This shows that homogeneity detectors were not a rare exception in our recordings; 20 out of 45 measured cells had an iso-rate form factor smaller than unity, indicative LY2109761 nmr of the characteristic nonconvex iso-rate curve. Interestingly,

these small iso-rate form factors occurred almost exclusively for cells with large receptive fields (Figure 3I), supporting the idea that homogeneity detectors form a particular subclass of ganglion cells. In order to search for the mechanisms underlying the observed nonlinear features of stimulus integration, we first probed the spatial scale at which these occur. To this end, we spatially interleaved the two stimulus components by arranging them in a checkerboard fashion with various sizes of the checkerboard squares. We then measured selleck inhibitor iso-rate curves for these interleaved stimulus components. We found that stimulus integration generally became linear if the squares were sufficiently small (Figure 4): the thresholding of nonpreferred positive contrasts disappeared (Figure 4A and 4B), and homogeneity

detectors lost the nonconvex shape of their iso-rate curves (Figure 4B). These data are consistent with a subunit model (Hochstein and Shapley, 1976, Enroth-Cugell and Freeman, 1987, Victor, 1988 and Crook et al., 2008), in which the receptive field is composed of linear subunits whose outputs are nonlinearly combined. To obtain an estimate of the spatial scale of the subunits, we quantified the amount of rectification depending on the size of the stimulus squares (Figure 4C). The calculated slope values of the tail ends of the iso-rate curves were

near unity for small stimulus squares, indicating linear integration, and dropped to a value slightly above zero, indicating strong, though not always complete rectification. The transition roughly occurred over a range up to about 150 μm, suggesting that the spatial scale of subunits is also approximately in this range. For a more quantitative analysis, we compared the experimental data to results of a simple model Carnitine dehydrogenase simulation that uses circular subunits with a rectifying nonlinearity under checkerboard stimulation. From the model, we also calculated the slopes of the iso-rate curves and found that subunits with diameter from about 50 to 100 μm best captured the course of the experimental data (Figure 4C). This spatial scale corresponds well to the typical diameter of bipolar cell receptive fields (Wu et al., 2000 and Baccus et al., 2008), making the direct excitatory input from bipolar cells a good candidate for the source of the nonlinearities. Indeed, nonlinear signal transmission from bipolar cells has been suggested to contribute to nonlinearities in ganglion cell receptive fields (Demb et al., 2001, Ölveczky et al., 2003, Baccus et al., 2008, Gollisch and Meister, 2008 and Molnar et al.