Colour my world

Share on FacebookShare on Google+Email this to someoneTweet about this on Twitter

Colour does not exist. Not out in the world at any rate. All that exists in the world is a smooth continuum of light of different wavelengths. Colour is a construction of our brains. A lot is known about how the brain does this, beginning with complicated circuits in the retina itself. Thanks to a new paper from Greg Field and colleagues we now have an even more detailed picture of how retinal circuits are wired to enable light to be categorized into different colours. This study illustrates a dramatic and fundamental principle of brain wiring – namely that cells that fire together, wire together.

Colour discrimination begins with the absorption of light of different wavelengths. This is accomplished by photopigment proteins, called opsins, which are expressed in cone photoreceptor cells in the retina. Humans have three opsin genes, which encode proteins that preferentially absorb light of different wavelengths: short (S, in what we perceive as the blue part of the spectrum), medium (M, green) and long (L, red). Each cone expresses only one of these opsin genes and is thus particularly sensitive to light of the corresponding wavelength. However, by itself the response of a single cone cell cannot be used to determine the colour (wavelength) of incoming light. The reason is that each cone is responsive to both the wavelength and the intensity of the light – so an M-cone would respond equally to a dim green light or a strong red light.

Colour information only arises by comparing the responses of multiple cone cells. This is accomplished in two distinct channels – one which compares the inputs of L and M cones (the red-green channel) and one which compares the inputs of S cones to the combined inputs of L and M cones (the blue-yellow channel). The latter of these is the original, evolutionarily older system, dating back at least 500 million years. It is found in most mammals, in which there are only two opsin genes – an S opsin and one whose absorbance is midway between L and M.

The L/M system evolved much more recently, due to a gene duplication that occurred in the lineage of Old World primates, probably around 40 million years ago. The duplication of the primordial L/M opsin gene allowed the two resultant genes to diverge from each other in sequence, generating proteins with different absorption spectra, which could then be compared. Something similar can actually be achieved even in species with only one copy of the L/M gene. This gene is on the X chromosome, so females will carry two copies of it. Due to the random inactivation of one X chromosome in each cell in females, each cone will express only one of the two copies of this opsin gene. If the two copies differ from each other, encoding proteins with alterations in the amino acid sequence that affect their light absorbance, then what will arise is a set of L cones and a set of M cones.

All of this raises an important question – how are the inputs to these different cone cells compared? If the cells which express L and M cones are essentially the same, with the sole difference being that they express different opsin genes, then how is the wiring in the retina set up so that their inputs are distinguished, allowing their subsequent comparison? Cells in the retina are arranged in a series of layers. Cone cells connect, through bipolar and other cells, to retinal ganglion cells, which in turn convey visual information to the brain. Retinal ganglion cells integrate inputs from multiple cones, but in a very specialized way – some cones connect through ON bipolar cells (which are activated by light) and others through OFF bipolar cells (which are inactivated). Typically, one cone in the centre of an array of cells is connected to an ON bipolar cell, while surrounding cones connect to the same retinal ganglion cell target via OFF bipolar cells. The result is that the light signal hitting an array of cones is integrated – if the central cone is an L cell and the surrounding cones are M cells then the retinal ganglion cell will be most strongly activated by red light.

This has been known for quite a long time now. What has not been clear is how this system gets wired up during development. S, M and L cones are distributed randomly across the retina. S cones, which are the least frequent, are molecularly distinct from L/M cones in many ways and connect to a dedicated set of S channel bipolar and retinal ganglion cells. The development of the wiring that carries out the comparison between S and L/M cones is thus molecularly specified. This cannot be the case for the comparison between L and M cones, which differ only in the opsin gene they express.

The new study by Field and colleagues worked out in breathtaking detail the circuitry of the retina at a cellular level. Their results reveal the beauty and elegance of this circuitry but also resolve an important question relating to how L and M cone cells are wired. Each retinal ganglion cell in the centre of the retina receives ON inputs from a single cone and OFF inputs from the surrounding cones. In the periphery, however, the ON “centre” is composed of up to twelve cones. For the ganglion cell to discriminate colours there must be a bias in how many L or M cone cells wire up to it through the ON and OFF channels.

Their results reveal exactly such a bias and further show that it cannot be explained simply by random clumping of L or M cones in the photoreceptor array. What this indicates is that there is some additional mechanism whereby inputs from just one type of cone are strengthened in each of the ON and OFF channels. In effect, the L and M cones are competing for inputs in each channel, presumably through so-called “Hebbian mechanisms” whereby inputs to a cell are strengthened if they fire at the same time and asynchronous inputs are actively weakened. Despite their being no molecular differences between these cone cells, the brain is thus primed to wire them into distinct channels based on their patterns of activity.

A remarkable experiment performed a few years ago dramatically illustrates this principle. Mice are naturally dichromatic – they only have two opsin genes (S and L/M). Researchers in Jeremy Nathans’s group replaced one copy of the L/M gene with a version of the human L gene. This meant that female mice could be generated which carried one mouse opsin (L/M) and one human version (L). Cone cells could express one or the other of these genes. The result was astonishing – in visual tests, these mice could clearly distinguish between light of wavelengths which they were previously unable to discriminate. (They could now tell red from green). Despite normally having only two channels, their nervous system was clearly primed to perform this comparison.

Amazingly, this may extend to humans as well. The opsin genes in humans can also be polymorphic – each one comes in several different versions. Females who carry one version of, say, the L gene on one X chromosome, and another on the other X chromosome, can effectively have four different channels of absorption: S, M, L and L’. If the retina is primed to compare inputs based on their patterns of activity then one would predict that such females would be tetrachromatic – they should be able to distinguish between more colours than trichromatic individuals (just as trichromats can distinguish more colours than dichromats – people with a mutation in one of the L or M opsin genes, who are red-green colourblind).

This increased ability to discriminate colours is, apparently, indeed present in about 50% of females and can be revealed by a very simple test. Consider the picture of the colour spectrum shown below. If you print this out and mark on it with a pencil everywhere there seems to be a clear border between two distinct colours, then what you will find is that most trichromats mark out about 7 colour domains, while tetrachromats mark out between 9-10 (and dichromats about 5).

So, where a man may just see “green”, a woman may see chartreuse or olive. Realising that people literally see things differently (and not just colours) could avoid needless argument. (That said, the woman is clearly more right, and it is usually best to concede graciously).

Field GD, Gauthier JL, Sher A, Greschner M, Machado TA, Jepson LH, Shlens J, Gunning DE, Mathieson K, Dabrowski W, Paninski L, Litke AM, & Chichilnisky EJ (2010). Functional connectivity in the retina at the resolution of photoreceptors. Nature, 467 (7316), 673-7 PMID: 20930838

Jacobs, G., Williams, G., Cahill, H., & Nathans, J. (2007). Emergence of Novel Color Vision in Mice Engineered to Express a Human Cone Photopigment Science, 315 (5819), 1723-1725 DOI: 10.1126/science.1138838

Jameson KA, Highnote SM, & Wasserman LM (2001). Richer color experience in observers with multiple photopigment opsin genes. Psychonomic bulletin & review, 8 (2), 244-61 PMID: 11495112

Mirrored from


  1. @Kevin said…
    “Colour does not exist. Not out in the world at any rate. All that exists in the world is a smooth continuum of light of different wavelengths. Colour is a construction of our brains.”

    My understanding is that we (humans) don’t even see color, but see Metamers.

    That we (humans) perceive many different spectral power distribution as the same “color”. This is why color printers and color screen are even possible, using things like the RGB, CMYK, YCbCr, etc color spaces. (Computer screens and color printers, in general, don’t reproduce the spectral power distribution in nature, but produce a spectral power distribution that is a metamer, or “close” to being a metamer.)

    Now, from what I understand, it’s even more complex than this, in that depending on what other stimulus we can perceive a spectral power distribution differently. (For example, putting a “color” next to another “color” can have you perceive that color differently.)

    (But that’s going off on a tangent, so I’ll just stop there.)

  2. Interesting: I see ten colors there that all seem rather distinct to me. Yet I am a male.

    (As a kid, I also thought it was odd that we have just ‘blue’ and ‘purple’ to cover a range of clearly distinct colors (tho I suppose that the now more common use of ‘cyan’ and ‘magenta’ help split things up) and that chartreuse wasn’t always readily distinguished from yellow)

  3. Given that we are all looking at your sample spectrum on an RGB screen, is it really possible for anyone to see anything beyond the usual set of trichromat colors? Don’t you need an actual physical fourth color for that — something like an RGBU screen?

    Also, I remember reading in some science magazine that the vertebrates we evolved from did originally have 4 opsins, that mammals lost two of them (probably because they were nocternal), and that primates gained back one. However fish, birds, and reptiles never lost the original 4, so tetrachromatic color vision is common in those groups. (This is just from memory — I couldn’t find the article).

  4. jb: you’re quite right about most birds being tetrachromatic and it is indeed likely that early mammals lost a couple opsin genes. Many birds can see into the ultraviolet range thanks to the absorption spectrum of one of the opsin genes. Cone cells in many birds apparently also contain a droplet of coloured oil which acts as a filter to further sharpen the tuning of each cone cell. (Amazing what you can learn on Wikipedia!)

    As for the worry of looking at the spectrum on the RGB screen that should not be a problem – those colours combined can generate thousands of distinct shades at each pixel.

    The question is how many distinct “categories” we each see, which may be determined by our opsin gene repertoire.

  5. Many years ago I read Edwin Land’s retinex theory in Scientific American (back when it reported on science). What’s the status of his theory?

  6. kjmtchl: when I was very young I learned that there were three primary colors, and I wondered “why three? Why not some other number?”. Much later I learned it was because we have three types of color receptors in our eyes, so that three different frequencies of light is all it takes to stimulate all possible responses. In contrast our ears can distinguish thousands of frequencies, which is why a speaker that could only produce three frequencies would be totally useless.

    An RGB screen may be able to generate what appears to our eyes to be thousands of distinct shades, but each of those shades is actually just some combination of the same three color frequencies (R, G, & B). You can make chords on a piano with four tones that you simply cannot reproduce using three, and in the same fashion you can make tetrachromatic colors with four distinct frequencies of light that cannot be reproduced with three frequencies. For this reason I’m pretty sure that a true representation of tetrachromatic colors on a computer screen would require four distinct colors (RGBU being only one possibility).

  7. Four color printing used to be standard (CMYK) and there are many six and eight color desk top printers on the market. Of course, color printing is color by subtraction.

Leave a Reply