VISUAL ADAPTATION AND RETINAL GAIN CONTROLS
consideration because of the problem of light
scattering within the eye and of optical aberrations
in the lens and cornea. If the measured size of the
receptive field center and adaptive summing area
were greatly influenced by such imperfections in
physiological optics, one would have to discount the
conclusions about the site or sites of gain control(s)
which were based on the correlation of the spatial
extents o f these two different mechanisms.
However, direct measurements of the physiological
optics of the cat's eye (Bonds, 1974; Robson and
Enroth-Cugell, 1978) indicate that, except for rare
cells with the smallest receptive fields, the optical
effect on the measured size of the receptive field and
adaptive summing area is small. It is interesting to
note that the existence o f the p o s t u l a t e d
"amacrine" gain control of the Y cells is not subject
to any doubts based on optical blur or scatter,
because the adaptive summing area and receptive
field center of Y cells are so large. The presence of
neighboring X cells with fields ten times smaller in
area serves as a control on the optical contribution
to the size of the Y cells' fields. However, there
could be some question about whether the size of
the receptive fields of X cells, which can be quite
small in area, might be due to optical blur or scatter.
It becomes a quantitative question in the case of the
X cells. However, even for X cells, the optical blur
seems to be less than the neural summation area
(Robson and Enroth-Cugell, 1978) when the
physiological optics are optimized with best
refraction and a small artificial pupil. Throughout
this paper, arguments based on receptive fields'
sizes have only cited as evidence the results of
experiments in which the physiological optics were
optimized.
That some of the retinal gain - - setting
mechanisms must involve pooling of signals from
many receptors is the conclusion of a physiological
extension of Rushton's reasoning about the low
level o f b a c k g r o u n d s which p r o d u c e light
adaptation (Enroth-Cugell and Shapley, 1973a). As
discussed a b o v e , R u s h t o n f o u n d that the
psychophysical scotopic threshold was raised by a
factor of two from its dark adapted value when only
one rod in a hundred actually caught a quantum
of light. However, as stated before, this result
implies nothing about gain control in the retina. It
could be explained in terms of an increase in
PRR3-I
311
"noise" from the background. Applying Rushton's
reasoning to retinal gain, we measured how much
background light is required to reduce the gain of
a ganglion cell by a factor of two. In several cells
with large receptive fields, the required background
light yields one quantum absorbed per second per
hundred rods. The integration time of feline rods
is almost certainly shorter than a second; indeed,
other work indicates a maximum integration time
of 0.1 s. Thus, one rod in a thousand per
integration time receiving a quantum of background
light is enough to reduce the gain by half a log unit.
This is solid support for Rushton's conclusion that
signals from a pool of rods must set the gain for
signals from rods which have not themselves
received light quanta from the background.
In ceils with smaller receptive fields, the
luminance required to reduce gain was higher, so
that in the worst case approximately one rod in ten
received a quantum of light per integration time.
We now presume that these ganglion cells with small
fields were X cells, on the basis of the receptive field
center size. So again the very low level of
background luminance which is required to reduce
the gain may suggest that photoreceptor signals
must be pooled to set the gain in the cat retina. Note
that light scatter would not a f f e c t these
measurements because they were made with large
uniform backgrounds which were at least one
thousand times larger in area than the optical point
spread function at half height. Furthermore, the
force of these arguments is not affected by the fact
that not all photoreceptors which lie within the
spread of the dendritic tree of a ganglion cell project
to that ganglion cell. The statistical randomness of
the quantum catch makes the fraction of receptors
hit by quanta the same whether one considers the
entire population of photoreceptors, or only that
population which projects to the ganglion cell under
study, as long as the background is truly uniform.
That there are different transition levels, from
dark to light adapted, for scotopic receptive field
center and surround mechanisms (Fig. 26) also
supports the idea of a gain control proximal to the
photoreceptors. The same rods must drive the center
and surround. If only the rods adapted, the center
and surround would have to lose their gain in
parallel. Since this is not observed, we must
conclude that gain must be controlled at a site in