Changing behavioral state and the impact of correlated variability on neural population coding in auditory cortex

Correlated variability within neural populations, sometimes called noise correlation, substantially impacts the accuracy with which information about sensory stimuli can be extracted from neural activity. Previous studies have shown that changes in behavioral state, reflecting phenomena such as attention and/or arousal, can change correlated variability. However, the degree to which these changes impact neural encoding of sensory information remains poorly understood, particularly in the auditory system. To study this problem, we used linear arrays to record populations of single neurons in ferret auditory cortex while monitoring arousal via pupillometry and controlling engagement in a simple auditory discrimination task. During passive listening, spontaneous increases in arousal tended to decrease the overall degree of correlation. However, the decreased correlation did not consistently improve the accuracy of neural coding. Instead, changes in correlation occurred in a low-dimensional population subspace, and the alignment this space with sound-evoked responses determined its effect on sensory coding. In behaving animals, we saw an additional decrease in correlated activity with task engagement, and the amount of decrease correlated with the accuracy of task performance. However, these changes in correlation again did not consistently improve the accuracy with which task-related stimuli were encoded. Together, these results establish a clear link between behavioral state and correlated neuronal variability. However, the changes appear to reflect processes that are not simply related to the accuracy of sensory coding.