Download Principles of Sensory Coding

Principles of Sensory Coding
Sensory systems appear to be very diverse. Yet they all solve the same task:
they convert environmental signals into neural activity that can influence
the motor system of the animal.
The plan for this section of the course is to first give an overview of what
sensory systems do in general and then to focus on the operation of the
classic senses of vertebrates.
Methods for studying sensory processing:
1. Psychophysics- use behavioral testing to establish the sensitivity of a
sensory system and the “rules” of its operation.
2. Electrophysiological recording from the single neurons or small groups
of neurons along the sensory pathway to find out how the neural
circuitry gives rise to the perceptual abilities.
3. Imaging in humans that are doing perceptual tasks to identify the brain
areas responsible.
What do Sensory Systems Sense?
There are two main functions of any sensory system:
1. The detection of a signal. Weak signals can be detected without the
animal being able to finely discriminate any of its features.
2. Discrimination of some aspects of a sensory input. This is often
referred to as estimation.
What must be estimated from the input:
1. Qualitative features such as colour or odorant; this is
often referred to as modality- what is it?
2. Quantitative features such as magnitude- often referred
to as the intensity of a stimulus;
3. Temporal features such as duration or frequency of a
signal;
4. Spatial location of a stimulus- where is it?
Typically, all these aspects are estimated at once.
A common strategy of sensory systems is to have
separate neural pathways specialized for estimating
different types of stimulus features. For example, the
visual system analyzes colour, shape and movement
in different brain regions.
How are these attributes represented in the brain?
Modality: the most basic mechanism for identifying the nature of a
sensory input is via labeled lines. What this means is that input from the
optic nerve is always interpreted by the brain as visual input etc. This
extends to much finer discriminations: the connections of “pain” and
“touch” fibers in the somatosensory system are entirely different and
electrical stimulation of either leads to the appropriate sensation.
Intensity: the estimated intensity of a stimulus is not a linear function of
the actual intensity. As shown in the graph below, the relation can be
described as logarithmic or power law. The reason is intuitively easy to
understand. Increases of a weak signal generate a larger perceived
increase than increases of a strong signal- the percept saturates.
A power law relation best describes the
relation between stimulus strength and
perceived stimulus intensity.
Kandel, Schwartz and Jessel
Temporal features:
Onset time andDuration. The duration of a stimulus is estimated from the
onset of the neural response and its duration. Typically there are neurons
in sensory systems that only respond to the onset of a stimulus- these are
generally referred to as phasic responders and they are good for
estimating the time of occurrence of a signal. There are other neurons that
respond throughout the stimulus presentation- tonic responders- these
signal stimulus duration.
Frequency. The frequency of a signal may be very important is some
senses (audition). There are several ways to estimate this; one way is to
make the neural response precisely time locked to the signal.
Kandel et al.
Bear et al.
Location. A common principle is used for estimating where a stimulus is
located- topographic mapping. This means that points close together
on the sensory surface are represented close together in the brain.
In the somatosensory surface this is called “somatotopy”, in the visual
system “retinotopy” and in the auditory system “tonotopy”.
Skin regions that are close together are mapped to
adjacent regions of the cerebral cortex.
Bear et al.
Getting Sense Input into the CNS
Sensory input comes in many flavours. Information in the CNS all
comes in the same currency- action potentials or spikes.
The reason that the CNS uses only one way to transmit information is
simple: it allows integration of different types sensory input and
the connection of sensory input to motor output- all the neurons
dealing with these different systems use the same “language” of
spikes.
The problem becomes: how to translate the different kinds of sensory
input into spikes. In all cases this is done by specialized receptor
cells (or parts of cells) in a process called sensory transduction.
The transduction process depends on the nature of the signal.
1. Chemoreception.
2. Mechanoreception.
3. Vibration reception.
4. Light.
The initial transduction process causes the receptor cell to depolarize
and this leads to spike initiation in sensory afferent fibers that then
convey this information to the brain.
Sensory Transduction
Stretch gated channels (in the nerve
terminal of an afferent fiber) open and
cause it to discharge. The fiber projects to
the CNS.
A chemical binds to the receptors on an afferent
process, causes depolarization and discharge.
Kandel et al.
A hair cell is activated by vibration, becomes
depolarized and releases transmitter. The
transmitter excites an afferent fiber; it
discharges and this information reaches the
CNS.
A photoreceptor is activated by light; its release of transmitter is reduced. After a series of
complex interactions in the retina, ganglion cells discharge and send this information to
the brain.
The most important point is that the message sent by a receptor to the brain is in
the form of a sequence of spikes- a spike train. What is this message?
Coding of sensory input
All sensory input is represented in the firing patterns of populations of neurons.
This is the most remarkable conclusion of sensory physiology. The
representation is referred to as a population neural code.
What is this code?
The dominant theory since the Lord Adrian’s work in the early 20th century is
that information is carried in a rate code- often a linear rate code. The intensity
of a sensory stimulus is correlated with the firing rate (spikes/second) of a
sensory afferent or neuron. This theory is well supported by data from some
mechanoreceptors such as stretch receptors.
In this experiment the firing rate of a neuron
was recorded (over 1 second periods) while
delivering stimuli with varying intensities. As
you can see, the discharge rate varies in a
linear fashion with intensity.
This type of plot is often referred to as the
tuning curve of a neuron.
Kandel et al.
Other Possible Neural Codes 1
There are a number of problems with rate based codes. The most serious
difficulty is that they are slow. Imagine a neuron discharging at 20 spikes/s
in a fairly random (Poisson-like) manner (this is fairly typical for many
cortical neurons). If it is activated it may increase its rate to 50 spikes/s; its
target cells will have to wait for at least 1/2 second to figure out (reliably
decode) that there has been a significant increase in firing rate. Our ability
to detect novel input is far faster than this suggesting that other forms of
coding must exist.
I will list some candidate codes below in order of how firmly they are
established.
An important point is identifying a neural code requires more than
merely showing that the information is present in a neurons
discharge pattern. It must also be shown that the information is
actually used by downstream neural circuitry- that it is decoded. In
general this has proven to be very difficult to accomplish.
Other Possible Neural Codes 2
Coding with time: As you’ve already seen, neurons can become
phase-locked to a stimulus. In this case, it is the time of occurrence of
a spike that is the signal. This mechanism is well established in the
auditory system.
It is also possible that the time of occurrence of the first spike response
to a stimulus carries most of the information.
Coding with spike patterns: Many neurons (in thalamus, cortex and
elsewhere) will, in response to sensory stimulation, produce bursts of
spikes as well as trains of isolated spikes. It has been strongly argued,
though not proven, that spike bursts are a separate coding mechanism.
Coding with correlated activity in a neuronal population: It has
been suggested that synchronized or correlated spiking activity is an
important code. This type of code would permit a huge increase in
coding capacity, but it has been difficult to conclusively demonstrate its
occurrence.
Coding with Bursts of Spikes:
Salient Sensory Input
Spike bursts are brief, high frequency sequences of spikes. They are generated by different
biophysical mechanisms including low threshold Ca2+ channels (thalamus) and interactions of
a neuron’s soma and dendrites (pyramidal cells).
Despite this, they may serve a common purpose- detection of important, brief and
unexpected events.
Here are two examples. In the retina reversing the direction of an object’s motion- an
important and unexpected event, causes a synchronous spike burst of many retinal
ganglion cells.
In thalamus, neurons receiving input from the whisker afferents, burst when the rodent
twitches its whiskers- a form of exploratory behaviour.
Retina- motion reversal
Schwartz, 2007
Thalamus- whisker twitching
Fanselow, 2001
Coding with Bursts of Spikes:
Reward Associated Signals
Lin, 2008
Recordings from cells in basal forebrain; these cells
project to cortex and might be related to learning.
These cells respond to emotionally salient signals from
any modality- emotionally salient means that the signals
are associated with reward or punishment.
When a neutral tone is presented the cells do not
respond.
What the figure shows is that, when the same tone has
been paired with reward, it initiates a strong burst
response (Hit)- but only when the rodent actually goes
for the reward- there is no response to the tone if the
animal does not go for the reward (Miss).
Receptive Fields
I have been emphasizing neural coding as temporal patterns. Sensory
physiologists must also take into account the spatial dimensions of
neural coding.
This is typically done by estimating a neuron’s receptive field. The RF
is defined as that region of sensory space whose stimulation results in
a change in discharge (usually firing rate) of the neuron
Imagine that a patch of skin is represented at the
left. A given receptor will be activated when the
stimulus is presented in that region of skin where
its terminal branches are found. This is that
receptor’s RF.
Many of these receptors will converge onto a
neuron in the spinal cord or brainstem. This
neuron will be excited when the skin is touched
over a much larger region- it has a larger RF.
Adding inhibitory interneurons into the circuit can
reduce the size of the RF or complicate its
response.
Kandel et al.
The Overall Plan of Sensory Systems
For all sensory systems there is a common plan.
Peripheral receptors respond to a specific stimulus and convert it (directly or
indirectly) into a spike train. The afferent fibers end in lower brain regions where they
are processed. Quite often there are many parallel pathways present. When rapid
responses are required, the processed information might go directly to a motor
system. However, for more detailed analysis, the information proceeds to higher brain
levels for further processing. In mammals and birds, the sensory input reaches the
forebrain where it somehow results in the perception of complex patterns.
We will now look in more detail at the classic senses.
The ascending pathways of the
somatosensory system.
Kandel et al.
Parallel Pathways for General and Communication Signals
A very general principle of sensory coding is that communication signals have
their own separate channels.
An animal might encounter very different environments depending on where it
is born; for example, a city rat and a country rat will likely encounter very
different odors. So sensory systems need to adapt themselves to the
experiences of different animals.
In contrast, communication signals evolve over evolutionary time and are
highly conserved for each species. For example, many animals (including
rats) use pheremones to communicate gender etc. These are fixed and so the
olfactory system does not have to “learn” about different pheromones; their
detection can be built in.
This distinction is also evident in the auditory system: speech and song versus
environmental sounds.
It is also obvious in the visual system: expressions on faces versus trees etc.
The obvious differences in the nature of these signals is carried forward in the
nervous system and we’ll encounter this repeatedly in this course.
The Somatosensory System
The somatosensory system is subdivided into 3 subsystems:
Exteroreceptors. These are the familiar receptors found in the skin that mediate
the sub-modalities of touch, pain and temperature. These types of sensory input
can mediate both rapid responses (e.g. reflexes) and reach the cerebral cortex and
induce perception.
Proprioceptors. Proprioceptor afferents are found in muscles and at joints; they
mediate the detection of muscle stretch and the degree of extension at a joint. For
the most part ,this type of sensory input is involved in motor control rather than
perception. The two major classes of proprioceptive input come from (a) Golgi
tendon organs and (b) muscle spindles. The latter are highly specialized muscles
that act to detect the amount of stretch of the muscle. We will not discuss these
afferents any further in this section of the course.
Interoreceptors. Interoreceptors are found in internal organs and convey signals
that include distension of the stomach, carbon dioxide concentration in the blood
etc. This type of sensory input is obviously of great importance but difficult to study
experimentally (in the CNS). We will not discuss this submodality any further.
Exteroreceptors 1
The receptors that transduce light or sound are highly specialized cells; in contrast,
exteroreceptors are merely nerve endings in the skin. The cells that give rise to these
nerve endings are found in ganglia just external to the spinal cord (spinal ganglia) or
brainstem (trigeminal ganglion). Ganglion cells are bipolar: one of its axons extends to
the skin while the other enters the spinal cord. Each spinal ganglion innervates a strip of
skin known as a dermatome.
Bear et al.
Bear et al.
Bear et al.
Exteroreceptors 2
There are many forms of exteroreceptors. The most important distinction is between
touch or mechano-receptors and pain and temperature receptors.
The nerve endings of pain and temperature receptors are simple and the axons are
either unmyelinated (C type) or lightly myelinated.
In the case of touch fibers the nerve endings are usually associated with some
specialized structures such as hairs. The axons of touch fibers are heavily
myelinated allowing them to conduct action potentials rapidly.
The response of touch fibers depends on their associated structures.
Bear et al.
Response Properties of Fine Touch Fibers
The structure of the accessory
tissue of the nerve ending
determines its response.
High frequency Pacinian
receptors are good for
detecting texture. Ruffini
endings estimate the duration
of contact etc.
Bear et al.
Central Projections of Exteroreceptors
The central projections of somatosensory afferents allows this system to be
divided into two functionally distinct streams: (a) a phylogenetically older
spinothalamic system and (b) a more recently evolved (especially in
mammals) lemniscal pathway.
Kandel et al.
The Spinothalamic System
The spinothalamic system is a phylogenetically old system that conveys sensory input
related to pain, temperature and touch to the brainstem as well as cortex.
Unmyelinated fibers (C type) carry pain input
to the spinal cord. These afferents use
Substance P (SP) as a co-transmitter with
glutamate. When activated, these fibers
release SP both peripherally and in the spinal
cord. SP is very potent is activating spinal
cord neurons that transmit pain information.
There are also lightly myelinated fibers that
do not have SP and convey transient pain
signals.
Bear et al.
The Spinothalamic System 2
Bear et al.
Note that ST axons cross in the
spinal cord before they ascend.
So that pain and temperature
are represented on the
contralateral cortex.
Spinothalamic fibers terminate densely within the spinal cord itself. This is important
for reflex functions such as the withdrawal reflex. The ST fibers also end in the
brainstem; again this is important for both low level motor control and for activating
the reticular system- this causes heightened attention to potentially dangerous input.
Finally, the ST axons reach the thalamus. They reach both non-specific thalamic
regions that activate large parts of association cortex and the thalamic nuclei
specifically involved with perception of touch.
The Dorsal Column, Medial Lemniscal Fine Touch System
Bear et al.
The dorsal column ML system contains only myelinated axons conveying fine touch
information. The central axons give off collaterals in spinal cord with the main axon
ascending via the dorsal columns to the medulla where they terminate in the dorsal
column nuclei: n. gracils (trunk and legs) and n. cuneatus (mostly upper limb).
The Dorsal Column, Medial Lemniscal Fine Touch System 2
Ventroposterior nucleus of
the thalamus.
Note that lemnsical fibers
cross in the medulla so that
touch is represented in the
contralateral cortex.
Bear et al.
The Dorsal Column, Medial Lemniscal Fine Touch System 3
Touch for the face is conveyed by a
functionally similar system. The
afferent fibers come from the
trigeminal nerve and enter in the
pons. Pain and temperature fibers
from the face then descend to the
descending trigeminal nucleus. The
fibers that carry touch information
terminate in the principal trigeminal
nucleus. From this point on, the
touch afferents from trunk and face
merge.
Bear et al.
The Dorsal Column, Medial Lemniscal Fine Touch System 4
Representation of touch in cerebral cortex.
Bear et al.
The Dorsal Column, Medial Lemniscal Fine Touch System 5
The columnar organization of cortex (Mountcastle).
A column of S1 has neurons all responsive to the
same type of stimulus from the same region of the
body.
This appears to be a major principle for the
organization of at least sensory and motor cortices.
Bear et al.
The Rodent Vibrissae System:
A Model System for the Study of Active Touch
Ahissar, 2008
Figure 3. Cortical representation of the five whiskers of interest (experiment R20). In this
color scheme, white corresponds to no activity above the spontaneous level. Erchova, 2004
Rodents have prominent “whiskers” or vibrissae. These represent a dominant touch sense for these
animals that they use to localize and identify objects (walls, insects, etc) in their environment even in total
darkness.
There are a constant number of vibrissae in every rat; they are heavily innervated by trigeminal afferents.
Each whisker has muscles (motor trigeminus) at its base and it uses them to “whisk” – therefore this is an
active sense in which the rodent actively generates movements that result in sensory stimulation.
The whisker afferents reach cortex in a standard manner (via thalamus) and terminate in Layer 4. But each
whisker input terminates in the center of a discrete ring of cells and the entire column of cells is devoted to
that whisker- so this part of the rodent cortex is called “barrel cortex”.
Because it is very convenient to stimulate whiskers and record from barrel cortex, the vibrissae system has
become the ideal system to study somatosensory processing and the effects of peripheral damage
(removing a whisker) on cortical processing.
The Vibrissae Sense can Encode Texture
Arabzadeh, 2005
Sandpaper
Fine
Coarse
Vertical
Whisker Velocity
Horizontal
Peri-Stimulus Time Histogram
PSTH
Whisker Receptor Afferent
Barrel Cortex Pyramidal Cell
Peripheral and Cortical neuron respond with different spike patterns to different sandpaper
textures. This is presumably the basis of a rodent’s (and ours) ability to discriminate texture.
The Vibrissae Sensory System uses many encoding Schemes
Ahissar, 2008
The identity of the whisker
activated encodes the vertical
location of an object.
The Horizontal location is encoded by
timing of the spiking response of
whisker afferent cells.
The radial location is encoded by the
firing rate of the spiking response.
Horizontal Location
Vertical Location
The Response of Trigeminal neurons to Active Touch Differs from their Response to Passive Touch