Statistics of the maintained discharge of cat retinal ganglion cells.

Abstract

Action potentials were recorded from single fibres in the optic tracts of anaesthetized cats. Continuous records were obtained at various levels of scotopic and mesopic retinal illumination. In some cases, the light intensity was modulated by a pseudorandom Gaussian white-noise signal. The maintained discharge of on-centre neurones increased while the maintained discharge of off-centre neurons decreased with increased illumination of the receptive field centre. For both cell types, the coefficient of variation declined with increased rate of discharge. There was minimal short-term dependency in the firing patterns, and it was unaffected by the level of retinal illumination. Virtually all of the structure revealed by the normalized autocovariance functions could be attributed to the shape of the interval distributions. The first few coefficients of the serial correlogram were slightly negative. The magnitude of this negativity was not related to illumination. Long-term dependency in the firing pattern was also quite small; the standard deviations of the mean rate in samples of about 1 sec duration were only slightly less than would be predicted from the interval distributions. This dependency tended to increase at higher retinal illuminations. Neural discharges elicited by Gaussian modulation of the light were strikingly different from those elicited by steady light. Modulation caused the first coefficient of the serial correlogram to become more positive, while the next several coefficients became more negative. A corresponding pattern could be seen in the normalized autocovariance functions, and in the differences between the normalized autocovariance and normalized autoconvolution. Long-term dependency also increased dramatically, such that the standard deviations of mean rate were about 60% of what would be expected given the interval distributions observed. These results place a number of constraints upon the ways in which intrinsic noise in the retina may enter the visual processing network. Two alternative models consistent with the data are presented.