to represent a critical point at which the image is divided into separate and parallel streams.
From the earliest histologic studies of the vertebrate retina it has been apparent that each major class of cells— photoreceptor, bipolar, horizontal, amacrine, and ganglion— contains within it morphologically distinct subtypes (11). A major theme during the past century of retina research has been the identification of functional correlates for these morphologic differences (12). Among ganglion cells, one correlation that is now well established (and is perhaps not surprising) is between the area of the dendritic field and the area of the receptive field, the former appearing to coincide with and to determine the extent of the latter. Both dendritic field size and cell body size differ markedly between physiologically distinct ganglion cell types. For example, in the cat, X and Y cells correspond, respectively, to the medium (β) and large (α) cell types, and in the monkey, P- and M-type cells correspond, respectively, to the small (midget) and large (parasol) cell types (reviewed in refs.6 and 13). For P and M cells, both dendritic field and soma size increase progressively with increasing retinal eccentricity, and this increase is matched by a corresponding increase in the size of the receptive field. The eccentricity-dependent change in receptive field size accounts for the absence of an illusory dark spot in the one intersection of the Hermann grid upon which the observer fixates (Fig. 2). In the human retina, receptive field sizes have been measured psychophysically by determining the threshold for detection of a small test flash in the presence of a superimposed circular background of varying diameter and constant brightness (14). When the superimposed background is confined to the excitatory center of a center-surround receptive field it produces a persistent activation, thereby decreasing the sensitivity of the cell to dim test flashes. When the superimposed background is enlarged so that it also includes the inhibitory surround, the level of persistent activation is reduced and the sensitivity of the cell approaches that seen with the test flash alone. This psychophysical measure closely matches the eccentricity-dependent size of primate M-type ganglion cell dendritic fields (15) and receptive fields (16).
A second correlation between ganglion cell structure and function relates the level at which the ganglion cell dendrites arborize in the inner plexiform layer and the inputs that the cell receives. By examining the morphologies of individual ganglion cells after recording their light responses, it was discovered that ganglion cells with OFF centers have dendritic arbors in the outer part of the inner plexiform layer (IPL), whereas ganglion cells with ON centers have dendritic arbors in the inner part of the IPL (reviewed in ref. 12). Further subdivisions within the IPL are evident upon close examination of ganglion, bipolar, and amacrine cell dendritic morphologies ( Fig. 3). These are likely to be related, at least in part, to the segregation of chromatic inputs. In one well characterized example, the blue ON/yellow OFF color opponent type of ganglion cell has been shown to be bistratified (17). One dendritic tree is located at that level in the inner part of the IPL where the processes of blue cone bipolar cells terminate, and the second dendritic tree is located in the outer part of the IPL where it presumably receives inhibitory signals from bipolar cells driven by red and green cones.
A third structure–function correlation can be seen in the different projections made by retinal ganglion cells, with the result that distinct aspects of the retinal image are delivered to different destinations in the brain (reviewed in refs.6 and 7). The two principal projections from the retina are to the midbrain and to the dorsal lateral geniculate nucleus (LGN) of the thalamus, the latter projecting to the primary visual cortex. In amphibia and other lower vertebrates the midbrain projection (the retinotectal pathway) constitutes the major output pathway from the retina and mediates simple visually guided behaviors. In primates, the analogous pathway is devoted principally to the control of eye and head movements. Many ganglion cells that project to the midbrain exhibit receptive fields with a high degree of selectivity—for example, to movement in a particular direction.
Ganglion cell axons navigate with extraordinary precision to contact their appropriate targets within the brain. At the optic chiasm, most axons from the nasal but not the temporal half of each retina cross the midline to follow the contralateral optic tract. Central to the chiasm, ganglion cell axons in the primate retinothalamic tract undergo further segregation. Axons from M-type ganglion cells project to the ventral two layers of the LGN while axons from the P-type ganglion cells project to the dorsal four layers; axons derived from the contralateral eye innervate the first, fourth, and sixth layers of the LGN, while those derived from the ipsilateral eye innervate the second, third, and fifth layers; and within each layer of the LGN the pattern of innervation generates a precise retinotopic map that is aligned with each of the retinotopic maps above and/or below it.
The diversity of ganglion cell properties and the precision with which these properties are programmed invite numerous questions regarding underlying molecular mechanisms. We list some of these questions below.
What determines the synaptic specificity of each ganglion cell for the various classes of bipolar and amacrine cells? What attractive or repulsive molecules determine the levels in the IPL where ganglion cells and the various classes of bipolar and amacrine cells synapse? What molecules determine the dendritic field size for each type of ganglion cell?
How do different ganglion cell classes differ in the types of neurotransmitters they use and in the properties and