Abstract
Border ownership neurons in primate visual cortex (present especially in areas V2 and V4) encode which side of a border belongs to an object, even if the defining stimulus information falls well outside of the classical receptive field. These signals are thought to be critical for scene segmentation and object recognition, but it remains unclear how they are computed. Important missing pieces of information are the occurrence and timing of border ownership signals in different cortical laminar compartments. Here we addressed this issue by mapping border ownership selectivity across cortical layers using 32-channel laminar probes to record responses from well-isolated units in extrastriate visual area V4 in the rhesus macaque. By replacing the native dura with a transparent artificial dura we were able to position the probes orthogonally to the cortical surface. We used current source density analysis of stimulus-evoked local field potentials to identify the granular (input) layer as the current sink with the shortest latency, and locate the units recorded from to supragranular, granular or infragranular layers. We then measured spiking activity evoked by square stimuli presented on a uniform background. The square was positioned such that only one of its borders fell in the classical receptive field of each neuron. Border ownership was thus defined by stimulus features outside of the classical receptive field. We find that border ownership selectivity occurs first in the infragranular layers, before it appears in the granular and the supragranular layers. This suggests that border ownership signals are not inherited in a feedforward manner from upstream areas, but instead established de novo, or inherited through cortical feedback from downstream areas, in V4 infragranular layers. These infragranular layers include neurons that project to upstream cortical areas, raising the possibility that these signals sculpt border ownership selectivity in V2.