Architecture of neural circuits for vision: from cell types to computation
Petr Znamenskiy
Group Leader
The Francis Crick Institute, UK
Schedule a meeting: calendly.com/petr-znamenskiy
Web: znamlab.org

Résumé
Perception, cognition, learning and behavioural control are products of coordinated activity of specialised neuronal cell types, connected into networks according to wiring rules we are only beginning to reconstruct. Using the visual system as a model, we aim to understand how the building blocks that make up its circuits are connected to each other, and how their connectivity gives rise to computations underlying visual perception.
Our focus is on one of the fundamental challenges of vision – estimation of depth of visual cues from the two-dimensional images formed on the retinae, motivated by the fact that depth perception is innate in most animals. This innate ability is thought to rely on motion parallax – visual motion resulting from animals’ movements. Using recordings in mice navigating in 3D virtual reality environments, we have found that neurons in the mouse primary visual cortex (V1) are selective for depth from motion parallax arising from integration of visual and locomotion-related signals. Consequently, V1 neurons have three-dimensional receptive fields – they are selective for both retinotopic location and depth of visual stimuli. Neuronal cell types across different layers of V1 differ in the ways they integrate optic flow and locomotion signals and show distinct preferences for depth. To determine how these coding differences relate to the molecular identity of individual neurons, we are using correlative in situ sequencing to measure gene expression patterns of the same neurons we profile in vivo.
Explaining how such computations arise from synaptic organisation requires linking in vivo responses of single neurons to their input patterns. However, conventional approaches for reconstructing synaptic connectivity, such as volume electron microscopy and paired patch-clamp recordings, are extremely laborious and typically limited to mapping connections between nearby neurons. To overcome these limitations, we developed a method that uses rabies viruses carrying random molecular barcodes to map both local and long-range monosynaptic inputs to hundreds of neurons in parallel. This approach reveals layer- and cell-type-specific connectivity rules as well as topographic organisation of long-range projections, enabling reconstruction of neural circuits at unprecedented throughput.

Invité par Stéphane Bugeon
Lundi 21 septembre 2026 à 11h – Salle de conférence de l’Inmed