Presentation of a sensory stimulus evokes a complex dynamic pattern of activity in sensory cortex. The nature of the evoked dynamics depends on the internal state of an animal's brain at the time of presentation. Here we describe a method for fitting dynamical system models for cortical population activity, and evaluating the fit of these models by their ability to predict the response to individual sensory stimuli.
The state of a neural circuit changes over multiple timescales. At short timescales (up to hundreds of ms), the sensory response of a circuit may be affected by its immediate history of activity. At longer timescales, behavioral states such as alertness and the sleep-wake cycle modulate cortical responsiveness, through mechanisms such as changes in neuromodulatory tone.
The global pattern of neocortical activity is often divided into two states. The synchronized state, typical of inalertness or slow-wave sleep, exhibits diverse, globally coordinated activity patterns, characterized by spontaneous transitions between UP states of widespread depolarization and spiking, and DOWN states of generalized silence. The desynchronized state, typical of alert wakefulness and REM sleep, is characterized by low amplitude, high frequency local field potential (LFP) power.
To investigate how global cortical state and sensory input interact to shape sensory responses, we recorded LFPs and population spike trains from the auditory cortex of urethane-anesthetized rats using multi-site silicon microelectrodes. 1ms noise click stimuli were presented, and intervals of silence were used to study spontaneous activity. We found enormous variability in stimulus-evoked responses; this variability was highly correlated to brain state, although it is often dismissed as noise.
We introduce a simple dynamical systems model of cortical state to characterize the dynamics of neural activity over a wide range of states. We find that this model, fit to spontaneous activity data, can accurately explain the variety of stimulus-evoked responses observed across different cortical states. The model also indicates that activity in the synchronized state is characteristic of a highly nonlinear, self-exciting system, whereas the desynchronized state exhibits approximately linear dynamics. Our results suggest that global features of internally generated dynamics in thalamocortical circuits suffice to explain essential aspects of the interaction between (coarse) sensory stimuli and ongoing network activity.
