These cells come in an astonishing variety of shapes and use an impressive number ofneurotransmitters. There may be well over twenty different types. They all have in common, first, their location, with their
cell bodies in the middle retinal layer and their processes in the synaptic
zone between that layer and the ganglion cell layer; second, their connections,
linking bipolar cells and retinal ganglion cells and thus forming an alternative,
indirect route between them; and, finally, their lack ofaxons, compensated
for by the ability of their dendrites to end presynaptically on other cells.
Amacrine cells seem to have several different functions, many of them
unknown: one type of amacrine seems to play a part in specific responses
to moving objects found in retinas of frogs and rabbits; another type is
interposed in the path that links ganglion cells to those bipolar cells that receive
Amacrines are not known to be involved in the center-surround organization of ganglion-cell receptive fields, but we cannot rule out the possibility.
This leaves most of the shapes unaccounted for, and it is probably fair to
say, for amacrine cells in general, that our knowledge of their anatomy far outweighs our understanding of their function.
BETWEEN BIPOLAR CELLS
AND GANGLION CELLS
We have seen that the main features of the ganglion-cell receptive fields are to be found already in the bipolar cells. This leaves open
the question of what transformations of information occur between bipolars and ganglion cells. It is hardly likely that nothing happens, if the complexity of
the synaptic layer between the middle layer and ganglion-cell layer is any indication,
for here we often find convergence between bipolars and ganglion cells in
the direct path, and also the intervention of amacrine cells, whose functions
are not well understood.
The synapses between bipolar cells and ganglion cells are probably all
excitatory, and this means that on-center bipolar cells supply on-center ganglion cells, and off-center bipolars supply off-center ganglion cells. That
simplifies the circuit: we could have had on-center cells supplying off-center cells through inhibitory synapses, for example. We should be thankful for small mercies.
Until 1976, it was not known whether on-center cells and off-center cells differed in their shapes, but in that year Ralph Nelson, Helga Kolb, and
Edward Famiglietti, at the National Institutes of Health in Bethesda, recorded intracellularly from cat ganglion cells, identified them as on- or off-center,
and then injected a dye through the microelectrode, staining the entire dendritic tree. When they compared the dendritic branchings in the two cell types
they saw a clear difference: the two sets of dendrites terminated in two distinct sublayers within the synaptic zone between the middle and ganglion-cell
layers. The off-center-cell dendrites always terminated closer to the middle
layer of the retina, the on-center dendrites, farther. Other work had already
shown that two classes of bipolar cells, known to have different-shaped synapses
with receptors, differed also in the position of their axon terminals, one
set ending where the on-center ganglion-cell dendrites terminated, the other, where
the off-center dendrites terminated. It thus became possible to reconstruct
the entire path from receptors to ganglion cells, for both on-center and off-center systems.
One surprising result of all this was to establish that, in the direct
it is the off-center system that has excitatory synapses at each stage,
from receptors to bipolars and bipolars to ganglion cells. The on-center path
instead has an inhibitory receptor-to-bipolar synapse.
The separation of bipolar cells and ganglion cells into on- and off-center categories must surely have perceptual correlates. Off-center cells respond
in exactly the same way to dark spots as on-center cells respond to bright
spots. If we find it surprising to have separate sets of cells for handling dark
and light spots, it may be because we are told by physicists, rightly, that darkness
is the absence of light. But dark seems very real to us, and now we seem to find
that the reality has some basis in biology. Black is as real to us as white,
and just as useful. The print of the page you are reading is, after all, black.
An exactly parallel situation occurs in the realm of heat and cold. In
high school physics we are amazed to learn that cold is just the absence of
heat because cold seems equally real—more so if you were brought up,
as I was, in frigid Montreal. The vindication of our instinct comes when we learn that
we have two classes of temperature receptors in our skin, one that responds
to the raising of temperature, and another to lowering. So again, biologically,
cold is just as real as hot.
Many sensory systems make use of opposing pairs: hot/cold, black/white,
head rotation left/head rotation right—and, as we will see in Chapter
yellow/blue and red/green. The reason for opposing pairs is probably related to the way in which nerves fire. In principle, one could imagine nerves
with firing rates set at some high level—say, 100 impulses per second—and
hence capable of firing slower or faster—down to zero or up to, say, 500—to
opposite stimuli. But because impulses require metabolic energy (all the sodium that enters the nerve has to be pumped back out), probably it is more
efficient for our nerve cells to be silent or to fire at low rates in the absence
of a sensory stimulus, and for us to have two separate groups of cells for any given
modality—one firing to less, the other to more.
SIGNIFICANCE OF CENTER-SURROUND
Why should evolution go to the trouble of building up such curious entities as center-surround receptive fields? This is the same as asking
what use they are to the animal. Answering such a deep question is always difficult,
but we can make some reasonable guesses. The messages that the eye sends
to the brain can have little to do with the absolute intensity of light shining
on the retina, because the retinal ganglion cells do not respond well to changes
in diffuse light. What the cell does signal is the result of a comparison
of the amount of light hitting a certain spot on the retina with the average
amount falling on the immediate surround.
We can illustrate this comparison by the following experiment. We first
find an on-center cell and map out its receptive field. Then, beginning with
the screen uniformly and dimly lit by a steady background light, we begin
turning on and off a spot that just fills the field center, starting with the
light so dim we cannot see it and gradually turning up the intensity. At a certain brightness,
we begin to detect a response, and we notice that this is also the brightness
at which we just begin to see the spot. If we measure both the background
and the spot with a light meter, we find that the spot is about 2 percent
brighter than the background. Now we repeat the procedure, but we start with
the background light on the screen five times as bright. We gradually raise
the intensity of the stimulating light. Again at some point we begin to
detect responses, and once again, this is the brightness at which we can just
see the spot of light against the new background. When we measure the stimulating light, we find that it, too, is five times as bright as previously,
that is, the spot is again 2 percent brighter than the background. The conclusion is that
both for us and for the cell, what counts is the relative illumination of
the spot and its surround.
The cell's failure to respond well to anything but local intensity differences may seem strange, because when we look at a large, uniformly lit spot,
the interior seems as vivid to us as the borders. Given its physiology,
the ganglion cell reports information only from the borders of the spot; we see the
interior as uniform because no ganglion cells with fields in the interior are
reporting local intensity differences. The argument seems convincing enough, and
yet we feel uncomfortable because, argument or no argument, the interior
still looks vivid! As we encounter the same problem again and again in later
chapters, we have to conclude that the nervous system often works in counterintuitive ways. Rationally, however, we must concede that seeing the
large spot by using only cells whose fields are confined to the borders—instead
of tying up the entire population whose centers are distributed throughout
the entire spot, borders plus interior—is the more efficient system:
if you were an engineer that is probably exactly how you would design a machine. I
suppose that if you did design it that way, the machine, too, would think the
spot was uniformly lit.
In one way, the cell's weak responses or failure to respond to diffuse
light should not come as a surprise. Anyone who has tried to take photographs without a light meter knows how bad we are at judging absolute light
intensity. We are lucky if we can judge our camera setting to the nearest
f-stop, a factor of two; to do even that we have to use our experience, noting
that the day is cloudy-bright and that we are in the open shade an hour before
for example, rather than just looking. But like the ganglion cell, we
are very good at spatial comparisons—judging which of two neighboring regions
is brighter or darker. As we have seen, we can make this comparison when
the difference is only 2 percent, just as a monkey's most sensitive retinal
ganglion cells can.
This system carries another major advantage in addition to efficiency.
We see most objects by reflected light, from sources such as the sun or
a light bulb.
Despite changes in the intensity of these sources, our visual system
preserves to a remarkable degree the appearance of objects. The retinal ganglion
cell works to make this possible. Consider the following example: a newspaper looks roughly the same—white paper, black letters—whether
we view it in a dimly lit room or out on a beach on a sunny day. Suppose, in each of
these two situations, we measure the light coming to our eyes from the white paper
and from one of the black letters of the headline. In the following table
you can read the figures I got by going from my office out into the sun in the
Harvard Medical School quadrangle:
The figures by themselves are perfectly plausible. The light outside
is evidently twenty times as bright as the light in the room, and the black
letters reflect about one-tenth the light that white paper does. But the figures,
the first time you see them, are nevertheless amazing, for they tell us that the
black letter outdoors sends twice as much light to our eyes as white paper
under room lights. Clearly, the appearance of black and white is not a function
of the amount of light an object reflects. The important thing is the amount
of light relative to the amount reflected by surrounding objects.
A black-and-white television set, turned off, in a normally lit room,
is white or greyish white. The engineer supplies electronic mechanisms for making
the screen brighter but not for making it darker, and regardless of how
it looks when turned off, no part of it will ever send less light when it is
turned on. We nevertheless know very well that it is capable of giving us nice rich
blacks. The blackest part of a television picture is sending to our eyes at least
the same amount of light as it sends when the set is turned off. The conclusion
from all this is that "black" and "white" are more than physical
concepts; they are biological terms, the result of a computation done by our retina and
brain on the visual scene.
As we will see in Chapter 8, the entire argument I have made here concerning black and white applies also to color. The color of an object is
determined not just by the light coming from it, but also—and to just as
important a degree as in the case of black and white—by the light coming from
the rest of the scene. As a result, what we see becomes independent not only of
the intensity of the light source, but also of its exact wavelength composition.
And again, this is done in the interests of preserving the appearance of
a scene despite marked changes in the intensity or spectral composition of the
The output of the eye, after two or three synapses, contains
information that is far more sophisticated than the punctate representation
of the world encoded in the rods and cones. What is especially interesting
to me is the unexpectedness of the results, as reflected in the failure
of anyone before Kuffler to guess that something like center-surround
receptive fields could exist or that the optic nerve would virtually
ignore anything so boring as diffuse-light levels. By the same token, no one made any guesses that even closely
approximated what was to come at the next levels along the path—in the
brain. It is this unpredictability that makes the brain fascinating—that plus
the ingenuity of its workings once we have uncovered them.