Abstract
V1 neurons, unlike their LGN afferents, appear to be tuned essentially randomly for color. One explanation for this apparent disorder is that previous attempts to estimate the preferred color directions (cone weights) of V1 neurons may have been distorted by threshold nonlinearities. Under a linear-nonlinear cascade model, the chromatic tuning of a V1 neuron is determined by a weighted sum of signals from the three classes of cone photoreceptor, and the spiking probability is related to the weighted sum of cone signals via a static nonlinear output function. Purely linear analysis techniques ignore this nonlinearity and thus may yield distorted cone weight estimates. To avoid this problem, we estimated the cone weights to V1 cells using a reverse correlation technique that is robust to static output nonlinearities. Cone weights of simple cells were estimated by spike-triggered averaging. Consistent with previous findings, the cone weights of V1 simple cells were broadly distributed with a few discernable clusters. Complex cells have rectified responses, so we estimated their cone weights by spike-triggered covariance rather than averaging. Every complex cell we studied responded strongly to non-opponent modulations of the L and M cones we found no evidence of full-wave rectified cone-opponent V1 neurons. Some complex cells responded to both non-opponent and opponent modulations in the stimulus sequence, demonstrating that some complex cells have multiple preferred directions in color space, and are poorly described as full-wave rectified linear mechanisms.
HHMI, McKnight Foundation