Ethological variability shapes the structure and content of sensory population codes

Abstract

Established theories of sensory coding posit that sensory systems are adapted to the statistics of natural stimuli that originate in the external environment. In natural, ethological conditions, however, sensory populations are simultaneously influenced by multiple modulatory processes. During behavior, the statistics of population responses already at the sensory periphery reflect not only external signals, but also internal and behavioral states such as locomotion, attention, and arousal. Theories of sensory coding in real-world conditions must therefore explain adaptation not only to natural scenes, but also to the structured variability induced by these internal processes and behavior. Here, we investigated population activity in the dorsal lateral geniculate nucleus (dLGN), a critical early stage of the visual pathway. Using a simple probabilistic model and information-theoretic approaches, we characterized this variability through the statistics of the population messages conveyed downstream. Our results show that population variability is multidimensional: even when one behavioral factor is constant, responses can vary substantially as other modulatory factors fluctuate. Under natural conditions, sensory population codes can therefore change their statistical structure and content, i.e., the stimulus features they encode along multiple dimensions simultaneously. These findings highlight the need for sensory coding theories that account not only for adaptation to natural stimuli, but also to the statistics of internal and behavioral states.