Séminaire de Andrea Benucci (RIKEN Center for Brain Science)

Andrea Benucci (RIKEN Center for Brain Science)

“Distributed motor and choice information in the mouse posterior cortex” 

Vendredi 15 avril à 14h30, Salle Henri Gastaut, INT

Abstract

During natural behaviors, visual cortical networks are activated by a multiplicity of extra retinal signals, for example motor- and cognitive-related. This observation has challenged the traditional view of feedforward, hierarchical processing of retinal inputs, suggesting that signals from other brain regions continuously shape the dynamical landscape of even early sensory areas. I will present our findings on the principles that govern the interaction dynamics of multiplexed signals in the visual cortex of mice. As animals engaged in a challenging “relative” orientation-discrimination task, we recorded widefield multi-area activations (GCaMP signals). We found that, although visuo-motor interactions were prominently non-linear, these variables could be linearly separated in a reduced-dimensionality space of multi-area activations. Furthermore, the separability between visual and motor signals increased during heightened states of sustained attention. Choice information had a characteristic spatial signature along the ventral visual stream, and it was found to be near orthogonal to both visual and motor components. The attention-dependent representational geometry of these activations was well recapitulated by a recurrent neural network model trained to reproduce the behavioral choices of the mice. Together, these findings define novel principles of signal interactions that shape the dynamical landscape of visual cortical networks, fundamentally constraining the underlying effective-connectivity architectures.