Bottomup Part 1: transduction, inhibition, neural coding
PSY380H1S L2 July 07, 2014 Try to find articles from Science, Nature – more relevant Brain.psynup.com – extra marks!
Bottomup • 2 ways to look at bottom: o Raw data o Analysis of subsystems into subsystems • First step in visual perception o Retine gives info about light in envt
Use other nuclei to control muscles in eye based on input from eye & other brain regions
What is the output of the eye? Difficult to answer • SubQ: Does the eye just convert light signals to physiological signals, or does it also perform meaningful computations on that info?
What is the input to the eye? Light! • carried by units/waves of energy called photons • Released when something (sun, light bulb) expels energy o When electrons move btwn orbitals • The eye : regions to which the eye signals, transduction, types of eye movements (and relevance for visual perception), • More energy higher wave freq photoreceptor cells, opsins (photopsin, rhodopsin, melanopsin), o Dif colours reflect dif energy levels optogenetics, compound eyes, rods, cones (L, M, S), retina, fovea, • Photons always travel at the same speed (299,792 km/s) optic nerve, optic disc, blind spot, bipolar cells (ON & OFF), o Anything traveling faster is presumably going back in time amacrine cells, ganglion cells, neuron coding, receptive field, firing rate, centersurround, lateral inhibition, resolution Photons are either absorbed or reflected from matter. • Retinal computation : how neurons/circuits compute, what retinal • Colour/texture determined by source of light (diffused or focused) circuits compute, circadian rhythms, temporal versus rate codes, or surface of object (scattering light – matte, or bouncing light in a feedforward, feedback precise direction – specular) • Lateral geniculate nucleus: parvocellular, magnocellular, • Objects w colour absorb only some wavelengths. koniocellular o Something red absorbs every wavelength but red, and reflects only red Part 1: transduction, inhibition, neural coding • The direction of reflection depends on the surface The eye • What is the bhvr of the eye? Where does the output of the eye go?
Input is not just light, but moving light • Our eyes are moving too
• • • • •
Multiple brain regions receive info from eye Pretectum (in midbrain): controlling pupil constriction, lens accommodation (stretching) o Controlled by ANS (EW nucleus) Suprcahiasmatic nucleus (in hypothal) o Regulates circadian rhythms Lateral geniculate nucleus o Thalamic nucleus, relays info from retina to cortex Superior colliculus o Very old, evolutionarily o Controlling eye movements & other bodily movements o Reflex loops in visual system
Types of eye movements: • physiological nystagmus: tremors in eye muscles o if you eliminate these tremors, the visual world will disappear within a few seconds! o We are always moving o So the eye depends on the fact that we move around o Computer image that moves w our tremors no sight • saccades: rapid, abrupt eye movements o 150500 ms fixations broken by ~30 ms saccades o don’t perceive changes during saccade perception inhibited or blocked • smooth pursuit movements: track objects in a smooth, continuous way • vergence movements: eyes move in different directions
• •
tracking an approaching or receding object ex. trying to be crosseyed vestibular movements: eyes move when head turns o anytime your vestibular system detects head rotation eyes automatically move in opposite direction optokinetic movements: eyes move when world moves
o o
How does light affect the eye? • Transduction: The conversion of signals from one type of energy into another. o Light energy physiological energy • Light transduction takes place in cells called photoreceptors, using molecules called opsins Opsins (a brief glimpse into molecular biology) • Proteins in outer layer of cells (membrane) • Attach to a vitaminA derivative o Carrots have vitamin A • Change shape when photons absorbed o change of events … physiological signals • Evolved more than once: o Type 1 : very old; found in microscopic organisms; normally fast acting (when directly linked to ion channels or pumps) o Type 2 : found in animals, normally slower acting (when indirectly linked to ion channels) • Many families w different specializations
• • • • •
Why opsins are trendy: optogenetics You use a virus to inject the DNA for certan types of opsins Can inject them in a way that the opsins only get expressed in particular types of cells Then you can put a laser or LED light to trigger those neurons or inhibit them So now we can precisely (cell type, location, temporally) stimulate or inactivate neurons o Very powerful
• • • •
2 main types of photoreceptors: rods & cones Outer segment: opsins found Cones: Photopsin Rods: Rhodopsin
Photoreceptor cells: rods vs. cones • Cones are less sensitive to light than rods o Rods become relatively more important when the light is very low • Cones are more dense at the fovea (centre of retina), less dense in the surrounding retina o Rods provide more peripheral vision info • Cones & rods contain dif opsins (photopsins vs. rhodopsin) o Several types of photopsins Dif types of cones transmit info about dif wavelengths of light Start of colour perception o Color info is transmitted by the set of cones & rods that are stimulated 3 types of cones
Photoreceptor cells
• • • •
S: detect 400450nm wavelengths Rods btwn S & M M: green L: longer wavelengths
Acquiring info from the eyes • “eyespots” 570500 mya o Detect if light present or not • Cuplike eyes o directionality • pinhole eyes o light enters from specific directions o more directionality o get a sense of an image if have enough recs • Complete closure o Some species • In some species: duplication of eye (compound eyes) • Addition of lens to focus light o Human eyes o So don’t need precise pinhole for directionality – the lens helps us focus
• •
As you move, pay attention to the black dot (attending to something that you aren’t looking at is called “covert attention”) At some point (when your eyes are ~3 x the distance between the cross and dot) the black dot will disappear. o It is in your blind spot—your visual system is “filling in” the area w background
The retina
Biological Psychology 6e, Figure 10.10 • Light passes thru vitreous humour ganglion cells bipolar cells photoreceptor cells • So we get less photons coming in because it has to go thru other cell layers o Less blocked in fovea, more direct
The human eye: anatomy
• • • •
Fovea: centre of retina, where world converge Optic disc: axons of retina ganglion cells converge travel to other areas of brain, beginning of optic nerve o Blind spot Optic nerve: bundle of axons carrying info to other brain regions Retina: inner surface of eye where light is projected onto
1) Close your left eye 2) Stare straight at the cross with your right eye 3) Move your head closer to the image
Cajal (1911) – detailed drawings of brain circuits from microscope; theorizing purposes • Photorecs, Horizontal Cells photorecs, bipolars • Bipolar cells (Layer 2) • Amacrine cells o bipolar cells, ganglion cells • Ganglion cells (Layer 3): outputting to brain Gollisch & Meister, Neuron, 2010 Sequence of events in the retina – the circuit of the retina
•
Light always inhibits photoreceptors
o
•
Lack of signal activates Bipolar ON cells (light on)
Light off photoreceptors constantly fire
o
•
•
A neuron’s receptive field: the variable space the neuron codes for o Can literally mean retinal space o In other areas of brain, real coordinates Ex. Hippocampus: fire when body in particular area of space A neuron’s firing rate: the # of APs per unit of time
Centersurround codes Activate Bipolar OFF • Can have CenterONSurroundOFF or CenterOFFSurroundON • Below: CenterONSurroundOFF cells o Center = excitatory, outside = inhibitory
Photoreceptors (rods & cones) • Input : o light (inhibitory) o horizontal cells (inhibitory) • Output : o Bipolar cells (excitatory or inhibitory) o Horizontal cells (excitatory) Bipolar cells (ON & OFF) • Recs determine if on or off • Input : o either cones or rods – not both o excitatory (OFF) or inhibitory (ON) • Output : o amacrine cells interneurons inhibit bipolar & ganglion o or ganglion cells (excitatory) Ganglion cells • (midget, parasol, bistratified, ip, etc.) • Input : o bipolar cells o amacrine cells • Outputs : o multiple brain nuclei
•
•
Neither photoreceptors or bipolar cells have APs o Signal continuously in graded fashion o More light – more changes Ganglion, amacrine – have APs
• •
•
Photoreceptors & lateral geniculate nucleus have receptive fields If show small spot of light in visual space o By chance, 1 AP appears after
Light in center of receptive field o APs in reaction
Info in the vertebrate NS is transmitted via signals btwn cells. • Neurons transmit signals when they experience an electrical event known as an action potential (AP) • When the rates or patterns of APs changes in relationship to a particular variable (ex. a sensory stimulus in a region of space), the neuron can be said to code for or represent that variable. Neuron coding • A neuron codes for (or represents) a variable if changes in its signaling, typically measured in the form of APs, are correlated w • that variable
Increase size of light
o
More APs
• • • •
Present light to just one photoreceptor (Ommatidia – compound eye) neuron fires If present light to surrounding areas decrease in activity Discovered before NS inhibition was discovered Fn’al inhibition
What is the fn (consequences) of centersurround fields? • (I.e., what is the consequent circuit bhvr/computation) • Increases resolution of spatial info!
•
Increase size of light more o Less APs than before because some light in surround
•
Light in surrounding area o No APs
https://www.youtube.com/watch?v=9qg9nBjUTc • Hubel & Wevil • Using electrical methods to probe visual sysem • Mapping receptive field on neurons in Lateral Geniculate Nucleus o Electrode in cat’s LGN o High freq pops, amplified
• • • • • •
Blue circles = photoreceptors If all photorecs activated excite all neurons o (no little dots) If put little dots in act laterally sharpen the receptive field via inhibition o Via Lateral Inhibition Top: Photoreceptors Middle: Horizontal Bottom: Bipolar cells
How might centersurround receptive fields explain the Herman Grid What is the mechanism of the centersurround receptive field? Lateral Illusion? inhibition • Not detecting as much light in the intersections because we have center surround codes
•
If rec field at intersection: too much inhibition from white areas in surround see black dots
o •
Btwn black squares: inside white, outside black
o • Haldan Keffer Hartline (~1958) • Studied eye of horseshoe crab – many eyes • Real eyes are lateral eyes – have dif photoreceptors
The outer inhibition overcomes the inner excitation
So no conflict, see white lines
Ultimately, our perceptions are computations
Summary part 1: the eye acquires light information.
• • • •
It can identify where light is coming from due to its design (image passes through a small hole & lens to project onto the 2D retina surface) It can identify which wavelengths are arriving (color) due to its diversity of photoreceptors It can sharpen the spatial resolution using lateral inhibition (Not discussed: it can also adjust to lighting conditions, due to photoreceptor “bleaching,” and inhibition by horizontal cells) o If exposed to light, less sensitive to light immediately after; also due to inhibition by horizontal cells, info from brain (ex. pupil dilation)
What computations take place in the retinal circuit? • The retina has over 50 types of neurons! • >20 amacrine cells o In general inhibitory but much variation • 1020 dif types of ganglion cells carry the output of dif types of circuits. o Specialization of cell types for dif circuit fns • Each dif circuit may extract unique info (features) that get transmitted in parallel to the brain.
Unanswered Qs from part 1: 1) 1) All of this can be accomplished at the level of photoreceptors & horizontal cells. What purpose do bipolar, amacrine, and ganglion cells serve? 2) What are the outputs (ie. the bhvrs, or computations) of the retina?
Part 2: neuron computation, computations of retinal circuits • How do neurons compute? • Each neuron is a processor. o It integrates signals (thru chemical & electrical o
• • • •
What kinds of features are extracted?
transmission) If those signals reach a threshold, it generates a new signal that gets transmitted to a new set of neurons.
Synapses from neurons A & B to neuron C (black) If A or B are sufficient on their own to fire: o OR. operator o Info generalized If A & B have only one connection each: o Have an AND operator Lateral inhibition: A & NOT.B fire o Need more specific conditions to become active
Actually more complicated: • Neurons have dif cellular properties that determine their bhvr. • Certain chemicals (neuromodulators) dynamically change how neurons respond to signals. • Circuits often show “emergent properties” that aren’t obvious from this description
1) Color • Redgreen opponency is transmitted by some “Midget” • • •
ganglion cells Some have Centersurround fields, created by horizontal cells o Midget bipolar cells further transmit their own center surround fields Blueyellow opponency is transmitted by small bistratified ganglion cell Opponency: seeing one colour inhibits seeing the opponent colour
2) Directional motion
• • •
Burst of act pots as light moves across receptive field Raster plot: More spikes when you move according to the preferred direction (up) No spikes when you move down (opposite to preferred direction)
Ganglion cells fire before the moving dot gets to the end of the screen
Other fns: • Overcoming noise (from movement & spontaneous signaling) o Noise: anything that isn’t signal, isn’t useful o Ex. eye muscle twitching = noise, spontaneously signal when no stimulation (like if light on signals, even tho should be inhibited) • Inhibition output during saccades o Ex. during high speed • Contrast adaptation o Ex. dif light conditions, bleaching of photoreceptor during light, inhibition via amacrine cells o Sharp contrasts on sunny days o Eyes accommodate a bit in fog • Adaptation to patterns o Ex. envt w many horizontal lines – your retina will adjust a bit relative to an envt w all verticals o Higherorder computation, impressive Some ganglion cells are intrinsically photosensitive (ip) • use melanopsin o Activated by light • widely branching dendrites o Covering as much of retina as possible to detect how much light is coming in • project to suprachiasmatic nucleus in hypothal (controlling circadian rhythms) •
• • •
Starburst amacrine neurons inhibit directionselective ganglion cells. Internal, cellular processes in the dendrites of starburst cells may make them “intrinsically” directionselective The starburst cells also appear to inhibit ganglion neurons in a spatiallyselective way, amplifying the direction selectivity. o Ex. Dendrites on the side of the ganglion cell connected to starburst amacrine are inhibited from signals by a bipolar, excited only if stimulated on other side on another bipolar (where no amacrine connection to ganglion dendrites)
What kind of features are extracted? 1) Color 2) Directional motion 3) Texture 4) Approaching motion o If light getting bigger, certain ganglion cells will pick up on this 5) Anticipated position o Evolutionarily advantageous to be processed early in retina o Critical because we convert light into physiological signals slowly via Type 2 Opsins (act indirectly within cell) o Computer screen: dot moving across screen, then stop
Temporal Coding: Rate vs. timing of action potentials: • Take longer for Bipolar cells to be inhibited by photoreceptors, so takes longer for ON Bipolar cells to be activated o So cells that respond to darkness are activated faster • Ganglion cells may signal lighter vs. darker image regions by precisely when they respond. o By knowing how quickly ganglion cells are activated in dif areas, we know what areas are darker • Take about 100ms to detect what’s going on in a visual scene o Not much time for physiological changes o So this is much quicker • The temporal code doesn’t require signal integration, so quicker o Over time you can integrate more to get a clearer image
(Image from Gollisch & Meister, Neuron 2012)
•
This is often called a temporal code & can be contrasted from rate codes (# of APs per unit time)
Summary of part 2: retinal circuits do compute • Although there is 1 “type” of retinal circuit (5 basic cell types), • LGN : nucleus of thal where most info from retina goes there are many variations of the circuit that use dif combos of cell o primary visual cortex types o Not just simple transducing The lateral geniculate nucleus is madeup of 2 magnocellular, 4 parvocellular, & 6 koniocellular layers • Each type extracts dif types of info across each area of retina • Thus, computations take place in the retina thru the stepby step (serial) processing of many sidebyside (parallel) • Computations in the system are performed thru serial & parallel processing Important concept list 1) Both topdown & bottomup processes contribute to visual perception 2) Computations in the visual system (and NS in general) take place thru serial & parallel processing.
•
•
• • •
Feedforward serial processing: the signal is stepping from one processing step to the next Feedback: a signal is acting back on an early processing step (Only makes sense if there’s a feedforward)
We still don’t know precisely how the circuits are connected & how they fn. But we may learn more from gamers playing Eyewire (MIT). Eyewire allows you, as a game player, to help identify the wiring, and ultimately fn, of retinal circuits
• • •
2 Magnocellular layers: o Large neurons 4 Parvocellular layers: o Smaller neurons 6 Koniocellular layers o Btwn M&P layers
Ganglion cell (info)
LGN layer
•
Midget (redgreen)
parvocellular
•
Parasol (noncolor)
magnocellular
•
Bistratified cells (blue)
koniocellular •
Specifically KIII & KIV, & koniocellular cells scattered in the other layers)
Part 3: The lateral geniculate nucleus as a bridge to cortex Neurons in the LGN receive more synapses from the cortex than from the retina! • Primary relay centre for bottomup info • Gets more topdown info than bottomup
o
What the cortex receives is already very biased by what the cortex has
Bottomup • 2 ways to look at bottom: o Raw data o Analysis of subsystems into subsystems • First step in visual perception o Retine gives info about light in envt
Use other nuclei to control muscles in eye based on input from eye & other brain regions
What is the output of the eye? Difficult to answer • SubQ: Does the eye just convert light signals to physiological signals, or does it also perform meaningful computations on that info?
What is the input to the eye? Light! • carried by units/waves of energy called photons • Released when something (sun, light bulb) expels energy o When electrons move btwn orbitals • The eye : regions to which the eye signals, transduction, types of eye movements (and relevance for visual perception), • More energy higher wave freq photoreceptor cells, opsins (photopsin, rhodopsin, melanopsin), o Dif colours reflect dif energy levels optogenetics, compound eyes, rods, cones (L, M, S), retina, fovea, • Photons always travel at the same speed (299,792 km/s) optic nerve, optic disc, blind spot, bipolar cells (ON & OFF), o Anything traveling faster is presumably going back in time amacrine cells, ganglion cells, neuron coding, receptive field, firing rate, centersurround, lateral inhibition, resolution Photons are either absorbed or reflected from matter. • Retinal computation : how neurons/circuits compute, what retinal • Colour/texture determined by source of light (diffused or focused) circuits compute, circadian rhythms, temporal versus rate codes, or surface of object (scattering light – matte, or bouncing light in a feedforward, feedback precise direction – specular) • Lateral geniculate nucleus: parvocellular, magnocellular, • Objects w colour absorb only some wavelengths. koniocellular o Something red absorbs every wavelength but red, and reflects only red Part 1: transduction, inhibition, neural coding • The direction of reflection depends on the surface The eye • What is the bhvr of the eye? Where does the output of the eye go?
Input is not just light, but moving light • Our eyes are moving too
• • • • •
Multiple brain regions receive info from eye Pretectum (in midbrain): controlling pupil constriction, lens accommodation (stretching) o Controlled by ANS (EW nucleus) Suprcahiasmatic nucleus (in hypothal) o Regulates circadian rhythms Lateral geniculate nucleus o Thalamic nucleus, relays info from retina to cortex Superior colliculus o Very old, evolutionarily o Controlling eye movements & other bodily movements o Reflex loops in visual system
Types of eye movements: • physiological nystagmus: tremors in eye muscles o if you eliminate these tremors, the visual world will disappear within a few seconds! o We are always moving o So the eye depends on the fact that we move around o Computer image that moves w our tremors no sight • saccades: rapid, abrupt eye movements o 150500 ms fixations broken by ~30 ms saccades o don’t perceive changes during saccade perception inhibited or blocked • smooth pursuit movements: track objects in a smooth, continuous way • vergence movements: eyes move in different directions
• •
tracking an approaching or receding object ex. trying to be crosseyed vestibular movements: eyes move when head turns o anytime your vestibular system detects head rotation eyes automatically move in opposite direction optokinetic movements: eyes move when world moves
o o
How does light affect the eye? • Transduction: The conversion of signals from one type of energy into another. o Light energy physiological energy • Light transduction takes place in cells called photoreceptors, using molecules called opsins Opsins (a brief glimpse into molecular biology) • Proteins in outer layer of cells (membrane) • Attach to a vitaminA derivative o Carrots have vitamin A • Change shape when photons absorbed o change of events … physiological signals • Evolved more than once: o Type 1 : very old; found in microscopic organisms; normally fast acting (when directly linked to ion channels or pumps) o Type 2 : found in animals, normally slower acting (when indirectly linked to ion channels) • Many families w different specializations
• • • • •
Why opsins are trendy: optogenetics You use a virus to inject the DNA for certan types of opsins Can inject them in a way that the opsins only get expressed in particular types of cells Then you can put a laser or LED light to trigger those neurons or inhibit them So now we can precisely (cell type, location, temporally) stimulate or inactivate neurons o Very powerful
• • • •
2 main types of photoreceptors: rods & cones Outer segment: opsins found Cones: Photopsin Rods: Rhodopsin
Photoreceptor cells: rods vs. cones • Cones are less sensitive to light than rods o Rods become relatively more important when the light is very low • Cones are more dense at the fovea (centre of retina), less dense in the surrounding retina o Rods provide more peripheral vision info • Cones & rods contain dif opsins (photopsins vs. rhodopsin) o Several types of photopsins Dif types of cones transmit info about dif wavelengths of light Start of colour perception o Color info is transmitted by the set of cones & rods that are stimulated 3 types of cones
Photoreceptor cells
• • • •
S: detect 400450nm wavelengths Rods btwn S & M M: green L: longer wavelengths
Acquiring info from the eyes • “eyespots” 570500 mya o Detect if light present or not • Cuplike eyes o directionality • pinhole eyes o light enters from specific directions o more directionality o get a sense of an image if have enough recs • Complete closure o Some species • In some species: duplication of eye (compound eyes) • Addition of lens to focus light o Human eyes o So don’t need precise pinhole for directionality – the lens helps us focus
• •
As you move, pay attention to the black dot (attending to something that you aren’t looking at is called “covert attention”) At some point (when your eyes are ~3 x the distance between the cross and dot) the black dot will disappear. o It is in your blind spot—your visual system is “filling in” the area w background
The retina
Biological Psychology 6e, Figure 10.10 • Light passes thru vitreous humour ganglion cells bipolar cells photoreceptor cells • So we get less photons coming in because it has to go thru other cell layers o Less blocked in fovea, more direct
The human eye: anatomy
• • • •
Fovea: centre of retina, where world converge Optic disc: axons of retina ganglion cells converge travel to other areas of brain, beginning of optic nerve o Blind spot Optic nerve: bundle of axons carrying info to other brain regions Retina: inner surface of eye where light is projected onto
1) Close your left eye 2) Stare straight at the cross with your right eye 3) Move your head closer to the image
Cajal (1911) – detailed drawings of brain circuits from microscope; theorizing purposes • Photorecs, Horizontal Cells photorecs, bipolars • Bipolar cells (Layer 2) • Amacrine cells o bipolar cells, ganglion cells • Ganglion cells (Layer 3): outputting to brain Gollisch & Meister, Neuron, 2010 Sequence of events in the retina – the circuit of the retina
•
Light always inhibits photoreceptors
o
•
Lack of signal activates Bipolar ON cells (light on)
Light off photoreceptors constantly fire
o
•
•
A neuron’s receptive field: the variable space the neuron codes for o Can literally mean retinal space o In other areas of brain, real coordinates Ex. Hippocampus: fire when body in particular area of space A neuron’s firing rate: the # of APs per unit of time
Centersurround codes Activate Bipolar OFF • Can have CenterONSurroundOFF or CenterOFFSurroundON • Below: CenterONSurroundOFF cells o Center = excitatory, outside = inhibitory
Photoreceptors (rods & cones) • Input : o light (inhibitory) o horizontal cells (inhibitory) • Output : o Bipolar cells (excitatory or inhibitory) o Horizontal cells (excitatory) Bipolar cells (ON & OFF) • Recs determine if on or off • Input : o either cones or rods – not both o excitatory (OFF) or inhibitory (ON) • Output : o amacrine cells interneurons inhibit bipolar & ganglion o or ganglion cells (excitatory) Ganglion cells • (midget, parasol, bistratified, ip, etc.) • Input : o bipolar cells o amacrine cells • Outputs : o multiple brain nuclei
•
•
Neither photoreceptors or bipolar cells have APs o Signal continuously in graded fashion o More light – more changes Ganglion, amacrine – have APs
• •
•
Photoreceptors & lateral geniculate nucleus have receptive fields If show small spot of light in visual space o By chance, 1 AP appears after
Light in center of receptive field o APs in reaction
Info in the vertebrate NS is transmitted via signals btwn cells. • Neurons transmit signals when they experience an electrical event known as an action potential (AP) • When the rates or patterns of APs changes in relationship to a particular variable (ex. a sensory stimulus in a region of space), the neuron can be said to code for or represent that variable. Neuron coding • A neuron codes for (or represents) a variable if changes in its signaling, typically measured in the form of APs, are correlated w • that variable
Increase size of light
o
More APs
• • • •
Present light to just one photoreceptor (Ommatidia – compound eye) neuron fires If present light to surrounding areas decrease in activity Discovered before NS inhibition was discovered Fn’al inhibition
What is the fn (consequences) of centersurround fields? • (I.e., what is the consequent circuit bhvr/computation) • Increases resolution of spatial info!
•
Increase size of light more o Less APs than before because some light in surround
•
Light in surrounding area o No APs
https://www.youtube.com/watch?v=9qg9nBjUTc • Hubel & Wevil • Using electrical methods to probe visual sysem • Mapping receptive field on neurons in Lateral Geniculate Nucleus o Electrode in cat’s LGN o High freq pops, amplified
• • • • • •
Blue circles = photoreceptors If all photorecs activated excite all neurons o (no little dots) If put little dots in act laterally sharpen the receptive field via inhibition o Via Lateral Inhibition Top: Photoreceptors Middle: Horizontal Bottom: Bipolar cells
How might centersurround receptive fields explain the Herman Grid What is the mechanism of the centersurround receptive field? Lateral Illusion? inhibition • Not detecting as much light in the intersections because we have center surround codes
•
If rec field at intersection: too much inhibition from white areas in surround see black dots
o •
Btwn black squares: inside white, outside black
o • Haldan Keffer Hartline (~1958) • Studied eye of horseshoe crab – many eyes • Real eyes are lateral eyes – have dif photoreceptors
The outer inhibition overcomes the inner excitation
So no conflict, see white lines
Ultimately, our perceptions are computations
Summary part 1: the eye acquires light information.
• • • •
It can identify where light is coming from due to its design (image passes through a small hole & lens to project onto the 2D retina surface) It can identify which wavelengths are arriving (color) due to its diversity of photoreceptors It can sharpen the spatial resolution using lateral inhibition (Not discussed: it can also adjust to lighting conditions, due to photoreceptor “bleaching,” and inhibition by horizontal cells) o If exposed to light, less sensitive to light immediately after; also due to inhibition by horizontal cells, info from brain (ex. pupil dilation)
What computations take place in the retinal circuit? • The retina has over 50 types of neurons! • >20 amacrine cells o In general inhibitory but much variation • 1020 dif types of ganglion cells carry the output of dif types of circuits. o Specialization of cell types for dif circuit fns • Each dif circuit may extract unique info (features) that get transmitted in parallel to the brain.
Unanswered Qs from part 1: 1) 1) All of this can be accomplished at the level of photoreceptors & horizontal cells. What purpose do bipolar, amacrine, and ganglion cells serve? 2) What are the outputs (ie. the bhvrs, or computations) of the retina?
Part 2: neuron computation, computations of retinal circuits • How do neurons compute? • Each neuron is a processor. o It integrates signals (thru chemical & electrical o
• • • •
What kinds of features are extracted?
transmission) If those signals reach a threshold, it generates a new signal that gets transmitted to a new set of neurons.
Synapses from neurons A & B to neuron C (black) If A or B are sufficient on their own to fire: o OR. operator o Info generalized If A & B have only one connection each: o Have an AND operator Lateral inhibition: A & NOT.B fire o Need more specific conditions to become active
Actually more complicated: • Neurons have dif cellular properties that determine their bhvr. • Certain chemicals (neuromodulators) dynamically change how neurons respond to signals. • Circuits often show “emergent properties” that aren’t obvious from this description
1) Color • Redgreen opponency is transmitted by some “Midget” • • •
ganglion cells Some have Centersurround fields, created by horizontal cells o Midget bipolar cells further transmit their own center surround fields Blueyellow opponency is transmitted by small bistratified ganglion cell Opponency: seeing one colour inhibits seeing the opponent colour
2) Directional motion
• • •
Burst of act pots as light moves across receptive field Raster plot: More spikes when you move according to the preferred direction (up) No spikes when you move down (opposite to preferred direction)
Ganglion cells fire before the moving dot gets to the end of the screen
Other fns: • Overcoming noise (from movement & spontaneous signaling) o Noise: anything that isn’t signal, isn’t useful o Ex. eye muscle twitching = noise, spontaneously signal when no stimulation (like if light on signals, even tho should be inhibited) • Inhibition output during saccades o Ex. during high speed • Contrast adaptation o Ex. dif light conditions, bleaching of photoreceptor during light, inhibition via amacrine cells o Sharp contrasts on sunny days o Eyes accommodate a bit in fog • Adaptation to patterns o Ex. envt w many horizontal lines – your retina will adjust a bit relative to an envt w all verticals o Higherorder computation, impressive Some ganglion cells are intrinsically photosensitive (ip) • use melanopsin o Activated by light • widely branching dendrites o Covering as much of retina as possible to detect how much light is coming in • project to suprachiasmatic nucleus in hypothal (controlling circadian rhythms) •
• • •
Starburst amacrine neurons inhibit directionselective ganglion cells. Internal, cellular processes in the dendrites of starburst cells may make them “intrinsically” directionselective The starburst cells also appear to inhibit ganglion neurons in a spatiallyselective way, amplifying the direction selectivity. o Ex. Dendrites on the side of the ganglion cell connected to starburst amacrine are inhibited from signals by a bipolar, excited only if stimulated on other side on another bipolar (where no amacrine connection to ganglion dendrites)
What kind of features are extracted? 1) Color 2) Directional motion 3) Texture 4) Approaching motion o If light getting bigger, certain ganglion cells will pick up on this 5) Anticipated position o Evolutionarily advantageous to be processed early in retina o Critical because we convert light into physiological signals slowly via Type 2 Opsins (act indirectly within cell) o Computer screen: dot moving across screen, then stop
Temporal Coding: Rate vs. timing of action potentials: • Take longer for Bipolar cells to be inhibited by photoreceptors, so takes longer for ON Bipolar cells to be activated o So cells that respond to darkness are activated faster • Ganglion cells may signal lighter vs. darker image regions by precisely when they respond. o By knowing how quickly ganglion cells are activated in dif areas, we know what areas are darker • Take about 100ms to detect what’s going on in a visual scene o Not much time for physiological changes o So this is much quicker • The temporal code doesn’t require signal integration, so quicker o Over time you can integrate more to get a clearer image
(Image from Gollisch & Meister, Neuron 2012)
•
This is often called a temporal code & can be contrasted from rate codes (# of APs per unit time)
Summary of part 2: retinal circuits do compute • Although there is 1 “type” of retinal circuit (5 basic cell types), • LGN : nucleus of thal where most info from retina goes there are many variations of the circuit that use dif combos of cell o primary visual cortex types o Not just simple transducing The lateral geniculate nucleus is madeup of 2 magnocellular, 4 parvocellular, & 6 koniocellular layers • Each type extracts dif types of info across each area of retina • Thus, computations take place in the retina thru the stepby step (serial) processing of many sidebyside (parallel) • Computations in the system are performed thru serial & parallel processing Important concept list 1) Both topdown & bottomup processes contribute to visual perception 2) Computations in the visual system (and NS in general) take place thru serial & parallel processing.
•
•
• • •
Feedforward serial processing: the signal is stepping from one processing step to the next Feedback: a signal is acting back on an early processing step (Only makes sense if there’s a feedforward)
We still don’t know precisely how the circuits are connected & how they fn. But we may learn more from gamers playing Eyewire (MIT). Eyewire allows you, as a game player, to help identify the wiring, and ultimately fn, of retinal circuits
• • •
2 Magnocellular layers: o Large neurons 4 Parvocellular layers: o Smaller neurons 6 Koniocellular layers o Btwn M&P layers
Ganglion cell (info)
LGN layer
•
Midget (redgreen)
parvocellular
•
Parasol (noncolor)
magnocellular
•
Bistratified cells (blue)
koniocellular •
Specifically KIII & KIV, & koniocellular cells scattered in the other layers)
Part 3: The lateral geniculate nucleus as a bridge to cortex Neurons in the LGN receive more synapses from the cortex than from the retina! • Primary relay centre for bottomup info • Gets more topdown info than bottomup
o
What the cortex receives is already very biased by what the cortex has
