JOURNALOF NEUROPHYSIOLOGY
Vol. 60, No. 5, November 1988. Printed in U.S.A.
Dissection of the Neuron Network in the Catfish
Inner Retina. I. Transmission to Ganglion Cells
HIROKO M. SAKAI AND KEN-ICHI
National InstituteSforBasicBiology, Okazaki 444, Japan
SUMMARY AND CONCLUSIONS
2. To characterize the signal transmission
from amacrine to ganglion cells, and to iden-
tify the filter that transforms amacrine-cell re-
sponsesinto ganglion-cell spike discharges,
an extrinsic current, either sinusoidally or
white-noise modulated, was injected into an
amacrine cell and the resulting extracellular
spike dischargeswere recorded from a neigh-
boring ganglion cell. For the sinusoidal in-
puts, PST (poststimulus time) histograms
were produced; for the white-noise inputs,
first- and second-order Wiener kernels were
computed by a cross-correlation process.
2. Extrinsic current injected either into a
type-N (sustained) amacrine cell or a type-C
(transient) amacrine cell modulated the spike
dischargesof nearby ganglion cells, whether
of the “ON,” “ ON-OFF” or “OFF” types. We
identified two modes of signal transmission,
fast (probably monosynaptic) and slow
(probably polysynaptic) transmission. Signal
transmission from amacrine to ganglion cells
of the same responsepolarity i.e., from type-
NA (depolarizing, sustained) amacrine to ON-
ganglion cell and from type-NB (hyperpolar-
izing, sustained) amacrine to OFF-ganglion
cell, was either fast or slow. Similarly, the sig-
nal transmission from type-C to either ON- or
OFF-ganglion cells was either fast or slow.
3. The signal transmission from amacrine
to ganglion cell of the opposite responsepo-
larity, i.e., from type-NA to OFF-ganglion cell
and from type-NB to ON-ganglion cell, was
always slow.
4. Fast transmission from type-N ama-
crine to a ganglion cell of the same polarity,
or from type-C to either ON- or OFF-ganglion
cells was always sign-noninverting. The
0022-3077/88 $1 SO Copyright 0 1988 The American
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.
NAKA
transfer function was lowpass, with a cutoff
frequency of 30 Hz.
5. Slow transmission from any type of
amacrine cell (either type-NA, -NB or -C) to
ON-ganglion cells was always sign inverting,
whereas from any amacrine to OFF-ganglion
cells was always sign-noninverting. The
transfer function for the slow transmission
was narrow bandpass,with a cutoff frequency
of 30-40 Hz.
INTRODUCTION
A major goal of our studies on catfish ret-
ina is to dissect neuronal circuitry by identify-
ing the transfer function between neurons,
i.e., to recover the filter that transforms the
responsesof the presynaptic neuron to those
of the postsynaptic neuron. The most direct
means of dissecting complex circuitry and of
identifying such a filter is to inject a known
signal into one circuit element and to record
the resulting and transformed signals in an-
other element. In the case of white-noise in-
put, cross-correlation between the input and
the output reveals the filter between neurons.
(cf. Fig. 1 in Sakai and Naka, Ref. 29).
In this series of papers, we will describe
neuronal filters studied in the catfish inner
layer by the use of the current injection
method; in this article, filters intervening be-
tween preganglionic cell and ganglion cell
spike dischargesare described, and in the fol-
lowing article (30) filters between ganglion
cells themselves are characterized. An extrin-
sic current modulated by a sinusoidal signal
was used to generate a PST (poststimulus
time) histogram whereas that modulated by a
white-noise signal was used to produce first-
and second-order Wiener kernels.
Physiological Society 1549
1550 H. M. SAKAI
Recently, we have shown that cross-corre-
lation between the (analog) white-noise input
and the resulting spike discharges (trans-
formed into unitary pulses) produced first-
and second-order Wiener kernels similar to
those computed from the intracellular slow
potentials (28, 32). Within the second-order
approximation, spike discharges (point pro-
cess) carry as much information as carried by
the intracellular (analog process) potentials.
The kernel thus derived characterizes the lin-
ear and non-linear components involved in
the signal transmission and Fourier transfor-
mation of the first-order kernel gives the fre-
quency and phase characteristics of the linear
component.
The analysis of light-evoked first- and sec-
ond-order kernels of ganglion and pre-gangli-
onic cells in the catfish retina has produced
two major conclusions: 1) horizontal and bi-
polar cell modulation responses are linearly
related to the input modulation; signal pro-
cessing in the outer retina is piecewise linear
(29, 33). The second-order nonlinearity ap-
pears in amacrine cells (28, 34, 35); and 2)
linear signals from bipolar cells and second-
order nonlinear signals from amacrine cells
are transmitted to ganglion cells and encoded
into spike trains without major modification
(28,32). The signal transmission from a bipo-
lar and an amacrine cell to a ganglion cell of
the same response polarity must, therefore,
be fast and sign-noninverting to preserve the
kernel waveforms (28).
The above conclusions are different from
those drawn from studies of mudpuppy and
tiger salamander retinas in which (transient)
amacrine cells were assumed to provide in-
hibitory inputs to ganglion cells (6, 16, 39,
46). Results of current injection experiments
will provide us with direct evidence on the
polarity of the transmission from amacrine to
ganglion cells; the polarity must be sign-in-
verting if a depolarization of an amacrine cell
depresses spike discharges of the ganglion
cells.
We conclude that 1) an extrinsic current
injected into an amacrine cell modulates
spike discharges of ganglion cells; 2) there are
two modes of signal transmission from ama-
crine to ganglion cells, fast (probably mono-
synaptic) and slow (probably polysynaptic)
transmission; and 3) transmission between
cells of the same response polarity is often but
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.
AND K.-I. NAKA
not always fast, whereas that between cells
having opposite polarities is always slow.
These conclusions significantly differ from
those derived in the earlier current injection
experiments by Naka (18). We will discuss
the discrepancies and possible synaptic path-
ways for the two types of transmission.
MATERIALS AND METHODS
The eye-cup preparation of the channel catfish
(Ictalurus punctatus) was used. Simultaneous re-
cordings were made from a pair of retinal neurons
using two electrodes; one was a glass electrode
filled with 2 M potassium citrate for intracellular
recordings and current injection, and the other a
platinum-coated tungsten electrode to record ex-
tracellular spike discharges from a ganglion cell.
Two electrodes were positioned by a pair of hy-
draulic manipulators (MO- 15, Narishige Sci. In-
struments, Ltd., Tokyo, Japan). Intracellular re-
cording as well as current injection were made
with a S-7 1OOA system, using the S-707 1A elec-
trometer module (WP Instruments, Inc., New Ha-
ven, CT). Extracellular recording was accom-
plished with a WPI DAM 50 amplifier. Glass pi-
pettes were pulled on a Model 700C puller (David
Kopf, Tojunga, CA). The tips of the two electrodes
were separated by 200-400 pm. The experimental
arrangement is shown schematically in Fig. 1.
Extracellular spike discharges were located first.
Then a glass pipette was introduced to record in-
tracellularly from a nearby neuron. When an in-
tracellular recording was established, two light and
two current stimuli were given successively as fol-
lows: I) a flash of diffuse white light, 30 pW/cm2
and -300 ms in duration, was presented in the
dark to evoke responses simultaneously from the
two cells; 2) a Gaussian white-noise modulated
light, 30 pW/cm2 in mean luminance, which cov-
ered the entire retinal surface, was given for 40-
60 s to define the cells’ response dynamics; 3) a
sinusoidal current of either 5 or 10 Hz was injected
for 1O-20 s through the intracellular electrode and
the resulting spike discharges were recorded from
the ganglion cell. The current intensity was man-
ually adjusted between a range of 4- 12 nA, peak-
to-peak until the spikes were clearly modulated;
and 4) a Gaussian white-noise modulated current,
typically with a standard deviation of 2-5 nA, was
injected for 20-40 s through the intracellular elec-
trode to evoke discharges from the ganglion cell.
The power spectrum of the white-noise modulated
current was flat from 1 to 100 Hz and that of the
white-noise modulated light was flat from 1 to
50 Hz.
Extrinsic current was delivered in the dark, but
in some experiments, current was injected in the
presence of a steady luminance. The first-order
 TRANSMISSION TO RETINAL GANGLION CELLS. I 1551

CROSSCORRELATION
INPUT OUTPUT
r

SPIKE
DISCHARGES

FIG. 1. Schematic drawing of the experiment. Two electrodes were used; an intracellular one for recording and
current injection into a preganglionic cell and an extracellular one to record current modulated spike discharges from
a neighboring ganglion cell. The tips of the two electrodes were separated by -200-400 pm. For white-noise analysis
spike discharges were transformed into unitary pulses and a cross-correlation was made between the current signal
and the pulses.

kernels for current injection performed in the dark
and in the presence of a steady luminance were
generally of similar amplitude although the ker-
nels obtained in the presence of a steady lumi-
nance had a slightly shorter peak response time
and their waveform was more differentiating.
similar finding was reported in a previous paper
for the discharges evoked by a current injected into
horizontal cells (Fig. 8 in Ref. 19). These differ-
ences were, however, minor and did not interfere
with the conclusions drawn in this paper. Results
shown in this paper are, therefore, obtained by
modulating the “dark” potential of preganglionic
cells.
The light source was a glow tube (R- 113Cl En-
glish Electrical Valve, Essex, England). The Gaus-
sian white-noise signals for both current and light
stimuli were obtained from a signal generator
(WG-667 white-noise generator, NF Design
Block, Tokyo, Japan). The light stimulus was
monitored by a photodiode (S 1406, Hamamatsu
Photonix, Hamamatsu, Japan). The current signal
was obtained from the current monitor of the S-
707 1A module.
Electrical signals were stored initially on analog
tape with a SONY NFR-300 data recorder (Sony
Magnescale, Inc., Tokyo, Japan) and transferred
off-line onto the memory of a VAX 1 l/780 com-
puter (Digital Equipment Corp., Maynard, MA),
with a digitization rate of 1 KHz for white-noise
analysis and 4 or 10 KHz for response waveform
display. Analyses were performed through a soft-
ware system, STAR, running on a combination of
VAX 11/780 computer and an AP 120B array pro-
cessor (Floating Point Systems, Portland, OR). Ex-
tracellular spike discharges were transformed into
A unitary pulses of l-2 ms duration (Fig. 1). First-
and second-order Wiener kernels were computed
by crosscorrelating the white-noise signal against
a train of unitary pulses (32). Time resolution of
analysis was 1 ms for the current-evoked and 2 ms
for the light-evoked kernels.
White-noise vs. pulsatile or
sinusoidal st im uli
In neurophysiology, a pulse or a step input has
been the most popular stimulus to define signal
transmission (11, 13,22, 23, 26,40). Results from
such experiments show: I) the polarity of signal
transmission is sign inverting or sign-noninvert-
ing. If a presynaptic depolarization produces a
postsynaptic depolarization or an increase in the
spike discharges, the transmission is sign-nonin-
verting; if a presynaptic hyperpolarization pro-
duces depolarization in the postsynaptic cell, the
transmission is sign inverting; and 2) the responses
can be classified either as sustained or transient.
A pulse of current, however, contains high fre-
quency components at its onset and offset, which
often contaminate evoked responses through the
cross-talk between two electrodes. In current in-
jection experiments, such artifacts become prob-

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

1552 H. M. SAKAI
lematic because they are often indistinguishable
from a true extracellularly recorded spike dis-
charge of < 1 mV in amplitude.
A sinusoidal current, another choice of stimulus
(12, 18), is a gradually changing signal and does
not contain high frequency signals as does the
pulse of current. However, as is shown in RE-
SULTS, A2 and B2 of Figs. 9 and 10, neurons re-
spond differently to inputs of different frequencies.
To test all the necessary frequencies, sinusoidal
currents at many single frequencies have to be
tested, a very time consuming process, or a current
has to be modulated by a sum of sinusoids (42). In
the case of a single test stimulus, the frequency of
a sinusoidal input, or duration of a pulsatile input,
has to be carefully chosen to evoke the optimal re-
sponse from a cell. In this study, we have used a
single-frequency sinusoidal current of either 5 or
10 Hz; we know from the white-noise analysis that
neurons in the inner retina respond vigorously to
the input frequency.
White-noise modulated current has a flat spec-
trum (up to 100 Hz in this series of experiments)
and varying intensities of a Gaussian (i.e., normal)
distribution. Wiener kernels can be computed by
cross-correlating the input current and the output
spike discharge from an experiment of 20-40 s in
duration. The kernel thus obtained is interpreted
as a filter (in time domain analysis) that transforms
presynaptic neural responses into postsynaptic
neural responses and Fourier transformation
the kernel defines the transfer function (in fre-
quency domain analysis) between two neurons.
To compute kernels, the spike discharges have
to be converted into an analog form. Marmarelis
and Naka ( 14) achieved this by producing a PST
histogram by repeating the same stimulus several
times and by recording the resulting spike dis-
charges in a sequence of bins. This method, al-
though it has produced many valuable results, has
two difficulties: I) it is time consuming, and 2) the
process of “binning” introduces artificial, high fre-
quency components.
Recently, Sakuranaga et al. (32) have shown
that a direct cross-correlation between analog whi-
te-noise input against the resulting spike dis-
charges (transformed into unitary pulses) pro-
duces first- and second-order Wiener kernels sim-
ilar to those obtained by a cross-correlation of the
analog white-noise input and the resulting intra-
cellular responses from ganglion cells. The first-or-
der kernel produced by this direct method is iden-
tical to that produced by the trigger (reverse) corre-
lation (3) although time runs in the opposite
direction.
This methodology allows us to equate the spike
discharges with the intracellular responses with a
recording time of ~40 s. The penalty is that the
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.
AND K.-I. NAKA
analysis is limited only to the first- and second-or-
der components (32). The first-order kernel thus
obtained is the best linear approximation of the
postsynaptic potentials (PSPs) evoked by a brief
depolarizing current injected into a presynaptic
neuron. Similarly, the second-order kernel is inter-
preted as the best second-order nonlinear correc-
tion to the linear estimate of the PSP evoked by
the same stimulus.
In trigger correlation the first-order kernel is the
best linear approximation of the stimulus wave-
form that produces a spike discharge (3). Naka
( 18) computed kernels by crosscorrelating the whi-
te-noise current injected into a preganglionic cell
and the resulting spike discharges from a ganglion
cell, but he did not give any specific interpretation
to the kernels.
Abbreviations
H Horizontal cells. In catfish, there is
only one class of cone horizontal
cells whose maximum sensitivity
is at 625 nm (17).
B Bipolar cells. Bipolar cells are sub-
divided into BA, ON-center, and
BB, OFF-center cells (2 1).
G Ganglion cells. We identify three
kinds of ganglion cell in the cat-
fish retina; GA (ON-center), GB
of (OFF-center), and GC (ON-OFF
transient) cells (28).
Type-N cells A class of amacrine cells that pro-
duce a sustained response to a
flash of light presented in the
dark, and, therefore, have been
called “sustained amacrine cells”
by other investigators ( 10). Type-
N cells are subdivided into type-
NA, depolarizing type, and type-
NB, hyperpolarizing type (2 1,
28).
Type-C cells A class of amacrine cells that pro-
duce ON-OFF transient depolar-
izations to a flash of light (2 1,28)
and, accordingly, have been
called “transient amacrine cells”
by other investigators (9, 44).
Type-C cells are generators of
second-order nonlinearity, which
is represented by the “four-eye”
structure [cf. Sakuranaga and
Naka (35), Fig. 5B].
PRT Peak response time is measured
from time 0 to the peak/valley of
the major phase of a kernel. This
is an indication of the “transport
delay”.
 TRANSMISSION TO RETINAL GANGLION CELLS. I

RESULTS
Identzjication of’cell
. types
The essentialfeatures of the modulation re-
sponsesof catfish retinal neurons can be re-
constructed with a reasonabledegree of accu-
racy by the first- through third-order Wiener
ms
kernels. Moreover, a large part of the nonlin-
ear components can be accounted for by the
second-order rather than by the third-order
model (34, 35). Thus the cells can be identi-
fied functionally basedon their first- and sec-
ond-order kernels (28). In this study, there-
fore, we classified ganglion cells and other in-
terneurons into several subtypes based on I)
the responsesevoked by a flash of light given
in the dark and 2) the first- and second-order
kernels produced by cross-correlation of the
white-noise modulated light and the resulting
cellular responses.
For the extracellular spike discharges,both
GA and GB cells produced well-defined bi- I-----l w
0 100 ms 200 0 IOOms
phasic first-order kernels with an initial posi-
tive phase in GA cells (Fig. 2Al) and a nega- FIG. 2. The first-order kernels (41, BI, and CI) and
tive phase in GB cells (Fig. 2Bl). The noisy second-order kernels (42, B2, and C2) computed by a
first-order kernel of GC cells was usually a cross-correlation of white-noise modulated light against
small two-trough negative wave followed by spike discharges of a ganglion cell. Kernels in A are typi-
cal of a GA cell, those in B of a GB cell, and C of a GC
a positive wave (Fig. 2Cl). Typical second- cell. The peak amplitude of the first-order kernel is 0.6
order kernels generated by GA and GB cells (A), 0.6 (BI). and 0.2 (<‘I) spikescm*/pW. The ampli-
are shown in Fig. 2A2 (GA cells) and in B2 tude of the second-order kernel measured from the valley
(GB cells). These second-order kernels are to the peak was 0.5 (n2), 0.4 (B2), and 0.8 (C2) spikes
s- cm4/pW2.
similar to those produced by type-NA and
-NB cells (cf. Fig. 11 in Ref. 33, Fig. 9 in Ref.
28). GC cells always produced the character-
well as by their characteristic second-order
istic second-order kernel with a “four-eye”
kernels (28, 34). Type-C cells were identified
signature seenin Fig. 2C2, which is identical
by their transient ON-OFF responsesto a flash
to that produced by a type-C cell (28, 35). A of light and by the second-order kernels of the
small number of GA and GB cells generated
“four-eye” signature (28, 35). Bipolar cells
a second-order kernel similar to that shown
produced linear modulation responseswith a
in Fig. 2C2. The similarity of the second-or-
MSE of the first-order model of ~20% (28,
der kernels which characterize ganglion and
33). Any ambiguous cells were excluded from
amacrine cell responsessuggeststhat nonlin-
the analysis.
earity generated by the amacrine cells is
transmitted to ganglion cells without much Transm iss ion -/km amacrine
modification (28). to ganglion cells
In intracellular recordings, ganglion cells Current injected into an amacrine cell in-
were recognized by their spike discharges al- variably modulated the spike dischargesof a
though, as already described by Chan and neighboring ganglion cell. Fig. 3AI shows re-
Naka (2), spike discharges often were lost sponsesof a type-NA cell (upper trace) and a
during intracellular recording. Type-N cells GA cell (lower trace) evoked by a flash of
were identified by oscillatory wavelets of 30- light. Figure 3ri2 shows a lo-Hz sinusoidal
40 Hz superimposed on the depolarizing current of 9 nA peak-to-peak, injected into
phaseof the light-evoked responses(7,28) as the type-NA cell (upper trace) and the result-

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

1554 H. M. SAKAI AND K.-I. NAKA

Al BI
NA NB

GA GB

LIGHT --mm 102s

-

A2

FIG. 3. A 1: simultaneously recorded responses evoked by a flash of light. The bar (LIGHT) indicates the approxi-
mate timing of illumination. Upper trace, type-NA cell; lower trace, spike discharges recorded extracellularly from a
neighboring GA cell. Upper trace in A2: a lo-Hz sinusoidal current injected into the type-NA cell. Lower trace: the
resulting spike discharges from the GA cell. Most of the spikes were evoked at the rising (depolarizing) phase of the
current. A3: the PST histogram of the resulting spike discharges of the GA cell (lower trace) to the current injected
into the type-NA cell (upper trace). Twenty periods were averaged to produce the PST histogram. The unsmooth
contour was due to the limited periods of the summation. The shape of the PSTs is roughly proportional to the input
sinusoid with a phase shift of about + 10” (leading phase). BI: simultaneous recordings from a type-NB cell (upper
trace) and extracellular spike discharges of a neighboring GB cell. Light flash evoked their responses. Upper trace in
B2: a IO-Hz sinusoidal current injected into the type-NB cell. Lower trace: resulting spike discharges of the GB cell.
Most spikes were evoked during the rising (depolarizing) phase of the current. B3: the PST histogram of the resulting
spike discharges from the GB cell (lower trace) for the current injected (upper trace).

ing spike discharges of the GA cell (lower
trace). Spikes were generated during the de-
polarizing phase of the sinusoidal current.
Figure 3A3 shows a PST histogram of the
spike discharges (or spike density distribu-
tion) evoked by a lo-Hz sinusoidal current.
About 20 periods of the shaped spikes were
summed. The waveform of the histogram
roughly followed that of the sinusoid with a
phase shift of about + 10” (leading), but the
negative phasewas clipped. The unsmoothed
contour of the histogram was due mainly to
the limited number of spikes in the summa-
tion and also due to synchronized spike fir-
ing. The maximum amplitude of the histo-
gram was 160 spikes/sto a sinusoidal current
of 9 nA, peak-to-peak.
Figure 3Bl shows responsesof a type-NB
(upper trace) and a GB cell (IOWU trace)
evoked by a flash of light. A sinusoidal cur-
rent injected into the type-NB cell (upper
trace in Fig. 3B2) modulated spike discharges
of the GB cell (lower trace in B2). Spikes were
induced during the depolarizing phase of the
sinusoidal current. The waveform of the PST
histograms was roughly proportional to the
input sinusoid with a phase shift of about
+ 10” (leading) and the maximum amplitude
of the histogram was 160 spikes/sto a current
of 7.5 nA, peak-to-peak (Fig. 3B3).
For both pairs, the signal transmission
measured by a lo-Hz sinusoidal current
was sign-noninverting, i.e., the depolarizing
phase of the injected current evoked spike
dischargesfrom ganglion cells. To character-
ize the transmission in all frequency ranges,

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

 TRANSMISSION
we replaced the sinusoidal current with a
white-noise modulated current and com-
puted the kernels.
The kernels shown in Fig. 4 are for the
transmission from type-NA to GA cells (A)
and type-NB to GB cells (B). Three kernels
from different pairs were superimposed in A
and similarly four kernels in B. The kernel
amplitudes (the values along the vertical axis)
were normalized. The kernels in Fig. 4 are all
biphasic with an initial positive phase (PRT
of -4 ms) followed by a smaller negative
phase. The first-order kernels thus obtained
are the best linear approximation of the post-
synaptic potentials (PSPs) evoked by brief de-
polarizing currents injected into presynaptic
neurons, i.e., a brief depolarizing current
pulse injected into a type-N cell evoked maxi-
mal spike discharges4 ms after the injection
and was followed by a phaseof spike suppres-
sion.
The two pairs of cells shown in Fig. 5 are
type-C and GA cells (A) and type-C and GB
cells (RI). A ~-HZ sinusoidal current injected
into the type-C cells (upper truce in A2 and
B2) modulated the spike dischargesof the GA
(lout trace in A2) and the GB cells (lower
I I 1 I 1 I
b- 50 100 ms 150 0 50 100
FIG. 4. A: first-order kernels produced by a cross-corre-
lation of the white-noise modulated current injected into
a type-NA cell against extracellular spike discharges
(transformed into unitary pulses) from a neighboring GA
cell. Three kernels from different cells were normalized
and superimposed. B: kernels derived from four different
experiments in which spike discharges from GB cells
were evoked by current injected into a type-NB cell. Ker-
nels in A and B are similar with an initial positive (depo-
larizing or an increase in spike discharges) phase followed
by a small negative (hyperpolarizing or a decrease in the
spike discharges) phase. The peak response time is -4
ms. The first-order kernel thus produced is interpreted as
the best linear approximation of the postsynaptic poten-
tials (PSP) induced by a brief depolarizing current in-
jected into a presynaptic amacrine cell. The average
of the peak amplitude of the kernels was 1.44 spikes.
nA-‘.s-’ for /i, and was 1.48 spikesnAp’ .
s-’ for B.
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.
TO RETINAL GANGLION CELLS. I 1555
trace in B2). The PST histograms of the GA
(A3) and of the GB cells (83) are nearly pro-
portional to the input sinusoid with a phase
shift of about + 10” (leading), as in the caseof
the histograms generated by a current in-
jected into type-N cells (Fig. 3, A3 and B3).
Figure 6, A and B, showsthe first-order ker-
nels from GA and GB cells generated by a
white-noise current injected into a type-C
cell. The kernel waveforms were biphasic
with a positive phase (PRT of 4 ms) followed
by a small negative phase and are almost
identical to those shown in Fig. 4. The kernel
waveform is more differentiating (larger neg-
ative phase) for the transmission from type-C
to GA cells (A) than that from type-C to GB
cells (B).
In addition to the fast transmission be-
tween amacrine and ganglion cells illustrated
in Figs. 3-6, we often observed a slower, more
complex form of transmission. In Fig. 7, in-
tracellular recordings were made from a type-
NA (A 1, upper truce), and from a type-NB
cell (BI, uppc-‘r tract); extracellular spike dis-
charges were recorded simultaneously from
two cells, a GA cell with small-amplitude
spikes and a GB cell with large-amplitude
spikes (lmvr truce in Al and BI). A sinusoi-
dal current injected into the type-NA cell (A2,
upper truce) and the type-NB cell (B2, upper
trace) each modulated spike discharges of
both ganglion cells (42 and B2, lower truce).
In both Fig. 7, A2 and B2, most of the small-
amplitude spikes from the GA cell were
evoked during the hyperpolarizing phase of
I the sinusoid current whereas most of the lar-
ms 150
ge-amplitude spikes from the GB cell were
evoked during the transition from depolariz-
ing to hyperpolarizing phase. The PST histo-
grams of spike discharges evoked by currents
injected into type-NA and -NB cells are
shown in Fig. 7, A3 and B3. Note that the
waveforms as well as the phasesof the histo-
grams were similar for the GA cells, with a
phase shift of about -2OO”, and for GB cells
with a phase shift of about -20” (both phases
lagging). The particular patterns of the cur-
rent-evoked dischargesobtained were depen-
dent upon the type ofganglion cell but not on
the type of amacrine cell into which current
was injected. This conclusion will be verified
by comparing current-generated kernels. In
all four cases, the amplitudes of the PSTs
1556 H. M. SAKAI AND K.-I. NAKA

GA GB

LlGHT

A2

W.-L.

A 83

FIG. 5. A I: light-evoked responses recorded simultaneously from a type-C cell (upper trace) and from a GA cell
(lower trace). BI: simultaneous recordings from a type-C cell (upper trace) and from a GB cell (lower trace). The bar
(LIGHT) indicates an approximate timing of the stimulus, a flash of light given in the dark. A2 and B2: the ~-HZ
sinusoidal current injected into the type-C cell (upper traces) and resulting spike discharges of a neighboring GA (A2,
lower trace) and of a GB cell (B2, lower trace). Most of the spikes were evoked at the rising (depolarizing) phase of
the current. A3 and B3: the injected current into a type-C cell (upper trace) and a PST histogram of the resulting spike
discharges from GA (A3, lower trace) and from GB cells (B3, lower trace). The shape of the response histograms
roughly correspond to the input sinusoid with a phase shift of about + 10”.

were - 150 spikes/s to a sinusoidal current of
10 nA, peak-to-peak, and waveforms of the
histograms were roughly proportional to the
input sinusoids. The results illustrated in Fig.
7 demonstrate that:
I) A type-N cell, either of type-NA or -NB,
could simultaneously modulate spike dis-
charges of both GA and GB cells.
2) The phase relationship between the in-
jected current and the resulting spike dis-
charges is not as simple and straightforward
as in the case of the fast transmission shown
in Fig. 3. There are, therefore, two kinds of
transmission from type-N to ganglion cells,
one fast and simple transmission, occurring
in -25% of the measured cell types, but
never between pairs of opposite sign, and the
other slow and complex transmission, occur-
ring in the remaining pairs.
3) The fast transmission was more efficient
(larger gain) than the slow transmission.
A sinusoidal current injected into a type-C
cell simultaneously modulated spike dis-
charges of both GA and GB cells. The PST
histograms show that a phase shift of about
-200” occurred for the transmission from
type-C to GA cells and about -20” for the
transmission from type-C to GB cells. Thus,
as was found for type-N cells, there were two
kinds of transmission from a type-C to GA
and GB cells; one a fast, sign-noninverting
transmission (Fig. 6) and the other a slow and
complex transmission (Fig. 8).
Figure 8 shows the first-order kernels for
the slow signal transmission from amacrine
to ganglion cells. A current injected into
types-NA, -NB, and -C cells produced similar
kernels from GA cells (column A); kernels are

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

 TRANSMISSION TO RETINAL
0 50 100 ms 150 0 50 100 ms 150
FIG. 6. First-order kernels produced by a current in-
jected into a type-C cell and the resulting spike discharges
from a neighboring GA cell are shown in A and those
from a neighboring GB cell is shown in B. Four kernels
from different cells were superimposed. The kernels are
similar to those shown in Fig. 4 with an initial positive
phase followed by a smaller negative phase although the
kernels in A are more differentiating (larger negative
phase) than in B. The average of the peak amplitude of
the kernels was 1.16 spikesnA-* .s-’ for A and 1.52
spikes. nA-’ s-’ for B.
all triphasic with a large negative phase (PRT
of 18-20 ms). A current injected into these
amacrine cells also produced similar kernels
from GB cells (column B); the kernels were
all triphasic with a large positive phase (PRT
of 18-20 ms).
The triphasic kernel with a positive peak
in ccllumn B indicates that maximum spike
dischargeswere evoked from a GB cell 18-20
ms after the injection of a brief depolarizing
current into an amacrine cell; the triphasic
kernel with a negative peak in column A indi-
cates that spike discharges were maximally
depressed 18-20 ms after the current injec-
tion. These findings are in agreement with
our observations made on the PST histo-
grams in Fig. 7, A3 and B3.
Fourier transjbrmat ion of the
-first -order kernels
Fourier transformation of the first-order
kernel gives the gain and phasecharacteristics
of the linear component of transmission. Fig-
ure 9AI shows the fast kernels derived for
transmission from amacrine to ganglion cells
of the samepolarity; the kernel in solid line is
for type-NA + GA cell transmission, while
the kernel in dashed fine is for type-NB -+
GB cell transmission. The two gain curves
(42) are of a lowpass character, with a cutoff
frequency of -30 Hz. The phase curve (A3)
showsthat the phase lag is near zero up to 10
Hz input frequency and becomes progres-
sively larger at higher frequencies. The phase
shift at 5- 10 Hz input is about + lo”, which
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.
GANGLION CELLS. I 1557
matches the phase shift observed in Figs. 3,
A3 and B3, and 5, A3 and B3.
Kernels shown in Fig. 9BI were derived for
slow transmission from type-NB + GA cells
(solid line), and from type-NA + GB cells
(dashed line). The gain curves (B2) are of a
narrow bandpass type; the bandwidth mea-
sured at -3 dB points was -20 Hz, with a
sharp cutoff frequency at -30 Hz (solid line)
LIGHT
A2-
A3
NA
GA
GB
7. Al: light-evoked responses of a type-NA cell
FIG.
(upper ~racc), and of GA and GB cells simultaneously
recorded through a single tungsten electrode (10~~
trace). Similarly, B2 shows light-evoked responses of a
type-NB cell (upper trace), and of GA and GB cells. In
both extracellular recordings (lower trace in A I and BI),
smaller spikes were from a GA cell and larger spikes were
from a GB cell. A ~-HZ sinusoidal current injected into
the type-NA cell (upper trace in A2) and the type-NB cell
(upper trace in 82) modulated spike discharges of both
GA and GB cells (lower trace in A2 and B2). In both A2
and 82, the large-amplitude spikes from GB cells were
evoked at the falling (hyperpolarizing) phase of the cur-
rent whereas most of the small-amplitude spikes from
GA cells were evoked at the trough of the sinusoidal cur-
rent. This was confirmed by the PST histograms in A3
and B3. A3 and B3: the PST histograms produced by a
current injected into a type-NA (43) and type-NB (B3)
cell (upper trace) and of the resulting discharges from GA
(middie trace) and GB cells (bottom trace). Four histo-
grams were obtained from four different current injec-
tion experiments. The phase of the PST histograms of
GA cells is -200” shifted, whereas that of GB cells is -20”
shifted to the sinusoidal current.
1558 H. M. SAKAI AND K.-I. NAKA
A with biphasic waveforms showing a PRT of
I I
0 100 mS 200 0 100 ms 200
FIG. 8. A shows first-order kernels produced by cross-
correlating a white-noise modulated current injected into
a type-C cell (zqper row), type-NA cell (middle yaw), and
a type-NB cell (bottom TOW)against the resulting spike
discharges from a GA cell (column A) and from a GB cell
(dumn B). Four typical kernels are superimposed for
each pair. The kernels produced from GA cells (column
A) have a triphasic waveform with a large negative (hy-
perpolarizing or suppression of the discharges) phase re-
gardless of the type of amacrine cells into which an ex-
trinsic current was injected. Similarly, the kernels pro-
duced from GB cells have a triphasic waveform with a
large positive (depolarizing or an increase in discharges)
phase (column B). The average of the kernel peak ampli-
tude was 0.94 spikesnA%-’ for A, and 1.23 spikes.
nA-’ K’ for B.
and 40 Hz (dashed line), respectively. The
two phase curves (B3) differ by 180”, but the
slopesare similar and become steeper above
10 Hz. The phase shift is about -20” at 5 Hz
for the NA + GB cell transmission and about
-200” for the NB --) GA cell transmission,
which matches the results shown in Fig. 7, A3
and B3.
In order to provide a reference point for
results reported in earlier papers (18, 22), a
white-noise current was injected into a bipo-
lar or a horizontal cell to evoke spike dis-
charges from ganglion cells. The first-order
kernels and their Fourier transformations are
shown in Fig. 10. In Fig. lOAl, the kernel
in solid fine is derived for transmission from
BA -+ GB cells, whereasthe kernel shown by
dashed line is a transmission from BB --) GB
cells. The two kernels are almost identical
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.
6 ms.
The kernels shown in Fig. 10BZ are derived
for transmission from horizontal to ganglion
cells. The kernel is predominantly positive
for a GA cell (solid line) but predominantly
negative for a GB cell (dashed line). One
waveform is approximately the inverse of the
other. These kernels are similar to those re-
ported in an earlier paper ( 18).
The gain curves shown in Fig. lOA for
transmission from bipolar to ganglion cells
(B --) G) are more of a low-pass type with a
cutoff frequency of 30 Hz, whereas the gain
curves in Fig. 10B2 for the transmission from
horizontal to ganglion cells have a sharper
cutoff frequency at 30 Hz. The phase lag in
A3 is small, whereas that in B3 is large, espe-
cially at higher input frequencies. The two
phasecurves in B3 differ by 180”.
In summary, Figs. 9 and 10 show the fol-
lowing.
I) The kernel for signal transmission from
bipolar to ganglion cell is similar to that for
the fast transmission from amacrine to gan-
glion cells; Transfer functions are lowpass
with a cutoff frequency at 30 Hz, and the
phase lag is minimal.
2) Signal transmission from amacrine to
ganglion cells of the opposite polarity (NA --+
GB and NB + GA) is always slow and com-
plex.
3) The gain curve for the slow transmission
from amacrine to ganglion cells of the oppo-
site polarity (NA + GB or NB -+ GA) is
characterized by a narrow bandpass with a
sharp cutoff, whereas that from a horizontal
to ganglion cells is of a more low-pass nature
and has a more gradual cutoff.
Figure 11 showsthe PRTs of the first-order
kernels counted in 2-ms bins for the signal
transmission from amacrine cells (366 ker-
nels in A), from bipolar ( 14 kernels in B) and
horizontal cells (60 kernels in B) to ganglion
cells. The PRTs for the amacrine + ganglion
cell transmission fall into two clusters that are
clearly demarcated. One cluster is for the fast
transmission with an average PRT of 5 ms
and the second cluster for the slow transmis-
sion with an average PRT of 2 1 ms. PRTs for
these two clusters did not overlap. Note that
the ordinate scales (number of pairs) are
different for the two clusters in A. The PRTs
 TRANSMISSION TO RETINAL GANGLION CELLS. I 1559

1 I 1 1
0 50 100 ms 150 0 50 100 ms I50

A2 1 B2
1
m m
E 0"
nt N

1 lrlrl 1 IIIrlrn t 1 rrr,,,,, 1 II,,,q-l

I IO Hz 100 I IO Hz 100
A3
1

-180
1

-360
!
t I Ill I l-l I I I’1’Il 1 LlIwl
I IO Hz 100 I IO Hz IO0

FIG. 9. Al: kernels derived for transmission from type-NA to GA cell (solid line) and from type-NB to GB cell
(dashed line). Two kernel waveforms are almost identical. Fourier transformation of the kernel gave the gain (A2)
and phase (A3) characteristics. The gain curves of both kernels have a 30-Hz cutoff frequency (42). The phase lag is
small up to 10 Hz input frequency, whereas the curve becomes steep at higher frequencies. B 1 shows kernels derived
for transmission from type-NB to GA cells (solid line) and from type-NA to GB cells (dashed line). The gain curves
of both kernels are bandpass limited with a sharp cutoff frequency of 30-40 Hz (B2). but the phase (B3) is 180”
different from each other and the curve is steeper than that in A3.

for the bipolar + ganglion cell transmission The table is composed of four sections that
was comparable to those for the fast ama- correspond to the four PRT clusters in Fig.
crine --) ganglion cell transmission. Horizon- 11. We observed 55 casesof fast amacrine-
tal to ganglion cell transmissions, on the to-ganglion cell transmission, including those
other hand, showed PRTs slightly shorter from type-C to GA and GB cells; and all were
than those for the AM + G slow transmis- represented by biphasic first-order kernels.
sion. The PRTs are similar to the transmis- We observed 3 11 examples of slow transmis-
sion reported by Naka ( 18). sion from amacrine to ganglion cells, each of
Table 1 summarizes all the results of this which had a triphasic first-order kernel. Po-
study. The first-order kernels are character- larity “+” indicates that the major peak of the
ized by their polarity, PRT, and amplitude. kernel was positive (depolarizing or an in-

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

1560 H. M. SAIL41 AND K.-I. NAKA

BI
B-G

\ I
, I
\ ,
\ 1
I
I :
\ I
\ I
\ I
‘s’

I I I I I I

0 50 100 ms 150 0 50 100 ms 15 0

A2 B2

1 rrrr,,,, , 1 IIn

I IO Hz 100 I IO Hz 100

A3 B3
1 1
go -
b
%
- 180 -180-----
1

1 -360I
-““iw I IO Hz ’ 100
t

I
, , 1, ,111,

IO
, , 1

Hz
,rr,rj

I00

FIG. 10. The kernels in AI are derived by transmission from BA to GA cell (solid line) and from BB to GB cell
(dushed line). The waveforms of 2 kernels are almost identical. The gain curves of both kernels have a 30-Hz cutoff
frequency (A2). The phase lag is small, up to 10 Hz, and becomes slightly steeper (43). Kernels in BI are obtained
for transmission from horizontal to GA cell (solid line) and to GB cell (dashed line). The peak response of the kernel
is positive for a GA cell, whereas negative for a GB cell. The gain curves (B2) of the 2 kernels (solid and dashed lines)
have a sharper cutoff frequency of - 30 Hz. The phase curves (B3) of the 2 kernels differ from each other by 180”.

crease in spike discharges) and polarity “-”
indicates that the peak was negative (hyper-
polarizing or a decrease in spike discharges).
One of the remarkable results of this study
was that, in essentially all of the pairs exam-
ined, extrinsic current injected into an ama-
crine cell modulated spike discharges of a
neighboring ganglion cell; very few failures
were recorded. About one-fourth of the pairs
of amacrine to ganglion cells of the same re-
sponse polarity (NA + GA, NB --) GB) and
of type-C + GA or GB cells produced a fast
biphasic kernel with an average PRT of -4.8
ms. The remaining three-fourths produced a
slow triphasic kernel with an average PRT of
21.5 ms.
A kernel produced by transmission from
amacrine to ganglion cell of opposite polarity
(type-NA + GB, type-NB + GA) was always
triphasic with a slow PRT; none of these pairs
produced a biphasic fast kernel. The polarity
of the peak phase of the triphasic kernel was
always negative for GA cells but invariably
positive for GB and GC cells. The peak am-

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

 TRANSMISSION TO RETINAL GANGLION CELLS. I 1561

Evaluation oj’nonlinearities
-200
. . .
:::. In order to evaluate nonlinearities that
.:::.
... Am-G
.. : .: I.: .: *. .:a
.

.:.I.:.
..:::
**:<
-...........I.:...
.. . .
.......:::.
..-.*.a
......
*.
.:::
*.*::
.*...:::.
.*::: . ...
.:.*.a
-::: :::.
:::.
m
21
PEAK RESPONSE
FIG. Il. A: distribution of PRTs (peak response time)
of the first-order kernels counted in 2-ms bins for signal
transmission from amacrine to ganglion cells (Am LG);
B: from bipolar to ganglion cells (B + G) and from hori-
zontal to ganglion cells (H + G). Peak response times
of the kernels for amacrine-to-ganglion cell transmission
fall into 2 clusters. Note that in A the scale of ordinate
(number of pairs) is different for the two subgroups.
plitude of the kernel produced by the fast
transmission was larger than that produced
by the slow transmission. For both fast and
slow kernels, the peak amplitude produced
by transmission to GA cells was, in average,
smaller than that produced by the transmis-
sion to GB cells, i.e. the former transmission
was more efficient (larger gain) than the lat-
ter one.
The average PRT of the kernel was 16 ms
for horizontal to GA cells and 14 ms for hori-
zontal to GB cells, while the average PRT for
bipolar to ganglion cells was 6.5 ms. These
results are similar to those reported by Naka
(18). Note that kernel amplitude was largest
for fast transmission from amacrine to gan-
glion cells and for transmission from hori-
zontal to ganglion cells (- 1.5 spikes/nA/s).
As may seenin Figs. 4, 6, and 8, the PRTs
derived from the same transmission pairs
varied to some extent; the kernel waveform
and the degree of its differentiation also vary.
might be produced during the signal trans-
mission from preganglionic cells to ganglion
cell spike discharges, second-order kernels
.I00
were computed for each pair of cells. Regard-
lessof the type of preganglionic cell, well-de-
fined second-order kernels were derived from
.O
most of the transmissions to GB cells,
ms whereas many of the transmission to GA cells
failed to produce well-defined second-order
3n
t-c” kernels; i.e., the nonlinearity produced dur-
ing the transmission had a very small second-
order component. Figure 12 shows examples
of the contour plot of the second-order kernel
-IO obtained by a crosscorrelation of the spike
discharges of a GB cell against a current in-
jected into a type-NB cell (A) and into a type-
NA cell (B). The first-order kernel for the pair
1 -0
ms 31
in A was a biphasic, fast type, whereas that
in B was a triphasic, slow type. The contour
TIME
indicates the amplitude of the kernel; the
solid fines are for an increase in spike dis-
chargesand the dotted lines are for a decrease.
The second-order kernel in Fig. 12A was
characterized by a single peak followed by a
small valley elongated perpendicular to the
diagonal. The second-order kernel in Fig.
12B was a four-eye signature with the maxi-
mum magnitude at the second peak. A trans-
mission from BB to GB cells produced a sec-
ond-order kernel similar to the one shown in
Fig. 12A.
The characterization and interpretation of
the second-order nonlinearity produced dur-
ing the transmission from a preganglionic cell
to a ganglion cell was attempted by compar-
ing the PST histograms with responsespre-
dicted from the kernels, by convolving the
kernels with the stimulus input. As the shapes
of PST histograms show (Figs. 3, 5, and 7)
the relationship of the input sinusoid and the
output spike discharges was approximately
linear. We found that the small amount
of nonlinearity produced was attributable
mainly to two properties of spike discharges;
the firing rate is positive only (rectification),
and it has an upper limit (saturation). Re-
gardlessof the type of preganglionic cells, the
contribution of the second-order nonlinear-
ity was classified into two kinds as is shown
in Fig. 13, A and B. In Fig. 13A, a ~-HZ sinu-
soidal current injected into a type-NA cell

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

 1562 H. M. SAIL41 AND K.-I. NAIL4

TABLE 1. Current-evoked kernels

Peak Response Time, Amplitude,
Pair n Polarity ms Spikes. nA-’ . s-’

I. Amacrine to Ganglion Cells (biphasic kernel waveform)

NA-GA 20 + 4.0 IL 1.8 1.4 * 0.22
NB-GB 25 + 5.7 t 0.6 1.5 + 0.65
C-GA 4 + 4.0 + 0.0 1.2 + 0.02
C-GB 6 + 4.5 2 0.9 1.5 + 0.41
II. Amacrine to Ganglion Cells (triphasic kernel waveform)
NA-GA 64 - 20.8 + 3.1 0.9 + 0.59
NB-GA 19 - 22.3 X!I3.1 0.9 + 0.47
C-GA 22 - 21.9 + 3.2 0.9 + 0.38
NB-GB 57 + 20.7 + 3.2 1.2 + 0.68
NA-GB 109 + 22.1 + 3.4 1.O + 0.52
C-GB 16 + 21.4 + 1.6 1.5 + 0.74
NA-GC 12 + 21.7 + 2.4 0.7 k 0.43
NB-GC 3 + 19.7 + 0.5 0.8 IL 0.13
C-GC 9 + 21.5? 1.7 0.9 + 0.26

III. Bipolar to Ganglion Cells

BA-GA 5 + 6.6 I!I 0.9 1.o * 0.35
BB-GB 8 + 6.4 IL 1.7 1.2 & 0.23
BA-GB 1 + 22.0 1.2

IV. Horizontal to Ganglion Cells

H-GA 34 + 16.4 + 2.8 1.5 + 0.32
H-GB 26 + 14.4 -+ 2.4 1.5 + 0.28

(top trace marked “INPUT”) produced a
PST histogram from a GA cell (bottom trace
marked “PST”). The response predicted
from the first-order kernel (marked “ 1st Pre-
diction”) and by the second-order kernel
(marked “2nd Prediction”) are depicted. The
sum of the predicted first-order and second-
order responsesis plotted in the bottom trace
(dotted lines) together with the measured PST
histogram (solid lines). They match well. The
second-order nonlinearity in the prediction
A B
INPUT
w w
1st PredictIon

2nd PredIctIon m

EPr-edlctlon m E
,0.2s,

FIG. 13. A and B: INPUT (~01)) indicates a ~-HZ sinu-

0 50 ms 10’0 0 50 ms Id0 soidal current injected into a presynaptic amacrine cell
71 71 and PST (hottoln) indicates a PST histogram from spike
discharges of a ganglion cell. 1st Prediction indicates a
FIG. 12. A: shows a contour plot of the second-order response to the sinusoid predicted from its first-order ker-
kernel produced by a signal transmission from type-NB nel and 2nd Prediction indicates a response predicted
to GB cell. The first-order kernel of the transmission was from its second-order kernel. The sum of the response
a fast biphasic type. B: second-order kernel produced by predicted from the first- and second-order kernel is plot-
a transmission from a type-NA to GB cell. The first-order ted in &tt& lines ( I+2 Prediction) together with a PST
kernel of the transmission was a slow triphasic type. The histogram in the bottom. A pair in A is derived by a trans-
amplitude of the second-order kernel measured from the mission from a type-NA cell to GA-cell spike discharges
valley to the peak was 3.2 (A) and 2.8 (B) spikes and a pair in B is from a type-NA cell to GB-cell spike
nAm2. sm2, respectively. discharges.

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

 TRANSMISSION TO RETINAL GANGLION CELLS. I 1563

contributes to saturating the peak of the PST ing transmission was obtained from some
histograms. The small second-order nonline- type-C to GA and GB cells.
arity observed in some of the GA cells was 2) Slow transmission was measured be-
attributed to the saturation of the spike dis- tween the remaining 75% pairs of the same
charges. The MSEs of the first-order predic- response polarity (type-NA + GA, type-
tion was 18% and that of the second-order NB ---) GB). Similar slow transmission was
prediction to the PST was 0.6%. obtained between all cells of the different po-
In Fig. 13B, on the other hand, a sinusoidal larity (type-NA + GB, type-NB + GA) and
current injected into another type-NA cell from some type-C to GA and GB cells.
(INPUT) produced a PST histogram from a We found that in many pairs oftype-NA to
GB cell (hottom truce marked “PST”). The GA cells and of type-NB to GB cells exhibit-
sum of the predicted first- and second-order ing fast, sign-noninverting transmission, the
responses to the sinusoid is superimposed light-evoked responses characterized by the
with the PST histogram. The prediction is first- and second-order kernels were similar.
shown in dash& line marked “ 1+2 Predic- Figure 14 illustrates this comparison. In Fig.
tion.” In this case, the second-order predic- 14A, the first-order kernel (s~/id line) is com-
tion shows a rectification of the response. The puted from the white-noise modulated light
MSEs of the first-order prediction was 20% and the responses from a type-NA cell; the
and that of the second-order prediction to the dotted lint shows a comparable kernel from a
PST was 2.2%. Similarly, in most of the pairs
of horizontal, bipolar or amacrine cells to GB
cell spike discharges, the contribution of the
second-order nonlinearity to the total re-
sponse was accounted for by a half-wave rec-
tification.
The PST histogram from GB cells elicited
by sinusoidal current was half wave rectified
(Figs. 5B3 and 7B3) whereas the histogram of
the GA cells was slightly saturating (Figs. 543
and 7A3). This difference results because
most of GA cells fired spontaneously whereas 0 100 200 ms 300
GB cells were usually silent in darkness, dur-
ing which current injections were carried out. BI B2

DISCUSSION
8-
1 0
Whenever an intracellular recording was
firmly established, current injected through
the intracellular electrode into a pregangli-
onic cell almost always modulated spike dis-
charges from a nearby ganglion cell. Indeed, 0 50 IOOms I50 0 50 100 ms I50
ganglion-cell spike discharges could be gener- FIG. 14. The light-evoked kernels of a type-NA cell
ated by current injected successively into sev- and extracellular spike discharges ofa GA cell are shown.
eral preganglionic cells located with a radius This pair produced a current-evoked kernel with a bipha-
of 0.4 mm. This was a remarkable finding sic waveform of 4-ms PRT, similar to the one shown in
Fig. 4. The light-evoked first-order kernel in A (solid lint)
suggesting that neurons in the inner retina are
and the second-order kernel in BI were produced from
extensively interconnected. a type-NA cell: the light-evoked first-order kernel in .4
In this set of experiments, we found the fol- (dushdinc) and the second-order kernel in B2 were pro-
lowing. duced from spike discharges of a GA cell. The peak am-
I) Fast and sign-noninverting signal trans- plitude of the first-order kernel in w/id lint was 0.8
mV+cm’/~W. and that in dotted line was 0.5
mission was found between ~25% of pairs of spikescm’/pW. The amplitude of the second-order ker-
the same response polarity (type-NA + GA, nel, measured from the valley to the peak was 0.9
type-NB + GB). Similar fast, sign-noninvert- mVs’~cm”/~W’(HI), and 0.8 spikcsscm4/~W2 (BZ).

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

1564 H. M. SAL41
GA cell. The kernel waveforms are damped
oscillation of similar shapes having an initial
positive phase, although they differ as to the
degree of differentiation. Both the type-NA
and GA cells generated similar but not identi-
cal second-order kernels characterized by a
series of alternating valleys and peaks whose
major axes were perpendicular to the diago-
nal (BI for type-NA and B2 for GA cell). A
fast, sign-noninverting transmission was re-
corded between these type-NA and GA cells,
as shown in Figs. 3 and 4. Thus linear and
second-order nonlinear dynamic compo-
nents produced by the type-NA cell were
transmitted to the GA cell without much al-
teration.
Based on the facts that only a relatively
small number of amacrine and ganglion cell
pairs showed fast transmission and that these
pairs of cells generated almost identical first-
and second-order kernels, we speculate that
responses from these ganglion cells produced
by a simple light stimulus such as a tempo-
rally modulated field of light are determined
by a subset of amacrine cells having the same
response characteristics. The contribution of
light-evoked inputs from amacrine cells of
other subtypes was small or absent. A possi-
ble synaptic organization accounting for this
transmission may be either I) synapses made
by a small number of amacrine cells close to
the spike initiation site determine the gan-
glion-cell response characteristics or 2) a large
number of synapses were made onto a gan-
glion cell by a small number of amacrine
cells.
In the following paper (30) we provide evi-
dence that ganglion cells of the same response
polarity, i.e., GA + GA and GB --) GB cells,
interact with each other through a fast trans-
mission pathway. At least some of such trans-
missions are likely to be monosynaptic. How-
ever, the synaptic organization between gan-
glion cells seems to be different from that
between amacrine and ganglion cells.
Anatomical evidence in catfish and other
retinas shows that chemical synapses are
made by amacrine and bipolar cells onto gan-
glion cells (4, 5). Therefore, it is reasonable to
assume that the PRT of 2-8 ms accounts for
the transmission through one chemical syn-
apse. Naka (18) assigned a PRT of 7.5 ms to
a monosynaptic transmission.
Sakuranaga and Naka (33) have shown
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.
AND K.-I. NAKA
that the signal transmission between horizon-
tal cells via gap junctions was instantaneous
and the PRT for such transmission was <2
ms, which was the limit of the time resolution
of their analysis. The average PRT of the ker-
nel for horizontal to ganglion cells was 14 ms
(for GB cells) and 16 ms (for GA cells). Ana-
tomical and physiological evidence suggests
either two (27) or three (29) chemical syn-
apses are involved in this transmission.
Slow transmission was observed in many
pairs with the same light response polarity
(type-NA -+ GA, type-NB + GB) and type-
C to either GA or GB cells. The first- and sec-
ond-order kernels generated by these pre- and
postsynaptic cells were different, whereas the
kernels were similar for the pairs showing fast
transmission. The transmission between the
cells of different response polarity was always
slow, and we found no exception to this rule.
The kernel waveforms for such slow trans-
missions were similar regardless of the types
of amacrine cell into which extrinsic current
was injected. The polarity of the transmission
was determined by the ganglion cells; the
transmission was sign inverting for GA cells
and sign-noninverting for GB cells. The long
PRT of the kernel (an average of 21.5 ms)
suggests that at least three chemical synapses
are involved in this transmission. The follow-
ing evidence in the catfish retina suggests that
the transmission must be mediated by type-
C cells.
Anatomical findings in catfish retina (2,
24) are that the dendrites of type-C amacrine
cells are bistratified or diffusely distributed in
the inner plexiform layer whereas the den-
drites of type-N cells are monostratified and
tend to be restricted to a single sublayer. In
the catfish retina, Naka and Christensen (20)
have shown that type-C cells are electrically
coupled via gap junctions. In the carp retina
electrical coupling among type-C cells is sug-
gested by the Lucifer yellow dye-coupling oc-
curring among transient (type-C) amacrine
cells (38). A similar “dye-coupling” was also
established previously in horizontal cells (37).
It is likely that type-C cells form a syncytial
compartment in the inner retina as the hori-
zontal cells do in the outer retina (25). Thus
type-C cells are one of the possible candidates
that mediate interactions between type-NA
and type-NB cells through the bistratified
dendrites and the electrically coupled syncy-
 TRANSMISSION TO RETINAL
tium. Here we assume that synaptic contacts
are made reciprocally between type-C and
type-N cells; morphological evidence for such
contacts is available (Sakai and Naka, unpub-
lished observation). Likewise, type-C cells
can mediate, through their electrical cou-
pling, interactions between type-N cells of the
same polarity but separated by some distance
(NA -+ type-C + NA and NB + type-
C + NB).
On the bases of these anatomical and elec-
trophysiological findings, we speculate that a
current injected into a type-N cell may spread
into type-C cells, which, in turn, drive other
type-N cells that make extensive or highly
efficient synapses onto the recorded ganglion
cell. In some cases, type-C cells may evoke
spike discharges of GA and GB cells directly
without intervening type-N cells, as is sug-
gested by the light-evoked kernels (28).
Another important finding in this study
was that we rarely failed to record changes in
ganglion-cell spike discharges to current in-
jected either into a horizontal or a pregangli-
onic cell. This may be attributed to the S-
space formed by horizontal cells in the outer
retina and to another space formed by type-C
amacrine cells via gap junctions in the inner
retina. Current injected into a horizontal cell
spreads across many horizontal cells (33) via
gap junctions (45,47) which in turn polarize
the membrane of a large number of bipolar
and amacrine cells resulting finally in the pro-
duction of spike discharges in multiple gan-
glion cells. Similarly, extrinsic current in-
jected into an amacrine cell of any type may
be transmitted to type-C amacrine cells,
which in turn polarize the membrane of a
large number of type-NA and type-NB cells
and elicit spike discharges in many ganglion
cells of GA, GB, and GC types. Thus, type-C
cells may be extensively involved in the signal
interaction between on-type and off-type
amacrine and ganglion cells.
Results of this study have confirmed the
following findings by Naka (18).
1) Current injected into a horizontal cell
modulates spike discharges of both GA and
GB cells. Transmission to GA cells is sign-
noninverting and that to GB cells is sign in-
verting.
2) Current injected into a bipolar cell mod-
ulates spike discharges of the ganglion cells of
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.
GANGLION CELLS. I 1565
the same polarity, and the transmission is
sign-noninverting.
3) Current injected into a type-NA cell
modulates spike discharges of both GA and
GB cells. Transmission from a type-NA cell
to a GA cell is sign inverting whereas to a GB
cell is sign-noninverting.
There are, however, several important dis-
crepancies between the results obtained by
Naka (18) and those presented here: I) Naka
failed to record fast, sign-noninverting trans-
mission from amacrine to ganglion cells of
the same polarity, 2) Naka failed to record the
effects on ganglion cells from current injected
into a type-C cell, and 3) the sign of signal
transmission from a type-NB cell to a gan-
glion cell in Naka (18) is opposite to that
found in the present study, i.e., Naka re-
ported that signal transmission from a type-
NB cell to a GA cell was sign-noninverting,
and from a type-NB to a GB cell was sign-
inverting. These differences may be ex-
plained by the improvements made during
10 yr in the recording as well as cell identifi-
cation procedures. Naka (18) identified the
cells based on criteria established in an earlier
study by Naka et al. (2 1) in which the type-N
cells were assumed to produce linear re-
sponses without any characteristic nonlinear
components. Recent papers (28, 34) showed
that these cells produced a very marked non-
linear response. In the earlier white-noise
analysis, the retina was not fully light
adapted, a condition necessary for evoking a
steady state dynamic response (cf. Fig. 6 in
Sakai and Naka, ref. 28). In the case of type-
NB to ganglion cell transmission, the identi-
fication of type-NB cells by Naka (18) might
be incorrect. It is probable that he identified
incorrectly type-NB cells as horizontal cells
because the light-evoked responses shown in
the figure lacked the oscillatory wavelets
characteristic of type-NB cells and are sus-
tained like horizontal cell responses are.
In this study, we have analyzed the degree
as well as characteristics of the nonlinearity
produced during signal transmission from
amacrine and bipolar cell to ganglion cells.
Predictions from the first- and second-order
kernels have shown that large portions of the
PST histograms generated by sinusoidal in-
puts are made of the first-order linear compo-
nent and that its small second-order compo-
nent is attributable either to response satura-
1566 H. M. SAKAI AND K.-I. NAKA

tion (spontaneously firing cells), or to half-
wave rectification (silent cells). Similar con-
clusions were obtained in the analysis of spike
discharges in the goldfish retina (36, 41).
However, it remains to be seen whether such
a second-order nonlinearity can entirely be
accounted for in this way.
It is often assumed that the major role of
(transient) amacrine cells is to provide inhibi-
tory inputs to ganglion cells (6, 16, 39, 43,
46). However, from the analysis of the first-
and second-order kernels produced by white-
noise modulated light, we have concluded
that the signals from amacrine cells were
transmitted to ganglion cells with little alter-
ation, i.e., the transmission must be fast and
sign-noninverting (28). The results from our
current injection experiments have provided
direct evidence for this conclusion. Note that
there are several lines of evidence in other ret-
inas, as well, that some amacrine cells pro-
vide excitatory inputs to ganglion cells (1,
8, 15).
In the catfish retina, the center-surround
receptive-field organization of ganglion cells
is established by horizontal-bipolar cell inter-
actions (22, 23, 33). This notion also is sup-
ported by certain results of the present study,
viz, I) the signal transmission from bipolar to
ganglion cells is sign-noninverting and 2) the
signal transmission from horizontal cell to
GA cell is sign-noninverting, whereas to GB
cell it is sign-inverting (cf. diagram, Fig. 1 1 in
Ref. 30). Thus our more general conclusion
is that, in the catfish retina, amacrine cells are
generators of second-order nonlinearities (28,
29,34,35). Their functional role is to provide
nonlinear signals to ganglion cells.
Sakuranaga and Naka (34, 35) reported
that current injected into horizontal cells pro-
duced characteristic linear as well as nonlin-
ear responsesfrom the amacrine cells. Their
results showed that signals in the horizontal
cells were transmitted to amacrine cells. On
the other hand, observations of the waveform
of the first-order kernels in this study haspro-
vided two sets of evidence: I) one-fourth of
the transmission from amacrine to ganglion
cell was fast, whereas the remaining three-
forths was slow and triphasic and 2) the mode
of signal transmission from the horizontal to
ganglion cells was only of one kind. It is very
likely, therefore, that current injected into the
horizontal cells reaches the ganglion cells ei-
ther directly through the bipolar cells or
through the fast pathway via amacrine to gan-
glion cells. Although there must be several al-
ternate or parallel pathways of signal trans-
mission from horizontal to ganglion cells, our
results indicate that, under certain condi-
tions, signals may reach the ganglion cells
through a given pathway. An intriguing pos-
sibility is that there may be a switching mech-
anism in the neuron network to select one
pathway out of many available parallel ones
according to the conditions.
Another intriguing possibility is that sig-
nals in the ganglion cells are transmitted back
to the amacrine cells. Morphological evi-
dence is available for such a pathway (3 1). We
will addressthis question in the next seriesof
papers in which two intracellular electrodes
will be used to identify the mode of signal
transmission among neurons in the catfish
inner retina.
ACKNOWLEDGMENTS
We thank P. Witkovsky and M. Sakuranaga for their
valuable comments on the manuscript and Y.-I. Ando
for his programming assistance.
Received 20 October 1987: accepted in final form 7
June 1988.

REFERENCES
1. ARIEL, M. AND DAW, N. W. Effects of cholinergic Synaptic structure. J. Nerrrophysio/. 32: 3 15-338,
drugs on receptive field properties of rabbit retinal 1969.
ganglion cells. J. Phjkol. Land. 324: 135- 160, 1982. 6. FRUMKES, T. R., MILLER, R. F., SLAUGHTER, M.,
2. CHAN, R. Y. AND NAKA, K.-I. The amacrine cell. AND DACHEUX, R. F. Physiological and pharmaco-
Vi.yiun Rcs. 16: 1 1 19- 1 129, 1976. logical basis of GABA and glycine action on neurons
of mudpuppy retina. III. Amacrine-mediated inhibi-
3. DE BOER, E. AND KUYPERS, P. Triggered correla-
tory influence on ganglion cell receptive-field organi-
tion. IEEE Trans. Biomed. Eng 15: 169-179, 1968.
zation: a model. J. N~wophysiol. 45: 783-804, 198 1.
4. DOWLING, J. E. AND BOYCOTT, B. B. Organization 7. HOSOKAWA, Y. AND NAKA, K.-I. Spontaneous
of the primate retina: electron microscopy. Proc. R. membrane fluctuation in catfish type-N cells. Vision
SW Lond. B Biol. Sri. 166: 80- 1 1 1, 1966. Rus. 25: 539-542, 1985.
5. DOWLING, J. E. AND WERBLIN, F. S. Organization 8. JENSEN, R. L. AND DEVOE, R. D. Ganglion cells and
of retina of the mudpuppy, Necturus maculosus. I. (dye-coupled) amacrine cells in the turtle retina that

Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.

 TRANSMISSION TO RETINAL GANGLION CELLS. I 1567

have possible synaptic connection. Brain Res. 240: 28. SAKAI, H. M. AND NAKA, K.-I. Signal transmission
146-150, 1982. in catfish retina. IV. Transmission to ganglion cells.
9. KANEKO, A. Physiological and morphological iden- J. Neurophysiol. 58: 1307- 1328, 1987.
tification of horizontal, bipolar and amacrine cells 29. SAKAI, H. M. AND NAKA, K.-I. Signal transmission
in goldfish retina. J. Physiol. Lond. 207: 623-633, in catfish retina. V. Sensitivity and circuitry. J. Neu-
1970. rophysiol. 58: 1329- 1350, 1987.
10. KANEKO, A. Receptive field organization of bipolar 30. SAKAI, H. M. AND NAKA, K.-I. Dissection of the
and amacrine cells in the goldfish retina. J. Physiol. neuron network in the catfish inner retina. II. Inter-
Land. 235: 133-l 53, 1973. actions between ganglion cells. J. Neurophysiol. 60:
11. MAKSHIMOVA, Y. M. Effect of intracellular polar- 1568-1583, 1988.
ization of horizontal cells on the activity of the gan- 31. SAKAI, H. M., NAKA, K.-I., AND DOWLING, J. E.
glion cells of the retina of fish. Biophysics 14: 570- Ganglion cell dendrites are presynaptic in catfish ret-
577, 1970. ina. Nature Land. 3 19: 495-497, 1986.
12. MANGEL, S. AND MILLER, R. F. Horizontal cells 32. SAKURANAGA, M., ANDO, Y.-I., AND NAKA, K.-I.
contribute to the receptive field surround of ganglion Dynamics of ganglion-cell responses in the catfish
cells in the rabbit retina. Brain Res. 4 14: 182-l 86, and frog retinas. J. Gcn. Physiol. 90: 229-259, 1987.
1987. 33. SAKURANAGA, M. AND NAKA, K.-I. Signal trans-
13. MARCHIAFAVA, P. L. Horizontal cells influence mission in catfish retina. I. Transmission in the outer
membrane potential of bipolar cells in the retina of retina. J. Neurophysiol. 53: 373-389, 1985.
the turtle. Nature Land. 275: 14 l- 142, 1978. 34. SAKURANAGA, M. AND NAKA, K.-I. Signal trans-
14. MARMARELIS, P. Z. AND NAKA, K.-I. Nonlinear mission in catfish retina. II. Transmission to type-N
analysis and synthesis of receptive-field responses in cells. J. Neurophysiol. 53: 390-4 10, 1985.
the catfish retina. I. Horizontal cell to ganglion cell 35. SAKURANAGA, M. AND NAKA, K.-I. Signal trans-
chain. J. Neurophysiol. 36: 605-6 18, 1973. mission in catfish retina. III. Transmission to type-
15. MASLAND, R. H., MILLS, J. W., AND CASSIDY, C. C cells. J. Neurophysio/. 53: 4 1 l-428, 1985.
The functions of acetylcholine in the rabbit retina. 36. SCHELLART, N. A. M. AND SPEKREIJSE, H. Dy-
Proc. R. Sot. Land. B Biol. Sci. 223: 12 I- 139, 1984. namic characteristics of retinal ganglion cell re-
16. MILLER, R. F. AND DACHEUX, R. F. Synaptic orga- sponses in goldfish. J. Gen. Physiol. 59: l-2 1, 1972.
nization and ionic basis of on and off channels in 37. TERANISH, T., NEGISHI, K., AND KATO, S. Dopa-
mudpuppy retina. III. A model of ganglion cell re- mine modulates S-potential amplitude and dye-cou-
ceptive field organization based on chloride-free ex- pling between external horizontal cells in the carp
periments. J. Gen. Physiol. 67: 679-690, 1976. retina. Nuture Lond. 30 1: 243-246, 1983.
17. NAKA, K.-I. Computer assisted analysis of S-poten- 38. TERANISHI, T., NEGISHI, K., AND KATO, S. Corre-
tial. Biophys. J. 9: 845-849, 1969. lations between photoresponse and morphology of
18. NAKA, K.-I. Functional organization of the catfish amacrine cells in the carp retina. Neurosci. Res.
retina. J. Neurophysiol. 40: 26-43, 1977. Sz4ppl. 2: 2 1 l-226, 1985.
19. NAKA, K.-I., CHAN, R. Y., AND YASUI, S. Adapta- 39. THIBOS, L. N. AND WERBLIN, F. S. The response
tion in catfish retina. J. Neurophysiol. 42: 44 l-454, properties of the steady antagonistic surround in the
1979. mudpuppy retina. .I. Physiol. Land. 278: 79-99,
20. NAKA, K.-I. AND CHRISTENSEN, B. N. Direct electri- 1978.
cal connections between transient amacrine cells in 40. TOYODA, J.-I.-AND TONOSAKI, K. Effect of polariza-
the catfish retina. Science Wash. DC 2 14: 462-464, tion of horizontal cells on the on-center bipolar cell
1981. of the carp retina. Nuture Land. 276: 399-400, 1978.
21. NAKA, K.-I., MARMARELIS, P. Z., AND CHAN, R. Y. 41. VICTOR, J. D. The dynamics of the cat retinal X cell
Morphological and functional identification of cat- centre. .I. Physiol. Land. 386: 2 19-246, 1987.
fish retinal neurons. III. Functional identification. J. 42. VICTOR, J. D. AND SHAPLEY, R. M. The nonlinear
Ncurophysiol. 38: 92- 13 1, 1975. pathways of Y ganglion cells in the cat retina. .I. Gen.
22. NAKA, K.-I. AND NYE, P. W. Receptive-field organi- Physiol. 74: 67 l-689, 1979.
zation of the catfish retina: are at least two lateral 43. WERBLIN, F. S. Regenerative amacrine cell depolar-
mechanisms involved? J. Neurophysiol. 33: 625- izations and formations of on-off ganglion cell re-
642, 1970. sponse. J. Physiol. Lond. 264: 767-785, 1977.
23. NAKA, K.-I. AND NYE, P. W. Role of horizontal cells
44. WERBLIN, F. S. AND DOWLING, J. E. Organization
in organization of the catfish retinal receptive field.
J. Neurophysiol. 34: 785-80 1, 197 1. of the retina of the mudpuppy, Necturus maculosus.
24. NAKA, K.-I. AND OHTSUKA, T. Morphological and II. Intracellular recording. J. Neurophwv.siol.32: 339-
355, 1969.
functional identifications of catfish retinal neurons.
II. Morphological identification. J. Neurophysiol. 45 WITKOVSKY, P., OWEN, W. G., AND WOODS-
38: 79-91, 1975. WORTH, M. Gap junctions among the perikarya,
25. NAKA, K.-I. AND RUSHTON, W. A. H. The organiza- dendrites, and axon terminals of the luminosity-type
tion and spread of S-potentials in fish (Cyprinidae). horizontal cell of the turtle retina. J. Comp. Neural.
J. Physiol. Land. 194: 259-269, 1967. 216: 359-368, 1983.
26. NAKA, K.-I. AND WITKOVSKY, P. Dogfish ganglion 46. WUNK, U. F. AND WERBLIN, F. S. Synaptic inputs
cell discharge resulting from extrinsic polarization of to the ganglion cells in the tiger salamander retina.
the horizontal cells. J. Physiol. Lond. 223: 449-460, J. Gen. Physiol. 73: 265-286, 1979.
1972. 47. YAMADA, E. AND ISHIKAWA, T. The fine structure
27. SAKAI, H. M. AND NAKA, K.-I. Synaptic organiza- of the horizontal cells in some vertebrate retinae.
tion of the cone horizontal cells in the catfish retina. Cold Spring Hurhor Symp. Quunt. Biol. 30: 383-
J. Comp. Ncurol. 245: 107- 1 15, 1986. 392, 1965.
Downloaded from www.physiology.org/journal/jn at Macquarie Univ (137.111.162.020) on February 14, 2019.