Figure 10.
Model of input-dependent surround modulation. A, Two directional columns within adjacent “center” and “surround” hypercolumns. Each column consists of an excitatory (E) and an inhibitory (I) neuron. Excitatory and inhibitory neurons receive feedforward input from V1. Lateral excitation (•) between hypercolumns occurs via direct connections between excitatory neurons. Lateral inhibition (○) is mediated via inhibitory interneurons. The level of feedforward input into the inhibitory neurons determines the efficacy of lateral inhibition. All connections are symmetrical so center and surround designations are arbitrary. B, Directional weighting of connections. Center and surround hypercolumns each consist of 8 excitatory (bottom) and 8 inhibitory (top) neurons. Redundant types of connections are omitted to avoid clutter. In particular, the pattern of feedforward inputs to center and surround hypercolumns (i.e., Wfi and Wfe) is identical but only the former are shown. Similarly, connections between hypercolumns (Wei and Wee) are symmetrical but are only those from the surround to the center are shown. Last, both hypercolumns have within-hypercolumn inhibition (Wie), but this is only shown for the surround hypercolumn. The feedforward connections to inhibitory neurons (Wfi) are not directionally weighted but all other weights are. Compare with A. C, Stimulus-dependent surround integration/antagonism. Inset, Directional tuning curve for a grating within the CRF. For this model neuron, P and PP directions are up to the left and up to the right, respectively (compare with Fig. 6). In this simulation, we compare responses to surround stimuli moving in the P and PP directions while the center stimulus moves upward (i.e., the local direction within the CRF). We examined these responses as we varied the level of the “V1” input into the center hypercolumn. With no center input, the surround stimulus has no effect and hence this model neuron would be classified as a no-control-response neuron. With maximum input spike rates of 20 and 40 spikes/s, the model reproduces surround integration like that seen in area MT neurons responding to contour and long-bar stimuli, respectively (compare with Fig. 6A,C). With an input rate of 60 spikes/s, surround modulation is absent as seen in response to short-bar stimuli (compare with Fig. 6D). With higher input rates of 80 and 100, antagonistic modulation emerges consistent with responses to dot stimuli (compare with Fig. 6B). Note that this neuron exhibits supersaturation, a property that may account for the finding that some neurons do not respond more to stimuli that elicit antagonism than to stimuli that elicit integration. The polar-plots above the Cartesian plot show responses to P and PP directions for a subset of the input values. D, Contrast-dependent suppression/facilitation. For low-contrast stimuli, increasing the size of a stimulus (moving in a neuron's preferred direction) so that it stimulates the surround results in facilitation. For high-contrast stimuli, the same size increase results in surround suppression. Contrast level was implemented by varying the strength of feedforward input.