Published online by Cambridge University Press: 02 June 2009
To investigate the importance of feedback loops in visual information processing, we have analyzed the dynamic aspects of neuronal responses to oriented gratings in cortical area V1 of the awake primate. If recurrent feedback is important in generating orientation selectivity, the initial part of the neuronal response should be relatively poorly selective, and full orientation selectivity should only appear after a delay. Thus, by examining the dynamics of the neuronal responses it should be possible to assess the importance of feedback processes in the development of orientation selectivity. The results were base on a sample of 259 cells recorded in two monkeys, of which 89% were visually responsive. Of these, approximately two-thirds were orientation selective. Response latency varied considerably between neurons, ranging from a minimum of 41 ms to over 150 ms, although most had latencies of 50–70 ms. Orientation tuning (defined as the bandwidth at half-height) ranged from 16 deg to over 90 deg, with a mean value of around 55 deg. By examining the selectivity of these different neurons by 10-ms time slices, starting at the onset of the neuronal response, we found that the orientation selectivity of virtually every neuron was fully developed at the very start of the neuronal response. Indeed, many neurons showed a marked tendency to respond at somewhat longer latencies to stimuli that were nonoptimally oriented, with the result that orientation selectivity was highest at the very start of the neuronal response. Furthermore, there was no evidence that the neurons with the shortest onset latencies were less selective. Such evidence is inconsistent with the hypothesis that recurrent intracortical feedback plays an important role in the generation of orientation selectivity. Instead, we suggest that orientation selectivity is primarily generated using feedforward mechanisms, including feedforward inhibition. Such a strategy has the advantage of allowing orientation to be computed rapidly, and avoids the initially poorly selective neuronal responses that characterize processing involving recurrent loops.