Abstract
Fully dark-adapted human observers are remarkably sensitive to light, coming close to the detection limits imposed by the discrete nature of photon absorptions and quantum uncertainty. In rod-mediated vision, classical experiments by Barlow and others revealed the existence of three laws relating detection threshold to adapting luminance. At very low light levels, threshold is independent of adapting luminance and is thought to be limited by neural noise. At high adapting luminances, threshold is set by the physiological gain mechanism responsible for Weber's law. Between these regimes, threshold rises in proportion to the square-root of the adapting luminance (Rose-deVries law). It is not known whether the square-root law is due to physiological gain control or simply to a tendency for the appearance of weak probe flashes to be masked by the larger quantal fluctuation noise associated with more intense adapting fields. We addressed this question using a combination of psychophysical and physiological methods. The psychophysical experiments were based on the technique of dichoptic brightness matching. A 10 msec, 0.55 deg., 492 nm flash of fixed intensity was presented to a rod-dominated region of one of the observer's two retinas, superimposed on a 7.4 deg adapting field producing an average of either 0.01, 0.1, or 1.0 photoisomerizations per rod-sec. Simultaneously in the observer's other eye, an otherwise identical flash of adjustable intensity was presented on an adapting field of fixed intensity. The intensity of the adjustable flash causing the two flashes to appear equally bright was determined using a staircase procedure. We found that the square-root law holds for brightness matching of superthreshold flashes having intensities as high as 1000 times threshold. This effect cannot be due to noise masking but must be due to monocular gain control. On half the trials, random flicker was added to the adjustable background to discover whether the added noise would affect the neural gain in that eye. The added noise slightly increased the brightness of superthreshold flashes at high background luminances. Since the direction of this effect is opposite to that expected if the neural gain was set by noise fluctuations, we tentatively conclude that the gain control responsible for the square-root law is deterministic. The existence of square-root gain control was further confirmed by preliminary intra- and extra-cellular recordings from ON-parasol ganglion cells in intact macaque retina, indicating that the gain control results from a rod-mediated network adaptation in primate retina.