Colour perception

Question 1

Spectra and colour matches

Answer 1

Colour is perceived differently in different species, depending on their opsins - so reflected colour from a real lemon has a different spectrum than a metameric match made using RGB pixels only, which to our eyes looks identical. To an animal with different opsins, these two spectra would look different from each other.
Sunlight and tungsten lightbulbs emit light across a whole spectrum. Fluorescent light bulbs and lasers emit only at one wavelength.
The illuminance spectrum matters, because it is combined with the surface reflectance of an object to produce the colour signal.
It’s all about the /relative/ stimulation of different cone types.

Question 2

What is colour vision

Answer 2

The ability to detect different wavelengths independent of their brightness

Question 3

How do we detect colour? (briefly)

Answer 3

visual pigment = rhodopsin plus chromophore
chromophore is the biggest determinant of spectral sensitivity
Young Helmholtz Maxwell theory - trichromacy through three cone types of different sensitivity
Hering colour opponent theory - three cone types into a broadband and two subtracted colour channels - allowing black/white (for intensity), red-green, and yellow-blue.
Grasshoppers use corneal filters as the basis of their colour vision (they only have one visual pigment), and many insects have distal photoreceptors filtering the light for proximal.

Question 4

How can we prove colour vision in animals?

Answer 4

First, prove they have more than one opsin (although grasshoppers..)
Then, can they distinguish between a coloured object and a black/white object of the same intensity (i.e. reflecting the same number of photons?)

First achieved by Karl von Frisch in 1914 - put a sugar dish on a blue card surrounded by greys of varying intensities. The bee reliably went to the blue card, wherever it was put that trial, proving colour vision. Further, when the bee was allowed to return to the hive, it returned with many more bees, proving communication - von Frisch later proved they communicated via the ‘waggle’ dance.
The bees could not distinguish between the different greys, so von Frisch suggested bees use colour vision only. But Lehrer showed that motion perception in bees is achromatic. So bees use colour vision for specific tasks, and can disregard it for other tasks.

Many animals are the other way around, and use monochromatic vision for learning tasks (because it is higher resolution in most animals), and thus colour vision wasn’t proven in mammals until much more recently.

Experimenters need to adjust intensities used to the species’ spectral sensitivities. They can test these, but sometimes they vary depending on the conditions (goldfish had different sensitivities when learning to recognise a dark object vs a light one), and sometimes they’re subject to chromatic adaptation.

Question 5

Adaptations of and to colour vision

Answer 5

Trading off spatial resolution:
Nocturnal hawk-moth - Superposition eyes, with large lenses and short focal length, allow UV, blue and yellow light detection under starlight
Helmet gecko - no rods, and longer + more sensitive cones, allow distinguishing between blue and grey under moonlight.

Oil droplets to focus light:
Clear ones [lungfish] increase range of wavelengths that can be discriminated
Coloured ones [chicken] will focus particular colours of light better by interfering with periodicity - they absorb wavelengths of light below a critical value, narrowing the spectral sensitivity of cones function, hence reduction in overlap. This is only useful when there’s sufficient light that noise is not a problem, because it will reduce the overall quantum catch too.

Rhabdomere pigment filters:
Ogawa et al 2013 - the ventral portion of the Colias erate butterfly is sexually dimorphic - females can see red better, because the proximal photoreceptors in one of their three ommatidia types has a sensitivity 40nm lower than in the male, due to different perirhabdomal pigment
Stomatopod (mantis shrimp) has 4 blue filter types to tune 2 UV rhodopsins. This is the most complex eye on the planet, with 12 cone types.

Question 6

Colour vision defects in humans

Answer 6

38% of Caucasian men have variation in red opsin spectral tuning
1% of men have a missing red cone - protanopes
1% of men have a missing green cone - deuteranopes
0.001% of people have a missing blue cone [not sex-linked] - tritanopes
0.00001% of people are rod monochromats.

Question 7

Why have colour vision?

Answer 7

Likely evolved to deal with changing illumination conditions, but also:
Enhances contrast
Enhances object categorisation
Allows image segmentation
Improve saliency (more details to make memory trace more unique)
Break camoflage
Enhance own camoflage (note - this doesn’t require conscious processing of colour, e.g. cephalopods)
Improve intra-specific communication (e.g. UV butterflies, UV stripe evolved at same time as duplication of UV opsin)
Improve inter-specific communication (e.g. flowers colours coevolve with pollinator spectral sensitivities, can be species specific)
Allow colour constancy
Improve motion perception
Improve image compression (i.e. you only need certain aspects of a scene/object to recognise it)

Question 8

Colour ‘vision’ in cephalopods

Answer 8

Cephalopods (octopus, cuttlefish, squid) change colour as camouflage. It was assumed this relied on eyesight. But biopsies of the skin changed colour when a different light was shone on them, independent of brain/eye. They found that the skin responded most quickly to blue light (similar peak sensitivity to the opsin in the eye), then found expression of a gene for an opsin in the skin similar to the opsin in the eye, which was later found to be in the chromatophores themselves. That is, the light-sensing structure is in the colour-changing component.

Question 9

How do we measure colour sensitivity

Answer 9

Ideal is intracellular recordings, which can be achieved by making an 80 micron hole in the fly cornea and then impaling a single photoreceptor. Can measure voltage change over time exposed to light, and voltge changed over changing light intensity. If you use a monochromator, you can assess spectral sensitivity.
Old-style monochromators - xenon bulb, grating to split it into its component colours, then slit to direct one of those colours towards the animal. Xenon bulb subject to temperature fluctuations, and slow.
New style monochromator - different coloured LEDs in an array, a planar reflective diffraction gating acting as a wavelength combiner, and launching the combined light into a fibre. Only switch each LED on when its input is needed.
Alternatively, electroretinogram - one electrode on the eye, one on the body, recording summed responses of all the cells the light hits. May be hard in animals with many different cone types.
Visual pigment Microspectrophotometry - prepare cryo-sections of retina in the dark (or red light only, to avoid bleaching) and mount on microscope. Bleach with a particular wavelength light, then scan again. For each wavelength, average over many aquisitions, and repeat with many sectioned rhabdoms. Scans normally range 300-600nm, in 1-5nm steps. You’re measuring the absorbance, and estimating spectral sensitivity from that.

You can also estimate from the amino acid sequence
Ogawa et al 2013 used the reflectance spectra from the butterfly tapetum to estimate.

Question 10

Colour constancy

Answer 10

All insects tested before 2014 showed colour constancy. Important in identifying the same colour under different lighting conditions. Werner’s bee Mondrian showed they can do it, though not all illumination changes were compensated equally well
von Kries thought it could be due to adaptation - they get used to the illuminant light, so all lighting conditions become the same. But Lotto and Chittka showed that bees have colour constancy immediately (not slowly like adaptation), and can detect a colour under conditions A and an opposite colour under condition B - rather than adapting to the illuminant, they were using it. Also, bees alter spatial foraging in patchy illumination to avoid changing conditions!
Computer models found colour blind bees were the worst at foraging, but chromatic weren’t much better. Neither von Kries adaptation, white patch calibration (which assumes brightest part of scene must be white) or grey world assumption (which assumes the average of each channel is the representative grey of the illuminant.) were as good as perfect colour constancy.
Bees have lateral inhibition, in the lamina for example, which may underly it.

Question 11

Variation in visual pigments

Answer 11

Mammals use A1 derivative chromophores, reptiles and fish use A2, insects use xanthophyll derivatives. Using A2 causes a red-shift from A1, but most of the variation in spectral sensitivities comes from the opsin.
Most animals just use one type of chromophore, but fish and amphibians often combine A1 and A2 derivatives
Mice have two types of chromophore in the same opsin!

Question 12

Opsin categorisation

Answer 12

Were originally split into r-opsins (found in rhabdomere-type photoreceptors, depolarise via Gq. Vertebrate melanopsins included.) C-opsins (found in cilial photoreceptors, which hyperpolarise via transducin).
Then Eakin introduced the Cnidops (which predated the other two) and Group 4 (a big mish mash, but including the scallop Go linked, which may suggest the beginning of a new evolutionary line).
Remember not all opsins are image-forming! e.g. melanopsin, cephalopod skin opsin.

Question 13

Levels of colour vision - what are the levels?

Answer 13

1) Spectral discrimination independent of spatial detection (e.g. phototaxis)
2) Spatial and spectral inputs combined (wavelength-specific behaviour)
3) Innate colour preferences that require learning (i.e. animal has a conscious representation of colour)
4) Can use colour appearance (e.g. hue vs saturation) or colour categorisation to make behavioural decisions

Question 14

E.g. of 1st level of colour vision

Answer 14

Spectral discrimination independent of spatial detection:
Daphnia spp move towards yellowish water. This has been shown to be independent of intensity. Daphnia have 4 types of photoreceptor but no image forming eye. This behaviour is to maintain position in the water column

Question 15

E.g. of 2nd level of colour vision

Answer 15

Innate preferences for coloured objects:
Doesn’t require conscious colour perception, just ability to discriminate between wavelengths and guide behaviour by that. Similar idea to ‘blindsight’.
Fiddler crabs show sexual preference for longer wavelengths of colour on claw. They approach yellow claws over grey claws regardless of intensity, but find it hard to distinguish between yellow claws of varying intensities.
Papilio butterflies have eight photoreceptor types, but only use three when looking for leaves to lay eggs on, and only one opponent mechanism on these, so are technically dichromat in this behaviour. However, when looking for a mate they are tetrachromatic.

Question 16

E.g of 3rd level of colour vision

Answer 16

Using colour in cognition and learning:
von Frisch’s bees learnt to go to the blue square. This was not an innate preference, because they didn’t go there immediately on the first trial.
Papilio butterflies can learn while mate-finding, but not while looking for leaves to oviposit on. This makes sense evolutionarily, since reproductive success is only determined after the female’s death - how can you learn from something if you’re not alive for the outcome?
Monarch butterflies can be trained to a certain colour flower for a sugar reward, then switch to a different colour when you alter the paradigm.

Question 17

E.g. of 4th level of colour vision

Answer 17

Colour categorisation and representation of colour qualities:
Was argued that these require and are determined by linguistics. But:
Train pigeons on two standard lights, of different wavelength. Give them a novel test light. They have to pick which of the standard lights it was closest to in wavelength. There were two test light wavelengths that were equally likely to be put in either category by the pigeons. These ‘ambiguous’ wavelengths were the same when the standard lights were blue-shifted by 20nm, indicating they’re on the boundary of pigeon colour categories, not simply intermediate wavelengths.

Chicks generalise colours - i.e. if you train them to like yellow and to like red, they will also like orange, but will not like far-red or green. Uncertain as to why they /preferred/ orange…

Honeybees, when trained at two wavelengths, have a pretty flat preference for wavelengths between the two.

Question 18

Grassmann laws of additive colour mixing

Answer 18

1) colour is specified as hue, saturation and brightness
2) mixing any two complementary colours will produce grey
3) mixing two colours of the same hue and saturation will produce another colour of the same hue and saturation
4) The total luminance of any mixture of light is the sum of each light’s luminance

Question 19

Additive and subtractive colour mixing

Answer 19

Additive coloured light (RBG) will produce yellow, cyan, magenta and white. Subtractive coloured pigments (YCM) will produce red, green, blue and black.

Question 20

Colour spaces

Answer 20

A colour space e.g Maxwell triangle is a graphical representation of how wavelength information appears to a given animal. That is, it plots the different spectral sensitivities of each pigment. It doesn’t include brightness, and is not perceptually uniform.
The colours along any straight line can be made by mixing the two colours at the end of that straight line
The edge of the colour space is pure monochromatic light at 100% saturation, and can be represented purely by a wavelength
McAdam tested people on accuracy of colour-matching, and found it varied from colour to colour
The colour space is about what we can see, not necessarily what we can perceive - it will ultimately be mapped onto a perceptual colour space, as shown by electrophysiology, but to compare perceptions between animals in an unbiased manner is hard. So we still use different species’ Maxwell triangles.

Question 21

Opponent circuits in visual system

Answer 21

Colour opponent signals arise from synaptic feedback mechanisms already in photoreceptors (e.g. lateral inhibition)
Magnocellular - large receptive field, fast responding, broadband (i.e. monochromatic), net convergence so high sensitivity low resolution. Peripheral vision, motion, coarse
Parvocellular - small receptive field, slow responding, colour-sensitive (red-green opponent, and blue-yellow opponent), net divergence so low sensitivity high resolution. Foveal vision, colour, fine grain
In humans, RGCs mainly terminate in LGN and superior colliculus (but also pretectum and hypothalamus).
Labelled lines continue to thalamus (PPPPMM layers in LGN, alternating right and left eye)
Fast pathway to the amygdala bypasses cortex, rest goes to V1, then split into dorsal ‘where’ and ventral ‘what’ streams.

Question 22

Proof of neural correlate to colour processing - primates vs insects

Answer 22

Full-field red-green flicker in marmosets shows black blobs in visual cortex using intrinsic imaging (imaging of oxygenation changes using near-infrared light)
In primates, blobs - colour processing - are in V2, but colour space respresentation is in V4, in the inferior temporal cortex.

In bees, colour opponent neurons supposedly start in inner medulla (comprised of retinotopic cartridges, colour comparisons, >50 neuron types), and opponency continues along pathway.
In each ommatidium, 6 of the photoreceptors have the same spectral sensitivity and project to the lamina, 2 (in flies) or 3 (in bees) have different spectra and project to the medulla. In the proximal medulla and layers 5 and 6 of lobula, colour-opponent cells have been found by electrophysiology.
Some colour-specific neurons have been identified, e.g. DM8 in Drosophila is involved in UV phototaxis

Question 23

Looking for a neural colour space map

Answer 23

Xiao et al 2014 - Intrinsic imaging (oxygenation changes under near-infrared light) of V2 in macaques found areas that responded to particular colours. There was large overlap, but a systematic shift with the colour of the stimulus. Interestingly, the spatial shift of the response peaks followed the perceptual hue order of the stimuli. The colour-responsive areas were in cytochrome oxidase thin stripes. They also found hue maps in V1 (in blobs), but an order of magnitude smaller.
But it could just be a relay from feedback areas, and they could just be labelled line (i.e. different colour is sent different places, yes, but it’s not perceptually rearranged in V2.)??

Later research looked at V4, comparing glob cell colour-tuning to various computer models and found that glob cells had a narrow hue response and could distinguish luminance from hue. This could be a perceptual colour map.

Categorisation may arise from the middle frontal gyrus, a linguistic rather than visual area

Question 24

Invertebrate eye mosaics

Answer 24

Expression of different colour receptors is genetically encoded, but stochastic in location of expression within the retina.
Dorsal 1/3rd of the eye has different complement of R7 and R8 cells, sensitive to UV and involved in phototaxis

Flies, humans and chickens are mosaic.
Zebrafish retina are highly organised. Function not understood

Question 25

Receptor patterning strategies

Answer 25

Mosaic - i.e. essentially random
Localised zones
Bands/territories - like localised zones, but more different zones.

Question 26

Colour illusions

Answer 26

Mainly due to centre-surround antagonism, and imperfecct colour constancy
–humans get colour constancy by comparing large parts of an image, not by von Kries adaptation, because studies suggest we can reliably detect illuminant conditions –
The comparison across neighbouring signals means an individual point can be sacrificed to perceiving the whole display
Humans can interpret the change of illumination, but can also be fooled by unnatural shading

Triggerfish are also subject to illusions - fish were trained on a Rubix cube illusion, first with all colours except orange and brown changed to grey, then with colour back in (and unnatural shading). Fish trained to go to the orange square went towards the brown square.

Question 27

Evolution of colour vision

Answer 27

Each visual pigment offers a selective advantage, and the vast variety of pigments have evolved to optimise niche exploitation while avoiding predation, in a variety of light environments.
It is most likely that the existing variety evolved from an initial four photoreceptor types, through duplication and deletions (reviewed in Kelber, 2016). It’s unlikely that insects have a cognitive perception of colour, more that colour information is directed to different brain regions to direct different behaviours, e.g. anterior optic tubercle receives polarity, intensity and chromatic information, combines it, and sends it to parts of the brain involved in flight direction (i.e. acts as a sky compass).
Opponency probably started for monochromatic spatial vision, to enhance edges, then was co-opted for colour vision - when we gave monkeys and mice a new photoreceptor, they developed new colour vision, presumably using opponency circuitry already in place

Gilad et al 2004 - acquisition of colour vision in old world monkeys, apes and the howler monkey coincided with loss of olfactory receptors (60% of human olfactory receptor genes are pseudogenes, with an interrupted coding sequence)
In lemurs and new world monkeys, there’s allelic trichromacy, with frequency-dependent selection. This may be because it’s beneficial to have some monkeys dichromats (better at seeing insects) and some trichromats (better at finding fruit)

Apart from butterflies, spectral sensitivities are similar within major taxa. This suggests they didn’t evolve to match food/environment etc. Yet they do seem to match, eg. bees are tuned to flowers they pollinate. Since bees colour vision evolved before angiosperm flowers, it must have been the flowers evolving to meet them, not vice versa.

Question 28

Mantis shrimp vision

Answer 28

Dorsal and ventral hemisphere of eye, split by midband. Hemispheres are for luminance. In two species, the midband has 6 rows of ommatidia - 1-4 are involved in colour processing, 5 and 6 in recognition of linear or circular polarised light (using UV photoreceptors)

12-20+ different photoreceptor types, multiple filters. Why so many, when three or four are good for colour discrimination?
Thoen et al 2014 - Mantis shrimps have surprisingly poor colour discrimination performance, suggesting they don’t use an opponent system. Authors compared performance on colour discrimination test to modelled opponent system (using noise-limited model than assumed analogue comparisons between adjacent spectral sensitivities). Mantis shrimps were worse, only reliably discriminating to within 25nm, suggesting they don’t make such comparisons.
Instead, they scan their eye across the object, producing a temporal pattern as it hits each of the cones. This pattern represents colour. It’s less precise, as it’s determined by the distance between different spectral sensitivities, but it’s quicker than processing onto a perceptual colour map.

So whilst they’ve got the most complex colour capture system (photoreceptor wise), they may not have the most complex colour processing system. Their spectral sensitivities look a lot like the tuning curves of V4 cells in macaques.

Question 29

Mantis shrimp uses for vision

Answer 29

mantis shrimp have a ‘meral spot’ that varies in UV reflectance with sex (so may be used in sex recognition/mate choice) and with habitat. During fights, they perform a meral spread that reveals the spot.
Franklin et al 2016 - When UV reflectance is reduced (by painting over meral spots with sunscreen), mantis shrimps are more keen to engage in a fight - that is, they underestimate strength and/or aggressive state of the opponent.
When ability to detect chemical signals is abolished (by dipping antennae in freshwater), mantis shrimp approach a refuge quicker (because they don’t know it’s occupied) and are less keen to engage in a fight - that is, they hesitate because they normally use chemical cues to judge strength of opponent.

Question 30

Butterfly colour discrimination

Answer 30

Papilio was the first invertebrate shown to be a tetrachromat (shown because testing spectral sensitivities reveals a discrimination minimum in the trough between two sensitivities)
Papilio’s is almost as good as a human’s in much of the range (discrimination threshold 1-2nm), better in some visible spectrum regions, and better in the ultraviolet (obviously). Diurnal hawkmoth is similar, and can use achromatic cues to distinguish between lights of long wavelength

Question 31

Structure of the bee eye

Answer 31

Each ommatidium contains 8 large receptors and 1 small basally located one
6 of the large ones are always L-receptors
In type I ommatidia (44%), the other 2 are M and S
In type II ommatidia the other two are both S
In type III ommatidia, the other two are both M
We don’t know what the small one is for
the 6 L receptors project to the lamina, R7 and R8 straight to medulla

Bees use L receptors (which detect into the UV) for brightness, with perhaps some contribution from M. Bees exhibit phototaxis across the whole spectrum, so it’s possible they use info from all three types for brightness.

The photosensitive parts of the photoreceptors are the microvilli, fused together in a waveguide like process parallel to the optical axis of the rhabdom. May help detect polarised light.

Question 32

Bees use of chromatic/achromatic signals

Answer 32

Bees were trained to approach disks with an outer and inner ring of different colours
When L-contrast was high, bees approached disks that subtended an angle of 5 degrees, i.e. hits 7 ommatidia.
When M or S-contrast was high, disks had to subtend 15°, sampled by 59 ommatidia. This suggested the chromatic system summates more, sacrificing spatial resolution for contrast sensitivity.

When the L-contrast subtended an angle more than 15°, it wasn’t recognised. This suggests Gaussian centre-surround antagonism, and more complex computation than linear summation. Could be achieved by laminar interneurons, connecting neighbouring cartridges
BUT disks were more likely to be detected when grouped with others, even if sufficiently far apart so that images did not merge. Also, detection didn’t increase when brightness was increased but L-contrast stayed the same, which the Gaussian model would have predicted.

Disks with a bright L-contrast centre and dim surround, or vice versa, were detected worse than single colour disks, the former especially - not clear why. Maybe edges - but why not outer edge? Maybe must be closer to limit of detectability - but why was bright centre dim outside worse?
M- or S-contrast disks were detected exactly as single colour disks. All we can say is it’s more complex than linear summation

Question 33

Bee pattern detection

Answer 33

Hempel de Ibarra et al 2002:
When patterns subtended a large visual angle, chromatic cues were used. There was a generalisation and hence innate preference towards yellow and orange centres. When patterns subtended a small visual angle (near limit of detection, so L-contrast had to be used), pattern detection was poorer.

[Using patterns where both colours stimulated the L receptors equally:
Pattern recognition was near perfect above and close to the chromatic threshold of angle subtended
Below the chromatic threshold but above achromatic threshold, patterns could not be distinguished
So despite the achromatic system having better spatial resolution, the chromatic system is used for pattern detection]

Note that L-contrast is necessary for distinguishing between closed shapes e.g. triangle and square, even at large angles

Question 34

Examples of vertebrate colour communication

Answer 34

Chameleons - darken as a submissive signal, in response to aggression. Aggression is decreased when opponent darkens.
Brown anoles - flashing dewlap, to signal ability to win contest for females and territory
Peacocks - 50% of reproductive success explained by eyespot colouration. Hue and irridescence of blue-green patch most important. Eyespots directly influenced mate selection (masking them reduced copulation success almost to 0). Males display at 45° to the sun’s azimuth, with female directly in front.

Question 35

Examples of vertebrate use of UV

Answer 35

Birds use UV cones for - magnetoreception (cryptochrome 1a found in robins and chickens only in UV cones, arranged in ordered bands along the membrane disks of the outer segment. This allows them to use magnetic field lines for navigation, via the Radical Pair process)

• Mate selection and altering sex ratio of offspring - in male blue tits, high UV reflectance of the crown predicts survival to the next breeding season (suggesting it’s a measure of fitness), and affects sex ratio of offspring. When UV reflectance is high more of the offspring are male. This effect is reversed when the crown is masked with sunblock
• finding berries - redwings preferred UV-reflecting bilberries to rubbed bilberries when there was UV illuminance. Naive (younger) birds showed no such preference, suggesting it’s a learnt thing

Arctic caribou have cornea and lenses that transmit more UV than most mammals. Retinae respond to UV light physiologically. The snowy environment scatters more UV which may decrease UV-contrast, but photos taken suggest that in fact plants stand out more in UV-images than when UV is blocked. So caribou use UV to enhance discrimination of plants, urine and fur (predators)

Question 36

Methods for determining spectral info processing

Answer 36

Electrophysiology - record and fill neurons, trace connections
Optogenetics and temperature sensitive block/activation - alter circuits in real time and observe effect on behaviour
fMRI - real-time imaging during a behavioural task. But in humans we can only image populations, not single neurons - use insects instead
Electron microscopy - to visually trace connections to form a synaptic and gap junction connectome
Genetic labelling - targeted and stochastic labelling and trace connections

Also?:
Virus vectors (can determine mono vs polysynaptic connection if you remove the glycoprotein coat so it can't reverse transcript and can only cross one synapse)
Takemura et al 2013 - used semi-automated pipeline of electron microscopy to determine connectome of a repeating unit in medulla, and identified a possible motion detector circuit there.
Filtered Back Projection - used for determining receptive fields of ganglion and bipolar cells on a multi-electrode array. Quicker than spike-triggered average, and allows detection of both 'on' and 'off' fields, and orientation selectivity
Simultaneous 2-photon imaging of a genetically encoded calcium indicator and light stimulation can show you spectral response curve and light intensity response curve. (but do not achieve single-spike resolution, especially when frequency is above a few hertz)