Cones

The mammalian retina contains two types of photoreceptor, rods and cones.⁴⁵ Rods account for 95% of all photoreceptors. They are numerous, with slender outer segments, densely packed and specialized for high sensitivity under dark or starlight conditions. Cones are larger, with tapering outer segments, and they are found in the top row of the ONL (see Fig. 15.3). Cones make up only ~5% of photoreceptors^28,45 but they provide high-acuity color vision in daylight conditions when photons are abundant. This versatile combination of rods and cones and their associated circuits covers an intensity range of around 10 log units from the darkest night to bright sunlight.⁴⁶ While the average visual scene has a range of intensities covering 2–3 log units, the continual adaptation of retinal sensitivity slides this operating range through the entire range of light intensities. This is a critical function of the retina because outside the normal operating range we are functionally blind. Common examples include the inability to see momentarily when entering a dark cinema or driving into the setting sun. Much of the adaptation takes place in photoreceptors but, as we shall see below, this is accompanied by major changes in the neural pathways through the retina. This is arguably the most important function of the retina, after light detection itself.

There are approximately 5 million cones in the human retina and 190 000 in the mouse retina.^47,48 Cones make up approximately 5% of the total photoreceptors in humans, compared to 2.8% in the mouse,⁴⁹ so we are all rod-dominated in terms of absolute numbers. One exception is the ground squirrel, which is truly cone-dominated. Importantly, cones are not evenly distributed. In human retina, there is a massive peak at the fovea (Fig. 15.3) where the density reaches around 200 000/mm², approximately 100 times the density in the periphery.^28,47 This is the region of maximum acuity, although the peak density slightly exceeds experimentally measured visual acuity due to blurring by the optics of the eye.^50–52 Where the ganglion cell axons gather to form the optic nerve there are no photoreceptors and their absence from this location is the cause of the blind spot (Fig. 15.3).

Cones support color vision and, in old-world primates and humans, there are three classes: blue, green and red. They are maximally sensitive to 430, 530, and 561 nm light, respectively.^53–55 Other mammals, including cats, rabbits, and rodents, have an evolutionary ancient form of color vision based on green and blue cones only. The presence of red and green cone opsins in a tandem array on the X chromosome is thought to be due to a recent gene duplication and underlies the preponderance of color blindness among males.⁵⁶ Using adaptive optics to correct for blurring in the lens and cornea, the distribution of red, green, and blue cones can be mapped in the living human eye.⁵⁷ Surprisingly, the distribution of cones was random and clumpy (Fig. 15.6). In addition, there appears to be enormous variation in the red/green cone ratio among individuals with normal color vision.

Fig. 15.6 The mosaic of red, green, and blue cones in the human retina. This image, taken from a human subject using adaptive optics, shows the distribution of the three cone classes. Blue cones make up a small fraction, <10%, but make a regular mosaic. Red and green cones have a clumpy, random distribution. In this subject, the red cones outnumber the green cones but this ratio is highly variable, even in subjects who have normal color vision.

(Courtesy of Austin Roorda; after Roorda A, Williams DR. The arrangement of the three cone classes in the living human eye. Nature 1999;397:520–2.)

Blue cones are present as a minority: they make up approximately 10% of the cone population (Fig. 15.6). This is not enough to support high acuity but calculations show that the blue cone density is sufficient to support the reduced resolution caused by visual aberration at the blue end of the spectrum.³⁰ At the very center of the fovea, blue cones are absent.^29,58 The fact that this is not readily apparent is due to the relatively poor spatial acuity of color vision compared to the luminance-driven pathways.

Red and green cones cannot be differentiated morphologically and the red and green cone opsins are so closely related that they are also indistinguishable. However, blue cones can be mapped with a selective antibody against blue cone opsin^29,59 or by their selective accumulation of certain dyes.⁵⁸

Rods

The ONL contains photoreceptor cell bodies, both rods and cones.⁴⁵ Even in the cone-dominated human retina, rods far outnumber cones, by a factor of 20 : 1, so they account for most of the ONL except at the fovea. The human retina contains approximately 100 million rods, and they pool signals to provide high sensitivity for dark-adapted vision, say starlight, which appears monochromatic. A lack of color vision is the hallmark of rod-mediated vision. Rods are absent within 350 µm of the fovea but reach a peak density in an annular region at about 20° eccentricity (Fig. 15.3). This does not match the area of maximum scotopic acuity, which occurs around 5°,⁶⁰ so it has been suggested that another component of the rod pathway, such as the AII amacrine cells, present at a much lower density, forms a bottleneck to limit acuity^61,62 (see below).

Cone pedicles and rod spherules

Cones and rods make contact with bipolar and horizontal cells in the OPL. Cone axons descend through the massed ranks of rod somas in the ONL to terminate in a two-dimensional array of cone pedicles at the OPL (Fig. 15.7). Near the fovea in primate retinas, the cone axons are splayed out radially so that the pedicles form an annular array around the center (Fig. 15.7B). Cones form two specialized contacts, ribbon synapses and flat contacts, with postsynaptic neurons. Ribbon synapses are so named because of electron-dense structures at invaginations where they contact depolarizing (or ON) bipolar and horizontal cells (Fig. 15.8). Cone pedicles also make flat contacts along the base of the pedicle with hyperpolarizing (or OFF) bipolar cells. Rod axons also descend to form synapses with a single type of depolarizing bipolar cell as well as horizontal cells. The terminals of rods are called spherules, and, similar to cones, they use a ribbon synapse. In contrast to cones, there is only a single invagination in each, containing two rod bipolar and two horizontal cell dendrites.

Fig. 15.7 Primate cones. (A) Vertical section of macaque retina shows cones stained with an antibody against cone arrestin. The outer segments of a few cones, <10%, have been double-labeled with an antibody against blue cone opsin. From each cone, an axon descends to the cone terminal or pedicle. (B) Whole-mount view shows how the axons run obliquely away from the fovea to terminate in an annulus of cone pedicles. (C) Close to the fovea, the cone pedicles form a tightly packed array connected by very fine telodendria. The dark regions within each pedicle are the invaginations penetrating from beneath. (D) More peripherally, the cone pedicles are widely spaced and connected by longer telodendria which are the sites of cone-to-cone coupling.

(Antibody donated by Peter MacLeish, images courtesy of J O’Brien and SC Massey.)

Fig. 15.8 Laminated postsynaptic structure of the cone pedicle. (A) Cross-section of a single cone pedicle contains synaptic ribbons and synaptic vesicles. Horizontal cell dendrites (red) invaginate the cone pedicle deeply and laterally. ON cone bipolar dendrites (green) assume a central position beneath the ribbon. (B) Whole-mount view, in the plane just below the multiple synaptic ribbons, shows the relative positions of horizontal and ON bipolar processes. The OFF bipolar contacts are beneath this plane of focus. (C) Electron microscopic view of a cone pedicle. Arrowheads mark synaptic ribbons. Horizontal cell processes (H) are lateral, ON bipolar dendrites (star) central. OFF bipolar cells (asterisk) form basal synapses away from the synaptic ribbon. Beneath each cone pedicle lies a snake’s nest of postsynaptic processes. There are also specialized desmosome-like contacts of unknown function between horizontal cell processes (small arrows).

(Reproduced with permission from Haverkamp S, Grunert V, Wassle H. The cone pedicle, a complex synapse in the retina. Neuron 2000;27:85–95.)

Each cone pedicle is 6–8 µm in diameter and, near the fovea, contains 20 synaptic ribbons and, in the periphery, around 40.⁶³ Cones release glutamate constantly in the dark and the synaptic ribbons are thought to support this high rate of release. As we will describe further below, the multiple ribbon and flat synapses on each cone pedicle form connections with many different types of bipolar cell.

A sense of the complexity of these synapses can be obtained by labeling some of the individual components. Antibodies reactive against a vesicle protein such as synaptophysin provide a way to label photoreceptor terminals in the OPL, because photoreceptor axon terminals are filled with vesicles. To visualize the synaptic ribbons, antibodies to kinesin II can be used. Finally, antibodies to the mGluR6 receptor, which is located on the dendritic tip of the depolarizing bipolar cells, mark one of the postsynaptic processes. When visualizing just the cone pedicle, there are a number of indentations or invaginations (Fig. 15.8). Within each invagination, horizontal cell dendrites are lateral and high, approaching the synaptic ribbons. ON cone bipolar cell processes are central but slightly lower at the synaptic ribbon. In contrast, OFF bipolar dendrites form basal synapses with the cone pedicle at a distinctly lower level. Staining a piece of retina for synaptophysin, kinesin II, and mGluR6, begins to show the complexity of the structures (Fig. 15.9). Rod spherules contain a single large horseshoe-shaped ribbon with two bipolar cell dendrites nestled within. The cones have numerous and slightly smaller ribbons with each also nestled against an mGluR6-labeled dendrite. What is not visible are the horizontal cell dendrites that also invaginate, but are slightly lower than the bipolar cells.

Fig. 15.9 Synaptic structure of rod spherules and cone pedicles. (A) The terminals of rods and cones are packed with synaptic vesicles and contain synaptic ribbons. An antibody against synaptophysin (green), a synaptic vesicle protein, labels rod spherules which appear round (some are outlined to the lower right) and cone pedicles which are polygonal (larger outlines). Each rod spherule contains a single large synaptic ribbon which is curved like a horseshoe, stained with an antibody against kinesin II (red). The cone pedicles contain a complex of smaller synaptic ribbons (also red). (B) The dendritic tips of rod bipolar and ON cone bipolar cells can be marked with an antibody against the glutamate receptor, mGluR6 (blue). The central invagination of each rod spherule contains a pair of mGluR6-positive rod bipolar dendrites. At each cone pedicle, there are a number of finer ON bipolar dendrites. (C) The rod bipolar dendrites (blue) are nestled within the horseshoe formed by the synaptic ribbon (red) at each rod spherule. At the cone pedicles, the ON bipolar dendrites also approach closely to the cone synaptic ribbons. (D) A triple-label image showing rod spherules and cone pedicles (green, some outlined), synaptic ribbons (red) and the postsynaptic bipolar dendrites tagged for the glutamate receptor mGluR6 (blue). Scale bars = 10 µm.

(Courtesy of W Li and SC Massey.)

Around 10–12 cone bipolar cells plus multiple horizontal cell dendrites contact each cone, so under the pedicle there may be as many as 200 postsynaptic processes.⁶⁴ In fact, the cone pedicle may be the most complex synaptic structure in the brain. The distribution of postsynaptic processes under the cone pedicle is laminated in a stereotyped fashion (Fig. 15.8).^63,64

The fundamental problem for rods is to detect a small brief hyperpolarization due to a single photon against a noisy background of thermal noise and the stochastic nature of transmitter release.^65,66 One strategy is to minimize background variation by maintaining a high sustained-release rate in the dark. The synaptic terminal of a rod, or rod spherule, is about 2 µm in diameter and contains a very large synaptic ribbon (Fig. 15.9), a specialization thought to be associated with a high rate of transmitter release. The spherule is packed with synaptic vesicles, containing the neurotransmitter glutamate, to support sustained transmitter release. There is a single invagination (imagine a closed fist and insert a finger between thumb and forefinger to make a pocket) containing a tetrad of processes, two from horizontal cells and two rod bipolar dendrites.⁶⁷ This structure brings the postsynaptic processes close to the release site and may prevent spillover to adjacent rods. These anatomical specializations appear to reduce variation in rod signaling so that small single-photon responses can be detected with high reliability.

Photoreceptor coupling

While the synapses between photoreceptors and second-order neurons are extremely complex, there are still additional connections between photoreceptors. These take the form of electrical coupling, mediated by gap junctions. Close to the fovea, the cone pedicles are densely packed, almost touching, and connected by very fine processes known as telodendria (Fig. 15.7C).⁴⁵ Rod terminals are either absent or displaced slightly higher, towards the outer segments, in this region. More peripherally, the cone pedicles are widely spaced and the telodendria are much more prominent (Fig. 15.7D). The contact points between telodendria of neighboring cones are the sites of connexin (Cx)36 gap junctions, which mediate electrical coupling between cones.^68,69 Recordings between neighboring cones also show that they are coupled.^68,70 This may seem puzzling at first because it should lead to a loss of acuity and blurring. However, the cone array actually oversamples the signal so this leads to little or no loss. Instead, the coupling is thought to reduce noise, which is random, while the light-driven signals, which are correlated between close neighbors, will be reinforcing. It has been calculated that this may improve the signal-to-noise ratio by 77% – a large gain for little loss.⁷¹ Coupling between red/green cones is indiscriminate, which may reflect the close absorption curves of red and green cones.⁷⁰ Morphologically, blue cone pedicles are slightly smaller than those of red/green cones and they have only a few withered dendrites, which rarely touch neighboring cones.⁷² Hence, blue cones are not coupled into the red/green network.^68,70 This may serve to preserve spectral information in the color pathways.

There are also gap junctions between cone pedicles and rods.⁷³ These allow rod signals to enter cones, and rod-mediated signals can be detected in second-order neurons that are thought to be connected exclusively to cones.⁷⁴ Responses with the signature of rod origin have also been recorded in primate cones.⁷⁵ Because rods far outnumber cones, the influence of many rods on a single cone may be substantial. It is now thought that rod–cone coupling forms an alternative rod-driven pathway that may be important at intermediate light intensities, in the mesopic range.^76,77 This influence may relate to the enhancement of cone electroretinogram (ERG) amplitudes in humans and mice by light adaptation, where the cone ERG gradually increases in amplitude following the onset of a steady adapting field.^78–81 Blocking gap junctions inhibits this effect in the mouse retina. In the human, however, the adaptation dependence of these changes has a photopic (cone) signature. Rod–cone coupling also is very dynamic and influenced by circadian rhythms, increasing at night and decreasing during the day.⁸²

Photoreceptors release glutamate in the dark

The first intracellular recordings from cones were surprising because they showed that photoreceptors hyperpolarize in response to light.⁴ This means they are relatively depolarized and release their neurotransmitter, glutamate, in the dark. From the viewpoint of signal information, the sign of the photoreceptor response makes no logical difference; photoreceptors produce graded responses modulated around the mean light level. When a photon is absorbed, the visual pigment is activated and then a cascade of other biochemical events is triggered.⁸³ This is known as phototransduction and it leads to the closure of a cGMP-gated cation channel in the cell membrane to produce hyperpolarization and a reduction in the release of glutamate.⁸⁴ The postsynaptic responses to light are in fact due to a decrease of glutamate release. Thus, glutamate uptake also plays a key role in retinal neurotransmission because the clearance of glutamate must be rapid to provide a fast postsynaptic response to light.⁸⁵ The hyperpolarizing light response of photoreceptors is now well established and it is supported by studies of vesicle turnover, which is much higher in the dark.⁸⁶ Furthermore, synaptic blocking studies produce the appropriate responses in the different second-order neurons and there is a stereotyped array of distinct glutamate receptors associated with specific cell types. Glutamate, with its library of postsynaptic receptors, seems particularly suitable to orchestrate the large variety of postsynaptic responses at the cone pedicle.

Rods operate in a manner essentially similar to cones: they hyperpolarize in response to light increases, albeit there are many molecular differences. However, everything about the rod pathway in the retina is designed for maximum sensitivity. This is reflected in the anatomical details and synaptic connections of rods.⁶⁶ Rods can respond to single photons, obviously the design limit, with a binary (all or nothing) signal, but visual threshold requires a signal in 5–10 rods. Depending on the species, about 20–100 rods converge on to a single rod bipolar cell^87,88 and this high convergence also contributes to the high sensitivity of the rod pathway. If 100 rods converge to a single rod bipolar cell and 100 rod bipolar cells converge to a ganglion cell, then the absolute threshold for vision is determined by the product, approximately 1 photoisomerization per 10 000 rods.

Midget bipolar cells

The primate retina is unusual in that the central retina is dominated by midget bipolar cells. Within the central 10 mm, there is one OFF midget bipolar cell and one ON midget bipolar cell for each cone. In this area, they account for more than 80% of all cone bipolar cells.^130,150 Midget bipolar cells have very small dendritic fields and receive input from single cones and make output to single midget ganglion cells. This is the so-called private line, one cone to one midget bipolar cell to one midget ganglion cell. The ON midget bipolar cells invaginate the cone pedicle, ramify in sublamina b of the IPL, and contact ON midget ganglion cells. OFF midget bipolar cells make flat or basal contacts with cones and terminate in sublamina a where they contact OFF midget ganglion cells. Most investigators think this specialization of the primate retina was designed to achieve maximum acuity at high cone density. It may also, by virtue of the single cone connections, which are automatically color-coded, serve red/green color vision.

Blue cone bipolar cells

In general, diffuse cone bipolar cells contact every cone within the dendritic field and this gives them a characteristic appearance. However, in the primate retina, one bipolar cell type is distinctly different in that it has long dendrites that bypass many cones to seek out only blue cones.^151,152 The dendrites are labeled for mGluR6 and the axon descends deep into sublamina b so the blue cone bipolar cells are ON cells.^17,153

Rod bipolar cells

In contrast to the cone bipolar cells, there is only one morphological type of rod bipolar cell. They are numerous, with a mop-head of fine dendrites that may receive input from as many as 80–120 rods in the rabbit retina (Fig. 15.16A).⁸⁸ This very high convergence contributes to the sensitivity of the primary rod pathway. A single dye-injected rod bipolar cell is shown in Figure 15.16B. The dendrites branch profusely but avoid the cone pedicles within the dendritic field. Instead, all the terminal dendrites invaginate a rod spherule and they are double-labeled for mGluR6. Only one terminal at each rod spherule comes from this rod bipolar cell. The other half of each mGluR6 doublet is claimed by another unlabeled rod bipolar dendrite. Thus, each rod contacts two different rod bipolar cells.⁹⁷ Each rod bipolar cell gives rise to a long slender axon that descends to sublamina 5 of the IPL (Fig. 15.16). Physiologically, rod bipolar cells give ON responses to light stimulation and this is consistent with the labeling for mGluR6 and the depth of stratification.^13,154 Rod bipolar cells do not usually contact ganglion cells directly. Instead, the primary output of rod bipolar cells is to AII amacrine cells, which pass on the rod signal, or to S1 and S2 amacrine cells^42,154 that provide a powerful negative-feedback signal to rod bipolar terminals (Fig. 15.17; see below). More than 90% of the rod bipolar output is on to these amacrine cells.

Fig. 15.16 Rod bipolar cells and contacts in the rabbit retina. (A) There is a single morphological type of rod bipolar cell that can be stained with antibodies against protein kinase C (PKC). (Courtesy of W Li and SC Massey.) (B) A single rod bipolar cell from the rabbit retina, filled with Lucifer Yellow (green), focus at the outer plexiform layer, has a spray of fine dendrites that contact many rod spherules (red puncta). (C) At higher resolution, mGluR6 receptors, which stain the dendritic tips of the rod bipolar cells (red), are doublets where they enter into the rod spherule invagination. The general rule is that each pair of rod bipolar dendrites at a rod spherule originates from different rod bipolar cells. A cone pedicle, marked by a cluster of dimmer, finer ON cone bipolar dendrites, is also outlined (dashed line).

(B, C, Reproduced from Li W, Keung JW, Massey SC. Direct synaptic connections between rods and OFF cone bipolar cells in the rabbit retina. J Comp Neurol 2004;474:1–12, with permission from the authors and Wiley-Liss, a subsidiary of John Wiley.)

Fig. 15.17 The primary rod pathway. (A) There is only one type of rod bipolar cell (rod B), which produces ON responses to light and terminates at the bottom of the inner plexiform layer (IPL) in sublamina 5. Rod bipolar cells express mGluR6 receptors and are hyperpolarized by glutamate released from rods at a sign-inverting synapse, marked with a minus sign. Rod bipolar cells release glutamate and are connected to AII amacrine cells and S1/S2 amacrine cells via excitatory receptors, marked by a plus sign. In turn, S1/S2 amacrine cells make inhibitory reciprocal synapses mediated by γ-aminobutyric acid, back on to the rod bipolar terminal. (B) Vertical section of macaque retina, the AIIs amacrine cells, stained for calretinin (green) descend around the rod bipolar cell axons (protein kinase C, red) as they descend through the IPL. Arrow marks fainter staining of cone bipolar cell DB4 at a different level in the IPL.

(Panel B reproduced with permission from Massey SC, Mills SL. Antibody to calretinin stains AII amacrine cells in the rabbit retina: double-label and confocal analyses. J Comp Neurol 1999;411:3–18.)

The high convergence allows the rod bipolar cell to collect signals from many rods but is also potentially noisy. However, the rod-to-rod bipolar synapse has a nonlinearity by which small signals are thresholded.¹⁵⁵ The price for this is that many small signals are rejected but the reduction in noise is worth it. Some near-threshold signals may be lost but when a photon signal is captured it has a high signal-to-noise ratio and is transmitted very reliably.¹⁵⁶

Multiple rod pathways

Rods are utilized under low light conditions and provide information over 5 log units. It is now clear that there are at least three pathways, referred to as primary, secondary, and tertiary, by which rod signals reach ganglion cells (Fig. 15.18). They are thought to process signals under slightly different conditions, the primary being the most sensitive and the tertiary being the least sensitive.

Fig. 15.18 Rod and cone pathways. The primary rod pathway (rod → ON bipolar cell → AII amacrine cell) provides an excitatory input to AII amacrine cells, which are electrically coupled to cone ON bipolar cells and inhibit cone OFF bipolar cells. The secondary rod pathway (rod → cone → bipolar cell) arises from electrical coupling between rods and cones and allows the flow of rod signals into the cones and glutamate release into the cone ON and OFF bipolar cells. The tertiary rod pathway (rod → OFF bipolar cell) is a direct glutamatergic input from rods to cone OFF bipolar cells. Note that the secondary and tertiary pathways are in principle independent of the primary pathway. Red arrows indicate sign-inverting synapses, and green arrows indicate sign-preserving synapses.

(Modified with permission from van Genderen MM, Bijveld MM, Claassen YB, et al. Mutations in TRPM1 are a common cause of complete congenital stationary night blindness. Am J Hum Genet 2009;85:730–6.)

The primary rod pathway utilizes rod bipolar cells, which make ribbon synapses with two postsynaptic amacrine cells in sublamina 5 of the IPL (Fig. 15.17). One of these postsynaptic neurons is the AII amacrine cell, and the other is either an S1 (A17 in the cat) or S2 amacrine cell (Fig. 15.17).^42,154 These two wide-field GABA amacrine cells both make reciprocal synapses with rod bipolar terminals and thus provide another stage of negative feedback. The conspicuous terminals of rod bipolar cells literally plug into holes in the meshwork of dendrites provided by these amacrine cell types. Rod bipolar cells, like other bipolar cells, release glutamate and the contact points with the AII matrix are covered with glutamate receptors of the AMPA subtype (Fig. 15.19).^157,158 The glutamate receptors of S1/S2 amacrine cells have not been completely clarified, but the A17 (probably the S1 equivalent in the rodent) uses a Ca²⁺-permeable AMPA glutamate receptor via L-type Ca²⁺ and Ca²⁺-activated potassium (BK) channels¹⁵⁹ to modulate feedback inhibition on to the rod bipolar cell and control glutamate release at this synapse. On the rod bipolar cell the postsynaptic targets are the GABA_A and GABA_C receptors.^160–162

Fig. 15.19 Rod bipolar input to AII is mediated by glutamate receptors. (A) The matrix of AII amacrine cells in the rabbit retina, stained for the calcium-binding protein calretinin (green), is decorated with punctate glutamate receptors (red) of the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) subtype. (B) The AII dendrites carefully wrap around the rod bipolar terminals (blue) at this level. Critically, the glutamate receptors only occur at the contact sites between rod bipolar cells and AII amacrine cells (red + green + blue = white). Intervening sections of AII dendrite, between rod bipolar terminals, have no glutamate receptors. This indicates that AMPA receptors are located at synaptic contacts between rod bipolar cells and AII amacrine cells.

(Reproduced from Li W, Trexler EB, Massey SC. Glutamate receptors at rod bipolar ribbon synapses in the rabbit retina. J Comp Neurol 2002;448:230–48, with permission from the authors and Wiley-Liss, a subsidiary of John Wiley.)

So, there are two obvious questions: how do rod signals reach ganglion cells and, if there is only one type of rod bipolar cell, how are both ON and OFF signals generated in dark-adapted conditions? The answer lies in the way the rod pathway is integrated into the cone pathways via the AII amacrine cell (Fig. 15.18). AII amacrine cells are glycinergic neurons and they also make conventional inhibitory glycinergic synapses with the axon terminals of OFF cone bipolar cells in sublamina a via alpha 1 glycine receptors.¹⁶³ In turn, the AII itself is modulated by a glycinergic input at synapses expressing alpha 3 glycine receptors.¹⁶⁴ Their distal processes make electrical synapses or gap junctions with ON cone bipolar cells in sublamina b and provide a direct, presumably sign-conserving, input signal to the ON cone bipolar cells.^43,165,166 The signal transfer via gap junctions is presumed to be sign-conserving. Thus, while the cone pathways split via different postsynaptic glutamate receptors in the outer retina, the rod pathway bifurcates at the level of the AII amacrine cell. It is often said that the rod signals piggyback on the cone pathways.

AII amacrine cells themselves are well coupled by gap junctions, as can be demonstrated by injecting a single AII amacrine cell with the diffusible tracer Neurobiotin, which passes through gap junctions to label all the coupled cells.^166,167 Figure 15.20 shows that many AII amacrine cells are coupled as well as an overlying group, consisting of four or five different types, of ON cone bipolar cells. The AII-to-AII gap junctions occur preferentially at dendritic crossings (Fig. 15.21) and AII coupling is absent in a Cx36 knockout mouse.¹⁶⁸ Coupling in this complex heterocellular network is modulated by dopamine and, perhaps more importantly, by light.¹⁶⁹ The underlying mechanism is by phosphorylation of Cx36, and this regulation can be achieved on a cell-by-cell basis.¹⁷⁰

Fig. 15.20 Coupling between AII amacrine cells and ON cone bipolar cells. This three-dimensional reconstruction from a series of confocal images shows the result of injecting a diffusible tracer, such as Neurobiotin, into an AII amacrine cell in the rabbit retina. A matrix of coupled AII amacrine cells (khaki), all with the characteristic morphology, is labeled. In addition, the tracer also spreads into a complex of ON cone bipolar cells (blue) which consists of 3–5 types.

(Courtesy of EB Trexler and SC Massey.)

Fig. 15.21 AII/AII gap junctions contain connexin (Cx)36. A confocal image of the AII dendrites (red) in sublamina b of the inner plexiform layer. Staining for Cx36 (green) shows that Cx36 plaques are abundant in the AII matrix. Image analysis indicates that there is a high probability of a Cx36 plaque when two AII dendrites cross.

(Reproduced from Mills SL, O’Brien JJ, Li W, et al. Rod pathways in the mammalian retina use connexin 36. J Comp Neurol 2001;436:336–50, with permission from the authors and Wiley-Liss, a subsidiary of John Wiley.)

These gap junctions are bidirectional. This means that electrical signals and tracers can pass through the gap junction in both directions.^171,172 One consequence is that glycine from the AII amacrine cell can enter ON cone bipolar cells by this route and indeed most ON bipolar cells contain glycine, even though they use glutamate as a neurotransmitter. The source of the bipolar cell glycine was definitively established by blocking gap junctions with carbenoxolone. This changed the labeling pattern for glycine, which was subsequently diminished in bipolar cells.¹⁶⁷ Another more relevant consequence of bidirectional coupling is that not only do rod signals pass into the cone pathways but, in fact, cone signals can pass from ON bipolar cells into the AII network. The implication is that the AII network can also influence cone signals.

The importance of these gap junction pathways has been elegantly demonstrated by the development of a Cx36 knockout mouse.¹⁶⁸ In these animals, filling an AII amacrine cell with Neurobiotin yielded only one cell: there was no coupling without the expression of Cx36. In contrast to the wild type, in the knockout animals, no rod signals are detectable in recordings from ON ganglion cells. In the absence of Cx36 gap junctions, rod level signals do not pass into the ON cone pathways. Of course, the OFF pathways are not dependent on transmission via gap junctions and so rod-driven OFF signals are maintained. One obvious explanation is the absence of AII/ON bipolar gap junctions, described above. However, the Cx36 knockout may also interfere with rod/cone coupling in the outer retina. In fact, both these pathways must be missing to eliminate the transmission of rod signals to ON pathways. In either case, this is the first time that a gap junction connection has been shown to be essential for the function of a neuronal pathway in the mammalian CNS.

In the rabbit retina, the gap junctions between one type of ON cone bipolar cell that is selectively labeled by antibodies against the calcium-binding protein calbindin also involve Cx36.¹⁷³ However, the properties of AII/bipolar gap junctions are different from AII/AII gap junctions,¹⁶⁶ suggesting that the bipolar side of the gap junction is different in some way, perhaps due to phosphorylation or even the expression of a different connexin by bipolar cells. There is some evidence that bipolar cells express Cx36.¹⁶⁸ However, there is convincing evidence that in some ON bipolar cells the coupling is mediated by Cx36 (AII)/Cx45 (BC) heterotypic gap junctions.¹⁷⁴

Secondary and tertiary rod pathways

The primary rod pathway is used under threshold conditions such as starlight^108,154 and it is absolutely dependent on synaptic transmission between rods and rod bipolar cells via mGluR6 receptors. However, when these synapses are blocked with APB it was still possible to record rod level signals in OFF ganglion cells.¹⁷⁵ This alternative pathway was attributed to rod/cone coupling, for which there is both anatomical and physiological evidence. This second rod pathway runs from rods via gap junctions to cones and then into cone bipolar cells and ganglion cells in the usual way (Fig. 15.18). It has been suggested that this pathway operates in mesopic conditions of intermediate light intensity.⁷⁶

This pathway has also been reported for the mouse retina. However, the rod-driven responses persisted even in a transgenic mouse line in which cones were eliminated.¹⁷⁶ The absence of cones would seem to rule out rod/cone coupling as a pathway. Therefore, a third rod pathway was proposed involving direct connections between rods and OFF cone bipolar cells and subsequently these connections were found.^177,178 This pathway was first thought to be a specialization of the rodent retina but it has now been identified in cats and rabbits.^97,179 It also is present in ground squirrel, where a specific type of OFF bipolar cell, the b2, was shown to contact rods and provide a means for rods to provide fast photoreceptor signaling.¹⁸⁰ Thus, this third rod pathway may be a general feature of the mammalian retina.

It has been suggested that the rod to OFF bipolar pathway operates at high mesopic/low photopic conditions, around the transition from rod to cone vision. In general, the different rod pathways may be designed to cover different intensity ranges and perhaps they are selectively connected to specific ganglion cell types but, as yet, there are no data on this point. There is physiological evidence from the mouse retina that different ganglion cell types have different intensity response functions but the ganglion cell types have not been identified and the contributing pathways are unknown.¹⁶⁸ The function of these novel retinal pathways under different light intensities is an important and active area of current research.

AII amacrine cells

The AII amacrine cell, also known as the rod amacrine cell, is the most numerous of the amacrine cells, accounting for 11% of the total. It is correctly written with a Roman numeral, which is retained from an early classification scheme, while other numbered amacrine cells use Arabic numerals. AII amacrine cells have a distinctive bistratified morphology, which makes them easy to identify, even in retinal slices.¹⁹⁹ The soma protrudes into the IPL and turns into a thick axon that descends to sublamina b of the IPL and branches into an overlapping matrix.²⁰⁰ This is the site of glutamate input from rod bipolar cells (Fig. 15.20) and output, via gap junctions to ON cone bipolar cells. There are also fine dendrites, which terminate in lobules in sublamina a. At this level, AII amacrine cells make glycinergic inhibitory contacts with OFF cone bipolar cells. They also receive input from OFF cone bipolar cells at this level, the function of which is unclear.¹⁵⁴

Primate AII amacrine cells are similar in morphology to those of other mammalian species.^61,62 The density of rods reaches a maximum at about 20° from the fovea but psychophysical measurements in humans indicate that, in dark-adapted conditions, the maximum acuity occurs at approximately 5° from the fovea, a substantial mismatch. The rod pathway for maximum acuity is thought to be rods → rod bipolar cells → AII → midget bipolar cells → midget ganglion cells (Fig. 15.18). There is good evidence that AII amacrine cells contact midget bipolar cells and other ganglion cell types are too sparse to support the maximum resolution. Now, rods and rod bipolar cells far outnumber AII amacrine cells so they will not present a limit to scotopic acuity. In addition, both ON and OFF midget bipolar cells and ON and OFF midget ganglion cells have a 1 : 1 : 1 correspondence with the tightly packed cones of the central retina. That leaves the AII amacrine cell as the lowest density, the bottleneck, in the pathway. In macaque retina, the peak density of AII amacrine cells is about 5000 cells/mm² around 5°, which matches the peak of dark-adapted acuity. If we consider the mosaic of AII amacrine cells and, from sampling theory, calculate the maximum resolution for such an array, we find that it closely matches the peak acuity from psychophysical experiments. Together, these results indicate that the density of AII amacrine cells sets the limit of dark-adapted resolution in central retina.

The AII network seems to be very important since it is found in all mammalian retinas. In general terms, the AII network is thought to reduce noise by canceling out random events which are not correlated and by reinforcing light-driven events that are synchronized.^201,202 Coupling is also strongly modulated by dopamine¹⁶⁹ and processes from the DACs make a ring around the neck of each AII where it protrudes into the IPL (Fig. 15.23). Of course, we think that dopamine is the general signal for light adaptation and, in fact, light itself seems to modulate the strength of coupling in the AII network.²⁰³ Disrupting the AII network interrupts the rod-driven ON pathways, as described above. However, even in the rod-driven OFF pathways, which are not routed via AII gap junctions, the sensitivity is reduced by approximately 1 log unit in the Cx36 knockout mouse.²⁰³ This indicates that coupling in the AII network adapts to ambient lighting conditions to optimize rod signaling over a large dynamic range.

Fig. 15.23 Dopaminergic amacrine cells form a dense matrix. The matrix of dopaminergic dendrites, stained for tyrosine hydroxylase (green), blankets the retina. There are many small varicosities surrounding AII amacrine somas, stained for calretinin (red).

(Courtesy of SC Massey.)

The AII amacrine cells also appear to have a day job.^204–206 During the day, dark objects that approach the viewer activate a particular class of ganglion cells. An important component of the circuit is the AII. However, unlike in the dark, the information flow is effectively reversed. Depolarization of cone ON bipolar cells results in depolarization of the AII through gap junctions, which increases inhibition by the AII on to OFF bipolar cells. This is a crossover inhibition pathway by which ON channels inhibit OFF pathways.²⁰⁷ The dual use of the AII amacrine depending on lighting condition maximizes circuit efficiency.

S1 and S2 amacrine cells

The S1 amacrine cell of the rabbit retina (A17 in the cat) is a wide-field GABA amacrine cell with straight radiating dendrites decorated with prominent varicosities (Fig. 15.24A). The S2 GABA amacrine cell is smaller, the dendrites are more tangled, and the varicosities are smaller but more numerous (Fig. 15.24B).^208,209 Both cell types contribute to a dense overlapping meshwork at the level of the rod bipolar terminals (Fig. 15.25). These large cells are relatively numerous so the dendritic overlap is huge with coverage factors as high as 500. Although the rabbit retina contains little endogenous serotonin, for some unknown reason these cells take up serotonin. Therefore, they are sometimes known as indoleamine-accumulating amacrine cells and this provides a simple way to label the entire population. Electron microscopy shows that S1/S2 amacrine cells make reciprocal synapses with rod bipolar terminals, i.e., they receive input at a rod bipolar ribbon synapse and nearby they synapse back on to the rod bipolar terminal.⁴²

Fig. 15.24 (A) S1 and (B) S2 amacrine cells filled with Lucifer Yellow, from the rabbit retina. These amacrine cells have prominent varicosities, which are synaptic structures, along slender dendrites.

(Reproduced with permission from Zhang J, Li W, Trexler EB, et al. Confocal analysis of reciprocal feedback at rod bipolar terminals in the rabbit retina. J Neurosci 2002;22:10871–82.)

Fig. 15.25 S1/S2 matrix. (A) The S1/S2 matrix (green), stained by serotonin uptake, from whole-mount rabbit retina. (B) Rod bipolar terminals (blue) fill holes in the S1/S2 matrix. (C) AII dendrites (red) also contact the same rod bipolar terminals. (D) Merged triple-label image shows S1/S2 processes and AII dendrites surrounding every rod bipolar terminal.

(Reproduced with permission from Zhang J, Li W, Massey SC. Confocal analysis of reciprocal feedback at rod bipolar terminals in the rabbit retina. J Neurosci 2002;22:10871–82.)

The varicosities contain presynaptic markers; they are wrapped around rod bipolar terminals (alternating with AII dendrites) and opposed by synaptic ribbons and GABA receptors (Fig. 15.26). In fact, the varicosities are synaptic sites. Confocal analysis of double-label material shows that every varicosity is in synaptic contact with a rod bipolar terminal.²⁰⁹ This means that the functional role of the S1/S2 amacrine cells is to provide a level of GABA-mediated negative feedback to rod bipolar terminals. This is their job and they do nothing else; they have no output to any other cell besides the rod bipolar cell.

Fig. 15.26 S1 synapse on a rod bipolar terminal. This three-dimensional reconstruction shows a single varicosity from an S1 amacrine cell (green) wrapped around a rod bipolar terminal (blue). The transparency has been adjusted so we can see through these structures. GABA_C (red) receptors are found on the surface of the rod bipolar terminal, including at the interface with the S1 (red + green + blue = white). This is an example of locating a synaptic structure by confocal microscopy.

(Courtesy of W Li and SC Massey.)

These cells provide a massive inhibitory input to rod bipolar terminals. S1/S2 amacrine cells carry about 300 and 500 varicosities, respectively, and the density of varicosities has been calculated as 330 000/mm². Each rod bipolar terminal receives input from about 25 S1 and 50 S2 varicosities. They are both apposed by GABA_A and GABA_C receptors.²¹⁰ Thus, the inhibitory input to bipolar terminals will have two components: a large sustained component from GABA_C receptors, which are only expressed by bipolar terminals, and rapid initial transient input from GABA_A receptors, which are more widespread.^{138,160,189,211–213} The S1 and S2 amacrine cells clearly have different spatial and coupling properties.¹⁹¹ Hence, the S2 input dominates within 200 µm and provides an inhibitory signal that matches the size of the surround recorded from AII amacrine cells. The larger, well-coupled S1 amacrine cell may provide a more distant network or global signal. This analysis suggests that the presence of both S1 and S2 amacrine cells is not redundant: each cell contributes different components of lateral inhibition in the rod pathway. Together, these components will summate to modulate the spatial and temporal properties of rod bipolar output. Further, each of the synaptic sites of S1 (A17) amacrine cells may operate independently of the whole. This is possible because each varicosity is electrically isolated from its neighbor. This was tested recently by Diamond and colleagues, who showed that at this reciprocal synapse of the rod ON bipolar cell and the A17 amacrine cell, BK channels control calcium influx via L-type voltage-dependent Ca²⁺ channels. In conjunction with the cable properties of the A17 neurites the outputs are compartmentalized, keeping individual reciprocal synaptic dyads independent.^159,188,214 Such local processing has an important consequence: it allows a single A17 to process upwards of 500 signals independently of each other. This represents a large increase in processing power based on what amounts to parallel circuits in a single cell.¹⁸⁸

It should be emphasized that the reciprocal feedback described at the rod bipolar terminal is, in fact, a general case. It is likely that the terminals of all bipolar cells make reciprocal synapses with GABA amacrine cells and bipolar terminals are literally surrounded by GABA-positive profiles. In the mouse retina, as well as other species, GABA_C receptors, a postsynaptic target of feedback inhibition, are expressed on cone ON and OFF bipolar cells. The current mediated by this receptor is slow and sustained, making it a likely target of feedback control on the cone bipolar cells, as it does in rod ON bipolar cell. The absence (in GABA_C receptors knockout mice) or pharmacological block of this receptor increases both the spontaneous activity and gain of ON-center ganglion cells relative to wild-type.^213,215,216

There seem to be GABA_C receptors on all bipolar cell terminals, although the density for each cell type varies.^211–213 If each bipolar cell type, with its unique address, had one GABA amacrine cell type, that would account for 10–12 of the amacrine cell types. The “sharing” of some broadly stratified amacrine cells would tend to reduce this number while those cases, like the rod bipolar cell, which receive feedback inhibition from two amacrine cell types, will inflate it. In either case, negative feedback provides greater stability, increased frequency response, and a wider bandwidth at the level of bipolar cell terminals. A major role for at least some of the multiple types of GABA amacrine cells is to provide negative feedback for all bipolar cells.

Dopaminergic amacrine cells

A small number of amacrine cells, less than 0.1%, contain dopamine. To make up for their low density, these cells have very long axon-like processes that can run for millimeters across the retina.²¹⁷ The net result is a dense high-coverage plexus of dopaminergic dendrites, which covers the entire retina. Most of the plexus is in sublamina 1, adjacent to the INL, but there are minor bands in sublaminae 3 and 5 of the IPL. In some species, particularly fish, the dopamine neurons project to the OPL. In other words, they are interplexiform cells. In the rabbit retina, a few stunted processes run towards the OPL but they do not form a plexus. In the macaque retina, some dopaminergic dendrites run into the INL where they surround the somas of AII amacrine cells (Fig. 15.23).⁶²

Electron microscopy shows that the dendrites in layer 1 receive direct bipolar input at ribbon synapses, many of which appear to be monads, as opposed to the usual dyadic arrangement with two postsynaptic processes.²¹⁸ Yet, recordings from GFP-labeled cells in the mouse retina have shown that DACs produce a variety of light responses: ON transient, ON sustained and light independent.^23,24 But how do these ON responses arise if most dopaminergic processes are stratified in the OFF sublamina? If DACs are ON cells, then they break the rules of stratification for the IPL.

Several different mechanisms have been proposed to overcome this difficulty. For example, it has been shown that ON bipolar inputs to the DACs occur on the minor bands in layer 3.²¹⁹ However, it is also clear that there are bipolar ribbon inputs to the DAC in sublamina 1.²¹⁸ Most recently, it has been suggested that the melanopsin-containing ipRGCs may provide an input to the DACs, because the sustained features of the DAC response and the spectral signature match those of the melanopsin ganglion cells.²⁴ Furthermore, the light-evoked DAC response is maintained when photoreceptors degenerate, although transient light responses are blocked by APB, suggesting they arise from the ON bipolar pathway.²³ In addition, melanopsin ganglion cell dendrites that ramify in sublamina a also have ON responses, thus breaking the stratification rules themselves.^25,27 Most recent evidence suggests that many ON cone bipolar cells make axonal ribbon or ectopic synapses as their axons descend through stratum 1 (Fig. 15.27). These axonal ribbons provide ON excitatory input via AMPA-type glutamate receptors to DACs (and to ipRGCs) in the OFF layer of the IPL. Thus, axonal ribbons break the stratification rules of the IPL and provide an additional accessory ON sublayer in the outer IPL.^26,27

Fig. 15.27 Dopaminergic amacrine cells, stained with an antibody against tyrosine hydroxylase (TOH: blue), receive ectopic, axonal ribbon synaptic input. (A) The dopaminergic dendrites run in sublamina a. At this level, ON cone bipolar cells, labeled with an antibody against calbindin (red), are passing through so the descending axons appear as dots. However, they are nearly all immediately adjacent to the dopaminergic dendrites (arrows). (B) At higher magnification, it can be seen that synaptic ribbons (green) also occur at these contact points. Thus, the arrows show ON synaptic input, via axonal ribbons, in sublamina a, the OFF layer of the inner plexiform layer.

(Reproduced with permission from Hoshi H, Liu WL, Massey SC, et al. ON inputs to the OFF layer: bipolar cells that break the stratification rules of the retina. J Neurosci 2009;29:8875–83.)

In the mouse retina, DACs also contain GABA²²⁰ but DACs from the rabbit retina do not display a GABA signature.³⁴ Dissociated DACs are spontaneously active²²¹ and, in contrast to GABA, dopamine seems to be steadily released in a paracrine fashion to diffuse throughout the retina.²²² Dopamine levels are higher in the daytime and seem to be under circadian control. The most recent work indicates that the dopaminergic cells themselves express clock genes.^223–226 This is appropriate because dopamine is intimately involved in the diurnal rhythms associated with the processes of light and dark adaptation. Dopamine acts at a variety of sites in the retina, many of which seem obviously linked to light or dark adaptation.²²⁷ For example, coupling in the AII network is dramatically reduced by dopamine^169,170 and, in the outer retina, dopamine acts to promote cone input to second-order neurons. Conversely, in the dark and absence of dopamine, rod inputs predominate. Specifically, by controlling rod/cone coupling, it seems like dopamine in the light, or its absence in the night, presets rod and cone pathways as appropriate for the lighting conditions.^82,228

A role for D1 receptors has recently been shown to be critical in the GABAergic control of the rod ON bipolar cell circuit that enhances light sensitivity and increases the dynamic range of their responses at dim-light levels.²²⁹ The location of this control remains to be discovered and could occur via a novel dopaminergic to GABAergic amacrine serial synapse in the IPL or a horizontal cell-mediated input modulated by dopamine in the outer retina. The retina is like a self-optimizing network controlled, at least in part, by the circadian release of dopamine.

Starburst amacrine cells

The ChACs, also known as starburst amacrine cells on account of their unique morphology, form two mirror-symmetric populations on either side of the IPL in all mammals.²⁰ They are the only source of ACh in the retina. In whole-mount view, they are radially symmetric with dendrites radiating from the soma in all directions (Fig. 15.28). The bipolar cell inputs, via AMPA-type glutamate receptors, occur all along the branches but the outputs seem to be restricted to the varicosities in the terminal third of each dendrite. The cells in the INL produce OFF responses and stratify in sublamina a while the displaced starburst amacrine cells are ON cells branching in sublamina b. They are relatively numerous wide-field amacrine cells. Consequently, they have a very high coverage factor and in cross-section their dendrites form two continuous bands in the IPL that look like train tracks in cross-section (Fig. 15.28). This is a very dense, narrowly stratified matrix of overlapping dendrites that are cofasciculated with the direction selective (DS) ganglion cell types.^230,231

Fig. 15.28 Starburst amacrine cells. (A) The somas of displaced starburst amacrine cells stained with 4’,6-diamidino-2-phenylindole (DAPI). This method is used to target this cell type. (B) A single starburst amacrine cell, filled with Lucifer Yelllow. (C) A section of retina stained for choline acetyltransferase (red) showing conventional and displaced somas and two bands in the inner plexiform layer. INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer.

(Courtesy of SC Massey.)

Starburst amacrine cells also contain GABA.²³² In fact, we should probably think of these cells as wide-field GABA amacrine cells, which also contain ACh. The presence of two classical neurotransmitters in the same cell is unusual and it has aroused a great deal of interest. The release of both transmitters seems to be dependent on calcium, which indicates a conventional mechanism using synaptic vesicles. But GABA and ACh release have a differential sensitivity to calcium. This implies that the two transmitters may be released from different populations of vesicles but whether cotransmission occurs at the same sites is unknown.²³³

The starburst amacrine cells have been a focus of attention because they seem to be directly involved in neuronal circuits that produce DS responses in certain ganglion cells. DS ganglion cells were discovered nearly 50 years ago and, although the mechanism has been under intense investigation, the details have still not been resolved. However, starburst amacrine cells are in the middle of the puzzle. Ablating starburst amacrine cells, in a mouse line engineered to respond to an immunotoxin, blocked DS responses in ganglion cells.²³⁴ Furthermore, optokinetic nystagmus, a type of reflex eye movement elicited by a moving stimulus, was also blocked. This combination of physiological and behavioral data provides convincing evidence that starburst amacrine cells are required for directional selectivity and optokinetic eye movements.

Starburst amacrine cells are found at the same depth as the bistratified ON/OFF DS ganglion cells, which are the most sensitive of all ganglion cells to ACh. But cholinergic input alone is not sufficient to produce DS responses.²³⁵ Rather, the mechanism seems to depend on asymmetrical GABA inhibition that comes from the starburst amacrine cells. In dual recordings, stimulating a starburst amacrine cell on the null side produced an inhibitory GABA input to DS ganglion cells.^233,236

Furthermore, calcium imaging by two-photon microscopy showed that individual dendrites of starburst amacrine cells can themselves produce directional responses, even though recordings from the soma do not.³³ The dendrites respond better to a centrifugal stimulus, moving away from the soma, than a centripetal stimulus, towards the soma. The directional signals are due to the intrinsic morphological properties of starburst amacrine cells and a voltage gradient generated by the asymmetric distribution of voltage-dependent channels.^237,238 It has also been suggested that a chloride gradient generated by the regional expression of chloride transporters may contribute to the directional asymmetry.²³⁹ In addition, there is GABA-mediated starburst-to-starburst inhibition, which may enhance or fine-tune the DS responses.²³³ These important results confirm a long-held suspicion that there is a great deal of local processing among amacrine cell dendrites and it suggests that starburst amacrine cells are the source of directional signals in the retina.

Given that starburst amacrine cells generate directional signals, they release GABA, and they synapse directly with ON/OFF DS ganglion cells, then the final requirement for specific wiring is that individual starburst dendrites with the appropriate directional asymmetry are connected to ON/OFF DS ganglion cells with the same orientation of the null/preferred axis. (Remember, there are four subsets of ON/OFF DS ganglion cells with different axes, roughly aligned with the four main compass points.) By a combination of calcium imaging to identify the directional preferences combined with serial block-face electron microscopic imaging to reconstruct the underlying retinal circuits, the specific synaptic connections have now been identified.²⁴⁰ Starburst dendrites with a specific orientation make synaptic connections, identified by varicosities containing presynaptic machinery, with ON/OFF DS ganglion cells with an antiparallel null axis. Thus, preferred excitation in starburst dendrites can provide the GABA input for null inhibition to ON/OFF DS ganglion cells. Often, the starburst dendrites and the ON/OFF DS ganglion cell dendrites are oriented nearly parallel and it is well known that these two cell types are cofasciculated at the level of the two cholinergic bands. Once a local DS response is generated in the ganglion cell, dendritic spikes propagate to the soma with a high probability of generating somatic spikes.²⁴¹ After 50 years, the neuronal circuitry underlying directional selectivity has finally been revealed. How this arises developmentally and what the role is for cholinergic input are further important questions that remain.

Finally, the network of starburst amacrine cells appears to play a key role in retinal development.²⁴² Imaging of neonatal retina shows the presence of calcium waves, which propagate across the retina and elicit bursts of firing in all ganglion cells. These waves apparently are not required to establish directional selectivity, which arises independently of both waves and visual experience.²⁴³ However, in one model, wave activity is thought to promote the development of ganglion cell connections in a retinotopic manner. The cholinergic plexus is present surprisingly early in development and, in the early stages, the calcium waves are blocked by cholinergic antagonists.^244–246 This suggests that spontaneous activity in the ChACs may initiate the calcium waves. This was thought to be light-independent; however, it now appears that light input via ipRGCs (light-sensitive ganglion cells: see below) is required for aspects of the spiking that are important for retinogeniculate segregation.²⁴⁷

Does each ganglion cell type represent a visual channel?

The simple answer to this question is yes, we think so. However, we still do not have a definitive value for the number of channels and, for those ganglion cell types that have been classified morphologically, it is not always clear which physiological class they belong to. The correspondence between functional anatomy and physiological type is only partially complete.²⁷³ A problem that has yet to be adequately resolved is what features of the functional responses of the ganglion cells are the ones that set individual classes apart. In fact, function alone is probably not sufficient and projection pattern, axon diameter, and terminal morphology may also be important. Regardless, it is widely believed that the retina, with the ganglion cells as the output, produces a large set of neural codes, most of which have not been deciphered, that are transmitted to the brain in parallel both on a temporal scale and to a variety of central structures. These are the visual channels, some of which carry different versions of Mona Lisa (Fig. 15.1), where we began. Below are some examples of different ganglion cells/visual channels that we are beginning to understand.

In a few cases, a particular ganglion cell type seems to have a specific role. Midget ganglion cells in the primate retina are a good example. There are two types, ON and OFF, which are broadly stratified towards the middle of the IPL. Midget ganglion cells are numerically dominant in central retina, where they may account for as many as 95% of all ganglion cells.¹⁵⁰ Importantly, we know that in central retina there is a 1 : 1 : 1 correspondence from cones → midget bipolar cells → midget ganglion cells. Beyond 2 mm or 10°, the midget ganglion cells form separate dendritic clusters that receive input from multiple midget bipolar cells. Sampling theory, based on the density of midget ganglion cells, closely predicts the curve describing a psychophysical measure of human acuity, except close to the fovea where the cone pedicles are laterally displaced.²⁷⁴ Thus, it seems clear that midget ganglion cells transmit a high-acuity version of the visual scene from the center of your gaze, the central retina. The 1 : 1 : 1 ratio also means that individual midget ganglion cells will be color-coded according to the type of cone at the beginning of the chain. Midget ganglion cells project to the parvocellular layers of the lateral geniculate nucleus which also carry color signals. Midget ganglion cells are a specialization of the primate retina, perhaps the final point in the relentless survival-driven goal of maximum acuity. There is no exact match in other mammalian species but an equivalent role could be played by beta ganglion cells in the cat retina. In the rabbit retina, there are neither midget nor beta ganglion cells but an ON and OFF pair of relatively small ganglion cells, G4, which are rather broadly stratified toward the middle of the IPL.²⁵⁶ These are at least morphologically similar. Consistent with the idea that the midget/beta cells are a specialization of animals with a fovea or foveal-like specialization, the rodent retina lacks these types of ganglion cells.

Some of the most convincing evidence that ganglion cells carry specific channels of visual information comes from comparative anatomy, where it can be seen that certain ganglion cell types recur frequently across species. Perhaps the best example is the alpha ganglion cell, which has been reported in many mammalian retinas.²⁷⁵ Alpha ganglion cells have large somas, come in paramorphic ON and OFF pairs, make up less than 5% of all ganglion cells, and have a large dendritic field with a characteristic radiate branching pattern (Fig. 15.31). In the rabbit retina, alpha ganglion cells are narrowly stratified just outside the cholinergic bands. Physiologically, they have transient nonlinear (Y-like) properties, they respond to stimuli of high temporal frequency, and project to the magnocellular layers of the lateral geniculate nucleus. Their responses are not color-coded. Alpha ganglion cells seem well suited for detection of low luminance contrast over large areas of the visual field, such as a blurred copy of the Mona Lisa. Their occurrence over such a wide range of species suggests they play a fundamental role common to the visual needs of many animals, e.g., detection of motion. Parasol ganglion cells in the primate retina and A-type cells of rodents have similar properties, although whether they correspond exactly to alpha cells is an open question. There appear to be wide-field ganglion cells in the primate retina that are morphologically closer to alpha ganglion cells in other species.²⁶⁰

Fig. 15.31 Dye-injected ganglion cells in the rabbit retina. The white cell is an ON/OFF directionally selective ganglion cell surrounded by four much larger transient ON DS ganglion cells. These two ganglion cell types have obviously different dendritic morphology. In addition, the transient ON DS ganglion cells form a nonrandom mosaic. There are also some dye-coupled amacrine cells in the background.

(Reproduced with permission from Hoshi H, Tian LM, Massey SC, et al. Two distinct types of ON directionally selective ganglion cells in the rabbit retina. J Comp Neurol 2011;519:2509–21.)

The ON/OFF DS ganglion cells of the rabbit retina, reported by Barlow and Levick nearly 50 years ago,²⁷⁶are one of the best characterized of the retinal ganglion cells. In these ganglion cells, a bar moving in the preferred direction through their receptive field produces a strong burst of spikes, but the same stimulus moving in the opposite direction produces little or no response. As first shown by Amthor and colleagues,²⁷⁷ bistratified cells fasciculate extensively within the two bands of starburst amacrine cell processes and they have a distinct retroflexive branching pattern (Fig. 15.32). DS ganglion cells are exquisitely sensitive to nicotinic cholinergic agonists but, surprisingly, directional selectivity is not dependent on this cholinergic input.²³⁵

Fig. 15.32 ON/OFF directionally selective ganglion cell. A single ON/OFF directionally selective ganglion cell was dye-injected with Neurobiotin and then, using the confocal microscope, the two dendritic fields were imaged and color-coded. Note the retroflexive branching pattern which fills most of the available space, in contrast to the branch pattern of the alpha ganglion cell above.

(Reproduced with permission from Kittila CA, Massey SC. Pharmacology of directionally selective ganglion cells in the rabbit retina. J Neurophysiol 1997;77:675–89.)

Paired recordings showed that starburst amacrine cells on the null side of the DS ganglion cells provide asymmetric GABA inhibition to DS ganglion cells.^236,278 The source of the GABAergic input is from starburst amacrine cells, which contain GABA as well as ACh.²³² Individual dendrites of the starburst amacrine cells generate directional responses due to their intrinsic properties and this is the origin of directional signals in several other types of DS ganglion cells.³³ This difference is consistent with recent morphological results using an extensive three-dimensional reconstruction of functionally identified DS ganglion cells by serial block-face electron microscopy that showed specific inhibitory inputs from starburst dendrites with the appropriate orientation.²⁴⁰

At present, four ganglion cell types with different null/preferred axes have been described, one of which is dye-coupled and shows a precise dendrite-to-dendrite tiling pattern with a coverage close to 1.²⁷⁹ DS cells with the other preferred axes are thought to tile the retina independently so that if the dendrites of two DS ganglion cells overlap, they must have different preferred directions. They can also be differentiated by molecular markers and have slightly different central projections.²⁸⁰ The four null/preferred axes are aligned with the extraocular muscles²⁸¹ so it is often suggested that the ON/OFF DS cells play a role in the control of eye movements. Cells with a similar morphology have been noted in mouse,^261,262 rat, cat (the iota cell),²⁸² and even primate retina.²⁶⁰ In the mouse two posteriorly tuned, essentially morphologically identical DS cells have been defined that have subtle receptive field differences, and different central projection patterns.²⁶⁸ If each compass point is duplicated in the same manner, this would predict there should be a total of eight ON/OFF DS ganglion cells. It is not unreasonable to assume a similar situation will exist in most other mammals.

There is another DS cell in the rabbit retina that has interesting properties. Its dendrites are monostratified in the ON starburst band and have a similar retroflexive branch pattern. They also receive specific null-side inhibition from starburst amacrine cells.²⁸³ However, these ON DS cells respond to much slower movements and its three variants have different null/preferred axes, which are aligned with the semicircular canals.²⁸⁴ Consistent with the hypothesis that DS ganglion cells are critical to eye movement control, the ON DS ganglion cells project to the medial terminal nucleus of the accessory optic system.^284,285 Ganglion cells with a strikingly similar morphology have been reported in mouse, rat, and primate retina.²⁶⁰ The value of this comparative anatomy becomes clear when we consider an experiment in the mouse retina. When starburst amacrine cells are ablated, DS responses in ganglion cells are abolished and the optokinetic nystagmus is also blocked.²³⁴ This is a type of reflex eye movement involved with tracking moving objects with the same velocity range as the ON DS ganglion cell. It suggests that ON DS cells provide the input to a system for tracking or stabilizing the retinal image. At least one gene associated with an eye movement disorder is exclusively expressed in starburst amacrine cells, which provide the directional input to the ON DS ganglion cell.²⁶⁷

A second type of ON DS ganglion cell has recently been identified in the rabbit retina.^285–287 This cell type, originally reported by Ackert and colleagues,²⁸⁸ is distinct from the cell described above in terms of morphology and dendritic stratification, just above the lower band of ChAC processes. If the starburst amacrine cells are the source of directional signals in the retina, it is not clear how DS signals are generated in this cell type. Finally, in a stunning example of a cell type-specific molecular marker, a whole population of ganglion cells from the mouse retina was stained using a junctional adhesion molecule B (JAM-B)-driven GFP mouse line.²⁶⁵ This procedure identified a uniform population of ganglion cells with asymmetrical dendritic trees aligned dorsal to ventral. A similar OFF cell type, G3 in the Masland catalog, has been reported for the rabbit retina.^256,289 In the mouse, this cell type has been characterized as OFF DS, although in the rabbit retina it was identified as orientation-selective.²⁹⁰ If it is truly DS, then the mechanism may be unusual because it does not stratify with the starburst amacrine cells. Together, these observations suggest that there may be multiple circuits to generate directional signals.²⁹¹

A ganglion cell for the control of pupil diameter and circadian rhythm

It has long been known that certain visually driven activities persist in mice where rods and cones have completely degenerated. These apparently blind mice retain pupillary light reflexes as well as circadian rhythms and a similar profile has been reported in totally blind humans.²⁹² It was concluded that these reflexes are driven by connections separate from the usual image-forming pathways. This issue was resolved when a set of ganglion cells were found to express an additional visual pigment called melanopsin. They form a relatively sparse mosaic (only 2000 per mouse retina) with loopy dendrites (Fig. 15.33) in sublamina a and b and they respond with a slow sustained discharge dependent on the light intensity.³⁶ Even when these cells are isolated, they still produce light responses.²⁹³ In other words, the melanopsin-containing ganglion cells are intrinsically photosensitive and they are referred to as ipRGCs.

Fig. 15.33 Melanopsin ganglion cell ectopic inputs. (A) Low-resolution picture shows melanopsin ganglion cells in the rabbit retina have sparse randomly arranged dendrites. (B) Staining certain ON cone bipolar cells with an antibody against calbindin showed that the descending axons were frequently adjacent to the melanopsin-positive ganglion cells. These are the sites of axonal ribbon inputs which provide ON input to melanopsin ganglion cells that stratify in sublamina a of the inner plexiform layer.

(Reproduced with permission from Hoshi H, Liu WL, Massey SC, et al. ON inputs to the OFF layer: bipolar cells that break the stratification rules of the retina. J Neurosci 2009;29:8875–83.)

The melanopsin-containing ganglion cells project to the suprachiasmatic nucleus (SCN), the brain’s circadian pacemaker.^36,294 Labeled axons were also found in the olivary pretectal nucleus (OPN), which is involved with the circuit controlling pupil diameter. These are referred to as nonimage-forming pathways. In mice lacking melanopsin, photoentrainment of the circadian rhythm and the control of the pupillary light reflex were attenuated at high light intensities.²⁹⁵

However, melanopsin ganglion cells also receive the usual rod and cone inputs, which can drive these circuits at low light intensities. The M1 subtype is stratified in sublamina a but it still has ON-driven light responses due to the presence of ectopic or axonal ribbon synapses from ON cone bipolar cells as they traverse sublamina a (Fig. 15.33).^27,28,296 In crossbred mice with photoreceptor degeneration as well as no melanopsin, the pupillary light reflex and the photoentrainment of the circadian clock were completely eliminated.²⁹⁷ The conclusion is that, at low irradiance, the normal rod/cone inputs drive these circuits but at higher light levels the intrinsic melanopsin system drives these accessory visual functions. These experiments provide anatomical, physiological, and behavioral evidence that the ganglion cell type defined by the expression of melanopsin forms a specific channel in the mammalian visual system that is connected to the SCN and the OPN among other targets. This is a prime example that different ganglion cell types provide parallel channels in the visual system that serves specific functions.

Based on a combination of morphology and physiology, five subtypes of melanopsin-containing ganglion cells, M1–M5, have been identified in the mouse retina.^298,299 The M1 subtype is the only one stratified in sublamina a; it contains the most melanopsin and has strong intrinsic photoresponses and relatively weak rod/cone input. M2 cells have dendrites in sublamina b but less melanopsin and weaker intrinsic responses. M3 cells are bistratified and do not form a uniform population.³⁰⁰ M4 and M5 cells have very low levels of melanopsin and correspondingly weak intrinsic responses.²⁹⁸ Using combinations of markers, the M1-class ipRGCs can be further divided into two types, one projecting to the shell of the OPN, which is required for the pupillary light reflex, and one targeting the SCN, required for circadian photo entrainment.³⁰¹ This Brn-3b-negative group of M1 ipRGCs comprises approximately only 10% of all melanopsin ganglion cells. Thus, the specific nonimage-forming visual pathway that resets the circadian clock every morning consists of only 200 ganglion cells.³⁰¹ It is very surprising that a distinct visual pathway can be formed by such a small number of cells.

The roles of non-M1 types are still unclear but they may contribute to light avoidance in mouse pups, sleep regulation and, in humans, the light sensitivity of migraine and seasonal affective disorder.²⁹⁸ The activity of ipRGCs is also important for retinogeniculate segregation, which is now known to involve light sensitivity prior to function of the photoreceptors.²⁴⁷

Color vision and ganglion cells

Humans are trichromatic, with red, green, and blue cones. Color vision provides an additional variable to improve visual discrimination. It is organized into two opponent systems, red-green and blue-yellow (where yellow = red + green).³⁰² Hence, there are no red/green hues and no mixtures of blue and yellow; they are said to be color-opponent. The blue-yellow system is found widely in mammals and is thought to be older on an evolutionary timescale.⁵⁶ In primates, once the midget system had reached maximum resolution in the 1 : 1 : 1 pathway, a comparatively recent gene duplication produced red and green pigments. By reason of the connection with a single cone, midget ganglion cells are automatically color-coded.

Midget ganglion cells are concentrically organized as red-green or green-red. The center input is spectrally pure because of the 1 : 1 : 1 connection with single cones. There has been some controversy concerning the origin of the color-opponent surround, whether it is spectrally pure, and whether it arises from the outer or inner retina. Recent recordings from midget ganglion cells in the primate retina indicate that the surround is chromatically mixed. In other words, it is the average of the surrounding cones. On average, the ratio of red to green in the surround was approximately 50%.¹¹⁰ Furthermore, both center and surround inputs modulated an excitatory conductance with a reversal potential around 0 mV. This means that the surround signal does not arise from amacrine cell inhibition in the inner retina. Rather, the color opponency must be generated in the outer retina by horizontal cell feedback.¹¹⁰ Horizontal cells are suitably wired to provide the spectrally mixed surround signal because they contact both red and green cones indiscriminately.³⁰³ Enhanced buffering to block pH-dependent horizontal cell feedback eliminated both the spatial and chromatic surround of midget ganglion cells. Although the exact mechanism of horizontal cell feedback is controversial, this strongly indicates that color opponency is produced by horizontal cell feedback in the outer retina.^110,304 Finally, neither GABA nor glycine antagonists blocked the surround inputs. Thus inhibition in either the outer or inner retina is not required. Nor does horizontal cell feedback in the primate retina depend on GABA or glycine.¹¹⁰

Large-scale multielectrode array recording has also been used to analyze the color inputs to primate ganglion cells.³⁰⁵ Hundreds of ganglion cells may be recorded simultaneously and then sorted by spike size and shape to identify individual ganglion cells. Types of ganglion cell may be identified by their receptive field sizes, center/surround organization, and the mosaic properties by which they tile the retina. Using a fine-grained textual stimulus, local hot spots were found within each receptive field that matched the distribution of the underlying cone array. The spectral characteristics of each hot spot, recorded in several nearby ganglion cells, were used to identify each cone locus as red, green, or blue and then the contributions of each individual cone were mapped to the overlying array of ganglion cells. Parasol ganglion cells sampled uniformly from red and green cones but not blue cones. In contrast, many midget ganglion cells showed color-opponent responses. As above, the surrounds were spectrally mixed but the center inputs to both ON and OFF midget ganglion cells had a small nonrandom tendency to sample from more red or green cones.³⁰⁵Obviously, in central retina, where midget bipolar cells may contact a single cone, this would provide a spectrally pure center with a mixed color-opponent surround. Intellectually, we understand that tracing back from ganglion cells through the retinal circuitry must lead to cones. However, the direct demonstration of ganglion cell sampling from the complete array of identified cones is an experimental triumph. These experiments indicate that parasol ganglion cells are not color-coded while color signals are carried by the midget ganglion cells.³⁰⁵

In contrast to the red-green system, the circuits underlying blue-yellow opponency have been relatively well described. Some of this success is due to the morphological details which make cells in the blue pathway more recognizable. This begins with blue cones themselves, which can be labeled with antibodies against blue cone opsin.⁵⁹ Red and green cone opsins are so close they cannot be discriminated by current antibodies. In addition, blue cone pedicles are noticeably smaller, with fewer telodendria.⁷² The blue cone bipolar cells can also be recognized because of their long dendrites, which bypass many cones in search of the blue ones.^152,306,307 Blue ON cone bipolar cells have direct input to a small bistratified ganglion cell that gives blue ON/yellow OFF responses.³⁰⁸ Blue-yellow small bistratified ganglion cells have also been identified in multielectrode array recordings from primate retina. These experiments suggested that most of the color opponency arose in the outer retina.^309,310 Further physiological evidence suggests that blue-yellow color opponency originates in the outer retina.³¹¹ Indeed, blue cones themselves are blue-yellow color-opponent, indicating that the yellow-opponent signal is generated by horizontal cell feedback.³¹²

In nonprimate mammalian retina, blue-driven ganglion cells were reported as monostratified ON cells.³¹³ This is quite different from the small bistratified blue-yellow ganglion cell of the primate retina. The color opponency of this monostratified ganglion cell could arise in the outer retina, as suggested for primate retina.^309,312