Adult Retinal Circuits

Contrary to popular misconception, the retina does not “take a picture” of the world and send this picture to the brain for viewing. Rather, the retina, as an extension of the brain, actively processes the visual input. The retina extracts specific kinds of information from each point in space.

Furthermore, the retina encodes this information in a condensed manner before sending it to other brain centers. This encoding must occur because of the limited number of fibers in the optic nerve. The encoding also allows the retina to perform such functions as gain control, edge enhancement, and complex feature detection (such as motion direction). An understanding of these retinal functions could directly assist in the development of a retinal prosthesis. Moreover, it could lead to treatments for retinal pathologies and guide in the development of artificial vision.

Here, we provide the following three examples of studies in our laboratory of adult retinal function: First, some ganglion cells, the output cells of the retina, respond better to some directions of motion than to others. These are the directionally selective ganglion cells. We have been studying both the possible functions of these cells for animals and the circuitry underlying directional selectivity. One of our most interesting discoveries is that while the neurotransmitter acetylcholine is not necessary for directionally selective responses to edge motions, it is necessary for textured motions.


Figure 1 Figure 1: Loss of Retinal Directional Selectivity Due to Blockade of Acetylcholine.
Each pair of traces contains responses of rabbit retinal ganglion cells to moving gratings. The grating, which resembles a sine wave in space, first moves in the preferred direction of the cell (top trace) and then in the opposite direction (bottom trace). In control condition, the cell is directionally selective (top pair). If acetylcholine is blocked (second pair), then directional selectivity in response to moving gratings disappears. The same does not happen if one blocks glutamate, another major neurotransmitter in the retina (third pair). Finally, the effect of acetylcholine blockade is not poisoning, because if one removes the drugs, directional selectivity reappears (fourth pair).

Second, we discovered that, if simultaneously stimulated by a moving edge, some ganglion cells show a strong correlation in the timing of their individual spike responses. In other words, if one cell fires a spike to the moving stimulus, then the other cell has a greatly increased chance of firing. This joint firing is almost simultaneous (within 2 ms). Although both cells may fire with other stimuli, such as a large flashed spot, spike correlations will not occur. They will occur however, if the inhibitory neurotransmitter GABA is pharmacologically blocked during the stimulus presentation. Thus, it appears that a dedicated mechanism in the retina permits spike correlations for certain stimuli. In our example, the correlations may encode the continuity of the moving edge.

Third, we began investigating ways to understand responses of ganglion cells to natural images. One difficulty in this investigation is that these images are complex and it is hard to know to which of their aspects a cell responds. In addition, many cells respond nonlinearly, that is, they respond to a complex combination of image features. To overcome these complexities, we have been developing new methods of analysis, which allow building models for cells under these circumstances. These models are linear and nonlinear filters, standard in fields as Electrical Engineering. The filters resemble “receptive fields,” which are structures widely used in visual neuroscience. Our new methods to estimate these filters treat natural images statistically, that is, do not attempt to get answers for individual images but over their ensemble. These methods use state-of-the-art estimation techniques from engineering.


Figure 2 Figure 2: Example of Estimation of Receptive Field from Natural Images.
In this example, we used a realistic model from the literature for a cortical simple cell to simulate its responses. We stimulated the cell with two-thousand natural images. In this image, we show the linear filter obtained for the cell (technically called first-order Volterra Kernel). This filter says how much the cell will respond if stimulated in each position of space. The filter appears in perspective (top) and contour (bottom) plots, and we show both the actual filter (left) and the estimated one (right). The estimate filter gives an excellent approximation to the actual filter.