Proc. Natl. Acad. Sci. USA
Vol. 85, pp. 8321-8325, November 1988
Bipolar cells in the turtle retina are strongly immunoreactive
B. EHINGER*t, 0. P. OTTERSENt, J. STORM-MATHISEN*, AND J. E. DOWLING*
*Department of Cellular and Developmental Biology, The Biological Laboratories, Harvard University, Cambridge, MA 02138; tDepartment of
Ophthalmology, University of Lund, S-22185 Lund, Sweden; and tDepartment of Anatomy, University of Oslo, N-0162 Oslo 1, Norway
Contributed by J. E. Dowling, July 21, 1988
served by both light and electron microscopy in bipolar cells of
the turtle (Pseudemys scripta elegans) retina after postembed-
ding immunohistochemistry. Virtually all bipolar cells showed
strong labeling, on average 18 times that of the Muller (glial)
cells. The data suggest that both on- and off-center bipolar cells
are glutamatergic. Photoreceptors were also labeled, but with
a labeling intensity about half that of the bipolar cells. Other
types of retinal neurons showed less immunoreactivity, except
for a small population of strongly labeled amacrine cells.
Strong glutamate immunoreactivity was ob-
Bipolar cells carry visual information from the outer to the
inner retina and are the first cells along the visual pathway to
be divided into separate on- and off-channels. Furthermore,
they show a center-surround organization similar to that
observed in retinal ganglion cells and other neurons in the
visual system (1, 2). Surprisingly, little is known about the
transmitters employed by the bipolar cells. Physiological
evidence has indicated that bipolar cells are excitatory to
ganglion cells (3-6), but firm evidence for the presence ofan
excitatory transmitter in bipolar cells has not been forthcom-
Both amacrine and ganglion cells possess receptors spe-
cific for the acidic amino acids, and therefore it has been
proposed that L-glutamate or a similar excitatory amino acid
is a neurotransmitter in bipolar cells (7-11). Attempts at
localizing endogenous glutamate in the retina have been only
partially successful, and at times contradictory, most likely
because of the indirect nature of the methods available (12).
Recently, a technique has been developed to localize
glutamate immunohistochemically by applying a purified
antibody to etched plastic sections and demonstrating the
binding site of the first antibody with a second antibody
tagged with small (15-nm) colloidal gold particles (13, 14).
The method gives agood signal-to-noise ratio, and since only
a small fraction of the glutamate of the cell is available for
detection by the primary antibody (i.e., that at the section
surface), the gold particles do not obscure the cytological
characteristics of the labeled neurons. We have used this
technique to localize glutamate in the turtle retina.
MATERIALS AND METHODS
For light microscopy, small pieces from the posterior pole of
light-adapted eyes of the turtle Pseudemys scripta elegans
were fixed in 2% glutaraldehyde in 0.1 M phosphate buffer
(pH 7.4) for 2 hr at room temperature, dehydrated, and
embedded in Durcupan ACM (Fluka). For electron micros-
copy, small pieces of retina were fixed at room temperature
for90 min with4% glutaraldehyde, 1% formaldehyde, and 0.2
mM CaCl2 in 0.1 M phosphate buffer (pH 7.4). After washing
and postfixation for 1 hr in 1% OS04 in the same buffer, the
specimens were dehydrated, embedded in Durcupan ACM,
and cured at 520C.
For light microscopy, 0.5- to 1-gm tissue sections were
processed according to Somogyi et al. (13, 15), using the
peroxidase-antiperoxidase (PAP) technique (13-16). For
electron microscopy , a modification (14) of the immunogold
procedure of Somogyi and Hodgson (17) was used that
employed Janssen AuroProbe 15-nm gold particles coated
with goat anti-rabbit antibodies. The antiserum, 13 Glu, was
purified by immunoadsorption on three different Sepharose
columns, one bearing bovine serum albumin treated with
glutaraldehyde and the others bearing the same protein to
which glutamine or y-aminobutyrate had been conjugated
with glutaraldehyde (14, 18, 19). The antiserum, which has
been characterized (14, 17, 20), was diluted 1:800 for both
light and electron microscopy.
Three kinds of controls were undertaken. First, the tissue
was processed as described above, but the primary antibody
was omitted from the reaction mixture. As expected, no
specific labeling was subsequently observed. Second, the
antiserum was absorbed with glutamate (200AM)that had
been treated with glutaraldehyde (19). This also abolished the
labeling. On the other hand, absorbing the antiserum with
glutaraldehyde-treated glycine, aspartate, taurine, p-alanine,
or glutamine did not appreciably diminish the reaction (19).
Third, various amino acids were added to brain tissue
homogenates that had been extensively dialyzed to remove
all free amino acids. The resulting mixtures were fixed with
glutaraldehyde and embedded in Durcupan (14). Electron
microscopy of this material showed a 95-fold higher density
of label when glutamate was added to the homogenates as
compared with glutamine and even higher density ratios in
comparison with the other substances. This agrees well with
previously published data on this antibody; for example, it
does not react with glutathione or a number of other small
The degree of labeling was assessed by counting the
numberofgrains in identified cell processes in a large number
ofelectron micrographs. Process area was measured with the
aid of a digitizing pad connected to a small computer.
Because the grain density with the colloidal gold procedure
is low, small processes often showed no or only a few
particles. The number of small processes was usually con-
siderable, and statistics based on observed grain densities are
therefore quantized and not distributed normally. Conse-
quently, nonparametric confidence and tolerance limits were
calculated with the quantile test, and probabilities of differ-
ences in median values were obtained with the nonparametric
median test (21).
Background labeling of pure plastic was negligible (<0.05
grain perjum2),and the grain count in Muller cells was the
lowest of all cells in the retina. The labeling index used was
the labeling for the different retinal neurons relative to the
Muller cells and is introduced to take into account the fact
that some glutamate is likely to be found in all cells. The
upper 95% tolerance limit ofthe 90th percentile ofthe Muller
The publication costs ofthis article were defrayed in part by page charge
payment. This article must therefore be hereby marked "advertisement"
in accordance with 18 U.S.C. §1734 solely to indicate this fact.
Neurobiology: Ehinger et al.
cell grain density was found to be 3.3 grains perAum'(a
labeling index of 4.1). Only profiles with higher grain densi-
ties were considered significantly labeled.
In the light microscope, bipolar cells were the most promi-
nently labeled of all the retinal neurons (Fig. 1). Bipolar cell
perikarya were found mainly in the middle or outer halfofthe
inner nuclear layer. They were often readily identified by
their Landolt club processes (Fig. 1, arrows), which extend
from the dendritic arborization ofthe bipolar cells to between
the photoreceptor inner segments (22, 23). Many of the
bipolar cells also showed an axonal process that typically
took a very oblique course through the inner nuclear and
inner plexiform layers (Fig. 1, arrowhead) (23). Numerous
strongly immunoreactive profiles were also observed in the
inner plexiform layer; many of these were likely to be bipolar
cell processes or terminals (see below).
Photoreceptor terminals were also consistently labeled,
although they were never as strongly labeled as the bipolar
cells. Most amacrine cells showed no or only moderate
labeling, but a few strongly labeled cells were observed (Fig.
1, open arrowhead). Most ofthe cells in the ganglion cell layer
exhibited modest staining, and an occasional horizontal cell
showed some immunoreactivity. Some staining in the optic
nerve fiber layer was seen.
by the peroxidase-antiperoxidase procedure. Strong labeling is seen
in bipolar cells (B) and their Landolt club processes (arrows). The
centripetal processes ofthe bipolar cells have a characteristic oblique
course through the inner nuclear layer (large arrowheads). There are
numerous strongly labeled processes in the inner plexiform layer
(IPL). Photoreceptor terminals are labeled in the outer plexiform
layer (small arrowheads). There is also a well-labeled amacrine cell
(open arrow), whereas most other cells at the same level (A) are only
moderately or weakly labeled. Ganglion cells (G) show some label,
as do the bundles of optic nerve fibers (NF) embedded in unlabeled
Muller cell processes (M). The small white spots in the inner
plexiform layer are holes in the section caused by the etching
combined with the vigorous peroxidase reaction. Ph, photoreceptors;
OPL, outer plexiform layer. (Phase-contrast micrograph; x 540.)
Glutamate immunoreactivity in the turtle retina, shown
In the electron microscope, the bipolar cells and their
processes were the most prominently labeled of any of the
retinal elements. Relative to the Muller cells, they had an
average labeling index of 18 (Table 1), and individual bipolar
cell processes or terminals had indexes as high as 40-45. Fig.
2a shows a heavily labeled bipolar cell terminal in the inner
plexiform layer. This process can be confidently identified as
a bipolar cell terminal because of the synaptic ribbon it
contains (arrow) (24). Fig. 2b shows another bipolar cell
terminal (B) in the inner plexiform layer, showing a more
average labeling density. Gold grains are clearly seen in this
terminal (arrowheads), but they are not nearly as numerous
as in the terminal shown in Fig. 2a. Furthermore, the label is
not evenly distributed throughout the terminal. A Muller cell
process (M) ofapproximately the same area as the bipolar cell
terminal is also present in this micrograph. It is devoid ofgold
All parts of the bipolar cells appeared to be labeled to the
same extent. Displaced bipolar cells were observed occa-
sionally lying among the photoreceptor cell perikarya and
terminals, and the glutamate immunoreactivity was analyzed
in five such cells. The labeling of the displaced bipolar cells
could not be distinguished from that of other bipolar cells.
Some bipolar cell processes showed no gold grains or only
one or two grains. In most of these cases the bipolar cell
profiles were small, and thus these may represent labeled
processes in which the label was not distributed evenly and
the section passed through an unlabeled region (see Fig. 2b).
However, a few larger bipolar cell processes with no or only
a few grains were noted. We cannot exclude, therefore, that
there may be a small percentage of bipolar cells with no or
very low glutamate immunoreactivity.
The density of labeling in -25O bipolar and Muller cell
profiles is shown in Fig. 3. The average sizes of the bipolar
and Muller cell profiles whose grain densities were analyzed
for this figure were approximately the same. The figure
shows clearly that the two cell types are distinctly different
in terms of grain density (median test, P < 0.001). The
overwhelming majority ofthe Muller cell processes had a <4
grains per,um', whereas the grain density ofall but four ofthe
bipolar cell profiles was between 4 and 38 grains per 1Xm2. The
distribution ofgrain density in the bipolar cell profiles did not
appear Gaussian; indeed the distribution appeared to fit
better a bimodal distribution with one peak at about 12-16
grains per Am2 and the other at about 22-26 grains per Am2.
Photoreceptor terminals were also consistently labeled.
Fig. 4 shows a typically labeled cone photoreceptor terminal,
identified by its position in the outer plexiform layer, char-
acteristic shape, and long synaptic ribbon. Although the
average labeling density in the photoreceptors was about half
that of the bipolar cells (Table 1), it was comparable to that
of labeled amacrine cells in the retina. Variation in labeling
Glutamate immunoreactivity (grain counts) in
Data were obtainedfrom three sections on a single grid. The degree
of labeling of the three cell types was significantly different in all
cases (P < 0.001). The background (<0.05 grain perAm2)was not
Proc. Natl. Acad. Sci. USA 85(1988)
Proc. Natl. Acad. Sci. USA 85 (1988)
4v . s@-
with average label density (arrowheads) in b. For clarity, the cell boundaries have been outlined in b. The synaptic ribbons (arrowina) identify
the processes as bipolar cell terminals. Note the absence of gold particles in the Muller cell process (M) in b. (a, x45,000; b, x34,000.)
Glutamate immunoreactivity in terminals of bipolar cells (B) in the inner plexiform layer of turtle retina, denselylabeled in a and
density was observed in photoreceptor terminals, but signif-
icant labeling was seen in both rod and cone photoreceptor
Most amacrine cell perikarya had alow density oflabel, but
a few amacrine cells showed strong labeling. Of45 amacrine
cell perikarya analyzed in the electron microscope, 24 had a
labeling index of <4 (i.e., were unlabeled), 13 had an index
in the range 5-11 and 8 had a labeling index averaging about
14. The strongly labeled amacrine cell perikarya most often
had a relatively clear and voluminous cytoplasm, and their
nuclei were indented.
Most amacrine cell processes identified in the inner plexi-
form layer had either no grains or only one grain, showing
that there is a large population of unlabeled amacrine cell
processes. However, a number of clearly labeled amacrine
cell processes were also observed, and an example is shown
in Fig. 5. Amacrine cell processes in the inner plexiform layer
typically show scattered synaptic vesicles and little cytoplas-
mic density. Furthermore, they make conventional-type
synaptic contacts. The synapse made in Fig. 5 is onto a
labeled process containing numerous synaptic vesicles and
showing significant cytoplasmic density, suggesting that this
postsynaptic profile is likely to be a bipolar cell terminal.
Labeled amacrine cell processes were observed to contact
many different elements in the inner plexiform layer. They
were presynaptic to ganglion cell dendrites, pre- and postsy-
naptic to labeled or unlabeled amacrine cell processes, and
pre- and postsynaptic to labeled bipolar cell terminals.
Gold grains per ,uM2
distributions of profile sizes of the two cell types were sufficiently similar to allow acomparisonto be made in theplot.Note thatvery nearly
all bipolar cell profiles had grain densities higherthan that of the Muller cells and thattheytend to form more than onepeak.
Percentages of 133 Muller cell and 121bipolarcellprocess profilesin the innerplexiform layerwith differentgraindensities. The
Neurobiology: Ehingeret al.
Neurobiology: Ehinger et al.
!? ,.' .'':: '
by its characteristic morphology and position in the outer plexiform
layer. The colloidal gold grains are indicated by arrowheads.
Reciprocal synapses between labeled amacrine cell pro-
cesses and bipolar cell terminals were not seen, however.
Most horizontal cell profiles showed little labeling, with an
average labeling index of 3.2 and 2.6 in 30 and 49 cell bodies
and axon terminals, respectively. However, afew cell bodies
and axon terminals did have a somewhat higher grain density
(labeling index around 8). At least some of these labeled cell
bodies appeared to have cytoplasmic characteristics of the
H1 horizontal cell type (26).
Many of the perikarya in the ganglion cell layer showed a
moderate amount of label (average labeling index of 6.5),
although some cells showed very little. Since many of the
cells in the ganglion cell layer may be displaced amacrine
cells, it is difficult to decide the type of cell showing label.
However, of 97 ganglion cell nerve fibers examined, about
60% had a labeling index above 4.1, indicating that a
significant number ofganglion cells were moderatelylabeled.
The most striking observation in this study was the strong
glutamate immunoreactivity observed inbipolarcells. These
results suggest that there is a significant amount of endoge-
that makes a synapse onto a labeled process, most likely abipolar cell
terminal. The gold grains in the amacrine cell process are indicated
by arrows. As can be seen here, clusters of gold grains were
occasionally observed. Clustering may be artifactual (42); thus, each
cluster was counted as a single grain. (x50,000.)
Glutamate immunoreactivity in an amacrine cell process
nous glutamate in bipolar cells, and thus they support the
notion that glutamate may be a bipolar cell neurotransmitter.
That the immunoreactivity demonstrated in the present study
does represent tissue glutamate is established beyond rea-
sonable doubt by the controls performed and by previous
investigations (13, 14, 18, 20). Furthermore, recent studies
(29) with the postembedding electron microscopic immuno-
gold method have shown that the grain density over gluta-
mate-glutaraldehyde-brain protein conjugate particles is ap-
proximately proportional to the concentration ofglutamate in
the conjugate, at least in the higher biologically relevant
concentration range. Thus, the relative labeling densities
reported here are likely to reflect real differences is endog-
enous glutamate concentrations.
Slaughter and Miller (30) provided evidence a few years
ago that on-center bipolar cells use an excitatory amino acid
as theirtransmitter, and anumberofother studies (7-11) have
provided indirect evidence that both on- and off-center
bipolar cells are likely to employ an excitatory amino acid as
a neurotransmitter. The present study provides direct evi-
dence for high levels of glutamate in bipolar cells and their
terminals. It is of particular relevance that the glutamate
concentration is high in the bipolar terminals, as is expected
of a substance used as a neurotransmitter by a neuron.
There are two main functional types of bipolar cells,
on-center cells and off-center cells, and they appear to be
present in roughly equal numbers in many retinas (1).
Because there can be at most only a very small minority, if
any, of the bipolar cells in turtle that are not glutamate
immunoreactive, we conclude that both on- and off-center
cells contain L-glutamate and are likely to be glutamatergic.
It has been shown that some bipolar cells in certain retinas
contain serotonin (31-34) as well as immunoreactive glycine
and y-aminobutyrate (35, 36).
significance of these observations may be, but it seems
possible that bipolar cells, like many neurons in the brain,
may contain more than one neuroactive substance.
The finding that all photoreceptor cell terminals in the
turtle retina show glutamate immunoreactivity is not surpris-
ing, because there is now extensive evidence that photore-
ceptors in many vertebrates use L-glutamate as their neuro-
transmitter (37-39). Why the labeling of the bipolar cells and
terminals was significantly greater than that of the photo-
receptor terminals is not clear. It may be that the glutamate
content of a cell or terminal relates in some way to its
physiology. Forexample, it is known that photoreceptor cells
release transmitter in the dark, whereas on-center bipolar
cells appear to release transmitter only in the light.
The majority of amacrine cells in all retinas contain
y-aminobutyrate or glycine, and amacrine cells are generally
regarded as inhibitory (40). However, there is compelling
evidence that some amacrine cells are excitatory (41). The
present results suggest that at least some ofthe amacrine cells
in the turtle retina are glutamatergic. The input into bipolar
cells from amacrine cells is often at reciprocal synapses, and
the transmitter of the amacrine process at such junctions is
presumed to be inhibitory. In agreement with this, no labeled
amacrine process was seen to make a reciprocal synapse with
a bipolar cell. However, the observation of glutamate-
immunoreactive amacrine cell processes making synapses
onto bipolar cell terminals, other amacrine cell processes,
and ganglion cell dendrites suggests there may be excitatory
glutamatergic input by amacrine cells onto all of these
It is uncertain what the
Valuable statistical advice was given by Dr. Armando Garsd.
Some of the turtles were supplied by Dr. Alan Adolph. We thank
Gordon Fain, Richard Masland, and Stephen Yazulla for providing
comments on the manuscript. The study was supported by Grant
EY00811 from the National Eye Institute (J.E.D.); by grantsfrom the
Proc. Natl. Acad. Sci. USA 85(1988)
Proc. Natl. Acad. Sci. USA 85 (1988)
Swedish Medical Research Council (project 2321), the Elsa och
Torsten Segerfalks Stiftelse, the Hoch L Nilssons Stiftelse, and the
Faculty of Medicine, University of Lund (B.E.); and by grants from
the Norwegian Council for Science and the Humanities (O.P.O. and
Werblin, F. S. & Dowling, J. E. (1969) J. Neurophysiol. 32,
Kaneko, A. (1970) J. Physiol. 207, 623-633.
Miller, R. F. & Dacheux, R. F. (1976) J. Gen. Physiol 67, 661-
Miller, R. F. & Dacheux, R. F. (1976) J. Gen. Physiol. 67, 679-
Naka, K.-I. (1976) Invest. Ophthalmol. 15, 926-935.
Naka, K.-I. (1977) J. Neurophysiol. 40, 26-43.
Slaughter, M. M. & Miller, R. F. (1983) Nature (London) 303,
Slaughter, M. M. & Miller, R. F. (1983) J. Neurosci. 3, 1701-
Kato, S., Teranishi, T. & Negishi, K. (1985) Brain Res. 329,
Bloomfield, S. A. & Dowling, J. E. (1985) J. Neurophysiol. 53,
Lukasiewicz, P. D. & McReynolds, J. S. (1985) J. Physiol.
(London) 367, 99-115.
Ehinger, B. & Dowling, J. E. (1987) in Integrated Systems of
the CNS, Part I, Handbook of Chemical Neuroanatomy, eds.
Bjorklund, A., Hokfelt, T. & Swanson, L. W. (Elsevier,
Amsterdam), Vol. 5, pp. 389-446.
Somogyi, P., Halasy, K., Somogyi, J., Storm-Mathisen, J. &
Ottersen, 0. P. (1986) Neuroscience 19, 1045-1050.
Ottersen, 0. P. (1987) Exp. Brain Res. 69, 167-174.
Somogyi, P., Hodgson, A. J., Smith, A. D., Nunzi, M. G.,
Gorio, A. & Wu, J.-Y. (1984) J. Neurosci. 4, 2590-2603.
Sternberger, L. A. (1979) Immunocytochemistry (Wiley, New
York), 2nd Ed.
Somogyi, P. & Hodgson, A. J. (1985) J. Histochem. Cytochem.
Storm-Mathisen, J., Leknes, A. K., Bore, A. T., Vaaland,
J. L., Edminson, P., Haug, F.-M. S. & Ottersen, 0. P. (1983)
Nature (London) 301, 517-520.
Ottersen, 0. P., Storm-Mathisen, J., Madsen, S., Skumlien, S.
& Str0mhaug, J. (1986) Med. Biol. 64, 147-158.
Ottersen, 0. P. & Storm-Mathisen, J. (1984) J. Comp. Neurol.
Conover, W. J. (1980) Practical Nonparametric Statistics
(Wiley, New York), 2nd Ed.
Hendrickson, A. (1966) Invest. Ophthalmol. 5, 484-496.
Dacheux, R. F. (1982) J. Comp. Neurol. 205, 55-62.
Dowling, J. E. & Boycott, B. B. (1966) Proc. R. Soc. London
Ser. B 166, 80-111.
Kolb, H. & Jones, J. (1982) J. Comp. Neurol. 209, 331-338.
Kolb, H. & Jones, J. (1984) J. Neurocytol. 13, 567-591.
Ohtsuka, T. (1985) Science 229, 874-877.
Ohtsuka, T. (1985) J. Comp. Neurol. 237, 145-154.
Ottersen, 0. P. (1988) Neuroscience (Suppl.), in press.
Slaughter, M. M. & Miller, R. F. (1983) Science 219, 1230-
Bruun, A., Ehinger, B. & Sytsma, V. M. (1984) Brain Res. 295,
Witkovsky, P., Eldred, W. & Karten, H. J. (1984) J. Comp.
Neurol. 228, 217-225.
Weiler, R. & Schutte, M. (1985) Cell Tissue Res. 241, 373-382.
Brunken, W. J., Witkovsky, P. & Karten, H. J. (1986) J.
Comp. Neurol. 243, 1-12.
Agardh, E., Ehinger, B. & Wu, J.-Y. (1987) Histochemistry 86,
Pourcho, R. G. & Goebel, D. J. (1987) J. Neurosci. 7, 1189-
Slaughter, M. M. & Miller, R. F. (1981) Science 211, 182-185.
Ishida, A. T. & Fain, G. L. (1981) Proc. Natl. Acad. Sci. USA
Lasater, E. M. & Dowling, J. E. (1982) Proc. Natl. Acad. Sci.
USA 79, 936-940.
Miller, R. F. (1979) in The Neurosciences: Fourth Study Pro-
gram, eds. Schmitt, F. 0. & Worden, F. G. (Massachusetts
Institute of Technology Press, Cambridge, MA), pp. 227-245.
Masland, R. H., Mills, J. W. & Cassidy, C. (1984) Proc. R.
Soc. London Ser. B 223, 121-139.
De Mey, J. (1983) in Immunocytochemistry: Practical Appli-
cations in Pathology and Biology, eds. Polak, J. M. & Van
Noorden, S. (Wright-PSG, Bristol, U.K.), pp. 82-112.
Neurobiology: Ehingeret al.