Microcircuitry and Mosaic of a Blue–Yellow Ganglion Cell in the Primate Retina

Ganglion cells associated with S cone bipolar cells

Underlying the territory of these circuits were six or seven tiers of cell somas, including 195 ganglion cells, the dendritic arbors of which we traced into the inner plexiform layer (Fig. 1). Approximately 150 of these arbors belonged to midget ganglion cells, corresponding to one ON and one OFF midget cell for each M and L cone terminal (Calkins et al., 1994). Of the 45 additional ganglion cell arbors, 18 branched in both the ON and OFF strata and, in this sense, were bistratified. Six of these 18 (3% of all the ganglion cells) had a distinctive morphology.

The somas, 10–13 μm in diameter, were set in a middle tier of the ganglion cell layer (Fig.2A). Each produced a single dendritic stalk, ∼3 μm in diameter, that snaked between overlying somas to reach the inner plexiform layer. This primary stalk branched in the ON stratum at a depth of 65–100% and sent multiple dendrites into the OFF stratum at a depth of 15–50% (Fig.2A; see Fig. 5). These ganglion cells, spaced 10–15 μm between nearest neighbors (Fig. 2B), were more diffusely branching than their peripheral counterparts (Dacey, 1993;Dacey and Lee, 1994) and resembled, to a remarkable degree, the shrub ganglion cell drawn from Golgi impregnations by Polyak (1941) andBoycott and Dowling (1969).

The ganglion cell dendrites in the ON stratum associated intimately with the axon terminals of S cone bipolar cells (Fig.2B). We identified four of these terminals by tracing them back to somas and thence to their dendritic tips, which were contacted exclusively by S cones. It is well established that the cones contacting this bipolar cell type are indeed S cones (Mariani, 1984;Kouyama and Marshak, 1992; Wässle et al., 1994). Four additional bipolar cells could not be traced this way, because their dendrites left the series, but their axons displayed the same morphology, stratification, and association with the bistratified ganglion cells. Furthermore, all eight bipolar cell terminals shared distinctive synaptic patterns described below, so we infer that all belong to S cone bipolar cells.

Synapses in the ON stratum from S cone bipolar cells

The axon terminals of three complete S cone bipolar cells formed 39, 42, and 44 ribbon synapses. Two others that were nearly complete formed at least 34 and 36 ribbons (see Fig. 6A). These terminals associated intimately with dendrites of the bistratified ganglion cell and provided contacts at ribbon synapses (Fig. 3). The two complete ganglion cells received, respectively, 34 and 32 contacts from these bipolar cell terminals. Contacts were absent from the primary dendritic stalk but clustered on smaller dendritic twigs (Figs.4A,B, 5). These were the only bipolar cell contacts to these ganglion cells in the ON stratum. Thus, the morphology of this ganglion cell type (bistratified dendritic arbor) and its connections in the ON stratum (synapses from S cone bipolar cells) force the conclusion that it is the bistratified B–Y ganglion cell (Dacey and Lee, 1994).

Two or three S cone bipolar cells contributed synapses to each B–Y ganglion cell (Fig. 2B), but a single bipolar cell predominated. Thus one ganglion cell received 71% of its S cone bipolar cell contacts from one terminal and 29% from another (Figs.2A, 4A,left); a second ganglion cell received 65% of its S cone bipolar cell contacts from one terminal, 25% from a second, and 10% from a third (Figs. 2A,4A, right).

We tried to determine whether all S cone bipolar cell synapses are directed to the B–Y ganglion cell or if other ganglion cells and amacrine cells are also _targets. For this we chose an S cone bipolar cell terminal at the middle of the series and traced every postsynaptic process until it was identified or lost. This terminal contained 42 ribbons; at one ribbon there was a single postsynaptic process (monad), and at 41 ribbons there were two postsynaptic processes (dyad), so there were 83 postsynaptic processes total.

Forty-two of the 83 processes postsynaptic to this S cone bipolar cell were confirmed to be ganglion cell dendrites. Thirty were traced back to two B–Y cells, but the remaining 12 left the series. However, 10 of these 12 arose from a single ganglion cell, the dendritic tree of which in the ON stratum resembled that of the B–Y cell. Moreover, in the pool of eight S cone bipolar cells, whenever we could trace a dendrite back to an identified ganglion cell, it always belonged to a B–Y cell (Fig. 2B). Thus we conclude that in the ON stratum the S cone bipolar cell contacts only the B–Y cell.

Twenty of the 83 postsynaptic processes were confirmed by their vesicle content and presynaptic densities to be from amacrine cells. This left only 21 postsynaptic processes that could not be identified for certain because they were too thin to trace. However, 12 of these processes shared a dyad with a confirmed ganglion cell dendrite, and because we never observed a dyad shared by two different ganglion cells, these were probably from amacrine cells. The remaining nine processes were probably ganglion cell dendrites, because they contained neither vesicles nor presynaptic densities. In short, the S cone bipolar cell directs ∼60% of its output to the B–Y ganglion cell and ∼40% to amacrine cell processes.

Approximately half (12 of 20) of the confirmed amacrine cell processes postsynaptic to the S cone bipolar cell terminal formed a synapse back onto the terminal (reciprocal synapse). Only rarely (3 of 20) did a postsynaptic amacrine cell process contact the ganglion cell member of the dyad. This arrangement differs strikingly from the midget bipolar cell circuit where the postsynaptic amacrine cell invariably feeds back to the bipolar cell terminal and also forward to the ganglion cell (Dowling and Boycott, 1966; Calkins and Sterling, 1996). Nevertheless, we identified 21 amacrine cell contacts on a B–Y cell, and these were about evenly distributed between dendrites in both ON and OFF strata (Fig. 4C). Possibly these synapses arise from amacrine cells that show dye coupling to the B–Y cell (Dacey, 1993).

S cone bipolar cell synapses in OFF stratum

Tracing three S cone bipolar cell axons back toward their somas, we observed that they contained synaptic ribbons in the OFF stratum just below the inner nuclear layer (0–10% inner plexiform layer depth). Two of the axons contained three ribbons (Fig.6A), and a third axon contained seven ribbons (Fig. 6B). The postsynaptic complex was structured such that the same process was often postsynaptic at several ribbons (Fig. 6B). We traced about 10 postsynaptic processes until they left the series or were lost because they were exceptionally thin. These processes, which contained no vesicles or presynaptic densities, were probably ganglion cell dendrites. They cannot belong to the B–Y cell, because its dendrites do not reach this level of the OFF stratum (Figs. 2A,4A). Apparently the S cone bipolar cell provides a few synapses in the OFF stratum to a second type of ganglion cell.

Synapses from S cones to the S cone bipolar cell

The primary dendritic stalk of the S cone bipolar cell coursed through the inner nuclear layer and branched horizontally on reaching the outer plexiform layer (Fig.7A). On reaching an S cone terminal it branched profusely, sending fine twigs to invaginate the cone terminal (Fig. 7B). Each twig formed the central element of a triad. One S cone bipolar cell, branching beneath a single S cone terminal, formed the central triadic element at 10 of its 22 ribbons (Fig. 7B) (Kouyama and Marshak, 1992; Wässle et al., 1994). However, because three or four of the dendritic twigs of this cell under the S cone could not be traced, the actual number of central elements it contributes is probably ∼15. Two other fine dendrites snaked horizontally beneath surrounding M and L cone terminals, reaching for additional S cones beyond the series. The central elements of all S cone triads are occupied by S cone bipolar cell dendrites (Herr et al., 1996), and each S cone diverges to one to five of these bipolar cells, whereas each bipolar cell generally collects from one to three S cones (Kouyama and Marshak, 1992; Wässle et al., 1994). Thus, the S cone bipolar cell in the fovea collects at least 70% of its input from one S cone and the rest from one or two others.

This result, coupled with our observation that ∼70% of the synapses to the B–Y ganglion cell in the ON stratum are from a single S cone bipolar cell, indicates that the short-wavelength input to the ganglion cell from a single S cone is far greater than that from either of one or two others. However, this circuit clearly differs from that of the midget ganglion cell, in which there is no convergence at all from cone to ganglion cell (Calkins et al., 1994).

Synapses from diffuse bipolar cells in the OFF stratum

The dendritic arbor of the B–Y ganglion cell in the OFF stratum was centered over its arbor in the ON stratum, but the OFF arbor was narrower (Fig. 4B); this is similar to the dendritic tree of more peripheral cells (Dacey and Lee, 1994). Correspondingly, the OFF arbor received fewer than half as many ribbon synapses as the ON arbor; for the two cells in Figures 2A and 4, there were 15 and 13 ribbon synapses. The B–Y ganglion cell dendrites generally contributed only a single element to the dyad; the other element belonged to a different type of ganglion cell or an amacrine cell.

The OFF bipolar cell synapses to the B–Y ganglion cell were contributed by two different types of bipolar cell, as defined by Cohen and Sterling (1990, 1991). One type had a soma low in the inner nuclear layer (Fig. 8). Its thin axon, 0.5–0.7 μm diameter, led to a terminal that stratified narrowly at 35–50% inner plexiform layer depth and contained 63 synaptic ribbons. The second type had a soma at midlevel in the inner nuclear layer. Its thick axon, 1.0–1.5 μm in diameter, led to a terminal that arborized diffusely at 10–50% inner plexiform layer depth and contained 53 ribbons. These two bipolar cell types correspond by morphology and axonal stratification to DB3 and DB2, named by Boycott and Wässle (1991).

Two DB2 cells provided four synapses, and a single DB3 cell provided 11 synapses to one B–Y ganglion cell (Figs. 2A,4A,left). DB2 cells provided at least five synapses to another B–Y ganglion cell (Figs. 2B,4A, right). The remaining eight synapses might be DB2 or DB3, but they could not be identified because their terminals left the series. Thus, based on two complete B–Y ganglion cells, ∼30% of the OFF bipolar synapses arise from two or three DB2 cells, and ∼70% arise from one or two DB3 cells.

Cone input to diffuse OFF bipolar cells

In the fovea, DB2 dendrites receive input from M and L cones at both semi-invaginating and basal contacts, whereas DB3 dendrites receive input at only basal contacts (Calkins et al., 1995). This difference permitted us to identify neighboring DB2 and DB3 cells, although their axon terminals left the series, and thus to determine the number of cones that converge on them. Two neighboring DB2 cells and one DB3 cell collected, respectively, from 12, 10, and 10 cone terminals (Fig. 9). The three bipolar cell dendritic trees overlapped considerably, so together they collected from 18 cone terminals. Because in aggregate three or four DB2 and DB3 cells contact a B–Y ganglion cell, we estimate that the total input from OFF bipolar cells arises from ∼20 M and L cones.

Anatomical wiring explains receptive field of the B–Y cell

When the B–Y receptive field was assigned by intracellular recording and tracer injection to a bistratified ganglion cell, there were strong predictions about the underlying circuitry (Dacey and Lee, 1994). Excitation at onset of a blue light was attributed to synapses from the ON S cone bipolar cell, and excitation at offset of a yellow light was attributed to synapses from OFF M + L cone bipolar cells. The synaptic circuitry described here strongly confirms these predictions (Fig.10). The bistratified ganglion cell in the fovea (perhaps Polyak’s shrub cell) does indeed receive many synapses (∼33) from the S cone bipolar cell and fewer synapses (∼14) from diffuse OFF bipolar cells. Because all bipolar cell synapses apparently release glutamate (Massey, 1990), and all ganglion cells probably express ionotropic glutamate receptors (Cohen and Miller, 1994; Zhou et al., 1994; Peng et al., 1995; Qin and Pourcho, 1995; Lukasiewicz et al., 1997), both the S cone bipolar cell and M + L cone bipolar cells probably excite the B–Y ganglion cell.

It might seem surprising that the S and M + L cone receptive fields of this ganglion cell are spatially coextensive (Wiesel and Hubel, 1966;Zrenner, 1983a,b; Dacey, 1996), in view of their great difference in cone convergence. Whereas the ON dendritic arbor collects mainly from one S cone with smaller contributions from one or two others (Fig. 7) (Kouyama and Marshak, 1992; Wässle et al., 1994), the OFF dendritic arbor collects from a total of ∼20 M and L cones (Fig. 9). However, there may be no real discrepancy. Because chromatic aberration strongly blurs the retinal image at short wavelengths (Williams et al., 1993; Marimont and Wandell, 1994), the S cone receptive field is effectively broadened, corresponding approximately to the dendritic spread of the B–Y cell in the ON stratum and thereby matching that of 20 M and L cones (Fig. 10B).

Site and molecular basis for blue–yellow hue opponency

The antagonism between S cone and M + L cone signals manifested by the B–Y ganglion cell most likely arises at the cone→bipolar cell synapses. The molecular mechanism seems assignable to the expression of different glutamate receptors by two classes of bipolar cell. The invaginating dendrites of the S cone bipolar cell express metabotropic glutamate receptor 6. We know this because the invaginating dendrites at all cone terminals in macaque retina express this receptor (Vardi et al., 1998). The noninvaginating dendrites of the diffuse bipolar cells probably express ionotropic glutamate receptors (Calkins et al., 1996; Euler et al., 1996;Hartveit, 1997; Vardi et al., 1998). On suppression of glutamate release by light, the metabotropic receptor opens cation channels, whereas the ionotropic receptor closes cation channels.

Consequently, when glutamate concentration decreases at both receptor types, their currents are opposed. However, the S cone bipolar cell should have about twice the effect at the B–Y ganglion cell as the M + L cone bipolar cells, because it provides approximately twice as many synapses. Allowing for this in subtracting (M + L) from S signals predicts a neutral point near 500 nm for the B–Y ganglion cell. That is, the steady firing rate of the cell should be unaffected by modulating the intensity of a 500 nm spectral light, and that is what the recordings show (Fig. 10C) (Zrenner, 1983a,b). This 500 nm light we see as neither blue nor yellow; i.e., it is perceptually in blue–yellow equilibrium (Larimer et al., 1975). So, the key neural mechanism that causes perceptual opponency may reduce simply to signals of opposite sign generated at the first synapse by different classes of glutamate receptor.

Spectral antagonism represents efficient packaging of color information

Why should the spectral antagonism that underlies hue opponency be established at the first visual synapse? Perhaps because the great overlap in spectral sensitivity between different cone types causes the spectral component of the postreceptoral retinal image to vary little in space or time (Buchsbaum and Gottschalk, 1983). This redundancy would tend to degrade transmission of the spectral component, because neurons are both noisy and of limited dynamic range (Laughlin, 1989;Tsukamoto et al., 1990; Atick, 1992). An efficient coding scheme would reduce spectral redundancy by using the difference between the S and M + L cone signals to fill the dynamic range of the ganglion cell (Laughlin, 1989).

Indeed, this is exactly how the excitatory inputs are arranged. Seventy percent of the excitatory synapses to a B–Y cell code the presence of one spectral band, whereas 30% code the absence of a different spectral band (Fig. 10B). Consequently the dynamic range of the the B–Y cell is best filled when the local ratio of S/(M + L) is at least 2:1. Perhaps this ratio is optimized for commonly encountered ratios of these wavelengths in natural scenes (Laughlin, 1981; Webster and Mollon, 1997). Put another way, the quantum catch of the S cone in broad-band light is relatively low; therefore, boosting its gain places the neutral point of the B–Y cell nearer to middle wavelengths, where it would be most useful (Fig. 10) (Buchsbaum and Gottschalk, 1983). Thus the reason we perceive a blue–yellow equilibrium for a spectral light near 500 nm under neutral adaptation (Hurvich and Jameson, 1957; Larimer et al., 1975) may simply be that it is the most efficient way to code the natural scenes inhabited by our ancestors.

The B–Y ganglion cell mosaic can support blue–yellow color discrimination

One reason to doubt that the B–Y cell constitutes the first stage of a pathway for hue perception has been that ganglion cells with spatially coextensive ON and OFF receptive field regions are rarely encountered by microelectrodes (Lee, 1996). Indeed, we found the B–Y cell to form only 3% of all foveal ganglion cells, but it does not follow that this small fraction cannot support hue discrimination. Even when chromatic aberrations are corrected, humans can discriminate blue–yellow gratings no finer than 7–14 cycles/degree (Stromeyer et al., 1978; Williams et al., 1983; Mullen, 1985; Sekiguchi et al., 1993). This limit corresponds to the Nyquist limit for the densest region of the S cone mosaic, which occurs at 1° eccentricity (Williams et al., 1981; Curcio et al., 1991). We find in this region of macaque retina that the spatial density of the B–Y cell equals that of the S cone. So, if the human circuit resembles that of the macaque, the B–Y ganglion cell mosaic would support discrimination of the finest blue–yellow grating.

Conceivably, additional ganglion cell types could contribute to the perception of blue and yellow. For example, there are reports of ganglion cells with an S-OFF/(M + L)-ON receptive field (Lee, 1996), and anatomy shows that every S cone connects, via a midget bipolar cell, to a midget OFF ganglion cell (Klug et al., 1992, 1993). Thus, the spatial density of this type must equal that of the B–Y cell described here. The S-OFF center of this midget ganglion cell would be opposed by the H2 horizontal cell that connects to S, M, and L cones (Dacey et al., 1996). Whether this cell with mixed antagonism, i.e., S − (S + M + L), contributes to hue perception or simply to S cone-mediated spatial perception remains to be determined.