[00:00:05]
>> Thank you so much time and thank you all of those coming in I mean I guess the the rain didn't stop you and the food helps I'm sure I'm looking forward to checking that out too but thank you for coming anyway. So before I get into navigation I mean I think the classic picture of the fly that many people have is it's sort of this reflex machine right things happen in the environment on the fly just react stood just response to it and so if there's one thing I'm hoping that you leave the talk with it's the notion that maybe there's more going on here than just that I mean reflexes are an incredibly wonderful part of insect behavior I mean they are remarkably fast they're robust and so on and so forth but insects have to do a lot more than just that and so hopefully you'll see that of the course of the dog so to start off with do flies even really do much navigation and so Michael riser who's a colleague of mine in Geneva had this collaborative study which Zuker a few years ago where they did what's the coolant of the rodent moderates water meets tasks and this is a task that I'm sure there's a few working rats know well rode into sport in a pool of milky white water it's you know maybe this large maybe a bit larger there's a hidden platform submerged under the water it can't see the platform you put the rodent in there over time the rodent kind of figures out where the platform is doesn't like to be in open water too much kind of swims over to it hangs out there and it uses visual cues to get there faster and faster over time to make a razor was wondering if he could do the same thing in flies and so of course he didn't put them in water that doesn't work so well but what he did do was he came up with this idea of having one cool spot in an otherwise hot floor OK So there's one spot that at a reasonable temperature this is super hot if you put a bunch of flies there this is what happens the Flies run around like crazy this is a speed up video so they're really running around like crazy eventually they find their way to the school spot which is going to be marked with this yellow dotted line most of them do anyway.

[00:02:05]
Spot moves but you'll note that the visual panorama moves with it so the question is will the Flies get faster and faster over time to get to this target area as they start to make the connection between the presence of this these landmarks and the particular spot that gives them safety so not all of them get it right I mean some of them clearly don't but you'll notice the next move I mean some flies to school directly to that spot I mean almost immediately they're going to the right spot using those cues and so they did a bunch of control experiment you know with just the thermal cues none of the visual cues and the flies cannot solve the task and they really need the visual cues in order to be able to do this and of course they did what's called the probe trial where there are actually nor term cues whatsoever but the entire floor is hard and they just ask where the flies end up where they go and they you know they have no term accused to help them and they tend to hang out on the 1st try the locations near where the cool spot is supposed to be sort of that does make sense essentially you use visual cues to get around find your way to see 50 flights tend to be able to do that in this task these are not flies that have been trained over weeks this is literally the 1st time they've seen the stars and in minutes they kind of get it.

[00:03:16]
So a more recent task where it's not really a task it's just a condition in a way that flies a Dane So if flies are just put in an empty arena like this they'll often wonder about the sides of the arena but if as in Michael Dickinson's experience I mean c'mon Michael dickens is the consent experiment you put a little droplet of food there something that the fly grabs onto and then kind of consumes and it's gone flies will initiate a very different sort of behavior they do this thing called a local search so they just wind around this area leave that spot search around come back to that original spot and go around and around again especially if it's a nutritive food like sugar or something that's what they do so it's as if they're searching but with a home base in there and what this gave them an opportunity to investigate was whether or not flies use what's called integration so the idea is if I knew that this was a spot I needed to come back to and let's say I didn't have visual cues would I be able to kind of wander off like this maybe make a turn keep track of how far I'm going come back in roughly with some security so that I don't smash into it stop here and I'm pretty close right so the way I did that was I kept track of all my movements I have sort of an internal representation of how far I'm walking how fast I'm walking I'm kind of able to keep track of my turns and I can come back to the same spot even though I'm not using visual cues turns out flies can do that so if that's the central spot what Michael Dickinson's sure is that flies can take these trajectories.

[00:04:41]
As they go further away from that central spot they change the angles at which they walk and the angles at which they come back so that they can come back to that original spot and they showed this in darkness and light and so on and so forth even without order cues there the flies can kind of find their way back to that spot and so there's a notion then that somewhere in the fly brain clearly it must keep track of its location not not like you know not location in 2 dimensional space here but still look a sion enough that it knows how far it's gotten away from something in which angles of talk around so that something that we call an internal representation the way you and I do it is using an internal representation or so it's thought and people have done experiments and rodents where they've recorded from these neurons called Head direction says and so if you're trying to keep track of angles Well it's helpful to kind of know have an orientation in your head of different angles in the space you're in and so flies tend to do this where not flies and so your audience tend to do this thing where they have these neurons in particular polemic regions and other parts of the brain as well that are very sensitive and they spike a lot so they fire a lot to particular directions in heading space so if they're pointed like this maybe one set of neurons is going another one another one in the queue and based on the heading direction that the animals in and so there's been a ton of work primarily theoretical to explore how you might generate an activity pattern that so localized in heading and so the kinds of models people have used are so-called Ring attractor models and there's again a long history of this from the sampling ski group on to many many others may be proposing a network of the sort but the general idea is that if you wanted to keep track of heading and you wanted a unique representation of having one representation that does your point in this direction versus this versus this and so on down the line well you could do better than I mean you couldn't do better than arrange neurons in what is logically and I state this again topology was a circle where each neuron is connected excited.

[00:06:41]
Certainly to the neurons near it so in other words if I put this direction I lightly excite the ones that are too into this direction this direction OK but I really inhibit the neurons that are tuned to the opposite direction because I don't want to confuse my head in representation of the one that's dramatically opposite you could see you know so essentially you have inhibition on all these units that have heading preferences in the opposite direction and excitedly coupling to the ones that are in the same direction OK.

[00:07:07]
So this is just an idea that these are these are kind of computational models beautiful models this is coloring a tractor but what we've been able to kind of investigate is well maybe the fly kind of has something like this and this is where the story is going to go and we're going to talk about roughly how this kind of thing works in the fly to some extent we're starting to have insights there are some others and then this particular brain region is the one we're going to focus on today which is where you find networks like this in the fly and so this is kind of your quick introduction to the parts of the fly brain that are relevant to the talk today so there's the visual system inputs from the visual system come into these central brain regions but I should see more struggle inputs go elsewhere and many of them go directly down to the motor areas and control lots of really cool reflexive behaviors as well as well as behaviors that we would call effects of what are also on full very quickly this region on the other hand it's where things Adelir and a little bit so there is currently connected circuits here things kind of unfold a little bit more slowly there and that's the region we're going to focus on primarily.

[00:08:12]
So this region is called the central complex and it is very central and increasingly clear that it was well named because it's pretty complex as well so this region is called the ellipsoid body this circular thing here this is called the proto cerebral bridge and I should say that there are many other sub regions within the central complex these aren't the only parts of the brain in the central complex that are others but for today we're going to focus primarily on these 2 this is something that you find across many filers it's and it's in lobsters is something called a central body it's in flies it's in dung beetles it's in Monarch butterflies and grasshoppers and locust.

[00:08:50]
So work in a bunch of these systems suggested this region might be involved in navigation and when I say navigation for a monarch butterfly we're talking about hundreds and hundreds of kilometers and not just something in a small space so each of these animals has a different kind of little thing that it navigate in nice that it navigates within and in this case it's not little it all again.

[00:09:10]
But it's shared across these animals and so one reason why we thought we could come in and maybe do something interesting here is because from the work of these animals it's clear that there are sort of sensory maps in these brain regions so there's basically a map of what's around you seems to be in there and so we thought well we could come in and maybe look at exactly how those maps work and the key technology that we had to develop in order to be able to study this properly was to put the flies in a situation where they were actually able to behave while we were recording from them and so this is decade ago now.

[00:09:45]
But what the kinds of separations we use are the flies on an air supported ball we call it a book or a ball after Eric Booker When the seventy's came up with this very nice idea and said it's an air supported ball the fly kind of can run on it and then we track the movements of the fly on the ball and we feed that back into enclosed loop into a visual arena so we can operate it in different ways open loop or closed loop but essentially imagine a virtual reality for the fly OK So one more is when it's walking a different mode might be it's flapping its wings or Simon does these kinds of experiments in Martir But basically the Flies flapping its wings and we track the wing beat and if the fly wing beats suggest to us that the fly intends to turn in this direction we move the world this way opposite direction we move in the opposite way and so all the control is yours so unlike some of you here who may be in the Mars or doing 2 dimensional tracking and virtual reality here we are doing one dimension in the fly and 2 dimensions in walking right so what if we focus on the 1st started to look in the ellipse side body at a particular class of neurons don't let the names bother you too much E.P.D. just means it goes from the body to the proto cerebral bridge to the sense of procedural bridge and another and another region called the galley but we won't talk about that yet so what do these neurons do each neuron has its dendrites primarily then right in one little slice of this donut shaped things so this is the ellipse body and each neuron.

[00:11:10]
Each different color this is a stochastic be labeled sample here so each color is a separate neuron and they sort of tile this donut OK Each one is in a slice and so what your Highness Seelig in the lab did was put these put G. cam which is a calcium indicator in these neurons and you'll see why we call them compass neurons because this is the population activity of this population each neuron again has arbors in just one slice of this but the sort of organize into this little cluster into this what we call a bump and so the activities localized and in this particular case when the fly is walking on the ball and it's looking at this visual scene so the visual scene I should say is that actor on the fly so it's not just a little square pattern.

[00:11:51]
In front of the fly and you see this bump of activity tracking the Flies orientation in that scene so again the scene is moving based on the Flies movement and hopefully you can see that the bump tracks the Flies movements OK So once again basically just to be clear this is not just a literal representation of what's on the visual display here right so this is not just a there's no topic map of what's there this really is something that represents something else and and we have good reason to believe that it is in fact the orientation of the fly.

[00:12:27]
And one reason we know that is we can take the neural activity use the fact that we know how these different neurons carve up this structure divided up into about that many little wedges and then compute from that a population vector average So this is just an old fashioned kind of way of doing things you compute the population vector average based on the activity profile in these different areas and you come up with a net vector and you call that kind of the population Victor average representation of orientation Well if we take that orientation estimate and we open up this.

[00:13:00]
Elliptical thing this circular thing into one through 16 like that so basically you're looking at $123.00 all the way down 16 but now we're going to track it over time what's the activity profile over time so you can see a plot where basically we're going to track this a long time if we could actually watch what happens over time what you see is that the population vector estimate this brown line really nicely tracks the orientation in that panorama OK So basically we could say that if the fly were using this kind of representation in order to keep track of its orientation it would do pretty darn well right so now let's put it in darkness how does a fly do when you take away its visual cues and just give it nothing to look at basically turns out the fly does just fine or of the structure does just fine so this shouldn't surprise you you already saw in the earlier slide I told you that in the Michael Dickinson R.C. flies were able to do this thing where they were able to come back to the original spot they were at in darkness thereby you should assume that they're keeping track of the orientations in darkness as well and so here's a representation then that works in light with visual cues works in darkness as well kind of like you and me being able to track how much we're turning you know we're going after opinion or doing something else.

[00:14:20]
Further what was kind of remarkable was not only do they do this if the fly just happens to be chilling out just standing there in darkness in future darkness its eyes are black and it really can't see anything it's just sitting there OK generally grooming but basically sitting there sometimes we have clearly prove that this thing can actually persist through all that 30 seconds 35 seconds just as long as the flap and to be standing there this representation persists so here's where really you should think hard about you know when you think about that insect in front of you that you're looking to squash I mean yes it can escape that that attempt to squash it but while it's sitting there there's a lot going on in there I mean it knows where it is right that's kind of amazing if you think of this tiny little brain creature and what it's capable of and so for us this is been kind of the basis of a bunch of studies and so this is this is older work but where are we going to go on with this is to ask how do you generally these kinds of representations because the challenge in other systems is not to sure that such representations exist like I said in the rodent people have recorded from Head direction cells now here you can see the complete population which is kind of cool but still fundamentally what we think the fly gives us an opportunity to do is to really get to the nitty gritty of how you create a representation like this and so that's what we've been doing for the last several years and so the 1st little bit I'm going to tell you about is how it works in darkness How do you move the bump around basically and so the key behind that we think is a recurrent loop with neurons that actually come back from this structure called the part of cerebral vision right it's a comeback Well thanks to Tanya Wolf's work where we know about these neuron populations is so there's the E.P. Gene urines that we thought at least at the time take input from the ellipsoid body and then convert it into I mean take it to the proto suitable bridge.

[00:16:07]
But there are a complementary set of neurons called the P E N neurons that we really latched onto as a gabby memories lab at Rockefeller and Jonathan Green in particular is then grad student but what we both latched onto was that there's a curiosity with these neurons that Tanya Wolfe discovered which is that these guys get their input in the same places that runs drop it however they come back with a shift OK so if you look at these 2 sides of the procedural bridge imagine handlebars of a bicycle the input from here goes to some column there these are very studio type morphologically the connections can be plastic and stuff but this morphology superceded by the P.N. neurons pick up the input there and bring it back but which shifted to one side if it's an E.P. Gene you're on that goes in that direction it comes back going to the other side OK so you've got this crisscrossing the current loop with the shift why would you have some.

[00:17:00]
These are each of these you can think of as a merry list like a compartment with axons and dendrites in there mixing up the bodies in the insect and the invertebrate systems in general tend to be away from all the sites of the actions of the actions kind of in these little compartments.

[00:17:15]
And so here is basically a little compartment where the outputs of these set of neurons interact with the inputs of the other the den rights of the others and this guy goes back but with a with an offset shift why would you do that but if you imagine continued imagine this as handlebars of a bicycle if whenever the bicycle turned one way or the other either this handle went up or that handle went up so it's a century it's like them turning one way we do this effect on the other way you do this if that were the case then this these neurons would potentially be providing angular velocity input to this structure and specifically maybe to these neurons that we time that say this goes up that means angle of the last input comes to this side and if the bumper originally here you could move the bump down this direction every time the flight on one way if this side went up which meant the angular velocity was the opposite Well then this thing would basically get input and the bump would kind of move down the other way does that make sense if you have started to bumpy or if I turn one way or the bump move down here turn the other way the month moves on there it's like a compass needle shifting around based on which direction the Animas moving much easier to just show you in an animation so Emily made this animation these are the new runs again they tiled the entire structure we're going to be focusing on these central ones that go to these columns in the bridge these are the P.N. runs getting their input from exactly those columns were coming back which shifts to the left and right to the clockwise and counterclockwise side we're now going to give the E.P. Jr on the bump of activity just like you saw in the in the videos earlier that propagates up to these parts of the proto city built bridge and so that you'll see a tiny tinge of blue here that represents the B.N. year on getting activated but every time the animal turns one way maybe this half of the bridge gets activated provides input here drags the compass needle down one way if on the other hand the flight turns in the other direction it would be this half getting activated.

[00:19:06]
And that would provide input neighboring to the to the original spot of the bump and it would keep moving if the animal keeps turning it keeps moving and the bum keeps moving around so that makes sense essentially you have a compass needle that is allowed to kind of move around based on the 3 current loop with the shift in a velocity dependent way that basically makes sense yes.

[00:19:26]
Right so how do we test this kind of conceptual scheme so the way that we sort of test it was decided well let's actually patch these neurons sort of. From these neurons in the fly while it's walking and so she took her electrodes attached these neurons these neurons and asked Well if this is true if this model is true we should see angular velocity coding and we should see also according for heading because I'm claiming this is input to these neurons comes from E.P. gene your arms and from some source of angular velocity and it should be unidirectional kind of in its tuning so when she recorded from these neurons she found there was indeed angular velocity tuning so she looked at you know whenever the animal moved one way versus the other and asked if there were more spikes then versus not and she quantified this into an inker of and so every time she passed on to neurons from one half of the proceeds will be to the runs here she go tuning curves like this so these are spikes more when the animal turns in this direction in the counter clockwise direction and she got tuning goes in the other direction when she patched neurons from the other half of the prophecy road bridge basically suggesting that the velocity is in the clockwise direction so you now have angular velocity tuning but you'd have noticed that it looks like there's a lot of variability at these spots where is that coming from where again they receive not just angular velocity input in the scheme but they also receive heading information so it's a common and Oriel thing conjunctive thing that gives rise to their tuning and so if you place this in a 2 dimensional tuning map so basically this is angular velocity this is heading Well they need a particular combination of both heading and angular velocity in to in order to fire maximally So if that if it's not the right heading they won't fire that much which is the blue you see over here OK So does this make sense essentially you need to be in the right headings space and the right angle of a lot of the space in order for these neurons to go so it looks like you have a representation then that basically provides you with this model the thing you need for this model and so every ruin your.

[00:21:26]
Rist came up with a nice little model for this which included just in the P N's just the same 2 populations have been talking about connected recurrently But with this offset but in order to get a bump a unique representation and not have this representation be spread out he also needed inhibition in the network something that was from a different study that he was importantly involved in as well and so these we part come from a certain population there are multiple sources of it but one population we thought with these in a between neurons you don't need to remember the name but just they have a provocative names I mean it's called Delta sevens because of peculiarities we can get into at lunch or something but basically there's another set of neurons that we thought provided inhibition and so with this model basically this is a ring attract a model he can get something that very closely mimics the kind of heading simulated heading looks very close to what the heading representation should actually look like when you feed it realistic velocities the statistics of actual turns the animal makes feeds that into this kind of model and we can kind of recapture a good number of those statistics just like you'd expect further if we were to say well is this stuff really that important and we kill that input as downturn or Evans or Dante did if you kill the input coming back that recurrent loop coming back we could show that at this is a genetic manipulation so if you silence this in the output of these neurons at high temperature you could show that the bump amplitude so the amplitude of this heading representation the compass thing just dropped as a consequence so it doesn't mean they're the only things that are responsible for this of this representation but clearly it's important so in principle of going to show you that there's this unique representation of heading in the circuit the heading is updated in darkness using angular velocity inputs and it persists in the absence of external input and I've just shown you well this model can kind of do all those things and the parts of those that I haven't shown you just trust me they can and so in principle one is tempted to put up a sign like this and it's a temptation that of course in situ.

[00:23:26]
Ations one must resist but in this case it turns out as well it's not over and it turns out well there's a bunch of little predictions that these things make that haven't been verified yet so one is you'd expect there to be excitedly connections from the Z.P.G. neurons up there excitedly connections from these P.N. young because that's what we assume inhibition from these Delta 7 year olds how do we go about testing this so we've started to really dig deep is the mechanistic part of this starting to get to our nice sequencing of these neuron types and asking are these neurons indeed excited to re an inhibitor in the ways that we expect and it turns out to some extent yes so these neurons do exist these broad inhibitory neurons that could potentially help make this happen they are in a bit dory there are cholinergic acid neurons still testing the other ones but it looks like that's not so bad that looks consistent with the model are they actually structurally connected in the way that we expect or for that we go to electron microscopy and so we're asking there well if you look at neurons that go from the ellipsoid body to the proto Siebel bridge you can sort of see these structures here they're not actually blue right I mean this is just my coloring for you to see it easier there's what it looks like an electron microscopy well but this is a collaboration with Davi Bach his lab and a whole bunch of undergrads who've been tracing neurons and so on and so forth and so what we saw there was in this in this stack if we were to trace neurons of this is the entire stack or not the entire stack slices from that stack and we can literally manually go in a software called Cat made that.

[00:24:57]
But Cardona and others developed we can basically go in and trace out these are input neurons that will come to later called ring you're on these blue ones but the purple ones are the neurons we've been talking about of the compass neurons and I'm just showing you one little piece this is the ellipsoid body I'm just showing you one piece and they go up to the produce we build bridge talk to the red neurons which are the.

[00:25:17]
Ones come back so they're synopses we can count this and as we can map it down to the nitty gritty level and ask are these neurons really connected in the way that we expect. And so it turns out the Delta 7 they really do talk to these neurons the DNS in the dark mutually and so on and so in principle we can verify some of these connections are these connections there well it turns out yes so the E.P. G.'s do talk to the P N's in this village the ellipsoid body there are connections between the P N's and the E.P. G.'s the Delta 7 is do indeed inhibit actually the in a bit both the P G's and the P N's all looks well with the world and then there are these minor inconveniences there's a whole bunch of minor inconveniences I'm going to tell you 11 of which is that these things aren't really your input and output that we have told you it's actually that there's back and forth chatter between these neurons even within this structure you'd say well why does that matter I mean it doesn't matter right because in some sense the recurrent loop isn't really needed in order to keep this activity going there may be chatter within these things so the compass the bump of activity may persist partly because of interactions even within this this has consequences for Ha stable the whole thing is how more bio this bump is going to be when you push it because if there's strong back and forth activity that keeps it there but how do you get the compass needle to move and so we're trying to explore the drift that you might have so what I mean by drift so if I am actually I start to drift out of view of the many forks I'm going to push myself back here so if I if I were to just keep track of my orientation or just you know my heading by just you know I shut my eyes and I just keep track of it my representation is it going to drift away and I'm going to I'm going to be wrong essentially in my estimate of if you were to ask me a few seconds later well that's sort of the problem the fly needs to solve you know if that bump drifts away it's essentially an incorrect estimate of its orientation and so we're trying to figure out do these kinds of connections these back and forth connections potentially help stabilize the bump in a particular place when the flood does not done so these are things that are ongoing I wouldn't say that we've by any chance finished anything there so I'm going to actually say so this is an ongoing thing that's still going but I'm going to close this part of the story and quickly tell you about another so I don't know how far I'll get through this but we'll see so so far.

[00:27:26]
I've told you about flies the compass in darkness but now we're going to see what happens in different visual scenes so I just entered this room a little while ago I already have a pretty clear sense of my heading in this room if you're outside you have a pretty clear sense of your heading there as well but you're using different cues in order to get your heading in outside than inside right how do you quickly map essentially a scene on to some representation of heading out to you map a scene onto a bump and it turns out the flies will actually manage just fine in this scene and that scene because we've tried it so we actually put these flies in virtual reality it's a much more spectrally impoverish situation we put them in but nonetheless the Flies do develop a stable bump that tracks their heading in this scene which is pretty scene from Genelia and this scene which is also from Deni Loubert is in the woods nearby OK And so if you can develop a stable representation into different vision environments or in a ton of visual environments clearly somehow you're able to kind of make the system plastic So how do you take a visual scene and map it to a bump that's what we want to ask we already know that if you you know from from our rig experiments from before the different flies even will have a different relationship of their scenes to the bump OK So this guy has as bump or this gal in this case because we do most of our experiment in female flies the bump is over here relative to the visual scene exact same visual scene and this other fly might have a bump over there in fact if you look across different flies the offset of this bump relative to the visual scene is different for different flies OK So in other words there's some flexibility there and the flexibility is such that individual flies tend to be pretty stable in these offsets like so like so like so but sometimes every now and then we saw evidence that individual flies would also have show signs of plasticity in this particular mapping between the scene and where the compass representation forms right so this is an arbitrary reference and then you're going to moving after that arbitrary reference but that's an arbitrary thing for real even within a single flight so yeah.

[00:29:33]
Yes. That is our assumption that the question is is there a parallel remapping of the motor side and we don't know that yet this is one of the things we're actively investigating but it's a totally irrelevant question if it's arbitrary here is is there a relationship to soon so we don't know yet so for the 1st part of the thing the visual part where does the visual input even come in well this particular visual pathway is coming into these structures into the central complex from regions like the.

[00:30:01]
Above the names don't matter the point is there's visual pathways coming in there and a multiple people have investigated it not just from my lab and my group but from other labs as well but in this particular case and sort of started looking at their visual inputs in pretty fine detail and more recently the sun in collaboration with the wonderful theorist colleague of mine engineer and Hermann Stead's lab they've sort of looked even more deeply at it and what it turns out to be is if you throw on white noise stimuli and just flash these little dots at the fly what you realise is that flies this particular set of neurons in the bud have a particular kind of feature detector like response Profiles In other words they respond to light and dark in particular patterns in different places different neurons have different places that they respond to further there's a complex relationship for those or SEPTA fields that are affected by things the other things in the environment not just the thing in the receptive field and they're affected by when things are presented so if you present for example a stimulus now like this and then another stimulus there you get a different response and if you show them both simultaneously OK So there's all these complexities to the visual input but I don't know when the level you can assume that basically in this region the inputs of the fly is getting is just like just for now just assume a simplification is literally like the position of something determines whether the neuron goes or not and so you're seeing different Lemaire lie again so these are little regions or denigrates and X. ons interact.

[00:31:27]
But this is a population of neurons and these guys are quiet now but as this dog gets lower you'll see these guys get activated to different neurons basically have different deceptive field properties again they're more complex than this but just for simplicity think of this as a mapping of the visual scene onto a population of neurons not get on to the bump right just a population of neurons that captures visual features in the fly surroundings.

[00:31:50]
So we think that's the input coming in and you know you get different flies have bunch of recognizable feature detectors that you can see in fly off the fly so where does the school well we know from electron microscopy these are the neurons you sort of saw earlier we've traced them and they've talked directly to the compass neurons and sons who Kim was basically interested in seeing well that communication is that where the flexibility might lie potentially and is is this how visual scenes get mapped onto a compass representation So just again just to clarify we're talking about each of these neurons sort of specializing for features in different parts of the visual surroundings and these purple neurons we're seeing just give you one bump right the population maps to one bump So somehow this scene is transformed into an abstract representation of my heading in this piece and so this idea has been suggested before Bruce McNaughton's lab again in theory work in this case conceptual work suggested visual inputs might come in to a bunch of neurons that would carry sort of land that would carry heading information and then they also suggested that there might be velocity neurons and move the bump around in darkness or those can ideas like this around and more recently as well.

[00:32:57]
As a lab in work that's battling to our own suggested that there might be plasticity in synapses again they were talking about good cells but still the general concept was the same Ultimately you want to be able to adjust things adjust this mapping so that when I've done this much the angular velocity part of the system is mapped well to the visual part of the system so that essentially the bump moves to the same place determined both by mangler velocity as well as by the visual scene movement Yeah that makes sense that's how you're going to have a consistent representation of the world in your head and so this is what's on some sort of look at and so the general idea was pretty simple it was like well let's put the fly in a simple word he's done it with more complex words as well where there's just a single straight and now let's say this is the population and the bump is over there right now because the bump as we know it can survive in darkness as well but not as a visual scene and maybe some particular ring neuron gets activated the ring neurons of these neurons are.

[00:33:50]
There so this new one gets activated and so the idea is well maybe because of the color activation of these you can use heavy and rules and see some association forms between the activity of these neurons and that one neuron OK now let's say that you move the scene and you say this bar is now over here a different during your on might get activated and it's actually all populations which will discover the simple theming for now and the bump moves to a different position in this in this ellipsoid body and so you'd say well OK maybe the association between these 2 this somehow being rule in operation there and maybe you create an association there and so over time as the fly explores the scene presumably you can map the world through this thing neurons eyes neuron populations eyes onto heading representation and so that was the general idea and so we thought well can we test this directly causally And so the way we thought sort of tested as we said well if we can opt to genetically induce a bump of activity in this part of the ellipsoid body can we then create an association between this part now and this set of thing Iran substantially can we arbitrarily create an association between a visual scene and heading representation OK And so what I'm going to show you is his experiments to do that but 1st you need to know how he does that so he uses an opportunistic region called C.S. crimson It's a channel results and very.

[00:35:11]
And so what he does is while doing 2 photon imaging he's basically stimulating a small set a little bit harder so he's running the 2 foot on a laser pass the whole thing imaging but he's stimulating one little piece a little bit harder and what that effect is is this is the bump originally but it moves there because he's opportunistically smashed it there and there's only one bump in the system so as soon as he creates a new bump the old bump just dies OK So essentially what you're seeing is when it's red is when he stimulates and then the bump sort of survives and is happy there after so as you do one more time in this case the fly is flying but it doesn't matter in this case.

[00:35:48]
Yeah so new bomb creatives and it turns out mutual inhibition is what keeps it there but that's OK we'll leave that aside for now so now here's the experiment initially there's a certain mapping which is in this case this is the where the visual bar is again the visual bar is moving all around the fly and it's mapped to this spot in the fly brain in the body what song Sue is going to do so this is the mapping right now this is the compass needle This is where the cue is sense who is going to create a new different mapping and so here he's opportunistically stimulating here to create an artificial bump and this is where the visual cue is and you'll see that the Cuninghame in the relationship is not different but he's going to hard core it in because he can specify where the bump needs to be and he can hard code in as well this is open loop now you can he's putting the visual stimulus in different positions and hard coding in a new association potentially and this is the relationship now it's a 90 degree separation and so if he now tests it back in closed loop again the fly now controls things and he looks at what this system is OK I'm running short on time a little bit so I'll go through this a little fast what he sees is that he can move the shift the bomb from its original thing on this side of the visuals bar to anti-clockwise counter-clockwise he can then shifted back so again he's not trying to change change in the opposite direction you get the basic cleaning protocol you're showing visual stimuli and smashing the bump into one particular please and now any tests it later.

[00:37:22]
It's again 90 degrees the other direction so basically they're a little bit more than 90 so again it started out here he was able to move it you know by forcing it to 90 degrees he ended up getting married 80 or so and then he goes in the other direction moves it maybe $110.00 if he does this experiment after experiment this is what the data looks like so this is how much he attempted to shift the bomb this offset from 0 to minus $360.00 plus $360.00 And this is where it's actually moved the diagonal tells you he's doing pretty well this is in contrast to where if it has no opportunistically agent and he just uses the exact same protocol of heating the flights at etc This is what he gets OK So basically you get nearly a flat line no matter what you attempt to shift it out so you get the basic idea essentially we can shift the bump which means there is plasticity in the system and the general idea seems to be a minute these are inhibitory neurons so we think actually that they weaken their connections every time they call active and what that gives rise to the simple learning rule kind of for have been individually have been you can kind of get this but in this case that I should you we kind of hammer is a system that we use off the genetics we really drive it hard can you see this even in natural behavior well so you can ask well if I put the fly in this world right now there's a single stripe in this world there's a bump tracking it.

[00:38:39]
You'll agree that the association is going to want the one I mean it's really locked on right now what if you expose the fly to a world with 2 bars that are ambiguous so it's basically like you know there's a bar there there's a bar there and the bump is really nicely locked you'd think that this based on this information you'd see the locking is one to one except except well at the same time you're potentially training on the other one too right if the training principles I just gave you are true then exposure to a scene where there's a bump in exactly the opposite thing should or should kind of also in called something and so we can ask did a ton of exposure to this like many minutes of exposure to this in this case actually just 3 or 4 minutes does it actually make it more likely that the bump actually hops to the other place the new offset and indeed it does so as you can see it here doesn't happen all the time it's like a fraction of flies that you see this for the amount of time we expose it but every now and then what you get is essentially a remapping based purely on visual experience OK so it starts in the police but then exposure to the environment in this case an ambiguous one basically creates another mapping to yeah.

[00:39:49]
So in this particular case we expected to be confused we haven't done behavioral tests that show that demonstrate that it's confused but we are doing experiments are asking about it later because there are experiments where we're doing learning in the fly and that gives us a clear goal that the fly has to go through and then we can ask is it more or less able to achieve that goal in the absence of the kind of compass that we have OK So essentially then what I've kind of shown you is our attempts to go from you know behaviors that we suggest would mean that the fly has internal representations of where it is in its surroundings stable to take multi sensory input potentially associated with this internal representation learning it cetera is where we're hoping to go but the end result we're hoping to get to is a mechanistic explanation they take these models of how these things might be done and ask if they really are done like that in the fly or if so in what precise ways are they done which allows us to examine properties of the system the stability and so on towards the 2 D. thing we've kind of started making some progress in.

[00:40:50]
The former grad student and now for a short while of course stuck in the lab Luckily for me has been doing these experiments where she can operate genetically tween these animals so the brightening of a head was basically we heat the animals as they call Close to one particular kind of landmark in their surroundings and we cool them when they go to another except the heating and cooling is done opera genetically so it's a heat receptor so they're not actually being heated but they feel the heat and so essentially she can she can do these experiments and train flies and I'll skip over that so that I can save time for Q. and A She can train flies on these protocols where she starts them off just to get their naive preferences between let's say a triangular conical shave landmark and a cylindrical one between them for 20 minutes in an environment where the cylinders are punished and then asked you know severely during the training protocol you see that they avoid the cylindrical parts of the world so there's a periodic world with many of these kinds of patterns and they all kind of go selectively to the the conical object.

[00:41:49]
The question is does it retain those preferences that it develops over time after the after the training and so you can see the switch here so there's a market preference for cylinders over corns that switches to us reasonable preference for corns over cylinders after after the training so this is kind of the difference in visits between the cones in the cylinders so where are we going then is something we can hopefully link these representations with also looking for more representations and just angular representations we think there's representations also for our how much how far you're walking or fast you're walking etc and so we're trying to link that now to to the navigation learn navigation and that's kind of a long term thing in the lab and when I say the lab these are the key people that I should particularly point you do you want to see league who is now group leader group leader Caesar and bon did the basic experiments to assure that the compass was even there Stephanie regular Did the electrophysiology expensive pretty challenging 100 of those D.V.R. experiments sung suited the opera genetics driven plasticity experiments as well as the one showing mutual inhibition all bunch of other people I need to thank but particularly point out Tania Wolf who's a fabulous anatomist doing tons of really beautiful work that drives a lot of lads not just mine and be collaborated a lot with and home instead of tourists in the lab not in the lab and so he has their own lab in addition we benefit from a ton of advice and discussions from others engineers so the fellowship extends out quite a bit too many of my colleagues Davi Bach and Michael riser I mention in particular but we benefit from a whole bunch of others as well and so with that I'll just close with a couple of statements so when Simon introduced me he stumbled through the mechanistic Cognitive Neuroscience which many people can look at me like What the heck is that anyway and so I'd be happy to talk about it later if you like but essentially it's a program where we're planning to use different systems many in that I mean right now many of you know we already have fish and rodents and flies and we're going to start with those systems maybe our.

[00:43:48]
There's But the general idea is to find unifying themes and principles for higher level behavior internal representations regarding network dynamics that maybe we can use theory to unify some of these different circuits motifs in different system maybe you identify the same circuit multis but they were differently in different systems so we will see but it's basically a theory experiment collaboration that maybe many of you here and enjoy as well but to specifically tackle questions that are a bit beyond like in the higher level cognitive domain I would encourage you to come visit some of you do visit Ginny learn the truth which is great but please do come visit there's lots of different things that will allow you to visit so please check them out and see us there thank you thank you.

[00:44:35]
Yeah. Yeah. It's a really good question we don't know it's a good question and it's one that we're curious about so basically does the amplitude of the bump and the fuzziness of the bomb represent something related to the uncertainty that the fly has about where it is that's kind of the thing you're asking right and so it is something we wondered we nor that that region of the fly brain the body also receives input from all kinds of other things like state like the state of you know attention attentiveness of light I mean attention for C. but basically something where how aroused is the fly how weak is it we know a good sleep input we know it gets said tired the information so it's entirely possible that there's a certain model ation of this whole system based on state behavioral state and internal state so we haven't explored that and that might well impact how certain the flies are much it cares even about knowing where it is.

[00:45:42]
Yes I'm. 20 seconds in a new environment maybe sometimes it takes a minute or something if you're switching it from a strongly and corded one another maybe it could take a minute to minute sometimes but it's often really quick like it just maps immediately not immediately but very quickly and so that raises questions of you know if a fly can map a scene onto a bump in tens of seconds what are the plasticity rules in operation what is the mechanism for that and we don't know yet but we assume it's not something that requires great structural changes because it's supposed to be flexible but we don't know what it is all whole I mean even if they have been in the dark there's already a preexisting relationship there but you'd expect it would die we haven't explicitly looked at that whether it's longer after the darkness of darkness we don't know your question back then.

[00:46:57]
That was instant I mean basically 3 minutes 3 minutes 3 minutes 3 minutes so it's literally 3 minutes of Paula that's enough 3 minutes is definitely enough and for the opportunity it's that's certainly enough but even in natural settings I mean if you just look at 2 different scenes like I showed you those 2 different scenes from Genelia So essentially if you take those 2 scenes it takes you a few minutes and then it's there so.

[00:47:22]
More questions. You know so we do them in our closed loop because I mean so that was the attempt was to see in the some bigger setting if you just let the fly do its thing. How ambiguous is it in terms of the representation and so it's closed loop the fly is just determining where it goes in that the assumption would be if it jumps in this way I mean we've seen that even in walking experiments where the flies walking in and Vironment where there's only 2 cues and they're basically ambiguous as one there and one behind and it's exactly symmetric every now and then the bump does jump and kind of switch so it's it's feels like something where if you're in a room and there's only 2 cues and after a while you lose your sense of where you are and you kind of confuse being in one point in one direction with the other rethink this happens naturally to us as well we would guess but certainly for the fly the representation does which naturally with no opportunity eggs in open loop to straight close you more questions.

[00:48:29]
Thank you.