Unlocking the Mind: How a Single Brain Circuit Captures Our Memories of Places and Events

Almost 50 years ago, neurobiologists identified cells within the brain’s hippocampus that retain memories of specific places. These cells also significantly contribute to preserving memories of events, termed episodic memories. Although the functionality of how place cells encode spatial memory has been extensively detailed, the process by which they encode episodic memories has remained unclear.

A recent model established by MIT researchers elucidates how these place cells can be recruited to create episodic memories, even in the absence of a spatial dimension. According to this model, place cells, alongside grid cells located in the entorhinal cortex, serve as a framework that can be utilized to anchor memories as an interconnected series.

“This model represents a preliminary draft of the entorhinal-hippocampal episodic memory circuit. It provides a groundwork to build upon to comprehend the essence of episodic memory. That’s what I find truly exciting,” notes Ila Fiete, a professor of brain and cognitive sciences at MIT, a member of MIT’s McGovern Institute for Brain Research, and the senior author of the new research.

The model effectively mirrors several characteristics of biological memory systems, including a substantial storage capacity, gradual deterioration of older memories, and the capability of individuals who participate in memory competitions to retain vast amounts of information in “memory palaces.”

MIT Research Scientist Sarthak Chandra and Sugandha Sharma PhD ’24 are the principal authors of the study, which is published today in Nature. Rishidev Chaudhuri, an assistant professor at the University of California at Davis, is also a contributor to the paper.

An index of memories

For encoding spatial memory, place cells in the hippocampus collaborate closely with grid cells — a unique kind of neuron that activates in various locations, organized geometrically in a consistent pattern of repeating triangles. Together, a group of grid cells forms a mesh of triangles that portrays a physical area.

In addition to aiding us in remembering places we’ve visited, these hippocampal-entorhinal circuits also assist us in navigating unfamiliar areas. Evidence from human patients shows that these circuits are also essential for forming episodic memories, which may include a spatial aspect but primarily consist of experiences, such as how you celebrated your last birthday or what you had for lunch yesterday.

“The same hippocampal and entorhinal circuits are employed not only for spatial memory but also for general episodic memory,” Fiete states. “The question arises: what is the relationship between spatial and episodic memory that allows them to reside within the same circuit?”

Two theories have been proposed to explain this functional overlap. One theory suggests the circuit is specialized for storing spatial memories, as these types of memories — recalling where food was located or where predators were spotted — are vital for survival. According to this theory, the circuit encodes episodic memories as a secondary function of spatial memory.

An alternative theory posits that the circuit is primarily tailored to store episodic memories, but also encodes spatial memory because location is one element among many in episodic memories.

In this study, Fiete and her colleagues introduced a third possibility: that the unique tiling structure of grid cells and their interactions with the hippocampus are equally crucial for both types of memory — episodic and spatial. To formulate their new model, they built upon computational models that her laboratory has been developing over the last decade, which simulate how grid cells encode spatial data.

“We reached a stage where I felt we had a sufficient understanding of the grid cell circuit mechanisms, so it seemed like the appropriate time to explore the interactions between the grid cells and the broader circuit that includes the hippocampus,” Fiete remarks.

In the new model, the researchers proposed that grid cells interacting with hippocampal cells can function as a framework for storing either spatial or episodic memory. Each activation pattern within the grid represents a “well,” and these wells are regularly spaced apart. The wells do not contain the content of specific memories; instead, each one functions as a pointer to a particular memory, which is stored in the synapses between the hippocampus and the sensory cortex.

When the memory is later triggered from fragmentary pieces, interactions between grid and hippocampal cells drive the circuit state into the nearest well, and the state at the bottom of the well connects to the corresponding region of the sensory cortex to fill in the details of the memory. The sensory cortex is significantly larger than the hippocampus and can store extensive amounts of memory.

“Conceptually, we can view the hippocampus as a pointer network. It operates like an index that can be pattern-completed from a partial input, and that index subsequently directs toward the sensory cortex, where those inputs were originally experienced,” Fiete explains. “The framework does not include the content, it solely comprises an index of abstract scaffold states.”

Moreover, sequentially occurring events can be interconnected: Each well in the grid cell-hippocampal network efficiently retains the information necessary to activate the next well, facilitating the recall of memories in the correct order.

Modeling memory cliffs and palaces

The researchers’ new model remarkably replicates several memory-related phenomena more accurately than existing models derived from Hopfield networks — a type of neural network capable of storing and recalling patterns.

While Hopfield networks provide insight into how memories may be formed by strengthening neuron connections, they do not perfectly emulate how biological memory functions. In Hopfield models, each memory is recalled with complete accuracy until capacity is reached. At that juncture, no new memories can be created, and more detrimentally, trying to add further memories results in erasing all prior ones. This “memory cliff” does not accurately reflect what occurs in a biological brain, which tends to gradually lose details of older memories while continuously incorporating new ones.

The newly developed MIT model embodies findings from decades of recordings of grid and hippocampal cells in rodents made while the animals explore and forage in various environments. It also aids in elucidating the fundamental mechanisms behind a memorization method known as a memory palace. A common task in memory competitions involves memorizing the shuffled sequence of cards in one or several decks. Participants often achieve this by assigning each card to a specific location in a memory palace — a recollection of a childhood home or another familiar environment. When they need to retrieve the cards, they mentally walk through the house, visualizing each card in its designated location along the way. Counterintuitively, the added memory burden of linking cards with locations enhances recall strength and reliability.

The computational model developed by the MIT team was able to effectively perform such tasks, indicating that memory palaces leverage the memory circuit’s own strategy of associating inputs with a scaffold in the hippocampus, but on a different level: Long-accumulated memories reconstructed in the broader sensory cortex can now be utilized as a framework for new memories. This allows for the retention and retrieval of many more items in a sequence than would typically be feasible.

The researchers now intend to expand on their model to investigate how episodic memories might be transformed into cortical “semantic” memory, or the memory of facts separated from the specific context in which they were obtained (for instance, Paris is the capital of France), how episodes are defined, and how brain-inspired memory models could be incorporated into contemporary machine learning.

The research received funding from the U.S. Office of Naval Research, the National Science Foundation under the Robust Intelligence program, the ARO-MURI award, the Simons Foundation, and the K. Lisa Yang ICoN Center.