Segregation and Integration of Visual Channels: Layer-by-Layer Computation of ON–OFF Signals by Amacrine Cell Dendrites

Morphological diversity of amacrine cells

Figure 1A shows the stacked confocal fluorescent image of an AC (Fig. 1A, a), a sketch of the same cell on a schematic background of a retina with the IPL divided into 10 strata (Fig. 1A, b), and the light-evoked current responses to a 0.5 sec light step at various holding potentials (Fig. 1A, c). Typical AC features include a soma (cell body) centered in the proximal half of the IPL, and dendritic processes ramifying laterally in the IPL. The AC in Figure 1A is an example of what we termed a “broadly monostratified” cell (cells with dendrites ramifying in two to three contiguous strata, in this case in strata 4–6). By using this protocol, we studied a total of 164 ACs. Forty-nine (30%) were “narrowly monostratified” (with dendrites ramifying in one of the ten strata of the IPL), 32 (19%) were broadly monostratified (dendrites ramifying in two to four contiguous strata), 24 (15%) were “narrowly bistratified” (dendrites ramifying in two noncontiguous strata), 8 (5%) were “broadly bistratified” (dendrites ramified in two bands and at least one band ramifying in two or more contiguous strata), 4 (2%) were “narrowly tristratified” (dendrites ramified in three noncontiguous strata), 11 (7%) were “broadly tristratified” (dendrites ramified in three bands and at least one band ramifying in two or more contiguous strata), and 36 (22%) were “diffuse” (dendrites ramified in four or more bands or in five or more contiguous strata). The distribution of AC dendrite stratification levels in the IPL is given in Figure 1B. Each vertical column represents an AC and the short vertical bars in individual strata indicate the presence of dendritic processes in that stratum. Cells were arranged in the order of number of their dendritic strata (narrowly monostratified cells first, within which cells were arranged in the order of their dendritic strata 1–10). In this sample of 164 ACs, approximately one-half of the ACs are monostratified, three-fifths of which are narrowly ramified in a single stratum. The second largest group is the diffuse ACs (22%), among which the majority have dendrites distributed contiguously in strata 2–9. There are no clear patterns of dendritic distribution within bistratified and tristratified cells in our data pool.

In addition to the difference in levels of dendritic stratification in the IPL, ACs vary with respect to the width of their dendritic field (W_D). In our sample of 164 ACs, there are 39 (24%) narrow-field cells [N cells;W_D < 125 μm (MacNeil and Masland, 1998)], 77 (47%) medium-field cells (M cells; 125 μm <W_D <400 μm), and 48 (29%) wide-field cells (W cells; W_D > 400 μm)s. The dendritic field width of each of the 164 cells is labeled in Figure 1B (bottom row). There is no clear correlation between the dendritic field width and the levels of dendritic stratification in the IPL. One exception is that we did not find any wide-field diffuse ACs: all wide-field cells were monostratified, bistratified, or tristratified. It is important to note, however, that our voltage-clamp recordings were made from the AC cell bodies, and therefore we might not have space-clamped all dendrites of the ACs, especially those of the wide-field cells. Thus the light-evoked responses of the wide-field cells may be dominated by the synaptic inputs to the dendrites close to the cell bodies. Another potential source of inaccuracy in our measurements is that we recorded ACs from retinal slices. Although we chose to record from ACs several layers under the slice surface (which had more intact dendrites as assessed by image rotation with the confocal microscope; see Materials and Methods), some dendrites may still be cut off during slicing. Therefore, we may have underestimated the AC dendritic field size, especially the W_D of the wide-field cells.

Light-evoked current responses of morphologically identified amacrine cells

Figure 2 shows stacked confocal fluorescent images (Fig. 2A,C) and light-evoked current responses at various holding potentials (Fig.2B,D) of eight ACs. Cells a–dare narrowly monostratified at strata 1, 3, 7, and 9, respectively; cell e is broadly tristratified at strata 3, 6, 7, and 9; cell f is narrowly bistratified at strata 3 and 8; cellg is broadly monostratified at strata 9 and 10; and cellh is diffuse across strata 1–9. Cells b,e, and f are narrow-field cells; cellsa, c, and h are medium-field cells; and cells d and g are wide-field cells. To a 0.5 sec light step measured at −60 mV, the excitatory light-evoked cation current response (ΔI_C) of cella was a transient inward current at light offset (OFF response), the ΔI_Cs of cellsb–d, f, and g were transient inward currents at light onset (ON response), and the ΔI_Cs of cells e andh were transient inward currents at both light onset and offset (ON–OFF response). The inhibitory light-evoked chloride current responses (ΔI_Cls) measured at 0 mV of cells a and f–h were ON–OFF transient outward currents, and those in cells b–e were ON transient outward currents. The current responses measured at +40 mV and −100 mV in each cell were mediated by the light-evoked cation and chloride conductance changes (Δg_C and Δg_Cl, respectively).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Stacked confocal fluorescent images (A,C) and current responses evoked by a 0.5 sec light step (500 nm, −3.3) at various holding potentials (40, 0, −60, and −100 mV) (B, D) of eight (a–h) ACs. The ten strata of the IPL in each image are marked by scales on the right. Scale bars, 25 μm. The excitatory light-evoked cation current response (ΔI_C) of each cell was measured atE_Cl (−60 mV), and the inhibitory light-evoked chloride current response (ΔI_Cl) was measured atE_C (0 mV).

Layer-by-layer analysis of excitatory and inhibitory light-evoked ON and OFF signals in narrowly monostratified amacrine cells

To determine the stratum-by-stratum excitatory and inhibitory synaptic inputs, we analyzed the ΔI_Cand ΔI_Cl of narrowly monostratified ACs, because these cells make synaptic contacts with bipolar cells and other ACs in one IPL stratum. Figure3 shows the ΔI_C and ΔI_Cl (Fig. 3, bottom panel) of 10 (marked as cells 1–10, corresponding to the stratum each cell ramifies) narrowly monostratified cells whose dendrites ramified in each of the 10 strata of the IPL. The top panel of Figure 3 shows sketches (based on stacked confocal fluorescent images; same procedure as shown in Fig.1A) of the AC bodies and a portion of the narrowly monostratified dendrites near the soma (extended portions of dendrites of medium- and wide-field cells were cut off to save space) on a schematic background of the retina. The dendritic thickness (in the dimension parallel with the photoreceptor axes) of the majority (75–80%) of the narrowly monostratified cells is approximately one-tenth the thickness of the IPL, whereas ∼10% are slightly wider than one-tenth of the IPL and the rest are somewhat thinner than one-tenth of the IPL. At −60 mV, cells 1, 2, and 4 exhibited an OFF ΔI_C; cells 3 and 7–10 gave rise to an ON ΔI_C; and cells 5 and 6 gave an ON–OFF ΔI_C. This pattern of excitatory light responses was consistent in all narrowly monostratified ACs within a given stratum, regardless of their dendritic field size (for example, cells a–d in Fig. 2). The average ± SD peak amplitudes of ΔI_C in various types of narrowly monostratified ACs are given in Table 1, and they suggest that the excitatory inputs to strata 1, 2, and 4 are predominantly mediated by hyperpolarizing bipolar cells (HBCs); those to strata 3 and 7–10 are predominantly mediated by the depolarizing bipolar cells (DBCs); and those to strata 5 and 6 are mediated by both HBCs and DBCs.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Top, Sketches based on stacked confocal fluorescent images (same procedure as shown in Fig.1A) of 10 (cells 1–10, corresponding to the stratum each cell ramifies) narrowly monostratified ACs. Extended portions of dendrites of medium- and wide-field cells were cut off to save space. PRL, Photoreceptor layer;OPL, outer plexiform layer; INL, inner nuclear layer; GCL, ganglion cell layer.Bottom, Current responses evoked by a 0.5 sec light step (500 nm, −3.3) at holding potentials of −60 mV (ΔI_C) and 0 mV (ΔI_Cl).

At 0 mV, cells 1, 2, and 4–6 showed ON–OFF ΔI_Cls and cells 3 and 7–10 exhibited ON ΔI_Cls, indicating that ACs making inhibitory synaptic contacts with cells 1, 2, and 4–6 (those with OFF ΔI_Cs) gave rise to ON–OFF light responses, and ACs making inhibitory synaptic contacts with cells 3 and 7–10 (those with only ON ΔI_Cs) gave ON light responses. This pattern of inhibitory light responses is also consistent in all narrowly monostratified ACs within a given stratum, regardless of their dendritic field size (for example, cells a–d in Fig.2). The average ± SD peak amplitudes of ΔI_Cl of various types of narrowly monostratified ACs are also listed in Table 1, and they suggest that the inhibitory inputs to strata 1, 2, and 4 (strata with OFF ΔI_Cs) and strata 5 and 6 (strata with ON–OFF ΔI_Cs) are mediated predominantly by ON–OFF ACs, and that those to strata 3 and 7–10 (strata with ON ΔI_Cs) are mediated predominantly by ON ACs.

In addition to differences in ON and OFF inputs, the kinetics of ΔI_C and ΔI_Cl in various types of narrowly monostratified ACs was different, and the average ± SD rise time (τ₀) and decay time constant (τ_D) of ΔI_Cand ΔI_Cl are listed in Table 1. The kinetics of cells 1, 2, and 10 was faster than that of the rest. The mean decay time constant (τ_D) of cells 1, 2, and 10, for example, ranged from 27 to 56 msec, whereas that of cells 3–9 varied from 150 to 725 msec. Because the axon terminals of rod-dominated HBCs ramify primarily in strata 1 and 2, and axon terminals of rod-dominated DBCs ramify primarily in stratum 10 (Wu et al., 2000), this result suggests that the postsynaptic responses of rod-dominated bipolar cells in ACs are more transient than the postsynaptic responses of cone-dominated bipolar cells. It is not clear why rod-dominated inputs are more transient in ACs. Because the light responses of rod-dominated bipolar cells are not more transient than the cone-dominated bipolar cell responses (Yang and Wu, 1997; Wu et al., 2000), the difference in AC response kinetics is probably mediated by differences in the amacrine–amacrine and amacrine–bipolar synaptic network. Additional studies are needed to clarify this issue.

Layer-by-layer computation of ON and OFF signals in broadly monostratified and multistratified amacrine cells

We subsequently examined whether the rules segregating ON and OFF channels in various IPL strata set forth by the narrowly monostratified cells are followed by broadly monostratified and multistratified (bistratified, tristratified, and diffuse) ACs. Figure4 shows the morphology (top panel) and ΔI_Cl and ΔI_C (bottom panel) of 10 (cells 11–20) ACs. Cell 11 was broadly monostratified in strata 3 (ON stratum) and 4 (OFF stratum), and ΔI_Cs and ΔI_Cls were ON–OFF. Cell 12 was broadly monostratified in strata 8 (ON stratum), 9 (ON stratum), and 10 (ON stratum), and ΔI_Cs and ΔI_Cls were ON. Cell 13 was broadly monostratified in strata 1 (OFF stratum) and 2 (OFF stratum), the ΔI_C was OFF, and the ΔI_Cl was ON–OFF, whereas cell 14 was narrowly tristratified in strata 3 (ON stratum), 7 (ON stratum), and 9 (ON stratum), and ΔI_Cs and ΔI_Cls were ON. Cell 15 was narrowly tristratified in strata 1 (OFF stratum), 7 (ON stratum), and 9 (ON stratum), and ΔI_Cs and ΔI_Cls were ON–OFF, and cell 16 was narrowly bistratified in strata 2 (OFF stratum) and 4 (OFF stratum), the ΔI_C was OFF, and the ΔI_Cl was ON. Cells 17 and 18 were diffuse across all strata (ON and OFF strata), and ΔI_Cs and ΔI_Cls were ON–OFF. Cell 19 (wide dendritic field) was broadly monostratified in strata 9 (ON stratum) and 10 (ON stratum), and ΔI_Cs and ΔI_Cls were ON, and cell 20 was broadly monostratified in strata 1 (OFF stratum), 2 (OFF stratum), and 3 (ON stratum), and ΔI_Cs and ΔI_Cls were ON–OFF.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Top, Sketches based on stacked confocal fluorescent images of 10 (cells 11–20) broadly monostratified (cells 11–13, 19, and 20), narrowly bistratified (cell 16), narrowly tristratified (cells 14 and 15), and diffuse (cells 17 and 18) ACs.PRL, Photoreceptor layer; OPL, outer plexiform layer; INL, inner nuclear layer;GCL, ganglion cell layer. Bottom, Current responses evoked by a 0.5 sec light step (500 nm, −3.3) at holding potentials of −60 mV (ΔI_C) and 0 mV (ΔI_Cl).

In addition, as shown in Figure 2, cell e was broadly tristratified in strata 3 (ON stratum), 6 (ON–OFF stratum), 7 (OFF stratum), and 10 (ON stratum), and ΔI_Cs and ΔI_Cls were ON–OFF. Cellf was narrowly bistratified in strata 3 (ON stratum) and 8 (ON stratum), and ΔI_Cs and ΔI_Cls were ON–OFF. Cellg was broadly monostratified in strata 9 (ON stratum) and 10 (ON stratum), and ΔI_Cs and ΔI_Cls were ON, and cell hramified diffusely across strata 1–9 (ON and OFF strata), and ΔI_Cs and ΔI_Cls were ON–OFF. We also examined the correlation between ON–OFF ΔI_Cand ΔI_Cl responses and patterns of dendritic stratification of the remaining 101 multistratified ACs listed in Figure 1B. All but five diffuse ACs (which showed ON instead of ON–OFF ΔI_C and ΔI_Cl) followed the ON–OFF rules for the ΔI_C and ΔI_Cl responses: cells with dendrites in strata 1, 2, and 4 exhibited an OFF ΔI_C and an ON–OFF ΔI_Cl; cells with dendrites in strata 3 and 7–10 gave rise to an ON ΔI_Cand ΔI_Cl; and cells with dendrites in strata 5 and 6 or with diffuse dendrites gave rise to an ON–OFF ΔI_C and ΔI_Cl.

It is clear from these results that the rules correlating ON–OFF channels and dendritic stratification in the IPL set forth by narrowly monostratified cells (Fig. 3 and Table 1) are closely followed by all 79 broadly monostratified, bistratified, and tristratified ACs, and by the vast majority (31 of 36) of diffuse ACs. Although the amplitudes of the broadly monostratified cells and the responses of the multistratified ACs did not necessarily correlate with the amplitudes of narrowly monostratified AC responses, the kinetics of ΔI_C and ΔI_Cl was very similar to that of the narrowly monostratified ACs ramifying in the strata in which the broadly monostratified and multistratified cells have dendritic branches. These results suggest that ACs with dendrites in multiple strata compute their signals in a layer-by-layer (stratum-by-stratum) manner. At least for the ON–OFF signals, responses of broadly monostratified and multistratified cells seem to result from summing up responses of the individual dendritic strata of the cells.

Amacrine–amacrine synaptic circuitry in the IPL

In the salamander retina, the dark membrane potential of ACs is near −80 mV (Yang et al., 2002), andE_Cl values in amphibian ACs are in the range of −50 to −70 mV (Miller and Dacheux, 1983). Therefore the light-evoked voltage responses are mediated by both ΔI_C and ΔI_Cl, and thus the ON–OFF pattern of voltage responses at −80 mV (ΔV) of each AC is determined by the ON–OFF pattern of ΔI_C and ΔI_Cl (the I–Vrelationship of the light-evoked current is approximately linear between 0 mV and −100 mV) (Yang et al., 2002). Consequently, if all ACs have dark membrane potentials near −80 mV, then narrowly monostratified cell types 1, 2, and 4–6 may receive inhibitory inputs from themselves, from other ACs with an ON–OFF ΔV, or from at least one type of AC with an ON–OFF ΔV and some ACs with an ON ΔV. However, narrowly monostratified cell types 3 and 7–10 can receive inhibitory inputs only from themselves or from other ACs with an ON ΔV. These amacrine–amacrine synaptic connections in the 10 IPL strata are summarized by a logic circuit diagram (Tokheim, 1994) in Figure5. This logic circuit gives all possible connections made by ACs on narrowly monostratified ACs in each stratum. The number of possible connections is very large, because for each OR (any or all) gate of five inputs, there are 31 possible connections (whereC  is the number of combinations with k out of five strata), and for the AND (all or nothing) gate with two inputs from the OR gates, there are 31× 31 = 961 possible connections. Therefore there are 31 + 961 = 992 possible connections that mediate amacrine cell inputs (ΔI_Cl) to the narrowly monostratified ACs in strata 1, 2, and 4–6 and 31 possible connections that mediate amacrine cell inputs (ΔI_Cl) to the narrowly monostratified ACs in strata 3 and 7–10. Figure 5 clearly indicates that the amacrine–amacrine synaptic circuitry is asymmetric in favor of the ON channels: ACs with ON responses may make synaptic connections with ACs in all 10 strata, whereas ACs with ON and OFF responses may make synaptic connections only with ACs in strata 1, 2, and 4–6.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Logic circuit diagram of amacrine–amacrine synaptic connections in the IPL. ΔI_C, ΔV₋₈₀, and ΔI_Cl are light-evoked excitatory current, light-evoked voltage response at −80 mV, and light-evoked chloride current, respectively. Open circles are ON responses,filled circles are OFF responses, andhalf-open/half-filled circles are ON–OFF responses.  and  are logic OR and AND gates respectively.

The logic circuit in Figure 5 is based on the finding that most ACs in dark-adapted salamander retina have a dark membrane potential near −80 mV (Yang et al., 2002). However, for those ACs with dark membrane potentials near −60 mV or above (Vallerga, 1981), the ON–OFF patterns of the voltage responses of the cell (ΔV) are different from the ΔV₋₈₀ listed in Figure 5, and more complex logic circuits are needed. In addition, analysis of cells e–h in Figure 2 and cells 11–20 in Figure 4 shows that the layer-by-layer rule for the ON–OFF responses applies for amacrine–amacrine inputs (ΔI_Cl) in broadly monostratified and multistratified ACs; therefore ΔI_Cls of ACs with dendrites ramified in multiple strata may be computed as sums of ΔI_Cls of individual strata.

Bipolar cell inputs to amacrine cells are organized in a layer-by-layer manner in the IPL: segregation and integration of ON and OFF signals

A major finding of this study is that the light-evoked excitatory synaptic inputs from ON and OFF bipolar cells to ACs, ΔI_C, are organized in a layer-by-layer manner with 10 strata in the IPL. This functional architecture resembles many parallel information pathways in higher visual centers, such as the columnar segregation of neurons registering different orientations in the visual cortex (Hubel and Wiesel, 1968,1969). Our data show that ∼30% of ACs are narrowly monostratified (Fig. 1A), and they ramify in each of the 10 strata of the IPL (types 1–10 for strata 1–10). Responses of these cells demonstrate that the ON and OFF channels are segregated in 10 strata in the IPL: ΔI_Cs in strata 1, 2, and 4 are OFF responses (predominantly mediated by the OFF bipolar cells); ΔI_Cs in strata 3 and 7–10 are ON responses (predominantly mediated by ON bipolar cells); and ΔI_Cs in strata 5 and 6 are ON–OFF responses (mediated by both ON and OFF bipolar cells). Among the 12 types of bipolar cells in the tiger salamander retina, only five types have monostratified axon terminals, whereas the axons of others are bistratified, tristratified, or pyramidally branching (Wu et al., 2000). Therefore, some strata in the IPL may contain axon terminals primarily from one type of bipolar cell, whereas other strata may contain axon terminals from several types of bipolar cells. What our data suggest, nevertheless, is that the inputs from bipolar cells to ACs in each stratum are the same within that stratum: ON, OFF, or ON–OFF.

Our data suggest that narrowly monostratified ACs carry segregated bipolar cell signals in the inner retina. In contrast, broadly monostratified and multistratified cells integrate bipolar cell signals through layer-by-layer summation: ACs with dendrites ramified in multiple strata exhibit ΔI_C that are sums of ΔI_Cs of individual strata (Fig. 4). This suggests that the ON and OFF bipolar cell inputs to ACs (monostratified, multistratified, narrowly stratified, or broadly stratified) in each stratum are very similar, and that ΔI_Cs from each stratum are transmitted to the soma in a parallel (and thus additive) manner. This is consistent with previous studies that have shown that most bipolar cells, either ON (DBCs) or OFF (HBCs), have synaptic terminals ramifying in a restricted stratum (or strata) of the IPL (Wu et al., 2000). Although it is difficult to match all bipolar cell light-response attributes with ΔI_Cs in ACs, our results demonstrate a clear stratum-by-stratum rule for the ON–OFF response channels (set forth in Fig. 3 and Table 1) in the IPL. This rule primarily agrees with the sublamina A/B rule for the OFF–ON cells established by previous studies in the cat retina (Famiglietti and Kolb, 1976; Nelson et al., 1978) and salamander retina (Pang et al., 2002). An obvious exception is that we found that the ΔI_C in stratum 3 (located in the middle of sublamina A) exhibits ON instead of OFF responses. Previous studies in salamanders and turtles have shown that there are DBCs that bear axon terminals ramifying in two strata in the IPL, one in sublamina A and the other in sublamina B (Ammermuller and Kolb, 1995;Wu et al., 2000). Recent results have shown that antibodies against G-protein α-subunit stain DBCs in the salamander retina, and in addition to heavy labeling in sublamina B, there is a clear band near stratum 3 in sublamina A (J. Zhang and S. M. Wu, unpublished results). It is possible, at least in the tiger salamander retina, that the ON channels mediated by DBCs activate synapses not only in sublamina B but also in the middle of sublamina A (stratum 3). This may partially explain why the vast majority of ACs and ganglion cells in the salamander retina are ON–OFF or ON cells, and only ∼5–10% of these third-order cells are OFF cells (Hensley et al., 1993; Gao and Wu, 1998; Yang et al., 2002).

At first glance, the division of the IPL into 10 strata appears arbitrary. However, the radial dendritic thickness of the majority of the narrowly stratified cells (either monostratified or multistratified) is very close to one-tenth the thickness of the IPL (only ∼10% are slightly wider than one-tenth of the IPL, and ∼10–15% are thinner than one-tenth of the IPL). This provides the anatomical rationale for the 10 strata scheme. In addition, our physiological data also support this scheme. For example, cells 2 and 4 in Figure 3 (narrowly monostratified in strata 2 and 4, respectively) are OFF cells, but cell 3, which has dendrites ramified approximately one-tenth of the IPL thickness away from the dendrites of cells 2 and 4, is an ON cell. Table 1 and Figure 4 show that this clear physiological difference can be observed in all monostratified and multistratified cells that bear narrowly stratified dendrites in strata 2–4. Another example is that although cells 9 and 10 (Fig. 3) are both ON cells, their ΔI_C kinetics is quite different. Table 1 shows that the ΔI_C decay time constant (τ_D) of cells in stratum 9 is 725 ± 78, and that the decay time constant of cells in stratum 10 is 56 ± 11. Therefore, the morphological and physiological data we obtained support the 10 strata scheme for classifying ACs in the tiger salamander retina.

Amacrine–amacrine synaptic interactions in the IPL: asymmetric synaptic circuitry in favor of the ON channels

Our data show that the inhibitory synaptic inputs to ACs are more complex than the excitatory inputs. Responses of narrowly monostratified ACs suggest that the light-evoked inhibitory inputs, ΔI_Cl, to strata 1, 2, and 4 (strata with OFF ΔI_Cs) and strata 5 and 6 (strata with ON–OFF ΔI_Cs) are ON–OFF (mediated predominantly by ON–OFF ACs), and that those to strata 3 and 7–10 (strata with ON ΔI_Cs from DBCs) are ON (mediated predominantly by ON ACs). This indicates that the narrowly monostratified cells receiving excitatory inputs from OFF bipolar cells (types 1, 2, and 4) must receive inhibitory inputs from ON and OFF channels mediated by both ON and OFF bipolar cells. However, narrowly monostratified cells receiving excitatory inputs from ON bipolar cells (types 3 and 7–10) receive inhibitory inputs only from the ON channel mediated by the ON bipolar cells. Therefore, at least in the tiger salamander, the amacrine–amacrine synaptic circuitry in the IPL seems asymmetric in favor of the ON channels. The logic circuit diagram in Figure 5 clearly demonstrates this asymmetry: ACs with ON responses may make connections with ACs in all 10 strata, whereas ACs with ON and OFF responses may make connections only with ACs in strata 1, 2, and 4–6. Although it is unclear how such an asymmetrical synaptic network is organized at the ultrastructural level, the logic circuit in Figure 5begins to suggest some computational algorithms for amacrine–amacrine synaptic wiring in the inner retina.

Morphological diversity and classification of retinal amacrine cells

We have shown that ACs in the tiger salamander, like their counterparts in other vertebrates, exhibit extreme morphological diversity (Vaney, 1990; MacNeil and Masland, 1998; Masland, 2001). ACs differ from one another in their shape and size of cell bodies, dendritic width and thickness, patterns of dendritic branching, and level(s) of dendritic stratification in the IPL. Certain morphological features are closely correlated with different physiological responses (e.g., the levels of dendritic stratification in the IPL of narrowly monostratified cells closely correlate with the ΔI_C ON and OFF responses), whereas other morphological features are not [e.g., the bistratified cell in strata 3 and 8 (cell f in Fig. 2) and the monostratified cell in stratum 7 (cell c in Fig. 2) have very similar light responses]. In addition, some ACs with dendrites ramifying in the same strata exhibit very different light responses (e.g., cell 17 and cell 18 in Fig. 4, both with diffuse dendrites ramified in strata 1–10, exhibit different ΔI_Cs and ΔI_Cls).

From these results, it is evident that AC classification is a complex task. In addition to levels of dendritic stratification in the IPL and the polarity and kinetics of ΔI_C and ΔI_Cl, there are many other morphological and physiological parameters that define AC types. These parameters include the receptive field size, presynaptic and postsynaptic partners, neurotransmitter contents, and patterns of dendritic branching of the cells. Each of these additional parameters may “lump” or “split” the various cell types set forth by the previous parameters (e.g., types based on the different levels of dendritic stratification in Fig. 1B) into a new set of AC types. For example, cell 13 and cell 20 in Figure 4 resemble the dopaminergic (tyrosine hydroxylase-positive) ACs, and cell 12 and cell 19 in Figure 4 and cell g in Figure 2 resemble the serotonin cells (Watt et al., 1988; Watt, 1992). Therefore the additional parameter (neurotransmitter content) lumps cells 13 and 20 into one type and cells 12 and 19 into another type. However, cell types 1 and 2 (Fig. 1B) both contain narrow-, medium-, and wide-field cells; thus, the addition of receptive field size as a parameter splits cell types 1 and 2 into six types of ACs. Therefore, because of the extreme morphological diversity and complex physiological and neurochemical variations, classification of retinal ACs is highly dependent on the parameters chosen. It may be impractical to expect a simple universal classification scheme for such a diverse population of interneurons in the retina. It is perhaps more useful, instead, to determine computational rules for individual parameters, such as the layer-by-layer rule for the ON–OFF bipolar cell inputs described in this article, and then integrate them into a comprehensive framework that process various attributes of visual images presented to the eyes.