Coordinated neural activity: Mechanistic origins and impact on stimulus coding

Abstract

How does the activity of populations of neurons encode the signals they receive? Since neurons in vivo are inherently variable, each fixed input to a population will elicit not a deterministic response, but rather a probability distribution of states of the individual neurons. Traditional theories of neural coding rely on single-cell tuning curves that describe the average response of each neuron to stimulus features. Adding complexity to this neuron- by-neuron view is the fact that neural activity is not independent: it is often correlated, reflecting shared input and connectivity. Such "coordinated" activity can have diverse and potentially strong impacts on how neural circuits encode stimuli. In this dissertation, we combine dynamical and statistical tools to examine how single-cell and network properties dynamically generate coordinated neural spiking, and how this affects stimulus coding in populations of cells. First, we show how feedforward connectivity leads to the emergence of a neutrally stable subspace that allows information about input rates to be transmitted through layers. At this critical parameter regime, neural activity is characterized by higher-order interactions, meaning that the activity cannot be described by minimal models including only the lower-order moments (mean and pairwise interactions). Interestingly, recent experiments have also demonstrated the existence of higher-order correlations in the neural activity patterns in retina and cortex. Using maximum entropy techniques, we show that in general populations, higher-order correlations can facilitate the encoding of stimulus information in neural activity patterns. We propose a statistical model for fitting neurophysiological data that incorporates only the most significant higher-order interactions. We apply this model to analyze the statistics of population firing patterns in the lateral geniculate nucleus of awake mice. Finally, we analyze dendritic nonlinearities as a novel mechanism by which intrinsic cell properties can generate higher-order correlations. Together, these results work towards determining the origins of coordinated spiking, understanding its impact on neural coding, and building better tools for quantification in electrophysiological data.