Modeling the non-linear spatial integration in V1 supragranular layers through asymmetric horizontal connections

Understanding how our brain integrates information across the visual space is crucial for understanding how visual information is encoded. In the supragranular and infragranular layers of the primary visual cortex (V1) of higher mammals, orientation-biased horizontal connections span several millimeters and connect neurons that encode different portions of the visual space [1], while connections in the granular layers remain local.We presented disks of varying radii and annuli of varying inner radii to anesthetized cats to investigate laminar differences in spatial integration of the stimuli in V1. Our intracellular recordings show that the spatial integration of the input in the supragranular layers of V1 is highly non-linear, whereas we observe a more linear spatial integration in the granular and infragranular layers.We investigated these properties in a large-scale spiking neural network model of cat V1 [2]. We found that increasing the spatial range of the horizontal connections targeting the excitatory neurons in layer 2/3, or decreasing the range of those targeting the inhibitory neurons, induces a highly non-linear spatial integration for the neurons of this layer. Making the horizontal connections to excitatory neurons more orientations specific than those targeting the inhibitory neurons have a similar effect. Our model therefore predicts a higher asymmetry, either spatial or functional, between the targeting of excitatory vs. inhibitory neurons by horizontal connections in supragranular layers relatively to infragranular layers of V1.[1] Buzás, P., et al., (2006). Journal of Comparative Neurology, 499(6), 861-881.[2] Antolík, J., et al., (2018). BioRxiv, 416156.