When we first think about sensory perception,
it seems trivial.
We look at the world around us, and we see objects.
We listen and we hear sounds.
We sniff, and we smell fragrances.
And it all fits together in one unified world view:
our perceived reality.
However, when we start thinking about
the underlying neuronal mechanisms,
it turns out that we're only just beginning to get our first clues
as to how sensory perception works.
Sensory perception is an active process in at least two different ways.
Firstly, motor control is very important for sensory perception.
We actively gather sensory information.
And that's what we discussed in the last video.
We make eye movements and head movements
to see specific parts of the visual field.
And we move our hands and fingers to touch objects,
actively get tactile sensory information.
It turns out that sensory perception
is active in another way of thinking also,
in which our neuronal activity actively generates our sensory percepts.
That sensory percepts are subjective,
can easily be seen by looking at images like this
that evoke bistable percepts.
We have one set of photons that fall upon the retina,
but if you stare at this image for long enough,
you'll see two alternating percepts.
Either you see the two white faces looking at each other,
with their noses almost touching,
or you see the black cup in the middle.
And it's one set of light that falls on the retina,
but the brain generates two distinct percepts from it.
Clearly, sensory percepts are internal constructs
that are generated by neuronal activity.
And right now, neuroscientists think
that it's largely the activity of neurons in the neocortex
that generate our sensory percepts.
And so for that causal and mechanistic insight
into how sensory percepts occur,
we need to look at the activity of individual nerve cells
and how they talk to each other,
perhaps with a particular focus upon what's going on in the neocortex.
So sensory percepts are not just out there in the world,
but they're actively generated by the internal activity of our brain.
Sensory percepts are also learned through experience.
And it's because we've seen many silhouettes before,
that we recognize this as a silhouette of a human.
And so the reason we recognize objects
is because we've seen many closely related images before,
images of faces and people from many different angles.
And our brain has managed to extract the appropriate
statistics that associate that information.
And so, as we are all aware of from watching young babies,
we don't immediately know how to perceive the world,
but we learn to see the world around us.
And the same is true of course, for all of our other sensory inputs.
So how are we going to investigate subjective, learned, sensory percepts?
Well, we clearly need to be able to deliver
well controlled sensory stiumuli.
That's then going to evoke neuronal computations inside the brain.
And those are going to depend upon context, learning,
and also the goals of the particular moment.
And if we're going to experimentally investigate the subjective percept,
then we also need the subject to report what it is that it's feeling.
And so we need to have a motor output
where behavior reports the subject to percept.
And so if this were a human psychophysics experiment,
we would have two buttons.
One that the human observer would push if he sees two faces,
and another button if he sees the black cup.
And he would then be pushing in alternation
these two buttons
as a percept flickered from one to the other.
And so, an important concept in terms of experimental investigation,
is that sensory percepts must be reported through motor ouput.
And we might consider this, all together, as a learned, goal-directed,
sensory motor transformation.
And that then, might represent the minimal, essential core feature
of sensory perception,
that it has to form a learned,
perhaps relatively abstract sensory motor transform.
Now, sensory perception is rather complicated,
and it might be a good idea to start with very simple forms
of sensory perception,
if we want to get to the detailed, mechanistic and causal insight
into how it actually works.
The simplest form of sensory perception is detection.
Simply to say, is there a stimulus or not?
And that's what we're going to investigate here,
in the whisker sensory system.
We implant metal head holders onto the head of mice,
so we can head-restrain them,
make recording chambers that allows us to do
electro-geological and imaging experiments at high resolution.
The animal sits inside a cardboard roll
where it feels relatively comfortable,
and in addition, we have a plate here in front
that restrains the movement of its forepaw,
so it can't for example, scratch its nose.
Onto the whisker we place a small piece of metal.
And the whole animal site on top of a electromagnetic coil.
We can then drive current pulses through that electromagnetic coil
that generates magnetic fields,
and those magnetic fields, act upon the metal particle
and cause a force to be applied.
And we can make these magnetic pulses relatively short,
about one millisecond in duration,
and so a one millisecond impulse is applied to the whisker,
and that occurs at a random time.
The animal has no cue to tell us when that stimulus is coming.
So the animal is waiting, waiting, waiting,
and every now and then,
there's a brief one millisecond pulse on the whisker,
and what the animal has to learn in this behavior,
is that that one millisecond whisker stimulus
means that reward is available from this licking spout.
And if it licks immediately after the whisker stimulation,
then it gets a droplet of liquid reward.
And we monitor the tongue contacting the spout here
through so called piezo film,
that then gives us a little electrical signal
when the tongue touches the spout.
And we can threshold that, and use that to open a valve
and deliver liquid water.
And so for the trials in which we stimulate,
there are two different outcomes.
If the animal licks during the reward window,
which we set at one second,
then we open the valve and the animal gets a reward.
On other trials, we give the same stimulus
but the animal fails to lick.
And this is then a miss trial and of course,
there's no reward that's delivered.
And so as the first indicator, we can quantify performance
as a so called hit rate--
the fraction of trials where the animal licks,
compared to the total number of trials where we deliver a stimulus.
We can also look at another trial type where we don't stimulate,
and sometimes the animal will spontaneously lick.
We can think of those as false alarm trials.
And these might also be very interesting to analyze.
Perhaps the animal is dreaming of a whisker stimulus being delivered.
And something in the neuronal activity
then makes it lick.
And we can think of other trials where we don't stimulate--
the animal doesn't lick,
and those are then correct projection trials.
On the first day, when the animal goes into the behavioral apparatus,
the animal knows that licking is a good thing,
that it sometimes gets reward when it licks,
but it doesn't know when.
It has no idea that the whisker stimulation is a cue to initiate licking,
and that when there's no stimulation,
there's no reward that's going to happen.
And so it licks just as much on stimulation trials,
as on no stimulation trials.
Through trial and error and reward-based learning,
the animal learns that licking after a stimulus is a good thing,
and it begins to pick up something like 70 to 80 percent
of the trials in which we stimulate, and licks in response to that stimulus.
It also learns that when there's no stimulation,
there's no point in licking.
And so it learns to suppress the licking in the absence of a stimulus.
And it 's really the divergence of these two curves,
that tells us that the animal has learned the behavior.
We can also look at the psychophysics of this behavior
where we plot the stimulus strength against performance of the animal.
And as we turn the stimulus strength down,
the fractional trials in which the animal perceives a stimulus
and gives a motor output in licking, goes down.
And somewhere here, we're at about half way in performance,
and that's a so called psychophysical threshold.
As we decrease the stimulus even further,
there's no licking above the false alarm rate,
and so here the stimulus is undetectable.
And so this s a standard psychometric function
for any given behavior.
Now, we're interested in the neurobiology of this behavior.
What's the neuronal activity that causally drives
this learned, goal-directed sensory motor transformation?
We've already looked a bit at the sensory pathway
where whisker deflection causes depolarization of sensory nerve terminals,
and action potential firing down the trigeminal nerve,
information crosses the glutamatergic synapses in the brain stem,
and action potential takes information to the thalamus,
and other glutamatergic synapse then brings the thalamocortical
action potential to the neocortex
in the primary somatosensory cortex.
And then in the primary somatosensory cortex,
we have this exquisite one-to-one mapping
of the sensory periphery where each whisker
is individually represented in an anatomical unit,
a so called barrel.
And so if we put sensory information here on the C2 whisker,
we might then expect that if the neocortex is involved,
that the C2 barrel column might be a key area
for where this behavior might depend upon.
So we can test that by inactivating this part of the brain,
and see if that makes any impact upon behavioral performance.
So here we have the animal sitting in the behavioral apparatus,
it's performing the task,
it's picking up something like 80 percent of the stimuli,
and licking in response to them,
and at sometime we inject tetrodotoxin into the C2 barrel column
of the mouse barrel cortex.
And that causes a dramatic and rapid decrement in behavioral performance.
TTX is a block of voltage-gated sodium channels.
It prevents action potential firing.
And so here we've blocked action potential firing,
in the primary somatosensory cortex, and that's caused a deficit in behavior.
So apparently, action potential firing in the C2 barrel column,
is essential in order to perform this task.
We can inject other agents like CNQX and APV
to block ampa and NMDA receptors.
And we see that there's a similar deficit in performance.
And so apparently, glutamatergic synaptic transmission,
is also essential to carry out this behavior.
As control experiments, we can inject the ring of solution
that dissolves these agents and that has no impact.
Or we can inject the tetrodotoxin toxin into another brain area here,
into the primary somatosensory forepaw representation
about a millimeter or so away from the C2 barrel column.
And that also doesn't cause a deficit.
So there's some degree of specificity to these pharmacological inactivations.
And it seems that action potential firing and glutamatergic synaptic transmission
are necessary in order to perform the detection task.
We can then see what happens in the other direction.
Can we substitute the whisker stimulation
by directly stimulating this area of the brain?
And we can do that by expressing channelrhodopsin
through injection of viruses that expressed channelrhodopsin here,
coupled with a fluorescent protein,
so we can see where the area of the brain that's been injected.
Here you see it in post-mortem sections,
and there's a small localized area of the brain
that expresses channelrhodopsin,
and that of course corresponds to the C2 barrel column.
And here through mass genetics, we've expressed the channelrhodopsin
specifically in excitatory neurons.
After injection of the virus, it takes a few weeks
before the channelrhodopsin expresses.
And during that time, we can train the animal,
in the standard whisker learning tasks.
And so here the animal never sees blue light flashes
and we just train it exactly as before,
we stimulate the whisker, and the animal learns to lick in response.
Then, once the animal performs this task at a high rate,
then on a given transfer test day,
we ask if we now-- instead of stimulating the whisker,
we now make an optigenetics stimulation of cortex,
what happens?
Does the animal interpret that stimulus as the same as a whisker stimulation?
Does it perceive the same?
Does that generate the licking motor output?
And remarkably, in most of the animals in which we tested,
the very first, five millisecond blue light flash
drove licking behavior.
Here's the light flash.
Here's what we're seeing on the piezo trace--
the animal's tongue contacts the spout,
Here we open the valve and the animal is now
enjoying its reward.
This turned out to be a robust phenomenon,
where we could replace the sensory peripheral stimulus
of whisker with a light stimulation directed at the neocortex.
And so we can directly replace sensory stimuli at the periphery
by internal stimulation of neurons in the neocortex.
We can also examine the learning in the opposite direction.
We can train animals first, in the optigenetic detection task,
where we now deliver five millisecond blue light flashes
to stimulate excitatory neurons in the S1 cortex,
the animal learns to lick in response to the S1 stimulation,
and after the animals learns that,
we then give the very first whisker stimuli.
And we then ask, does the animal now
interpret that stimulus in the same way,
and transform that into a goal-directed, licking motor output?
And again, the answer is "yes".
We stimulate the whisker, the animal licks,
and the replacement from optigenetics stimulation
to a real peripheral stimulation is extremely good.
And so it seems we can optigenetically program the behavior from S1 cortex.
And so these data then tell us that activity in S1 cortex
is both necessary and sufficient in order to generate
the perception of a whisker stimulus that then gets transformed
presumably by downstream brain areas,
into a licking motor ouput in order to obtain a reward.
It's therefore clearly interesting to find out what types
of neuronal activity are taking place in the C2 barrel column
during task execution.
And here we're making membrane potential recordings
from layer two, three parameter neurons in the C2 barrel column.
We stimulate the whisker,
and this upper trace is from a hit trial,
the animal licks, it gets a reward,
and so by definition, this is a hit trial.
If we now look at the membrane potential trace,
shortly after stimulation we see there is a depolarizing sensory response
that is presumably the sensory response driven by
the feet forward thalamocortical circuits
then at a later time, there's a late depolarization here,
with a couple of action potentials riding on top of it,
and that late depolarization, nonetheless precedes licking
by hundreds of milliseconds.
And so there could be a causal relationship
between late depolarization and driving the licking motor output.
Down below we see a miss trial.
We give the same identical stimulus,
it drives a depolarizing sensory response,
but the late component here seems to be less obvious,
and of course, there's no licking and no reward.
It's a miss trial.
These are just two individual trials.
But they turn out to be representative of the average trials
for this particular recording.
And so here we take all the hit trials together
and average them into the black trace.
We take all the miss trials, average them together into the red trace,
and here at time and stimulus, you see that both hit and miss trials
generate large, obvious sensory responses,
and there's no obvious difference between the hit and the miss trial.
So there's a early, reliable sensory response
And that's probably a good thing when primary somatosensory cortex
and so you'd expect there to be a good, reliable representation
of the periphery.
At later times these two traces diverge,
with depolarization being more prominent on hit trials than miss trials.
And that's accompanied by enhanced action potential firing
on hit trials compared to miss trials,
and all of that precedes the licking by some hundredths of milliseconds.
So here in this late component of activity
we have something that correlates with the subject of percept of the animal,
as reported by licking.
We think that late depolarization, correlating with subjective percept,
is interesting, because similar results have been found in humans.
And so here in the work of Stanislas Dehaene
and coworkers,
he has made measurements of the human EEG signal
across early visual cortex, and have given visual stimuli
that are just at around the threshold for conscience perception.
And so some trials are perceived,
and that's indicated by the black trace,
and other trials are not perceived,
and that's represented as a red trace.
And what Stanislas Dehaene and his colleagues suggest,
is that there's an early, reliable, sensory response in the visual cortex,
and that there are late changes to activity in the human brain
that correlate with consciously perceived, or not perceived, stimulii.
And so it looks like similar might be happening
in the human brain, compared to what
we've seen in the mouse.
Now, what's difficult in the human situation,
is to get a causal perspective on what this late activity might be doing.
And that's something that we can do in the mouse system.
And so what we need to do then, is specifically inactivate
late periods of activtity.
And we can do that by using channelrhodopsin optigenetics,
and stimulating the GABAergic inhibitory neurons.
And so here we take the parvalbumin
expressing GABAergic neurons that you might remember
are the ones that inhibit perasematically.
We put channelrhodopsin specifically inside those neurons
using mouse genetics,
we can then get blue light flashes
and then we can stimulate action potential firing here
in the parvalbumin expressing GABAergic neuron.
That's going to then release GABA, hyperpolarize the post synaptic cells,
the excitatory parameter neurons and then we'll then get
a hyperpolarization in the membrane potential
of these excitatory neurons.
And so we'll be able to shut down primary somatosensory cortex
by shining blue light on GABAergic neurons
expressing channelrhodopsin.
So let's have a look at that in the situation of our experiments.
Here's our C2 whisker deflection,
here's our sensory response,
the black curve is a normal control condition,
and if we now turn the blue light on,
immediately around the time of stimulation,
we get this light blue curve here,
we get rid of the early sensory response
So now we shut down S1, just at the time
where sensory processing of the stimulus occurs,
and when we look at the behavioral performance,
that causes a deficit in the detection of the sensory stimulus.
And so that's really very similar to what we found
with the pharmacological inactivation experiments
where we injected TTX or [sinqx] and APV into S1 cortex,
and found that caused a deficit in detection performance.
Here we see the same thing
except now, we've done the inactivation
on this time scale of 50 milleseconds
and so we know that it's a neuronal activity
immediately surrounding whisker stimulus
that drives sensory perception.
The question that we were really interested in,
is what happens if we leave that early sensory response in tact,
and we only hyperpolarize at late times.
And that's what you see in this purple trace.
We turn the blue light on,
we stimulate the channelrhodopsin,
after a hundred milliseconds that causes hyperpolarization
at late times, leaving the early sensory response in tact.
And the question now is, if we just get rid of
the late depolarization
does that make any difference in behavior?
And indeed, it does.
It causes a significant deficit in detection.
And so that late depolarization not only correlates
with the subjective percept of the animal,
but it also contributes to driving that sensory percept.
And so we're slowly beginning to understand mechanistically,
the types of neuronal activity
that occur in the mouse brain during subjective sensory percepts.
In this video we've seen that head-restrained mice
can be trained to perform simple, perceptual tasks.
And over the coming years, we'll get a detailed and mechanistic
understanding of how simple, sensory motor loops are performed.
In the case of the detection task,
we've seen that there appear to be two phases to the sensory processing.
An early, reliable response, and a late depolarizatoin
that seems to correlate with the subjective percept.
In the years to come, it will be important to understand
the mechanisms that generate that late depolarization.
And also the learning mechanisms that are involved in tying together
the sensory input, with the goal-directed motor output.
In the next video we'll begin to examine
the basis of the learning process.