Glutamate receptors at rod bipolar ribbon synapses in the rabbit retina

Glutamate Receptors at Rod BipolarRibbon Synapses in the Rabbit Retina

WEI LI, E. BRADY TREXLER, AND STEPHEN C. MASSEY*

Department of Ophthalmology and Visual Science, University of Texas Medical School,Houston, Texas 77030

ABSTRACTIn the mammalian retina, maximum sensitivity is achieved in the rod pathway, which

serves dark-adapted vision. Rod bipolar cells carry the highly convergent rod input and makeribbon synapses with two postsynaptic elements in the inner retina. One postsynaptic neuronis the AII amacrine cell, which feeds the rod signal into the cone pathways. The otherpostsynaptic element is either an S1 or S2 amacrine cell. These two wide-field GABAamacrine cells both make reciprocal synapses with rod bipolar terminals but their individualroles are unknown. AII and S1/S2 dendrites come in close together and form a dyad opposingthe presynaptic ribbon, which is the site of glutamate release. Therefore, two postsynapticneurons sense the very same neurotransmitter yet serve different functions in the rodpathway. This functional diversity could be derived partly from the expression of differentglutamate receptors on each postsynaptic element. In this study, we labeled all pre- andpostsynaptic combinations and a signal-averaging method was developed to locate glutamatereceptor subunits. In summary, GluR2/3 and GluR4 are expressed by AII amacrine cells butnot by S1/S2 amacrine cells. In contrast, the orphan subunit �1/2 is exclusively located on S1varicosities but not on AII or S2 amacrine cells. These results confirm the prediction ofdivergence mediated by different glutamate receptors at the rod bipolar dyad. Each differentamacrine cell type appears to express specific glutamate receptors. Finally, the differentialexpression of glutamate receptors by S1 and S2 may partly explain the need for two wide-fieldGABA amacrine cells with the same feedback connections to rod bipolar terminals. J. Comp.Neurol. 448:230–248, 2002. © 2002 Wiley-Liss, Inc.

Indexing terms: retina; rod pathway; AII amacrine cell; indoleamine accumulating amacrine cell

The mammalian retina has a broad functional range,over 10 log units of light (Sterling, 1998). Cone pathwaysprovide high acuity and color discrimination, whereas thehigh sensitivity rod pathway is critical for dark-adaptedvision in dim ambient light.

In dark-adapted conditions, the small signals producedin a group of one to five excited rods can reach conscious-ness in humans (Hecht et al., 1942; Baylor, 1987). Conse-quently, the rod signals must be amplified as they tra-verse the retina and insulated from noise. Rod signalsfollow a specific pathway through the retina, which hasbeen well described (for review, see Bloomfield and Da-cheux, 2001). Rod bipolar cells receive highly convergentinput from approximately 50 rods and make ribbon syn-apses with two postsynaptic elements in sublamina 5 ofthe inner plexiform layer (IPL; Fig. 1A). One postsynapticneuron is the AII amacrine cell, which feeds rod signalsinto the cone pathways by means of inhibitory glycinergicsynapses with OFF cone bipolar cells and by means of gapjunctions with ON cone bipolar cells (Strettoi et al., 1992).The other postsynaptic element is either an S1 or S2

amacrine cell. These two wide-field GABA amacrine cellsboth make reciprocal synapses with rod bipolar terminalsand thus provide local negative feedback and a substratefor lateral inhibition (Sandell et al., 1989; Strettoi et al.,1990; Massey et al., 1992). A schematic drawing of theultrastructure of the synapse (Fig. 1C) shows two postsyn-aptic elements, an AII dendrite and an S1 varicosity, bothopposing a common presynaptic ribbon. The increasingscale of these panels corresponds approximately to con-

Grant sponsor: NEI; Grant number: EY 06515; Grant number: EY10608; Grant sponsor: NEI Vision Core Grant; Grant number: EY 07024.

*Correspondence to: Stephen C. Massey, Department of Ophthalmologyand Visual Science, University of Texas Medical School, 6431 Fannin, MSB7.024, Houston, Texas 77030. E-mail: [email protected]

Received 4 September 2001; Revised 21 December 2001; Accepted 8January 2002

DOI 10.1002/cne.10189Published online the week of May 20, 2002 in Wiley InterScience (www.

interscience.wiley.com).

THE JOURNAL OF COMPARATIVE NEUROLOGY 448:230–248 (2002)

© 2002 WILEY-LISS, INC.

ventional light microscopy, confocal microscopy, and elec-tron microscopy.

Rod bipolar cells produce a depolarizing response tolight as do both postsynaptic neurons from which we havedefinitive recordings in the rabbit retina, the AII amacrinecell and S1 cell (Raviola and Dacheux, 1987; Bloomfield,1996). The cat A17 amacrine cell, homolog to the rabbitS1, is also depolarizing (Nelson and Kolb, 1985), but thereare no confirmed recordings as yet from S2. However, S2 isprobably also depolarizing, because otherwise, the recip-rocal synapses could provide a destabilizing positive feed-back signal. Rod bipolar cells contain high levels of gluta-mate and exogenous glutamate agonists depolarize bothAII and S1 (Boos et al., 1993; Menger and Wassle, 2000;Zhou and Dacheux, 2001). Thus synaptic transfer from rodbipolar cells to both postsynaptic neurons is sign conserv-ing or excitatory by using glutamate as the neurotrans-mitter.

This also means that two different neurons with verydifferent functions respond to glutamate release at a rib-

bon synapse, perhaps by means of different receptors.Previously, postsynaptic pairs at ribbon synapses havebeen shown to express different glutamate receptors, andthis finding suggests a substrate for functional diversity(Brandstatter et al., 1997). This is the case for OFF conebipolar cell subtypes, which express different glutamatereceptors that each produce stereotyped postsynaptic re-sponses (DeVries, 2000). Ionotropic glutamate receptorsare pentameric and a large number of glutamate receptorsubunits have been identified (Michaelis, 1998). SubunitsGluR1-4 associate to form AMPA receptors, whereas kai-nate receptors are composed of GluR5-7, KA1, KA2 sub-units. The so-called orphan subunits �1 and �2 share20–30% sequence identity with the glutamate receptorfamily but do not form functional channels by themselves,although the �2 subunit is expressed in certain neuronssuch as cerebellar Purkinje cells (Wollmuth et al., 2000).All these glutamate receptor subunits have been found inthe retina, in the two plexiform or synaptic layers (Brand-statter et al., 1997; Thoreson and Witkovsky, 1999).NMDA receptors are also present in the inner retina, butthey do not appear to be associated with rod bipolar ter-minals (Fletcher et al., 2000). In addition, there is a largefamily of G-protein-coupled metabotropic glutamate re-ceptors, which act by means of intracellular biochemicalpathways (Pin and Duvoisin, 1995).

Surprisingly, it is reported that input to AII amacrinecells was resistant to CNQX although the surround input,perhaps provided by means of S1 and S2 amacrine cells,was blocked by CNQX (Bloomfield and Xin, 2000). Thisfinding may suggest that AII amacrine cells express glu-tamate receptors with unusual pharmacology and thatdifferent postsynaptic receptors are involved. Therefore,we sought to identify the glutamate receptors for bothpostsynaptic components of the rod bipolar dyad. Previouselectron microscopy (EM) studies showed that AII ama-crine cells were stained by antibodies to GluR2/3 andGluR4 (Qin and Pourcho, 1996, 1999a,b; Thoreson andWitkovsky, 1999). The so-called orphan subunit �1/2 wasalso reported as asymmetrically located on the AII side ofthe dyad (Brandstatter et al., 1997). However, no gluta-mate receptor subunits have been found on the S1/S2 sideof the dyad.

In this study, we used confocal microscopy to localizeglutamate receptor subunits to different postsynaptic ele-ments at the rod bipolar dyad. This kind of resolution isusually thought to require EM analysis. However, by us-ing cell-specific antibodies, such as PKC for rod bipolarcells (Young and Vaney, 1991), calretinin for AII amacrinecells (Wassle et al., 1995; Massey and Mills, 1999), andserotonin for S1/S2 (Vaney, 1986; Sandell and Masland,1986), we were able to label all pre- and postsynaptic cellsselectively for triple-label confocal microscopy. In addi-tion, a signal averaging method was developed and ap-plied to repeated neuronal structures in a given image.Despite the close proximity of the three structures at rodbipolar dyads, we were able to conclusively assign theexpression of several glutamate receptor subunits to ei-ther AII or S1/S2 amacrine cells. Furthermore, we foundevidence that S1 and S2 amacrine cells also differ in theirglutamate receptor components, despite the similarity oftheir reciprocal connections with rod bipolar terminals.

Fig. 1. Schematic drawing of dyad synapse at the rod bipolarterminal A: A rod bipolar cell is highlighted from a cross-sectionalimage of rod bipolar cells in the rabbit retina. Synaptic connections atone of the rod bipolar (RBP) terminals at the level of sublamina 5(shaded area) are encompassed by a square, which is magnified in B.OPL, outer plexiform layer; INL, inner nuclear layer; IPL, innerplexiform layer. B: Rod bipolar terminals are contacted by dendritesfrom AII cells and varicosities from both S1 and S2 cells. C: Schematicdrawing of the ultrastructure of dyad synapse at rod bipolar terminal.S1 and S2 amacrine cells feed back to rod bipolar axon terminals byreleasing GABA, shown here as dark vesicles. Scale bars � 2.0 �m inB, 0.2 �m in C.

231GLUTAMATE RECEPTORS AT ROD BIPOLAR TERMINALS

MATERIALS AND METHODS

Preparation of retinas

As described in detail previously (Massey and Mills,1999), the isolation of the rabbit retina is briefly summa-rized as follows. Under a protocol approved by the Insti-tutional Animal Welfare Committee, adult New ZealandAlbino White rabbits of either sex (2–3 kg) were deeplyanesthetized with urethane (loading dose, 1.5 g/kg, i.p.)and the orbit was infused with 2% lidocaine hydrochloridebefore enucleation. The eye was then removed and hemi-sected. The retina was isolated from the inverted eyecupwhile immersed in oxygenated Ames medium (Ames andNesbett, 1981). The isolated retina was mounted ontofilter paper, ganglion cell side up and fixed for 20–30minutes with 4% paraformaldehyde in 0.1 M phosphatebuffer (pH 7.4) for further immunocytochemical processes.For experiments with intracellular dye injection, the ret-ina cells were prelabeled with 4,6-diamino-2-phenylindole(DAPI) by incubating the retina in Ames medium with 5�M DAPI for 30 minutes. Sometimes the retina was incu-bated in 10 �M serotonin for 30 minutes to label theindoleamine accumulating S1 and S2 amacrine cells (San-dell and Masland, 1986; Vaney, 1986).

Injection of neurobiotin

Retina pieces prelabeled with DAPI were visualized onan Olympus BX-50WI microscope (Tokyo, Japan) withepifluorescence. Cells were impaled under visual controlby using pipettes tip filled with 4% Neurobiotin (VectorLaboratories, Burlingame, CA) and 1% Lucifer Yellow-CH(Molecular Probes, Eugene, OR) in ddH2O and backfilledwith 3 M LiCl. The electrodes had resistance approxi-mately 150 M�. The impaled cells were then injected witha biphasic current (�1.0 nA for 100 msec and �1.0 nA for100 msec) for 10–15 minutes. After the last injection, theretina pieces were fixed in 4% paraformaldehyde for 30minutes before further immunocytochemical experiments.

Immunocytochemistry

After fixation, the tissues were washed extensively with0.1 M phosphate buffer (PB, pH 7.4) and blocked with 3%donkey serum in 0.1 M PB with 0.5% Triton X-100 and0.1% NaN3 overnight. The antibodies were diluted in 0.1M PB with 0.5% Triton X-100 and 0.1% NaN3 containing1% donkey serum. The tissues were incubated for 3–7days at 4°C and, after extensive washing, incubated insecondary antibodies overnight at 4°C. After washing with0.1 M PB, the tissues were mounted on slides for obser-vation. The primary antibodies used in this study includedthe following: rabbit polyclonal antibodies to glutamatereceptors (GluR): GluR1 (1:100), GluR2/3 (1:100), GluR4(1:100), and �1/2 (1:100; Chemicon, Temecula, CA); rabbitpolyclonal antibodies to GluR6/7 (1:500) and KA2 (1:500)(Upstate Biotechnology, Lake Placid, NY); goat polyclonalantibodies to calretinin (1:5,000; Chemicon) and serotonin(1:1,000; DiaSorin, Stillwater, MN); mouse monoclonalantibodies to PKC� (1:500; Transduction Laboratories,Lexington, KY) and kinesin (1:100; Covance, Richmond,CA). The secondary antibodies used were donkey anti-rabbit Cy-3 (1:200), donkey anti-mouse Cy-5 (1:200; Jack-son ImmunoResearch Laboratories, West Grove, PA), anddonkey anti-goat Alexa-488 (1:200; Molecular Probes).Neurobiotin was visualized by Alexa-488–conjugatedstreptavidin (Molecular Probes).

Confocal microscopy and three-dimensionalreconstruction

Images were acquired on a Zeiss LSM-410 (Zeiss,Thornwood, NY) confocal microscope with a krypton/argonlaser by using 488-, 568-, and 647-nm lines and a 63�objective (N.A. 1.4). Alignment for all three channels andresolution were checked at �8 zoom by using 1 � fluores-cent spheres (Molecular Probes). The XY resolution of theinstrument was 200–300 nm and all three channels weresuperimposed. Digital images were processed in AdobePhotoshop (Adobe Systems, San Jose, CA). For three-dimensional (3D) reconstructions, confocal image serieswere acquired at 0.2 �m intervals and processed in Amirasoftware (TGS, San Diego, CA).

Signal averaging analysis

Digital images were analyzed with custom software thatallowed the level of association between two labeled struc-tures to be distinguished from chance. Repeating struc-tures of interest, such as clusters of GluR labeling or S/S2varicosities, were clipped from the image by centering asampling box on the structure. Alignment and averagingof these boxes produces a plot of color intensity vs. pixelpositions that reveals the spatial distribution of the im-munofluorescence around a specific type of structure. Astrong level of association between two channels is re-vealed by two correlated peaks from the average intensityplots of two colors representing them, whereas a caldera-like plot from one signal that is associated with a peakfrom another signal represents an anti-correlation, i.e., anexclusive relationship between the structures. A 3 � 3median filter was sometimes applied for smoothing. Theaveraged image was divided into annuli of equal radii;Duncan’s test was used to establish the level of signifi-cance for each annulus. Control images were produced byrotating one color channel out of phase, which usuallyproduced a flat pattern without correlated peaks, repre-senting random overlap.

RESULTS

GluR2/3 is expressed on AII dendrites butnot S1/S2 varicosities

Whole-mount rabbit retinae were immunolabeled withantibodies to (1) PKC, which is a specific marker for rodbipolar cells; (2) calretinin, which prominently labels AIIamacrine cells; and (3) GluR2/3. This combination of anti-bodies reveals pre- and postsynaptic components and thereceptor subunits that reside on the postsynaptic cells.Images were taken from whole-mount preparations at thelevel of rod bipolar terminals, sublamina 5 in the IPL. InFigure 2A, the punctate staining of the GluR2/3 receptorsubunit antibody (red) along the surface of the rod bipolarterminals suggests that GluR2/3 may associate with dyadsynapses at rod bipolar terminals (blue). When theGluR2/3 signal and the calretinin signal are superimposedinstead (Fig. 2B), it is evident that most of the GluR2/3puncta are located on the AII dendrites (green). In Figure2C, where all three labels are presented, it is clear thatmost of GluR2/3 puncta appear on the AII dendrites at thecontact sites with rod bipolar terminals.

Before further analysis of the receptor localization, wefirst confirmed that GluR2/3 staining occurs at postsynap-

232 W. LI ET AL.

Fig. 2. The localization of GluR2/3 receptor subunits was deter-mined in triple-labeled tissue by using antibodies to GluR2/3 (red)together with calretinin (CR), which labels AII amacrine cells (green),and PKC, which labels rod bipolar cells (blue). A: Focused at thebottom of the inner plexiform layer in a piece of whole-mount retina,the image shows the punctate staining of GluR2/3 (red) that sur-

rounds the rod bipolar terminals (blue). B: In the same frame of theimage shown in a A, almost all the GluR2/3 puncta (red) are on AIIamacrine cells stained with calretinin antibody. C: An image of thesame area with all three labels visible illustrates that GluR2/3 punctaare localized primarily to the contacts between AII dendrites and rodbipolar terminals. Scale bars � 5 �m in B (applies to A,B), 5 �m in C.


tic locations. Kinesin II is a protein associated with pre-synaptic ribbons (Muresan et al., 1999, Li et al., 2001).Therefore, we used an antibody to kinesin II as a presyn-aptic marker for rod bipolar synaptic ribbons. As shown inFigure 3A, most of the GluR2/3 clusters (red) have a green

kinesin punctum associated with them. However, they arenot perfectly colocalized. The spatial separation of the twosignals is not due to the misalignment of excitation wave-lengths because the signals from the two wavelengthscould not be superimposed. The tissue was also labeled

Fig. 3. GluR2/3 puncta on AII amacrine cells are postsynaptic, asshown by their close proximity to kinesin labeling of synaptic ribbons.A: Most of the red GluR2/3 clusters have a corresponding greenkinesin structure, but they are not perfectly colocalized. B: A triple-labeled image of the same frame as A shows that GluR2/3 clusters(red) reside on AII dendrites (blue), with kinesin (green) representing

presynaptic ribbons nearby. C,D: A three-dimensional reconstructionof GluR2/3 (red), kinesin (green), and AII dendrites (blue) surround-ing a putative rod bipolar terminal shows the same result observedfrom A and B more clearly with higher magnification. Scale bars � 5�m in A (applies to A,B), 2 �m in C (applies to C,D).

234 W. LI ET AL.

with an antibody to calretinin (Fig. 3B), demonstratingthat the GluR2/3 clusters, but not the ribbons, reside inAII dendrites (blue). A higher magnification 3D recon-struction was generated from a series of confocal images toshow several GluR2/3 clusters associated with synapticribbons stained for kinesin II (Fig. 3C). Most of theGluR2/3 clusters are embedded in the AII dendrites thatenclose an unstained rod bipolar terminal, whereas thesynaptic ribbons are located toward the center of the holein the AII matrix, presumably within the rod bipolar ter-minal (Fig. 3D). This figure shows that GluR2/3 clusterson AII amacrine cells are postsynaptic to rod bipolar syn-aptic ribbons.

Then we examined the colocalization of GluR2/3 clus-ters with AII dendrites and rod bipolar terminals. If

GluR2/3 subunits are part of the AII postsynaptic recep-tor, receiving input from rod bipolar terminals, then allthree signals should be colocalized. To demonstrate this,we signal averaged all of the pixels surrounding GluR2/3clusters that were located on the surface of the rod bipolarterminals. As shown in Figure 4A, sampling boxes (55 �55 pixels) were placed on the image, each centered on aGluR2/3 cluster. We chose only those receptors associatedwith the blue rod bipolar signal and, to avoid bias, thegreen AII signal was turned off. After alignment and av-eraging, the three channel intensities were plotted againstthe location of pixels in the x-y plane, revealing the spatialdistribution for each channel around the GluR2/3 clusters(Fig. 4C–E). Because GluR2/3 clusters were chosen as thecenter of the sampling boxes, there is a sharp central peak

Fig. 4. Signal-averaging analysis was used to localize GluR2/3clusters. A: Boxes of 6 �m (50 pixels) were centered on the GluR2/3clusters (red) that are adjacent to rod bipolar terminals (blue). Toavoid sample bias for clusters contacting AII amacrine cells, thecalretinin signal (green) was turned off before sampling. B: Aftersampling, calretinin signal was turned back on and the intensities ofthe red, green, and blue pixels within the box were read by thesoftware and averaged according to the two-dimensional location ofeach pixel in the box. C: The averaged GluR2/3 signal is presented asa surface plot of the intensity of red color (z axis) vs. the location of thepixels (x and y axis). The plot shows a sharp peak at the center, asexpected due to sampling bias. D: The plot of blue representing therod bipolar signal (PKC) shows a broad hump with a peak at the

center, consistent with the sampling bias. The sampling boxes werecentered over GluR2/3 clusters that were adjacent to rod bipolarterminals. The large size and random orientation of the rod bipolarterminals yields a less sharp peak than the GluR2/3 clusters (C).E: The average green signal, which represents AII dendrites, was freefrom sampling bias. However, it also shows a peak at center, whichindicates that AII dendrites also colocalize with GluR2/3 clusters thatare adjacent to rod bipolar terminals. F: As a control, the green imagein B was rotated 90 degrees, and the analysis was performed again.On average, the rotated AII signal shows no correlated peak withGluR2/3 signal, indicating that the peak seen in E is not due to chancecolocalization.


in the average GluR2/3 signal (Fig. 4C). The average rodbipolar signal yields a less-defined central peak, becauseof the random orientation and large size of the rod bipolarterminals (Fig. 4D). Importantly, the AII signal, whichwas free from sampling constraints, showed a sharp andprominent central peak that coincides with the GluR2/3peak (Fig. 4E). In other words, when a GluR2/3 clusteroccurs postsynaptic to a rod bipolar terminal, there isalmost certainly an AII dendrite at that point. As a con-trol, the green AII channel was rotated 90 degrees out ofphase and the analysis was repeated (Fig. 4F). Althoughthere were many apparent AII contacts with rod bipolarterminals due to random overlap, the loss of the centralAII peak indicates that these controls are not associatedwith GluR2/3 clusters. In summary, this is a quantitativedemonstration that GluR2/3 clusters occur where rod bi-polar terminals contact AII dendrites.

In addition to AII dendrites, S1/S2 amacrine cells formthe second postsynaptic component at rod bipolar dyads.Therefore, we also analyzed the spatial distribution ofGluR2/3 clusters at contact points with S1/S2 varicosities.S1 and S2 cells were injected with Neurobiotin, and thetissue was further immunolabeled with antibodies toGluR2/3 and PKC. An example of an S1 varicosity (green),which wraps around a rod bipolar terminal (blue), isshown in Figure 5A. The varicosity was outlined, andthere are no GluR2/3 puncta within the varicosity, butthere are two close to the edge (Fig. 5B, arrows). These arelikely associated with a neighboring AII dendrite, notvisible here. For a more detailed view of the spatial rela-tionships between the three structures, a series of highmagnification confocal images were taken and used togenerate a 3D reconstruction of an S1 varicosity (Fig.5C,D). In Figure 5C, the rendering of the rod bipolarterminal (blue) is translucent, revealing the contact areabetween the S1 varicosity and the rod bipolar terminal.The volume was rotated (Fig. 5D), and the rod bipolar wasremoved to depict the lack of GluR2/3 staining on thesurface of the S1 varicosity. Although the S1 contact pad islargely free of GluR2/3 subunits, there is one outlyingcluster at the edge (Fig. 5D, arrow), which may be associ-ated with a neighboring AII dendrite.

To confirm that GluR2/3 receptor subunits do not colo-calize with S1/S2 varicosities, signal-averaging analysiswas performed on images clipped from triple-labeled tis-sue (Fig. 6A,B). The sampling boxes were centered overthe varicosities from both S1 and S2 cells, respectively. Wechose to sample the varicosities because they are clearrepeated structures in these images, suitable for signalaveraging. The surface plot of the color intensity shows abroad, yet well-defined peak for S1 varicosities (Fig. 6C).However, the average GluR2/3 signal shows a volcano-likepattern with a hole, rather than a correlated peak at thecenter (Fig. 6D). This finding indicates that the two struc-tures are anti-correlated, rather than colocalized. Thesame result was obtained for S2 varicosities and the av-erage GluR2/3 signal (Fig. 6E,F), except the S2 peak andthe GluR2/3 hole are smaller, reflecting the smaller sizeof S2 varicosities. Thus, GluR2/3 receptor subunits arenot located on S1/S2 varicosities but adjacent to them,on the other component of the dyad, a neighboring AIIdendrite.

GluR4 is associated with AII dendrites butnot S1/S2 varicosities

We found the distribution pattern of GluR4 is verysimilar to that of GluR2/3. As shown in Figure 7A, most ofthe GluR4 clusters (red) are on the AII dendrites (green)and, thus, appear yellow. When rod bipolar terminals arealso labeled (Fig. 7B), it is clear that the GluR4 clustersare located at contacts between the AII dendrites and therod bipolar terminals. As a more quantitative measure ofGluR4/AII colocalization, we performed signal-averaginganalysis as with GluR2/3 in Figure 4. In the absence of thegreen AII signal, 2.5-�m-square boxes were centered overGluR4 clusters that were associated with rod bipolar ter-minals. After alignment of the sampling boxes, the aver-age intensity for all three signals was calculated. As ex-pected, the average GluR4 signal yields a sharp centralpeak (Fig. 7C). The average AII signal also shows a centralpeak that coincides with the GluR4 signal (Fig. 7D), indi-cating there is a high probability of finding an AII dendriteat GluR4 clusters. Rotation of the AII signal by 90 degreesabolished the central peak (not shown), confirming thatAII/GluR4 colocalization is much greater than randomchance.

The distribution of GluR4 at the non-AII side of thedyad was examined by injecting individual S1 and S2 cellswith Neurobiotin. Figure 8A shows a segment of an S1dendrite bearing a varicosity (green). GluR4 clusters (red)and AII dendrites (anti-calretinin, blue) are also labeled.The inset shows an example of an S2 varicosity. Both S1and S2 varicosities were outlined and the green signalremoved (Fig. 8B), demonstrating that there are no GluR4clusters within the limits of the two varicosities shown.Signal averaging centered on either S1 (Fig. 8C,D) or S2(Fig. 8E,F) varicosities show there is no GluR4 signalcoincident with the average S1/S2 varicosity. These dataindicate that GluR4 receptors are not associated with theS1/S2 varicosities.

�1/2 subunits are expressed on the S1varicosities but not on AII dendrites

After demonstrating that GluR2/3 and GluR4 subunitsare components of the glutamate receptors on AII den-drites postsynaptic to rod bipolar terminals, we sought toidentify the glutamate receptor subunit types on the otherpart of the dyad, the S1 and S2 varicosities. Pieces ofretina were preincubated in 10 �M 5-HT (serotonin) toreveal the entire S1/S2 population after staining with anantibody to serotonin (Sandell and Masland, 1986; Vaney,1986). The retinae were also labeled with antibodies to�1/2 receptor subunits and PKC to label rod bipolar ter-minals. In Figure 9A–C, the same triple-label field isshown but with only two of the three signals in each panel.Figure 9A shows the general relationship of the S1/S2matrix (green) and rod bipolar terminals (blue). The var-icosities from both S1 and S2 dendrites cluster together,forming holes which invariably contain rod bipolar termi-nals. Arrows mark several examples of individual varicos-ities that contact prominent rod bipolar terminals. When�1/2 (red) and PKC (blue) are presented together (Fig. 9B),it is obvious that most of the �1/2 labeling is associatedwith rod bipolar terminals. Notice that each arrow nowpoints to a red punctum, suggesting that �1/2 may belocated on the S1/S2 varicosities. When the �1/2 clusters(red) are presented with the S1/S2 matrix (green), much of

236 W. LI ET AL.

the �1/2 labeling appears yellow, suggesting colocalizationwith the S1/S2 matrix. In high-resolution 3D reconstruc-tions, it appears that �1/2 clusters are on large, probablyS1, varicosities occupying the inside surfaces where thevaricosities wrap around the rod bipolar axon (arrowhead)or terminal (Fig. 9E,F, arrows).

Previous studies have shown that in the middle of IPL,there are occasional varicosities along the descending den-

drites of S1 or S2 amacrine cells that make synaptic con-tacts with the axons of rod bipolar terminals (Fletcher andWassle, 1999; Massey et al., 1999). Because they are abovethe dense matrix surrounding rod bipolar terminals, theseisolated varicosities are particularly suitable for imaging.In Figure 9D, an image of whole-mount retina focusedhigh in the IPL clearly shows �1/2 staining (Fig. 9D,arrowhead) at the contact point between a large S1/S2

Fig. 5. GluR2/3 is not localized on S1 varicosities. A,B: A triple-labeled image was generated with antibodies to GluR2/3 (red) andPKC (blue) and a Neurobiotin-injected S1 cell (green). A varicosity ofthe S1 cell (outlined in a white ellipse), wrapping around the rodbipolar terminal, is located in between two red GluR2/3 puncta. Re-moval of the S1 cell (B) shows that there are no GluR2/3 puncta withinthe limits of the varicosity. Two outlying puncta are marked witharrows. C,D: Three-dimensional reconstructions of another rod bipo-

lar terminal (blue) with an associated S1 dendrite bearing a varicosity(green) and the GluR2/3 puncta (red) give a clearer picture of the lackof GluR2/3 labeling on S1/S2 cells. C: Most of the Glu2/3 puncta (red)are associated with the rod bipolar terminal (blue). D: The GluR2/3puncta are not on the surface of the varicosity of S1 cell, but there isone punctum close to the edge of it (arrow). Scale bars � 5 �m in B(applies to A,B), 2 �m in D (applies to C,D).


varicosity (Fig. 9D, green) and the descending axon of arod bipolar cell (Fig. 9D, blue). These larger varicositiesusually arise from S1, as opposed to S2, amacrine cells.This labeling pattern also provides further evidence thatthese contacts with rod bipolar axons, above the axonbranching point, are synaptic.

For a clearer picture of individual varicosities and theirspatial relationship with �1/2, we injected individual S1and S2 amacrine cells with Neurobiotin. Figure 10A con-tains several S1 varicosities and a few coupled dendrites(green). The inset shows a higher power image of a singlerod bipolar terminal (blue) wrapped by two S1 varicosities(green), one from the injected cell and one from a coupledS1 amacrine cell. There are red �1/2 clusters apposed toevery S1 varicosity which are visible, but more prominent,when the green channel is removed (Fig. 10B). The arrows

indicate the location of S1 varicosities, and now point toindividual �1/2 clusters. Thus it appears that �1/2 is acomponent of glutamate receptors on S1 varicosities.

Analogous to our analyses for GluR2/3 and GluR4 anti-bodies, we examined the distribution of �1/2 labeling onAII dendrites in addition to S1/S2 varicosities. To see bothcell types and the receptor staining, we injected S1 cellswith Neurobiotin and then labeled the retinae for �1/2 andcalretinin to label AII amacrine cells. One example of suchtriple labeled tissue is shown in Figure 11A, in which anS1 dendrite bearing a varicosity (Fig. 11A, green)traverses the matrix of AII dendrites (Fig. 11A, blue). Anarrow points to an S1 varicosity that contains a red �1/2cluster, but the rest of the �1/2 puncta are near but not onAII dendrites (Fig. 11A,B, blue). The lack of overlap be-tween AII (calretinin) and �1/2 staining is clearer in a 3Dreconstruction from a series of high-magnification confocalimages. In Figure 11D, the vast majority of the red �1/2puncta are slightly away from the AII dendrites (Fig. 11D,blue). However, when the green channel of the S1 cell isadded (Fig. 11C), the red cluster marked by an arrowclearly lies within the confines of the S1 varicosity. Signal-averaging was performed with sampling boxes centered on�1/2 clusters (Fig. 11E). In striking contrast to Figure4C,E, which shows a central AII peak that coincides withthe GluR2/3 signal, the distribution of AII labeling aroundthe �1/2 clusters forms a caldera (Fig. 11F,G). There is adepression at the center of the AII plot that coincides withthe peak of the average �1/2 signal. This anti-correlationof the two signals indicates that �1/2 punctate are not onAII dendrites at contact points with rod bipolar terminals.

�1/2 is only on the S1 varicosities but noton S2 varicosities

S1 and S2 amacrine cells are morphologically similarand both make reciprocal GABA synapses with rod bipolarterminals. We have shown that �1/2 is expressed on S1varicosities (Fig. 10), but we also wished to determinewhether the S1 and S2 amacrine cells have the sameglutamate receptors. S2 cells were injected with Neurobi-otin and processed in the same way as S1 cells describedabove. In an example from triple-labeled material, den-drites from two S2 cells (Fig. 12A, green), �1/2 (Fig. 12A,red), and rod bipolar terminals (Fig. 12A, blue) are la-beled. Five varicosities are marked by arrows, and with-out exception, they all contact rod bipolar terminals. How-ever, none of them contain �1/2 clusters, as shown inFigure 12B where the green signal has been removed. Allthe arrows that pointed to S2 varicosities now point toareas on rod bipolar terminals where there is a discernibleabsence of red clusters (Fig. 12). For example, at the areamarked by the arrow farthest to the right, the S2 varicos-ity occupies a space devoid of �1/2 staining, yet it is sur-rounded by three red �1/2 clusters, presumably associatedwith S1 varicosities (Fig. 12). These images demonstratethe lack of colocalization between �1/2 clusters and S2varicosities.

Thus, it appears that S1 and S2 amacrine cells usedifferent glutamate receptors. However, to verify this re-sult, it was necessary to show differential labeling in thesame material. This verification was achieved by injectingneighboring S1 and S2 cells with Neurobiotin. For illus-tration purposes, the S1 dendrite was coded green and theS2 dendrite was coded in yellow with �1/2 clusters in redand PKC labeling of rod bipolar cells in blue (Fig. 12C,D).

Fig. 6. Signal-averaging analysis confirms the separation ofGluR2/3 clusters and S1/S2 varicosities. A,B: An image taken from atriple-labeled material with antibodies to GluR2/3 (red) and PKC(blue) and a Neurobiotin-injected S1 cell (green) shows an example ofS1 varicosity contacting a rod bipolar terminal. For signal-averaginganalysis, the sampling boxes were centered over the S1 or S2 varicos-ities, rather than the GluR2/3 clusters. Twenty varicosities wereaveraged for both S1 and S2 cells. C,D: A surface plot of S1 varicos-ities show a peak in the average green signal, representing the vari-cosity (C), which corresponds to the “hole” in the caldera-like patternof the average red signal, representing GluR2/3 clusters (D). E,F: Asurface plot of the average of S2 samples yields a peak, as with S1varicosities, except the peak of the average S2 signal and hole of theaverage GluR2/3 signal are smaller, reflecting the smaller size of S2varicosities compared with S1. Scale bar � 2 �m in B (applies to A,B).

238 W. LI ET AL.

The varicosities of each cell are outlined by a white ellipseand, although there are two red �1/2 clusters within theboundary of the S1 varicosity, there are none associatedwith the nearby S2 varicosity (Fig. 12C,D).

Finally, we performed a signal-averaging analysis forboth types of varicosity. When fourteen S1 varicositieswere selected, the average �1/2 signal formed a well-defined coincident peak (Fig. 13A,B). This finding indi-

Fig. 7. GluR4 is localized to AII dendrites. A: A double-labeledimage shows that most of the red GluR4 clusters reside on AII den-drites (green) and appear yellow. B: In the same frame as A, atriple-labeled image shows GluR4 (red) clusters are located on AIIdendrites (green) where they contact rod bipolar terminals (blue).

C,D: Signal average analysis by using GluR4 clusters for samplingproduces a central peak in the average AII signal (D), which coincideswith the central peak of the average GluR4 signal (C). Scale bar � 5�m in A (applies to A,B).


cates that �1/2 subunits are present on S1 varicosities.However, when twenty S2 varicosities were sampled, theaverage �1/2 signal had no central peak and the distribu-tion was essentially random. This finding indicates that�1/2 subunits are not expressed by S2 varicosities. Insummary, these data show the differential expression ofglutamate receptors between S1 and S2 amacrine cells.

Other GluRs are not associated with the rodbipolar terminal dyad

As listed in Table 1, we examined the labeling pattern ofmany other glutamate receptor subunits. These antibod-ies have all previously been reported to work well for otherretinal material by means of Western blot and immuno-cytochemistry (Qin and Pourcho, 1999a, b; Brandstatter etal., 1997). All the subunit antibodies generated a punctatepattern in the IPL, OPL, or both, as expected for synapticlabeling, but they were not associated with rod bipolarterminals or their postsynaptic elements.

DISCUSSION

Localizing receptor expression to oneelement of the postsynaptic dyad

At the ribbons of rod bipolar cell terminals, two postsyn-aptic elements, the AII dendrites and the varicosities ofS1/S2 amacrine cells come in close contact with eachother, forming a dyad, opposing the same presynapticribbon (Fig. 1C). It is reasonable to think that, becausethese two elements are so close to each other, only the highresolution of EM would be able to distinguish the locationof subcellular structures, such as receptor subunits, to oneor the other postsynaptic cell. However, this is not an easytask, because with immuno-EM images, it is difficult toassign cell types to the two postsynaptic components (forreview, see Thoreson and Witkovsky, 1999). Furthermore,EM studies are laborious and time consuming. It is usu-ally impossible to reconstruct an entire cell, thus limitingthe number of synapses investigated.

Fig. 8. GluR4 is not on either S1 or S2 varicosities. A,B: Triplelabeled image (A) shows an example of a Neurobiotin injected S1 cellvaricosity (green) and an S2 varicosity (upper left inset, green), GluR4labeling in red, and AII dendrites labeled by calretinin in blue. BothS1 and S2 varicosities are outlined in B, and there are no GluR4clusters enclosed. In contrast to Figure 7C,D, signal averaging anal-

ysis by using S1 varicosities for sampling shows no central peak of theaverage GluR4 signal (D) that correlates with the peak of the averagesignal of the S1 varicosities (C). E,F: The same analysis performedwith S2 varicosities for sampling yields similar results. Scale bar � 5�m in B (applies to A,B).

240 W. LI ET AL.

Fig. 9. The glutamate receptor subunit �1/2 is associated with S1,S2, or both. Whole-mount retinae were triple labeled with antibodiesto GluR �1/2 (red), serotonin (green), and PKC (blue). A: In thisexample, focused at the bottom of inner plexiform layer (IPL), thedendrites of S1/S2 form a dense matrix (green) with rod bipolarterminals (blue) located in the holes surrounded by S1/S2 varicosities.Arrows mark three examples of varicosities. B: From the same frameas A, �1/2 clusters (red) are located close to the rod bipolar terminals(blue). The arrows in A are repeated, now pointing to three large �1/2clusters. C: Most of the �1/2 clusters (red) are also on the varicosities(green), as seen in the examples marked by the same three arrows asin A and B. D: When focused higher in the IPL, some varicosities

(green) from the descending dendrites of S1 or S2 contact the axons ofrod bipolar cells (blue) above the branching point. Here, a largevaricosity contacts a rod bipolar axon, and there is an intervening �1/2cluster (arrowhead). E: A three-dimensional reconstruction shows�1/2 clusters (red) on large varicosities (green) at the bottom of IPL.An example is shown by an arrow. Notice the varicosity high in theIPL (arrowhead) also contains a �1/2 cluster. F: The same reconstruc-tion with a see-through view of the rod bipolar terminal demonstratesthe location of �1/2 clusters where varicosities contact a rod bipolarterminal (arrow) or an axon (arrowhead). Scale bar � 10 �m in C(applies to A–C), 2 �m in D.


Fig. 10. The varicosities of Neurobiotin-injected S1 cells contain�1/2 where they make contacts with rod bipolar terminals. A: Intriple-labeled tissue, the green signal of a Neurobiotin-injected S1cell, as well as coupled S1 cells, are visible together with rod bipolarterminals labeled with a PKC antibody (blue) and the red punctatestaining of the �1/2 antibody. Several �1/2 clusters are on the vari-cosities of S1 cells (marked by arrows) where they contact rod bipolar

terminals. The inset at upper right is a higher magnification image ofa single rod bipolar terminal (blue) with two S1 varicosities and theirassociated �1/2 puncta marked by arrowheads. B: For clarity, thegreen S1 signal has been removed to aid in visualization of the �1/2clusters at contacts between S1 varicosities and rod bipolar terminals.Scale bar � 10 �m in B (applies to A,B), 2 �m in the inset of B (appliesto both insets).

Fig. 11. �1/2 is not on AII dendrites. A: A triple-labeled imageshows the �1/2 clusters (red) are not on AII dendrites (calretinin,blue). A lone dendrite of a Neurobiotin-injected S1 cell traverses theimage, and an arrow marks a single varicosity that contains �1/2staining. B: With removal of the S1 dendrite, the staining of a �1/2cluster that lies on the S1 varicosity is more visible (arrow). C,D: Thecolocalization �1/2 clusters (red) with an injected S1 varicosity

(green), but not the surrounding AII dendrites (blue) is shown moreclearly in this three-dimensional reconstruction from a series of con-focal images. E–G: Signal averaging analysis was performed on A,with sampling boxes centered on �1/2 clusters (E). A sharp peak of theaverage �1/2 signal (F) corresponds to the “hole” of the average AIIsignal (G), indicating that the two signals are anti-correlated. Scalebars � 10 �m in B (applies to A,B), 5 �m in D (applies to C,D).

By using immunocytochemistry and confocal micros-copy, we were able to pinpoint glutamate receptor sub-units specifically to one element of the dyad. First, multi-ple specific cell markers such as calretinin (AIIs),serotonin (S1/S2), and PKC (rod bipolar cells) labeled allthe pre- and postsynaptic structures. Together with Neu-

robiotin injections of individual S1 and S2 amacrine cells,we were able to solve the problem of cell type identifica-tion. In addition, this method provides a large number ofpotential synapses for analysis.

However, does light microscopy have enough resolutionto distinguish receptor subunit expression between the

Fig. 12. �1/2 subunits are not components of glutamate receptorson S2 varicosities. A: In a triple-labeled image generated with anti-bodies to �1/2 (red) and PKC (blue) and a Neurobiotin-injected S2 cell(green), several S2 varicosities are marked by arrows. B: There are nored �1/2 clusters associated with the S2 varicosities, which have beenremoved (arrows). C: Neighboring S1 and S2 cells were injected withNeurobiotin. The S1 dendrite is coded in green, and S2 dendrite is

coded in yellow. The varicosities of each that contact different rodbipolar terminals (blue) are outlined by white ellipses. D: The S1 andS2 dendrites are removed, revealing that �1/2 clusters are localized tothe ellipse surrounding the S1 varicosity, with none in the S2 vari-cosity. Scale bars � 5 �m in B (applies to A,B), 5 �m in D (applies toC,D).

244 W. LI ET AL.

two postsynaptic cells? The limit of resolution of a lightmicroscope is a little less than 0.5 micron, and a well-tuned confocal microscope is slightly better, approxi-mately 200–300 nm. Receptor clusters routinely exceededthis limit with dimensions in the range of 0.5–1.0 �,whereas the S1 varicosities were much larger, up to 5 � inlength.

Our confocal images showed that the glutamate recep-tor subunits are preferentially expressed on one side of thedyad but not on the other side. The ability to assign theexpression of receptor subtypes to one cell is probably dueto the extent of the GluR staining to a distance of at leastseveral hundred nanometers from the contact point of thetwo postsynaptic membranes. In fact, as shown by theimmuno-EM images from previous studies (Brandstatteret al., 1997; Qin and Pourcho, 1999a), the immunolabelingof glutamate receptor subunits is not restricted to the sitewhere the two parts of the dyad meet at the ribbon releasesite. Rather, the staining extends within the host cell, inmany cases over 500 nm. If we consider that the gluta-mate receptor staining in the high magnification confocalimages represents the signal summation of whole receptor

clusters demonstrated in the EM images, then the centerof each cluster could be 300–400 nm away from the borderof the two postsynaptic components. Thus receptor stain-ing on either postsynaptic cell may be distinguishable atthe resolution limit of the light microscope. This is alsoconsistent with the labeling of kinesin, which is a markerfor presynaptic ribbons (Fig. 5). The presynaptic ribbonextends closely to the presynaptic membrane, yet there isa visible separation of the kinesin-labeled structure andthe postsynaptic receptors. This finding also suggests thatpostsynaptic receptor labeling extends for several hun-dred nm from the ribbon.

Still, with this limited resolution, several criteria mustbe met before one may conclusively localize subunit ex-pression to one cell type. Because of the intimate contactbetween two elements of dyad, it is not fully convincing toassign the location of a given receptor subunit by simplyshowing the overlap of receptor labeling and one postsyn-aptic cell. Rather, for each receptor subunit, we also pro-vide evidence that labeling is absent from the other part ofthe dyad (Figs. 5, 8). Additional evidence may be obtainedby examining receptor labeling in relation to the entiredendritic matrix. Thus in addition to obtaining high-power images at the limit of confocal resolution, we exam-ined the distribution of receptor clusters with the den-dritic arborization of the candidate cell from a relativelylow-power view to include many cells. This method pro-vided confirmatory evidence for the localization of GluR2/3and GluR4 to AII amacrine cells (Figs. 2B, 7A). For exam-ple, in Figure 7A, it is evident that GluR4 red puncta arefound along the length of AII dendrites, especially wherethey curve around and contact rod bipolar terminals. Forcomparison, no such overlap exists in the staining patternfor �1/2 in the AII dendritic matrix (Fig. 11A).

Application of signal-averaging software

Immunocytochemical studies have a shared problem ofmaking sure the labeling actually belongs to a certainneuron. As pointed out by Malsand and Raviola (2000),most researchers have relied on simple inspection, whichis not satisfactory. For instance, without careful analysis,an image like Figure 1A could easily lead to the conclusionthat GluR2/3 is found on rod bipolar terminals, the pre-synaptic cell. In this research, we took high-resolutionconfocal images and extracted, aligned, and averaged re-peated structures within the image. This signal-averagingtransfers the impression of an observer into a quantitativeplot, derived from a large group of like samples. Thesurface plots of the three color channels reflect, on aver-age, the spatial relationships between the structures rep-resented in each channel. An earlier version of this soft-ware was used to show that connexin 36 gap junctionsoccur at a specific structure, dendritic crossings betweenAII amacrine cells (Mills et al., 2001).

There are three critical issues regarding this analysis.The first concerns selection criteria. For a triple-labeledimage like Figure 4B, clusters of GluR2/3 that are associ-ated with rod bipolar terminals were chosen as samplingpoints, because the question of interest is whether theclusters are on the AII dendrites at contact points with rodbipolar terminals. This selection criterion sets the AIIsignal free from sampling constraints. However, the situ-ation is different for the S1/S2 varicosities. Unlike the AIIcell, in which case the entirety of the dendrite around arod bipolar terminal could be synaptic, S1/S2 varicosities

TABLE 1. Other Glutamate Receptor Subunits Tested

GluRSubunit OPL IPL Associated with Rod Bipolar Terminals?

GluR1 Yes Yes NoGluR2 Yes Yes Much less signal than GluR2/3GluR6/7 Weak Yes Big puncta associated with rod bipolar axons in

sublamina b; No convincing association withrod bipolar terminals

KA2 Weak Yes Little signal at sublamina 5 of IPL, and noconvincing association with rod bipolarterminals

Fig. 13. Signal averaging analysis of relationships between �1/2and S1/S2 varicosities. A: Fourteen S1 varicosities were sampled andaveraged, yielding a characteristic central peak. B: A central peak in�1/2 signal is correlated with the central peak of the average S1varicosity (A). C: Twenty S2 varicosities were averaged. D: In contrastto S1 sampling results (B), the �1/2 signal shows no central peak thatcorrelates with the average S2 varicosity.


are the only synaptic sites (Sandell et al., 1989; Massey etal., 1999). Even though the entire matrix could be labeled,it is not suitable for the analysis, because there is toomuch irrelevant signal, which would greatly increase thebackground noise. Therefore, we visualized the individualvaricosities more clearly by injecting S1 and S2 amacrinecells with Neurobiotin and chose individual varicosities asthe sampling points to detect the relative distribution ofGluR clusters.

The second concern is contamination from random over-lap, yielding false-positive results in the colocalizationanalysis. Because the neuronal networks in this workhave a high density, they automatically have a high like-lihood of overlapping with other structures. To control forrandom overlap, we rotated the other channels out ofphase from the sampling channel and generated addi-tional intensity plots (i.e., Fig. 4F). The sampling boxeswere not moved, but the intensity plot changed from anunmistakable central peak to a random, nearly flat distri-bution. This change indicates no spatial relationship inthe control image.

Third, because we were comparing expression patternsof GluRs between two cells that come in close contact at apostsynaptic dyad, we examined, for each glutamate re-ceptor subunit, its spatial relationship to both elements ofthe dyad. This approach was necessary to confirm thatexpression of a given GluR was localized to one side of thedyad but not the other. This analysis served as a negativecontrol for all GluRs tested. Furthermore, the same com-parison was applied for the S1 vs. S2 amacrine cells,which indicated that they have different receptor compo-nents (Fig. 13).

Glutamate receptor components of AIIamacrine cells

AII amacrine cells are one of the most heavily studiedneurons in the retina, and extensive research has uncov-ered many details such as their neuronal connections,light responses, and receptive field organization (for arecent review, see Bloomfield and Dacheux, 2001). Rela-tively less information is known about the AII synapticconductance and receptor properties. Boos et al. (1993)recorded from AII amacrine cells in a rat slice preparationand found evidence for the existence of AMPA/kainatereceptors but not NMDA receptors. This finding is consis-tent with the lack of NMDA receptor labeling associatedwith rod bipolar terminals (Fletcher et al., 2000). How-ever, there was no further differentiation between AMPAand kainate types.

The present results indicate the glutamate receptors onAII amacrine cell are probably of the AMPA type. AmongGluR1-4, we found the presence of GluR2/3 and GluR4 onAII cells by signal-averaging analysis of high-resolutionconfocal images. This finding is consistent with the EMresults of Qin and Pourcho (1999a,b), who also foundGluR2/3 and 4 on AII amacrine cells of the cat retina. Insome preliminary experiments, an antibody to GluR2showed relatively little labeling on AII dendrites (data notshown) compared with the GluR2/3 antibody and GluR2was not expressed by AII amacrine cells in the cat retina(Qin and Pourcho, 1999a). This finding suggests that acombination of GluR3 and GluR4 subunits form the glu-tamate receptors on AII amacrine cells. This same combi-nation of subunits was found in endbulb synapses of theanteroventral cochlear nuclei, in which rapid neurotrans-

mission is required for the auditory system (Petralia et al.,2000). We did not find evidence for the presence ofGluR6/7 and KA2 subunits on AII dendrites, althoughthey have been reported at unidentified processes postsyn-aptic to rod bipolar cells in rat retina (Brandstatter et al.,1997).

In light of the above results and those of Qin andPourcho (1999a,b), which suggest that rod bipolar to AIItransmission is mediated by AMPA receptors, it is sur-prising that CNQX did not block the center response ofdark-adapted AII amacrine cells in the rabbit retina(Bloomfield and Xin, 2000). Furthermore, AII responses toexogenous glutamate were blocked by quinoxalindiones(Boos et al., 1993). This finding suggests that the lightresponse of dark-adapted AII amacrine cells is not entirelymediated synaptically by means of the rod bipolar cell. Isit possible that a portion of the response comes from conebipolar cells by means of gap junctions with the AII net-work? Under dark-adapted conditions, this mechanismwould suggest that rod signals enter the cone pathways bymeans of rod/cone coupling or by means of occasionaldirect cone bipolar contacts with rods (Vardi et al., 1998;Soucy et al., 1998). The participation of NMDA receptorsat this synapse is unlikely because NMDA receptor sub-units are not associated with rod bipolar terminals(Fletcher et al., 2000), AIIs are not responsive to NMDA(Boos et al., 1993), and the light response was only par-tially reduced by an extremely high dose of the potentNMDA antagonist MK-801 (Bloomfield and Xin, 2000).Bloomfield and Xin (2000) reported that AII responsescould in fact be enhanced by CNQX. A similar effect wasreported from primate AII amacrine cells (Stone et al.,1997). This could occur if the reciprocal GABA input fromS1/S2 amacrine cells is blocked by CNQX, potentiating asmall residual input (Bloomfield and Xin, 2000). It willtake some further experimentation to sort through thesecomplex and interdependent pathways. Nevertheless, it isclear that GluR2/3 and GluR4 are expressed on AII ama-crine cells, forming AMPA-type receptors that may func-tion in the rapid transmission of rod signals.

GluR� subunits are expressed by S1amacrine cells

The presence of GluR� subunits on S1 varicosities isunexpected. To date, they remain an “orphan” glutamatereceptor, because there is no evidence of their formingchannels with other known GluR subunits, or for theirbinding to normal GluR agonists (for review, see Holl-mann and Heinemann, 1994). However, accumulating ev-idence indicates that � subunits play important functionalroles in the central nervous system. GluR�2 knockoutmice showed impairment of motor coordination, Purkinjecell synaptic formation, and cerebellar long-term depres-sion (Kashiwabuchi et al., 1995). These functional deficitscorrespond well with previous studies localizing �2 expres-sion to cerebellar Purkinje cells (Araki et al., 1993; Mayatet al., 1995). Moreover, �2 subunits with a mutation of oneamino acid (Lurcher mutation, Zuo et al., 1997) form con-stitutively active cation channels in both the oocyte ex-pression system and cerebellar Purkinje cells. This findingindicates that �2 subunits could be competent to formhomomeric channels. Therefore, the failure to observechannel activity in expression systems may reflect theinability to activate the channel properly rather than theirlack of channel forming ability. It has been proposed that

246 W. LI ET AL.

an additional undetected subunit may be needed for li-gand binding and channel gating (Heintz and DeJager,1999). Functionally, a study of GluR�2 channels with theLurcher mutation in an expression system identified �2 asan AMPA/kainate-like channel potentiated by extracellu-lar calcium (Wollmuth et al., 2000).

Besides the cerebellum, GluR� subunits have also beenfound in many sensory structures in the brain, includingcochlear nuclei (Petralia et al., 2000), hair cells in theauditory and vestibular systems (Safieddine and Went-hold, 1997), and the olfactory bulb (Mayat et al., 1995).Therefore, it is perhaps not surprising to find �1/2 sub-units expressed in the retina. Brandstatter et al. (1997)first reported the labeling of �1/2 in the IPL of rat andmouse retina, but they found it selectively distributed onAII amacrine cells, which is different from the presentresults but could perhaps result from a species difference.We found that �1/2 is expressed exclusively by S1 varicos-ities, but not by AII or S2 amacrine cells. Moreover, thedistribution of �1/2 differentiates between S1 and S2 var-icosities. The absence of S2 labeling needs to be inter-preted carefully to rule out false negatives. However,when neighboring pairs of S1s and S2s were dye injectedso that the S1s could provide a positive control in the sametissue, only S1 varicosities expressed �1/2 subunits. Thedifferential expression between S1 and S2 cells is consis-tent with the previous finding that the synaptic distribu-tion of GluR� subunits is very selective (Landsend et al.,1997). The lack of physiological information on the S1 celland the GluR� channel itself makes matching of functionand anatomy difficult. However, it will be intriguing tostudy the physiology of the S1 amacrine cell, which may inturn provide more information about � subunits.

S1 and S2 amacrine cells express differentglutamate receptors

The need for two similar amacrine cells such as S1 andS2 with the same reciprocal connections to the rod bipolarterminal has always been a puzzle. In this work, we haveshown that S1 and S2 amacrine cells may be differenti-ated by their glutamate receptor expression. The presenceof different glutamate receptor subunits may in turn affectsuch functional properties as desensitization, time course,or sensitivity for these two amacrine cell types (DeVries,2000). A question that remains unanswered is the identityof glutamate receptor subunits on S2 cells. We found that�1/2 is preferentially expressed by S1 but not S2 varicos-ities. Evidence has also been presented that S2 amacrinecells do not express GluR2/3 and GluR4. Furthermore, itseems that GluR1 is not associated with dyad synapses atthe rod bipolar terminals (Table 1, data not shown). Thusthe S2 glutamate receptor is unlikely to be an AMPA type.

For kainate receptor subunits, Brandstatter et al.(1997) showed the presence of GluR6/7 and KA2 at one ofthe processes postsynaptic to a rod bipolar terminal butcould not identify the cell type. We also observed punctatestaining for GluR6/7 and KA2 at the bottom of IPL. Nev-ertheless, there was no consistent pattern, and the num-ber of clusters was not large enough to correlate with thenumber of S2 varicosities (Table 1, data not shown). Othersubunits such as GluR5 and KA1 still need to be examinedwhen more specific antibodies are available. However, it isunlikely that NMDA receptors carry the S2 signal. NMDAreceptors are not associated with rod bipolar dyads(Fletcher et al., 2000), and NMDA antagonists did not

block the light response of rat A17 amacrine cells (Mengerand Wassle, 2000). Finally, we note that certain metabo-tropic glutamate receptors, such as mGluR2/3, are ex-pressed by A17 amacrine cells in the rat retina (Cai andPourcho, 1999). Their exact function is unknown, but inthe olfactory bulb, mGluR2 coexists with ionotropic gluta-mate receptors (Hayashi et al., 1993). In summary, wehave shown for the first time that S1 and S2 amacrinecells may be differentiated by the expression of postsyn-aptic glutamate receptors, despite their morphologic sim-ilarities and reciprocal connections with rod bipolar ter-minals.

CONCLUSION

We have shown that the two postsynaptic elements atrod bipolar dyads express different glutamate receptors.AII amacrine cells express GluR2/3 and 4, typical of anAMPA receptor serving fast synaptic transmission. In con-trast, there was no evidence for these subunits on theother postsynaptic element, an S1 or S2 amacrine cell. Theglutamate receptors of S2 amacrine cells were not identi-fied but S1 amacrine cells expressed the �1/2 subunit.Thus each different amacrine cell in this study expressesdifferent glutamate receptors. Different glutamate recep-tors produce specific postsynaptic responses, and theseresponses must contribute to the functional diversity ofretinal neurons. It has long been a puzzle why there aretwo reciprocal amacrine cells with such similar morphol-ogy and connections. The above evidence, that S1 and S2express different glutamate receptors, indicates theyprobably play different functional roles in the rod path-way.

ACKNOWLEDGMENTS

This work was supported by an unrestricted grantfrom Research to Prevent Blindness to the Departmentof Ophthalmology and Visual Science. E.B.T. was sup-ported by an NEI Vision Training Grant. S.C.M. is agrateful recipient of fellowships from HWK (Hanse-Wissenschaftskolleg) and Research to Prevent Blind-ness. S.C.M. is the Elizabeth Morford Professor of Oph-thalmology.

LITERATURE CITED

Ames AD, Nesbett FB. 1981. In vitro retina as an experimental model ofthe central nervous system. J Neurochem 37:867–877.

Araki K, Meguro H, Kushiya E, Takayama C, Inoue Y, Mishina M. 1993.Selective expression of the glutamate receptor channel delta 2 subunitin cerebellar Purkinje cells. Biochem Biophys Res Commun 197:1267–1276.

Baylor DA. 1987. Photoreceptor signals and vision. Proctor lecture. InvestOphthalmol Vis Sci 28:34–49.

Bloomfield SA. 1996. Effect of spike blockade on the receptive-field size ofamacrine and ganglion cells in the rabbit retina. J Neurophysiol 75:1878–1893.

Bloomfield SA, Dacheux RF. 2001. Rod vision: pathways and processing inthe mammalian retina. Prog Retin Eye Res 20:351–384.

Bloomfield SA, Xin D. 2000. Surround inhibition of mammalian AII ama-crine cells is generated in the proximal retina. J Physiol 523:771–783.

Boos R, Schneider H, Wassle H. 1993. Voltage- and transmitter-gatedcurrents of all-amacrine cells in a slice preparation of the rat retina.J Neurosci 13:2874–2888.

Brandstatter JH, Koulen P, Wassle H. 1997. Selective synaptic distribu-


tion of kainate receptor subunits in the two plexiform layers of the ratretina. J Neurosci 17:9298–9307.

Cai W, Pourcho RG. 1999. Localization of metabotropic glutamate recep-tors mGluR1alpha and mGluR2/3 in the cat retina. J Comp Neurol407:427–437.

DeVries SH. 2000. Bipolar cells use kainate and AMPA receptors to filtervisual information into separate channels. Neuron 28:847–856.

Fletcher EL, Wassle H. 1999. Indoleamine-accumulating amacrine cellsare presynaptic to rod bipolar cells through GABA(C) receptors.J Comp Neurol 413:155–167.

Fletcher EL, Hack I, Brandstatter JH, Wassle H. 2000. Synaptic localiza-tion of NMDA receptor subunits in the rat retina. J Comp Neurol420:98–112.

Hayashi Y, Momiyama A, Takahashi T, Ohishi H, Ogawa-Meguro R,Shigemoto R, Mizuno N, Nakanishi S. 1993. Role of a metabotropicglutamate receptor in synaptic modulation in the accessory olfactorybulb. Nature 366:687–690.

Hecht S, Shlaer S, Pirenne MH. 1942. Energy, quanta, and vision. J GenPhysiol 25:819–840.

Heintz N, DeJager PL. 1999. GluR delta 2 and the development and deathof cerebellar Purkinje neurons in lurcher mice. Ann NY Acad Sci868:502–514.

Hollmann M, Heinemann S. 1994. Cloned glutamate receptors. Annu RevNeurosci 17:31–108.

Kashiwabuchi N, Ikeda K, Araki K, Hirano T, Shibuki K, Takayama C,Inoue Y, Kutsuwada T, Yagi T, Kang Y, et al. 1995. Impairment ofmotor coordination, Purkinje cell synapse formation, and cerebellarlong-term depression in GluR delta 2 mutant mice. Cell 81:245–252.

Landsend AS, Amiry-Moghaddam M, Matsubara A, Bergersen L, Usami S,Wenthold RJ, Ottersen OP. 1997. Differential localization of deltaglutamate receptors in the rat cerebellum: coexpression with AMPAreceptors in parallel fiber-spine synapses and absence from climbingfiber-spine synapses. J Neurosci 17:834–842.

Li W, Trexler EB, Massey SC. 2001. Distribution of ribbon synapses in therabbit retina. Invest Ophthalmol Vis Sci 42:S670.

Masland RH, Raviola E. 2000. Confronting complexity: strategies for un-derstanding the microcircuitry of the retina. Annu Rev Neurosci 23:249–284.

Massey SC, Mills SL. 1999. Antibody to calretinin stains AII amacrine cellsin the rabbit retina: double-label and confocal analyses. J Comp Neurol411:3–18.

Massey SC, Mills SL, Marc RE. 1992. All indoleamine-accumulating cellsin the rabbit retina contain GABA. J Comp Neurol 322:275–291.

Massey SC, Liu S, Zhang J, Li W. 1999. Confocal analysis of the rodpathway in the rabbit retina: connections with the matrix of indoleam-ine accumulating amacrine cells. Invest Ophthalmol Vis Sci 40:S437.

Mayat E, Petralia RS, Wang YX, Wenthold RJ. 1995. Immunoprecipita-tion, immunoblotting, and immunocytochemistry studies suggest thatglutamate receptor delta subunits form novel postsynaptic receptorcomplexes. J Neurosci 15:2533–2546.

Menger N, Wassle H. 2000. Morphological and physiological properties ofthe A17 amacrine cell of the rat retina. Vis Neurosci 17:769–780.

Michaelis EK. 1998. Molecular biology of glutamate receptors in the cen-tral nervous system and their role in excitotoxicity, oxidative stress andaging. Prog Neurobiol 54:369–415.

Mills SL, O’Brien JJ, Li W, O’Brien J, Massey SC. 2001. Rod pathways inthe mammalian retina use connexin 36. J Comp Neurol 436:336–350.

Muresan V, Lyass A, Schnapp BJ. 1999. The kinesin motor KIF3A is acomponent of the presynaptic ribbon in vertebrate photoreceptors.J Neurosci 19:1027–1037.

Nelson R, Kolb H. 1985. A17: a broad-field amacrine cell in the rod systemof the cat retina. J Neurophysiol 54:592–614.

Petralia RS, Rubio ME, Wang YX, Wenthold RJ. 2000. Differential distri-bution of glutamate receptors in the cochlear nuclei. Hear Res 147:59–69.

Pin JP, Duvoisin R. 1995. The metabotropic glutamate receptors: structureand functions. Neuropharmacology 34:1–26.

Qin P, Pourcho RG. 1996. Distribution of AMPA-selective glutamate re-ceptor subunits in the cat retina. Brain Res 710:303–307.

Qin P, Pourcho RG. 1999a. AMPA-selective glutamate receptor subunitsGluR2 and GluR4 in the cat retina: an immunocytochemical study. VisNeurosci 16:1105–1114.

Qin P, Pourcho RG. 1999b. Localization of AMPA-selective glutamatereceptor subunits in the cat retina: a light- and electron-microscopicstudy. Vis Neurosci 16:169–177.

Raviola E, Dacheux RF. 1987. Excitatory dyad synapse in rabbit retina.Proc Natl Acad Sci U S A 84:7324–7328.

Safieddine S, Wenthold RJ. 1997. The glutamate receptor subunit delta1 ishighly expressed in hair cells of the auditory and vestibular systems.J Neurosci 17:7523–7531.

Sandell JH, Masland RH. 1986. A system of indoleamine-accumulatingneurons in the rabbit retina. J Neurosci 6:3331–3347.

Sandell JH, Masland RH, Raviola E, Dacheux RF. 1989. Connections ofindoleamine-accumulating cells in the rabbit retina. J Comp Neurol283:303–313.

Soucy E, Wang Y, Nirenberg S, Nathans J, Meister M. 1998. A novelsignaling pathway from rod photoreceptors to ganglion cells in mam-malian retina. Neuron 21:481–493.

Stone S, Buck SL, Dacey DM. 1997. Pharmacological dissection of rod andcone bipolar input to the AII amacrine in Macaque retina. InvestOphthalmol Vis Sci 38:S689.

Sterling P. 1998. Retina. In: Shepherd GM, editor. The synaptic organiza-tion of the brain. New York: Oxford. p 205–253.

Strettoi E, Dacheux RF, Raviola E. 1990. Synaptic connections of rodbipolar cells in the inner plexiform layer of the rabbit retina. J CompNeurol 295:449–466.

Strettoi E, Raviola E, Dacheux RF. 1992. Synaptic connections of thenarrow-field, bistratified rod amacrine cell (AII) in the rabbit retina.J Comp Neurol 325:152–168.

Thoreson WB, Witkovsky P. 1999. Glutamate receptors and circuits in thevertebrate retina. Prog Retin Eye Res 18:765–810.

Vaney DI. 1986. Morphological identification of serotonin-accumulatingneurons in the living retina. Science 233:444–446.

Vardi N, Morigiwa K, Wang TL, Shi YJ, Sterling P. 1998. Neurochemistryof the mammalian cone “synaptic complex.” Vision Res 38:1359–1369.

Wassle H, Grunert U, Chun MH, Boycott BB. 1995. The rod pathway of themacaque monkey retina: identification of AII-amacrine cells with an-tibodies against calretinin. J Comp Neurol 361:537–551.

Wollmuth LP, Kuner T, Jatzke C, Seeburg PH, Heintz N, Zuo J. 2000. TheLurcher mutation identifies delta 2 as an AMPA/kainate receptor-likechannel that is potentiated by Ca2�. J Neurosci 20:5973–5980.

Young HM, Vaney DI. 1991. Rod-signal interneurons in the rabbit retina:I. Rod bipolar cells. J Comp Neurol 310:139–153.

Zhou C, Dacheux RF. 2001. Voltage- and ligand-gated currents of AIIamacrine cells in the rabbit retina. Invest Ophthalmol Vis Sci 42:S674.

Zuo J, De Jager PL, Takahashi KA, Jiang W, Linden DJ, Heintz N. 1997.Neurodegeneration in Lurcher mice caused by mutation in delta2 glu-tamate receptor gene. Nature 388:769–773.

248 W. LI ET AL.

Documents

Glutamate receptors at rod bipolar ribbon synapses in the rabbit retina