The Dynamics of Orientation Tuning in Macaque V1.

Dario L. Ringach, Michael J. Hawken, and Robert Shapley
Nature, volume 387, 15 May 1997.


Abstract: Orientation tuning of neurons is one of the chief emergent characteristics of the primary visual cortex, V1. LGN cells that comprise the thalamic input to V1 are not orientation tuned, but the majority of V1 neurons are quite selective. Two main classes of theoretical models have been offered to explain orientation selectivity: feedforward models, in which inputs from spatially aligned LGN cells are summed together by one cortical neuron; and feedback models, in which an initial weak orientation bias due to convergent LGN input is greatly sharpened by intracortical feedback. To test these models we studied the dynamics of orientation tuning - how the orientation tuning of a neuron evolves with time - using a novel method: reverse correlation in the orientation domain. We find that the broad orientation tuning seen in the input layers 4C-alpha and 4C-beta is associated with very simple dynamics. However, sharper orientation tuning in supragranular and infragranular layers is accompanied by more intricate dynamical features, such as ``rebound'' responses, delayed secondary peaks, and sharpening of orientation tuning with time. Simulations of feedforward networks yield plain dynamic responses. In contrast, many of the dynamical features observed outside layer 4C arise naturally in feedback models. Therefore, our experimental results imply that the relatively broad orientation bias seen in 4C-alpha and 4C-beta may be computed by a feedforward network, but that cortical feedback is responsible for sharpening orientation-selectivity and causing intricate dynamical responses in macaque V1.


The goal of this research is to study the detailed dynamics of orientation tuning in V1 neurons. The new method we developed, illustrated below, consists of presenting a visual stimulus that is a fast sequence of sinusoidal gratings having a fixed spatial frequency (optimal for the cell) but random orientations and spatial phases. The images are shown at a rate of 60 frames/sec. We present this stimulus to cells for 15 min. During this time the cell responds with a spike train that we record with an accuracy of 1msec.

To analyze these data we estimate the probability that a particular orientation was present in the input image sequence (independently of its spatial phase) tau msec before a spike occurred at the output. We call this quantity r(theta,tau). Note that for each time tau, r(theta,tau) is a probability density function on the orientation domain. We can now study how this density function evolves over time. This is how we obtain data about the dynamics of orientation tuning. To see some results on-line click here.