Our stimuli were random binary sequences in which either a preferred stimulus (P, which drove the cell maximally) or an anti-preferred stimulus (A, giving minimal response) was shown on each video frame at 100Hz. We also used ternary random sequences which included a neutral stimulus (N). Typically, P was an optimized sine grating, A was counterphase (LGN and V1) or orthogonal to P (V1), and N was mean gray. For direction selective (DS) cells, P moved in the preferred direction, A moved in the opposite direction, and N was static. We tested the center and surround in LGN, the classical receptive field and surround in V1, and DS cells in V1 and MT/V5.
Almost always, the response latency for on-to-off (P-->A) transitions was shorter than that for off-to-on (A-->P) transitions. In particular, the decrease in firing rate for P-->A transitions began sooner than the rate increase for A-->P transitions, and the delay for the A-->P response was often less when A was present longer. With ternary stimuli, the N-->P and N-->A responses showed little or no timing difference for the LGN and for DS cells in V1 and MT/V5. However, for orientation selectivity in V1, the timing asymmetry persisted for N-->P and N-->A responses. Our results are consistent with the notion that anti-preferred stimuli place neurons in a hyperpolarized state, delaying the response when a preferred stimulus is applied. The delay depends on both the duration of A and on the strength of P. An integrate-and-fire model can reproduce the observed latency asymmetry; however, variations in the typical behavior across classes of neurons reveal differences in mechanisms that process different visual features.