Institute of Ophthalmology, University College London, London, UK
The visual representation of an object is distributed amongst multiple, functionally specialized areas of the cerebral cortex. When several objects are represented simultaneously, what mechanism preserves the identity of a single object as its features are progressively separated at higher levels of processing? We address this question—known as the ‘binding problem’—by examining combinatorial feature-selectivity of neurons in area V2 of the anaesthetized macaque monkey. We find that dual selectivity for chromatic and spatiotemporal attributes is common enough in the superficial and deep layers that receive feedback connections from higher areas, but disproportionately sparse within the middle layers (4 & 3) that relay ascending signals. In other words, for example, the neurons in V2 relaying motion information to area V5 are not colour-sensitive, and the colour-selective neurons in V2 communicating with area V4 are not direction-selective. Neurons that do combine colour and direction sensitivity are found in layers that do not send output to higher centres such as V4 and V5, but do receive feedback from them.
How does this arrangement promote binding? Firstly, because cortical feedback pathways are the likely conduit for top-down attentional modulations. Secondly, because in terms of cognitive modelling, attention is instrumental in binding by acting to select one single object for higher representation, and filtering out competing objects. We propose that dual-selective neurons perform a ‘bridging’ function, mediating the transfer of feedback-induced bias between feature dimensions. Hence, for instance, if attention were directed toward a particular colour, only dual neurons tuned to that colour, and to the motion direction physically coupled to it in the visual stimulus, would show feedback enhancement. This modulatory effect on activity could then be propagated through V2 (by translaminar & transcolumnar intrinsic connections) in order that the selected feature combination be reflected in the pattern of activity across unimodal output neurons in layer 3. And, subsequently, across their fields of influence in higher level areas—such that representations of a single attended object come to dominate in multiple feature maps, as envisaged by the ‘integrated competition’ model of attention. In short, we postulate that the bridging function of dual-tuned neurons in V2 acts to unify the outcome of parallel object-selective processes taking place along specialised visual pathways diverging from V2.