The Synaptic Organization of the Retina

The basic architecture, signal flow, and neurochemistry of signaling through the vertebrate retina is well-understood: photoreceptors, bipolar cells (BCs), and ganglion cells (GCs) are all thought to be glutamatergic neurons and the fundamental synaptic chain that serves vision is photoreceptor → BC → GC. But, our understanding of detailed signaling is far from adequate and a complete description of synaptic interactions or signaling mechanisms is lacking for any retinal network. For example, GCs express different mixtures of ionotropic glutamate receptors (iGluRs) and each receptor can be composed of many different subunits leading to a vast array of possible functional varieties. At a larger scale, network topologies are too numerous to resolve with current physiological or pharmacologic data. Each GC contacts many different amacrine cells (ACs) and a full description of the inputs to any given GC does not yet exist. Physiology can screen only a limited parameter space for any cell. Pharmacology is still an emergent field with many incomplete tools and an immense diversity of neurotransmitter receptor subunit combinations, modulators, and downstream effectors remains to be screened for any cell type. Molecular genetics, despite its power to modulate signaling elements, remains an ambiguous tool for analyzing retinal networks. Morphology, augmented by immunochemistry and physiology, remains the core tool in discovering new details of retinal organization.

Nothing has been as powerful as transmission electron microscopy for discovering retinal networks. Mammalian night (scotopic) vision is a prime example. Its unique pathways were described by Kolb & Famiglietti using electron microscopy. Subsequent physiological analyses ^, provided clarification of how the network functions but would not have yielded the correct network architecture. Further complexities have been discovered by anatomical studies, including the fact that the network rewires in retinal degenerations ( Box 21.1 ). But, electron microscopy has not kept pace with the demands for high-throughput imaging until recently. We are now on the verge of a new era in imaging that will provide a deluge of new information about retinal circuitry. Finally, the basics of retinal development and new findings in neuroplasticity are beyond the scope of this chapter, ^, but the implications should be held in mind throughout: the connections we have long considered as static or hard-wired in retina display many of the same molecular attributes as plastic pathways in brain.

Box 21.1

Retinal Remodeling in Retinal Degenerations

Primary photoreceptor or RPE degenerations leave the neural “inner” retina deafferented
The neural retina responds by remodeling in phases, first by subtle changes in neuronal structure and gene expression and later by large-scale reorganization
In Phase 1 , expression of a primary insult activates photoreceptor and glial stress signals
In Phase 2 , ablation of the sensory retina via complete photoreceptor loss or cone-sparing rod loss triggers revision in downstream neurons
Total photoreceptor loss triggers wholesale bipolar cell remodeling
Cone-sparing degenerations trigger BCs reprogramming, down-regulating mGluR6 expression and up-regulating iGluR expression
Loss of cone triggers Phase 3 : a protracted period of global remodeling, including:
- neuronal cell death
- neuronal and glial migration
- elaboration of new neurites and synapses
- rewiring of retinal circuits
- glial hypertrophy and the evolution of a fibrotic glial seal
In advanced disease, glia and neurons may enter the choroid and emigrate from the retina
Retinal remodeling represents the pathologic invocation of plasticity mechanisms
Remodeling likely abrogates or attenuates many cellular and bionic rescue strategies
However, survivor neurons are stable, healthy, active cells
It may be possible to influence their emergent rewiring and migration habits

The basic signal flow in retina is overlaid on a well-studied cell architecture ( Fig. 21.1 ). Retinal ON and OFF BC polarities are generated in the outer plexiform layer and mapped onto the inner plexiform layer into largely separated zones. The distal sublamina a receives inputs from OFF BCs and therein the dendrites of OFF GCs collect signals via BC synapses. The proximal sublamina b receives inputs from ON BCs and therein the dendrites of OFF GCs collect signals via BC synapses. ON-OFF GCs thus collect inputs from both sublayers.

Kinds of neurons

The retina is a thin, multilayered tissue sheet … an image screen … containing three developmentally distinct, interconnected cell groups that form signal processing networks:

Class 1 :: sensory neuroepithelium (SNE) :: photoreceptors and BCs
Class 2 :: multipolar neurons :: GCs, ACs, and axonal cells (AxCs)
Class 3 :: gliaform neurons :: horizontal cells (HCs)

These three cell groups comprise over 60–70 distinct classes of cells in mammals and well over 100–120 in most non-mammalian retinas.

The SNE phenotype includes photoreceptors and BCs. These cells are polarized neuroepithelia with apical ciliary-dendritic and basal axonal-exocytotic poles. They form the first stage of synaptic gain in the glutamatergic photoreceptor → BC → GC → CNS vertical chain. This aggregates photoreceptor signals into BC receptive fields and amplifies their signals. The basal ends of the BCs form the inner plexiform layer. There are at least 12 kinds of BCs in mammals ^, and BCs delimit different functional zones in the IPL, suggesting nearly 1 micron precision in lamination. Both photoreceptors and BCs use high fusion-rate synaptic ribbons as their output elements, fueled by hundreds to thousands of nearby vesicles. The retina is the only known tissue where SNE cells are arrayed in a serial chain.

As summarized in Figure 21.2 , most mammals possess three classes of photoreceptors: rods expressing RH1 visual pigments, blue cones expressing SWS1 visual pigments, and green cones expressing Long-Wave System green (LWSG) visual pigments. Conversely, the most visually advanced and diverse vertebrate classes (teleost fish, avians, reptiles) possess up to seven known classes of photoreceptors (RH1 rods, SWS1 UV/violet cones, SWS2 blue cones, LWSR and RH2 green members of double cones, LWSR and RH2 green single cones).

Figure 21.2, Photoreceptor cohorts and connections in vertebrates.

Similarly, the diversity of BCs in mammalians is lower (10–13) than non-mammalians (>20). This reduced diversity is a result of the Jurassic collapse of the mammalian visual system, where over half of the visual pigment genes, half of the neuronal classes and almost two-thirds of the photoreceptor classes were abandoned to exploit nocturnal niches. In addition, the disproportionate proliferation of rods in the mammalian retina was accompanied by the loss of mixed rod–cone BCs in mammals and their replacement with pure rod BCs. How this occurred is unknown, but it cannot be due to an absolute selectivity of rod BCs for rods, as they will readily make contacts with cones when rods are lost in retinal degenerations. As we will see, the mammalian retina has exploited a re-entrant use of synapses to enhance scotopic vision. The relationship between BCs and photoreceptors is still unclear, but there is both anatomical and molecular evidence that BCs were initially photoreceptors. For example, many non-mammalians possess BCs Landolt clubs, which are apical extensions extending from a BC primary cilium, extending past the outer plexiform layer into the outer nuclear layer, and containing packets of outer-segment-like membranes. Whether they are photosensitive has never been determined. Further, SWS1 blue cones and cone BCs share some SWS1 cis -regulatory sequences.

The multipolar neuron phenotype

The multipolar neuron phenotype includes ACs, AxCs, and GCs. Multipolar neurons can be further divided into axon-bearing (GCs, AxCs) and amacrine cells (ACs). Mammals display ≈ 30 kinds of ACs. The 15–20 kinds of mammalian GCs ^, are classical projection neurons. GCs are postsynaptic at their dendrites and presynaptic at their axon terminals in CNS projections. So far, all are presumed to be glutamatergic. ACs are local circuit neurons similar to periglomerular cells in the olfactory bulb. ACs lack classical axons and often have mixed pre- and postsynaptic contacts on their dendrites, though some ACs partition inputs and outputs into different parts of their dendritic arbors. Most ACs are GABAergic and the remainder are glycinergic. Several classes of ACs are dual transmitter cells, expressing both acetylcholine and GABA, serotonin and GABA (in non-mammalians) or peptides and GABA or glycine. In between are the AxCs, also known as polyaxonal cells and intraretinal GCs, which have distinct axons that project within the retina. One dramatic example of the AxC phenotype is the TH1 dopaminergic AxC. This cell releases dopamine at unknown but probably axonal sites and likely glutamate at others, similar to nigrostriatal neurons. Some polyaxonal cells are GABAergic. There is no evidence for a glycinergic AxC. Multipolar neurons are characterized by numerous neurites branching in the plane of the retina, most collecting signals from BCs. Multipolar neurons are among the earliest to develop in the retina and quickly define the borders of the IPL and its stratifications. Multipolar neurons all manifest somewhat classical “Gray”-like synapses, generally with small clusters of less than 200 vesicles.

The gliaform cell phenotype

This phenotype contains the horizontal cells (HCs), whose somas and processes are restricted to the outer plexiform layer. Though HCs are multipolar, neuron-like, and may display axons, they do not spike. Further, they express many glial features such as intermediate filament expression and very slow voltage responses. Further, HCs produce high levels of glutathione and make direct contact with capillary endothelial cells in some species, suggesting they play homeostatic roles similar to glia. Even so, HCs clearly mediate a powerful network function, collecting large patches of photoreceptor input via AMPA receptors and providing a wide-field, slow signal antagonistic to the vertical channel. The mechanism of HC antagonism remains a matter of uncertainty and debate. HCs do make conventional-appearing synapses onto neuronal processes in the outer plexiform layer, and in fishes these synapses are made onto dendrites of glycinergic interplexiform cells, a form of AxC. However, these are so sparse in all species and contain so few vesicles that they cannot be the source of the large sustained opponent surrounds of retinal neurons that HC generate. HCs must use some other mechanism.

The phylogenetics of HCs has been thoroughly reviewed. HCs in mammals are postsynaptic to cones at their somatic dendrites. One class of HCs common in mammals (foveal type I in primates, type A in rabbits and cats, and absent in rodents) contacts cones alone. A second class of HCs (extrafoveal type I in primates, type B in rabbit and cats, and the only known HC in rodents) displays axons several hundred microns long that branch profusely and form massive arborizations contacting hundreds to thousands of rods. Another class of primate HC (type II) has axon terminals contacting cones and rods. Importantly, the axon of HCs appears to be electrically inactive and these somatic and terminal regions are believed to act independently. HCs also appear to be early-developing pioneer cells that define the outer plexiform layer. After the GCs and HCs define the layout of the inner and outer plexiform layers respectively, photoreceptors and BCs mature and search for connections.

True glia and vasculature

The neurons of the retina are embedded in an array of vertical Müller glia that span the entire neural retina, forming one-third to one-half of the retinal mass and generating high-resistance seals at the distal and proximal limits of the retina. Most mammalian retinas are vascularized in three capillary beds: at the GC-inner plexiform layer border, the AC-inner plexiform layer border, and the outer plexiform layer. Squirrels (Sciurids) display two beds (at the GC-inner plexiform layer and AC-inner plexiform layer borders; and rabbits (Lagomorphs) have none at all, similar to all other non-mammalian vertebrates. The GC layer of many species also displays classical astrocytes, though their role remains unclear. In brain, astrocytes carry out some of the operations attributed to retinal Müller glia, including transport of spillover K ⁺ and glutamate, and glucose supply via vascular > glial cell > neuron transcellular transport. Why and how most vertebrate retinas function without vasculature remains uncertain, but it is likely that Müller glia act as a surrogate vascular system with the added ability to accumulate large glycogen stores (like hepatocytes) as part of a glucose-skeleton homeostasis. The segregation of retinal astrocytes away from the inner plexiform layer remains a mystery.

Basic synaptic communication

With the discovery of the signaling mechanisms of the neuromuscular junction decades ago, one might have thought that the archetypal synaptic format had been discovered. Yet it has become clear, especially in retina, that every kind of synapse is subtly different, with diverse physics, topologies, and molecular mechanisms leading to very different forms of synapses, most of which do not follow the single presynaptic “bouton” → single postsynaptic target pattern of brain. Further, the arrangement of these systems into synaptic chains in retina is unlike any other known network, including olfactory bulb. In retina, the first stage of synaptic signaling is a direct SNE → SNE synapse ( Fig. 21.3 ): photoreceptor → BC. No other instance of this topology has been discovered in any organism. There are at least six modes of presynaptic–postsynaptic pairing in retina.

Figure 21.3, Basic organization of mammalian photoreceptor synaptic terminals.

1 Photoreceptor ribbon synapses: small-volume multi-target signaling

It is thought that all photoreceptor signaling is glutamatergic, but sporadic indications of cholinergic physiology and molecular markers have been found in many non-mammalians. Glutamate release from photoreceptors is effected by high rates of vesicle fusion at active sites on either side of a large synaptic ribbon positioned close to the pre-synaptic membrane. The presynaptic zone is a protrusion or ridge with vesicle fusion sites positioned on the slopes of the ridge ( Fig. 21.4 ). The releasable vesicle pool is so large that photoreceptors and BCs are capable of maintaining continuous glutamate release in response to steady depolarizations. This, among other things distinguishes photoreceptors and BCs from ACs, which have very small presynaptic vesicle clusters.

Figure 21.4, A detailed schematic of synaptic organization at cone (left) and BC (right) ribbon synapses.

Various vertebrate rods and cones differ greatly in the number of ribbons and postsynaptic targets arrayed within the presynaptic terminals. For example, most mammalian and teleost fish rods have small grape-like presynaptic spherules ≈ 3 µm in diameter with a small entrance aperture leading to an enclosed extracellular invagination or vestibule in which thin postsynaptic dendrites are contained ( Fig. 21.3 ). Importantly, glial processes are excluded from the interior of the spherule and any glutamate release must diffuse out of the spherule to reach the Müller glia. However, mammalian rods express the EAAT5 glutamate transporter and likely regulate their own intrasynaptic glutamate levels. Each spherule contains one or two synaptic ribbons and a few postsynaptic targets. In fishes, the postsynaptic targets are the dendrites of roughly five kinds of mixed rod–cone bipolar cells and one kind of rod horizontal cell. Thus each ribbon serves no less than six different types of postsynaptic targets. In mammals, only two targets are common: the dendrites of one kind of rod BC and the axon terminals of HCs. There are some instances of sparse OFF BC contact in mammals, but this seems to vary with species and may be an evolutionary relict with variable expression rather than a major signaling pathway. In sum, rod spherules form a sparse-ribbon → small volume, sparse-target architecture.

Cones and rod terminals in some non-mammalians (e.g. urodele amphibians) adopt a different topology, with the presynaptic ending expanding to form a foot-piece or pedicle some 3–5 µm wide shaped either like a cupola (fishes) whose broadly concave interior admits some 50–100 or more fine dendrites served by roughly 12 synaptic ribbon sites; or like a true pediment (e.g. primate cones) whose shallow concavity is studded with up to 50 ribbon sites ( Fig. 21.3 ). Cone pedicles in primates target at least ten different kinds of BCs and at least two kinds of HCs. Mouse cone pedicles are smaller but still target 11 kinds of BCs and one kind of HC. In sum, cone pedicles form a multi-ribbon → small volume, multi-target architecture .

2 BC ribbon synapses: semi-precise target signaling

Like photoreceptors, BC signaling is generally considered glutamatergic. Sporadic evidence of exceptions exists. In mammals (especially primates) and amphibians, some BCs contain biomarkers of GABA-related metabolism. In contrast to photoreceptors, BC synaptic endings are topological spheroids, usually multiple (depending on BC type), with dozens to hundreds of ribbons abutting the surface. BCs form no invaginations, so there is no restricted volume into which glutamate is injected by vesicle fusion. In most cases each ribbon is directly apposed to a pair of postsynaptic targets, usually ACs. This is termed a dyad and, while monads, triads and tetrads do occur, dyads dominate. Large BC terminals such as those found in teleost fishes can drive up to 200 distinct processes. Mammalian BCs drive many fewer targets and most BCs have elaborate, branched terminals with connecting neurites often as small as 100 nm. In contrast to photoreceptors, the targets of BCs are focal. BC terminals are largely fully encapsulated by neuronal processes at their release sites to which they are presynaptic or postsynaptic, with rarely direct contact between the terminal and Müller glia near the synaptic release zone. This means that any glutamate that escapes from the synaptic cleft may travel some distance before glial glutamate transporters can clear it. Thus the potential for glutamate overflow at BC synapses is substantial. This may be particularly important for the activation of NMDA receptors, as they are suspected to be displaced from primary AMPA receptors. Thus, BCs form multi-ribbon → semi-precise target architectures.

3 AC and AxC conventional fast synapses: precise presynaptic → postsynaptic signaling

ACs and AxCs are the only retinal cells that make synaptic contacts resembling CNS “Gray”-like, non-ribbon conventional synapses. ACs target BCs, GCs, or other ACs. The targets of most AxCs are not well known but appear likely to be ACs and GCs. Though each AC may make many hundreds of synapses, each synapse contacts only and only one postsynaptic target, similar to classical multipolar neurons in CNS and spinal cord. The dominant fast transmitters of AC systems are GABA and glycine, with GABAergic neurons making up half to two-thirds of the AC population depending on species. Additional transmitters such as acetylcholine, peptides, or serotonin (in non-mammalians) are also associated with GABAergic (in most cases) or glycinergic systems. ^, Acetylcholine (ACh) is a fast excitatory transmitter and is found in paramorphic starburst ACs in mammals and also uses conventional synapses. However, we know of no distinguishing anatomical differences between GABA- and ACh-utilizing synapses in retina.

You're Reading a Preview

Become a Clinical Tree membership for Full access and enjoy Unlimited articles

Become membership

If you are a member. Log in here