A Model of Synaptic Normalization and Heterosynaptic Plasticity Based on Competition for a Limited Supply of AMPA Receptors

摘要

Simple models of Hebbian learning exhibit an unconstrained growth of synaptic efficacies. To avoid this unconstrained growth, some mechanism for limiting weights needs to be present. There is a long tradition of addressing this problem in neural network models using synaptic normalization rules. A particularly interesting normalization rule scales synapses multiplicatively such that the sum of a neuron's afferent exctiatory synaptic weights remains constant. One attractive feature of such a rule, next to its conceptual simplicity, is that in combination with Hebbian mechanisms it can give rise to lognormal-like weight distributions as observed experimentally. While such a normalization mechanism is not considered implausible, its link to neurobiology has been tenuous. A full understanding of such synaptic plasticity arguably requires the development of models that capture its complexities at the molecular level. Inspired by recent findings on the trafficking of neurotransmitter receptors across a neuron's dendritic tree, I propose a mathematical model of synaptic normalization and homeostatic heterosynaptic plastictiy based on competition between a neuron's afferent excitatory synapses for a limited supply of AMPA receptors. In the model, synapses on the dendritic tree of a neuron compete for a limited supply of AMPA receptors, which are produced and distributed across the dendritic tree and stochastically transition into and out of receptor slots in the synapses, or simply disintegrate at a low rate. Using minimal assumptions, the model produces fast multiplicative normalization behavior and leads to a homeostatic form heterosynaptic plasticity as observed experimentally. If the production rate of AMPA receptors is controlled homeostatically, the model also accounts for slow multiplicative synaptic scaling. Thus, the model offers a parsimonious and unified account of both fast normalization and slow scaling processes, which it both predicts to act multiplicatively. It therefore supports the use of such rules in neural network models. Because of its simplicity and analytical tractability, the model provides a convenient starting point for the development of more detailed models of the molecular mechanisms underlying different forms of synaptic plasticity.