A fundamental property of neurons in primary visual cortex (V1) is their selective response to bars of light at particular orientations within their receptive fields . Neurons with similar orientation preference and receptive field location are organized into vertical columns [20,22,34]. In addition to radial connections between neurons in a column, horizontal connections within V1 link columns of similar orientation preference that can lie up to several mm apart and represent distinct regions of the visual field [7,15,26]. Such long range connections form the likely substrate for mediating influences on neurons from outside their ``classical'' receptive field .
The precise nature of this influence remains unresolved, however. Several studies have demonstrated that the presence of strong stimuli beyond the classical receptive field suppresses the response to an optimal stimulus within the classical receptive field [3,18,19,24,29]. In contrast, stimulating the extra-classical receptive field in conjunction with weak or no classical receptive field stimulation has been shown to facilitate responses [16,18,23,24]. The most effective stimuli for eliciting both types of effect are oriented, have a similar orientation to that of the preferred stimulus, and are of high contrast. The orientation dependence of the effects indicates that they are cortical in nature and mediated by long-range connections. However since the same surround stimuli are maximally effective for producing both effects, the data paradoxically suggests that the same set of connections should underlie these opposing effects. We propose a computational model of V1 that resolves these apparently conflicting data. The model suggests that the role of long range horizontal connections is a dynamic one, and can be either suppressive or facilitative depending on stimulus context (i.e., on the relative level of excitatory drive from the classical and extra-classical receptive field). The dynamic modulation that results is not due to changing synaptic strengths, as these are fixed in the model. Rather the modulation occurs because of changes in the gain control properties of the local circuitry. This local gain mechanism is based on a mechanism that we utilized, in a short-range, recurrent excitatory model of orientation selectivity , to generate contrast saturation and contrast-invariant orientation tuning bandwidths.
In addition, in this paper we experimentally demonstrate the predicted bi-phasic response dynamics in cortical cell populations using optical imaging of activity and single unit recordings in cat V1. Presentation of high contrast surround stimuli produce orientation-specific increases central activity, when center activation is low; however, the same surround stimuli suppress responses for high center activation.