Robust spatial working memory through homeostatic synaptic scaling in heterogeneous cortical networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Persistent neural activity, often bell-shaped, is key for representing analog information like spatial location.
• Prior models assumed uniform neural networks, which is biologically unrealistic.
• Network heterogeneities can disrupt the precise tuning required for memory functions.

Purpose of the Study:

• To investigate the impact of cellular and synaptic heterogeneities on neural network models of spatial working memory.
• To explore how homeostatic mechanisms can counteract these disruptive effects.
• To assess the biological plausibility of recurrent attractor networks for spatial memory.

Main Methods:

• Developed a network model incorporating cellular and synaptic property variations.
• Simulated the network's ability to store spatial information under heterogeneous conditions.
• Introduced a homeostatic mechanism to scale excitatory synapses.
• Analyzed the resulting tuning curves and information loss.

Main Results:

• Heterogeneities rapidly degraded stored spatial information.
• Homeostatic scaling of excitatory synapses restored accurate spatial encoding.
• The model generated diverse neural tuning curves, consistent with experimental observations.
• Recurrent attractor networks with homeostatic regulation proved robust.

Conclusions:

• Network homogeneity is an unrealistic assumption for persistent neural activity models.
• Homeostatic mechanisms are essential for maintaining spatial working memory in heterogeneous neural networks.
• Recurrent attractor networks coupled with homeostatic regulation offer a biologically plausible framework for spatial memory.