Integrating behavioral state into models of auditory neural population coding

Abstract:
Sensory encoding models provide a powerful tool for understanding how sensory information is represented and transformed at successive stages of processing by the brain. In the auditory system, a family of encoding models have been developed, based the classical linear spectro-temporal receptive field (STRF). It has also been established that changes in large-scale brain activity – either spontaneous or driven by changes in behavioral state – can influence neural auditory-evoked responses. However, less is known about how properties of encoding models are influenced by these changes in behavioral state. 

We recorded the activity of single neurons in auditory cortex and considered two aspects of internal brain state that can influence coding: spontaneous fluctuations in the activity of neurons recorded simultaneously from the same cortical column (n=20-40 units) and global arousal state, measured using pupillometry. We used a generalized linear modeling framework to measure how fluctuations in these internal state variables explained variability in neural activity that could not be explained by an exclusively auditory encoding model. The activity of the neighboring neural population had significant influence on the activity of most units. Moreover, population activity could be projected into a low-dimensional state space that captured much of its influence. Some dimensions of this state space correlated strongly with pupil-indexed arousal, while other dimensions showed no correlation. These results support a model that auditory cortical coding is influenced by multiple latent variables whose effects arrive through distinct neural subpopulations. Future experiments will explore the interaction of state variables reflecting arousal with those reflecting engagement in auditory behavior.