Neurons, synapses, and the instinct of survival | 91TV
Transcript
- Thank you to my very dear colleague Greg Jeffreys for nominating me and to the colleagues who wrote
- in support of the nomination, and thank you, all of you, for coming tonight and also for
- the people who have joined online. I'm going to continue and start by thanking people. I've
- never done science by myself. I've done always science as part of the team, and so I'd like
- to acknowledge all members of my lab, past and present, who contributed, all have contributed,
- to the body of work that this award is for. I thought I'd also list all of the people who I
- published papers over the years just to drive the point home of how international and collaborative
- science is. There's more than 130 people listed here from 25 different nationalities.
- I'm going to single out one person, which is Domingos Henrique, which was my scientific advisor
- in Portugal who took me into his lab when I was a second-year medical student, a rather useless one,
- actually, and he just let me have fun in the lab. He taught me how to plan experiments,
- think about experiments, read papers and think about problems. Then he also really pushed me
- to come to the UK to join the Wellcome Trust PhD programme at UCL that ran for many years,
- which was a really important step in my career. Now, this lecture is about neurons, synapses and
- the instinct of survival, as John said, and I'm going to spend a little bit of time introducing to
- you instinctive behaviours, which are behaviours that animals do without prior experience.
- They're behaviours that animal doesn't have to learn, and therefore they constitute an expression
- of innate knowledge. The animals are born knowing how to do these behaviours and they can be very
- simple, a simple reflex, or they can be complex series of actions, but they all have in common the
- fact that they have evolved to promote survival. There are many things that animals have to do in
- order to survive, and evolution has hard-coded some of these things in the DNA to give animals
- a head start when they are born. To illustrate some of these instinctive behaviours, I'm going to
- show you a video, a sequence of videos from that BBC Natural History Unit has very kindly shared
- with us. The video should be coming up in a little bit. Can we have the video, please? There we go.
- So this video follows one of the most remarkable instinctive behaviours in the animal kingdom,
- which is the salmon run, where every year there are millions of salmons that leave the ocean to
- go and find a river that where they were born to spawn and to eventually die. The particular
- salmon run we're going to follow here happens in the northwest coast of America. It starts
- in the Pacific Ocean and then salmon will migrate about 5000 kilometres to the rivers
- of British Columbia. Not only do they have to navigate to the coast. When they get there,
- they have to navigate this maze of rivers in order to find the exact one where they were born,
- which they can do with really remarkable spatial precision. So this illustrates two key instincts,
- the instinct of migration, where animals migrate over long distances to find places
- with better resources - here the resources are the shallow waters of the rivers which are safe
- place to put eggs - and also the instinct of navigation. Animals are very good at knowing
- where they have to go and getting there. Now, this journey takes many months,
- and during this journey they will face a lot of challenges. The first challenge that they
- face are coastal predators. The most deadly one is this, is the bald-headed eagle, which captures and
- predates on fish with a really remarkable degree of skill and precision, and this illustrates one
- of the most important instincts in the animal world, which is predation, which shapes all other
- behaviours. The top priority of all animals is not getting killed. You might go out to find food, and
- if you don't find food that's fine. You can come back the other day. If you fail to avoid predator,
- you're probably not going to come back the next day. This is the topic of today's lecture,
- avoiding predators. Now, for the fish that make it, and make it past the predators - and
- there's a lot of fish that make it past the predators - the challenge is not really over
- yet because this is what they face, right? It's the proverbial uphill battle where fish
- will have to swim upstream, jumping over rapids and rocks, but which they can do with really
- remarkable degree of physical prowess, just like if they have been practising this all their lives,
- but they haven't. This is the first time they're swimming upstream, yet they do it,
- and they do it because the result of millions of years of evolution. Now there's another problem,
- however, is that there are some bears that are waking up and they're really hungry. So grizzly
- bears have been hibernating here at the top of these snowy mountains. They have given birth,
- and now they are going to start a journey of their own that will last many months to
- go and meet the salmon run and get some food. They do this, and this illustrates first again
- this really important instinct of navigation where animals are really good at knowing where they have
- to go and getting down there, but also another one which is, these bear cubs, this is the first time
- that they've been out of their den and yet here they are going down a really steep hill, perhaps
- not as elegantly as their parents - certainly much more elegant than me going down a ski slope - but
- they do it. This is a very scary descent, but they do it because they have to in order to survive. So
- this journey will take them a lot of months, and then they will eventually go to the rivers where
- the salmon are and what they have to do… Well, the first thing they find is that there's a bunch of
- other bears that had the same idea and they'll have to fight for positions. Here they are.
- Then they have to catch the fish, and catching a flying fish, a very slippery
- fish that's flying past you, is not easy. This illustrates another very important aspect of
- instinctive behaviours which is, these animals might have been born with a drive to catch fish,
- even maybe a set of basic actions to catch these fish, but then they have to refine this skill over
- years through experience. Bear cubs, for example, are really bad at this. They can't do it. This is
- true for most instinctive behaviours. They provide a basic blueprint that sets a basic set of actions
- that is then refined through experience to give animals the best possible chance of adapting to
- the version of the world that they were born in. Now, a lot of fish do make it past it, and then
- eventually they reach the shallow waters here of the rivers that they were born, and they'll engage
- in a more peaceful set of behaviours, courtship, nesting and eventually spawning, and the start
- of a new generation which will grow up here for about a year, a year and a half go back down, and
- then four years later come all the way back up. Now, these behaviours, some of these behaviours,
- are very species-specific right. Not all species travel 5000 kilometres in order to breed,
- but there was one that I mentioned that is really common to all species,
- which is avoiding predators. This is because from the moment there was life there were there were
- always species predating on each other, and a very strong evolutionary pressure to develop behaviours
- or evolve behaviours to avoid predators. One of the most universal behaviours for avoiding
- predators is just to run away, to escape, which is a behaviour that has been reinvented time and time
- again during evolution. At its most basic level, escape is a locomotor action that moves an animal
- away from a threat, ideally towards safety, but because of the enormous diversity of species,
- of their habitats, of their predators escape, can be very varied. You can escape by flying,
- by jumping, by running, by swimming. Despite all of this diversity,
- escape can be decomposed into a set of common elements. If you think about it, all animals,
- the first thing you have to do in order to escape is, they have to detect a threat. They have to
- use their senses to figure out whether there's a threat there or not. If there is a threat,
- they have to choose or decide what to do next. They might decide to escape or they might decide
- to do another defensive action, such as stay very still to avoid being detected or even engage in
- the fight if the predator is too close. If they do choose to escape then they will have to execute
- this escape as fast and as accurately as possible. They should really know where they're going,
- not getting cornered, for example, and then they have to terminate the escape,
- ideally as soon as possible as well. Escape is very costly, not only in terms of energy but also
- the cost of missed opportunities. When you're escaping, you're not doing many other things
- that you also need to do in order to survive. As John mentioned, all of these processes are
- modulated by experience. For example, you might learn that something that you thought was really
- scary is not scary anymore, or you might find that as a new safe place in the environment that you
- should get to, that you can get to more fast, for example, or faster, that you should use. The goal
- of our research is to understand how the brain implements all of these steps. We think that the
- answer to this might be interesting, because not only we'll learn the biological basis of
- instinctive behaviours, of behaviours of survival, but also because many of these steps are common to
- many other behaviours. For example, the decision to escape is a decision, but we make decisions
- every day. We choose actions every day. We execute actions every day. So maybe by understanding how
- the brain implements this process in a simple behaviour, we can learn something about how
- the brain solves wider problems in general. Now, ideally we'd like to understand the human
- brain, because the brain that we care about the most is the most powerful thing that evolution
- has produced, and it would be great to be able to fix it when it goes wrong. Our research, however,
- is done at the very cellular level. We want to understand the biological processes behind
- each computation. This means that we need to get quite messy with the brain. We need to do invasive
- recordings that give us the spatial and temporal resolution that we need. We want to take things
- out of the brain, put artificial things into the brain. Clearly, this is something that we're not
- going to do in a living human. So we go down the phylogenetic tree and we stop at the mouse, where
- we can be a little bit more invasive while making sure that the animals are happy - we go to great,
- great extents to ensure that all animals are very happy - but where we can have this level
- of resolution and precision that we need. Now, the mouse brain is smaller than the
- human brain, has about 1000 times less the number of neurons, but if I slice these brains now like
- this - you're seeing the brains from the top - if I slice them and then flip them and I blow
- up the mouse brain, you can see that the anatomy is different, but actually many of the regions
- that exist in human brain can be identified in the mouse brain. For most of this talk,
- I'm going to be focusing on a part of the brain called the midbrain, which has a very clear
- equivalent between the human and the mouse. If we look at the neurons actually of a human, if you
- compare human neurons with mouse neurons, they're actually even more similar. So their physiology,
- the way they work is very similar. Their anatomy is very similar. The main difference is that the
- mouse neuron is about a third of the size of the human neuron. These ones here are drawn to scale.
- So, for you to be able to follow this lecture I need to explain to you how a single neuron works,
- which is something that I think we understand really well after more than 100 years of research,
- and which I'm going to try and summarise for you in about four minutes. So this is a cartoon
- of a neuron. The first thing you can notice is that it's aa very asymmetric cell type. It's very
- non-uniform. It has at the centre a cell body. The main role of the cell body is keeping the neuron
- alive. Then at the top it has these protrusions that form a tree-like structure that John already
- mentioned to you that are called dendrites. The main job of dendrites is receiving input. The
- input arrives into neurons via synapses. Synapses are specialised junctions between neurons,
- and so the dendrites have a bunch of synapses to receive input from other neurons. At the
- other end we have this long thin protrusion that is called an axon, which is responsible
- for generating output and passing information to the next set of neurons, and at the end
- of this axon there's also some synapses. So a neuron has synapses at the dendrite
- to receive input and synapses at the axon to transmit output to the next set of neurons. Now,
- the most important property of neurons is that they are electrical devices. There is
- a membrane potential difference between the outside and the inside of the neuron. So if
- we look if we record this potential when the when nothing much is happening, the neuron is
- kind of bobbing around what we call the resting potential. It's just kind of sitting there. Now,
- if the neuron starts receiving a lot of input, the potential changes, and at some point it's going to
- hit a threshold that is indicated here by this dotted line and generate what we call an action
- potential, which is a very large and brief change in the membrane potential which is responsible
- for passing down information. So these action potentials are going to activate the synapses.
- Where is this? Ooh, this is tricky. Wow, okay. Anyway, the action potential goes like this and
- then activates the synapses. So what happens at the synapse? This is a blow-up of the synapse.
- You can see the axon. You can see the dendrite. The main property of the synapse is that it has
- these vesicles that are filled with molecules of neurotransmitter, which are typically amino acids,
- for example glutamate. When an action potential arrives at the axon it causes these vesicles to
- fuse with the membrane, release neurotransmitter and it's going to then act on receptors on the
- side of the dendrite and cause a potential change at the dendrite. So, the result of
- this is you get a potential down the axon, you get this little magic of synaptic vesicles, and
- then the potential propagates to the next axon. Now, one very important aspect about synapses
- that I'm going to need you to remember for the duration of this talk is that most of the times
- an action potential comes down the axon nothing happens. Because of biological constraints the
- probability of actually releasing a vesicle is very low. For example, in the cortex, the
- most advanced part of the brain, you need about five action potentials to release one vesicle,
- so the probability of release is about 20 per cent. During my PhD working with Yukiko Goda and
- Kevin Staras at UCL we developed some methods to measure this release probability directly using a
- bunch of different techniques, including electron microscopy. This is what I'm showing you here, is
- a reconstruction of a synapse in the hippocampus where you can see the dendrite and the axon,
- and the synaptic vesicles inside the axon. What we did was to develop methods where
- we could deliver a fixed number of action potentials to this synapse, to a single synapse,
- and then turn the vesicles that have been released black, literally black. Then we can go into the
- electron microscope and just count the number of vesicles. We knew the number of action potentials.
- So we can compute directly the probability of release. Using this and other methods,
- what we found is that the release probability is a very dynamic property of the neuron that can be
- changed up or down and depending on what's going on in the neuron. If the neuron is really active,
- speaking really loudly, release probability goes down. If the neuron is really quiet,
- release probability goes up to make the neuron louder. So it keeps it in this dynamic range
- where it can integrate and pass down information. So what happens when a vesicle is released? So I
- told you that a vesicle acts on receptors, causes a change in membrane potential,
- but the catch is here that the change in membrane potential is very small. So in order for neuron to
- actually generate any kind of output, you need to add up a bunch of these synaptic vesicles
- together. The way the neuron integrates this information defines the input/output function
- of the neuron. So, what does the neuron do with the input it receives? Here, for example,
- I'm illustrating you a straight line which is a linear input/output function, meaning that the
- neuron is just adding up synapses at a constant rate. During my postdoc working with Michael
- Hausser and Beverly Clark, one of the things that we set out to do is measure these input/output
- functions in a bunch of different neurons. What we found is actually, if you look at the
- synapses here, we when we shifted synapses… We decided to activate a bunch of synapses at the
- very tip of the dendrite all clustered together. What we got was an input/output function that was
- really super-linear. So we started getting much more activity than you would expect
- if everything was linearly summating. This was exciting because we did a bunch more experiments
- to show that this kind of property allowed neurons to do very powerful computations,
- such as discriminating between temporal sequences, and we were able to actually pin down the exact
- mechanism to the activation of a specific ion channel called an NMDA receptor. This,
- together with the work of many other colleagues, painted a picture of neurons as really powerful
- computation devices that can generate a variety of input/output functions. That means performing
- a bunch of different computations just depending on the type of ion channels that they express,
- and also the spatial-temporal pattern of activation of their synapses.
- So this is this is how a neuron works. At least, I think so. So now that you know this, how do these
- neurons how these neurons get together and work together to compute escape? This is what John was
- alluding to in his introduction, and it's really spot on. We know a lot about how neurons work,
- but what we need to know now is how are these properties actually used to compute anything
- useful in the living brain? How do they serve behaviour? I'm going to walk you through this
- and I'm going to tell you two stories, one story about the decision to escape - how
- do mice decide to escape depending on what's going on in the world - and then a second one
- on mechanisms of how mice navigate to shelter. This first project was led by Dominic Evans and
- Vanessa Stempel, and for all the experiments we do that I'm going to show you today,
- and for most of the experiments that go on in the lab, we take mice that we put in an arena,
- which is a round table about this high. It has to be high enough so that they don't jump out.
- These mice have never been in this arena. The arena has a little hut at one end that
- mice find through their natural exploration. They get in there and actually quite like it,
- and from there they start making outings to explore the rest of the environment. In some
- of these outings we present them with threats, and John already mentioned one threat that we
- like to use is a shadow that that comes from above that mimics maybe a predator, maybe an object in
- collision course, but that mice and actually all animals, including humans, innately react
- to. So these animals have never seen these stimuli but they're going to react to it.
- This is an example. This mouse is going to see one of these looming stimuli, and then turns around,
- goes to the shelter. That's the behaviour we want to understand. It's pretty simple. This
- is the behaviour rendered from above. This blob there is the mouse, and then it goes.
- All right. So we want to understand how animals come to the decision of escaping or not. So the
- first thing we did was a behavioural experiment where we varied the contrast of this of this
- threat stimulus. This is kind of the equivalent of a mouse having to figure out if there's some
- bird of prey coming at it when the sky is really clear and it's sunny, versus when it's cloudy,
- it's rainy and they can't really see properly. So we're just going to vary the saliency of the
- threat and to see what happens to the decision. So I'm going to show you… This is the same mouse
- exposed to three different contrasts. And the videos are aligned to when the stimulus is
- presented. Okay, so mouse is coming out of the shelter, and when it gets to the end here it's
- going to see the stimulus. See what happens. So you can see that there's a very clear
- modulation of the behaviour. When the contrast is very high, the mouse turned around and ran
- straight to the shelter. When the contrast is very low it really took its time, and it had to
- see the thing five times in order to be convinced that, 'Yes, I should get out of here.' So if we
- plot this data so the probability of escaping goes up as the contrast increases - this is
- really funky. It's extremely nonlinear as well. Wow - and the reaction time goes down. So when
- the contrast is really high they react really quickly. So the next thing we did was to model
- this. Can we come up with a simple model that describes this? What we did was to actually
- take models that have been developed for human decision-making, to model human decision-making,
- and apply it to this behaviour. This type of model is called the drift diffusion model or diffusion
- to bound, where the gist of it is that there's a threshold for you to make the decision and you
- keep integrating evidence towards or against it. So you're sitting there thinking about, 'Okay,
- should I go or should I stay?' At some point if there's enough evidence then you
- cross the threshold. You go. If you never cross the threshold you never go. So the
- way we model is that we thought, okay, the animal is integrating some sort of
- threat and the threshold is the threshold for escaping. So when the contrast is really high,
- this threat variable rises very quickly and hits the threshold very quickly, so you always escape,
- and you escape very early on. If the contrast is low it will take some time to drift up.
- So it increases and decreases a bit and then increases, and if you see enough eventually you
- hit the threshold. Sometimes you don't get it and sometimes you do. When you do get it, you hit it
- late, so that's why the time to react is longer. So then we went into the brain to try and
- understand how the mouse brain might be implementing this type of model. We started
- in a place called the superior colliculus. As I mentioned before, it's in the midbrain. It's
- at the back of the brain. It's a region that's very well studied. It receives direct input from
- the eye, which is important for us because we're showing them a visual stimulus , and
- this input arrives to the superior colliculus via these cells, called retina ganglion cells
- that make synapses onto some of the neurons in the colliculus. Work from many colleagues
- has shown that there are neurons in the superior colliculus that are sensitive to looming stimuli,
- so they react very strongly to looming stimuli. So we thought this is a pretty
- good candidate for processing this type of threat. The second region we focused on is just below it,
- still in the midbrain which is called the periaqueductal grey. I might call it PG sometimes,
- and superior colliculus I might refer to as SC sometimes. This periaqueductal grey sends axons to
- the motor centres, so it activates motor commands, and it has long been known to be important for
- defensive behaviour. So if you electrically stimulate it you get a mouse to jump. Actually,
- interestingly, in humans, if you put a human in a scanner and you scare it - and the scare
- here is the threat of receiving a little electric shock that is kind of painful, so you don't want
- to have it - if the threat is really imminent, the periaqueductal grey gets activated. Also,
- if during a neurosurgery you activate the periaqueductal grey, the patients report a sense
- of dread, of being chased, like a panic attack. Also, patients with post-traumatic stress disorder
- have an overactivation of the periaqueductal grey. So this is clearly a region that is very
- important for defensive behaviours, and it has a clear homology between humans and mice.
- So we want to see what happens, what goes on in these two regions during escape, to do this we use
- a very nice technique developed by colleagues at Stanford where you can mount tiny microscopes on
- the head of the mice. So the microscopes are about the size of a coin. We're going to take advantage
- of the fact that when an action potential happens, so that large potential difference that sends
- information out, calcium gets into the cells. So what we can do is, we put a molecule that turns
- fluorescent when you see calcium. So we can just sit there, and whenever a cell lights up
- that means that the cell is active. So we point our microscope to the superior colliculus. We
- point the microscope in other animals to the periaqueductal grey, and we use a virus to
- genetically modify these cells and express a calcium indicator. Then we see what happens.
- So this is the activity profile in the superior colliculus. So when a threat comes the activity
- starts increasing and then animal escapes and activity in the colliculus peaks, more or less,
- during escape. Now, when we look at what's happening in the periaqueductal grey the
- profile is quite different. So during the parts of the threat nothing happens, and then you have
- a jump in activity just before the animal starts to run. When we look at trials or situations where
- the animal didn't escape for whatever reason, for example when the contrast is low, we see that the
- superior colliculus is actually still activated and the periaqueductal grey is dead silent.
- Nothing happens. So if we think back to our drift diffusion model, our decision-making model,
- what this suggested to us is that the activity in the superior colliculus might be representing
- the level of threat that is being thresholded, and the activity in the periaqueductal grey represents
- the result of this thresholding operation. This tells us whether you should go or not.
- So to further test this and to establish a causal link between these activity profiles and the
- behaviour of the mouse, we used another technique called optogenetics where we can play the same
- tricks, where we use a virus to genetically modify the cells, but now we're going to express a new
- molecule that makes the cells sensitive to light so that whenever we shine blue light we
- can activate these cells. The experiment we're going to do is, so we express this optogenetic
- tool in both of these regions, and then we're going to activate a small number of cells,
- a bigger number of cells, more and more and more, and see what happens to the behaviour. We can do
- this just by changing the intensity of the laser that we apply to the brain. So what I'm going to
- be plotting here is the speed of the mouse. These lines are different trials of the speed
- of the mouse during representation. If the speed is above this line here it means that the animal
- has escaped, and if it's not above this line it means that animal is just walking around.
- So we activate some cells in the colliculus. Nothing happens. We activate more cells,
- we get one escape. We activate more, we get more escapes. As you keep cranking up the stimulus,
- we get progressively more and more and more escapes. So if we plot this,
- the probability of escaping smoothly goes up as we activate more and more cells in the colliculus,
- and this smooth curve is really similar to the curve that you get when you actually deliver
- the real threat stimulus. Now let's do this for the periaqueductal grey. We activate some cells.
- Nothing happens. We activate more cells. Nothing happens. We activate the next level and boom,
- you always get escape, and then again always escape and always escape. So if we plot this, we
- get a really steep curve, meaning that we either don't get escape or we get escape all the time.
- First we learn one thing, which is activity in these cells is really causally related,
- is sufficient to cause the animal to escape. These animals are not seeing any threat.
- We are just hacking into the system to make the animal run. It also tells us, cements our view,
- that the activity in the superior colliculus might be the variable that is being thresholded,
- and that the activity in the PG represents the command to initiate escape. Because of it we
- thought, okay, maybe the thresholding, that line actually sits between the superior colliculus and
- the periaqueductal grey. So the first thing we then wanted to know was whether the two,
- these two, these two populations of cells are connected to each other. To do this,
- the first thing we did is use a really wonderful technique developed by several colleagues,
- including Marco Tripodi who is here, where we can use a virus to infect a group of neurons.
- Here we infected a group of neurons in the periaqueductal grey. We've turned them blue.
- Then what this virus is going to do is going to travel back up one synapse and one synapse only,
- and turn the other neurons in a different colour. So we're going to ask whether we get neurons that
- are in the superior colliculus, and indeed we do. This is one example. We can see that
- all the magenta neurons are the ones that give input to the blue neurons. This tells us that
- the superior colliculus gives a lot of input to the periaqueductal grey. So the next thing
- we did then was to record from these individual neurons. So we're going to record what's going on
- inside the cell. We're going to express our optogenetic tool in the superior colliculus
- so we can activate the colliculus while we're recording from the periaqueductal grey neuron.
- So this is illustrated here. We're going to shine light. We're going to activate the pink cell,
- and we're going to record what's happening in the blue cell, which is in the PG. This is what
- happens. The first thing that this tells us is that, okay, yes, they are connected,
- but this is a tiny, really crappy response. The reason why this response is very small
- is because the synapses have a very low release probability. So remember what I told you. Most
- of the times when action potential comes you don't get any release, and these synapses have
- a really low release probability. Now when you activate the synapse connection a ton of times
- you can overcome, just because you have a lot of them. You can generate a lot of activity
- in the downstream neuron, and we think this is pretty cool because this is the equivalent of,
- seeing what happens in these neurons, the equivalent of low contrast being the single input.
- So if the contrast is really low, nothing much is happening because the probability
- is really low. In order to actually get a lot of input and get an animal to escape, you have
- to have a tonne of activity in the superior colliculus that you only get when you have
- really high-contrast stimulus. This is basically a biophysical implementation of this thresholding
- operation that is done at the synaptic level. So the final thing we wanted to do here for
- this project was to test whether actually this pathway is really necessary for the behaviour,
- because that's our model. So the prediction is, if we block the synaptic connection the animal should
- not escape. So in order to do this we use yet another trick where this time, instead of using
- light to change the activity of the neurons, we use the tool. It's called a chemogenetic tool,
- where we express another molecule in the pink neurons, in the superior colliculus neurons,
- that makes neurons sensitive to a drug. So when they see the drug the neurons get silenced.
- So here we're just going to check if this tool works. So we're recording activity in the
- colliculus. We see these action potentials, and then this is what happens in the periaqueductal
- grey, so the cells are connected. Then we're going to deliver the drug. We're going to
- block the synaptic connection, and then we can see that the response is gone. So what
- happens to the behaviour? This is a mouse where we're doing this block. Now comes the stimulus,
- and this is what happens. So the first thing you should notice, and the most important thing,
- is this animal is not escaping. So this tells us that this synaptic connection is really necessary
- for escaping, but there's another interesting aspect here, which is that the animal actually
- reacted to the stimulus but reacted with a different type of defensive response.
- This animal is actually freezing, which is interesting because it suggests to us that
- actually the synaptic connection is necessary for evoking escape, but the threat information
- is actually being routed to another circuit that that can initiate defensive freezing,
- so that when we block escape it goes via another pathway that generates freezing. So the summary
- here is that we think that the core of the circuit for computing escape decisions is you
- have this layer of superior colliculus neurons that integrate threat information. We have the
- PG neurons that initiate or command the initiation of escape, and then we have this threshold that is
- implemented by a synaptic connection. The key here is a very low release probability that implements
- this threshold. Now, once you decided to escape, you should really know where you're going. In
- mice I've already shown you several times that what they do is, they escape to a shelter. It's
- a very sophisticated thing to do because shelters are by definition places where the attacker can't
- get to you, so they offer long-term survival. So we decided to investigate this process and
- actually understand the mechanisms by which mice navigate to shelter. This is work that
- was started by Ruben Vale. The first thing that Ruben did was to note that - I've showed you
- this already - you scare a mouse and it runs in a straight line to the shelter, which is
- a trivial finding in many ways. So regardless of whether you use a looming stimulus, a loud sound,
- you put the looming stimuli in different places, the animals always run in a straight line to the
- shelter, and this will become more important. What we wanted to understand next is, well,
- how do mice get to the shelter? What strategies do they use to navigate to the shelter and mice? Mice
- and people can navigate to places using a lot of different strategies. Perhaps one of the simplest
- things that these animals could do was to look to see where the shelter is and just run there.
- So we did a lot of experiments to test whether this is the case,
- but I'm going to show you the results of one which I think is very informative. We let the mouse know
- where the shelter is. It goes there, it escapes there. Then, one of the times when it comes out,
- we sneak up on the mouse and we move the shelter to a different place. Then we're going to see
- what happens. So the shelter is now moved. The mouse could just look at it, but we scare it and
- this is what it does. It goes to where the shelter used to be and kind of stays there hanging around,
- searching, and it does not ever go to the new one. So this is an extremely robust effect. When we do
- this rotation experiment, all mice will escape to where the shelter used to be and not where
- it is. We've done lots of other experiments like doing this in the dark, for example,
- where animals can do perfectly fine. The overall conclusion from this set of
- experiments is that we think that mice are forming a memory of where the spatial location is of the
- shelter, and use this memory to navigate there. Now the spatial world is not as simple as this.
- There's often lots of complications in order to get where you want to go. One of the complications
- we thought we'd introduce is a very simple one, but it's just a simple obstacle between the mouse
- and the shelter. This is work done by Philip Shamash, who then asked, 'Well, how do mice
- behave in under these conditions?' The first thing we found is that, with a little bit of experience,
- mice are pretty good at this. Instead of now turning and running to the shelter what they
- do is, they turn to the edge of the obstacle and get to the shelter, which is the most efficient
- way of getting to the shelter. So you target the edge, you turn, and then you go to the goal. So
- the idea is that there might be using this this edge as a subgoal. They know that they need to
- get there in order to reach the final goal. So what strategies do they use to navigate
- in this situation? So again we did several experiments, but we did one which is kind of
- conceptually similar to the one where we moved the shelter which is, we let animals learn that
- there's a barrier there and then we're going to lower the barrier. Then we're going to see
- what the animal does. The idea is that if the animal could see that the barrier is not there,
- they should just run straight. You can probably guess what happens. The threat's going to come and
- the mouse goes around as if the barrier was there, but it's not there anymore. Again, this is a very
- robust finding. So here we're plotting just the trajectories of the mouse. The blue lines is when
- the mouse targets the obstacle edge. There's lots of blue here. When the obstacle goes down there's
- still quite a lot of blue here. A lot of them still choose the obstacle edge, and this suggested
- to us again that they're using a memory-based process to know where the edge of the obstacle is.
- So overall, if we look carefully at both of these behaviours, what happens is that the very first
- thing the animals do is, they orient to where they have to go. So when there's a shelter,
- the first thing that the mouse will do is orient to the shelter and then run. This is why the
- escape trajectories are all in straight lines because when they start running they're already
- facing where they have to go. This is a good strategy, because if you started running where
- you're facing you end up going around-about. So you turn in place and then you go. This is true
- also for when there's an obstacle. When there's when there's an obstacle, they target the edge
- and then they go. So the basic blueprint for escaping is turn and then go, so this suggested
- to us that mice are probably keeping constant track of a vector of where they have to go.
- The vector is defined by the angle between the heading direction of the mouse and the
- shelter - so, 'How much do I have to turn to in order to get to the shelter?' - and the distance.
- 'How much do I have to cover?' And an important observation here is that we never told these mice
- that they were going to get scared. When we do it, they immediately go to the shelter. So during
- exploration it means that there's a very strong drive for mice to know where safety is, an innate
- drive. They found it as a safe place and they will always keep track of that in case they get scared.
- This also makes the prediction that if we're going to look inside the mouse brain, we should
- see a representation of where the mouse thinks the shelter is during the during exploration.
- So we decided to see if that's the case. We know a lot about spatial representations
- in the brain. John O'Keefe got the Nobel Prize for finding that there are cells in the brain
- that represent where you are at any given time point. They're called place cells. We could have
- started there and tried to understand how these systems might be representing where shelter is,
- but to tease John a little bit, we ignored place cells for the time being,
- and actually instead we decided to start from the action side. Our reasoning was, 'All right,
- the first thing the mouse does is turn the head to where it has to go. So let's look at the place
- of the brain that makes the head turn, and then ask what makes the head turn to the right place?'
- So I've told you a lot about the superior colliculus, and everything that I told you
- about is in this middle part of the colliculus. The superior colliculus is better known actually
- for its role in orienting the head and orienting the eyes, and even orienting the body sometimes,
- and when the head moves in this direction it's carried out by this more lateral part of the
- calculus. There's a lot of work showing that. This is true for all vertebrates from lampreys
- to humans. So we thought, 'Okay, the superior colliculus is probably doing the head turn,
- so what tells the colliculus where to turn the head to? Now we know that it can't be sensory
- input because I've already shown you that this is a memory process, so it can't be input from
- the eye, for example. So we looked for places that could convey some sort of spatial memory
- into the colliculus, and based on anatomy we focused on this part called the retrosplenial
- cortex that I might refer to as RSP, that is known to be very rich in spatial representations and
- sends axons straight onto the superior colliculus. So this is work led by Ruben Vale and Dario
- Campagner. So we went then to look for spatial representations of where the shelter direction
- is while the mouse is exploring. So this time we really wanted to have very high spatial precision,
- so instead of using the head-mounted miniature microscopes we used a silicone probe that we can
- also stick on the brain and record the voltage directly with very high temporal precision. This
- is what it looks like if you kind of slice the mouse brain coronally like this. This is where the
- probe was. With one silicone probe we can record activity in the retrosplenial cortex, which is
- right above the colliculus, and in the colliculus itself. So we're going to ask whether there are
- neurons that have information about the shelter angle, which is the angular distance between
- where the mouse is heading and the shelter. We define here zero as the mouse is heading
- at the shelter. So indeed we find these cells. What I'm showing you here is the activity of a
- single neuron as a function of where it's facing. Each dot here represents one action potential.
- The point that I want you to take from this is that there's a lot of yellow dots. There's a lot
- of spikes or action potentials when a mouse is facing the shelter. This is represented here. So
- this neuron cares the most about… This neuron tells the mouse when it's actually facing the
- shelter. This is another way of representing this, when this arrow just tells you which direction
- does this neuron care the most about. Now, to really make sure that these neurons care about
- the shelter and not something else, like some something else behind the shelter, what we do is,
- we're now going to rotate the shelter. This time we're actually going to give time for the animal
- to learn where the shelter is, so after a while they actually figure out, 'Okay, the shelter has
- moved. I should start going there,' which we can confirm by seeing the mouse escape there. Then
- we're going to see what happens to these cells. What we find is that this firing profile rotates
- in space, but the neurons are still firing when the mouse is facing the shelter, except now,
- instead of facing east, they have to face north. So these neurons care about not north,
- west or east. They care about their relative position to where the shelter is. Now, our
- model here that we're testing is, we think that the cortex, the RSP, tells the colliculus where
- shelter is. So we're going to do an experiment where we're going to inactivate these cells,
- these neurons that project to the colliculus, and see what happens to the information there. We use
- the same type of chemogenetic strategy where we can use a drug to inactivate neurons. So we do
- this in a retrograde manner, which means that we we're going to only be inactivating the neurons
- in the cortex, these blue neurons here, that send axons to the superior colliculus. This is
- an example of a superior colliculus cell that is tuned again to pretty much where the shelter is.
- Now we're going to inactivate the blue neurons, and this is what happens. This this cell loses
- its tuning. So this is again a very robust effect. About 60 or 70 per cent of the neurons lose their
- tuning, and this tells us that the superior colliculus really does need to hear from the
- retrosplenial cortex where the shelter direction is. Now again, similar to the previous project,
- does this matter for the behaviour? So to test this we did again the same type of experiment.
- This time we're actually, instead of activating the whole neurons, we're going to deliver the
- drug right on top of the synapses. We can do using little cannulas. So we specifically inactivate the
- synapses, and we're going to see what happens to the mouse behaviour. So this is a little guy that
- has had the synaptic connection inactivated, and what hopefully you saw from this movie is that
- it reacted when the threat came, but instead of turning around and going to the shelter, it just
- started running in a pretty random direction. The reason why they do this is because they
- perform the wrong orientation movement. Instead of orienting to the shelter, they orient to some
- random place and start running, and then turn around and start running. Eventually they actually
- stop in the middle of the arena without reaching the shelter. If this was a mouse being chased by a
- predator, this would be the last mistake that this mouse does, so this tells us that this synaptic
- connection is really fundamental for the animal to perform this orienting movement and escape
- successfully. Now, the final thing we wanted to do is to know how information about the shelter
- direction is moved from the retrosplenial cortex to the superior colliculus. So again we looked at
- the connectivity first. Actually, I haven't told you about inhibitory neurons, but the brain has
- both excitatory and inhibitory neurons. Excitatory neurons excite. Inhibitory neurons inhibitory.
- We wanted to know whether the cortex is connected to either excitatory or inhibitory neurons in the
- superior colliculus, or perhaps both. So we use the same type of viral strategy to look at this,
- first in excitatory neurons. So we infect with a virus a bunch of excitatory neurons here, and then
- we see where the virus goes. It goes to a bunch of places including the retrosplenial cortex. We do
- the same thing for the inhibitory neurons and we see that the profile is really very, very similar.
- So this tells us that the cortex sends its accents and makes synaptic connections with both
- excitatory and inhibitory neurons. We then went and again recorded from these individual neurons.
- So we're going to record from the superior colliculus neurons while activating input from the
- retrosplenial cortex. This is what happens when we record from excitatory neurons. We confirm they're
- connected, and the connections are pretty small. We do the same for inhibitory neurons and actually
- see these connections are actually much stronger, and so it's much easier to actually drive the
- excitatory inhibitory neurons in the SC to fire action potentials than excitatory ones. Then in
- several more experiments we actually found out that not only the retrosplenial cortex
- sends axons to the SC, to the excitatory and inhibitory neurons. The inhibitory neurons
- inhibit the excitatory ones. So that is what we call the feedforward lateral inhibition.
- The reason why I'm bringing this up is because what this means is, actually,
- when you activate the retrosplenial cortex, the most what you do is actually inhibit the whole SC,
- because it's much easier to activate the superior colliculus neurons, the inhibitory neurons, and
- the inhibitory neurons shut down excitatory ones. This is a little bit puzzling because we're
- thinking that the retrosplenial cortex is actively putting information in the superior colliculus.
- What we're finding here is actually what it's doing is shutting down the whole thing. So
- how does this work? To try and shed some light onto this we did a bunch of modelling. We took
- all of our data, our experimental data, and we built models of this of this circuit trying to
- understand what model fits the data. The upshot of this is that what we think happens is that
- we have a layer of neurons in the retrosplenial cortex that is tuned to different directions, and
- it probably computes the directional tuning from things like place cells, for example, and other
- types of cells. Then these neurons send their axons to the excitatory neurons in the superior
- colliculus which just inherit their tuning. So when the mouse is facing in a certain direction
- you have this particular set of neurons activated, and when it turns the head you have another one.
- So you can go back and forth and the superior colliculus always knows where it's facing because
- the retrosplenial cortex is saying it's so. So what does inhibition do? What we think
- happens is that the inhibitory neurons have an inverse connectivity pattern - and we have
- some evidence for this - such that when the mouse is facing in a particular direction,
- what happens is that the inhibitory neurons inhibit all of the excitatory neurons except
- the one that's pointing in the right direction for the shelter. The result of this is that you
- completely clean the representation of shelter direction in the superior colliculus so that there
- is no doubt of where you have to go in case you have to. So I'm going to finish here. What have
- we learned? Well, we learned that instincts are innate behaviours that are encoded in the genome
- that set out a blueprint of actions and drives that is critical for survival. These actions
- and drives can be modified by experience to give the animal the best possible chance of
- adapting to the world they were born in. We've learned that neurons are very
- powerful computational units because of their properties and the properties of the synapses,
- and that we can put these neurons together to form very simple computational models such as
- the first one I talked about, where with only two layers you can build a little module for compute
- for computing instinctive decisions. This type of model can be completely hardwired. You can just
- encode this in the DNA, make these neurons, connect them in this way, and you've built a
- mouse that can make these type of decisions. I've also shown you that you can do this in make these
- more complicated networks that in this case, in the example I gave you, are important for
- keeping track of a particular location in space, and therefore they can represent an
- interface between learned and innate knowledge. I'm going to finish by thanking again all of
- the members of my lab. They're a really amazing group of people. I'm inspired by them every day.
- They're smart, they're creative, and they have the most important attribute that a scientists needs,
- which is a good sense of humour. So we have a lot of fun. I'm very grateful to the Sainsbury
- Wellcome Centre for the support, and everybody inside that institute for making it a really
- amazing place to work. Science is not cheap, and so I'm very grateful for the support that
- we receive from the Wellcome Trust, who have been supporting me since my PhD. The Gatsby Foundation
- have also been very long-term supporters. The European Research Council and the MRC
- Laboratory for Molecular Biology where I started my lab, and finally my family for all their love
- and support. I think my daughter is watching online with her cousins in Cambridge. Hi, I'm on
- YouTube. Don't eat too much pizza before bedtime. Thank you very much for coming tonight. Thank you.
- So thank you very much, Thiago, and congratulations to the committee for
- making a very good choice. Thiago has agreed to answer some questions. We have
- about a little under ten minutes, so I'll take questions first from the audience that's here,
- and then we'll see if there's any questions on the chat. So are there
- any questions? All the way in the back. There are microphones, and if I see any of the hands.
- Krzysztof Potempa from Braincures. It's a beautiful talk. I was wondering if you
- had any clues on what might be the molecular cues responsible for the structural stuff,
- in the structural circuits you've shown. The fear.
- The molecular cues for setting up the circuits? I do not have a clue. I do
- not know. No. So that's a very important and very interesting question, actually.
- It's actually spot on. So if you're going to build a system that that works in this way,
- you need to make sure that everything is wired properly and they go to the right places,
- to the right neurons with the right properties. We don't know. There might be developmental
- neurobiologist colleagues that will know, but we don't. Knowing that is really the
- key for understanding how you can wire a system up to develop this. Yes. Thank you.
- I guess the concept of concept of synaptic tagging, I guess, who you think would be the tag.
- So maybe. I mean, synaptic tagging normally is a mechanism that that we think is important
- for changing the weight of synapses which may well be important in this particular
- case for learning where the shelter is, for example. To establish the system,
- I guess you need to route the axons to the right places and make sure they're
- connected to the right dendrites with the right properties. That's a really interesting problem
- and also a very hard one, because you also want to know the function of the neurons afterwards.
- So let me take chairman's privilege and ask you a question. You've concentrated on the
- whole escape from threat circuit, and you very nicely laid out the components of it. Do you
- think this circuit can be used for other things like approach to reward, and if not
- the whole circuit - and I suppose we wouldn't want to think of the PAG as being involved in
- the function - but can you think of… I mean, are there some parts of it that can be repurposed?
- So it depends, I guess is the answer. So we did an experiment where we trained a mouse
- to go to a reward. So we play a sound and the mouse has to go and collect a reward,
- but we gave this animal ten seconds to get there. When we do this and we get rid of this pathway,
- the mouse is perfectly fine. When we look at what the mouse does, instead of doing this
- snap orienting and running, it starts running where it's going and it goes like this, right,
- and it takes its time there. For that, this pathway is not important. Now, what we think
- this pathway might be important for is maybe not necessarily just for escape, but when we have to
- get to the goal really short. So the way I think about it is, you have the hippocampus and the
- head direction cells doing these really complex computations. They put it there in retrosplenial
- cortex that puts this information in the superior colliculus, which will have limited capacity.
- Depending on what your context is, you keep track of the most relevant goal. So if you
- need to get there, you use this system to get there really quickly because you don't have
- to recompute the whole thing from scratch. So that I think can be repurposed for many
- other behaviours that need to be implemented really quickly. That's how I think about it.
- Good. Thank you. Other questions. Right here in the front. Scott.
- I'll just shout.
- Oh, you need a microphone.
- Thiago, there are so many variables possible [unclear words 0:58:24.3].
- I wonder if you had two shelters, and they were equidistant from where the animal is,
- but one time the animal had to go around the barrier to get to it and the other time it could
- go straight to it, the distance is the same, but maybe the effort required to turn isn't.
- All right, very good. That's a variant of that question that I've never heard. Very good. The
- answer is they would prefer… So the time is what would matter, we think. We've never done that
- exact experiment. It's a very good experiment. If the shelters are equidistant what matters a
- lot is where they're facing. They might be far away. They might have one close to their back,
- but if they're facing this way and there's one there, depending on the distance,
- they would prefer there. I think what matters is a mixture of effort and time. We're actually
- doing exactly that experiment, trying to come up with a model of what are the key variables
- that are important for the mouse to choose where it has to go. I think you get it right. Exactly,
- that mixture of effort or time that it takes to get there. Exactly.
- Hi. Can you hear me? That was amazing. I was just wondering. I was quite fascinated
- by the fact that when you turn off the circuit they no longer escape,
- but they start freezing. So there are alternative strategies,
- and clearly in your scenario there's really only one obvious strategy. Have you explored
- indecision or whether conflicting options… Is it a binary switch? Do you have to decide
- one and then stick to it, or can you look at what the circuits are that are choosing that?
- So that's a very good point. So when animals are faced with threat, there's actually a very
- well-established relationship between the imminence of the threat and the action they
- choose. If the threat is very imminent they will escape, but if the threat is at a perceived longer
- distance, they will actually freeze instead. Actually, when we do these experiments with
- low contrast the first thing that animals do is they freeze, and they freeze for a while,
- and then when another one comes eventually they run. So I think we have that decision there,
- that indecision that you're referring to already in this assay. So then the question is, how does
- it work? That's something that we've been trying to nail for a while. [?Yaris 1:00:46.8] here
- actually has been working on this for a long time. The model, the standard model, would be that the
- freezing… We know quite a lot about freezing, and it's supposed to be carried out by actually the
- more ventral parts of the PAG. So one model to solve this that we'd like to show or test - and
- we'd like it to be true - is that the superior colliculus sends axons to both the dorsal part
- that does escape and the ventral part that does freezing, and that the threshold for
- activating the freezing circuit is lower than for activating the other one. Maybe you could
- just have a higher release probability there, and that would actually implement that behaviour.
- It's going to be horrendously more complicated than that, but this would be a way of doing it.
- Congratulations, and thank you for allowing somebody like me to understand.
- Thank you.
- So what happened to the circuit when innate behaviour is ignored? For example,
- when an animal decides not to follow suit.
- Yes, that's a very good question actually. So work from Troy Margrie actually that we collaborated
- with has shown that these animals figure out that actually nothing bad is going to happen to them if
- you keep looming. If they stake out the loom and they stay, nothing bad happens. They do it again,
- nothing bad. They learn very quickly to actually suppress, and Troy has shown that
- actually they can suppress this for a very, very long time, for up to months. So the question is,
- how does it work? We don't know. I think what we know is that what we're looking
- at is - and Steve [?Lindsey 1:02:27.6] is here, he's doing these experiments together with Marcus
- Stephenson-Jones - we think that whatever happens there's a learning process that's
- going to change the activity in the superior colliculus, and it's going to make that part
- of the superior colliculus that cares about threat much less responsive to threat by inhibiting it.
- That's what we think is one of the key targets for modulating this instinctive behaviour,
- is just by acting on the threat detector. Of course you can act at many different levels.
- You could act at the synapse, you can act at the escape initiation. There seems to be
- a pathway there that's optimally poised to shut down and to decrease the sensitivity to threat.
- The final question.
- Thank you. My privilege.
- This is from my brother, by the way.
- So you give you the idea that these circuits are quite deeply conserved. Presumably there's
- animals that don't care about this behaviour as much, perhaps if you go up the food chain. Is
- that the case and, if so, how do you expect the circuits to change in those species?
- That's a very good point. So as you go up the food chain there's less and less predators.
- Humans are still scared, especially baby humans are still scared, by this looming stimulus,
- suggesting that this structure is still there. I think what happens there is that, I don't know,
- maybe the circuit has shrunk so that it's not important. What these higher species have is
- a really big cortex that has the ability to modulate all of the instinctive behaviours
- and probably act to shut down. The fact that, for example, an infant baby will react to this,
- and it takes much more for an adult to actually react, might have to do with the
- fact that you just learn to parse the sensory world in a more efficient way. That will come
- from the cortex that can just shut this down. I think that's one thing that definitely happens,
- but it's still possible that the circuit has evolved to be to be redundant.
- Well, very good. Thank you very much. It only remains for me to read the citation and to present
- Thiago with his prize. It's official. I have to read this. Doctor Thiago Branco is awarded the
- Francis Crick Medal and Lecture for 2023 for making fundamental advances in the molecular,
- cellular and circuit basis of neuronal computation, and for successfully linking
- these to animal decision behaviour. Thank you very much, Thiago. I've got something. All right.
- I've got a medal. Am I supposed to open this now? Okay. There's the medal.
- Thank you very much and good night.
- Thank you.
How do millions of connected neurons generate behaviour?
Francis Crick Medal and Lecture 2023 given by Professor Tiago Branco.
Professor Branco will discuss how his group is using mouse instinctive behaviours to answer this question. By recording and manipulating the activity of single neurons and their connections, the team is discovering the biological mechanisms behind instinctive decisions, such as when to escape from imminent threat. This work has uncovered molecular and cellular principles of how brains perform fundamental computations, laying important foundations for tackling psychiatric diseases.
About the Royal Society
91TV is a Fellowship of many of the world's most eminent scientists and is the oldest scientific academy in continuous existence.
/
Subscribe to our YouTube channel for exciting science videos and live events.
Find us on:
Bluesky:
Facebook:
Instagram:
LinkedIn:
TikTok: