Author: Mickey Rowe (rowe@lepomis.psych.upenn.edu) Title: Evolution of Color Vision EVOLUT

======================================================================
Author: Mickey Rowe (rowe@lepomis.psych.upenn.edu)
 Title: Evolution of Color Vision
======================================================================


EVOLUTION AND COLOR VISION I: OPSIN GENES

  Bob Bales brought up an interesting topic in a recent post (well, it
was recent when I started writing this :-).  The topic is "evolution
and color vision".  Bob is apparently under some misconceptions either
about color vision, or at least what evolutionary theory might predict
about it.  In a series of four posts beginning with this one, I want
to ramble on and on about some of the background you might want to
know if you were going to make some meaningful statements about
evolution and color vision.

  I'm going to start by describing a tiny fraction of what's well
known about the molecular biology and biochemistry involved in visual
transduction.  If you're familiar with the topic, you may want to skip
to the last couple of paragraphs in this post where I get to some data
illuminating the evolutionary origins of the visual pigments.  In
followup posts I'm going to describe a bit of comparative psychology
of color vision (to buttress the point that color vision systems are
not all the same).  In a third post I'm going to discuss some
comparative anatomy, with a focus on the visual systems of mammals.  I
intend to demonstrate how comparative anatomy makes sense in light of
what the fossil record tells us about the history of mammalian
evolution.  Finally, in a fourth post I will outline some of the steps
that would be required in order for an organism to acquire color
vision, with a discussion of how reasonable it is to suppose that such
systems could evolve multiple times.

  I no longer have access to Bob's post, but as I recall he was making
some sort of statement that the distribution of current species which
have color vision is at odds with what an evolutionary biologist would
expect.

  Let's dissect this claim with a little bit of thought and a look at
some data.  In the first place, I got the distinct impression from
Bob's post that he thought "animals with color vision" should form a
monophyletic group.  This is absurd.  In this context, saying that an
animal has color vision is like saying an animal has a tail.
Suggesting that two animals (say bees and humans) should be considered
more closely related to each other than two others (say cats and
humans) on the basis of the extent of their capacity for making color
discriminations is similar to suggesting that two animals (say
salamanders and lobsters) should be considered more closely related to
each other than two others (say salamanders and frogs) just on the
basis of which animals have tails.  I'm not going to leave this
discussion at this point, but I'll postpone the rest of it for my
second post while I begin with some basics.

  What does it mean to say that an animal has color vision?  The term
color vision is used in different contexts with somewhat different
meanings, but from our own perspective of what it means to see in
color, the best definition would go something like:

An animal has color vision if it has the capability of discriminating
lights (scattered light as well as light sources) on the basis of the
lights' spectral content, even when those lights are of equal
subjective brightness.

  The front end requirement for such a system is that the animal must
have at least two different spectral classes of receptor, where each
class is defined by the sensitivity of the receptor to light as a
function of wavelength.  This often leads to a looser definition of
color vision: an animal is declared to have color vision if it has at
least two spectral classes of photoreceptor operating at the same
time.

  Although there are a variety of ways in which different receptor
classes could be constructed, it seems that extant organisms use only
one.  

  The first step in the transduction of light energy to a neural
signal is the light-induced isomerization (change of shape) of a
chromophore, specifically a vitamin A derivative.  Each chromophore is
bound to a membrane protein called an opsin.  The main function of the
opsin is to change shape after light absorption triggers the
isomerization of the chromophore: the opsin is an enzyme that is
activated by the chromophore's isomerization.  However, because of the
linkage between the opsin and the chromophore, the opsin also serves
to tune the wavelength dependence of the light induced isomerization
reaction in the chromophore.  That is, the chromophore's sensitivity
to light at a given wavelength is established in part by the
opsin--different opsins (i.e. opsins with different amino acid
sequences) bound to identical chromophores will have different
absorption probabilities at each wavelength.  The result is that
photoreceptors which express the gene for only one type of opsin will
form a different class than photoreceptors that express a gene coding
for a different opsin.  Although there are other mechanisms that
animals could use to differentiate photoreceptor classes (most notably
some animals use more than one chromophore, and many vertebrates have
colored oil droplets that screen individual receptors) it seems that
the expression of only one of their possible opsin coding genes in
each receptor is the mechanism that all animals use.

  Now we have to throw in a slight wrinkle.  In a vast majority of
vertebrates, there are two different sets of photoreceptors, one that
operates during the day and another that operates in the dark.  Most
people are probably familiar with the distinction between rods and
cones--rods mediate night vision, cones day.  At night when the number
of photons around is low, visual systems don't go to such fancy
lengths to discriminate the light's spectral content, so there is
generally only one class of rod in any given animal (at least some
frogs are exceptions to this rule).  For all intents and purposes,
none of us have color vision when we're dark adapted.  Thus with
respect to vertebrates, the discussion of photoreceptor classes above
was more specifically a discussion of cone classes.

  Here we'll get to some interesting stuff by looking at the opsins
for which we have the most data.  DNA and peptide sequences for
various opsins have been determined.  In 1990, all of the then known
amino acid sequences were compared in order to infer a phylogeny for
the opsin molecules.  These sequenced proteins consisted of four
different opsins from drosophila, one from octopus, four from human
(one rod, three cone) and one rod opsin each from chicken, sheep, cow,
and mouse.  All of the opsins have similar sequences, but any good
evolutionary biologist could tell you that some should be more similar
to each other than to others.  Would anybody like to draw their guess
at the phylogeny determined for these thirteen proteins?  (Hint, it
appears that all opsins derive from a very ancient protein, since it
has homologs in bacteria* as well as in both invertebrates and
vertebrates.  Amongst vertebrates, the rod opsin seems to be the most
conserved; cone opsins have arisen principally by duplication and
subsequent mutation of the rod opsin gene.)  Suffice it to say that
these known opsins are not distributed in a mix and match fashion as
one might guess a designer would have distributed them.  If you'd like
to see the phylogeny, you can look up the Goldsmith paper listed in
the fourth post in this series.  Alternatively I guess I could make an
ASCII representation of it...

*I've recently stumbled onto a reference that claims that vertebrate
rhodopsin and bacteriorhodopsin are not part of the same gene family.
I'll reserve judgement until I've read more than just the abstract of
the paper.  E-mail me if you would like to see the reference yourself. 

   It should also be noted that many humans carry more than one copy of
the middle wavelength-sensitive cone opsin.  As this is grist for the
evolution of color vision mill, we're literally ripe for the addition
of a fourth cone class.  (This probably won't happen, though, because
people with a fourth cone class will be constantly trying to readjust
the color on television sets.  As a result of that such people will be
highly selected against in bars the world over :-)

  Since 1990, a few other opsins have been sequenced, specifically
opsins from a variety of monkeys.  I don't know as that they've been
compared with the others, but I'm willing to predict where they should
fit into the picture.  It's nice to have a theory that lets you do
that.*

(*Since I wrote that last paragraph, I've seen another phylogeny that I
think had more than twice as many opsin sequences as my best current
reference.  As far as I know, that work is still in press, and I no
longer have access to it.  From what I saw, though, the creationists
have even more reason to fold their hand on this one now than they did
two years ago...)


EVOLUTION AND COLOR VISION II: COMPARATIVE PSYCHOLOGY

  Prior to the advent of some nifty techniques in molecular biology,
people had to use less direct methods of classifying photoreceptors.
Among these methods are: direct measurement of the absorptive
properties of individual receptors, measurement of the electrical
responses of cells to monochromatic lights, and the conditioning of
learned behaviors.  Thus even without molecular biology, we knew (and
know) a lot about the pigments underlying color vision systems
(amazing isn't it, Chris? :-)

  Based on this sort of information, it's clear that most vertebrates
have at least two cone classes.  In fact, many birds, turtles and fish
have four or five.  Many invertebrates are similarly well endowed, and
last I heard, the mantis shrimp was the winner of the contest of who
has the largest number of photoreceptor classes.  Given that coral
reef animals and tropical birds often appear very colorful to us, it's
not surprising that they have well developed systems of color vision.
That different animals have different numbers of receptor classes
already tells us that color vision systems are not all equivalent (as
Bob might have us believe).  If we restrict ourselves to animals which
have the same number of receptor classes, might we expect that *their*
color vision systems are equivalent?

  The answer is a resounding no.  Let's compare the color vision
systems of two animals that both have three photopic (e.g. active
under bright illumination) photoreceptor classes.  One is the human,
the other is the honey bee (specifically the worker--I don't know how
the other castes are endowed).  Does anybody here think that what a
bee sees when it looks at a rainbow has the same appearance as what we
see?  We'll ignore optical polarization (which the bee is sensitive to
and we're not) and focus on what we can infer about "color" based on,
among other things, our knowledge of the bee's receptor classes.  To
begin with, at the inside of the rainbow where the violet-appearing
light fades off to invisibility for us, the bee will still see more
rainbow.  On the outside, where we see red, the bee would see nothing
for although bees have an ability to see what for us is UV, we have
the ability to see what bees might call infra-red.

   Now picture that rainbow: what you see appears to have discrete
bands of color.  Don't for a minute think that those bands arise from
there being anything discrete about the radiation emanating from that
patch of sky.  If you measured the radiation with a spectrophotometer,
you'd find that the wavelength of maximum intensity as a function of
the radial distance across the rainbow would decrease smoothly and
monotonically from the outside to the inside of the bow.  The apparent
discreteness is an artifact of our photopigments* and the neural
processing of our photoreceptors' outputs.  The bee too would probably
see discrete bands**.  However, just as the outer and inner borders
would be in different locations for us and bees (as described in the
preceding paragraph), the borders of each "color" would be placed
differently by the bee as well.

* Photopigments (aka visual pigments) == chromophore + opsin.

** We can't ever really know how the world appears to a bee, but given
what we can infer from doable experiments (I actually chose the bee in
part because its color vision has been studied about as much as any
other animal's, excluding the human's) the supposition that it would
see discrete "color" bands from a rainbow is reasonable.

  I can't claim that we have a good handle on why different animals
have different visual pigments.  There *are* some cases that are well
understood--most notably it was predicted some 20 years before
verification that marine fish that live just above the aphotic zone
would have only one pigment, and that that one pigment would have a
maximal sensitivity down around 450 nm (for us light at this
wavelength would appear blue).  It makes sense that if there isn't
much light around, an animal's photoreceptors will be adapted to
respond most strongly to the wavelengths of light most readily
available.  Bioluminescent fish and insects also tend to have pigments
that are adapted for maximal sensitivity to the wavelengths of light
emitted by their photophores (the molecules responsible for the
emission of light e.g. from the abdomens of fireflies).  The specifics
of what selective advantage other pigments in other environments might
convey are still somewhat mysterious (See the Lythgoe and Partridge
paper listed in the fourth post of this series for a discussion of the
topic).

 One thing is clear, however.  The best known predictor of what sort
of pigments will be expressed by any given animal, is the pigments
expressed by its nearest living relatives.  To an evolutionary
biologist this makes a lot of sense, of course.

 There are a lot of other differences (or similarities) between
manifestations of color vision systems in different animals.  I've
chosen to stick to a discussion of pigments here partly for
simplicity, and partly because the straightforwardness of analyzing
retinal receptors makes this the facet of color vision about which the
most data is available.  The point of this post is to say that it
makes no sense to use the presence or absence of color vision in
determining a phylogeny.  If you want to be serious about asking what
color vision and evolution have to say about each other, you have to
ask specific questions about what sort of color vision different
animals have.


EVOLUTION AND COLOR VISION III: MAMMALIAN DEFICITS

  In Bob's post, it was suggested that among mammals, color vision is
more or less exclusive to primates.  This isn't quite correct.  In
fact there are many other mammals with color vision.  For example,
diurnal squirrels and tree shrews have each been demonstrated to have
at least two photoreceptor classes, and behavioral studies indicate
that each meets the strict definition of color vision (the first
definition in the first post of this series).  Recent finds have also
indicated that some rodents are sensitive to ultraviolet light,
suggesting that they have a previously unknown class of photoreceptor.
However, color vision systems do appear less frequently and with less
complexity (i.e. with fewer photoreceptor classes) amongst species of
mammals relative to species of most other classes of animal.  To
understand why this might be so, let's examine the history of
mammalian evolution as evidenced by the fossil record, and conciliate
that information with some comparative anatomy.

  The lineage of animals which joins reptiles and mammals is often
touted here as an excellent example of a transitional series.  One
detail that might seem surprising to people is that this transition
occurred at the *beginning* of the Mesozoic era--the same time during
which other reptiles were transitioning into dinosaurs.  It's not
quite right to say that mammals replaced dinosaurs at the beginning of
the Cenozoic era, because mammals existed alongside of dinosaurs
during the dinosaurs' entire "reign".  However, during the Mesozoic
era, dinosaurs and other reptilian cousins (e.g. pterosaurs,
plesiosaurs and ichthyosaurs) were an extremely diverse group which
occupied most of the available niches.  The bush of life had only a
small twig representing the lineages that would later branch out into
all of the mammalian forms currently extant.  Mesozoic mammals were
small rodent-like creatures that were most probably nocturnal.

  Note that the last paragraph is based only upon what we can glean
from the fossil record.  If current species arise from the descent
with modification of pre-existing species, one might predict that the
above inferred history of mammals would leave clues in contemporary
mammalian anatomy.  Oddly enough such clues exist.

  You may recall from the first post in this series that pretty much
all animals are "color-blind" in the dark.  Consequently, if an animal
is only active at night a color vision system would be of little use,
much as eyes are of little use to cave fish, moles and other animals
which live in the absence of light.  So if modern mammals are just
descendents of the animals whose fossilized remains are found in
Mesozoic strata, you would expect that this would be reflected in the
makeup of their retinas.  Lo and behold, this expectation is born out
quite well.

  First off, comparative anatomy indicates that most mammals don't
have well developed color vision systems *not* because their line
didn't get around to developing it, but more likely because after our
ancestors evolved color vision it became superfluous and was lost.
The color vision of primates is not strictly homologous to the color
vision of fish, birds, turtles, etc.  Much of the machinery used for
primate color vision arose independently long after similar systems
developed (without being lost) in other vertebrate lineages.  At this
point the wary creationist might say, "Aha! So primate color vision
doesn't fit into the mammalian scheme, and could be construed as
evidence of a creator--in developing primates, the creator used a
feature similar to what he'd used in those other so-called
'lineages'".  I urge *anyone* who might think this to look more deeply
into comparative anatomy.  I will only briefly describe a few relevant
features here.

  Color vision is mediated by cones, so named because of the shape of
the receptive part of the cells.  If the history of mammalian
evolution described above were correct, you would expect to find
significant differences between mammalian cones and the cones of other
vertebrates.* Whether or not you expect it, this is exactly what has
been found.  There are several features that are quite common to the
cones of non-mammalian vertebrates, but that are completely lacking in
mammals.**

* The initial definition of "cone" vs. "rod" photoreceptors has some
kinks, because it is apparent (based on criteria other than the shape
of the receptive part of the cell) that some photoreceptors that
appear on first glance to be homologous are actually analogous.  For
example, the "rods" of nocturnal geckos (a type of lizard) are most
likely homologous to the cones of other animals--geckos did the
opposite of mammals.  In their development, geckos went through a
strictly diurnal phase, and hence lost some of the adaptations for
nocturnal vision.  They subsequently became nocturnal again, and thus
their cones faced some of the same adaptive pressures faced by the
rods of other vertebrates.

** As most of you might guess, there are exceptions to the "complete[]
lack" of the characters I'm about to describe.  I'll leave it as an
exercise to the reader as to which animals *are* exceptions.  I'll
tell anyone who guesses and provides with their guess a rationale for
why they guessed what they did.  That is, if you understand and accept
evolution, you would *predict* that if there are exceptions, they will
be found in particular animals.  If you don't understand or don't
accept descent with modification as the origin of current species, I
really would like to know what sort of reasoning you might use to
guess at the exceptions.

  Many vertebrates have oil droplets at the bases of the light
sensitive parts of their photoreceptors.  These oil droplets often
have pigments in them that absorb (i.e. filter out) some of the light
that would otherwise stimulate the cell.  What this does is to modify
the spectral sensitivity of the photoreceptor bearing that droplet.
This feature is not found in mammals.

  Many vertebrates have double cones--two cones that are joined along
their long axes by tight junctions, gap junctions or both.  Nearly all
classes of vertebrates have some variety of this form of receptor in
their retinas.  This feature is not found in mammals.

  The photoreceptors of many vertebrates perform a sort of circadian
dance.  During the day, the rods are extended on long stalks so that
their sensitive parts are buried in a layer of pigmented epithelium.
This epithelium shields the rods so that very little light reaches
them from the sides, and the cones basically shield them from axially
propagating light.  At night the cones are extened out into the
pigmented epithelium, and the rods are contracted back to where the
cones were during the day.  This feature is not found in mammals.

The conclusion that might be drawn from the above is that there are
many features of ancestral retinal anatomy that were retained in most
classes of vertebrates, but lost in mammals.  Elaborate color vision
is just one such feature.  The phylogenies of the opsin molecules that
I discussed in the first post of this series suggest that mammals have
always retained two cone pigments (a survey in 1981 indicated that
there aren't any vertebrates with only one cone pigment), but any
mammals that, like us, have more than two pigments (re)gained the
third relatively recently (for us probably around 63 million years
ago).  [For those from sci.bio, this was why I was suggesting that
squirrels might be trichromatic--I'm not willing to climb out on that
limb now.  Our short wave-sensitive or "blue" cone is probably
homologous to the UV cone of other (i.e. non-squirrel) rodents.  One
of the inferred pigments in the dichromatic squirrels has about the
same absorption spectrum as our short wave cone, so there isn't any
reason to suppose that the squirrel has a third.  That is, it is
reasonable to conclude on current evidence that the UV sensitive
rodents have the same cone classes as the squirrels, only their short
wave pigment has been shifted to absorb even shorter wavelengths than
those of most other mammals.]


EVOLUTION AND COLOR VISION IV: HOW TO SEE RED

  In this final segment I'd like to address some of the probable steps
required in the formation of a color vision system.  I do this in an
attempt to circumvent an argument via lack of imagination about the
improbability of such a system arising more than once.  The generation
or elaboration of color vision systems is not a terribly complicated
process (at least at the periphery).

  Let's presume that we're starting with an organism that already has
an eye of some sort.  The first step towards a color vision system is
the need for at least two visual pigments.  It should be obvious that
the addition of a pigment would be of immediate advantage even if the
new pigment was expressed in the same cells as the older pigment(s).
The reason is that there are some wavelengths of light where the new
pigment will respond more strongly than the old, so the addition of a
pigment will increase the animal's sensitivity over those wavelengths.

  The next step (conceptually anyways--it may be that this and the
first step typically occur simultaneously) is the sequestration of the
new pigment into a discrete population of photoreceptors.  (By
discrete, I don't mean spatially.  I just mean that each photoreceptor
should express only one opsin.)*  The advantage that this provides
comes in the form of visual contrast.  The lowest level of visual
information processing is the recognition that something is different
about a given region of space--i.e. that there is food or a predator
"over there".  To perform this function in habitats that are rich in
light of particular wavelengths (the short wave "blues" underwater, or
the mid wave "greens" of the tropical rainforest) it's best to have at
least two pigments, one matched to the dominant wavelengths and one
offset from those wavelengths.  With the matched pigment,
non-reflective objects have high contrast as dark areas on a bright
background.  With the offset pigment, reflective objects will apear
bright against a darker background.  Except in some extreme conditions
(i.e. just above the aphotic zone for marine environments) the
background probably isn't constant enough for that simplistic analysis
to hold, but it's easy to imagine that if an animal has more
photoreceptor classes it has a greater chance that one of them will
allow for the visibility of a given target under a given set of
background conditions.

*It's not yet clear how the expression of photopigments is regulated
in individual cells, but because of its accessibility, the retina is
frequently used in studies of developmental neuroanatomy.  Experiments
with transgenic animals have already given us some key pieces of
information about the regulatory mechanisms that determine what sort
of photoreceptor a retinal precursor cell will become.  Immuno-
histochemistry has also been used to show that the fate of a cell
(i.e. what sort of cell it will become) is established long before
morphological differentiation is apparent.  This is a hot area where
the rest of this century is sure to see incredible advances in our
knowledge base.

   The next step is the development of neural wiring in the retina
that segregates the signals from one population of cells from those of
the other(s).  (Oddly enough this isn't necessary for the previous
advantage, although as I'll describe below, there is good reason to
suppose that these two steps occur simultaneously as a result of the
mechanisms of neural development.)  The advantage of this is that it
allows the animal's retina to "draw" contours around an object (i.e.
to place "color" boundaries on the visual scene).

   The last phase in the development of color vision is the
arrangement of wiring in the brain that allows an animal to segregate
and classify objects according to their "color" (i.e. according to how
well the object stimulates the different receptor classes).  The
advantage of this adaptation is that it allows the animal to classify
objects according to "color".  For example, it has been argued
(although to me this is a just so story and may not be correct) that
color vision and the expression of pigments in fruit co-evolved.  That
is, it is to the plants' advantage to have its fruit remain un-eaten
until the seeds are ready for dispersal, so the color change in
ripening fruit is a signal that the plants are sending to the animals.
In turn the animal gets the greatest benefit from eating the ripened
fruit, so it is to the animals' advantage to recognize when the fruit
is ripe.  The sort of comparison necessary for discriminating ripe
from unripe fruit is easy if, for example, objects which reflect a lot
more long than short wavelength visible radiation bring about a
particular *quality* of sensation (e.g. what we call "red").

   The point of the above was to make explicit that even if an animal
were to develop color vision in steps, it's not hard to imagine a
sequence of steps in which each step confers some advantage which
would cause that step to be selected for.  However, there's a beauty
in the way that nervous systems are constructed which might lead you
to expect that rudimentary color vision can arise in a "color blind"
animal in only one or two steps.

   The addition of a new pigment arises from a gene duplication
followed by mutation of one (or both) of the copies.  As indicated in
the first part of this series, it seems pretty clear from gene
sequence data that this is exactly how new pigments have arisen in us,
fruit flies and a couple of other primates.  By inference it seems
likely that this is a widespread occurrence.

   I'll end now with a brief foray into neurobiology.  Animal nervous
systems, particularly the nervous systems of vertebrates are not
"hard-wired" at birth (or hatch or the end of metamorphosis...).
Decisions about which nerve cells should be connected to which other
nerve cells are made during a long space of time prior to adulthood,
and in some animals (though usually to a much more limited extent)
even during adulthood.  Genetics seem to specify (in unknown ways)
some of the gross features of connectivity--for example in mammals the
axons of ganglion cells in the eye mostly grow through the optic nerve
to a particular group of cells in the thalamus.  However, the fine
distinctions about, for instance, which ganglion cells connect to
which cells in the thalamus are made initially by the formation of a
lot of random connections.  Many of these connections are then pruned
back so that each ganglion cell stimulates only a small subset of the
cells it initially connected with.  The "rules" governing the pruning
back are largely based on correlations in the activity of different
cells--if two cells in the retina are generally active at the same
time, then they will probably end up being connected to the same cells
in the thalamus.

   This activity-dependent pruning of connections appears to be the
way that "maps" are created in higher brain areas.  The best indicator
of whether or not two cells in the retina will be simultaneously
active, is how close they are to each other in space.  Cells in the
thalamus thus form a map of cells in the retina according to their
activity, and hence their connectivity.  Now it's easy to imagine that
another determinant of whether or not two cells will be active at the
same time is whether or not they are connected to cells which express
the same pigment (within the retina, the same rules are followed in
the creation of connections, so ganglion cells will preferentially be
connected to cells which express the same pigment).  So in the
thalamus and other brain regions, there will be maps of the different
receptor types within the maps of retinal location.

   Of course neural development is a lot more complicated than I've
described here, but the take home message is that the way that nervous
systems develop in growing animals makes it easy to incorporate
changes at the periphery.  It *has* to be that way, or our nervous
systems would not be able to cope with changes which occur in our
muscles and sense organs as a result of growth.

   If I've been at all clear, you'll see that once an animal has
different photoreceptor classes the rest of the nervous system is
already prepared to take advantage of them.  An interesting case study
in this regard is new world monkeys.  In at least one species, two of
their opsin genes are (like our mid- and long- wave sensitive opsins)
on the X chromosome.  The monkeys' expression patterns are different
from ours, however, and it turns out that for one (or two depending on
how you think about it) class of photoreceptor, the females can
express the genes on each X chromosome.  The males naturally only have
one X chromosome.  In the population, there are two types of male
(depending on which allele they have on their X chromosome), and three
types of female (depending on whether they are heterozygous, or
homozygous for one or the other allele).  The monkeys' developing
nervous systems seem to take advantage of whatever photoreceptor
classes happen to be out there in that animal's retina.

   I'm sure I've left a lot of things unclear, but if anyone wants to
find out more, I can recommend a few references.  A good general
source for information about neurobiology, with several chapters on
vision is:

Principles of neural science 3rd Edition, edited by Eric R. Kandel,
     James H. Schwartz, Thomas M. Jessell, Elsevier, New York, 1991.

An excellent book that treats the principles necessary for
appreciating comparative studies of visual systems is: 

Lythgoe, J. N. (1979). The ecology of vision, Oxford University Press,
     Oxford (and also Clarendon Press, New York).

After reading that, you'll probably be prepared for:

Jacobs, G.H. (1981). Comparative Color Vision, Academic Press, New
     York. 

and a few papers:

Goldsmith, T.H. (1990). "Optimization, Constraint, and History in the
     Evolution of Eyes", Quarterly Review of Biology, 65(3):281-322.

McFarland, W.N. and Munz, F.W. (1975). "The Evolution of Photopic
     Visual Pigments in Fishes", Vision Research, 15:1071-1080.

Hemila, S., Reuter, T. and Virtanen, K. (1976). "The Evolution of
     Colour-Opponent Neurones and Colour Vision", Vision Research,
     16:1359-1362. 

Lythgoe, J.N., and Partridge, J.C. (1989). "Visual Pigments and the
     Acquisition of Visual Information", Journal of Experimental
     Biology, 146:1-20.

Mickey Rowe     (rowe@pender.ee.upenn.edu)