The neocortical network is primarily excitatory as most neurons and most synapses are excitatory. It has therefore generally been assumed that cortical computation is primarily shaped by the excitatory to excitatory connections. However, there is strong experimental evidence that the excitatory connectivity is dynamic even in the absence of explicit learning. This raises the question of how functionality is maintained over time. Here we quantified the excitatory connectivity remodeling by chronically imaging dendritic spines in the mouse auditory cortex, and investigated its computational implications in the framework of balanced networks. We found that, surprisingly, the ongoing firing pattern is stably maintained even in the presence of massive remodeling of the excitatory connectivity, while it is exquisitely sensitive to changes to the inhibitory synapses. Computationally, we show that in contrast to excitatory plasticity, inhibitory plasticity is both necessary and sufficient for large scale changes in the patterns of network activity. Moreover, it is information about the firing pattern of the inhibitory neurons, rather than that of the excitatory neurons that is communicated between the neurons. All these results are a direct consequence of the different distributions of firing rates of the excitatory and inhibitory neurons. Taken together, our findings suggest that cortical computation is dominated by inhibition and shaped by inhibitory plasticity.

Work conducted in the framework of the France Israel Laboratory of Neuroscience.