings extend that view by showing the hippocampus provides a richer, context-dependent prediction system that integrates memory and spatial information to forecast rewards.
Alzheimer’s disease patients experience early impairments not only in remembering past events but also in learning from new experiences and making effective decisions. By revealing how a healthy hippocampus converts memories into predictions about the future, this work suggests a mechanism that could fail in early Alzheimer’s. Understanding how predictive hippocampal signaling breaks down may point to novel approaches for protecting or restoring learning and decision-making abilities.
Editorial Notes
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full.
- Additional context was added by staff writers.
About this Hippocampus Research
- Source: McGill University
- Contact: Keila DePape
- Image: The image is in the public domain
Original Research: “Predictive Coding of Reward in the Hippocampus” by Mohammad Yaghoubi, Mark Brandon, and colleagues, published in Nature. DOI: 10.1038/s41586-025-09958-0. This research received support from Fonds de recherche du Québec – Santé and the Canadian Institutes of Health Research.
About the Brandon Lab
The Brandon Lab, established in 2015 at the Douglas Research Centre at McGill University by Professor Mark Brandon, studies the core mechanisms of memory—how memories are encoded, stored, and retrieved—and how these processes fail in Alzheimer’s disease. The lab aims to identify strategies to protect and restore memory by understanding the neural circuits that support learning and prediction.
Abstract
Anticipating future outcomes is a central function of the brain. This requires learning the states of the world and the transitions between them. In rodents, the hippocampal spatial cognitive map is considered one such internal model, but evidence of reward sensitivity and predictive coding indicates its role is broader than purely spatial representation. How reward-related hippocampal representations evolve with extended experience has been unclear. By tracking hippocampal activity over weeks as mice learned a reward-based task, the study shows that hippocampal representations become increasingly predictive of reward. Population-level encoding of reward and the proportion of reward-tuned neurons decreased with experience, while representation of features that precede reward increased. Longitudinal tracking of individual neurons revealed a gradual shift in activity from encoding reward itself to representing preceding task features, consistent with a backward-shifted reorganization that anticipates reward. A temporal difference model of place fields reproduced these effects. These results highlight the dynamic nature of hippocampal representations and their role in learning through prediction of future outcomes.