Fire together, wire together

Networks in rodent visual cortex become functionally specific during development. That is, naive networks with salt-and-pepper structure, where neurons with different stimulus preferences are non-selectively connected together, later during development tend to form subnetworks where neurons with similar functional properties tend to be coupled. It was long hypothesized that a Hebbian learning rule which connects neurons that respond to the same stimulus (“fire together, wire together”) could underlie such a refinement of connectivity.

This straightforward idea is, however, challenged by the complex recurrent dynamics of realistic cortical networks, with intense recurrent connectivity of excitatory and inhibitory neurons. In fact, harnessing plasticity in large-scale recurrent networks with balanced excitation and inhibition to build viable networks that remain stable and functional at the same time, has been very difficult to achieve so far. The new article by Sadra Sadeh and Stefan Rotter from the Cluster of Excellence BrainLinks-BrainTools and the Bernstein Center at the University of Freiburg, and Claudia Clopath from the Bioengineering Department at Imperial College London in the journal PLoS Computational Biology reports a recent breakthrough in this direction.

The study presents a network model which performs a simple, but important functional task, namely orientation processing as reported from recordings of the primary visual cortex (V1) in many mammalian species. The study extends and enhances previous modeling studies, which addressed similar questions in much simplified models. Among the new insight obtained by this study is the immensely important role of inhibition in this process. The authors report that both the innate emergence of orientation selectivity in primary cortical sensory neurons, as well as the lasting stability of the network during learning, depend in a crucial way on the functional dominance and the plasticity of inhibitory synapses in the network, respectively. Both mechanisms together reliably lead to the emergence of feature-specific connectivity in these cortical network models. The study therefore reveals how the intricate interplay of recurrent dynamics and plasticity in inhibition-dominated networks can naturally overcome problems that troubled previous models of balanced recurrent networks with synaptic plasticity. The new model, therefore, offers a new framework to study functional properties in balanced plastic networks throughout the brain.

Original Publication:
Sadeh S, Clopath C, Rotter S
Emergence of functional specificity in balanced networks with synaptic plasticity
PLoS Computational Biology 11(6): e1004307, 2015

Figure Caption:
In random networks (not shown) of excitatory (triangles) and inhibitory (circles) neurons, synaptic connections are established disregarding the stimulus selectivities (indicated by color) of pre- and post-synaptic neurons. In specific networks (shown here), synapses between neurons of similar preferred features are stronger, while dissimilar feature selectivity of pre- and post-synaptic neurons imply weaker synapses between them.