Eye Smarter than Scientists Believed

A new review in Neuron attempts to enumerate examples of interesting computation in the retina, arguing that retinal circuitry is much more than an image preprocessor that merely relays the visual world to higher brain areas. The authors, Tim Gollisch at Max Planck and Markus Meister at Harvard, instead argue that within the retina lies circuitry specifically designed to perform ethologically important computations, reducing the high dimensionality of the visual scene into easily utilizable signals such as "a moving object is located there" or "something is approaching from over there." Extracting this information in retina confers the advantage of speed to the visual system, allowing higher brain areas to react quickly to stimuli with high saliency, increasing the likelihood of the organism's survival. Eye Smarter than Scientists Believed: Neural Computations in Circuits of the Retina Tim Gollisch and Markus Meister Neuron, Volume 65, Issue 2, 150-164, 28 January 2010 http://www.cell.com/neuron/abstract/S0896-6273(09)00999-4

The computations the authors explore include object motion detection, approaching motion detection, motion extrapolation and prediction, omitted stimulus response (an error signal elicited at the termination of a train of periodic flashes), saccadic suppression, burst latency coding, and switchable circuits which dynamically gate information flow. Several of these computations employ precisely matched, convergent excitatory and inhibitory input to sharpen the ganglion cell response to the precise feature of interest. Nonlinear rectification of dendritic subunits followed by summation by the cell body achieves sensitivity to that feature over a wide receptive field without averaging away the signal. The review also shows "building-block" circuit models for how each computation is thought to occur.

Besides serving as an excellent review of what I believe ultimately makes retina worth studying, the paper carefully delineates the implications of this mode of thinking about retinal computation. For example, most of the phenomena discussed in the review are elicited by ethologically salient stimuli of the kind that would not occur in the white noise flicker used to derive LN models of the cells' responses. Acknowledging the positive impact of white noise analysis on the field, the authors recommend a return to probing the retina with more naturalistic stimuli.

Indeed, some early studies of retinal function were guided by much more ethological sensitivity and tried to relate ganglion cell responses to specific tasks (Lett- vin et al., 1959; Levick, 1967). The next generation overturned this anecdotal thinking and applied the rigorous new engineering tools from systems analysis (Rowe and Stone, 1980). Instead of showing the retina some arbitrary photographs of flies, these workers sampled the stimulus space systematically, feeding the retina with sine waves and white noise to measure its transfer function. Sadly, all the resulting ganglion cell receptive fields looked like Mexican hats with a biphasic time course, and the only possible conclusion was that the interesting visual computations are performed later in the brain (Stone, 1983). Now we understand that the problem was not with the ganglion cells, but with the stimuli. For example, the OMS [object motion sensitive] ganglion cells discussed above have perfectly bland-looking center-surround receptive fields when studied with white-noise flicker. Of course the particular condition that reveals their function—differential motion of an image patch and its background—never occurs during white-noise flicker, whereas it represents a common occurrence on the retina in real life.

Nevertheless, the ultimate goal of understanding the population response to authentic natural scenes remains out of reach. Consequently, a balance struck with a reductionistic approach and ethologically important stimuli will be most informative for the present.

Looking forward, the authors are optimistic that new technologies will help solve longstanding questions in the field as well as open up new lines of research. The ability to genetically target, label, and manipulate individual cell types is a promising direction for the field, as many of these interesting signals are output by distinct types of ganglion cells that are not easily morphologically resolved. Additionally, improved experimental access to interneurons, likely with optical reporters of neural activity, will allow direct testing of the circuit models hypothesized to implement the computations described in the review.