Slow cortical dynamics generate context processing and novelty detection

The cortex amplifies responses to novel stimuli, compared to those elicited by redundant stimuli—a function key to efficiently processing sensory information and building predictive models of the environment. Novelty detection is measured by the “Mismatch Negativity” (MMN) signal, the reduction of which represents the best functional biomarker of schizophrenia. To better understand the circuit mechanisms of novelty detection, we used an auditory “oddball” paradigm and two-photon calcium imaging to measure responses to simple and complex stimuli in neuronal populations across the mouse auditory cortex. Stimulus statistics and complexity generated differences in neural response profiles across contexts and auditory cortical subregions. At the population level, neuronal ensembles separately and reliably encoded basic auditory features, as well as temporal context. Interestingly, stimuli-evoked responses were particularly long-lasting, persisting after the stimuli ended and affecting responses to future stimuli. These slow network dynamics encoded stimulus history and temporal context, generating novelty detection. Recurrent neural network models trained on the oddball task exhibited slow network dynamics and recapitulated the biological data, including context selectivity, MMN, and stimulus-specific adaptation. We conclude that the slow dynamics of recurrent cortical networks underlies temporal processing of stimuli, a canonical computation that gives rise to context-specific encoding and novelty detection.

Methods
Methods of data collection, preprocessing, and analysis are described in detail in the associated article

Data was collected with two-photon calcium imaging in the Auditory Cortex of awake mice that were head fixed and moving on a circular treadmil. Mice were presented with auditory oddball stimuli, and many standards control variation. Raw data were motion corrected, and regions of interest (ROIs) corresponding to cells were demixed with CaImAn. The demixed data is provided here.