The spatial representations acquired in CA3 by self-organizing recurrent connections

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Neural computation models propose the dentate gyrus (DG) drives memory storage in the CA3 network, particularly for spatial memories.
• A key hypothesis is whether recurrent CA3 connections can self-organize into continuous attractors (charts) to store spatial information.

Purpose of the Study:

• To investigate if self-organized CA3 recurrent connections can form functional spatial representations.
• To contrast self-organized Hebbian plasticity with pre-wired synaptic matrices for continuous attractor models.

Main Methods:

• Utilized a simplified mathematical network model to simulate memory storage.
• Compared spatial representations formed by simulated Hebbian plasticity against pre-wired synaptic matrices.

Main Results:

• Both self-organized and pre-wired models formed granular quasi-attractors with drift, approximating continuous attractors only in infinitely large networks.
• Self-organized connections resulted in distorted spatial metrics, inadequate for accurate path integration, even at biological scales.
• Prolonged self-organization improved positional information but led to chart collapse and decreased context discrimination.
• Positional information was observed for unlearned environments, suggesting cross-environment interference.

Conclusions:

• The idealized CA3 continuous attractor model faces feasibility challenges.
• Findings suggest CA3 network specialization for episodic memory rather than path integration.
• Self-organization mechanisms may lead to inaccuracies in spatial representations crucial for navigation.