Hippocampal neurons shift their activity backward in time to anticipate rewards

Recent experimental findings suggest that the hippocampus, the brain region primarily associated with memory and navigation, actively reorganizes its neural patterns to anticipate future events. Researchers observed that as mice learned to navigate a complex task, the neural signals associated with a reward shifted backward in time to predict the outcome before it happened. These results were published in the journal Nature.

The hippocampus is a seahorse-shaped structure located deep within the temporal lobes of the brain. Neuroscientists have recognized for decades that this region is essential for forming new memories. It is also responsible for creating a cognitive map. This internal representation allows an organism to visualize its environment and navigate through space.

Biologists have traditionally viewed the cognitive map as a relatively static record of the environment. Under this view, the hippocampus encodes features such as landmarks, borders, and the location of resources. However, survival requires more than just a record of the past. An animal must use its prior experiences to predict where food or safety will be located in the future.

This necessity leads to the theory of predictive coding. This theory suggests that the brain is constantly generating models of the world to estimate future outcomes. When an outcome matches the prediction, the brain learns that its model is correct. When an outcome is unexpected, the brain must update the model.

While this theory is widely accepted in computational neuroscience, observing the physical reorganization of cells in the hippocampus over long periods has been a technical challenge. Most neural recording technologies can only track brain activity for short durations. This limitation makes it difficult to see how internal maps evolve as learning consolidates over weeks.

Mohammad Yaghoubi, a researcher at McGill University, aimed to bridge this gap. Working with senior author Mark Brandon at the Douglas Research Centre, Yaghoubi designed an experiment to track specific neurons across an extended timeframe. They sought to determine if the hippocampal map restructures itself to prioritize the prediction of rewards.

The research team employed a sophisticated imaging technique known as calcium imaging. They injected a modified virus into the brains of mice. This virus caused neurons to express a fluorescent protein that glows when calcium enters the cell, which happens when a neuron fires.

The researchers then implanted a gradient refractive index lens, a tiny microscope component, above the hippocampus. This setup allowed them to attach a miniature camera, weighing only a few grams, to the head of the mouse. The camera recorded the fluorescence of hundreds of individual neurons while the animal moved freely.

Because this method relies on optical imaging rather than physical electrodes, it is less invasive to the tissue over time. This stability allowed Yaghoubi and his colleagues to identify and monitor the exact same neurons day after day for several weeks. They could then correlate specific cellular activity with the animal’s behavior during learning.

The mice were trained to perform a task known as “delayed nonmatching-to-location” inside an automated chamber. The apparatus featured a touch-sensitive screen at one end and a reward dispenser at the other. The task required the mouse to initiate a trial and then observe a sample location lighting up on the screen.

After a short delay, the screen displayed the original location alongside a new, novel location. To receive a reward, the mouse had to ignore the familiar spot and touch the new location. The reward was a small amount of strawberry milkshake delivered at the opposite end of the chamber. This task is cognitively demanding because it requires the animal to hold information in working memory and apply a specific rule.

At the beginning of the training, the researchers noted that a distinct population of hippocampal neurons fired vigorously when the mouse received the milkshake. These cells appeared to be tuned specifically to the experience of consuming the reward. The neural map at this stage was heavily focused on the outcome itself.

As the mice repeated the task over weeks and their performance improved, the neural patterns began to change. The researchers observed a phenomenon described as backpropagation of neural tuning. The cells that originally fired only upon receiving the reward began to fire earlier in the sequence of events.

“What we found was surprising,” said Brandon. “Neural activity that initially peaked at the reward gradually shifted to earlier moments, eventually appearing before mice reached the reward.”

By the time the mice had mastered the task, these specific neurons were firing while the animal was still approaching the reward port. In some instances, the firing shifted all the way back to the moment the mouse made the correct choice on the touchscreen. The cells had transformed from sensors of the present reward into predictors of the future reward.

The study also analyzed the activity of the neuronal population as a whole. In the early stages of learning, a large percentage of the recorded cells were dedicated to encoding the reward location. This resulted in an over-representation of the reward site in the mouse’s mental map.

As the weeks passed, the proportion of neurons tuned to the reward itself decreased. Simultaneously, the number of neurons encoding the approach and the choice period increased. The brain appeared to be efficient. Once the reward was predictable, fewer resources were needed to represent it. The cognitive effort shifted toward the actions required to obtain it.

This reorganization supports the idea that the hippocampus acts as a predictive device. The backward shift in timing allows the brain to signal an upcoming event based on the current context. This predictive signal likely helps guide the animal’s behavior, reinforcing the actions that lead to a positive outcome.

The researchers confirmed that this shift was not due to simple changes in the animal’s speed or position. They used statistical controls to ensure that the change in firing timing was a true remapping of the cognitive representation. The consistency of the findings across multiple animals suggests a fundamental biological mechanism.

“The hippocampus is often described as the brain’s internal model of the world,” said Brandon. “What we are seeing is that this model is not static; it is updated day by day as the brain learns from prediction errors. As outcomes become expected, hippocampal neurons start to respond earlier as they learn what will happen next.”

There are limitations to the study that warrant mention. The research was conducted on mice, and while the hippocampus is evolutionarily conserved, human cognition involves additional layers of complexity. Further research is necessary to confirm if identical cellular mechanisms drive predictive learning in the human brain.

Additionally, the study focused on a reward-based task. It remains to be seen if the hippocampus utilizes the same predictive backpropagation for negative or aversive outcomes. Future experiments will likely investigate whether the brain rewires itself similarly to predict threats or punishments.

The findings may have implications for understanding neurodegenerative disorders. Individuals with Alzheimer’s disease often exhibit disorientation and difficulty learning from new experiences. If the predictive coding mechanism in the hippocampus is disrupted, it could explain why patients struggle to anticipate consequences or navigate familiar environments.

By demonstrating that memory circuits are dynamic and predictive, this study offers a new perspective on how the brain interacts with time. The hippocampus does not merely archive the past. It actively reconstructs it to prepare for the future.

The study, “Predictive Coding of Reward in the Hippocampus,” was authored by Mohammad Yaghoubi, Andres Nieto-Posadas, Coralie-Anne Mosser, Thomas Gisiger, Émmanuel Wilson, Sylvain Williams, and Mark P. Brandon.