Physiology of Horizontal Cells

[History] [Archetypical circuitry] [Passive electrical models] [Rod and cone contributions] [Spatial characteristics] [Gap junctions] [Functional roles] [Color opponency] [References]

1. History.

Fig. 1. S-potentials from fish retina (13 K jpeg image)

S-potentials were the first light-evoked electrical responses recorded intracellularly from nerve cells in vertebrate retinas (Fig. 1). Prior to the discovery of S-potentials retinal physiologists explored retinal function through light-evoked, extracellular field potentials such as the electroretinogram (ERG), which reflected the massed action of many retinal cells, or through the minute extracellular voltages evoked by the passage of individual action potentials along nearby optic nerve fibers. S-potentials may have been named in honor of their discoverer (Svaetichin, 1953). At first the location of the tiny, sub-micron tips of the hollow, electrolyte-filled, glass microelctrodes used to record these signals from inside neurons, was not accurately known, and the electrical responses were thought to arise from cones. Later intracellular marking techniques, in which dyes were ejected from electrode tips into the cytoplasm of the recorded neuron, revealed horizontal cells as the source (MacNichol and Svaetichin, 1958; Werblin and Dowling,1969; Kaneko, 1970). 'S-potential' has also come to mean 'slow-potential'. The light-evoked electrical responses of horizontal cells, and in fact many retinal neurons, are, characteristically, 'slow', requiring 10's to 100's of msec to fully develop. During the horizontal cell response to light, the trans-membrane potential increases or 'hyperpolarizes', from ~-30 mV in darkness to -40 to -70 mV in the presence of the light stimulus. This change is maintained for seconds- to minutes-long durations. Sometimes there are transient 'overshoots' of the final maintained levels both at stimulus onset and offset.

S-potential responses evoked puzzlement amongst neurophysiologists of the late 1950's. At that time neurons were pictured as being 'depolarized' by excitatory synaptic inputs, having their inside-negative resting membrane potentials reduced. This in turn induced action potentials, or nerve spikes, to transfer signals across the fetch of the long neuronal axon. S-potential responses, however, had neither light-induced depolarizations or nerve impulses.

Since first described in fish retinas, S-potentials have been recorded from retinal horizontal cells in all vertebrate classes, including mammals (Naka and Rushton, 1966; Byzov et al., 1968; Norton et al., 1968; Werblin and Dowling, 1969; Steinberg, 1969a, b; Baylor et al., 1971; Fuortes and Simon, 1974; Nelson et al., 1974; Dacheux and Raviola, 1982; Dacheaux and Raviola, 1990; Dacey et al., 1996).

Fig. 2. S-potentials from monkey retina (19 K jpeg image)

CLICK HERE to see a movie of the intracellular recordings from an horizontal cell in the vertebrate retina(117 K quicktime movie)

2. Archetypical circuitry patterns in rod-dominant mammals.

Two horizontal cell types are found in most mammalian retinas. One is an axonless type (often called A-type). This type contacts solely cones. The other (often called B-type) has a cell body and dendrites contacting cones, and an axon of several hundred microns length that arborizes into a terminal structure that contacts only rods (Cajal, 1892; Gallego, 1986; Kolb, 1970; Boycott et al., 1978; Kolb and Nelson, 1985). Horizontal cell dendritic tips always terminate at presynaptic ribbons in photoreceptor terminals at what are known as lateral elements (Fig. 3, le arrow).

Fig. 3. Dendrite of an A-type HC contacts a cone pedicle (57 K jpeg image)

3. Passive electrical models.

Electrical models of B-type horizontal cells suggest that horizontal cell axons are too long and thin to allow passive 'electrotonic' conduction of signals from one end of the cell to the other (Nelson et al, 1975). Such models are based on the anatomical dimensions and geometry of the neuron, and ohmic, linear properties, of cell membrane and cytoplasm. The latter appear appropriate for horizontal cells, which lack evidence of 'active' non-linear membrane properties such as nerve impulses. The inference from such models is that signals in cell bodies of either the A or B type horizontal cells are expected to reflect only the local synaptic inputs from cones while signals in axon terminals of the B-type horizontal cells are expected to reflect only the local synaptic inputs from rods.

4. Rod and cone contributions.

In agreement with electrotonic models, intracellular recordings of S-potentials from horizontal cell axon terminals in cat revealed mainly rod signals. Intriguingly, intracellular recordings of S-potentials from cone-connected A- and B-type horizontal cell bodies in cat retina revealed not only cone signals, but rod signals as well (Steinberg, 1969a, b; Neimeyer and Gouras, 1973, Nelson et al., 1975; Nelson 1977).

Intracellular responses of the three horizontal cell elements in cat reveal a mixture of rod and cone signal wave forms (Fig. 4). Cone signals are identified by quick repolarizations at stimulus offset, while rod signals linger, recovering more slowly. This latter effect, observed at high stimulus irradiances, was studied by Steinberg (1969) who identified it with rod action and called it the 'rod after effect'. The difference in offset kinetics for rod and cone signal-components provides a convenient assay for rod and cone signal composition of S-potential wave forms.

Fig. 4. Cat HCs and their S-potential responses (41 K jpeg image)

Responses to three different stimulus irradiances are superimposed in Fig. 4. The brighter evoke more hyperpolarized S-potential responses with longer-duration rod after-effects. In A- and B-type cell bodies rod and cone signals are about equal in amplitude, whereas the axon terminal of the B-type cell responds mainly with the slowly recovering wave form characteristic of rod signals.

Fig. 5. Rod and cone components of cat HCs (21 K jpeg image)

Further evidence of the mixing of rod and cone signal components in horizontal cells is seen during adaptation by background lights. When retinas are adapted by steady background lights, rod signals, which are highly sensitive to allow detection of very dimly illuminated objects, are overwhelmed by the background and any further responses are lost. But cone signals, which are faster and less sensitive, adapt to ambient lighting conditions and continue to respond. These properties of rod and cone signals are easily seen in responses from mammalian horizontal cells. S-potentials recorded in horizontal cell axon terminals are virtually abolished by light adaptation. This is consistent with a rod-dominated physiology. Although light adaptation eliminates the rod components in S-potentials recorded from horizontal cell bodies, large cone signals remain, as these cellular regions continue to respond well in the presence of background lights (Fig. 5).

Rod signals seen in the cone-connected horizontal cell body regions arise not from the synaptic circuitry of horizontal cells, but from electrical synapses or gap junctions that link rod and cone photoreceptors in the outer plexiform layer (see chapter on photoreceptors). These electrical synapses introduce rod signals into cones.

5. Spatial characteristics.

Retinal areas over which visual neurons respond to photic stimuli are called 'receptive fields'. Receptive fields can be simple with the cell giving the same response over a large circular area of stimulation, or consist of centers and surrounds and sub-regions mediating different sorts of responses. There is only a single, simple, receptive-field mechanism in horizontal cells. As small, focal stimuli are displaced progressively farther from the optimal position, or 'center', response amplitudes smoothly decrease. Horizontal cell receptive-field properties are modeled with 'electrical syncytium' or 'slab' models. In these models receptive field width is characterized by a single radius-like parameter known as the 'space constant'. Horizontal cells are affected by light stimuli over a much wider area than one might suppose based on the lateral extent of process arborization, or 'dendritic field'. The size of space constants fit to horizontal cell receptive-field data imply that photic currents generated in one cell propagate laterally amongst many horizontal cells, before gradually dissipating through leakage out of neural membranes. Lateral, electrical propagation of photic signals is a feature of receptive fields of many neural types in retina, including, surprisingly, cones, bipolar and amacrine cells.

Fig. 6. Horizontal-cell receptive fields for rod and cone signals.(36 K jpeg image)

In the above figure, the space constant for the horizontal cell cone field, as measured with red, cone-selective stimuli, is 250 Ám, while that for the rod field, as measured with blue, rod-selective stimuli, is 200 Ám. Both signals arise within overlapping, concentric spatial regions. The result is significant in that any rod signals arizing from horizontal cell axon terminals might be expected to be displaced by the length of the horizontal cell axon. No such displacement of the centers of rod and cone signals is observed. Local coupling of rods and cones is consistent with such overlapping of fields. Space constants of cat horizontal cell receptive fields range from 200 to 450 Ám (Nelson, 1977).

There is no simple relationship between receptive field and dendritic extent. It is apparent that receptive-field size depends on the amplitude criterion chosen (Fig.6). Nonetheless it is also evident that vigorous responses can be obtained to stimuli well outside a horizontal cell's dendritic field. The largest of horizontal cell dendritic fields corresponds to only one tic mark on the position axis of Fig. 6.

Linear integration of signals from different receptive field regions is also a hallmark of horizontal cell physiology. This is consistent with the 'electrical syncytium' or 'electrical slab' models which successfully characterize the wide spatial extent. Photic currents arising from various receptive-field regions summate linearly, giving rise to larger, summed responses.

Fig. 7. Center and surround responses for a cone and an horizontal cell (37 K jpeg image)

Above (Fig. 7) a central slit stimulus (~100 Ám) covers the dendritic field of the recorded cell and produces a low-amplitude S-potential response. A further 'surround' stimulus is added to this slit stimulation. This is composed of 2 hemi-disks flanking the slit. It evokes added hyperpolarization, arising solely from surrounding, electrically coupled cells. Signals from these cells (arrow) are linearly added to the signal arising from the central cell as more horizontal cells are recruited into forming the response. For comparison (below in, Fig.7), a cone response is not much influenced by addition of the large surround stimulus. For cones the slit stimulus is already wide enough to encompass most neighboring electrically coupled cones.

6. Gap junctions.

Horizontal cells, in all vertebrate retinas including mammalian, are characterized by large-surface-area gap junctions (Fig. 8) between dendrites of like-type neighboring cells (Kolb, 1977). These junctions allow lateral flow of electrical signals within a syncitial network of cells. Receptive field sizes are increased, as is the area over which horizontal cells integrate photic information.

Fig. 8 Gap junctions between A-type horizontal cell dendrites of cat retina (54 K jpeg image)

In addition to electrical currents, small molecules also pass through horizontal cell gap junctions. Lucifer dye or Neurobiotin, microelectrode stains used to mark and identify recorded cells, are such molecule. In Fig. 9 Lucifer dye injected into a rabbit A-type horizontal cell has spread to several neighboring cells, forming striking images of the interconnected horizontal cell network (Mills and Massey, 1994).

Fig. 9 Rabbit horizontal cell network revealed by dye injections. (31 K jpeg image)

7. Functional roles.


Horizontal cells send visual information back to cones through feedback synapses. This interaction takes place through processes invaginating photoreceptor synaptic endings (reviewed in Kolb and Nelson, 1995). Light responses of cones are antagonized by signals of opposite polarity arising from horizontal cell feedback synapses. Light stimulation is believed to reduce the release of neurotransmitter gamma amino-butyric acid (GABA) from horizontal cells onto cones, although GABA antagonists typically do not block feedback. In human retinas conventional synaptic contacts from horizontal cells onto rods and cones have been observed (Linberg and Fisher, 1988) though in other species such structures are lacking.

Color opponency in cones and C-type (chromatic) horizontal cells appears to be mediated by feedback, as discussed in another section of this chapter. A further consequence of feedback is 'spatial opponency'. In the spatial domain, horizontal cells respond over a much wider retinal area than do cones. Through the horizontal cell pathway, wide field stimuli exercise an indirect, delayed, depolarizing influence on cones, opposing the direct hyperpolarizing influence of light stimuli on the cone phototransductive machinery (Fig. 10, below). Feedback appears to be mediated by chloride ions, and is enhanced by high chloride concentrations in cones (Lasansky, 1981). Feedback onto rods has not been documented.

Fig. 10. Large-spots depolarize cones through horizontal-cell feedback. (34 K jpeg image)

Spatial opponency in cones is passed forward to bipolar cells and ganglion cells, contributing to 'concentric' or 'center-surround' organization of receptive fields. In experiments where horizontal cell membrane potentials are directly polarized by injections of current through the microelectrode, ganglion cells discharges characteristic of 'surround responses' or 'large-spot responses' are evoked (Naka, 1971; Mangel, 1991). This current-evoked response appears to be caused primarily by feedback. The glutamate agonist 2-amino-4- phosphonobutyrate (APB), selective for transmission between cones and ON-center bipolar cells, blocks not only the light responses of ON-center ganglion cells, but the current-evoked responses as well (Mangel, 1991). Such experiments indicate the current-evoked responses utilize the feedback pathway, and that this pathway is important in shaping the spatial organization of retinal neurons beyond the cones themselves.


Horizontal cells impinge synaptically on bipolar cells, feeding visual information forward to these second-order retinal neurons. Conventional synaptic structures occur between horizontal cells and bipolar cells in the retinas of several mammalian and non-mammalian species. As with cone feedback, the feedforward neurotransmitter appears to be GABA. OFF-center bipolar cells, like cones, are hyperpolarized by light stimuli. The influence of feedforward synapses on these cells is likely to be similar to that of feedback synapses on cones. Wide field visual stimuli induce delayed depolarizing signals in OFF-center bipolar cells. Such effects probably combine the effects of 'feedback' on cones and 'feedforward' onto bipolar cells to produce the characteristic 'center-surround' spatial organization of OFF-center bipolar cells.

ON-center bipolar cells are depolarized by light stimuli. Here the depolarizing effects of wide-field, 'feedforward' synapses from horizontal cells would not mediate spatial antagonism since they are of the same polarity as the direct input from cones. Logically feedforward contacts appear to facilitate ON-center responses. In ON-center bipolar cells, feedforward and feedback synapses tug in opposite directions. Nonetheless 'center-surround' spatial organization is also a characteristic of ON-center bipolar cells.

Fig. 11. Center and surround receptive fields of turtle bipolar cells (21 K jpeg image)

Examples of area effects in ON-center and OFF-center bipolar cells in turtle retina are illustrated above. These two basic types of bipolar spatial organizations can be recorded in a wide range of vertebrate retinas and the 'spatial opponency' of these cells is generally attributed to horizontal cell action though feedback and feedforward synapses (Werblin and Dowling, 1969; Kaneko, 1970; Naka, 1971, 1976; Lasansky, 1978; Marchiafava and Weiler, 1980; Saito et al., 1979, 1981; McReynolds and Lukasiewicz, 1989; Ammermuller and Kolb, 1985, 1986; Hare and Owen, 1990). Nonetheless inner retinal inputs may need to be considered also (Kolb and Nelson, 1993; Marc, 1989), particularly for ON-center signals originating from rods, as seen in ON-center mammalian rod bipolar cells (Dacheux and Raviola, 1986). In this case feedback onto rods is probably weak or absent, and feedforward synapses provide signals of inappropriate polarity.

Not all bipolar cells exhibit spatial antagonism. Some OFF-center bipolar cells in cat appear to lack surrounds (Nelson and Kolb, 1983). Like horizontal cells, bipolar cells, particularly of the OFF variety, (Hare and Owen, 1990) appear to be interconnected by electrically conductive 'gap' junctions. When sufficiently strong, such junctions may cause bipolar cell receptive fields to resemble those of horizontal cells, where the predominant signal is the hyperpolarizing 'center' mechanism, spread throughout a large, uniform receptive field area.


The gain and temporal characteristics of cone-to-horizontal-cell and cone-to-bipolar cell synapses are modulated by several influences, among them horizontal cell activity. When horizontal cells are hyperpolarized by wide-field photic stimuli, cone-signals evoked in second order neurons both increase in amplitude and become quicker in time course. Such effects are commonly observed when small spots are flashed in the presence or absence of large background fields (Chappell et al, 1985). Horizontal cell modulation effects may also explain suppressive rod-cone interaction (SRCI, Goldberg et al, 1983). In SRCI large, dim green stimuli, seen only by rods, cause observers to perceive focal, red stimuli, seen only by cones, as being brighter. Analogous physiological effects involving rod and cone signals occur in horizontal cells and bipolar cells themselves (Frumkes and Eysteinsson, 1987; Pflug et al, 1990; Nelson et al; 1990). In Fig. 12, flickering cone signals are evoked in a cat horizontal cell by flickering red stimuli of various sizes. When the stimuli are small, flicker amplitudes are increased by superimposing a steady blue background, which selectively excites rods. The amplitude of the cone flicker signals is 'modulated' through wide-field horizontal cell stimulation, probably through increased gain at horizontal-cell-to cone synapses. Area effects for SRCI suggest that a wide-field cell such as the horizontal cell mediates the modulation of gain.

Fig. 12. Supressive rod-cone interaction in HCs (49 K jpeg image)

Calcium entry into synaptic structures is thought to induce the release of neurotransmitter by promoting fusion of neurotransmitter-laden synaptic vesicles with presynaptic membranes. Calcium entry occurs thorough specialized membrane channels regulated both by voltage and by neurotransmitters. In a simple model of horizontal cell modulation of gain at cone synapses, depolarized, dark-adapted horizontal cells release a neurotransmitter (likely GABA) which down-regulates calcium channels, and thereby calcium entry to the cone synapse. This in turn has the effect of reducing the voltage-modulated calcium signal in the cone synapse, and the gain of the synapse. Light adaptation, which hyperpolarizes horizontal cells, reduces the release of the down-regulating neurotransmitter from horizontal cells, and increases cone synaptic gain.

Fig. 13. Model of feedback to cone pedicles (33 K jpeg image)

In the model (above), this modulation effect also serves to complete a negative feedback loop between horizontal cells and cones, which regulates the horizontal cell membrane potential. When horizontal cells become depolarized, calcium influx into the cone synapse is reduced, as is transmitter efflux from the synapse. The effect sends horizontal cell membrane potential in a hyperpolarized direction. The horizontal cell membrane potential becomes controlled by the network, rather than by factors intrinsic to the cell. Interestingly bipolar cells perceive this negative feedback effect and synaptic gain modulation as a further spatial opponency effect of horizontal cells. This is true for both ON- and OFF-center bipolar cells. The increase in transmitter release caused by horizontal cell hyperpolarization is just the reverse of the decrease in transmitter release which occurs as cones are hyperpolarized by small spots of light. Thus 'modulation' effects provide yet a third way in which horizontal cells may serve to mediate spatial opponency, or 'center-surround' effects.

[History] [Archetypical circuitry] [Passive electrical models] [Rod and cone contributions] [Spatial characteristics] [Gap junctions] [Functional roles] [Color opponency] [References]

Updated: January, 2001