Thanks for coming in today's lecture, which is about cognition and disease.

I will
first talk about what kinds what types of human memory that are. I will explain a
little bit about declarative and explicit memory. I will then focus on which area is
an episodic memory and how memories are storing information. Then I will
switch gears and I will talk a little bit about dementia, Alzheimer's disease and
cognitive impairment. After that, I will discuss which tests we're using to
diagnose dementia and how we study memory impairment in the lab. And finally,
I will discuss a little bit how we are now utilising artificial intelligence and
machine learning to the behavioural decoding. So it is not a general consensus in
behavioural neuroscience and in psychology this Yes. You can hear me. Okay.
Okay. One second. No. It's okay. No, no. It's okay. Um. Not this one. This one? No.
Maybe the one. Oh. Hello? One, two. Okay. Let's see. Yeah. Okay, good. So, shall I
start from the beginning? All right. Okay, so, um, the consensus in behavioural
neuroscience now in psychology is that we have three types of human memory.
One is the sensory memory, which is your senses. What you can touch, what you
can, hear, what you can't see. And it lasts for less than a second. The second
type of memory is short term memory, and we also call it working memory. And
this lasted less than a minute. And it's something that you can hold in your brain
for a very brief period of time, such as remembering a number or using it to
access information from your brain. The third type, which is more the third type,
is a long term memory and is more lifetime. So it's something that you will
always remember. And this is split in explicit memory, which is your conscious
memory and the implicit memory which is your unconscious. So the declarative
memory which is part of the explicit memory is what we call in everyday life
normal memory. So it's you remembering facts and events and you can actually
access them and remember them. The second part is procedural memory, and
it's a skill based knowledge that you can learn through repetition, such as riding
a bike or playing the piano. But you have no awareness of how you're learning or
access how you're learning. This is something that you can learn. And declarative
memory is further separated into episodic, semantic and perspective memory. So
these are different examples of memory that one can have. And I will explain
more. So this is episodic memory. This involves you for example going
somewhere specific and watching a play. And then you can talk about it. I went
to the Royal Albert Hall and I watched this concert. So this is your self memory
and semantic memory is your knowledge of facts, such as knowing that on the
$1 bill, the face that you see is George Washington. Um, the third type of
memory is what we call simple classical conditioning. And in this you can
associate a usually neutral stimulus, such as the mustard, in this case with an
outcome of a situation that can cause you, for example, pain or making your
sick. Um, in this case of the lady remembering that having must admit you're
sick so she doesn't want it anymore. Another example of the classical
conditioning is the experiments done by Pavlov, which was ringing a bell, and he
trained a dog that when he hears the bell he will eat. And then he would ring the
bell and the dog would salivate without even having presented the food. So this
is what we call conditioning. And we use that a lot in the lab. Um, procedural
memory is, for example, playing the piano. You don't know that you have that
knowledge. But then when you sit on the piano, even after years of you're not
playing, you can still remember it's your kind of muscle memory. And lastly,
working memory, which is, as I said, something that you can accept very briefly.
Um, for example, if I tell you the number 10,003,342, you will only hold it for a
few seconds in your brain. But if I ask it to remember it 5 or 10 minutes later,
most of you probably won't be able to recall it. And declarative memory is
basically the, the, um, memory that one, uh, is able to use to recount, um, the
details about the time, place and, and the circumstances. And this includes, as I
said, semantic memory, which is your knowledge as well as episodic memory
events, and you're able to access it to access this all the time are to request one
very important part of this, um, declarative memory is a prospective memory,
which is important for future planning. This is remembering I need to take my pill
at 8 p.m.. My to do list for my class or your supermarket list. So it's quite
important to have them all intact. On this image, you kind of see the different
medial temporal structures that are involved in them. Um, let me show it here.
Can you see my cursor? Okay. One second. I will do the. This one is faster. Okay.
Wondering. So these are the temporal structures that are involved in declarative
memory. So on this side you can see, uh, the hippocampus and the amygdala.
And these are buried deep into the cortical structures of the parahippocampal,
entorhinal and cortex. So they both exist on both areas of the brain is just one is
buried in the other. So for simplicity, we're just showing one. And this is the brain
of a monkey, a species that is very close to us, evolutionary. And we're seeing it
from the bottom up because from the top you can't access them. They're kind of
here. And the flow of information from the sensory cortices is, um, as it shows on
this diagram, on the right hand side, it goes from the sensory cortices to the
barrier island and hippocampal cortex. Then it travels to the internal cortex and
finally to the hippocampus. And then from the hippocampus through the intra
and hippocampal cortex, it goes back to the sensory neocortex. And that's how
the information is transmitted from the cortex to the hippocampus and back.
Now, the most famous um patient in the history of neuroscience is patient H.M..
This guy had severe seizures. And at the age of 15 or 16. He was unable to work
anymore because he said, it was so bad that they had to remove a part from his
temporal lobe, and that included the hippocampus. So in the 1950s, this guy
lived to a very old age. In the 1950s, the general consensus is that for episodic
memory, which is your memory of events and experiences, the only brain area
involved was hippocampus, based on knowledge from a patient, H.M.. And then
in the mid 1990s, after studying other patients, that they had bigger lesions of
the temporal, medial temporal lobe and other brain areas, they actually came to
a consensus that the areas that are involved into episodic memory are areas
here that are part of the medial temporal cortex, as well as prefrontal cortex and
all the areas that are associated with them. So they're connected. So nothing is
as simple as it looks. It's not the one brain area. It's many together that are
making episodic memory. So how do memory store information? One of the first
theories was from Arthur Miller in 1956. And this guy came up with this magic
number seven. And he said that one can hold in its brain seven items or seven
bits plus minus two. And he did lots of studies using words, using images, using
different things, starting working in humans. And he proposed that one can use
chunking. And this is something that the musicians are using a lot. And they
break everything in items of seven. And it's much easier for them to learn. And
then also that the brain is recording in seven chunks limits, because this is as
much as it can handle in the working memory, which, as I said before, is your
short term memory that only lasts for a brief period of time, but the brain is able
to then process more information and then one can learn. This kind of, um,
working memory. Move it into your long term memory through repeated
rehearsal and through basically, um, intense effort. You're able to store these for
a longer period of time. So that was in 1956, uh, in 1968, these, uh, these two
researchers, Atkinson and Schiff and Shiffrin, came with the multi store model,
which in many ways is still the one that most neuroscientists are following, um,
in this day. So they envision that the memory is in the three stores, uh, the
sensory storage, the short term memory and the long term memory storage. And
the way the information flow is, it goes from them. Um, the incoming sensory
information goes through the environment, through your sensory memory, and
then it stays there for a very brief period of time in the case. And it's lost. Some
of it if you pay attention, for example, can move into the short term memory,
which again can stay for a little bit longer. But again it's forgotten and it's only
through, um, what they called rehearsal, that you're able to retain that
information into the short term memory storage. But this can also go to the long
term memory through consolidation by repetition, for example. And also the
short term memory can access the long term memory through what we call
retrieval. So you can retrieve for a brief period of time some information that you
need. And this is very important. For example, um, for quick decision making for
IQ, for school learning, all of these aspects are very important for this retrieval.
So then vision that this short term memory can be auditory, verbal and linguistic.
And they found that there's levels of processing. So something that you will
process more it will stay there for longer. Something that you want to process. It
will decay and be forgotten. So when they say transfer from one um
compartment to the other, the memory is not forgotten from the previous stage.
So the memories are forgotten from the short term memory to the long term
memory. It's more copied. It's a bit like a computer. It goes from there on to the
hard drive. So in 1974, um, these, uh, these researchers at, uh, this model, uh,
the multi-step model is very simplistic. And they came up with their own model,
which they called the working memory model. So but in each Which were
researchers in York University. And they proposed that there are two components
in the kind of like working memory. One of them is the phonological loop that
briefly stores and rehearses verbal, verbal contain. And then it probably holds
the information and then it moves on. Um, the second part is the visual spatial
sketchpad. It's kind of like your inner eye or the previous one was the inner ear.
That's how they call it. So the inner eye is holding the visual information again, is
storing it is manipulating it. And then it's your mental canvas. And then it can
move on to this, um long term memory. So in 2000 they revised their theory and
they came up with another third component, which they called the episodic
buffer. And this was, um, the combination of sounds visual and long term
memories into short, temporary experiences with the limited capacity and all of
that, the vision that they are controlled by this central executive, this kind of
conductor that is controlling everything. And it can move things from one
compartment to the other and from the short to the long term memory. So in a
way, they built upon the previous the Atkinson model. Uh, but even to this day,
they are still writing papers. They had one, um, last month that they selected
that the multi store model is wrong. Um, so there are more theories on how
information can be stored or lost. Um, and you can find it in this link. One that I
found very exciting was number five. They work on effect on eyewitness
testimonies. So they did this study where they showed two people, someone
working out with a a knife that had blood on it and someone that didn't have
anything on him, and they ask the witnesses to describe the person. And the
people that saw the guy that had the bladed knife couldn't do it. There were so
focussed on the weapon that they couldn't describe the other person. And it's
quite interesting how an eyewitness testimony is especially in, in, in the court,
having a weapon at the scene can affect and make false testimonies from the
witnesses. So if you're interested, you can read more on how memory decays or
stored. So what happens when memory goes awry? We get dementia and
dementia is very prevalent. Um, these are the latest numbers from this study
that was published in 2024in the UK. Right now there are 982,000 people living
with dementia in the UK. A third of them is over the age of 85, which means most
of them are in care homes. Um, and this number is projected to increase to 1.4
million by 2040. That's a huge number. And why is this so important? Because
it's costing the UK economy £42.5 billion a year, and half of it is from unpaid
care. Which means that your mother or your aunt or your brother is looking after
this person that has dementia and they cannot work. So the government is is
losing this, this money. And there are their big concerns because this is projected
to increase to 90 billion. So you can imagine a lot of this is on the NHS. A lot of
this is on the people and on the care homes. So just to give a context, um, is
every three minutes someone in the UK develops dementia And one out of three
babies born in the UK this year will develop dementia in their life, and one every
two about 1 or 2 of us will either develop dementia or be affected by caring for
someone with dementia or both. So this is this is a big, big, big problem. So as
you heard this morning, Alzheimer's disease is the most common cause of
dementia in the aged population and is characterised by generalised brain
atrophy. Um and the two main pathologies are neurofibrillary tangles comprised
of hyper phosphorylated protein and with amyloid plaques. And it's more
prevalent in women, which in this is shown as blue. For some reason they
decided blue. And they studied. And it increases with age. And the disease is a
continuum. is not one stage, then the next, and the other is a continuum that
starts if you focus on the amyloid plaques shown. Um, sorry. So right there,
there's the disease starts a good 20 years before any manifestation of any
symptoms of dementia. And then it plateaus way before you have the dementia
onset or the memory impairment. The neurofibrillary tangles, that is the second
neuro pathological finding starts again earlier and it starts accumulating around
the mild cognitive impairment phase, which is when the person starts forgetting.
But it's not full blown Alzheimer's yet. And then it plateaus. Um, then you have
the brain atrophy, and then slowly the person will start getting dementia and
losing itself. So As Damian discussed, the stages are not perfect. Uh, but I'm
going to explain a little bit here. In this image on the right, you can see with the
red arrows the hippocampus. The more it progresses that it is progressing, the
more atrophy you have on the hippocampus, uh, shown with the red arrows. And
another area that is is very interesting and it's heavily impacted during the later
stages of the disease is the Bronwen area 19. This area together with um Brain
and Brockman area eight are part of the extra, uh parietal cortex. Striated
cortex. Excuse me. And they're involved into very complex visual information.
So, for example, a person that has Alzheimer's cannot drive because they can't
just, um, translate the distance or what they're seeing or where they're driving.
And this cartoon on the left hand side is basically showing that tau is there in the
medial temporal cortex. And you can see that then it starts moving towards the
medial prefrontal and the medial parietal cortex. And when it finds amyloid you
have cognitive decline. So the two of them together, what the scientists are
saying is that one is the bullet and the other one is they're gone. When they're
together, the brain just go, you know, demented. So there are quite a few
cognitive domains that are affected in Alzheimer's disease. And here on this
paper, they show two men of the same age, and they're showing their cognitive
decline here on the bottom left. And you can see how one is progressing fine and
its cognitive score. It's okay. And then the other one just declines with age. Um,
sorry. Let me. So the cognitive domain is about and how it manifests. This is for
example, uh, we have impairments in episodic memory, and this can be
expressed by amnesic episodes such as I forgot where I put my keys. I don't
know where I've left my purse. I don't know how to return home. Um, the second.
This is one of the very early signs of Alzheimer's disease. The second one is the
attention, uh, functioning problem. So quite early on, Alzheimer's patients cannot
focus. They can't hold their attention for brief for a long period of time. They're
scattered. They're just all over the place. And they also lose, um, a lot of their
executive functioning, which is manifested as a loss of inhibition, for example, is
not very, um. It's quite often that one can see its grandfather running naked in
the house, or some of them are even hypersexual, and they're showing it, and
their carries are quite frustrated because it's quite upsetting to see your loved
one in this state. Like I mentioned, there's the impairments in the visual spatial
functioning. One of the really nice days that the neurologists are doing, they
wave their hands about there and they ask the patient to rub it, and they kind of
do it. They go like this. It's kind of like they can calculate the distance of where
the neurologist hand is and where they're trying to grab. One other things about
the neurologists are doing is asking them to cross the two fingers, and they can't
do that either. So it's quite a sensitive task to do. And in later stages you have
impairments in language such as not finding the right word, saying. For example,
uh, not not being able to find the word cut and saying, oh, that pet that miaows
or they will use a similar word such as, I don't know, um, tiger instead of cat,
something that is similar but not what they mean. And then we have, um, lastly,
one of the um, later is the flood effect. So effect is a psychological, um term that
means that it's affecting the physical and the mental experiences of a person.
And it is when you see an undiscovered patient being a bit, uh, catatonic, they're
unable to express their emotions, they're unable to be happy or sad. They just
have almost no reaction. So I can tell them you have a new grandchild. And the
response is, oh, okay. So it's it's quite sad to see that. And. It's a bit stuck
moving. Okay. Okay. So how do I. How do I diagnose dementia? Uh, the first thing
that, um, one can do is go to the GP and they run this general practitioner
assessment of cognition tests. As you can see, here is five questions. It's very
simple. Um, they ask them to remember a name in an address and they will ask
about it later. They ask what is the date? Uh, they ask, uh, to draw a clock. And I
will show some images later. They ask them, can you tell me the latest piece of
news? And they don't expect to hear that they were somewhere. They expect
them to hear their say in such and such country. It has to be specific, so they're
aware that they know what time is today and what is happening around them.
And then remember I said that they give them a name and ask them to
remember them, and then ask them to record that name. And they just give
points whether they remember everything or nothing. So another examination
that is mainly used in the US is a monumental state examination that, again, is a
very short questionnaire of about 30 questions and one can feel it in 30 in 10
minutes. So again it's very quick. Very frequently they ask the patients to do
blood tests. Most often is to exclude another condition. For example, um an
undetected uterine tract infection can cause dementia surprisingly. So they want
to exclude it could be a question of giving them some antibiotics. And in this
case also in many cases also take blood for and search for biomarkers. So we
now know that quite a few biomarkers are very good at detecting in the blood.
For example, first for tau um that is high in the brain, it can get into the blood
and we kind of see it, and it correlates pretty well with the pathology in the brain.
Um, on the bottom left you can see some brain scans. These are different brain
scans, but we can do, uh, in the patients. These are quite expensive. So usually
neurologists will first refer you for an MRI. And then if the MRI is inconclusive or
there's more words, uh, a more worry. They refer to pet, which is very expensive.
And in the top left you can see it's, uh, fluorine 18 FDG. And this can detect
deficits in glucose metabolism. So, um, The you see here the medial temporal,
that is kind of like the hippocampus. And all of that is quite low. The second one
that they use is the carbon 11 nicotine. So this shows deficits in cholinergic
activity. So we know that the culinary neurones are the first ones to be lost. So it
can be picked up. The bottom left is Pittsburgh compound which it can detect
fibular amyloid, which as I mentioned before is the main component of the
amyloid plaques. And we kind of said and the last thing is a carbon 11 PMP that
is affecting high um acetylcholinesterase activity. So you can see that the only
area that has high activity is the striatum here that is not affected by Alzheimer's
disease. So here is the clock. So it goes from higher scores five and zero. So what
happens is that the neurologist will ask a patient to draw a clock at 11:10 and it
has to be. The arm has to be longer on the two and shorter on the 11. Then the
numbers have to be anchored at 12, six, three and nine, and the circle has to be
round. And you can see, for example, it progressively deteriorates. And you
know, these guys just can't do it. And at series they don't even draw a circle in
some occasions. So it's a very sensitive task. They have other tasks as well that
they ask you to draw. A shape, looking at it from the metre. And it's actually
quite hard to do it, so try it at home. Um, we also have other tests. Uh, this one
was actually developed here in UCL by the group of Neil Burgess. Uh, and this is
uh, the for mounting test and this tapping into the other centric spatial
hippocampal dependent memory. So understand what it means that you're
making inferences about your environment by looking at different spatial cues,
such as different buildings, landmarks, um, how the the sun is. And this is very
sensitive at picking up hippocampal impairment. So what happens in this task
you have this kind of like landscape. And they show them this first item and they
say okay can you see these mountains now. They allow them eight seconds to
see them. Then they cover it with a blank page for two seconds, and then they
show them for different views. And they said, can you tell me which one is this
mountain task? So can you guess which one is the right one? You're looking at it
from a different angle, but you have to like realise. So there is a bottom right. It's
quite hard, but you have to basically think that you're looking at the images and
you're making an inference from the clouds, the way the, the mountain is and all
of these things. So. Also, another group at UCL, led by Dennis Chan of the ICN, is
developing this virtual reality task that is tapping into the internal cortex and we
call it the path integration test. So what happens in this task? And it's very
sensitive to the lateral internal accumulation of tau. So it's again a very sensitive
early test that we can do. And actually the team of them is trying with General
Keith and Karen Dave. They are now testing it in a cohort in Colombia that they
have familiar large camera disease, to see if they can detect it very early on
before these people develop dementia at the age of 40. So what happens at with
this task is that you have to find the cone. So you have to go from cone one. So
this is kind of like the pointer that this person that is in the VR is kind of like
pointing to you find cone one. Then you have to travel to count to let the cone
three and then go back to one. And then they do different things such as
changing the conditions of the day. Then they remove the plantation or they
remove the mountains. And this is actually quite hard because this is an
egocentric kind of task. So you have to remember that I walk ten metres this
way. Right. And then I went this way and then to go there. So the right angle is
basically you have to walk back. So this angle of view turning from one direction
is very sensitive. So this is we measure the perfect angle which could be 30
degrees to go back to your to cone number one or for example, um, uh, you
turned 60 degrees and you made it there, but through a longer path. So you
made it back to the original path. And we can measure distance from the original
path, time that it took you to get to the original path or the angle. And these are
very, very sensitive. So we can use these metrics to find which people have
impairment in their internal cortex. So how do we study it in the lab. So I'm an
animal behaviourist and I use mouse models to study Alzheimer's disease. And
the first generation of mouse models that we have used were transgenic mice
that were overexpressing the human amyloid precursor protein with some, uh,
familial mutations. I can see your frustration. The slides are slightly different. I
will upload them after. I apologise for that. So, uh, what, uh, we did in the past,
we just use these methods that were overexpressing AP, and we found out that
actually there were some toxic By-Products that were from this, uh, from the
cleavage of this protein that we didn't want there. So a group in Japan led by
Takuma Sato developed this humanised app, no mouse models, in which they
changed three amino acids to make it more like human like. And this is express
under the mouse app promoter. So it's expressing normal levels. And to make it
develop amyloid plaques, they introduce three familial Alzheimer's mutations the
Swedish, the Arctic and Siberia. What does that mean? It means that you get a
lot of amyloid beta deposition in the brains of these mice, mainly in beta 42,
which is the more toxic species. and you get a lot of your inflammation, which is
the plus. And then you have activated associated microglia. That is the brain's,
uh, resilient immune cells that will attack any kind of pathogens. And then this
these guys also have memory impairment. So this is a picture of one of my mice
that has some staining for amyloid plaques, and is exactly the same staining that
the neuropathologist will use to find amyloid plaques. And these bright spots are
amyloid plaques and they're everywhere. So these mice are quite impaired. They
have also generated mice using other genetic factors such as Trem two. For
those of you who are interested in genetics, and these have been able to worsen
the pathology or the cognitive impairment. And there's a lot of studies now going
on. So the second player in Alzheimer's disease is protein marketing and the
mouse model. The mice normally would express three out of the successive
forms. So in humans you have six isoforms generated by alternative splicing of
them up to a gene. So what happens is that in mice you only have three that are
what we call the four hours. So these researchers again from the same lab led by
not what they generated, these are mapped in mice that have all six isoforms.
And again there's these mice that have no mutations. They humanised mapped.
They don't have any tangles. So what they did is they had to introduce some
mutations. So we don't have any time mutations in Alzheimer's disease. Um the
T just goes away by itself. And we're not entirely sure why, but we have
mutations in front of temporal dementia, and we utilise this mutation to generate
mice that have, as you can see here, hyper phosphorylated tau. Some of them
can get tangles like like humans. And we're trying to study them, um, in the lab,
trying to see how their memories affected, how their motor skills are affected
and how their pathology is affected by different interventions that we're doing.
We also use a lot of cognitive tests in the lab, and we basically put them in
different arenas. We let them run for a bit and then we assess their behaviour
using those videos. So one of these tasks is the open field. And in this we can
assess their locomotion and their signal taxis. So locomotion is just how much
the mouse is moving. Some of them can move a bit slower. Some of them may
be faster if they're a bit anxious. And they sigmoid axis is when a mouse is the
one we call wall hugging. So they're just running in circles close to the walls,
because either they're hyper aroused or they're too scared to go in the middle.
And we can again use this as a proxy for, um, anxiety. We have all this other
things, other countries anxiety, such as the elevated plasma shown here. So in
this task we have two closed arms and the mouse feels safe. And then we have
two open arms. The plasma is about 50cm from the floor. So even if the mouse
force is not going to um um hurt itself. And then we can see how much time
these mice can spend in the open arms, which means they're less anxious. And
this kind of task has been used a lot with anxiolytic drugs. So, um, it's it's quite a
good one. Then we use the image for spatial working memory. So the mouse has
to alternate between the three different arms and the mouth has a good what we
call special working memory. We'll go from the top to the bottom to the top left,
and it will keep alternating. And then we count how much correct alternations it
has done. And again that's a proxy for hippocampal lesions. Another task that we
like to use is the normal arm maze. So this one is a transparent one and it has
visual cues all around it. We close one arm. We let the mouse explore the other
two for about five minutes, take it out, put it back into its cage, and then open up
the third arm. Like I said, there are cues and the mouse. The mouse is curious.
It's going to go and explore the normal arm. So then we can I am characterised
there what we call the discrimination index, which is a percentage of how much
time they spend into the normal arm versus the other two arms will take as well
at our box. So imagine that you're in Wembley Stadium with loads of lights
shining on you, and you're scared and you're looking for a dark place to go and
hide because everyone is watching you. So the minds do. They say they just tend
to go into the dark box, and we can use how much time they spend into the light
versus the dark box to see how stressed or anxious they are. These at the
bottom left are some of my favourite tasks because they're very hippocampal
dependent and we can say spatial memory. One is the more water mate. So the
mice have to learn to find a hidden platform. This platform is about one
centimetre below the surface. We have milky water basically, and you put the
mouse in for up to a minute and it has to find the rain. And again you have visual
cues and the mouse. We have some amazing minds that you put them in the
water and they go like this, and they just run because they know where the goal
is, and they tend to learn it pretty quickly. So typically, um, a wild type model
learned within 10 to 15 days with five trials a day. So it's quite quick. Um, very
similar task is with teams. So the Team-mates again is tapping into their spatial
memory. In this we have one arm closed here, the most plastic on the left arm.
They collect some reward which is usually strawberry milkshake that they love.
And then we take them back into their start box and we open both arms. And
they must remember I collected the reward on this side. So now I'm going to go
and explore the other side to see if I can get more, because we know that I
consume all of it here. And that again is very sensitive to hippocampal lesions.
Another task that we do is object location. So you put two objects. The mice are
curious I'm going to explore them taking them out of the box. Put them into their
home cage, return them a little bit later and you displaced one, and the mice
would be like, oh, this was not here before. And we'll go and explore the
displaced object and quite sensitive tasks. And the last lecture says entorhinal
cortex, which is a normal object recognition which is tapping on recognition
memory. So in this task and Mars is exploring two identical objects, and then you
remove them and after some delay, we usually do between 15 minutes and 24
hours depending on what we want to see. Um, we return them back into the box
and we change one of the objects with another one that they've never said
before, and they're very excited to explore the normal object because it's a new
thing. So I've used these tasks in the past to test some of the mice cognition. So I
did a gene therapy intervention in Alzheimer's Most models. What I did is I had
wild type mice and AP 23, one of the transgenic overexpressing lines. So I
injected these mice with viruses that were carrying either green fluorescent
protein as control or PGC one alpha, which is a master regulator gene of the
mitochondria. And I was trying to see if it could be used as a therapeutic agent.
So what I saw when I ran the level of recognition is that DPP 23 milestone in blue.
They couldn't do the task. They couldn't remember that they had never seen
that object before or the familiar. They couldn't remember it. Whereas the and
the wild types, obviously they could do it irrespective of what virus they got. The
AP 23 mice that got the biggest one, they could do it, and they were even better
than the wild types, and they could discriminate between a novel and the
familiar object. Similarly, I did the object location task where this placed one of
the objects. And after 15 minutes the non displaced and the displaced object is
here. And then you can see the wife had mice to do it. The AP 23 mice that
received the control virus could not do it. But there was about received the
therapeutic could do it. And the brain areas that are involved did all of this injury.
So pretty much everything was involved in two of these tasks because you need
the prefrontal cortex intact. Like we said, in episodic memory, all the medial
temporal cortices have to be intact to be able to recall these memories. So
there's a lot of us about artificial intelligence. And I work at the Sainsbury
Welcome Centre, where we have the guts, the computational unit. We love
machine learning. So we started using this kind of analysis to Basically. Scholar.
Writer. So, um, you have this software, they're all open source. Anyone can use
them. They're pretty good fun, if you're interested. One of them is the block
cards, which is the one I mainly use. This one is sleep is, uh, for social
behavioural experiments, like two mice interacting with each other. And it is a
little bit that it can understand motifs in the different behaviour and is doing
some fancy unsupervised clustering. And then it can pull out differences between
two different genotypes for example or intervention. So usually the the workflow
is we extract frames from let's say 20 videos, uh ten frames from each video.
That takes a few hours because the computer has to go through this algorithm to
see this clustering. So it doesn't extract the same frames. And they have to be
different. It takes us about one day to annotate it. rotated. So we annotate the
the snout, the ears, the neck, the centre, whatever. You can do anything from 8
to 20 points. Um, then we trained the deep neural network. We usually leave our
computer to run overnight. It takes about 16 to 18 hours with a good GPU. Or
now you can use, for example, Google Colab on the cloud to do it. Um, then
usually we go back and see if our network performs, uh, fine. And we can
visualise the images that the computer has annotated. And if we need to refine
it, it will take another day. And then we run inferences in on your video so that it
never run before. And this could take about 20 minutes for 100MB Video. And
then we can compare it because we always care about the ground truth. We can
compare it with our, um, data. The one that we scored takes about a day to do
the analysis and the post-processing. So in total, this takes between 6 and 7
working days. In the past when we're doing mango plantation, this could take
weeks. For example, when I was manually annotating a ten minute behavioural
video for exploration, it would take me anything between 20 and 40 minutes. So
you can multiply this by hundreds and hundreds of videos that we have in the
lab, and this can immediately go up to weeks of us sitting there on a computer,
just watching a mouse and going frame by frame. So this was a revolution for
behavioural neuroscience. We also use a lot of automated home cages. And the
reason we do that is because sometimes when you pick a mouse from its cage
and you then put it into an arena, you get very stressed. So the best way of
running your experiment is to do it in a non stressful environment, which is our
home cage. So we'll have these specialised cages. Um one of them is the
phenotype for the phenome I said and they delegated to the main difference is
that the, the phenotype in the master can house only one mouse where the
intelligence has up to 16. In the intellect, the mice have an idea chip on them,
and they can do the different tasks just by RFID recognition. And you get really
nice results. You can do, for example, an aggressive task that you teach them
that the bottom on the bottom left has something that is bitter and they don't
like it was the top right has, uh, sugar that they love. And you kind of see how
much they are avoiding one or the other. Um, lastly, the smart cage is the one
that we currently use in the lab. It was developed by the group lab, um, at
Cambridge, at the University of Cambridge, and she started at the UK, DRI, UCL
And in this we can do different things. And I will explain shortly. So you kind of
see here a mouse. So the mouse can go in the left of the right hand side. And it
can perform what we call a smart team-mate. So the mouse goes on one side, it
will drink water, and then it can go on that segment in water. It has to go on the
other side. And we can also measure uh, creation, which is a proxy for sleep. We
can measure something more taxes. Is that going around in circles? That is very
um, is a typical behaviour that we see a lot in hippocampus and mice. And I can
play it again. Let's see if I can manage that. Um, we can also see how fast they
run and we can measure their mobility. And we have also the, the normal objects
that they can go and explore on each side to a different object. So there's quite a
few things that we cover in these tasks and we leave them in that cage for for
five weeks and then we can analyse lots of data. So in the original study here
they show, for example, quiescence. They run this um app in mice of Alzheimer's
disease, uh, at the age of 5 to 9 months that they're not that impaired. And then
at the very old age, 18 to 20 months. And as you can, um, I'll show you, because
I think this is not working well. So here's the activity device, okay. And you can
see this is the room shift and this is the control. So these are more hyperactive.
So red is more hot. So it's running. This one is not running. So these are
hyperactive. We can also study the quiescence. And we can see that they
became at the older age they have more fragmented sleep and they become
excessive activity. And you can see that the control tone change But if you can
hear pop music. So what this is showing us is that we can use, uh, metrics from
these tasks. And we can also do the team-mates. And you can, uh, for example,
appreciate here that at, uh, five and nine months, there's not much difference
between the controller and the mice. But in the old guys, the people gave the
count to the task. So after, because the mouse is deciding how long it's going to
take between drinking from one spot and the other, we can find, okay, when do
they start showing deficits or signal? Remember? So then if you have mice after
five minutes, they just can't do it without remembering. They keep going to the
same one where the the app controls. They start getting impaired, not
remembering if the in between. Um, it's a trial interval. It's over 12 minutes. So
again we can use different metrics and identify when our mice are getting
impaired. So now we have run different mice, different ages, different genotypes,
and we're not currently analysing. It's quite uh it's quite nice to see it. And you
know, I can log in from home and watch my mice. See what they're doing. So it's
quite nice. So this is kind of like knowledge recap I don't know if you want to
answer this question. So if you have any questions for me. Um, so we went
through all kinds of memory how dementia is affecting these types of memory.
And yeah, any questions? Feel free to either email me or come and ask me at
any time. And thank you for your attention.