January 14, 2026
Along with his team, Dr. Mark Brandon, a researcher at the Douglas Research Centre and Associate Professor at McGill University’s Department of Psychiatry, has just published an article investigating how the hippocampus learns from experience to predict future rewards. The article, entitled “Predictive Coding of Reward in the Hippocampus,” was published on Jan 14, 2026 in Nature. This work is a result of collaborations with fellow-Douglas researcher Dr. Sylvain Williams, and Dr. Cengiz Pehlevan’s group at Harvard University.
Key Takeaways
The study reveals that the hippocampus does far more than map spatial environments. It dynamically reorganizes its activity to anticipate future rewards. Over time, the representation of reward in hippocampal neurons decreases, while the representation of cues that predict rewards becomes stronger. Individual neurons gradually shift their activity backward in time, moving from encoding the reward itself to encoding the events that predict the reward. These changes are consistent with predictions from a Temporal Difference learning model, which suggests that hippocampal dynamics align with principles of reinforcement learning.
Why This Matters
Understanding how the hippocampus predicts future rewards is crucial for advancing our knowledge of learning and memory. This discovery bridges neuroscience and artificial intelligence by providing insights into how predictive coding supports decision-making and adaptive behaviour. It also has potential implications for understanding and treating disorders that involve reward processing, such as addiction and depression.
What We Did
Researchers recorded the activity of CA1 neurons in mice using head-mounted calcium imaging over several weeks. During this period, the mice performed a delayed nonmatch-to-location touchscreen task that required memory and decision-making. The team applied advanced analytical methods to track how representations of reward and pre-reward cues evolved over time. They also developed a computational model based on Temporal Difference Reinforcement Learning (TDRL) to account for the observed neural dynamics.
What We Found
The results show that reward encoding in the hippocampus declines with experience at both the population and single-cell levels. In contrast, the representation of reward-predictive cues, such as screen touches and approach movements, becomes increasingly prominent. Approximately 21% of neurons initially tuned to reward shifted their activity backward to encode predictive cues. Simulations using a Temporal Difference learning model reproduced these backward shifts, confirming that predictive coding could emerge from reward prediction error signals.
What It Means
These findings indicate that the hippocampus is not a static spatial map but a predictive engine that reorganizes its coding to anticipate future rewards. By shifting its representation toward cues that forecast rewards, the hippocampus optimizes learning and behaviour. This dynamic mirrors the reward prediction error signals observed in dopamine neurons, suggesting that predictive learning involves a broader network that includes both the hippocampus and dopaminergic circuits.
What’s Next
Future research will examine how these hippocampal dynamics interact with the dopaminergic system and other brain regions. Studies in humans could determine whether similar predictive coding occurs during complex learning tasks. These insights may inspire the development of biologically informed artificial intelligence models [SD1][MY2][MB3] that incorporate hippocampal-like predictive mechanisms. Additionally, understanding these processes could lead to new approaches for treating memory and reward-related disorders.
The Study
Yaghoubi, M., Kumar, M.G., Nieto-Posadas, A. et al. Predictive coding of reward in the hippocampus. Nature (2026). https://doi.org/10.1038/s41586-025-09958-0