UCSF

Cortical networks are thought to operate in a state of tightly balanced excitation and inhibition. A typical cortical neuron receives thousands of synaptic inputs from other cortical neurons. The majority of these inputs are excitatory and the network is stable only because strong recurrent excitation is balanced by similarly strong feedback inhibition. Theoretical models have shown that sparsely connected networks with strong, balanced excitation and inhibition exhibit chaotic activity that is consistent with the highly variable responses and large membrane potential fluctuations observed in cortex in vivo. In these models, and hypothetically in cortical networks, large noise is a consequence of large excitatory inputs being balanced by inhibitory inputs, keeping the mean membrane potential below threshold but leaving large fluctuations about this mean. Direct experimental evidence for this balanced state has been provided by measurements of synaptic conductances in vivo and in cortical slices. In chapter two we show that noisy fluctuations in membrane potential are an essential part of a mechanism by which neurons can modulate the gain of their responses. Gain modulation is common in cortical responses and is thought to play an important role in cortical computation.

Membrane potential fluctuations during spontaneous activity are not entirely random. In cat primary visual cortex spontaneous activity has been shown to exhibit spatial patterns similar to those evoked by a visual stimulus, in which neurons with similar preferred orientations are co-active. This suggests that, from unstructured input, cortical circuits selectively amplify activity patterns related to normal function. Current understanding of such amplification involves elongation of the lifetime of a neural pattern by mutual synaptic excitation among the neurons involved. In chapter three we describe a new mechanism by which recurrent networks with strong recurrent excitation and balancing feedback inhibition can amplify neural activity patterns without elongation of lifetime, which we call transient amplification. By better understanding the implications of balanced excitation and inhibition in cortical networks, we gain insight into the role of recurrent circuitry in cortical computation.