A new computational model sheds light on how place cells in the hippocampus can contribute to forming various types of episodic memories, even when spatial elements are absent.
Nearly 50 years ago, scientists identified cells in the hippocampus that encode memories related to specific locations. These cells also significantly contribute to episodic memories, which are memories of past events. While the encoding of spatial memory by place cells is well understood, their role in encoding episodic memories has remained unclear.
Researchers at MIT have created a model to explain how place cells might be engaged to create episodic memories without a spatial context. Their findings indicate that place cells, along with grid cells in the entorhinal cortex, function as a framework for interconnecting memories as a cohesive series.
“This is an initial model of the entorhinal-hippocampal circuit for episodic memory. It serves as a starting point for deeper exploration into episodic memory,” remarks Ila Fiete, a professor of brain and cognitive sciences at MIT, who is part of the McGovern Institute for Brain Research, and the senior author of the research.
The model effectively captures several characteristics of biological memory systems, including vast storage capability, the gradual fading of older memories, and the phenomenon where participants in memory competitions manage to retain huge amounts of information using “memory palaces.”
Lead authors of the study include MIT research scientist Sarthak Chandra and Sugandha Sharma PhD ’24. The study is published in Nature, with contributions also from Rishidev Chaudhuri, an assistant professor at the University of California at Davis.
A Memory Index
To encode spatial memory, place cells in the hippocampus coordinate closely with grid cells, a unique type of neuron that fires in a geometric pattern at various locations within a space. This collective of grid cells creates a triangular lattice structure that represents physical environments.
Besides aiding in recalling places we’ve visited, these neural circuits connecting the hippocampus and entorhinal cortex are vital for navigating unfamiliar locations. Evidence from patients shows that these circuits are crucial for forming episodic memories, which often include spatial elements but fundamentally consist of events—such as how one celebrated their last birthday or what they had for lunch yesterday.
“The same circuits in the hippocampus and entorhinal cortex are involved in both spatial and general episodic memories,” Fiete explains. “The intriguing question is what connects spatial and episodic memory that allows them to coexist within the same circuitry?”
There are two main theories regarding this functional overlap. One suggests that this circuit evolved to primarily store spatial memories, as remembering the locations of food or predators is crucial for survival. In this view, episodic memories are recorded incidentally as a side effect of spatial memory.
The second theory posits that the circuit is designed to store episodic memories, with spatial memory being one dimension of various episodic experiences.
In this study, Fiete and her team proposed a third theory: that the unique tiling pattern of grid cells and their connections to the hippocampus are equally essential for both episodic and spatial memory. They based their new model on previous computational models from the lab that replicate how grid cells process spatial information.
“We reached a level of understanding about how grid cell circuits function, which motivated us to explore how grid cells interact with the larger memory circuit that encompasses the hippocampus,” Fiete mentions.
In this novel model, the researchers propose that grid cells connecting with hippocampal cells may act as a foundational structure for storing both spatial and episodic memories. Each activation pattern in the grid signifies a “well,” positioned at regular intervals. While these wells do not contain the actual memory data, they serve as directional pointers to specific memories, which are stored in the synapses connecting the hippocampus and sensory cortex.
When a memory is triggered from partial inputs, the interactions between grid and hippocampal cells navigate the circuit to the nearest well, allowing the state at that well to connect with the relevant sensory cortex components, thereby retrieving the complete memory details. The sensory cortex has a greater capacity than the hippocampus, enabling it to retain extensive amounts of memory.
“Conceptually, we can visualize the hippocampus as a pointer network. It operates like an index that is capable of reconstructing a memory from partial information, which subsequently directs attention to the sensory cortex where the original experiences occurred,” Fiete suggests. “The scaffold solely contains an abstract index of these states, not the content itself.”
Also, sequential events can be interconnected: every well in the grid cell-hippocampal network effectively retains the necessary information to trigger the forthcoming well, enabling memories to be recalled in the proper sequence.
Modeling Memory Structures
This new model from the researchers replicates numerous phenomena related to memory much more accurately than traditional models based on Hopfield networks—which are a kind of neural network capable of storing and retrieving patterns.
Although Hopfield networks provide valuable insights into the formation of memories through strengthened neuronal connections, they do not precisely emulate biological memory processes. In Hopfield approaches, each memory is retrieved in fine detail until the memory limit is reached. Once full, no new memories can be stored, and adding further memories leads to the loss of all previously stored information. This issue, known as the “memory cliff,” does not reflect the natural brain’s tendency to gradually forget older memories while new ones are adopted.
The MIT team’s model successfully incorporates findings from extensive recordings of grid and hippocampal cells in rodents as they navigate and forage across diverse environments. Moreover, it provides insights into the mechanics behind a memorization method called the memory palace technique. Memory competition participants often memorize a shuffled deck of cards by associating each card with a specific location in a familiar environment, like a childhood home. When recalling the cards, they mentally traverse the space, visualizing each card in its designated location. Interestingly, associating the memory of cards with locations enhances recall strength and reliability.
The computational model developed by the MIT team adeptly carried out such tasks, indicating that memory palaces leverage the memory circuit’s strategy of linking inputs to the hippocampal scaffold. Essentially, older memories stored in the sensory cortex can serve as a foundation for new memories, allowing for the retention and recall of significantly more sequential items than would otherwise be feasible.
Looking ahead, the researchers intend to expand their model to examine how episodic memories transition into cortical “semantic” memories—knowledge of facts independent of the specific situations they were acquired in (e.g., knowing that Paris is France’s capital), the definition of episodes, and how brain-inspired memory frameworks can be integrated into contemporary machine learning.
This research was supported by the U.S. Office of Naval Research, the National Science Foundation’s Robust Intelligence program, the ARO-MURI award, the Simons Foundation, and the K. Lisa Yang ICoN Center.
