Temporal characteristics of gamma rhythm constrain properties of noise in an inhibition-stabilized network model

Abstract

Gamma rhythm refers to oscillatory neural activity between 30-80 Hz, often induced in visual cortex by presentation of stimuli such as iso-luminant hues or gratings. Further, the power and peak frequency of gamma depend on the properties of the stimulus such as size and contrast. Gamma also has a typical arch shape, with a narrow trough and a broad peak, which can be replicated by a self-oscillating Wilson-Cowan (WC) model operating in an appropriate regime. However, oscillations in this model are infinitely long, unlike physiological gamma that occurs in short bursts. Further, unlike the model, gamma is faster after stimulus onset and slows down over time. Here, we first characterized gamma burst duration in LFP data recorded from two monkeys while they viewed full screen iso-luminant hues. We then added different types of noise in the inputs to the WC model and tested how that affected duration and temporal dynamics of gamma. While the model failed with the often-used Poisson noise, Ornstein-Uhlenbeck (OU) noise applied to both the excitatory and inhibitory populations in the WC model replicated the duration and slowing of gamma, and also replicated the shape and stimulus dependencies. Therefore, temporal dynamics of gamma oscillations put constraints on the type and properties of underlying neural noise.