The variance of V1 neurons' responses saturates in brief counting periods

Purpose Previous analvses oi neuronal response variability typically counted spikes over a period of several hundred ms. We investigated the effect of reducing the counting porind to 50 ms, to see whether neuronal reliability would improve Visual responses nf single units were extracellularly recorded in the primary visual eortex of anesthetized macaque monkey. Responses to sine wave; gratings and to drifting square waves (edges) were compared. The response measure was spike count within a 50 ms window centered on the response peak: the inverse of the coefficient of variation was our measure of reliability. Results Within a counting window of 125 250 ms. variance increases proportionally with mean. However, within the 50 ms window, variance of response magnitude saturates, approaching an asymptotic level of 2 5 imp²/50ms for responses larger than 50 imp/s. This asymptotic varrance is similar to what wo find in LGN neurons. Simple cells are less variable than iomplex cells; neurons in layers -1 and 6 have smaller variance than neurons in other layers. The most reliably responding neurons were also located in lavers 4 and 6. In this regime of saturating variance (approximate additive noise), reliability of responscs increases linearly with response strength. The highest information/stimulus/50ms transmitted about an optimal stimulus was found to be ∼ 1 bit like the esiimat.es of other investigators (Optican & Richmond, 1990; Tovée et al. 1993). Conclusions Cortical variance depends on the temporal window of analvsis. In short periods, variance of optimal responses is additive, but over periods longer than 100 ms it. becomes multiplicative. Such a difference in response variance affects the neurons' information encoding capacity.