The On pathway is specifically required for temporal bandwidth adaptation.

A. Top. Contrast of a randomly flickering stimulus drawn from a Gaussian intensity distribution with a constant mean. Contrast values ranging from 3–35% were each presented to the retina for periods of 20 s. Middle. Example membrane potential recording lasting 100 s of a ganglion cell responding to the 5 different contrast levels shown. Spikes were digitally removed (inset shows segment of raw recording). Bottom. Expanded segment showing transition between higher and lower contrast. B. Linear filters (Left) and nonlinearities (Right) fit during low and high contrast to membrane potential recordings in control solution (top) and with 30 μM APB (bottom). Filter amplitudes are normalized, with gain represented in the slope of the nonlinearity, and have units of s-1 so that the convolution over time of the filter with a dimensionless stimulus yields a dimensionless output. C. Temporal power spectrum of the response at high and low contrasts in the control (top) and APB (bottom) conditions for an example cell. D. Median frequency computed as the half-maximal value of the cumulative distribution function of the power spectrum as a function of contrast in the control and APB conditions for one cell. Lines indicate a linear fit. E. Shift in the median temporal frequency as a function of contrast, computed as the slope of the line in (D) for five cells, shown in the control and APB conditions. F. Gain of cells as a function of contrast, measured as the average slope of the nonlinearity, averaged over 5 cells. G. Delay in response, measured as the time to the first negative peak of the filter as a function of contrast. H. Offset in response measured as the mean membrane potential in 4 s segments, normalized by the standard deviation and computed during all contrasts and compared between the control and APB conditions.