THE JOURNAL OF COMPARATIVE NEUROLOGY 316:42’2446 (1992)

Polyaxonal Amacrine Cells of Rabbit Retina: PA2, PA3, and PA4 Cells. Light and Electron Microscopic Studies With a Functional Interpretation EDWARD V. FAMIGLIETTI Department of Anatomy and Lions’ Sight Centre, University of Calgary, Faculty of Medicine, Calgary, Alberta, Canada, T2N 4N1

ABSTRACT Polyaxonal (PA) amacrine cells are a new class of amacrine cell bearing one to six branching, axon-like processes that emerge from the cell body or dendritic trees within 50 pm of the cell body. These slender processes of uniform caliber branch at right angles and in many respects closely resemble the axons of Golgi type I1 cells found elsewhere in the brain. Of the four types of polyaxonal amacrine cell that we have recognized in rabbit retina, two have been described previously in brief communications. One of these, the PA1 amacrine cell with its interstitially displaced cell body, located in the inner plexiform layer (IPL), has been analyzed extensively in two preceding reports. This paper concerns PA2, PA3, and PA4 amacrine cells. Type 2 polyaxonal (PA21 amacrine cells, identified in Golgi preparations of whole-mounted rabbit retinas, have smaller cell bodies (9-14 pm) than the other three types and these are always displaced to the ganglion cell layer (GCL) or the inner border of the inner plexiform layer (IPL). The dendritic fields of PA2 cells are also smaller than those of other PA amacrine cells, and most of their sparse dendritic branching is narrowly stratified at the border of strata (S) 4 and 5 . Some members of this more heterogeneous amacrine cell “type” are bistratified, however, and more highly branched with terminal branches rising to end in S1. PA2 amacrine cells bear a scattering of small dendritic spines and may also exhibit complex dendritic appendages arising at the ends of terminal branches in proximal regions of the dendritic tree. PA2 cells emit one to three axons from the proximal dendritic tree, and about half of the cells bear a single axon. Type 3 polyaxonal (PA3) amacrine cells resemble PA1 cells in the large size of their cells bodies (11-16 pm) and dendritic fields, but differ from the latter in placement of cell bodies, which is in the GCL, and dendritic and axonal stratification, which is multistratified, ranging from S4 to S1, with a concentration in S3 or S4 and a variable contribution to S1. PA3 cells differ from PA1 cells in several other respects, including dendritic branching which occurs at higher frequency and is biased toward temporal retina, and in characteristic bristling dendritic spines, clustered in the intermediate regions of the dendritic tree, that are longer, more variable in appearance and more tightly clustered than the small, uniform spines of PA1 cells that are clustered on proximal dendrities. Type 4 polyaxonal (PA41 amacrine cells, characterized in Golgi and immunocytochemical preparations, have large cell bodies located in the amacrine cell sublayer of the inner nuclear layer and dendritic branching in S1; the straight dendritic branches are highly varicose bearing a sparse complement of spines. A single axon may be usual in PA4 cells. The axon is apparently confined to S1 for most of its course, but may contribute a few branches to S3 and S5 of the IPL. The axon initial segment of PA4 cells, like that of PA1 cells, exhibits one or more swellings along its course. Electron microscopic study of a PA2 amacrine cell reveals largely if not exclusively postsynaptic dendrites and an axon initial segment with ultrastructural features differentiating it from the dendrites. Certain similarities are noted in a comparison of the axons of polyaxonal amacrine cells and both developing and regenerating axons of the central nervous system.

Accepted September 19,1991,

o 1992 WILEY-LISS. INC.

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA

423

Hence polyaxonal amacrine cell axons may be viewed in terms of a flexible genetic repertoire governing the form and function of neuritic outgrowth. The likely neurotransmitters of polyaxonal amacrine cells are considered and the correspondances suggested are: PA1 cells-epinephrine; PA2 cells-somatostatin and/or substance P; PA3 cells-unknown; PA4 cells-dopamine. It is proposed that the family of polyaxonal amacrine cells plays a “fast” neuromodulatory role in light adaptation, mediating information at mesopic levels about changing, uniform background illumination. Key words: very wide-field amacrine cells, dopaminergic amacrine cells, axons, dendrites, neuromodulation, light adaptation, vision

Polyaxonal amacrine cells have been previously identified in rabbit retina (Famiglietti and Siegfried, ’80; Famiglietti, ’81b, Famiglietti, ’89b). Type 1 polyaxonal (PA11 amacrine cells have been described in detail in terms of their morphology and stratification in the first paper of the present series (Famiglietti, ’92a). The “axons” of polyaxonal amacrine cells, although they might be regarded by some as extensive dendritic (or somatic) appendages, were described as morphologically indistinguishable from the locally branching axons of Golgi type I1 neurons elsewhere in the brain (cf. Famiglietti and Peters, ’72). It was noted that the definition of “amacrine cells,” a term proposed by Cajal (1893), does not exclude the presence of “short axons,” as exemplified his description of “associational amacrine cells” in bird retina (Cajal, 1896). Following the description of their morphology (Famiglietti, ’92a), it was tentatively concluded that type 1(PA1) and other polyaxonal retinal neurons with similar “axons” should be regarded as amacrine cells. In the second paper of this series, describing the size and distribution of PA1 polyaxonal amacrine cells (Famiglietti, ’92b), additional cytological evidence was provided from Nissl preparations, supporting the amacrine cell nature of their large, but very lightly staining cell bodies in the inner plexiform layer (IPL), very different from those of occasional ganglion cells displaced to a similar interstitial position. Quantitative studies were also made of PA1 amacrine cells in Golgi preparations and their presumed counterparts in Nissl preparations. It was noted that similar cells have been identified in primate retina (Dacey, ’89), and a quantitative comparison was made with rabbit PA1 cells. In this, the third paper of the series, three other known types of polyaxonal amacrine cell are described, one of which (PA21 was recognized early (Famiglietti, ’81b), and the other two (PA3 and PA4). more recently (Famiglietti, ’89b, and unpublished observations). A retinal neuron similar to the PA3 cell was illustrated in a review of primate retina (Rodieck, ’881, and a neuron resembling the PA4 cell has recently been described in cat retina (Dacey, ’90a). In this report, the dendrites and initial segment of the axon of a PA2 amacrine cell are examined by electron microscopy, providing further evidence upon which to base the distinction between these two types of processes of polyaxonal amacrine cells. It is suggested that polyaxonal amacrine cells are polarized, with a recipient dendritic tree and transmissive axons, the latter conducting signals centrifugally from the perisomatic region of the dendritic tree, in a manner analogous to conventional axons. The probable neurotransmitters of polyaxonal amacrine cells are considered, including dopamine (in PA4 cells), and the likely modulatory role of polyaxonal amacrine cells in the retinal processing of visual signals is discussed.

MATERIALS AND METHODS Golgi impregnation Polyaxonal amacrine cells were impregnated by a modified Golgi-Kopsch-Colonnier method, described elsewhere (Famiglietti, ’85a; Famiglietti, ’92a). Briefly, eyes were removed from adult, pigmented rabbits under Fluothane anesthesia. They were rapidly hemisected; the vitreous humour was removed, and the eyecup was everted in dilute, buffered glutaraldehyde (0.5% to 1%). The retina was then freed from the pigmented epithelium, mounted on a glass slide, and fixed in more concentrated aldehydes (2% to 5%). In some cases, the retina was removed in oxygenated physiological salt solution, and fixed afterward in 3% glutaraldehyde on a glass slide. After 1 to 2 hours of fixation, retinas were immersed in one of several glutaraldehyddichromate mixtures, ranging in concentration from 2% to 5% glutaraldehyde, and 2% to 8%potassium dichromate, for at least 3 days, followed by 1%silver nitrate for 2 days. Retinas were dehydrated and cleared, and mounted whole and flat on a glass slide, either in DPX medium or in Epon-Araldite epoxy resin.

Immunocytochemistry Tyrosine hydroxylase immunocytochemistry was performed as follows. The anterior segment of enucleated rabbit eyes was removed with the vitreous humor, and the resulting eyecup was fixed for 2 hours at 4°C in phosphatebuffered 4% paraformaldehyde (pH 7.21, after which it was washed for 1 hour at 4°C in phosphate-buffered saline (PBS) containing 5% sucrose. The retinas were dissected from the eyecups on ice and then incubated in 10% normal goat serum (ngs) for 45 minutes at room temperature. The free-floating retinas were incubated for 6 days at 4°C in an antibody to tyrosine hydroxylase (W. Tank) diluted 1:500 in 1%ngs and 0.3% Triton-X, followed by 18 hours in a goat-anti-rabbit secondary antibody in the same diluent. A third, peroxidase-antiperoxidase (PAP), antibody was applied, and diaminobenzidine (DAB) was used as the chromogen. Retinas were flattened on a glass slide, mounted in PBS-glycerine (1:2), and coverslipped and the coverslips were sealed with nail polish.

Light microscopy Retinas were examined in a light microscope with conventional Koller illumination, and at the highest resolution with a Zeiss Neofluar oil immersion long-working distance objective, ~ 1 0 0(N.A. = 1.0) with a calculated depth of focus of 0.4 km. Drawings of cells and their processes were made with a camera lucida (Zeiss drawing tube) at initial magnifications ranging from x 100 to ~ 2 , 0 0 0 .

424

Most polyaxonal amacrine cells identified were completely impregnated in their dendritic fields and dendritic appendages. In most cases, the axonal fields of these cells were judged to be incomplete, however. No attempt was made to correct for shrinkage in the tissue, in calibrating magnification. Shrinkage was estimated to range from 5% to 25%. Relatively uniform shrinkage of individual retinas Golgi-impregnated using the sandwich technique is unavoidable, and is in fact desirable. When the retina is fixed to a slide, shrinkage perpendicular to the retinal layers can be 90%, preventing accurate determination of dendritic stratification, which is essential in the structural characterization of inner retinal neurons from a functional perspective. The issue of shrinkage in such Golgi preparations is discussed in the previous paper (Famiglietti, '92b) and elsewhere (Famiglietti, '85a).

E.V. FAMIGLIETTI retinas were made at x10 with the aid of projection microscopy; coordinates of reference objects were recorded and the drawing was transferred to millimeter graph paper. Points separated by as much as 15 mm were positioned within 0.5 mm of their true location. Locations of cell bodies, mapped onto the graphed retina, were specified in terms of linear distance dorsal (-1 or ventral (+) to the visual streak (dVS), either as the shortest distance to the midvisual streak, or as great circle distances, when samples are located significantly away from the projection of a midvertical meridian passing through the center of the optic nerve head.

Electron microscopy

Golgi impregnation of polyaxonal amacrine cells was performed as described above. After removal from the silver nitrate solution, sectors of retina containing cells for elecComputer graphic reconstruction tron microscopic study were immersed in 1% osmium The operation of the computer-controlled microscope tetroxide for 5-20 minutes, until some observable darkensystem, developed by the author and used in this study, is ing of the tissue had taken place. In addition, a modified described elsewhere (Famiglietti, '85a). The overall accu- gold-toning procedure was applied which has proved generracy of positioning logged points along the dendritic tree in ally unreliable in (aldehyde) Golgi-Kopsch-fixed material, three-dimensional Cartesian coordinates is conservatively because it removes the silver, but leaves very little gold in estimated to be 0.8 km. Thus in principle, about 20 levels its place. Nevertheless, in our particular example, processed could be distinguished in the IPL, which ranges in thick- generally as described by Fairen and colleagues ('771, many ness from 14 to 17 pm in the central retinal regions of these of the processes could be found by careful mapping of the light micrographic representation onto the thin sections, preparations. Stereo pairs were also constructed to allow visualization obtained as described below. It was a distinct advantage for of dendritic stratification,and to confirm the narrowness or the electron microscopical analysis, of course, not to have breadth of stratification. Images were tilted in a receding the silver deposits obscuring the internal structure of the plane directed upward and parallel to the horizon, in order polyaxonal cell's dendrites and axons. For electron microscopy, a relatively isolated polyaxonal to prevent excessive, obscuring superposition of stratified dendrites. The two views were then rotated about a vertical amacrine cell was selected, photographed in the wholeaxis in the plane of the paper, at 6" with respect to one mounted retina, and drawn at x 2,000 with the aid of a Zeiss another, a rotation determined empirically to give the best (camera lucida) drawing tube. A piece of the plasticembedded retina containing the cell was then cut out and stereoscopic effect. mounted with epoxy upon a blank chuck. It was trimmed, Measurement and sectioned at 1 pm until the section plane was parallel Cell body and dendritic field sizes of cells in Golgi to the retinal layers and the cell body and primary dendrites preparations were determined in the following manner. Cell bodies were drawn with the camera lucida at their , maximum contour in the plane of the retina at ~ 2 , 0 0 0and Fig. 1. Camera lucida drawings: type 2 polyaxonal (PA21 amacrine were enlarged to 4,000 diameters. Dendritic fields were drawn similarly at x 150, as convex polygons connecting the cells of rabbit retina. A and D are taken from the same retina, while B and C are from two other retinas. Dendritic branching (thicker outermost tips of terminal dendrites. Areas were traced processes) and axonal branching (thinner processes, labelled a at their twice on a digitizing tablet and averaged. Occasional pairs origins) are sparse. Dendritic trecs are believed to be completcly of measurements differing by more than 1% were per- impregnated, while most of the axonal processes here and in Figs. 2 and formed again. Diameters of cell bodies and dendritic fields 4-8 are presumed to be incomplete. In the more complete examples, it is were calculated from area measurements as diameters of evident that the axonal fields are much larger than the dendritic fields. Terminal dendritic branches are irregular in length, and short terminal circles of equivalent area. sometimes adorned with complex appendages, are present in Locations of the cell bodies of identified cells were branches, the proximal and intermediate regions of the dendritic tree (see Fig. 2). determined on a Zeiss rotary microscope stage, and esti- Axons (a) range in number from one to three, all originating proximally. mated to the nearest 0.05 mm by reading vernier scales Rarely, segments of dendrites take on a thinner and smoother axonal calibrated in tenths of millimeters. The center of the visual appearance (a') (see Fig. 3). Dendrites cross infrequently, but axons streak was estimated to within 0.05 mm, by checking and freely cross dendrites, and each other, as is unavoidable with rightestimating the minimum dendritic field diameters of previ- angle branching. The large cell body size of C is shared with other amacrine cells from that retina. Its larger-than-expected ously studied cells, either starburst amacrine cells (Famigli- polyaxonal dendritic field size may be due to individual variation coupled with etti, '85a), or ganglion cells, or else by examining the differences in tissue shrinkage; there is apparently some heterogeneity unstained ganglion cell bodies with asymmetric contrast among PA2 cells, however. All here are unistratified in substratum 501 illumination, and reading the coordinates of the center of or at the stratum (S1415 border, except for the end of one distal dendrite maximum ganglion cell density along the visual streak. in C that rises to terminate in stratum 1(indicatedby a l), and the PA2 in D which is bistratified with branches additionally in S1 (indicated This line, representing the midvisual streak, usually bro- cell by 1s) (see Fig. 4). The terminal dendritic branchingof D is clearly more ken or curved, due to the retinal distortion resulting from profuse than that of the other examples. Cells here and in Figure 5 are flattening the quasi-hemispherical retina, was plotted on a oriented with the visual streak parallel to the top of the page. Location: retinal map drawn to scale. Accurate drawings of the dVS = (A: 0.01, B: +2.5), (C: +5.6), (D: +8.5). Scale bar = 500 km.

425

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA

I

/

I I

D

Figure 1

Figure 2

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA

427

were in view. The block was then sectioned at about 80 nm until the cell and its dendrites were cut through (about 120 sections). The serial sections were viewed and photographed at ~ 3 , 6 0 0or ~ 7 , 4 0 0in a Philips 300 electron microscope. Montages were made for sections 20 through 59 at x 8,800 magnification and traced on a single sheet of paper. A final “maximum contour,” two-dimensional projection drawing was made from this composite drawing (Fig. 14B).

RESULTS Displacement of cell bodies The cell bodies of all PA2 and PA3 cells thus far identified are displaced to the ganglion cell layer (GCL), or else next to it in sublamina b of the IPL. As previously reported, the cell bodies of PA1 amacrine cells are generally displaced interstitially to the IPL (Famiglietti, ’81b), although they may sometimes lie in the inner nuclear layer (INL) or GCL (Famiglietti, ’92a). Thus, cell body displacement from the “normal” position of amacrine cell bodies seems to be a common feature of polyaxonal amacrine cells. Moreover, the displacement of PA2 and PA3 cell bodies in the GCL and their relatively large size (Fig. 10) raise the possibility of their confusion with ganglion cells. On the other hand, PA4 amacrine cells, to the extent that they can be identified with tyrosine hydroxylase-immunoreactive (presumed dopaminergic) amacrine cells (Figs. 8, 91, have cell bodies almost invariably positioned in the INL.

PA2 polyaxonal amacrine cells Light microscopy. The type 2 polyaxonal amacrine cell, PA2 (Figs. 1 4 ) , branches narrowly in sublamina b of the IPL, at the border of strata 4 and 5, and is sometimes partly bistratified, with a few dendritic branches in stratum 1. The dendritic branching of PA2 cells is more irregular and more heterogeneous than that of PA3 or PA1 cells, and occurs at a slightly higher frequency than the branching of PA1 cells. The number of terminal branches ranges from 10 to more than 20 (Fig. 1). Most dendrites take a relatively straight course, and terminal dendrites often bend at an angle of 45” to 90” 10-40 pm from the end. Some terminal dendrites take a long, unbranched course, extending radially 20-40 pm from the cell body to the dendritic field perimeter.

Fig. 2. Camera lucida drawing: detail of the PA2 amacrine cell of Figure 1C.The relatively thick, tapering dendrites branch at a narrow angle and give rise to daughter branches of smaller caliber. The dendrites stand in contrast to the much thinner, smooth axon (a)which branches at a wider angle, producing daughter branches of a caliber equal to the parent branches. One axonal branch reverses direction 180”, a not uncommon feature of axonal branching shared with PA1 cells (Famiglietti, ’92a). Small dendritic spines (small arrowheads), similar to those which form clusters on the dendrites of PA1 cells, are scattered along the proximal dendrites. Short dendritic branches terminate in the midregion of the dendritic field, some bearing complex dendritic appendages (large arrowheads). The initial segment of the axon lacks the swellings common to the initial segments of PA1 and PA4 cells (cf. Figs. 8 and 9). The uniform caliber of slender axonal branches is interrupted by small swellings, the boutons en passant. The arrow indicates a region of the dendritic tree, depicted in Figure 3, in which an axon-like transformation of a dendritic segment appears to take place (a‘), as it runs in parallel with a conventional dendrite. Small arrows indicate that the impregnation of axons is presumed to be incomplete, and that these processes continue past the apparent terminations. Scale bar = 100 pm.

Fig. 3. Camera lucida drawing: detail of the PA2 cell in Figure 2, showing co-stratification and contact with a wide-field type b cone bipolar cell axon terminal. The large arrow points toward the proximal dendritic tree of the PA2 cell (cf. Figs. 2 and lC). Of the two crossing dendrites, the thinner one on the right (a‘) mimics an axon with a small swelling en passant (small arrowhead), but resumes its dendritic character at the bottom of the figure, heralded by a small spine (double arrowhead). In two places (small arrows) the left-hand dendrite is in contact with the narrowly stratified axon terminal of a wide-fieldtype b (Wb) cone bipolar cell, represented as a filled contour. The cell body and dendrites of the Wb bipolar, lying in deeper planes of the retina, are drawn in outline. Scale bar = 100 pm.

Other dendrites originating proximally may take a shorter, more curved trajectory, and give rise to a cluster of short and somewhat curved terminal branches and long appendages (large arrowheads, Fig. 2). In other cases, branching frequency may be higher in some regions of the dendritic tree, both proximally and distally, particularly in examples which give rise to a significant contribution to stratum 1 (Figs. 2D, 4). PA2 amacrines exhibit both the very small, 1 km spines typical of PA1 amacrines (small arrowheads, Fig. 21, and larger, 3 p,m long spines. Both types of spine are more frequent in the proximal dendritic tree (Fig. 21, as with PA1 cells. In general, however, PA2 dendrites are less spinous than those of PA1 or PA3 cells, and proximal clustering of dendritic spines, typical of PA1 cells, is not obvious. Unique to PA2 cells are long dendritic appendages, 8-14 pm in length, which may occur proximally in the dendritic tree, as noted above, 100-200 pm distant from the cell body, near the ends of short terminal dendritic branches (large arrowheads, Fig. 2). The main dendritic tree of PA2 amacrine cells branches at the border of strata 4 and 5. This is of interest, because all class Ib ganglion cells (Famiglietti, ’871, and wide-field, type b (Wb) cone bipolar cells (Famiglietti, ’81a) branch precisely at this level. This point is illustrated in Figure 3, where end-to-side contact is shown between axon terminal branches of a Wb cone bipolar and a terminal dendrite of a PA2 amacrine (small arrows, Fig. 3). In addition to regions of more complex branching, noted above, this PA2 amacrine exhibits two other features of

E.V. FAMIGLIETTI

428

interest. First, running parallel to the dendrite contacted by the Wb bipolar, is another branch which arises 140 pm from the cell body in axon-like fashion, maintaining a thin caliber over the course illustrated in Figure 3. Just beyond the second dendritic crossing, however, it reverts to a thicker, spine-bearing dendritic morphology to its termination. This is an unusual back-transformation for polyaxonal amacrine cells, but typifies some heterogeneous-longprocess (HLP) amacrine cells (Famiglietti, ’90a). Secondly, one terminal dendritic branch of this cell rises to end in stratum (“1”in Fig. 1). The branching of PA2 amacrine cells in stratum 1 seems to be limited to short terminal dendritic segments, although it is possible that these are incompletely impregnated. The few PA2 cells in our preparations which lie near or in the visual streak (Fig. 1A,B) apparently produce no ascending processes, while those lying further away give rise to one (Fig. 1C) to seven (Fig. 1D) ascending processes, in the latter case constituting one-third of the terminal dendritic branches (Fig. 4). PA2 amacrines have one to three axons which, like those of PA1 amacrines arise from the cell soma or a proximal dendrite, branch at right angles, and exhibit auniform, thin caliber. In contrast to PA3 cells, about half the PA2 cells in this sample have but a single axon (cf. PA4 cells, below). Single axons are a rarity in the larger sample of PA1 cells (Famiglietti, ’92a,b). In all cases observed, at least one radially directed axon gives off a branch or turns at 90” in the middle third of the dendritic field (Fig. 1).It is perhaps more than coincidence that the orientation of the longest axonal segments in all members of our sample are directed dorsalward toward the visual streak and/or medially toward the zero vertical meridian, both from temporal and nasal locations. Occasionally, distal portions of PA2 axons bear spine-like appendages or boutons termineaux (Figs. lB, 13A), not observed on the axons of other types of polyaxonal amacrine cells. The axons of PA2 amacrine cells are not completely impregnated, and therefore a complete description of their distribution around the cell of origin, in particular their maximum extension and their areal extent, as well as the effect of distance from the visual streak on these parameters, is not possible. It is evident, however, from the better impregnated examples, that the axonal tree extends beyond the dendritic tree by a factor of at least 2-4 (Fig. 1A,B). While the branching seems sparser than that of the best impregnated PA1 cells (Famiglietti, ’92a), the extent of axonal processes may not be so different from that of PA1 cells.

PA3 polyaxonal amacrine cells Light microscopy. The third type of polyaxonal cell, PA3 (Figs. 5-71, first described in this report, is as large or larger in its dendritic field than PA1 amacrines and much larger than PA2 amacrines (cf. Fig. 11).The cell bodies of PA3 amacrine cells tend to be elongate or fusiform in shape, and larger than those of PA2 cells, but comparable in size to the cell bodies of PA1 cells (cf. Fig. 10). Cell bodies lie either in the ganglion cell layer or in the innermost portion of the IPL. The dendritic tree of PA3 amacrine cells is multistratified, branching in S1, S2, S3, and at the S4iS5 border. Considerable lengths of dendritic and axonal branching run parallel to the strata, and transitions from one stratum to

another occur over relatively short distances. These are reasons for classifying PA3 amacrine cells as “multistratified,” rather than “diffuse” in their branching. Transitions may occur from inner to outer strata, and back again to inner strata along the course of lengthy dendritic segments. Some branching occurs in stratum 4, and the S4lS5 border, but most is at the a / b sublaminar border (S2/S3). Extensions, particularly from peripheral dendrites, rise up to stratum 2, and terminal extensions project into stratum 1. Back transitions may occur to inner strata (S2-5). The dendritic tree of PA3 amacrine cells is more highly branched than either PA1 or PA2 amacrine cells, averaging 12 terminal dendrites, as compared to 8 for PA1 cells (cf. Famiglietti, ’92a). Some dendrites exhibit a curving trajectory proximally, and some crossing of dendrites occurs in the middle of the dendritic field. Most terminal dendrites are relatively straight, except for their terminal segments which are slightly wavy or gently curved in the plane of the retina. These are generally those parts of the dendritic tree which rise from stratum 3, the main stratum of branching, to terminate in stratum 1 or 2. The shortest terminal branches which arise in proximal and intermediate locations within the dendritic tree are longer than 100 km. In common with the majority of inner retinal neurons, the dendritic trees of PA3 amacrine cells are elongated parallel to the visual streak, and the more so, the closer they lie to the streak. A peculiarity not shared with other polyaxonal amacrine cells or with any other retinal neuron yet characterized is a bias in the dendritic tree toward the temporal retina (Fig. 5 ) . This effect was more pronounced in examples closer to the visual streak (Fig. 5A,B), and was discernable in all cases, lying in both the nasal and temporal retina, except in the dorsal peripheral retina, where a more pronounced dorsoventral bias was observed. Some degree of dorsoventral bias is also observed in an example from the ventral periphery (Fig. 5C). This topographic variation in the precision of dendritic orientation and direction is reminiscent of that reported for dorsally directed, vertical, asymmetrical amacrine cells (Famiglietti, ’89a). PA3 amacrine cells bear no complex dendritic appendages, like those found on PA2 cells. On the other hand, long spines may be found, and these may be clustered, particularly in intermediate regions of the dendritic tree. These longer spines are 4-5 pm in length, are sometimes branched, and in this case have stouter initial segments 0.2-0.25 pm in width. In general these spines have slightly stouter stems and comparable or smaller heads than the spines of PA1 cells (Famiglietti, ’92a). The clustering of these longer spines occurs further out on the dendritic tree than does the clustering of spines on PA1 cells, and thus farther from the origin of the axons. In the example located at dVS = t 4 . 8 5 mm, the clusters lie in a range 125-225 pm distant from the cell body (large arrowheads in Fig. 7). In the dendritic tree of a comparable PA1 cell, clusters of smaller spines lie closer to the cell body, typically about 50 pm distant (Famiglietti,’92a). Dendritic segments exhibiting high-frequency waviness (bounded by pairs of double arrowheads in Fig. 7 ) are typical of these cells and are not artefacts of preparation, since straight dendritic segments belonging to other types of cells are observed adjacent to some of these wavy segments. These wavy regions are also spine-bearing segments and suggest a special synaptic relationship at these sites with as yet unidentified, presumably presynaptic components. The smaller spines found on the dendrites of

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA

429

A

Fig. 4. Computer reconstruction and rotation of the bistratified PA2 amacrine cell of Figure 1D.A The cell is plotted in the orientation in which it was logged. All the axons (a)and except where marked all the dendrites lie in S4iS5. Those that lie in S1 are marked 1, and occasionally these turn back to S4iS5 (5). B: Rotation of -90" around the X axis. Boundaries of the IPL are logged and plotted in rotated

orientation with the dendritic tree as sectored, hexagonal, planar figures. Unfortunately, the retina in this location was wrinkled, producing a curvature which partly confounds the demonstration of histratification. Thus the axonal branches to the left should form a straight,horizontal line, rather than the upward drift which defines the plane of the S4iS5 border. Scale bar = 100 km.

E.V. FAMIGLIETTI

430

PA1 and PA2 amacrine cells are also present on PA3 amacrines, to a varying degree. The PA3 amacrines in our sample have two to four axons which arise from the cell body or proximal dendrites, in the fashion of PA1 and PA2 amacrine cells (cf. Figs. 1and 5 and Famiglietti, ’92a). These axons may branch near their site of origin (Fig. 7), like those of some PA1 amacrine cells. Axons travel in the main stratum of branching, but some also ascend to stratum 1or 2, joining dendrites of the same cells at that level. Due to incomplete impregnation, it is not possible to characterize the axons further. It is expected, however, that they branch like those of the PA1 amacrine cells and extend long distances from the cell body (Famiglietti, ’92a; Dacey, ’89).

PA4 polyaxonal amacrine cells Light microscopy. Thus far we have succeeded in impregnating only a single example of a PA4 amacrine cell with its axon(s) intact (Fig. 8). In this case a single axon emerges from the cell body and bifurcates 3 times at a wide angle, with the result that the distalmost branches head in opposite directions. The axonal branches are very slender and smooth and their fine caliber is interrupted at interval ranging from 10 to more than 100 pm by small boutons en passant. The impregnation of the axonal arborization is incomplete (note small arrowheads, Fig. 81, but the dendritic tree appears to be complete. The varicose dendrites are sparsely branched and irregular in appearance with a wide variation in branching angle. Some of the relatively straight terminal branches are oddly directed back toward the cell body. The dendrites bear an irregular scattering of spines that vary considerably in shape and length. The dendritic and axonal stratification of this example is narrow and strictly confined to the outer part of stratum 1. The relatively large cell body is located in the amacrine cell sublayer of the inner nuclear layer. The form of the Golgi-impregnated PA4 amacrine cell is consistent with that of dopaminergic amacrine cells characterized by tyrosine hydroxylase immunocytochemistry applied to whole-mounted retinas (Fig. 9), in regard to cell body size, dendritic stratification, and dendritic branching pattern. In this material the staining of relatively slender dendrites and much more delicate axons fades rapidly with distance from the cell body. In the larger and more robust cells of peripheral retina, one can occasionally observe evidence of very fine processes emerging from proximal dendrites of dopaminergic amacrine cells (small arrowhead in Fig. 9). In this case a varicosity (large arrowhead) typical of the initial segment of PA1 amacrine cells (Famiglietti, ’92a),and also found in the PA4 amacrine cell (arrowheads in Fig. 81, also marks this slender process as a PA amacrine cell axon. Small clusters of fine boutons can also be seen at intervals in the photomicrograph which is focused at the level of the outer part of stratum 1 (small arrows in Fig. 9). Focusing a t levels progressively nearer the ganglion cell layer reveals occasional scattered boutons in strata 3 and 5, indicating that occasional axonal branches, but not dendritic branches, of PA4, dopaminergic amacrine cells leave stratum 1 and innervate dendrites of unknown origin in these two inner strata (cf. Figs. 15 and 16).

Soma size and dendritic field diameter of polyaxonal amacrine cells Soma size. The cell body size of polyaxonal cells is large compared to that of other amacrine cells, and within 3 mm

of the visual streak on its ventral side, their size is also larger than that of the smallest ganglion cells, when measured in flat, whole-mounted preparations (Fig. 10). This has been well documented for PA1 cells (Famiglietti, ’92b), which are sufficiently numerous in a single preparation to reduce the scatter due to individual variation, which for cell bodies sizes of retinal neurons may be pronounced among unrelated individuals or animals of different strains (Famiglietti, ’85a). In the graph of Figure 10, a straight line fitted to the distribution of PA1 cells is plotted together with 95% confidence limits from the preceding paper Famiglietti, ’92b). In addition, the cell body sizes of ganglion cells in the same retina are also plotted as circles, and as stars in the case of “small tufted” ganglion cells. The PA2 and PA3 cells, plotted on the same graph as diamonds and squares, respectively, are relatively few in number in our preparations and are variable in size, due mainly to the fact that they are scattered among different retinas. Nevertheless, one may conclude from these small samples that the cell bodies of PA3 (and PA4) cells are about the same size as PA1 cells, whereas PA2 cells are generally somewhat smaller at the same retinal locations. Dendritic field diameter. As in the graph of cell body sizes (Fig. lo), a straight line fitted to the distribution of equivalent dendritic field diameters of PA1 cells is graphed in Figure 11,together with 95% confidence limits, from the preceding paper (Famiglietti, ’92b, Fig. 2). PA2 and PA3 cells are again plotted as diamonds and squares with one less PA3 cell, the dendrites of which could not be followed to their termination in a heavily impregnated region. The limited conclusions which may be drawn from this graph are that PA3 (and probably PA4) cells are comparable in dendritic field size to PA1 cells, while PA2 cells have significantly smaller dendritic fields. Evidence concerning their relative sizes in peripheral retina is evidently less certain than that obtained for central retina.

Electron microscopy of polyaxonal amacrine cells One example of a PA2 polyaxonal amacrine cell was selected for electron microscopic study (Fig. 1B). This neuron [with its cell body in the ganglion cell layer and emitting a single axonal process (Figs. 12, 13A)I expresses features most closely associated with ganglion cells. It thus appeared to offer an opportunity to clarify the character of the highly distinct axonal and dendritic processes of polyaxonal amacrine cells and possibly to qualify the classification of such cells as amacrine cells. As mentioned in the

Fig. 5. Camera lucida drawings: type 3 polyaxonal (PA31 amacrine cells of rabbit retina. A-C: Each example lies in a different left retina. As with PA2 cells, all cell bodies are displaced to the ganglion cell layer, but PA3 cells have larger cell bodies and larger dendritic trees which are more highly branched than those of PA2 cells. Numbers of axons (a) range from two to four. Note that most of the branching, including that of the poorly impregnated axons, is directed (or redirected) horizontally and rightward toward the temporal retina (see text). Dendritic stratification appears to be somewhat variable among PA3 cells with most of the branching in the middle strata of the IPL, either S3 or S4. Some dendrites remain in the middle strata, and some rise to stratum 1, remaining there, while others rise and then return to the middle strata. In general, the axons branch at the same levels as the dendrites (see Fig. 7). Cells are oriented with the visual streak parallel to the top of the page. Location: dVS = (A, -0.05, temporal), (B, -0.3, temporal), ( C , f4.85, nasal). Scale bar = 500 km.

C

Figure 5

i

Figure 6

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA methods, gold-toning left very little silver or gold in the tissue as a marker. The very thin and essentially unlabelled axon could not be followed beyond 50 pm of its length (Fig. 13B), to the more distal region of synaptic boutons (Fig. 13A). Similarly, only the proximal half of the dendritic tree was studied, including, however, a few terminal dendritic branches (Fig. 13B). The deimpregnated, faintly stained dendrites were followed discontinuously for a distance of about 100 pm from the cell body (Fig. 13B), beyond which they entered regions overlapping other impregnated cells (Fig. 12). These labelled dendrites could be distinguished from the surrounding neuropil by slightly darker staining and a coarse, more granular appearance (Fig. 14). Some dendrites contained scattered gold particles (Fig. 14a), while others contained 5 particles/lO pm length (Fig. 14b), or less. By comparison to neighboring dendrites, in the interior of deimpregnated dendrites, microtubules are seen to be well preserved, while floccular cytoplasmic matrix is more clumped, and intracellular membrane systems are minimally disrupted by impregnation (Fig. 14). Synapses, and particularly synapses originating from cone bipolar cells, are not as plentiful in stratum (S)5 and at the S4/S5 border as in strata 2-4 (Famiglietti, ’91). Therefore it is not surprising that relatively few synaptic inputs were identified on the dendritic tree of the PA2 cell. In a total of 325 linear pm of reconstructed dendritic length, eight synaptic inputs and a possible ninth were identified. Of the nine inputs, only one derived from a cone bipolar (B in Fig. 13B), and the rest originated from unidentified amacrine cells. Only two inputs occured on primary dendrites. Although no attempt was made to reconstruct and trace the bipolar input to its source, there is a significant possibility that it derives from a Wb cone bipolar cell (Famiglietti, ’81a) (see Fig. 3). A relatively stout mainstem, tertiary dendrite (D) is illustrated in Figure 14a, deimpregnated and containing a few scattered gold particles. It contains a regular array of microtubules and a small cluster of vesicles (v) of the size of synaptic vesicles. At the lower right, a synaptic bouton containing small round vesicles makes a synaptic contact with the dendrite on a shallow conical protrusion or “varicosity” of the dendritic shaft (small arrowhead). In size, shape, electron density, and vesicular content, this bouton resembles the neighboring “Ar” bouton, which forms part of a rod bipolar dyad in adjacent sections. In Figure 14b, the distal segment of a primary dendrite is displayed. It bears an input from an amacrine cell bouton

Fig. 6. Camera lucida drawing: detail of the PA3 amacrine cell of Figure 5C. Differences in caliber, taper, contour, and branching angle between dendrites and axons (a) are the same as noted for PA2 cells. The axons may emerge from dendrites with or without tapering initial segments, and like those of PA2 cells lack the swellings usually present along the axon initial segments of PA1 and PA4 cells (see Fig. 8). Small dendritic spines (small arrowheads), similar to those which form clusters on the dendrites of PA1 cells, are clustered very proximally, even to accumulation on the cell body of this example (small arrowheads). A few long, curved, and occasionally branched dendritic spines with small spine heads are seen proximally, but these are concentrated in clusters in the middle region of the dendritic tree (large arrowheads). These are the most distinctive detailed feature of the dendritic tree. Also conspicuous are dendritic segments in the middle and distal parts of the dendritic tree which exhibit a high frequency of waviness (double arrowheads) that appears not to be a consequence of tissue shrinkage. Scale bar = 100 km.

433

(small arrowhead) in a configuration very similar to that illustrated in Figure 14a. A cluster of small vesicles (vj is present in the postsynaptic varicosity. At the more distal dendritic swelling (upper left in Fig. 14b) a possible synaptic input from an amacrine cell bouton (small arrowhead and ?) exhibits parallel and more electron-dense membranes, but lacks presynaptic or postsynaptic dense material and close apposition of the vesicles with the plasma membrane. The same varicosity exhibits what may be a presynaptic relationship with an unidentified amacrine cell profile (large arrowhead and ?). Vesicles are closely apposed to the plasmalemma, although most of these are larger than conventional synaptic vesicles, and there is a suggestion of postsynaptic density, except that this density lies in a segment of obliquely sectioned membrane. It may be noted that in this proximal dendrite there are many membranous cisterns indicative of smooth endoplasmic reticulum (darker elongated and irregular profiles) as well as vesicles of various sizes and shapes. Thus it is uncertain whether the collection of vesicles in question is an artefactual coalescence of synaptic vesicles adjacent to an “active zone” or simply part of the smooth endoplasmic reticulum. The PA2 amacrine cell’s initial segment of the axon was identified (Figs. 12, 13B, 14c,d) in series of ultrathin sections cut parallel to the dendritic tree. The tangential plane of section affords the opportunity to compare the ultrastructural cytology of the axon initial segment and its dendrite of origin, after the branching, fortunately in a single section (Fig. 14c,d). This comparison is made against the background of ultrastructural studies performed upon axons and dendrites in other parts of the central nervous system. The ultrastructure of the initial segments of Golgi type I1 (short axon) cells is not well defined, because their initial segments are difficult to identify in random ultrathin sections and the ultrastructure is generally obscured in labelled preparations. Thus only a very unequal comparison can be made with the features of the large and well-defined axon initial segments of “long-axon” Golgi type I cells (Peters et al., ’90). The latter exhibit dense undercoating of the plasmalemma, a clustering of microtubules, in contrast to the orderly, parallel microtubular arrays of dendrites, and a higher ratio of neurofilaments to microtubules than exhibited by dendrites of similar cells. In longitudinal section, microtubules sometimes cross from one cluster to another yielding a more disorderly appearance of nonparallel microtubules. Like dendrites, but unlike the axon proper, axon initial segments may contain small clusters of ribosomes. The dendrite from which the PA2 axon arises (D in Fig. 14c) is a little larger in caliber than the axon (A), and more irregular in contour. At two points, including a varicosity, a parallel array of microtubules may be observed (double arrows in Fig. 14c). The internal structure of the axon initial segment is quite different and less replete with organelles. Where the membrane can be seen clearly, there is no unequivocal evidence for undercoating of the plasma membrane (Fig. 14c,d). Microtubules (double arrow in Fig. 14d) are scant and appear disorderly. Solitary neurofilaments (arrow) also seem unaligned and part of a crossed distribution of tubules and filaments. The slender caliber of the initial segment (outlined by arrowheads in Fig. 14d), about 0.16 pm in diameter, will not admit many microtubules, however, and there may be a lower limit below which

A

B

Figure 7

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA orderly arrays of microtubules cannot be formed in a neural process of any sort. Additionally, proximal dendrites leading toward the axonbearing tree were compared with the other two dendritic systems lacking axons, in regard to the ratio of neurofilaments to microtubules. This comparison was prompted by the observation that the dendrites of axotomized lamprey central neurons giving rise to axonal sprouts contained many more neurofilaments than the dendrites of unaffected cells (Hall et al., ’89; see discussion, below). In the short second order dendritic branch (not shown), leading to the axon-bearing tertiary branch (cf. Fig. 13), the number of microtubules was reduced and the number of neurofilaments increased in comparison to the primary dendrite of origin and the dendrites of the other two dendritic systems (Fig. 14a,b).Although continuity could not be traced along the few additional microns to the initial segment because of missing sections, it may be that the majority of neurofilaments in the more proximal dendritic segment were associated with the nearby origin of an axon.

DISCUSSION Character and diversity of polyaxonal amacrine cells The special character of polyaxonal amacrine cells of the retina has been recognized for a decade (Famiglietti and Siegfried, ’SO), together with the fact that, in their diversity, they exceed the boundaries of a single amacrine cell type (Famiglietti, ’81b). The distinctions that separate polyaxonal amacrine cells from retinal ganglion cells with axon collaterals in the retina, and from ganglion cells with axons apparently confined to the retina, have been explained in the first paper of this series (Famiglietti, ’92a). Also described there are the morphological characteristics that separate polyaxonal amacrine cells from an even more diverse group (HLP amacrine cells), also the principal subject of a recently published abstract (Famiglietti, ’90a). The differences between HLP amacrine cells and polyaxonal amacrine cells are clear and important to appreciate. The HLP (heterogeneous-long-process) amacrine cells that most resemble polyaxonal amacrine cells are the HLPa amacrines. They give rise to processes of relatively uniform caliber and slender diameter, in some respects similar to the axons of polyaxonal amacrine cells. The axons or axon-like processes of most HLPa amacrine cells are unbranched. Although they may occasionally arise within 50 km of the cell body, most arise in the distal dendritic tree, either as extensions emerging from the tips of the longest “terminal” dendrites or as daughter branches of preterminal dendritic segments, wholely different in appearance from their sib-

Fig. ?. Computer reconstruction and rotation of the PA3 amacrine cell of Figures 5C and 6. A: The cell is plotted in the orientation in which it was logged. The distal dendrites were truncated to magnify the display and to reduce the confounding factor of imperfect flattening of the retina on the rotated display in B. Numbers and boundary markers next to the dendrites and axons (a) indicate the strata of the IPL in which the various segments travel. Some impression of this multistratified branching can be obtained from rotation of the dendritic tree by -86“ in B, if one ignores the dendrites directed to the top of the page in A that lie in a curved portion of the “flat-mounted” retina. Note in A the tendency for the axons to lie more toward the inner strata than the dendrites, in the proximal and intermediate regions of the dendritic tree depicted here. Scale bar = 100 pm.

435

ling terminal branches. HLPa amacrine cells of rabbit retina thus include the so-called giant amacrine cells characterized by Cajal (1893) in lizard retina as bearing “fine branches which resemble axis cylinders.” Typical of mammalian HLPa amacrine cells is the “dopamine-accumulating” amacrine cell, identified by Dacey (’88) in cat retina, then supposed to be the dopaminergic amacrine cell. That amacrine cell is clearly different from the PA4 amacrine cell and tyrosine hydroxylase-immunoreactive cells described here, as well as the other polyaxonal amacrine cells. More recent evidence (Dacey, ’90a) reveals that it is not the dopaminergic amacrine cell of cat retina (see below). Distinctive common characteristics of polyaxonal amacrine cells can be reduced to three main features: 1) the presence of discrete dendritic trees, terminating distally at the field perimeter in relatively robust processes that are homogeneous in character with proximal dendrites; 2) the exclusively proximal origin of one to six axons within 50 p,m of the cell body; and 3) right-angle branching of axons, which extend well beyond the dendritic tree. It is evident from inspection of polyaxonal amacrine cells that items 1 and 2 above are related. Additionally, “displaced” cell bodies are characteristic of PA1, PA2, and PA3 amacrine cells, although only PA1 cells exhibit significant variability in cell body placement among the layers of the inner retina (Famiglietti, ’92a). The cell bodies of PA4 cells, on the other hand, are apparently placed in the “normal” position of amacrine cells at the inner border of the inner nuclear layer. PA1, PA3, and probably PA4 amacrine cells have cell bodies and dendritic fields of comparable size, while PA2 cells are smaller. Each of the four types of polyaxonal amacrine cell has distinctively different dendritic branching and a different complement of dendritic spines and appendages (cf. Table in Famiglietti, ’92a). If the dendritic contours of PA1 and PA2 cells are the smoothest, those of the PA4 cell are the roughest exhibited by polyaxonal amacrine cells. The irregular dendritic spines of PA4 cells are not as profuse and irregular as those of PA3 cells, nor generally as small as those of PA1 cells. A detailed comparison of axonal branching is not possible in the present material, because axonal impregnation is limited, particularly in the case of PA3 cells. PA1 and PA3 amacrine cells are each distinct cell types and homogeneous within each type. On the other hand, the PA2 cells described here are not homogenous. Our sample is too small to characterize this heterogeneity adequately, but PA2 cells are divisible into at least two types or subtypes, designated PA2 and PA2’, differing mainly in levels of dendritic stratification (Fig. 151, but also in branching pattern (Fig. 1). The most significant differences of functional importance among the four types of polyaxonal amacrine cell thus far identified concern the levels of dendritic stratification (Fig. 15).The processes of PA1 cells (not shown) stratify broadly in the middle of the IPL. They are concentrated in stratum 3, and are almost entirely confined to the region between the two starburst substrata (hatched bands in Fig. 15).PA2 cells, on the other hand, are narrowly stratified, either at the S4iS5 border (unistratified, PA2), or additionally in S1 (bistratified, PA2’) (cf. Figs. 1D and 4). PA3 cells are more broadly stratified than PA1 cells and could be termed “multistratified,” as dendritic segments may remain in a single stratum for long distances. PA3 cells could also

I

Fig. 8. Camera lucida drawing: type 4 polyaxonal (PA4) amacrine cell of rabbit retina. The dendritic tree is well impregnated and bears no distal axon-like processes. A single axon (a) emerges from the cell body and branches several times at right angles, travelling for some distance before the impregnation ceases (small arrows). The relatively large cell body lies in the inner nuclear layer, while the dendrites and single axon branch in the outer part of stratum 1. The cell body and dendritic field perimeter of a neighboring type a starburst amacrine cell are drawn here for comparison of sizes. The axon initial segment, like that of PA1 cells (Famiglietti, '92a1, bears swellings along its course (arrowheads). The boutons en passant, scattered along the axonal branches are smaller in size. It has a sparse, irregular branching pattern, with terminal dendritic branches pointing both toward and away from the cell body. The dendritic contour of this PA4 cell is rough, although the irregular dendritic spines are not profuse. The long arrow is drawn as a perpendicular to a line parallel to the visual streak, and indicates the direction of the latter with respect to this PA4 cell. Location: dVS = +0.2. Scale bar = 100 km.

137

Fig. 9.

Pliotl~tiiicr~~graph [ i f a presumed dopaminergic and PA4

aniacrints rell. stained by tyrosinc hydroxylase ('PHI imniuniiryt ~ i c h e m istry The rclati\dy largr crll body iif' t h i s example from peripheral retina is typical of 'I'H-immunoreactive !IR! cells. A very slrnder process emerges from a proximal drndritr (stiiall arrowhead! and gives

risr t o a swelling !large arrowhead) like t hat found on thr woii initial s t p i e n t of t h e PA4 cell icf. Fig. 8).A frw clusters CII' sinall houtons

justifiably be termed "diffuse," as t h e dendrites, taken together, generally travel in all t h e four outer strata with occasional extensions into S5.Nevertheless, t h e greatest concentration of dendritic extension lies in the middle strata, either S3 o r S4 (Fig. 15). PA4 cells. like 1'AS cells, a r e narrowly stratified (Fig. 15). To t h e extent that PA4 cells can be identified with dopaminergic aniacrine cells. t h e following generalizations can be made. Their dendritic trees a r e unistratified in S1, while their axons, generally co-stratified with t h e dendritic trees, nonetheless give rise to some branches which travel in 53 and finally to the S4,'S5 border (Fig. 15).None o f t h e axons of PAS cells have heen ohserved to leave t h e region of t h e S4iS5 border, however. As a general rule, with t h e exceptions noted, the axons and dendrites of individual polyaxonal amncrine cells are co-stratified. I n this respect, a s well as in several others relating to t h e form of t h e axons and axonal branching, polyaxonal amacrine cells resemble Golgi type I1 cells of t h e brain (e.g.,Famiglietti and Peters, '721.

I'olyaxonal amacrine cells of other mammals I t has been noted elsewhere (Famiglietti, '92a,b) t h a t t h e apparent homologue in primates of t h e PA1 amacrine cell in rabhil retina has heen identified by means of intracellular stajiiing (Dacey, '89). 'I'hc primate homologue of t h e PA3 amacrine cell also appears to have been identified recently in Golgi preparations (Rodieck, ' 8 8 ) . I n t h e latter case, the obvious similarity relates mainly to the characteristic appearance of t h e dendritic spines which a r e relatively long, complex, and clustered i Fig. 6 1.

ismall arrnws!. prc7suinahly representing nodes of intersection of axon5 from distant TH-IR cells. punctuate the neuropil of outer htratuni 1. i n focus here with the unstained cell hodics of amacrine cells lining the hrirdrr [ i f t he ll'13. A f'ew scattered boutons of this t.ypc can be ohhervcd a t different focal planes in strata 3 and 6 . I,ocat ton: tlVS = 9.0 Scale bar = 2 0 bni.

Recently, dopaniinergic amacrine cells, identified hy tyrosine hydroxylase immunocytochemistry, have been stained intracellularly in cat retina (Dacey, '90a). They exhibit a morphology very similar to that of t h e PA4 cell of Figure 8, with a single branching axon t h a t travels for many millimeters in s t r a t u m 1. Similar cells may have been identified in primate retina as well (Dacey, '90b). No homologue of t h e PA2 cell has yet been characterized in a speciw othor than rabbit.

Electron microscopy of polyaxonal amacrine cells T h e limited ultrastructural evidence obtained in this study indicates t h a t t h e dendritic tree of PA2 polyaxonal amacrine cells receives synaptic input primarily from amacrine cells, but also from cone bipolar cells. Dendrites give rise to few if any synapses, even though t h e dendrites o f PA2 amacrine cells evidently contain clusters of vesicles t h a t resemble synaptic vesicles. Starburst amacrine cells also exhibit segregation of outp u t sites (Famiglietti, '83b, ,911, in this case in t h e distal zone of a radial, bifurcating dendritic tree. Although i n those studies, silver-chromate deposits often obscured t h e dendritic ultrastructure, it was often possible to see clusters of vesicles t h e size of synaptic vesicles in t h e most proximal dendrites (cf. Figs. 5, 6, and 12b, in Famiglietti, '91 I. An even more conspicuous segregation of chemical synaptic output in amacrine cells is exhibited by thr: rod ( A 11) aniacrine cell. Electron microscopic study and reconstruc-

438

33

E.V. FAMIGLIETTI 25

20

0

15

10

3

.d

w5

5

1

I

I

I

I

1

- 7 - 6 - 5 - 4 - 3 - 2 - 1 0

I

I

I

1

I

I

I

I

1

1

2

3

4

5

6

7

8

9

Distance from Visual Streak (mm) Fig. 10. Graph of the variation of cell body diameter of polyaxonal amacrine cells with distance from the visual streak. A straight line fitted to the distribution of PA1 cells is plotted together with 95% confidence limits from the previous paper (Famiglietti, '92b). In addition, the cell body sizes of ganglion cells in the same retina as the majority of PA1 cells are also plotted as circles, and as stars in the case

of "small tufted" ganglion cells (dotted line). PA2 and PA3 cells, scattered among seven retinas, are plotted as diamonds and squares. The PA4 cell of Figure 8 is plotted as a filled triangle. The cell bodies of PA3 (and PA4) cells are about the same size as PA1 cells, whereas PA2 cells are generally somewhat smaller at the same retinal locations.

tion of rod amacrine cells from serial sections resolved the ganglion cells (e.g., Nelson et al., '78; Peichl and Wassle, puzzling finding of dendrites in sublamina b of the IPL '81). A potentially large collecting area seems, however, to containing synaptic vesicles but giving rise to no synapses. be mated to a much larger axonal area of distribution, as It was demonstrated that the chemical synaptic output of noted below. Examination of the axon initial segment of the PA2 cell polarized rod amacrine cells occurs only at the distinctive lobular appendages in sublamina a (Famiglietti and Kolb, indicates that cytoskeletal components are organized differ'75). By analogy, the vesicles in the dendrites of PA2 cells ently than in dendrites of the same cell, which contain a could be the result of indiscriminate transport, but synaptic greater concentration and a more orderly array of microtuutilization primarily in the axonal tree. bules. These findings are consistent with the proposal that Presynaptic dendrites of ganglion cells have never been the axonal and dendritic trees of polyaxonal amacrine cells observed in mammals (e.g., Famiglietti, '85b; Freed and are different from one another, both in form and function. Sterling, '881, but in the retinas of some fish, even the Among several alternative models for the functional organiwell-differentiated distal dendrites of some wide-field gan- zation of polyaxonal amacrine cells (Famiglietti, '92a), the glion cells may give rise to synaptic output (Sakai et al., idea has been advanced that polyaxonal amacrine cells (PA1 '86). The evidence of the present electron microscopic study cells in particular) collect input principally within the of a PA2 amacrine cell indicates that, like mammalian dendritic tree and transmit their output much more widely ganglion cells, the dendrites of polyaxonal amacrine cells across the retina by means of their axons (Famiglietti, are primarily the recipients of input. The fact remains, '92b). The ultrastructural evidence obtained from the PA2 however, that this input derives principally from amacrine cell is consistent with this idea, but study of the axon cells, a conclusion similar to that reached concerning inputs terminals is required to put this proposal on a firmer basis. to dopaminergic (presumed PA4) amacrine cells (Dowling Production of axons and flexibility of gene and Ehinger, '781, and these presynaptic cells are likely to expression in development and regeneration include wide-field amacrine cells. This raises the possibility that the receptive field of a polyaxonal amacrine cell is If there is a requirement in vertebrate retina for very much wider than its dendritic field, in contrast to many long-distance lateral communication within the inner ret-

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA

23

1100

900

700

\

\

5 00

300

100 - 7 - 6 - 5 - 4 - 3 - 2 - 1 0

1

2

3

4

5

6

7

8

9

Distance from Visual Streak (mm) Fig. 11. Graph of the variation of dendritic field diameter of polyaxonal amacrine cells with distance from the visual streak. As in Figure 10, a straight line fitted to the distribution of equivalent dendritic field diameters of PA1 cells is graphed together with 95% confidence limits, from the previous paper (Famiglietti, '92b). PA2 and

PA3 cells are again plotted as diamonds and squares, and the PA4 cell as a triangle. PA3 (and probably PA41 cells are comparable in dendritic field size to PA1 cells, while PA2 cells have significantly smaller dendritic fields.

ina, greater than 1 or 2 mm, it may seem curious that amacrine cells rather than ganglion cells would be selected and specialized for this task. It is possible of course that ganglion cells perform such functions as well, either by means of collaterals of axons which ascend to the brain (Marenghi, '01; Dacey, '85), or by associational, monoaxonal ganglion cells wholely contained within the retina (Drager et al., '84). The latter cells are reported to be very rare, however, and the density of the former is apparently low. On the other hand, polyaxonal amacrine cells constitute a family of neurons which shares cytological features with amacrine cells (Famiglietti, '92b), and which populates and covers the retina topographically in a more or less regular fashion (Famiglietti, '92b). Evidence has been provided that some amacrine cells may share a common early developmental course with ganglion cells in mammalian retina (Hinds and Hinds, '78). The proposal advanced by Hinds and Hinds to explain their results is that most of the earliest developing amacrine cells migrate from the ganglion cell layer to the inner nuclear layer. Hinds and Hinds suggest that this process of reverse migration, and its incompleteness in some cases, may account for the presence of displaced amacrine cells. Preceding the reverse migration is a postulated loss of primitive

axons. If this proposal is correct, and if polyaxonal amacrine cells follow such a developmental course, then their axons would not be strictly homologous with the axons of ganglion cells. Hinds and Ruffett ('73) have studied the ultrastructure of developing axon initial segments in mitral cells of the olfactory bulb, which at the earliest developmental stages exhibit certain features in common with the axon initial segments of polyaxonal amacrine cells. At the earliest stages, separation of subnuclear cytoplasm and the initial segment is not well marked. A thick initial segment and even blebs of perinuclear cytoplasm are present. Reduction in numbers of ribosomes, fasciculation of microtubules in the initial segment, and development of a continuous dense undercoating of the plasmalemma occur in sequential stages. The PA2 polyaxonal amacrine cell initial segment appears comparable to an early stage in which microtubules are disorderly but not yet fasciculated. If fasciculation of microtubules facilitates axonal transport (Palay et al., '68), then the blebs or swellings of the initial segments of PA1 and PA4 amacrine cells, may represent "front end" inefficiency in the transport mechanisms of polyaxonal amacrine cells, unmasked in those types which may have more extensive intraretinal axons.

110

Fig. 1%. Photomicrograph of t he PA2 amacrine cell of Figurc 18, clioscn [or electron micrirsccqiic study. A single. very slcnder axun (ax) nrises from B dendrite, about 30 p m from t he cell body, and hranchrs a t the uppcr right-hand ciirnc'r

(arrowbl. T h e

synaplic t i o u L o n s iasttarisksi

'I'hc invcnc of thc, dcvelopinental process of axonal elimination postulated by Hinds and Hinds ( ' 7 8 ) for displaced amacrinr cells. t h r production of new axons, occurs at least in mamrnalian gariglion cells in a regenerative mode stimulated by injury. Axotomized adult hamster ganglion cells, under the influericr of a peripheral nerve graft, can develop a second axon, sprouting from a dendrite (as do most of the axons of polyaxonal amacrine cells), while at the same time maintaining the orignal axon (Cho and So, '89; Fainiglietti et al., '91 1. These regenerating ganglion cells were stained particularly well with a neurofibrillar (reduced silver) method, suggesting the overproduction of cytoskeletal elements, particularly ncwrofilaments which characterized mature axons. This plasticity of mammalian ganglion cells could reflect the exploitation by niechanisms of neural regeneration of a genetic repertoire held in common with developniental mechanisms. This same genetic repertoire may be available to and normally exploited by polyaxonal aniacrine cells. Evidence that polyaxonal amacrine cells also react to injury with a neurite-proliferative response has been provided in a report on tyrosine hydroxylase-immunoreactive amacrine cells of mouse retina, some of which temporarily produce new processes that extend throughout most of sublamina a in response to administration of the neurotoxin l-methyl-4plienyl-1,2,3,6-tetrahydropyridineMPTP (Tatton et al., '901. It i s not yet known whether these processes have the character ofaxons or dendrites. In this connc:ct,ion, the results of in vivo axotoiny on reticulospinal neurons of lamprey is interesting ( I lnll et nl., '891. Normally, the dendrites of these neurons are only postsynaptic. After axotoniy close to the cell body, however, the dendritic tree gives rise to long sprouts which have the characteristics of axons: containing neurofilanients in preference to microtubules and giving rise to synapses. These axlxon-like processes also contain large clusters of vesicles in the range o f sizes occ;ision;illy seen in the dendrites of PA2 amacrine cells (Fig. 14 1. In view of this flexibility of gene expression, stimulated hot h by developmental programs and regenerative signals, the fact that some retinal amacrine cells are capable of

could not hc studicd h r ti.chnical reasons. Its initial s i g n i i m t [ i n box c ) is illustrated in the electron micrographs of Figure 14c and d. Two dendritic segments 1 hoxcd arcah a and h) are illoslriitid 111 F1guI-I. 1 4 1 and b. Also see Fipire 1.X Scale bar = 50 pm.

producing dendritic trees of conventional form, resembling those of typical ganglion cells (Cajal, 18931, renders less surprising the capacity of some to produce axonal arhorizations. It would be interesting to know to what extent the available genetic repertoire of polyaxonal cells may he held in common with ganglion cells (e.g., Hinds and Ilinds, '781, and thus affords common structural solutions iincluding axonal branching, cytoskeletal features, and possibly transport processes) to distinctly different problems of information processing in the visual system.

Neurotransmitters of polyaxonal amacrine cells r 1

1he neurotransmitters of polyaxonal am;icrinr? cr:lls are likely t o differ with each type. The most certain identification is that of PA4 cells with dopaminergic amacrine cells, characterized by means of tyrosine hydroxylase immunocytochemistry (Figs. 8, 9).In rabbit retina, they are the same in cell body position and size, dendritic stratification, dendritic branching pattern, and axonal morphology. In cat retina, this relationship has been demonstrated directly by double labelling (Dacey, '90a, see above). In primate retina, amacrine cells with thc variable cell body position of PA1 cells appear to contain c a t c ~ h o h mines, for they can be revealed by formaldehyde-induced fluorescence (Ehinger and Falck, '69). These can also be faintly stained with an antibody to tyrosine hydroxylase, and 35% of their cell bodies are located in the IPL (Mariani and Hokoc, '88). Similar cells in rat retina that also branch in the middle of the IPL hut rarely exhibit displaced cell bodies (Nguyen-Leposet al., '831 are pheny1ethanoleamine~N-methyltransferase (€'NM'I')-immunoreactive, and may therefore contain epinephrine (Hadjiconstantinou et al., '84; Versaux-Botteri et al., '86).I n primate retina, arnacrine cells which branch principally in S3 have been shown to contain suhstnnce P, and some of these have cell bodies displaced to the ganglion cell layer (Erecha, '8.3).In rahhit, however, substance P-immunoreactive cells provide very few branches to S3 (Farniglietti et al., 'go), and are clearly not the same a s PA1 ainacrine cells.

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA

441

A

\

Fig. 13. Camera lucida drawing and electron micrographic reconstruction of the PA2 amacrine cell of figures 1B and 12. A: Detailed camera lucida drawing revealing few dendritic spines and a single axon (ax)which, in unusual fashion, gives rise to several spine-like appendages, as well as the usual boutons en passant. The cell body lies at the border of the ganglion and inner plexiform layers and the dendritic and axonal branches all lie in the outer part of stratum 5. An axonal branch is apparently unirnpregnated (large arrow). B: Electron micrographic

reconstruction of the serially sectioned dendritic tree in which reconstructed segments are filled in and the contours of connecting segments are drawn in dotted lines. Arrows indicate the locations of sparse inputs, all but one (B,for cone bipolar cell) from amacrine cells, and one possible output. Question marks indicate questionable synapses (see Fig. 14b). Circled lower case letters indicate regions illustrated in the montage of Figure 14.Scale bar = 100 pm.

PA2 amacrine cells may contain the neuropeptide somatostatin (SS),for they resemble neurons characterized in rabbit retina by SS immunocytochemistry in at least three respects: cell body position in the ganglion cell layer, the presence of axons, and probably dendritic stratification (Sagar et al., '86). SS-immunoreactive (SS-IR) processes stratify in S1 and even more prominently in the outer part of S5, as viewed in retinal cross sections (Sagar et al., '86).

The slight discrepancy between our assessment of PA2 cell stratification at the S4lS5 border and the stratification of SS-IR cells could be due to the greater difficulty in our whole-mounted material of establishing the inner boundary of the IPL with the ganglion cell layer. The axon terminal of the wide-field type b cone bipolar cell, found to co-stratify with one example of a PA2 amacrine cell (Fig. 3), appears to be a reliable marker for the S4/S5 border, however.

Figure 14

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA The curious absence of SS-IR cells in all but the peripheral rim of dorsal rabbit retina and the presence of axons throughout the dorsal retina (Sagar, ’87; Rickman and Brecha, ’89) indicate that the axons of SS-IR cells travel long distances of many millimeters across the retina. This is a dramatic example consistent with the morphological character of polyaxonal amacrine cells (Famiglietti and Siegfried, ’80; Famiglietti, ’81b, ’89b), and with functionally differentiated axonal and dendritic trees (Famiglietti, ’92a,b) (see below). SS-IR cells have also been identified in human retina where they have been called “associational ganglion cells” (Sagar and Marshall, ’88). If, as seems likely, SS-IR cells are polyaxonal amacrine cells, then this is a denomination that incorrectly applies the term ganglion cell to neurons which are amacrine cells on cytological grounds, according to our analysis of Nissl-stained retinas (Famiglietti, ’92b) and of electron micrographs of PA2 cells (Fig. 14). The possibility that PA2 cells could be included among substance P-immunoreactive (SP-IR) amacrine cells needs to be addressed, because the latter, like SS-IR amacrine cells, also stratify principally in S1 and S5 in rabbit retina (Famiglietti et al., ’80). SP-IR cells are more diverse than SS-IR cells appear to be, with cell bodies on both sides of the IPL, and two or three distinct morphologies among the SP-IR amacrine cells (Famiglietti et al., ’80). One of these has its cell body in the ganglion cell layer and its principal branching in S5. No evidence for contributions to S1 from this cell type could be found. It is therefore possible that one or both of these peptides are contained in PA2 cells, separately or co-localized in the same cells. The neurotransmitter of PA3 cells is presently unknown. If the poorly impregnated axons of our PA3 cells do follow

Fig. 14. Electron micrographs of serial sections from the deimpregnated, gold-toned PA2 amacrine cell of Figures lB, 12, and 13. a: A scattering of small gold particles and a darker cast mark the deimpregnated PA2 mainstem dendrite (D). In comparison to neighboring dendrites, microtubules are seen to be well preserved, while floccular cytoplasmic matrix is more clumped, and intracellular membrane systems are minimally disrupted by impregnation. It contains a regular array of microtubules and a small cluster of vesicles (v) of the size of synaptic vesicles. At the lower right, a synaptic bouton containing small round vesicles makes a synaptic contact with the dendrite on a shallow conical protrusion or “varicosity” of the dendritic shaft (small arrowhead). In size, shape, electron density, and vesicular content, this bouton resembles the neighboring (Ar)bouton, which forms part of a rod bipolar dyad with a dark process of a rod amacrine cells (RA) in adjacent sections. The synaptic ribbon (r) in the rod bipolar (RB) cell profile points to the two postsynaptic profiles. b The distal segment of a primary dendrite (D) bears an input from an amacrine cell bouton (small arrowhead) very similar to that in a. A cluster of small vesicles (v) is also present in the postsynaptic varicosity. At left, a possible synaptic input from an amacrine cell bouton (small arrowhead and ?) occurs on a varicosity which in turn exhibits what may be a presynaptic relationship with an unidentified amacrine cell profile (large arrowhead and ?) (see text). c: The tangential plane of section affords the opportunity to compare the the axon initial segment (A) and its dendrite of origin (D) in a single section, as they course on opposite sides of a Muller cell process (MI. The dendrite is a little larger in caliber than the axon, and more irregular in contour. At two points, a parallel array of microtubules may be observed (double arrows). The internal structure of the axon initial segment is quite different and less replete with organelles. d: The plasma membrane of the fragment of initial segment, about 0.16 pm in diameter, is outlined by arrowheads. Where the membrane can be seen clearly, there is no unequivocal evidence for undercoating of the plasmalemma. Microtubules (double arrow) are scant and appear disorderly. Solitary neurofilaments (arrow) also seem unaligned and part of a crossed distribution of tubules and filaments. Scale bars = 0.5 pm.

443

the dendritic stratification of the dendritic trees and represent the bulk of PA3 cell volume, then the matching neurotransmitter-selective staining pattern would have to be multistratified, and concentrated in strata 1 and 3 or 314.

Neurochemical stratification and function of modulatory amacrine cells All of the neurotransmitters mentioned here as likely candidates mediating the synaptic action of polyaxonal amacrine cells, whether catecholamines or neuropeptides, fall into the category of neuromodulators, and this correlation may be an indicator of the functional role that polyaxonal amacrine cells play in the retina. It is probably no coincidence that the dendritic and axonal stratification of polyaxonal amacrine cells and amacrine cells containing the neuromodulatory agents mentioned here, as well as others, including substance P (Famiglietti et al., ’80; Brecha, ’83), occurs in strata 1,3,and 5 or 415. These three zones of the IPL are marginal to the two sublaminae of the IPL (a and b), in which the signals for OFF and ON responses are processed (Famiglietti and Kolb, ’76). In rabbit retina, the middle regions of sublaminae a and b are marked by the processes of starburst (cholinergic) amacrine cells (Famiglietti, ’83a; Famiglietti and Tumosa, ’871, apparently modulators of a different kind, increasing the intensity-response function of many ganglion cell types (Ariel and Daw, ’82).Starburst amacrine cells are presynaptic principally to ganglion cells (Famiglietti, ’83b; Famiglietti, ’91). In contrast, dopaminergic amacrine cells are presynaptic only to other amacrine cells (Dowling and Ehinger, ’781, particularly the rod amacrine cells (Pourcho, ’82; Voigt and Wassle, ’87). Extrapolating to other polyaxonal amacrine cells, one may therefore suppose that their modulatory influence upon ganglion cells is exerted in general at the remove of one or two synapses, and perhaps mediated by narrow-field, broadly stratified or multistratified amacrine cells (see Famiglietti, ’92a). The effects of dopamine and its agonists and antagonists on retinal ganglion cells are well studied in rabbit retina (Jensen and Daw, ’84, ’86).Dopamine appears to act in an inhibitory manner, to fashion the surrounds of rod (MI) amacrine cells (Daw et al., ’89) that distribute rod bipolar input to both ON-center and OFF-center ganglion cells (Famiglietti and Kolb, ’75). In fact, D, receptor antagonists reduced or abolished the surround responses of both ON and OFF ganglion cells (Jensen and Daw, ’84, ’86). Dopamine is also modulatory, altering the spontaneous activity and depressing the sensitivity of ganglion cells. Thus it seems that an apparently selective effect, alteration of the center-surround balance of ganglion cells, and a relatively nonselective effect, the depression of ganglion cell sensitivity, may both contribute to the mechanisms of light adaptation (cf. Daw et al., ’89). If both “modulatory” and “selective” effects are mediated by a single type of amacrine cell, as seems possible in the case of dopaminergic amacrine cells, then the classification of individual amacrine cells by functional type may be more difficult than expected. The demonstration in the present studies, however, of a set of wide-field, polyaxonal amacrine cells built on a common morphological plan, when coupled in the future with secure knowledge of their synaptic transmitters, may serve to illuminate a range of common functions, now only dimly understood in relationship to dopaminergic amacrine cells. The range of polyaxonal amacrine cell types and their axonal stratification

E.V. FAMIGLIETTI

444

a b PA2 PA2’ Fig. 15. Diagram of polyaxonal amacrine cell stratification in the inner plexiform layer (IPL). The IPL is divided into two sublaminae, a and b, containing the neural circuitry of OFF and ON pathways, respectively. Each of the five strata may in turn be subdivided into semiarbitrary substrata (a,p, y) to serve as a proportional scale of measure for the level and extent (widthinarrowness) of dendritic stratification in the IPL. The entire thickness of the IPL is about 25 pm. Hatched bands indicate the levels of branching of type a and type b starburst (cholinergic) amacrine cells, which serve as marker cells and “substrata of reference” for the middle of sublaminae a and b, respectively. For each type of polyaxonal cell the levels of dendritic and axonal branching are indicated by dark bands. The width of the bands is

PA3

PA4

approximately proportional to the fractional length of dendrite contributed to each level. Data from Famiglietti (’92a) for PA1 cells is represented here. The stratification of PA1 cells, also illustrated in Figure 9 of Famiglietti (’92a), is focused mainly at the S2iS3 border, with some extension up through S2p, and rarely to S1, and extension downward to 4p. Two subvarieties of PA2 cell are recognized, one of which is more highly branched and bistratified (PA2’).The multistratified branching of PA3 cells appears to be variable, and two variants are illustrated here. In the case of the PA4 cell, contributions to S3 and 55 appear to be made only by the axons, based upon the evidence of TH immunocytochemistry.

a b

Fig. 16. Summary drawing of the neural organization of polyaxonal amacrine cells and related neurons in rabbit retina. The wide-field (presumed blue cone) bipolar (CB) cells (Wa and Wb) and stratified amacrine cells (e.g., A) constitute the neuronal substrate for the interaction between the bi-sublaminar organization of ON and OFF pathways, embodied here in the rod amacrine cell (€%&, and a tristratified organization of modulatory polyaxonal amacrine (PA) cells. Individual PA cells are presumed to influence certain co-stratified retinal neurons over a very wide retinal area by means of their branching axons (a). PA4

cells are apparently dopaminergic amacrine cells, known to make contact with RA cells, and PA2 cells may contain somatostatin and/or substance P. PA1 cells, which may contain epinephrine, generally have their cell bodies in the middle of the IPL, and branching in stratum 3, at the a/b sublaminar border. Modulatory influences acting on RA cells, and perhaps other amacrine and bipolar cells, thus indirectly affect the output of ganglion cells, such as the large-bodied, class I, type b (Ib) ganglion cell. Crosshatched bands mark the stratification of starburst amacrine cells in the middle of sublaminae a and b, as in Fig. 15.

POLYAXONAL AMACRINE CELLS OF RABBIT RETINA suggest that these functions may be distributed in parallel, and in some instances convergently, upon a range of ganglion cell types.

A speculation on the role of polyaxonal amacrine cells in rabbit vision A particularly notable aspect of the dendritic stratification of PA2, PA3, and PA4 cells is their co-stratification in S1 and S4f5 with the axon terminals of wide-field cone bipolar cells (Fig. 31, which branch at these two levels (Fig. 16; and Famiglietti, ’81a) and with class Ib ganglion cells (Fig. 16).It has been argued elsewhere that wide-field cone bipolars are connected to blue cones (Famiglietti, %la, ’gob), and there is evidence that blue cones provide excitatory input to X-ON ganglion cells (Caldwell and Daw, ’78), yielding more sustained responses than longer wavelength inputs (Famiglietti, unpublished observations). Detection of airborn predators, as small objects on a blue background, would seem to be an important factor in the survival of rabbits. Evidence was discussed above that dopaminergic amacrine cells may regulate the sensitivity of ganglion cells as well as center-surround balance in the receptive field through rod amacrine cells. An intriguing observation has recently been made on the distribution of the cell bodies and the dendritic trees of SS-IR amacrine cells in rabbit retina. Nearly all are located in the skyward-looking ventral retina (Sagar, ’87; Rickman and Brecha, ’89).A related and dramatic observation from the different and darker visual world of the rat is that all PNMT-immunoreactive amacrine cells (possibly homologous with PA1 cells in rabbit) are confined to the dorsal retina (Versaux-Botteri et al., ’86).If these observations on the hemiretinal distributions of modulatory amacrine cells and the results of dopamine pharmacology are relevant to polyaxonal amacrine cells generally, then the possibility exists that one of their principal roles in rabbit is to enhance the detectability of small objects against a background of sky. Polyaxonal amacrine cells might then mediate a neural adaptation of retinal ganglion cells, accessed in their diversity through multiple pathways in the inner retina, which could be particularly active in the crepuscular phases of the day when lighting conditions are changing rapidly in the mesopic range.

ACKNOWLEDGMENTS I thank E. Siegfried and B. Ferguson for technical assistance. This work has been supported in part by an Alfred P. Sloan Foundation Fellowship, a National Eye Institute Vision Training Grant in the laboratory of N.W. Daw, the Medical Research Council of Canada, and the Alberta Heritage foundation for Medical Research. E.V.F. is an AHFMR Scholar.

LITERATURE CITED Ariel, M., and N.W. Daw (1982) Effects of cholinergic drugs on receptive field properties of rabbit retinal ganglion cells. J. Physiol. (Lond.) 324:135160. Brecha, N.C. (1983) Retinal neurotransmitters: Histochemical and biochemical studies. In P.C. Emson (ed): Chemical Neuroanatomy. New York: Raven Press, pp. 85-129. Cajal, S. Ramon y (1893) La retine des vertebres. Cellule 9t121-225. Cajal, S. Ramon y (1896) Nouvelles contributions a l’etude bistologique de la retine et a la question des anastomoses des prolongements protoplasmiques. J. Anat. Physiol. Paris 321481-543.

445

Caldwell, J.H., and N.W. Daw (1978) New properties of rabbit retinal ganglion cells. J. Physiol. (Lond.) 276257-276. Cbo, E.Y.P., and K.-F. So (1989) De novo formation of axon-like processes from axotomized retinal ganglion cells which exhibit long distance growth in a peripheral nerve graft in adult hamsters. Brain Res. 484:371-377. Daw, N.W., W.J. Brunken, and R.J. Jensen (1989) The function of monoamines in the rabbit retina. In R. Weiler and N.N. Osborne (eds): Neurobiology of the Inner Retina. Berlin: Springer-Verlag, pp. 363-374. Dacey, D.M. (1985) Wide-spreading terminal axons in the inner plexiform layer of cat’s retina: Evidence for intrinsic axon collaterals of ganglion cells. J. Comp. Neurol. 242247-262. Dacey, D.M. (1988) Dopamine-accumulating retinal neurons revealed by in vitro fluorescence display a unique morphology. Science 240:1196-1198. Dacey, D.M. (1989) Axon-bearing amacrine cells of the macaque monkey retina. J. Comp. Neurol. 284275-293. Dacey, D.M. (1990ai The dopaminergic amacrine cells of cat retina. Invest. Ophthalmol. (Suppl.) 31:535. Dacey, D.M. (1990b) Distinct cell types in living macaque retina display an unexpected granular fluorescence. SOC. Neurosci. Abstr. 16:466. Dowling, J.E., and B. Ehinger (1978) Synaptic organization of the dopaminergic neurons in the rabbit retina. J. Comp. Neurol. 18Ot203-220. Drager, U.C., D.L. Edwards, and C.J. Barnstable (1984) Antibodies against filamentous components in discrete cell types of the mouse retina. J. Neurosci. 412025-2042. Ehinger, B., and B. Falck (1969) Adrenergic retinal neurons of some new world monkeys. Z. Zellforsch. Mikrosk. Anat. 100:364-375. Fair&, A., A. Peters, and J. Saldanha (1977) A new procedure for examining Golgi impregnated neurons by light and electron microscopy. J. Neurocytol. 6:311-337. Famiglietti, E.V. (1981a) Functional architecture of cone bipolar cells in mammalian retina. Vision Res. 21:1559-1563. Famiglietti, E.V. (198lb) Displaced amacrine cells of the retina. SOC. Neurosci. Abstr. 7:620. Famiglietti, E.V. (1983a) “Starburst” amacrine cells and cholinergic neurons: Mirror-symmetric ON and OFF amacrine cells of rabbit retina. Brain Res. 261:138-144. Famiglietti, E.V. (1983b) ON and OFF pathways through amacrine cells in mammalian retina: The synaptic connections of “starburst” amacrine cells. Vision Res. 23r1265-1279. Famiglietti, E.V. (1985a) Starburst amacrine cells: Morphological constancy and systematic variation in the anisotropic field of rabbit retinal neurons. J. Neurosci. 5:562-577. Famiglietti, E.V. (198513) Synaptic organization of ON-OFF directionally selective ganglion cells in rabbit retina. SOC. Neurosci. Abstr. 11:337. Famiglietti, E.V. (1987) The ganglion cells of rabbit retina: Shape, stratification of dendritic trees, and relationship to cholinergic amacrine cells. Invest. Ophthalmol. (Suppl.) 28:289. Famiglietti, E.V. (1989a) Structural organization and development of dorsally directed (vertical) asymmetrical amacrine cells in rabbit retina. In R. Weiler and N.N. Osborne (eds): Neurobiology of the Inner Retina. Berlin: Springer-Verlag, pp. 169-180. Famiglietti, E.V. (1989b) Polyaxonal amacrine cells of the retina. SOC. Neurosci. Abstr. 15:968. Famiglietti, E.V. (1990a) Three categories of very wide-field amacrine cells in rabbit retina: Polyaxonal, HLPa, and HLPp amacrine cells. Invest. Ophthalmol. (Suppl.) 31t37. Famiglietti, E.V. (1990b) A new type of wide-field horizontal cell, presumablylinked to blue cones, in rabbit retina. Brain Res. 535:174-179. Famiglietti, E.V. (1991) Synaptic organization of starburst amacrine cells in rabbit retina: Analysis of serial thin sections by electron microscopy and graphic reconstruction. J. Comp. Neurol. 309:40-70. Famiglietti, E.V. (1992a) Polyaxonal amacrine cells of rabbit retina: Morphology of PA1 cells. J. Comp. Neurol. 316:391405. Famiglietti, E.V. (1992b) Polyaxonal amacrine cells of rabbit retina: Size and distribution of PA1 cells. J. Comp. Neurol. 316:406421. Famiglietti, E.V., and A. Peters (1972) The synaptic glomerulus and the intrinsic neuron in the dorsal lateral geniculate nucleus of the cat. J . Comp. Neurol. 144:285-334. Famiglietti, E.V., and H. Kolb (1975) A bistratified amacrine cell and synaptic circuitry in the inner plexiform layer of the retina. Brain Res. 84.293-300. Famiglietti, E.V., and H. Kolb (1976) Structural basis for ON- and OFFcenter responses in retinal ganglion cells. Science 194: 193-195.

446 Famiglietti, E.V., and E.C. Siegfried (1980) The amacrine cells of rabbit retina. Invest. Ophthalmol. (Suppl.) 19:70-71. Famiglietti, E.V., and N. Tumosa (1987) Immunocytochemical staining of cholinergic amacrine cells in rabbit retina. Brain Res. 413:398403. Famiglietti, E.V., N.C. Brecha, and H.J. Karten (1980) Neural localization of substance P-like immunoreactivity in rabbit retina. SOC. Neurosci. Abstr. 6212. Famiglietti, E.V., S. Sharpe, and G. Thurlow (1991) Degenerative and regenerative responses of axotomized rabbit retinal ganglion cells. SOC. Neurosci. Abstr. 17:187. Freed, M.A., and P. Sterling (1988) The ON-alpha ganglion cell of the cat retina and its presynaptic cell types. J. Neurosci. 82303-2320. Hadjiconstantinou, M., A.P. Mariani, P. Panula, T.H. Joh, and N.H. Neff (1984) Immunohistochemical evidence for epinephrine-containing retinal amacrine cells. Neuroscience 13:547-551. Hall, G.F., A. Poulos, and M.J. Cohen (1989) Sprouts emerging from the dendrites of axotomized lamprey central neurons have axonlike structure. J. Neurosci. 9:588-599. Hinds, J.W., and P.L. Hinds (1978) Early development of amacrine cells in the mouse retina: An electron microscopic, serial section analysis. J. Comp. Neurol. 179277-300. Hinds, J.W., and T.L. Ruffett (1973) Mitral cell development in the mouse olfactory bulb: Reorientation of the perikaryon and maturation of the axon intial segment. J. Comp. Neurol. 151:281-306. Jensen, R.J., and N.W. Daw (1984) Effects of dopamine antagonists on receptive fields of brisk and directionally selective cells in rabbit retina. J. Neurosci. 4:2972-2985. Jensen, R.J., and N.W. Daw (1986) Effect of dopamine and its agonists and antagonists on the receptive field properties of ganglion cells in the rabbit retina. Neuroscience 17.437-855. Marenghi, G. (1901) Contributo alla fina organizzazione della retina. Boll. SOC. Med. Chir. Pavia, 1-33. Mariani, A.P., and J.N. Hokoc (1988) Two types of tyrosine hydroxylaseimmunoreactive amacrine cell in the rhesus monkey. J. Comp. Neurol. 276:81-91. Nelson, R., E.V. Famiglietti, Jr., and H. Kolb (1978) Intracellular staining reveals different levels of stratification of On- and OFF-center ganglion cells. J. Neurophysiol. 41:472483.

E.V. FAMIGLIETTI Nguyen-Legros, J.,A. Vigny, and M. Gay (1983) Post-natal development of TH-like immunoreactivity in the rat retina. Exp. Eye Res. 3723-32. Palay, S.L., C. Sotelo, A. Peters, and P.M. Orkand (1968) The axon hillock and the initial segment. J. Cell Biol. 38:193-201. Peichl, L., and H. Wassle (1981) Morphological identification of On- and Lond. Off-centre brisk transient (Y) cells in the cat retina. Proc. R. SOC. [Biol.] 212:139-156. Peters, A., S.L. Palay, and H. de F. Webster (1990) The Fine Structure of the Nervous System: Neurons and Their Supporting Cells. New York: Oxford. Pourcho, R.G. (1982) Dopaminergic amacrine cells in the cat retina. Brain Res. 252:lOl-109. Rickman, D.W., and N.C. Brecha (1989) Morphologies of somatostatinimmunoreactive neurons in the rabbit retina. In R. Weiler and N.N. Osborne (eds): Neurobiology of the Inner Retina. Berlin: SpringerVerlag, pp. 461-468. Rodieck, R.W. (1988) The primate retina. In H.D. Steklis and J. Erwin (eds): Comparative Primate Biology, Vol. 4. Neuroscience. New York: Alan R. Liss, Inc., pp. 203-278. Sakai, H.M., K.-I. Naka, and J.E. Dowling (1986) Ganglion cell dendrites are presynaptic in catfish retina. Nature 319:495497. Sagar, S.M. (1987) Somatostatin-like immunoreactive material in the rabbit retina: Immunohistochemical staining using monoclonal antibodies. J. Comp. Neurol. 266.291-299, Sagar, S.M., and P.E. Marshall (1988) Somatostatin-like immunoreactive material in associational ganglion cells of human retina. Neuroscience 27:507-516. Sagar, S.M., P.E. Marshall, S.T. Onesti, and D.M.D. Landis (1986) Somatostatin immunocytochemistry in the rabbit retina. Invest. Ophthalmol. 27:316-322. Tatton, W.G., M.M. Kwan, M.C. Verrier, N.A. Seniuk, and E. Theriault (1990) MPTP produces reversible disappearance of tyrosine hydroxylase containing retinal amacrine cells. Brain Res. 527:21-31. Versaux-Botteri, C., E. Martin-Martinelli, J. Nguyen-Legros, M. Geffard, A. Vigny, and L. Denoroy (1986) Regional specialization of the rat retina: Catecholamine-containing amacrine cell characterization and distribution. J. Comp. Neurol. 243t422-433. Voigt, T., and H. Wassle (1987) Dopaminergic innervation of A I1 amacrine cells in mammalian retina. J. Neurosci. 7:4115-4128.