V1 INHIBITORY POPULATION ACTIVITY IS LOW-DIMENSIONAL AND ENCODES BEHAVIORAL STATE

Understanding the function of a neuronal population requires understanding the patterns made by the conjoint activity of its neurons. While the structure of population activity has been well studied in excitatory neurons, less is known about inhibitory populations. We addressed this using two-photon calcium imaging to simultaneously record thousands of inhibitory and excitatory neurons in mouse V1. Head-fixed mice ran freely on a wheel while we monitored locomotion and facial movements. We also presented natural image stimuli to assess visual responses.

Inhibitory populations carried less information about visual stimuli, but more information about behavioral state compared to excitatory populations. Decoding of facial movements was more accurate from inhibitory than excitatory populations; for decoding of visual stimuli, the reverse was found. Consistently, encoding models predicted inhibitory activity more accurately from facial movements than from stimulus identity, but the reverse for excitatory activity.

Inhibitory activity occupied a lower-dimensional space than excitatory activity. The majority of inhibitory—but not excitatory—neurons were strongly correlated with leading neural dimensions, which in turn tracked spontaneous behavior.

For both cell types, spontaneous and visual activity largely consisted of independent patterns. There was one notable exception: neurons strongly participating in synchronous, quasi-oscillatory activity during low-arousal states were also strongly activated by natural images.

Thus, excitatory and inhibitory populations play complementary roles in cortical computation: excitatory populations encode sensory information, inhibitory populations modulate local activity according to behavioral state. Within each cell type, visual and spontaneous activity overlap along a single dimension, capturing synchronized activity across the population.