Vision: Two Speeds in the Retina.

The central (foveal) retina takes about 30 milliseconds longer to signal to the brain than the peripheral retina. In the natural world, a 30 millisecond delay could have real consequences. Why did evolution do it this way?

The types of retinal ganglion cells: current status and implications for neuronal classification.

In the retina, photoreceptors pass visual information to interneurons, which process it and pass it to retinal ganglion cells (RGCs). Axons of RGCs then travel through the optic nerve, telling the rest of the brain all it will ever know about the visual world. Research over the past several decades has made clear that most RGCs are not merely light detectors, but rather feature detectors, which send a diverse set of parallel, highly processed images of the world on to higher centers. Here, we review progress in classification of RGCs by physiological, morphological, and molecular criteria, making a particular effort to distinguish those cell types that are definitive from those for which information is partial. We focus on the mouse, in which molecular and genetic methods are most advanced. We argue that there are around 30 RGC types and that we can now account for well over half of all RGCs. We also use RGCs to examine the general problem of neuronal classification, arguing that insights and methods from the retina can guide the classification enterprise in other brain regions.

Automated computation of arbor densities: a step toward identifying neuronal cell types.

The shape and position of a neuron convey information regarding its molecular and functional identity. The identification of cell types from structure, a classic method, relies on the time-consuming step of arbor tracing. However, as genetic tools and imaging methods make data-driven approaches to neuronal circuit analysis feasible, the need for automated processing increases. Here, we first establish that mouse retinal ganglion cell types can be as precise about distributing their arbor volumes across the inner plexiform layer as they are about distributing the skeletons of the arbors. Then, we describe an automated approach to computing the spatial distribution of the dendritic arbors, or arbor density, with respect to a global depth coordinate based on this observation. Our method involves three-dimensional reconstruction of neuronal arbors by a supervised machine learning algorithm, post-processing of the enhanced stacks to remove somata and isolate the neuron of interest, and registration of neurons to each other using automatically detected arbors of the starburst amacrine interneurons as fiducial markers. In principle, this method could be generalizable to other structures of the CNS, provided that they allow sparse labeling of the cells and contain a reliable axis of spatial reference.

A genetic and computational approach to structurally classify neuronal types.

The importance of cell types in understanding brain function is widely appreciated but only a tiny fraction of neuronal diversity has been catalogued. Here we exploit recent progress in genetic definition of cell types in an objective structural approach to neuronal classification. The approach is based on highly accurate quantification of dendritic arbor position relative to neurites of other cells. We test the method on a population of 363 mouse retinal ganglion cells. For each cell, we determine the spatial distribution of the dendritic arbors, or arbor density, with reference to arbors of an abundant, well-defined interneuronal type. The arbor densities are sorted into a number of clusters that is set by comparison with several molecularly defined cell types. The algorithm reproduces the genetic classes that are pure types, and detects six newly clustered cell types that await genetic definition.

The neuronal organization of the retina.

The mammalian retina consists of neurons of >60 distinct types, each playing a specific role in processing visual images. They are arranged in three main stages. The first decomposes the outputs of the rod and cone photoreceptors into ∼12 parallel information streams. The second connects these streams to specific types of retinal ganglion cells. The third combines bipolar and amacrine cell activity to create the diverse encodings of the visual world--roughly 20 of them--that the retina transmits to the brain. New transformations of the visual input continue to be found: at least half of the encodings sent to the brain (ganglion cell response selectivities) remain to be discovered. This diversity of the retina's outputs has yet to be incorporated into our understanding of higher visual function.

The tasks of amacrine cells.

Their unique patterns of size, numbers, and stratification indicate that amacrine cells have diverse functions. These are mostly unknown, as studies using imaging and electrophysiological methods have only recently begun. However, some of the events that occur within the amacrine cell population--and some important unresolved puzzles--can be stated purely from structural reasoning.

Physiological clustering of visual channels in the mouse retina.

Anatomy predicts that mammalian retinas should have in excess of 12 physiological channels, each encoding a specific aspect of the visual scene. Although several channels have been correlated with morphological cell types, the number of morphological types generally exceeds the known physiological types. Here, we attempted to sort the ganglion cells of the mouse retina purely on a physiological basis. The null hypothesis was that the outputs of the ganglion cells form a continuum or should be divided into only a few types. We recorded the spiking output of 471 retinal ganglion cells on a multielectrode array while presenting 4 classes of visual stimuli. Five parameters were chosen to describe each cell's response characteristics, including relative amplitude of the ON and OFF responses, response latency, response transience, direction selectivity, and the receptive field surround. We compared the results of four clustering routines and judged the results using the relevant validation indices. The optimal partition was the 12-cluster solution of the Fuzzy Gustafson-Kessel algorithm. This classification contained three visual channels that carried predominately OFF responses, six that carried ON responses, and three that carried both ON and OFF information. They differed in other parameters as well. Other evidence suggests that the true number of cell types in the mouse retina may be somewhat larger than 12, and a definitive typology will probably require broader stimulus sets and characterization of more response parameters. Nonetheless, the present results do allow us to reject the null hypothesis: it appears that in addition to well-known cell types (such as the ON-OFF direction selectivity cells) numerous other cell classes can be identified in the mouse retina based solely on their responses to a standard set of simple visual stimuli.

Regular mosaic of synaptic contacts among three retinal neurons.

Retinal bipolar, amacrine, and ganglion cells contact each other within precisely defined synaptic laminae, but the spatial distribution of contacts between the cells is generally treated as random. Here we show that not to be the case. Excitatory inputs to inner retinal neurons were visualized by introduction of a plasmid coding for the postsynaptic protein PSD95-GFP. Our initial finding was that synapses on the dendrites of retinal ganglion cells are regularly spaced, at 2-3-µm intervals, along the dendrites. Thus, the presence of a PSD95 punctum creates a nearby zone from which other inputs appear to be excluded. Despite their great variation in size and different morphologies, the spacing is similar for the arbors of different retinal ganglion cell types. Regular spacing was also observed for the starburst amacrine cells. This regularity is mirrored in the spacing of axonal varicosities of the stratified bipolar cells, which have a regular, nonrandom interval consistent with that of the PSD95 puncta on ganglion cells. Thus, for each level of the inner plexiform layer all three cell types participate in a single 2D mosaic of synaptic contacts. These findings raise a new set of questions: How does the self-avoidance of synaptic sites along an individual dendrite arise and how is it physically maintained? Why is a regular spacing of inputs important for the computational function of the cells? Finally, which of the three players, if any, is developmentally responsible for the initial establishment of the pattern?

Structural remodeling of fibrous astrocytes after axonal injury.

Reactive astrocytes are a pathological hallmark of many CNS injuries and neurodegenerations. They are characterized by hypertrophy of the soma and processes and an increase in the expression of glial fibrillary acidic protein. Because the cells obscure each other in immunostaining, little is known about the behavior of a single reactive astrocyte, nor how single astrocytes combine to form the glial scar. We have investigated the reaction of fibrous astrocytes to axonal degeneration using a transgenic mouse strain expressing enhanced green fluorescent protein in small subsets of astrocytes. Fibrous astrocytes in the optic nerve and corpus callosum initially react to injury by hypertrophy of the soma and processes. They retract their primary processes, simplifying their shape and dramatically reducing their spatial coverage. At 3 d after crush, quantitative analysis revealed nearly a twofold increase in the thickness of the primary processes, a halving of the number of primary processes leaving the soma and an eightfold reduction in the spatial coverage. In the subsequent week, they partially reextend long processes, returning to a near-normal morphology and an extensive spatial overlap. The resulting glial scar consists of an irregular array of astrocyte processes, contrasting with their original orderly arrangement. These changes are in distinct contrast to those reported for reactive protoplasmic astrocytes of the gray matter, in which the number of processes and branchings increase, but the cells continue to maintain nonoverlapping individual territories throughout their response to injury.

Synaptic input of ON-bipolar cells onto the dopaminergic neurons of the mouse retina.

In the retina, dopamine fulfills a crucial role in neural adaptation to photopic illumination, but the pathway that carries cone signals to the dopaminergic amacrine (DA) cells was controversial. We identified the site of ON-cone bipolar input onto DA cells in transgenic mice in which both types of catecholaminergic amacrine (CA) cells were labeled with green fluorescent protein or human placental alkaline phosphatase (PLAP). In confocal Z series of retinal whole mounts stained with antibodies to tyrosine hydroxylase (TH), DA cells gave rise to varicose processes that descended obliquely through the scleral half of the inner plexiform layer (IPL) and formed a loose, tangential plexus in the middle of this layer. Comparison with the distribution of the dendrites of type 2 CA cells and examination of neurobiotin-injected DA cells proved that their vitreal processes were situated in stratum S3 of the IPL. Electron microscope demonstration of PLAP activity showed that bipolar cell endings in S3 established ribbon synapses onto a postsynaptic dyad in which one or both processes were labeled by a precipitate of lead phosphate and therefore belonged to DA cells. In places, the postsynaptic DA cell processes returned a reciprocal synapse onto the bipolar endings. Confocal images of sections stained with antibodies to TH, kinesin Kif3a, which labels synaptic ribbons, and glutamate or GABA(A) receptors, confirmed that ribbon-containing endings made glutamatergic synapses onto DA cells processes in S3 and received from them GABAergic synapses. The presynaptic ON-bipolar cells most likely belonged to the CB3 (type 5) variety.

The morphology and spatial arrangement of astrocytes in the optic nerve head of the mouse.

We evaluated the shapes, numbers, and spatial distribution of astrocytes within the glial lamina, an astrocyte-rich region at the junction of the retina and optic nerve. A primary aim was to determine how the population of astrocytes, collectively, partitions the axonal space in this region. Astrocyte processes labeled with glial fibrillary acidic protein (GFAP) compartmentalize ganglion cell axons into bundles, forming "glial tubes," and giving the glial architecture of the optic nerve head in transverse section a honeycomb appearance. The shapes of individual astrocytes were studied by using transgenic mice that express enhanced green fluorescent protein in isolated astrocytes (hGFAPpr-EGFP). Within the glial lamina the astrocytes were transverse in orientation, with thick, smooth primary processes emanating from a cytoplasmic expansion of the soma. Spaces between the processes of neighboring astrocytes were spatially aligned, to form the apertures through which the bundles of optic axons pass. The processes of individual astrocytes were far-reaching-they could span most of the width of the nerve-and overlapped the anatomical domains of other near and distant astrocytes. Thus, astrocytes in the glial lamina do not tile: each astrocyte participates in ensheathing approximately one-quarter of all of the axon bundles in the nerve, and each glial tube contains the processes of about nine astrocytes. This raises the mechanistic question of how, in glaucoma or other cases of nerve damage, the glial response can be confined to a circumscribed region where damage to axons has occurred.

Remodeling of cone photoreceptor cells after rod degeneration in rd mice.

We studied the survival of cone photoreceptors following the degeneration of rods in the rd mouse. Cones were visualized by selective expression of green fluorescent protein (GFP) following transduction with an adeno-associated virus (AAV) vector. As previously reported, many cones survive after the initial degeneration of the rods. Soon after the initial degeneration, they lose their outer segments and all but a vestigial inner segment; and they partially retract or lose their axon and synaptic pedicle. However, they retain many fundamental features of the cone phenotype, and for many weeks show a polarized morphology indicative of substantial regrowth of processes. The cells retain their laminar position, forming a cell row just distal to a much thinned outer plexiform layer. The somata subsequently enlarge. Most of the cells extend bipolar processes, recreating the original bipolar morphology of a photoreceptor cell--though now turned on its side relative to the native position. The cells express short- or middle-wavelength opsins, recoverin and connexin36. One or more of the polarized processes could often be shown to contain synaptic ribbons, as visualized by antibodies against RIBEYE. The cones do not express protein kinase C alpha, Go alpha, ChX10 or calbindin, markers of bipolar or horizontal cells. The partially differentiated cone morphology persists for at least several months, after which the processes begin to retract and there is slow loss of the cells. Thus, during the time following the loss of their rod-dominated microenvironment, the cones achieve a semi-stable state in which much of their normal phenotype is preserved. Cone photoreceptors in retinas of human RP donors appear from their morphology to undergo a similar progression. The therapeutic window for rescue of cone photoreceptors may be longer than would have been thought.

Restoration of visual function in retinal degeneration mice by ectopic expression of melanopsin.

The rod and cone cells of the mammalian retina are the principal photoreceptors for image-forming vision. They transmit information by means of a chain of intermediate cells to the retinal ganglion cells, which in turn send signals from the retina to the brain. Loss of photoreceptor cells, as happens in a number of human diseases, leads to irreversible blindness. In a mouse model (rd/rd) of photoreceptor degeneration, we used a viral vector to express in a large number of retinal ganglion cells the light sensitive protein melanopsin, normally present in only a specialized subset of the cells. Whole-cell patch-clamp recording showed photoresponses in these cells even after degeneration of the photoreceptors and additional pharmacological or Cd(2+) block of synaptic function. Interestingly, similar responses were observed across a wide variety of diverse types of ganglion cell of the retina. The newly melanopsin-expressing ganglion cells provided an enhancement of visual function in rd/rd mice: the pupillary light reflex (PLR) returned almost to normal; the mice showed behavioral avoidance of light in an open-field test, and they could discriminate a light stimulus from a dark one in a two-choice visual discrimination alley. Recovery of the PLR was stable for at least 11 months. It has recently been shown that ectopic retinal expression of a light sensitive bacterial protein, channelrhodopsin-2, can restore neuronal responsiveness and simple visual abilities in rd/rd mice. For therapy in human photodegenerations, channelrhodopsin-2 and melanopsin have different advantages and disadvantages; both proteins (or modifications of them) should be candidates.

The spatial distribution of glutamatergic inputs to dendrites of retinal ganglion cells.

The spatial pattern of excitatory glutamatergic input was visualized in a large series of ganglion cells of the rabbit retina, by using particle-mediated gene transfer of an expression plasmid for postsynaptic density 95-green fluorescent protein (PSD95-GFP). PSD95-GFP was confirmed as a marker of excitatory input by co-localization with synaptic ribbons (RIBEYE and kinesin II) and glutamate receptor subunits. Despite wide variation in the size, morphology, and functional complexity of the cells, the distribution of excitatory synaptic inputs followed a single set of rules: 1) the linear density of synaptic inputs (PSD95 sites/linear mum) varied surprisingly little and showed little specialization within the arbor; 2) the total density of excitatory inputs across individual arbors peaked in a ring-shaped region surrounding the soma, which is in accord with high-resolution maps of receptive field sensitivity in the rabbit; and 3) the areal density scaled inversely with the total area of the dendritic arbor, so that narrow dendritic arbors receive more synapses per unit area than large ones. To achieve sensitivity comparable to that of large cells, those that report upon a small region of visual space may need to receive a denser synaptic input from within that space.

Neurite arborization and mosaic spacing in the mouse retina require DSCAM.

To establish functional circuitry, retinal neurons occupy spatial domains by arborizing their processes, which requires the self-avoidance of neurites from an individual cell, and by spacing their cell bodies, which requires positioning the soma and establishing a zone within which other cells of the same type are excluded. The mosaic patterns of distinct cell types form independently and overlap. The cues that direct these processes in the vertebrate retina are not known. Here we show that some types of retinal amacrine cells from mice with a spontaneous mutation in Down syndrome cell adhesion molecule (Dscam), a gene encoding an immunoglobulin-superfamily member adhesion molecule, have defects in the arborization of processes and in the spacing of cell bodies. In the mutant retina, cells that would normally express Dscam have hyperfasciculated processes, preventing them from creating an orderly arbor. Also, their cell bodies are randomly distributed or pulled into clumps rather than being regularly spaced mosaics. Our results indicate that mouse DSCAM mediates isoneuronal self-avoidance for arborization and heteroneuronal self-avoidance within specific cell types to prevent fasciculation and to preserve mosaic spacing. These functions are analogous to those of Drosophila DSCAM (ref. 6) and DSCAM2 (ref. 7). DSCAM may function similarly in other regions of the mammalian nervous system, and this role may extend to other members of the mammalian Dscam gene family.