Source: Science daily
Researchers propose a groundbreaking hypothesis that brain cells called astrocytes play a crucial role in memory storage, potentially explaining the brain’s extraordinary storage capabilities.
The Hidden Role of Astrocytes in Memory
The human brain contains approximately 86 billion neurons, long considered the primary cells responsible for memory and information processing. However, another type of brain cell—astrocytes—may hold the key to understanding the brain’s vast memory capacity.
A new study from MIT researchers suggests that astrocytes, traditionally viewed as mere support cells, may actively participate in memory storage through a mechanism called dense associative memory. This model could explain how the brain stores far more information than previously thought possible with neurons alone.
From Support Cells to Memory Partners
Astrocytes are star-shaped glial cells with long extensions that interact with millions of neurons. Historically, they were believed to perform only housekeeping functions, such as:
- Cleaning up cellular debris
- Providing nutrients to neurons
- Regulating blood flow
But recent research indicates that astrocytes may do much more. Studies disrupting astrocyte-neuron connections in the hippocampus—a brain region critical for memory—have shown significant memory impairments, suggesting these cells play an active role in cognition.
How Astrocytes Communicate with Neurons
Unlike neurons, astrocytes don’t transmit electrical signals. Instead, they use calcium signaling to communicate with each other and influence neuronal activity. When neurons fire, astrocytes detect this activity and adjust their calcium levels, which in turn triggers the release of gliotransmitters—chemical signals that modulate synaptic strength.
This bidirectional communication creates a feedback loop between neurons and astrocytes, raising the question: What kind of computations do astrocytes perform with this information?
The Dense Associative Memory Model
To explore this, MIT researchers developed a computational model based on Hopfield networks, a type of neural network used to simulate memory storage. Traditional Hopfield networks have limited capacity, but a modified version—dense associative memory (DAM)—can store vastly more information by incorporating higher-order interactions between multiple neurons.
The challenge? In biology, synapses typically connect just two neurons (presynaptic and postsynaptic). So how could the brain achieve these multi-neuron couplings?
Astrocytes as the Missing Link
The MIT team’s breakthrough was recognizing that astrocytes form tripartite synapses, connecting to multiple neurons simultaneously. This allows them to act as hubs for information integration, enabling the dense associative memory needed for high-capacity storage.
In their model:
- Astrocyte processes function as computational units
- Calcium dynamics encode memory patterns
- Gliotransmitters relay information back to neurons
This architecture could theoretically store an unlimited number of memory patterns, limited only by the size of the network.
Why This Matters for Neuroscience and AI
The idea that astrocytes contribute to memory storage challenges long-held beliefs about brain function. If confirmed, this research could revolutionize our understanding of learning, memory, and cognition—while also inspiring next-generation AI systems. For example;
- Explaining the Brain’s Memory Capacity
- The model suggests astrocytes help the brain achieve far greater memory storage than neurons alone could provide.
- Each astrocyte process could act as an independent computational unit, making the system highly efficient.
- Implications for Artificial Intelligence
- The study bridges neuroscience and AI, offering new ways to design high-capacity memory systems.
- Variations of the model could improve machine learning architectures, such as attention mechanisms in large language models (LLMs).
As Jean-Jacques Slotine, an MIT professor and study co-author, notes:
“This work may be one of the first contributions to AI informed by recent neuroscience research.”