two intriguing questions: (a) What were the selective pressures underlying the evolution of partial trichromacy in the New World lineage and uniform trichromacy in the Old World lineage? (b) How did the retinal circuitry for color vision coevolve with the establishment of a second visual pigment locus on the X chromosome in the emergence of trichromacy in Old World monkeys?

The next set of papers dealt with the circuitry and development of the retina. Human color vision begins with signals from three types of cones that combine antagonistically to form blue–yellow and red –green opponent pathways. Dennis Dacey (7) showed how the circuits underlying opponency are being deciphered. The macaque monkey retina can be studied in vitro, and photoresponses can be recorded from cells identified by their morphology and binding of specific fluorescent markers. Blue–yellow opponency is mediated by a small bistratified ganglion cell that receives depolarizing inputs from a blue-sensitive “on” bipolar cell and a summed red and green-sensitive “off” bipolar cell. A different kind of circuitry underlies red–green opponency, which is signaled by midget ganglion cells. Dacey proposed that the receptive field centers of these cells get a simple cone input (either maximally red- or maximally green-sensitive), whereas the surround gets both types. This mixed-surround model provides a simple basis for the evolutionary transition from dichromacy to trichromacy.

The determination of cell fate in the vertebrate retina was discussed by Constance Cepko (8). Distinctive cell types are born in overlapping order. In the mouse, ganglion cells, cones, amacrine cells, and horizontal cells arise early in development. Rods come later, followed by bipolar cells and Müller glial cells. Retroviral vectors have been used to insert genetic tags (such as the β-galactosidase gene) for lineage analysis. The significant finding is that retinal progenitors are multipotent. As many as six cell types have been seen to arise from a single precursor. The overlapping birth order of retinal cell types and the multipotency of progenitor cells imply that extrinsic cues play key roles in directing cell fate in the vertebrate retina. Cepko proposed that retinal progenitors undergo a series of state changes that are accompanied by alterations in competence to respond to environmental cues to produce particular cell types. Each state of competence is transient and is endowed by expression of a combination of transcription factors.

The remarkable diversity of ganglion cell properties and the precision of their programming stimulated Jeremy Nathans (9) to pose a set of questions concerning the underlying molecular mechanisms: (a) What determines the synaptic specificity, neurotransmitter type, and dendritic field of each class of ganglion cells? (b) What are the guidance mechanisms that lead ganglion cell axons to precise locations in the midbrain and lateral geniculate nucleus? (c) What are the genetic regulatory circuits specifying ganglion cell type? There is much interest now in identifying transcription factors that control ganglion cell development. Four POU-domain transcription factors (homeodomain proteins) are attractive candidates because they are expressed in subsets of ganglion cells. One of them, Brn-3b, is abundant in P-type but not in M-type ganglion cells. P-type (parvocellular-type) cells have high spatial resolution and exhibit color opponency, whereas M-type (magnocellular-type) cells have high temporal resolution and can respond to small changes in contrast but are achromatic. The importance of Brn-3b is evidenced by the finding that retinas lacking the gene have 70% fewer ganglion cells then do normal retinas.

Ganglion cell axons from the two eyes terminate in adjacent but nonoverlapping eye-specific layers in the lateral geniculate nuclei of adults. By contrast, the inputs are intermixed in development. Carla Shatz ( 10) presented experiments that provide insight into how neural activity contributes to the emergence of eye-specific layers. Segregation takes place in utero before vision is operative but requires ganglion cell signaling. How is this accomplished? Spontaneous action potentials arising from as many as 100 ganglion cells were simultaneously recorded by use of a multielectrode array. The surprising finding was that neighboring cells fired in a concerted manner. Their action potentials occurred within 5 sec of each other, followed by a silent period of up to 2 min before firing resumed. The ganglion cell activity comprised a wave that swept across the retina. Optical recordings monitoring changes in intracellular calcium levels suggested that amacrine cells and ganglion cells act together in generating spontaneous synchronous activity in the developing retina. Shatz proposed that activity-dependent wiring may be generally used in the developing nervous system to help refine early neural connections.

The optic nerve is a severe bottleneck in visual signaling. All information captured by 125 million photoreceptor cells in humans is carried into the brain by only 1 million ganglion cell axons. How does the retina generate a highly efficient representation of the visual scene? Markus Meister (11) described recent experiments suggesting that the retina employs multineuronal coding to compress a large number of distinct visual messages into a relatively small number of optic nerve fibers. Simultaneous recordings of many ganglion cells with a multielectrode array showed that nearby ganglion cells have a pronounced tendency to fire synchronously (within 20 msec of one another). A particular ganglion cell can partake in several different concerted firing patterns. Hence, synchronous firing events, rather than individual action potentials, may be the fundamental symbols of the retinal code. A calculation based on a simple model shows that concerted firing conveys more information than does independent firing and therefore could be advantageously used to enhance spatial and temporal resolution. Meister suggested that multiplexed messages could be decoded in layer IVc of the visual cortex, which contains many more neurons than afferents from the lateral geniculate nucleus.

The last four papers considered higher-level processes in the visual cortex. Charles Gilbert (12) presented experiments showing that receptive field properties of cells in the cortex can be dynamically altered in times ranging from seconds to months. Focal retinal lesions were made at cognate positions in the two eyes to remove visual inputs destined for a particular area of the visual cortex. Over several months, the silenced cortical area regained functioning visual input by an expansion of the representation of the retinal region around the lesion. Furthermore, a transient blind spot could be generated by occluding part of a twinkling random dot pattern. This occlusion led within a few minutes to a reversible expansion of the size of the receptive field of the corresponding cortical cell. It will be interesting to learn the molecular mechanisms underlying this cortical plasticity. Gilbert suggested that experiencedependent changes in cortical function play essential roles in perception.

David Heeger, Eero Simoncelli, and Anthony Movshon conducted a workshop on “Computational Models of Cortical Visual Processing” (13). Their aim was to devise detailed quantitative models of neuronal function that capture the behavior of different classes of cortical neurons with a small number of measurable parameters. One of their models deals with simple cells in the primary visual cortex (V1), which are known to be selective for stimulus position, orientation, size, and direction of motion. The other model is concerned with pattern direction-selective neurons in the extrastriate visual area MT (V5), which have been shown to signal the movement of entire patterns by combining information from several orientations. Both models compute a linear combination of their inputs, rectify this sum, and then divide the neuron's response by a quantity proportional to the pooled activity of many neurons in the cortical neighborhood. Readers can

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement