Homeostatic structural plasticity can build critical networks

Van Ooyen, A., and Butz-Ostendorf, M. (2019). In: Tomen, N., Herrmann, J. M., and Ernst, U., eds. The Functional Role of Critical Dynamics in Neural Systems. Springer, pp. 117-137. [Full text: PDF]

Abstract

Many neural networks, ranging from in vitro cell cultures to the neocortex in vivo, exhibit bursts of activity ("neuronal avalanches") with size and duration distributions characterized by power laws. The exponents of these power laws point to a critical state in which network connectivity is such that, on average, activity neither dies out nor explodes, a condition that optimizes information processing. Various neural properties, including short- and long-term synaptic plasticity, have been proposed to underlie criticality.

Reviewing several model studies, here we show that during development, activity-dependent neurite outgrowth, a form of homeostatic structural plasticity, can build critical networks. In the models, each neuron has a circular neuritic field, which expands when the neuron's average electrical activity is below a homeostatic set-point and shrinks when it is above the set-point. Neurons connect when their neuritic fields overlap. Without any external input, the initially disconnected neurons organize themselves into a connected network, in which all neurons attain the set-point level of activity. Both numerical and analytical results show that in this equilibrium configuration, the network is in a critical state, with avalanche distributions described by precisely the same power laws as observed experimentally.

Thus, in building critical networks during development, homeostatic structural plasticity can lay down the basis for optimal network function in adulthood.