 και παρακολουθούν ότι είμαστε αρκετές. Γεια σας, καλύτερα και καλύτερα για έναν άλλο σημεριό της Σεμίναρς Σεμίναρς μας, όπως πάνω από την ΕΕΜΑ. Είμαι Γιώργος Καφαγιζ και είμαι πρόσφυρος της Γραδιούς από το Σόλλιερς Σλαβ και πρόσφυρος της Γραδιούς από τον Σόλλιερς Σαμίναρς. Και όπως είμαι your host για σήμερα, θα ήθελα να ξεκινήσω από πρόσφυρος της Γραδιούς και τον Πάνος Μποζέλος για να παίρνουμε την εξανότητα της Σεμίναρς Σεμίναρς για την ΕΕΜΑ και μεγαλύτερη Σεμίναρς Σεμίναρς. Είμαστε αρκετές, εξανότητας να πάμε back to the reason we all gathered here for today και να παίρνουμε τον εξανότητα μας από την Γιάννιλια's research campus, Σε 2010, καλύτερης, εξανότητας της Γιάννιλια's research campus, και σε κάποιο μήνες από σήμερα θα είσαι ένας οικονομικογραμμιστικός πρόσφυρος στην ΕΕΜΑΕ. Είμαστε ανοίγησης σε εξανότητας στον ΕΕΜΑ, θέλουν να δοξεθούν πρόσφυρος τεχνοί εξανότητας για να εξατασχεθούν τα εξανότητα και τα εξανότητα για την εξανότητα της Γραδιούς, και σήμερα θα δεχνήσουμε τα τέτοια, και εσύ εξανότητας σε εξανότητα, στο σημερινότητα της Γιάννιλιας και στον ΕΕΜΑ, και στον ΕΕΜΑ. Φυσικά, για να δούμε από μου, επειδή είμαι καλύτερα στον ΕΕΜΑ's δόξα. Σε πρόσφυρος, η στιγμή είναι εξανότητα όλοι οι εξανότητας. Ευχαριστώ πολύ για τη δύο που έγινε. Είναι καλύτερη να έδωσαι αυτό με εξάγησης. Λίγο, αν ξεχνάς τα τέτοια μου, δεν θα ανοίγουν ότι θα αρχίσω να πιστωθώ να μιλήσω το σαμμινό του δυρμουλού του Τ. Λεπθενού, που πήγε το 1959 και αφήνουν αυτό το αγόδιο της πρόγραμμας, το που το πιστωθώ σ' το πρόγραμμα της σπίδας. Η ιδια που έφυγα να το προσέβησαν στην φορία, ήταν να το πιστωθώ, το αγόδιο της σπίδας, ήταν δυνατή να κάνει εξηγία, και πατάς όλα αυτά's relevant information onto the central brain, as opposed to for example just taking a picture like a camera and passing all of the unfiltered pixels of light and dark onto the process later. In particular, if you read the paper in which they record from different optical fibers in the frog, they were very excited to find one particular type of neuron that they called a bug detector, as it seemed particularly sensitive to small convex moving shapes. To me, that really was one of the sort of first demonstrations of feature detectors and in fact just neuro-ethologically relevant feature detectors found in the brain and has really inspired some of the ways we think about approaching how animals are using visual information and processing that to guide behavior. And so what I'm going to tell you about today is what the lab has been working on along with some really close collaborators over the last five years or so, trying to build up a picture of what the eye might tell the brain and then trying to go beyond that to understand how these features distilled by the eye but then be combined and used to actually guide behavior. And to do this, we wanted to use a model organism that maybe was a little bit more compact than something like the frog or even something like the mouse where there's also been great strides made and understanding the encoding of retinal ganglion cells. And we in fact turned to the object of this frog's attention which is the fly. So one question we might ask is if the frog has a fly detector, does the fly in fact have a frog detector? And so the fly is a really nice, model organism to study, not just because of the genetic tools available to use and to manipulate individual neuronal cell types in the brain but because in fact there aren't that many neurons we have to look at something on the order of 300,000 across the whole central nervous system. As I mentioned, we can access these genetic tools one at a time almost for, in theory, any cell type we want to get to. And then the new piece that we've been able to add to this arsenal as a field is a connectome that is a diagram telling us how all of the neurons in certain parts of the brain and soon to be the whole nervous system are connected to each other. And so the question is how do we use all of these tools to understand this whole processing stream all the way from visual to motor. And so my lab has really been guided by an anatomical approach inspired by the sort of perspective of Lebson and others. And that is to try to look at bottleneck populations in this whole path from light to light movement to understand across those bottleneck populations what the representation of information is. And there's two in particular we focused on. One where information leaves the optic lobe of the fly and that is the area here in the eye. It has about four different types of information about four different neuro pill, four or five different neuro pill that process information. And then the other is where information leaves the central brain and goes down into the ventral nerve cord carried by a set of descending neurons. And there's only about actually we've revised that number upwards now, maybe 1300 total descending neurons from both sides in the fly. And so what I'm gonna tell you about today is trying to understand something about how visual information flows through the brain into the nerve cord. So here is a sort of cartoon of the fly's brain with some of the different relevant neuro pill mapped. And from the information we have now this will surely be refined and revised as we elaborate the connectome. It looks like there are about three major visual pathways coming out of the fly's eye. Carrying information across about 100 different cell types. And so the first of these is information going from the optic lobes into these kind of ventral lateral neuro pills here on the side of the fly's central brain. The second are some pathways that go from a little bit earlier in the optic lobe all the way to an area right in the middle of the fly's brain called the central complex through several relays here. And then finally there are some pathways that take visual information again, early visual information into an area called the mushroom body. And the central complex and the mushroom body of the fly I've been well studied actually in other insects as well. And they're known to play large roles in for example associative memory in the case of the mushroom body or navigation in the case of the central complex. The central complex actually has a set of cell types that seem to represent the heading direction of the fly in its environment. But what are these ventral lateral neuro pills for? Well, the hypothesis that I think has built up over the decades is that this area might be some kind of sensory motor switchboard area. This is where visual information comes in and has some very direct connections to descending neurons in order to more rapidly control some of the fly's behavior. So we're gonna explore some of those hypotheses today. Well, one of the things we can do once we have a connectivity graph of how neurons are connected to each other is we can just kind of take a very coarse look at what the motifs of that connectivity are. So if we just look at how three neurons might be connected to each other. These are sort of the 15 different ways in which they can hook up. And very roughly, again, this is kind of a very coarse analysis, you can imagine there are some motifs that are sort of more feed forward in nature, some that actually involve feedback from the neurons back to the ones that gave them input and then others which are more in this kind of loop structure are more recurrent. And so if we look at these three primary visual pathways, what we can see is that these ventral lateral neuro pills tend to have more of these feed forward structures in them and fewer of the recurrent or feedback pathways, especially in comparison to the mushroom body and the central complex. And you can see that again here in this pie chart in terms of what percentage of the different triplet motifs are represented by each of these sort of flavor of connectivity. And so this is just to give you a rough idea that this BLNP area at large might be somewhere where we could actually gain a lot of knowledge with a kind of feed forward analysis which is what I'm gonna largely talk about today. Now, in particular, a type of visual projection neuron that projects from the optic lobe into this BLNP are these so-called LC or lobular columnar neurons. And these are populations of anywhere from say 50 to 200 neurons that have these, each individual neuron is sort of has narrow dendrites, a small visual field and these individual dendritic visual fields tile retinotopic space here in the optic lobe. But then their axons as they project into the central brain bundle and form these clusters are optic glomerally. So each of these different colors here is representing the axon bundle of a different LC type. There's on the order of something like 20 LC types that have this kind of configuration. And with our collaborators, Michael Reiser and Jerry Rubin along with Alea Shnirn, Nathan Kopopi who is a postdoc in my and Michael's lab set out to examine the visual response properties of these LC neurons in more detail but across as many of them as he could get good driver lines to record from which is about half of them, about 10. And so again, this is what the structure of these LC neurons look like. Here are individual members of each of these different 10 types that he is going to examine. One thing to note is that these different LC neurons have different ramifications in different layers within this lobula neuro pill in the optic lobe indicating that they could be getting different kinds of the process visual information. And again, they output to these distinct glomerally. Okay, well roughly what he found is very similar to what they found in the frog which is that different ones of these LC types seem to be detecting specific different visual features. So here are some examples. There's a type called LPLC2 which is very sensitive to looming or expanding motion. And in fact, it's very specific just for looming. There's another type LC15 which seems to be very especially tuned to long narrow moving bars. So it's sort of a line detector. And a third type LC18 which looks like it's a point detector, a small object detector. And its response to points even potentially smaller than the visual resolution of the eye moving across its receptive field. In other words, one could interpret that in fact the fly has its own fly detector. I won't say more about this today but we could discuss the ecological implications of some of these. So okay, here are three specific examples but as I said Nathan did a whole survey of 10 of these LC types. And I'll cover this a little bit quickly because I think my colleague Michael Reiser actually went through some of this material in this forum about a year ago when it was just hot off the presses. But the way Nathan was able to get very precise data for the visual responses of these individual LC types was he found a place where he could isolate individual members of this population. That is the start of the axon as it's leaving the lobula and entering into the central brain. And so you can see here three different ROIs for three different individual cells of this particular cell type LC18. And what Nathan did is he would image from these neurons using the GCAN 6F and for every ROI, for every neuron that he was focused on he first did a kind of set of pre-stimuli where he found with a receptive field that that neuron was so that he could present all future stimuli as receptive field-centered visual presentations. And then he ran a battery of say 100 different kinds of visual stimuli trying to cover all the different basic kinds of motion, dark looming, bright looming, moving objects of all different sizes, moving bars and edges, wide field grading motion, et cetera. And what he found was that individual cell types in fact had a very unique tuning. So no two of these had identical responses across this whole battery of stimuli. However, to a sort of first approximation when we ran a sort of PCA analysis the first principal component really separated these cell types into two large clusters. One that was responsive to small objects and one that was responsive to looming. And so we were actually quite surprised to see that for the 10 of these LC types that we surveyed nearly half of them were responsive to looming. That's sort of saying that, you know, the half of the retinal ganglion cells in the frog might be fly detectors, right, at a first-order approximation. But as you'll see later, there's a reason for this which is that in fact, these are not all conveying exactly the same kind of looming information. Okay, the next thing Nathan observed was that not only do these cell types segregate into these clusters by response properties across this battery of visual stimuli, but in fact, the clusters physically segregate. That is the small object sensitive LC types tended to be in this kind of more dorsal cluster and the looming ones in this more ventral cluster. And so what that indicated to us is that perhaps feature detectors that are detecting somewhat similar features are more likely to be integrated downstream by downstream neurons that would want to integrate from spatially nearby partners. And so that's what Nathan looked at next. He looked at individual pairs of these different types of LC neuron, these visual projection neurons. This is what we can do in the fly that is very difficult to do and maybe larger brains without a connectome. We simply ask the graph of the connectome what is downstream of these pairs. And so what he found by looking at the downstream partners is that in fact, there were several clusters of particular LC types that tended to supply information to a downstream partner with other particular ones. So how do we read this graph? These are the different LC types. And here he was able to expand his connectome analysis beyond the time that he had surveyed with his imaging system. And so each square here represents a pair of these different LCs. And the color of the square is telling you basically on average how many synapses that particular pair will supply to downstream partners. And so where it's dark, those two LCs don't have any downstream partners in common and where it's light, they tend to have a lot of synapses they're giving in common to downstream partners. And you can see there's sort of roughly four clusters here and some of the stronger clusters were ones as expected where these are all the looming type LCs and they seem to be clustering together. So looming responsive visual projection neurons or providing information downstream with other looming sensitive visual projection neurons. Where are those postsynaptic neurons that are getting information from these visual projection neurons? Where are they going? Well, largely they seem to be staying within these ventral lateral neuro pills. So this is consistent with that kind of general information flow I showed you at the beginning in that for these ventral lateral neuro pills, by and large, they're not all sending projections then more deep into the central brain, processing is happening here within those ventral lateral neuro pills and then largely descending down into the ventral nerve cord. Okay, well, let me make this a little bit more concrete for you and reach back to a paper we published several years ago now but I think it illustrates how this information is being integrated. So I'll give you a particular example. Let's take LPLC2 and LC4 which are two of these looming sensitive projection neurons. They both synapse on a particular descending neuron called the giant fiber. And we actually did this study early before some of the automated connectome analyses could just give us a graph and so we actually had to go back to the original electron microscopy data from which these connectomes are derived and we manually traced the giant fiber, LC2 and LC4 neurons as well as all the other neurons providing input to the giant fiber but we really tried to get a complete view of this one sort of synaptic relay. So here what you're seeing are you're seeing profiles with individual neurons in the EM data set and then what you're going to see superimposed are the tracing. So by connecting those profiles in depth you can reconstruct whole neurons and so here in pink the giant fiber neuron is being reconstructed and in cyan and yellow the two visual projection LC types. And so what we learned from this is not just the synapse count coming from this data but also by looking at all of the inputs to the giant fiber we could know conclusively that in fact LC4 and LPLC2 are the only visual inputs direct from the optic lobe providing significant synaptic information to the giant fiber and so that's going to be very important in the next step where we then try to ask okay, so now we know the giant fiber is one of these post-synaptic neurons that's integrating a particular pair of looming LCs. Why is it integrating two and what is the particular information content of those two visual channels? And so because we know that the giant fiber is basically just getting it to direct the visual inputs of these two we could do the following experiment. We can do a wholesale patch plan recording from the giant fiber and that is guided by the fact that we have these cell-type specific genetic lines that let us target the giant fibers so much specifically. And so what you're seeing here is a head-fixed fly we stick our electrode in the back of the head and we target the fluorescing cell body and then this is a reporting of the giant fiber's membrane potential and we're going to show this fly a looming visual stimulus at the same time and so our data looks something like this in this case when the giant fiber fires a single spike that is when the fly takes off and would have launched itself off the ground for not head-fixed. Okay, well so this is the experiment we can now do. We can now silence LPLC2 we do this by again using genetic tools to express an inwardly rectifying potassium channel and LPLC2 this hyperpolarizes it effectively silencing its input or we can do the converse we can silence LC4 and in either case when one is silenced we're gonna record from the giant fiber and we're going to interpret our responses that we read in the giant fiber as being only through the active channel and not from the silenced channel. And so again long story short because this is work that's published so if you're interested in the details you're welcome to read further. What we found is that in fact LC4 input to the giant fiber seems to correlate almost exactly with the instantaneous angular velocity of the loom that is how fast the size of the looming object on the five retina is changing as expanding. And so you can see that here what these different colors are are these are responses from the giant fiber when only the LC4 channel was active responses to loom and the different colors are different speeds of loom. So it didn't matter whether the looming was slow or the looming was fast in all cases this response aligned very closely with the angular instantaneous angular velocity. Well if you try the same analysis with the same optic variable for the LPLC2 input when we silenced LC4 channel what you can see is for these different speeds of loom again represented in these different colors and I should say these are average traces of many many trials across many flies they don't overlay indicating that you don't get a faithful relay of the angular velocity of the loom through this LPLC2 channel and instead if you put on the x-axis just the instantaneous size of the looming object in this case it was a disc you see that now our different recordings our responses to different looming speeds do overlay in other words this is supporting evidence that LPLC2 is in fact representing to the giant fiber downstream the instantaneous disc size on the flyby. And these were optic variables we were excited to see encoded because they have been found in the brains of other animals called row cells or eta cells and for example logist or pigeon or cat many other types of animal have these particular types of looming calculation represented. And so we were able to show that simply by summing these row or eta computations we were able to recapitulate here you can see in pink our model laying over the black average trace of the giant fiber total response in the control fly So this was sufficient for us to feel like we had understood the complete primary visual input to the giant fiber and here you're seeing that it holds true for these four different speeds of looming which were what we tested. Okay, so that's telling us that one reason that you might want to have a lot of looming channels is in fact there are a lot of relevant different looming computations to do that you might want to assemble in different ways. In this case what's happening is that LC4 representing angular velocity and LPLC2 representing instantaneous looming that object size are combined in the giant fiber effectively creating the percept of an attacking predator. And when that happens when both of these channels are on so let me flip to this fast attack case when both of these channels are on then the giant fiber is driven to fire a single action potential and that is sufficient to activate in the fly a rapid take off in which it doesn't raise its wings it simply extends its legs and jumps off the ground. Now if I go back to this example and in the case where the looming stimulus is slow enough that it doesn't activate that velocity channel it only activates the size channel in that case the giant fiber is not brought to threshold doesn't necessarily fire an action potential and other descending pathways do get some of this information potentially processing it in a different way but they coordinate a different kind of take off in which the fly takes its time elevates its wings and is more coordinated in the air. And so I just like to show this video kind of illustrating how in the fly we really can go all the way from processing in the optic load to develop these features which I didn't talk at all about today there's plenty of people in the field working on that how these what these features are then that are being represented as a central brain and then how that's being carried further and actually integrated and so here's kind of the overview cartoon as it were a predator comes in and these two visual feature detectors for looming are activated in the optic lobe they come together to provide input to this giant fiber the giant fiber actually synapses directly on that yellow motor neuron which activates the jump muscle and the thorax of the fly that's what gets the legs to extend and then through an interneuron there was another set of motor neurons activated which drive the wings to press starting the flight motor and so that's how you get this kind of coordinated fast escape Okay, so I hope that gives you one sense of how in the fly we can take this fraud view of how features are represented and then go into individual examples trying to link all the way to behavior and ethologically relevant behavior of the fly and how they're actually being used So what I want to do is I want to spend the next part of the talk telling you about another specific example that looks at completely orthogonal information and how that might be being passed from the optic lobe to the central plane and then actually used by the fly That is, I just talked to you about specific spatio-temporal features that the fly might be interested to touch but what about object location? Where objects are in the world is incredibly important to animals for all sorts of reasons whether it's that they need to move towards them to find mates or food or whether it's that they want to move away from them in the case I just showed you of the fly evading the predator and so how is that spatial location actually being conveyed to these downstream circuits? Okay, let me tell you a little bit about why this might be a mystery or at least was when we started and that has to do with the retinotopy individual system So as I pointed out a little bit in passing when I introduced the lobular columnar neurons to you the lobular columnar neurons, which are these speech or detectors have these neurodendrites which look at these small portions of the visual field and they're arrayed within and slightly across these columns this columnar arrangement that's in the early visual system So the fly actually has a compound eye which means it's a composition of 7-800 individual lenses these lenses then this columnar arrangement is then passed down to subsequent neuro pill such that each of these different columns is processing information from a different part of visual space So that is all very well-organized and back here in the lobular as well as in the lobular plate which is pictured here in this nice picture from Axelborst the visual projection neurons, their dendrites are also aligned within these columns So that is our LC neurons are getting retinotopic information on the input side However, the work we did with Michael Reiser and Jerry Rubin and especially Alyosha Neuron who did this analysis back in 2016 looked at the output of these lobular columnar neurons in the central brain and what he found was that in fact there did not seem to be any retinotopy retained in these axon terminals in the central brain So that's illustrated here just for LP-LC2 You can see what Alyosha has done is he's colored different individual LP-LC2s that are retinotopathy arranged in this lobular structure in the optic lobe but then he's carried those colors forward into the central brain and you can see that they intermingle indicating there is no obvious retinotopic structure Now we will say that a few years later Michael Reiser's lab had a nice follow-up paper looking particularly at LC6 that did seem to maybe find some coarse retinotopy So there was an open question of how might spatial information be being conveyed Is there some organization here that we just weren't seeing at this sort of light level of analysis And so the work I'm going to talk to you about next has been a great collaboration between my lab and that of Larisa Persky It was originated by Martin Peake who was a really talented graduate student in the lab carried on in our lab with some electrophysiology by Jim Young Park and then really picked up by Mark Dombrovsky who is a postdoc with Larry at UCLA But in some ways the roots and the background of this particular approach we're going to take to look at this question came from my time as a graduate student in Michael Dickinson's lab where we were purely looking at the behavior of flies evading looming stimuli So I'm going to show you a video from that graduate work In this video, a looming stimulus is coming from the left-hand side of your screen And what you can see is this fly which you're seeing in two views because this is a prism so we can see the underside of the fly The fly is initially grooming As the disc expands on its retina it's going to stop that grooming and it's going to do a very precise motion of its middle legs here actually moving them towards the threat That means that later when it takes off its legs are already in position to push it away from the threat rapidly In other words, the early posture adjustment that the fly did was preparatory in order to actually control the direction that it escaped to allow it to escape away from the threat And I'd like to point out that this is a very sort of common behavioral algorithm Here is a man walking down the street He's about to encounter a very unexpected looming stimulus in the form of a car coming up on the sidewalk nearly running from over And what you'll see is that he actually performs exactly the same behavioral algorithm It's going to play back in slow motion in a second And you'll see that as this car expands on its retina he leans over such that his legs, his feet or the contact around are closer to the looming threat than his body And then he can push off away from the car So this is a sort of most postural adjustment is a very common way to control your direction How is the fly actually doing this? Well, in Michael's lab I looked at how the fly's center of mass was moving relative to its two middle legs which are the ones that are going to actually push off the ground and send it into the air And what we can see is that before the stimulus starts this average center of mass of the fly is basically just a little bit in the center and in front of those legs But if we look after the fly has been viewing the expanding looming stimulus for several hundred milliseconds right before it launches into the air to take off what you can see is that now the center of mass is distributed according to the direction that the fly is going to jump So that's what the color is here meaning So for example if the fly is going to jump to its left here in light green its center of mass is already over towards the left side of its stance And in fact, we could go into even more granular detail because we had these lovely high speed videos capturing this postural adjustment at 6000 frames per second And what we can see is in fact how the fly gets its center of mass to those ending locations which almost seem like target locations now depends on where its center of mass starts So this is a flow field in which the start of each black vector is telling you where the fly's center of mass started again relative to its two middle legs which are represented by this vertical axis here And you can see the black arrows point in all sorts of different directions but they sort of seem to be moving the fly towards this location where it is where it wants its center of mass if it's to take off away from this looming threat And we could do that for all of the different directions that the fly is getting threatened from And what you can see here in black I've just recapitulated those sort of target locations that we measured the fly moved its center of mass to before it jumped And you can sort of see generally at this point we didn't have as much data when I was doing this as a graduate student but in general these flow fields seem to point sort of towards these locations Okay, so the postural adjustment is how the fly is getting itself prepared to do this directional behavior And then just to reassure you that we can quantify the directional behavior and in fact not just in that one video but flies do in fact tend to escape away from the threat Here you're seeing each little green dot is a fly that took off This is now data that we were able to start capturing more numbers with because we have an automated system for showing flies looming stimuli The blue vector here is the looming stimulus direction and the red vector is sort of the mean escape direction And you can see that they're usually roughly opposing In fact we were able to just write out what the algorithm is in order to go from the blue vector to the red vector And that is if the fly were to simply be trying to jump directly away from the looming stimulus with some bias to also be jumping forward So we wait the forward direction a little bit and then that is what this is showing you here that completely recapitulates the actual direction that the fly goes Okay So that was a little bit of a digression back into sort of the background of this directional behavior the fly does to sort of preface our investigation of how the fly is doing this and how is this is actually getting this visual information about where the looming stimulus is coming from Well, I started by showing you that many of these different LC visual projection neurons are looming sensitive Here's a subset of four the five that we had found in our earlier study that responded looming and then I'd already told you the story about how two of these LPLC2 and LC4 actually signups directly on a descending neuron the giant fiber whose activation caused the take-up jump and the giant fiber is integrating looming information for these two Well, again, going back to our connectome it turns out that the giant fiber is not the only descending neuron in this ventral lateral neuropel region getting direct input from the looming responsive LC neurons In fact, there's a whole battery of them There's at least 9 or 10 and I'm showing you here the subset for which we have good genetic driver lines And what you can see here is that LC4 in particular and I'll remind you that LC4 was the one that was particularly responsive to this speed the velocity of the looming expansion That visual projection neuron in particular seems to connect to a lot of these descending neurons in some cases providing even a large percentage of that whole descending neuron's input that is what these pie charts are showing you here Here's a little glimpse at what those neurons actually look like So you can see they all actually have these are dendrites in the brain and they actually have fairly narrow dendrites And so what we did is we used intersectional genetic techniques in the fly to create driver lines that target individual ones of these descending neurons That was actually an entire project led by Shigihara Nemeke done about four years ago now And then we just simply activated each one of these descending neurons using those driver lines to see what would happen And so very first pass what you can see is just simply it does activation of individual ones of these descending neurons it does seem to drive some flies to take off not as much as activating the giant fibers that's what's shown here in pink giant fibers are DMP01 activation of the giant fiber almost always drives the fly to take off You can see the other ones are not quite as effective but if you start to combine them we just happen to have a couple driver lines that expressed in both DMP02 and DMP04 or 2,4 and 6 Now that seems to start to increase the take off rate a little bit And so I'll just take this opportunity for a slight aside for what I'm about to tell you which is that I'm going to tell you a slightly more detailed story about two of these DMP02 and DMP11 And I think often when we focus on individual ones of these neurons they have a particular effect it's easy to think that the fly's nervous system is structured like a set of labeled lines A looming detector connects to a drum detector and not the whole story But I think you should just keep this in the back of your mind that in fact we think that these particular neurons that we tend to focus on for the individual stories are actually just components of larger networks of population that are actually encoding information more broadly giving the animal even more flexibility And this is where we're going to scrap you know, in the future we've only sort of scratched the surface of this and so I won't talk more about populations but just kind of keep that in context Okay So maybe a little bit of drive to take off from activating these neurons but not a whole lot especially in the case of DMP02 But remember, we're looking for directional behavior and the directional takeoff of the fly is driven by these posture adjustments So how does activating these neurons affect the posture of the animal? First I'm going to show you DMP11 Here's a set of different individual flies at the bottom of your screen You're seeing the red light that goes on I should say, sorry, that we used our genetic driver lines to express a red-shifted channel redoxin such that we can determine a red light the cells are depolarized and so you can see when that red light is on here And what you're going to see is that this starting position of the fly is frozen in green and then the fly is movie is playing over time and so what you should see is a whole lot of abdomens of flies in green here indicating that most of these flies have actually shifted forward is what activating DMP11 did And there was a second one we found that had an effect on posture and that's DMP02 And so look what happens now when I activate DMP02 the flies again, the first frame was frozen in green and so what you're going to see remains behind as the flies move is actually their head In other words, the flies are shifting back Okay, let me quantify that for you a little bit We just tried to capture that with a single metric so we came up with this angle that we call the change into two-leg angle So just the angle between the center of mass and the two tarsal contact points at those middle legs If the legs are in front of the fly and the center of mass is behind if the fly is going to push backwards then that angle is less than 180° and if their back is greater than 180° So for these different descending neurons that we activated this is what that change in angle looks like over time And so what you can see is most of these lines aren't affecting this forward-backward posture but DMP11 drives the fly to move forward DNPO2 drives the fly to lean back and if we look at our combination lines with DNPO2 in them they also drive the fly to lean back And this is, we can see for imposing here so you can sort of see the difference compared to the control more directly Okay, so that's the posture adjustment Well, does this actually lead to the directional take-off? In fact, it does For DNP11 So this is now the take-off direction the fly, the fly is facing towards the top of the screen For dozens or a couple hundred flies depending on their genotype DNP11 activation causes the fly to shift forward and to jump forward For DNPO2 DNPO2 alone causes the fly to lean backwards but we need that co-activation with DNPO4 to actually get the fly to take off and in that case it takes off largely backwards But you can see here activation of DNPO4 alone doesn't drive the fly to jump backwards So we think that the way this is working is that PO2 is responsible for that posture shift but that you need that co-activation with some other DNs in order to actually get the fly to take off Okay, so these seem to be two of the descending pathways that are involved in the motor side of this directional behavior What about the inputs? So remember these particular descending neurons were chosen because they're ones that we know get synaptic input directly from LC4 However, LC4's optical marialess was not one that when we first looked at it had any really obvious retinotopy in it So how is it that information about the location of the stimulus which is what we hope is going to be hooked up to this motor program to jump forward or backward is actually being conveyed? While using our conical atomic data we looked at the synaptic connections between LC4 and these two different descending neurons And so what you're seeing here are individual LC4 neurons on the x-axis So there's 55 of them And they're ordered now in terms of the number of synapses they make onto DNPO2 That's the one that drives the fly to lean backwards And then the same identity of the LC4 neuron is retained But now in red we've plotted the number of synapses that particular LC4 neuron makes with DNP11 And so what you can immediately see is first of all individual LC neurons are not making the same number of synapses with DNPO2 There's a wide range all the way from nearly zero and then the case of onto DNPO2 all the way up to nearly 50 And the number of synapses they make onto the two different descending neurons are also very different and in fact they look like their opposite radians That is a particular LC4 neuron that makes very few synapses with DNPO2 makes a lot with DNP11 If we do the same kind of analysis with DNPO4s Now DNPO4s in red you can see you don't get exactly that same sharp opposing gradient So there seems to be something kind of special about PO2 and PO11 with regards to their connectivity to DNPO4 or sorry to LC4 So we wanted to look at this visualize this a little bit more directly And so what we're going to do is we're going to take these gradients of synaptic number and we're going to color each LC according to the number of synapses it makes onto a particular descending neuron And then the view we're going to take is we're going to go back to the optic lobe where the dendrites of those LC neurons are and we're going to look at whether there's any spatial arrangement of the dendrites according to the number of synapses they make onto the postsynaptic descending neuron Alright, so this is what this looks like In the case of DNPO2 we're now looking at a lateral view This is sort of straight on to the fly eye here of the lobula of the dendrites of LC4 and the lobula That's what's colored And remember the red color means a lot of synapses The blue color in this case means almost no synapses And so what you can see is that there's basically a hot spot in the interior portion of the lobula here in that LC4 neurons that have interior visual fields tend to have a lot more synapses with a descending neuron that makes the fly moving backwards And the converse is true for the LC4 neurons sorry, that synapse onto DNP11 which makes the fly jump forward They tend to get more input from the back of the eye Okay So that made a lot of sense and it was kind of exciting to see for us But of course there's still this lingering question A connectome is a great starting place but it really in the end tells you nothing except for maybe where to look because the question is Are these synaptic numbers actually synaptic weights? Does this actually get translated into the function of the physiology? And so we set out to try to measure that directly We have a set up which I showed you a little bit of a video of earlier in which the fly is head fixed and we stick an electric in the back of its head And this was Jinyoung Park who and Martin who are doing these recordings But what you can see I think illustrated nicely in this picture from Gabi and Michael Dickinson is that it's very difficult in this configuration to show the visual stimulus everywhere around the fly's head simply because of the block of this platform And so what we're going to do is we're going to show visual stimuli in as wide a range as we can but then we're going to use a model to see what part of the gradient we were in and whether we capitulated at and then to try to extrapolate what the functional response would be across the whole eye given the responses we measured across a more narrow portion of the eye And for that modeling part we were very grateful to team up with Art Zhao and Michael Reiser's lab who applied the same modeling technique that he had in a paper for Michael Reiser's lab in 2020 by my Morimoto So he reapplied this technique in which he could look at these other neurons coming from the medulla which he knows how they line up in an absolute visual space He could correspond that in our connectome in our EM data with the LC4 neuron dendrites that we were traced In order to be able to project onto visual space these are the outlines of the dendritic fields of every particular LC4 neuron And remember from the connectome we know for every one of these fields and now we know what it's looking we know how many synapses it makes onto the post-naptic descending neuron Okay, and then we're going to show a battery of looming stimuli going across this anterior-posterior axis and this square shows you the small portion of the actual visual field in which, given the constraints we were actually able to show visual stimuli and we're going to see if there's different responses Okay, so this is what our setup and our stimuli look like and here are some average traces from basically four different experiments One in which Jinyoung recorded from DNPO2 Oh, sorry, two different experiments One in which he recorded from DNPO2 but showed either anterior 32.5 degrees or more posterior visual stimuli and another in which he did the same for DNP11 So already you can see that these the red and the blue traces seem to have different heights depending on the location of the stimulus and we can summarize for you the results across many flies and many trials which is that if you look at the spike number or if you simply look at the area under this curve in both cases we seem to get the right trend that is DNPO2 was more responsive to these anterior looming stimuli and DNP11 was more responsive to the posterior looming stimuli Okay, so then to go forward with the modeling exercise that Art and Mecherizer's lab did Here are his predicted sort of representations of the synaptic radiance based on the locations of the actual LC4 dendritic field So dark is showing you where we predict the DNPO2 would be more responsive and light were to be less responsive and then here we're showing you for response of the model how it compares to the data So the models in solid and the data which is what I just showed you here the area under the curve is dashed and you can see that again across the inner portion of the visual field they seem to sort of roughly correspond and so then we're able to use the model to extrapolate then what if these weights are functional that would mean in terms of the response of these downstream neurons and sure enough matching the expectation for the gradients to be observed and running through this sort of physical spatial model of whether receptive where the visual fields are actually looking in space we can see that DNPO2 is predicted to respond much more strongly to anterior luminescimuli and DNP11 to posterior luminescimuli Okay, so we think that these synaptic number gradients are actually synaptic functional synaptic weight gradients and putting that together then for this particular story what we see are that we have two different descending neurons both getting input from LC4 which was responsible to looming velocity that they're filtering that information through these different spatial gradients which then are what is translating the location of the object effectively for the downstream neuron such that when you have a looming stimulus going through these filters you get out the appropriate behavior either jumping forward or jumping backwards and then in just the last couple minutes I just wanted to show you how Mark and Larry were able to extend this work asking the question well, is this actually a general phenomenon of this interface of what the eye is projecting to the brain through these LC neurons or is this very specific to this case escape forward or escape backwards is the tendency to think that that might be a very specialized case and so let me just show you some of what Mark was able to find he looked now at all 20 of these different LC types and he just simply asked if we use the connectome to look at their downstream partners do we see even connectivity between members of the given type with downstream partners or do we see preferential connectivity that is what I showed you with the DNPO2 and the DNPO11 an individual LC neuron is making more synapses one and fewer synapses with another and then he was going to visualize that by showing different clusters of how that differential connectivity might show up so what you're looking at here is you're looking at again lobular projections of the dendrites of individual LC types and they're colored now by clusters that tend to make the same number of synaptic connections with downstream partners so in the case of LPLC2 here these orange ones are all having similar profiles of projections on the downstream partners but they're distinct from these blue LPLC2 neurons and so this is simply to show that this kind of differential connectivity on the downstream partners is clearly a widespread phenomenon of these LCs it doesn't have to be there are a couple exceptions LC12 and LC17 which are sort of the two lowest ones here you can see are very mixed up so there isn't necessarily a reason a priori to think that you would have to get this kind of cluster differential of connectivity downstream but we looked further because this was indicative to us that in fact these gradients might be very common and basically anywhere Mark looked he could find more and more examples of cases where now you're again looking at the lobula and individual dots are showing synaptic weights of these individual LC neurons onto a particular downstream partner so in this case LPLC2 onto GF and so this is again a kind of gradient in which LPLC2 has stronger input to the giant fiber in the dorsal part of the lobula the dorsal LPLC2's do but the ventral LPLC2 neurons have much less and so here's examples across many of these different LC's of these kinds of spatial gradients being represented through the synaptic number onto downstream neurons and so we think this is a pretty general way in which the eye is communicating to downstream partners in the central brain where objects are located it's actually you can't necessarily see it in the retina topi but it's actually in the synaptic number and just to show you one more last kind of intriguing piece of this I said you can't always see it in the retina topi and Mark did a really beautiful analysis where in fact he looked for it and he did this by looking at either using the connectome data which is what's shown in these left two panels or actually doing flipped out imaging of individual LC neurons and following where they go he looked at anterior versus posterior or dorsal versus ventral members of these different LC groups to see then in the in their projections in the central brain and the glomerulus whether they were segregated or not and so he did find some examples and actually LC4 is one in which there is even though it's hard to see at a course level if you look in detail there is actually some retinotopic organization here you can see the red neurons that are on the anterior side they tend to terminate on one particular side of the glomerulus and the posterior ones on the other and that was true for several others LCs as well you get this nice retinotopic segregation at least along this one axis in the glomerulus but in fact these did prove to be the exception even carrying this more detailed level of examination into the glomerulus for these other glomeruli what he found is that the majority of the LCs have these non-retinotopically arranged again at least at the level of the deeper resolution organization of their axon terminals so here's the sort of summary of which ones are retinotopic and which ones aren't and so this is where we really think that again it's actually sort of hidden in that synaptic number that the object location is getting translated into information that the downstream central brain neurons can process further okay, let me just really quickly summarize what I talked to you about today stepping back, we looked at the pathway sort of through this ventral lab on our pills sorry my arrow got flipped it should go this way from the eye into the central brain focusing sort of particularly on this pathway and asking what does the eye tell the brain and then what does the brain do with that information starting out with this you know, a hypothesis that maybe there's some kind of feed forward sensory motor processing going on here that we could actually get purchase on and so I showed you that LC neural types do convey a set of 20 distinct visual features to the central brain that the tenities that we looked at more closely could be coarsely clustered by whether they detect small objects or looming these are sort of the first principal components and that they're also spatially organized in the central brain, they're terminals such that neurons detecting similar features are actually physically closer to each other neighboring in the central brain and then showed you an example of then what does the brain do with those features how is it integrated and how there's actually different computations being represented by these different features in one case of looming velocity another case of looming size and that these get integrated by downstream neurons like the giant fiber in order to enact specific rapid behaviors and we then talked about retinotopy well here's an orthogonal piece of information optical location how is that being communicated from the eye to the brain and we saw that retinotopy is actually only preserved in a few of these LC alpha primarily and it's usually lost however by using a sort of synaptic gradient mechanism so precise numbers of connections onto downstream neurons that information is retained and that I showed you then a particular example of not only how that information is being retained but how it's being used by the fly to direct appropriate actions and so I think what you were seeing there when you were looking at the LC neurons make these spatial gradient synapses onto these different descending neurons that come up forward about the jumping you're really seeing there the sensory motor interface the transformation of that visual information into the motor directive Okay, I want to make sure to thank everyone in the lab who I've had the pleasure to work with in my time at Janelia both the current lab and the past lab as well as all of our collaborators and the many, many great groups at Janelia that we've had the privilege to work with and also of course our UCLA collaborators Mark and Larry and I'm happy to take questions, thank you Thank you very much Gwyneth for this very interesting talk fantastic breadth and granularity of the pathways that you are so carefully dissecting and interrogating if I may use these words I will be posting the zoom room link that we are currently using so people can join us already if they wish and the first question that appears is from Tom it is related to the last part of your talk with this synaptic weight arithmetic and the architecture of the circuits and the question is the following the synaptic gradients look great but I wonder if numbers along are enough what about a synapse is actual weight can you measure that, does it matter and might that change over time adaptation, brain state, circadian and so on? Yeah, I mean it's always what was a little bit surprising to us about how cleanly it worked out in our one example, right you can imagine there's lots of ways to change synaptic weight the size of the actual synapse number vesicles you're docking all sorts of ways that you could affect that and at least in this one example the number, and at least again in the fly so this could be different in different animals the number really seemed to provide explanation for a large percentage of what we're seeing match the model very closely so sure, I agree with you that it's probably not the whole story but as we get more and more examples I think I would form the hypothesis that it's actually one of the primary parameters that the nervous system is twiddling at least in these innate circuits for controlling that synaptic weight The second one is from Simon Laughlin Do the gradients ensur that the centroid of an object is computed independent of size and position? Oh, yeah due to the sense of gradients So the interesting thing about the looming LC neurons in particular is that the object really does have to be centered on the receptive field just the way they integrate which we've looked in a little bit more detail requires that sort of precise location at the centroid and perhaps explains why the LPLC2 dendroids at least are very largely overlapping in the lobula so you get a lot of higher granularity of where that centroid is So, yeah, I mean, we can talk more about how that plays out with the different kinds of information in the brain process but yes, I mean, we do think that this is for precisely locating that centroid of where it's coming from Right, and before I'm moving on to the next question because you mentioned the overlap of the receptive fields of the dendroids so like this is in stark difference with what we observe let's say in vertebrate retinas where you have like its retinal ganglion cell type tiling the retina and the dendritic overlaps of the receptive fields are kind of marginal they do overlap but not so excessively as they do in what you saw so my question is like do you think you will have like regional differences with respect to the overlap degree? Yes, in fact, that was again one of the surprises when we did the modeling exercise with art so I kind of glossed over it because it's sort of a second order detail but in fact, when we look at the gradients onto DNPO2 and V11 they're an anterior posterior gradient but because there are more LC4 neurons in the dorsal hemisphere there are also sort of a hot spot that creeps up dorsally that wasn't as obvious until we actually looked at the locations of all of those overlapping dendritic fields so the tiling is somewhat uneven at least in the cases we've looked at in these LC neurons in the optic load Thank you for that So the next question appearing is from Munir Oslemsevik and I'm sorry if I'm mispronouncing the name Hi, great talk, RDPLP, PVLP and AOTU modulated by dopamine that may be related to behavioral state transitions Sure, yeah, you know, I didn't even get into all the behavioral state modulation that could be going on here and going on with these visual pathways I mean, I think that is the next part but I decided to get into not just sort of maybe population representation of some of these computations but how those representations may either be altered by modulators such as dopamine or octopamine or serotonin but also how that might help prioritize some of the pathways So this isn't exactly answering your question but because we haven't worked with that particular pathway but in pathways we have worked with so descending pathways we have tracked LC input that go down and drive landing behaviors we see that there is feedback which we think is at least partially octopiminergic that basically turn on and off those descending pathways according to the behavioral state of the animal So you can imagine if you're standing on the ground or you don't want to activate your landing pathways you sort of simply shut down those particular pathways So it will be interesting to see whether that's true for pathways or not but maybe you want to make them more prominent in terms of how they're getting integrated downstream and you could modulate that potentially with things like dopamine Right, and as there are no more questions appearing for the time being in the chat I will ask one last myself and then I will be terminating the live broadcast so we continue with people that are already here in a more informal fashion as we traditionally do So my question is like within the agglomerulus Do we have like interneurons that kind of control the signal flow in the sense if something is happening dorsally or ventrally and so on like switch this pathway or enhance it and switch the other off Each glomerulus has its own set of interneurons and then as you can imagine there's also interneurons that go between the glomeruli So I like that hypothesis Yeah, I don't think that we we certainly haven't looked I don't know that anyone has been able to look precisely yet at what those are doing but yeah great suggestion Because I'm just trying to think like what is the purpose of putting them in glomeruli if you don't maintain some retinotopic fashion in a big scale So by putting like segregating based on functionality by having some interneurons you can always create small maps that are feature detector based directly So like you kind of optimize like you save energetically on that perspective And the last one one last question that appear right now is from Merth, thank you for the nice talk If the retinotopy is largely lost at the action terminals of LC neurons how do you think the DNs seek out of the correct LCs to preserve the retinotopy functionally? Yeah, so that was actually the question that we initially went to Larry's lab with because this is a fascinating puzzle for development How do you actually if there's not a spatial queue that's allowing these to hook up How are you actually getting these gradients formed? So I will just say stay tuned because it's something that we're interested in investigating and they have some hypotheses we have that might be happening And with that I would like to thank you once again Gwyneth for this excellent talk and thank the audience for being here I'm posting once again the zoom room link in case people want to join us and at this point I will be terminating the live broadcast. Thank you Thanks everybody