How fly neurons compute the direction of visual motion
Date Posted:
February 26, 2024
Date Recorded:
February 14, 2024
Speaker(s):
Alexander Borst, Max-Planck-Institute for Biological Intelligence, Martinsried, Germany
All Captioned Videos Brains, Minds and Machines Seminar Series
Loading your interactive content...
Description:
Abstract: Detecting the direction of image motion is important for visual navigation, predator avoidance and prey capture, and thus essential for the survival of all animals that have eyes. However, the direction of motion is not explicitly represented at the level of the photoreceptors: it rather needs to be computed by subsequent neural circuits. The exact nature of this process represents a classic example of neural computation and has been a longstanding question in the field. Our results obtained in the fruit fly Drosophila demonstrate that the local direction of motion is computed in two parallel ON and OFF pathways. Within each pathway, a retinotopic array of four direction-selective T4 (ON) and T5 (OFF) cells represents the four Cartesian components of local motion vectors (leftward, rightward, upward, downward). Since none of the presynaptic neurons is directionally selective, direction selectivity first emerges within T4 and T5 cells. Our present research focuses on the cellular and biophysical mechanisms by which the direction of image motion is computed in these neurons.
TOMASO POGGIO: If you think about the short history of neuroscience, there have been a few milestones that had to do with understanding the basic operations that neurons and synapses can do from a computational point of view. I call this biophysics of computation. It's like corresponding to knowing what transistor and logical gates do in a silicon chip.
And so one of the first such discovery was, of course, Hodgkin and Huxley, having-- explaining how action potentials are generated and propagate. And of course, they got a Nobel Prize for it. And then it was Hartline explaining the role of inhibition, linear inhibition, in the limulus. And this was another Nobel Prize.
And then another such problem that I think is at the same level is understanding the biophysical mechanisms that underlie the computation of motion in vision. Motion is a very primitive operation. It's evolutionary conserved. It's very important for even developments in humans of object segmentation. And I think this problem has been solved now by Axel Borst, over-- with work over the last few years in almost all details. And we'll hear about it in a couple of minutes.
Just a few words about him. He was a bit younger than I am, and he arrived at the Max Planck Institute after I left the Max Planck Institute in '81 to come to MIT. You arrived there '87?
ALEXANDER BORST: '84.
TOMASO POGGIO: '84. So we missed each other, just by a little bit. But we met soon thereafter. And you studied at University of Wurzburg, and you got a PhD as a member of Martin Heisenberg group, which was-- Martin Heisenberg had a group also in Tübingen under Karl-Georg Götz.
OK. And you are now-- yes, professor in Berkeley, and then meantime, before becoming director of Max Planck Institute in Munich in the institute that was called Institute-- Max Planck Institute of Neurobiology. And now it's changed the name to--
ALEXANDER BORST: Biological Intelligence.
TOMASO POGGIO: Yeah. These are the times.
[LAUGHTER]
Axel.
[APPLAUSE]
ALEXANDER BORST: So thank you very much. So first of all, thanks, Tommy, me the invitation and for the lovely introduction. And as Tommy already told you, I arrived at the Max Planck Institute of Biological Cybernetics in '84 and was confronted there with the work of Werner Reichardt and his famous Reichardt motion detector model. And it was not my intention to study visual motion, but it was unavoidable. Within a few months, I turn from studying Drosophila olfaction-- all of a sudden, I studied motion vision.
And for a very long time, we've been busy with simulating this model and making experiments, behavioral experiments, with some recordings, but the neural circuit really underlying the implementation of this motion detector sort of evaded-- was elusive for decades.
And finally, within the last, I would say, 10, 15 years we learned at an amazing precision how this motion detector is implemented here.
But for those of you who are not familiar with the fly visual system, I wanted first to give you an introduction what the fly nervous system looks like. And here you see such a fly. And in blue, you see the nervous system. You may notice that it has a thoracic ganglion for motor control. And in the head capsule, it has the head ganglion or the brain.
And if you zoom in to the brain, and-- you will notice that the parts shown here in red are the optic lobes. And the optic lobes is the part of the brain that is dedicated to processing the visual information that it receives from the photoreceptors in the facet eye.
And the optic lobes are built in a columnar way that replicate the facet structure of the eye. And it's built in a retinotopic way, like most early visual processing areas, where neurons in neighboring columns process the information from neighboring locations in the image.
And in the fly, you have four such neuropils that make up the optic lobe. And it's called the lamina, the first stage of processing after the retina, followed by the medulla, and then you have the lobula and the lobula plate.
And within the lobula plate, this columnar organization is abandoned. And you find neurons with large dendrites, and they collect the signals of many hundreds of columnar motion-sensitive elements. And they are called tangential cells. And here, you see the reconstruction of three such tangential cells that are receiving horizontal motion information. So these are called the HS cells, and there are others that are called the VS cells that are sensitive to vertical motion.
Now, the central phenomenon that we want to explain is the following. If you take a bar, move it in front of the eyes of the fly to the right and to the left and you insert your electrode into a photoreceptor, it's just schematically here, you will see the same signal each time. So this is a nondirectional signal. Photoreceptor just looks at the luminance, and it's activated by light. It's different from the vertebrate photoreceptor, which has a dark current. So it's activated by light, but the essential point is that no matter whether you move it to the right or to the left, you get the same signal. So from looking at this signal, you can't tell in which direction the bar has moved.
Now, you take this electrode and move it literally three synapses downstream, put it into one of the tangential cells, you get this. All of a sudden, you are receiving a directional selective signal. So these cells depolarize in response to rightward motion, and they hyperpolarize in response to leftward motion. So this is directional. You can, by just looking at this signal, tell in which direction the bar has moved. So within a few steps you go from a nondirectional to a directional selective signal. And how does that come?
And that's exactly what the Hassenstein-Reichardt detector does. So Werner Reichardt and Bernhard Hassenstein in the '50s started off with behavioral experiments on a beetle that was walking tethered on a spherical y-maze. And they designed ingenious behavioral experiments, and from these experiments deduced this kind of motion detector model that is shown with all the bells and whistles here. Has numerous filters in there, but the essence is shown here.
It consists of two subunits which shares the same two inputs. These are the input photoreceptors that look at adjacent image points. And this is the one subunit shown in blue. And that's the other one shown in red. And in each subunit, the signal from one photoreceptor is delayed by a low pass filter with a time constant tau and subsequently multiplied with the instantaneous signal from the adjacent photoreceptors.
And that's done in a very symmetrical way in these subunits. The output of both subunits are subtracted. And what you get is a nice, directional selective signal, right?
So this-- it's important to note that this detector not only qualitatively replicates this phenomenon to give you direction-selective signals, but also in a quantitative way. So I spent the first couple of years together with my colleague, Martin Egelhaaf, in Tübingen, challenging this motion detector. So we were recording from these large field tangential cells.
And we were making computer simulations and then designed various kinds of stimuli and then always came up with predictions and then did the experiments and compared it. And one after the other, my conviction grew that this really is a very good way to describe what's going on in the fly visual system, right? It really strikingly provided these predictions.
And so after a while, of course, the next question was, how is it implemented, in neural terms? What is this delay line here? What is this multiplier? How is this subtraction realized? And all of that.
And so we started to look at the fly neurons. What kind of cell types do we find within the neuropil of the fly? And there's an amazing zoo of different cell types. We find about a hundred different cell types per column. And it's about the same variety of cells that we find in the retina of the vertebrates. And it's about a hundred cell types.
And somehow, somewhere in this neuropil, this computation has to happen. So that was a very long, very unsatisfying period for all of us in the field because we had this perfect formal description of input/output transformation in the Reichardt detector model, but all these neurons were way too small to record from them. And we had no idea about the visual response properties of any of these local columnar neurons here.
And so we were-- just had this sort of black box that-- we wanted to open it one day. And about the time when I came back from Berkeley, that was-- I started-- yeah, that was, like, 20 years ago, I switched from big flies to Drosophila and made use of the huge, genetic toolbox that we have in Drosophila. And that, after a couple of years, really helped us open this black box.
And in the following years, we made a couple of important discoveries. First, we discovered that as in the vertebrate retina, motion is computed in two parallel on and off channels. And some of my colleagues always teased me for a long time by saying, why did it take you so long to figure that out, Axel? [LAUGHS] Anyway, so-- but, you know, that was the first important discovery. And it sort of doubled our work with respect to finding the neural substrate for that.
Then, about 10 years ago, we discovered the T4 and T5 cells as being the primary motion-sensitive neurons, the T4 and on pathway and the T5 in the off pathway. And these were the first to respond direction selective to visual motion, so these are the key players now.
And then we found the LPi neurons to implement the subtraction stage here. And two years ago, we found a biophysical mechanism underlying this multiplication process in T4 cells. And I will tell you in more detail about it.
So just, again, a short, historic word about the T4 and T5 cells. It's like any other neuron that exists on Earth. It has been first observed by Ramon y Cajal. That's an eternal truth. He actually published a monograph in 1915 on the fly optic. Yeah, on the fly optic system. And he described moved and drew T4 cells with the dendrites here and the axon terminal here and T5 cells, and he saw them.
And somehow, they did not fit his expectation of a proper neuron. [LAUGHS] And so he called them [SPANISH], which means "strange things that give rise to a stick with two bushes on each side." Like, he was referring to this and that because he didn't know what a dendrite-- you know, he couldn't tell a dendrite and axon terminal from just looking at the cell. It looked strange to him.
I mean, meanwhile, we know that there exists four anatomical distinct subtypes of T4 cells and of T5 cells. And the four different types of T4 cells-- they all have their dendrites here in the medulla, in the layer 10, and have the axon terminals in one of the four layers of the lobula plate, so there are T4 a, b, c, and d. And the same thing for the T5 cells, except that they have the dendrites here in the lobula, and then, again, ramify-- have the axon terminals in the four layers of the lobula plate.
Now, meanwhile, we can do single cell flip out with Drosophila genetics. We can look at the dendrite of individual single-cell flip outs, top view, and it shows you that the cell actually has the dendrite covering about eight to nine columns. And we can also count the cells by just staining the somata. And if you know that a fly has 750 facets on each side of the head, then 750 columns with eight neurons per column-- this makes up 6,000 of these T4T5 cells on each side of the brain. And I counted them manually for a couple of weeks and confirmed this number.
So these cells were candidates for being motion sensitive for quite a long time. But we had no access, neither to the calcium signal nor to their voltage signal, because they were too small. And over the time, the calcium indicators developed in quality and sensitivity. And very importantly, Gerald Rubin and his colleagues at Janelia Farm provided the community with thousands of different driver lines, cell-specific driver lines.
And so when I heard that, I asked Gerry whether they have driver lines for the T4T5 cells. And sure enough, they send us this driver line. And that is a confocal image of this driver line showing you the dendrites of the T4 cells, the dendrite of the T5 cells. And you see those four layers where they send their axon terminals, too, right?
And then we expressed a calcium indicator in them. And Matt Maisak, a grad student, and Juergen Haag along here, long-term collaborator of mine, they took these flies under the two-photon microscope. And this is how this area looks under the two-photon microscope. And you can already see anatomically 1, 2, 3, 4 layers here in the lobula plate.
And then they stimulated the flies visually. And they saw the following. When they moved the grating to the right, they saw a very strong activity in layer one. Moved the grating to the left-- they saw very strong activity in layer two. Moved it upward, activity was in layer three. Moved it downward-- activity was in layer four, right?
So if we now assign a color to each pixel in this image according to which of the four directions this pixel responded the strongest, we got this. And that shows to you that the four subtypes of T4 and T5 cells are physiologically distinct from each other in their preferred direction. Those with rightward motion sensitivity are terminating in layer one. And those one with leftward in layer two, and then upward and then downward. So this is how they segregate functionally and how the four different subtypes differ from each other.
And that was, I think, a very important step forward to us. But of course, it did not tell us about the functional difference between T4 and T5 cells. And the suspicion was that maybe one is the on, maybe the other one is the off pathway. But we needed specific driver lines for them.
And we got them again from Gerald Rubin and his colleagues, Janelia Farm. So this is the driver line for T4 cells. That's the driver line for T5 cells. But in order to test their sensitivity for on or off, you can't use gratings. You have to use edges of a distinct polarity. So now we have eight different stimuli. On and off edge is moving to the right, to the left, upward, downward. And the same thing here.
And when we tested all that for T4 cells, we saw that the T4 cells respond to on edges very strongly and very little to the off edges, and the other way around was the case for T5 cells. They responded very strongly to the off edges and very little to on edges. And this remaining responsiveness is not-- unclear whether this is due to the selectivity of the driver line. Maybe there were a few T5 cells still active as well.
OK. So from this, we know that the T4 and the T5 cells are the primary motion-sensitive neurons, but the question was, Are these T4 cells and the T5 cells the only motion-sensitive neurons, or the primary-- the only primary ones, or do there exist other parallel pathways that extract motion?
And so the only way to find out about it is behavior. And so we've been using the optomotor response to probe whether or not these flies are motion blind when we block T4 and T5 cells. And so Amin Bahl and Georg Ammer, Tabea Schilling, a grad student at that time, they were in the lab. They put the fly on a Styrofoam ball and had it surrounded by a cylinder with stripes.
And then if you rotate this drum, flies, humans, fish, every animal that has eyes has this optomotor following response. And if you experience this once yourself in an IMAX theater, you know why, because this creates the illusion of self-motion in the opposite direction, and you want to stop that. So you go with the pattern in order to compensate this unwanted rotation in the opposite direction. And that's the optomotor following response.
And now in flies, if you rotate clockwise, you see a very strong following response. If you rotate clockwise-- counterclockwise, it's the opposite direction. If you measure that as a function of contrast, you see an increase and then a saturation.
Now, if we block these T4T5 cells, this response is gone, but it's really completely gone. It's zero over the whole contrast. There is no more optomotor response. Well, that could tell us that T4 and T5 cells are the only motion-sensitive neurons and everything else is sort of post-synaptic to them. But there might be a specific motion system for local motion and not for global motion.
So we also looked at the responses of the flies in behavior to a single bar. And so we replaced this periodic pattern with the single bar and moved it through a window on the right side of the fly.
And if you do that, then we had three types of stimuli. The first one was front to back motion. The second one was back to front motion. And the third one was only a luminance modulation that mimicked the luminance change while-- during the passage of the bar, right? So it has no directional component to it, simply the luminance modulation.
And you see that wild-type flies have a very strong response to front to back motion because they usually fixate the single bar, so they follow this bar and have very little response to back to front motion. They would-- in a closed loop situation, that would allow the bar to move in front of them. And they have this intermediate response to luminance modulation.
Now, when we block these T4T5 cells, flies respond the same in all three cases, literally the same. And I find that extremely amazing because it tells you to them, this moving bar is just as good as a luminance modulation at this place. They don't get the direction of motion. It's just-- it's the luminance modulation, right?
You know, I try to imagine how it would to be motion blind a few times, but this is the closest that you can get by looking at this. And we did further tests on these, behavioral tests on the flight control, et cetera, with these motion-blind flies. But for today, it's just important to state that T4T5 block flies are really completely motion blind. And so that's-- T4T5 is our primary motion sensors, and everything else that is depending on the direction of motion is post-synaptic to these cells. There is no parallel pathway.
So that brings me to the question of how these cells now become directionally selective. What is the circuit mechanism? What is the biophysical mechanism that brings direction selectivity to these cells?
And there have been two ideas in the literature of how you can become direction selective in an algorithmic way. First one is what Hassenstein and Reichardt proposed in '56. And that is the following. You delay the signal here, and you have multiplied with the undelayed signal from the neighboring. So that makes you being sensitive to motion to the right direction. And it's called preferred direction enhancement as a mechanism.
About 10 years later, Barlow and Levick in the vertebrate retina studying rapid retinal ganglion cells proposed null direction suppression, which is very similar, except that the delay is now on the different side, on the opposite side. And now you take this signal delayed, and when it coincides with this signal, it suppresses the response here, right?
And so the question was, are we dealing with preferred direction enhancement or null direction suppression in the fly? And in order to answer that, Juergen Haag, again, did apparent motion stimuli and used the telescope to place the luminance spot directly on the facet eye in the precise location. And he first measured the responses to the individual spots and then in the sequence.
And what he found was that if you have-- these are the individual responses to the individual spots. And this is the response to the sequence. And if you compare it to the linear expectation as the sum of the flicker responses, you see that when you go in preferred direction, you get an enhancement of the response. This is the nonlinearity. And so you have preferred direction enhancement. But if you go in the opposite direction, you have less than the linear expectation, so we also have null direction suppression.
So instead of one or the other, truth is that we're having both here. Preferred direction enhancement and null direction suppression both support direction selectivity in these T4 cells and in T5 cells as well. And that has also been confirmed by work from Tom Clandinin at Stanford.
Now, what is the advantage of having this dual process? One of them would do it by itself. And so if you look at, again, in algorithmic models, what is the directional tuning? And you plot it here as a polar plot. You see preferred direction enhancement gives you such a rather broad directional tuning.
Null direction suppression also gives you directional tuning, less broad than this one, but not as tight as we find it. And together, it's just better. So each process alone also provides direction selectivity, but together, it's more precise. It's tuned much tighter than the other one. So this is still algorithmic modeling, right? a times b to c.
So now what about the neural circuit that is underlying that? Now, for that, it's been very important that my colleagues at Janelia Farm started connectomics works already, like, 10 years ago. And they determined through volumetric EM analysis the precise connectivity of all the different neurons in the fly optic lobe. And that was a tremendous treasure chest for us because we knew which neurons to probe now.
And in case of the T4 cells, they found five different neurons to provide input, or actually six. But for the sake of this talk, I just concentrate on these three types here, Mi9 on the preferred side, and Mi1 as the central input, and Mi4 on the null side of the dendrite of these T4 cells.
And so we measured the response properties of these different cells. That was done by Alexander Arens, Michael Drews, Florian Richter, Georg Ammer in the lab a couple of years ago. And we first measured the spatial receptive field of the three different cell types and then the temporal properties. And these are here now approximated by linear filters.
And what we realized was that the receptive field of these different cells was about the size of one facet. It's about 6 to 7 degrees maximum half width, sometimes surrounded by an antagonistic surround. But curiously, for being the on pathway, Mi1 and Mi4 had on centers. Mi9 had an off center. That was, like, shrug my shoulders, right? So why would you have an off cell in the on pathway? Well, let's wait a bit.
Now, looking at the temporal profile, Mi9 and Mi4 qualified as delay lines because they were low pass properties, and the Mi1 in the center qualified as the unfiltered band pass fast input. So that was very nice.
And we could now model the T4 responses given these cellular responses of Mi9, Mi1, and Mi4 cells by just saying, OK, let's multiply 1 minus Mi9, because it's an off cell, with Mi1, and then divide by Mi4. And we could very much replicate the T4 responses with that.
But of course, this is-- now it's sort of cellular because we know the presynaptic cells, but this operation is still algorithmic, a times b divided by c. So we want to know what is going in the membrane in term of conductances and all of that. Are voltage-gated channels involved? And who's doing what? And so on and so forth.
And now, very important for that, taking this formal model to a biophysical model, is knowing the transmitters. And so we and others identified the transmitters of the various cells. And we learned that Mi9 is glutamatergic. Mi1 is cholinergic. And Mi4 is GABAergic. Well, that made sense in a certain way because Mi1 cholinergic, OK strong excitation. Mi4, GABAergic, OK, null direction suppression. That fits. Well, what about glutamate? You know, it should be excitatory as well.
So we looked through RNA sequencing at the different receptors expressed in T4 cells. And we found the regular nicotinic receptors, so no surprise here; the GABA A receptors, Rdl, and no surprise there. But there was the surprise. The glutamate receptor turned out to be the GluCl alpha. And GluCl alpha is a chloride conductance that is gated by glutamate, so that makes glutamate an inhibitory transmitter in this case.
And so I like that a lot, I have to, say because that sort of-- it's an off cell. You want to sign invert it. It makes it inhibitory. So yeah, minus times minus is plus. This is great, yeah?
So before then taking you to the next modeling stage, we also looked at the subcellular receptor localization in these different types of neurons. And there, the AM analysis done at Janelia showed that those different subtypes, T4 a, b, c, and d not only have their axon terminals in the four layers of the lobula plate, they also have their dendrites oriented against their preferred direction.
So by just looking at the dendrite orientation, you can tell in which direction this cell will respond to. So T4a cell-- for that, you have the dendrite like this. T4b cells have the dendrite oriented like this. And the upward-motion-sensitive neurons have the dendrite downward. And the downward-sensitive-motion neurons have the dendrite oriented upwards, right? It's amazing.
And two students in the lab, Sandra Fendie and Rene Vieira, they tagged GluCl alpha nicotinic alpha subunits 7 and the GABA A receptor subunit RDL and looked at the receptor distribution. And you can already see that GluCl alpha likes to be on the tip. There's more in the middle. And here, you see accumulation on the base.
And if you do that for all the different subtypes, and you always show from the tip to the base, you always find that the GluCl alpha are located primarily at the tips, the nicotinic alpha 7 in the center, and RDL at the base. So they really segregate on the dendrite as we saw the different input neurons, with their transmitter segregating their presynaptic terminals.
So now, that allowed me to come up with a proposal how preferred direction enhancement and null direction suppression could work in principle. So let's assume now these cells with their transmitters regulate postsynaptic conductances here, and these two ones are chloride conductances. And this one here is mixed sodium potassium reversal potential, so it's excitatory here.
Now, let's assume that the fly is in the dark. And then we flash a light in the center here onto Mi1. In the dark, Mi9 will be active because it's an off cell, and Mi4 will be suppressed because it's an on cell, right? So this permanent activity of Mi9 already decreases the input resistance to 50% here. And so the activation of Mi1 will lead to a further decrease of input resistance and give you a membrane potential response in the postsynaptic T4 cells of an intermediate size, right?
So now, imagine that we move an on edge, a bright edge, from the left to the right. What will that do? That will suppress Mi9. And the suppression of Mi9 will lead to a closure of the chloride conductance, so that will lead to an increase of the input resistance. And this increase of the input resistance now leads to an increase of the membrane potential response due to the excitation through Mi1, yeah? And then later on, Mi4 is activated, but the signal of Mi9 and Mi4 leave this window of opportunity for this excitation to arrive here.
Now, for-- this is for preferred direction. And it's a-- sort of an amplification because it's released from shunting inhibition. Now, for the opposite direction, the two signals of Mi4 and Mi9 overlap in time, so there is no such window of opportunity. And that leads to a further decrease of input resistance. And that leads to a decreased membrane response. So you see that by purely passive membrane properties, you get an amplification in the preferred direction and a suppression in the null direction. There is no voltage-gated process involved.
And I was not the first to propose such a mechanism. Tommy here already in '78 proposed null direction shunting inhibition to underlie direction selectivity. And that is a very beautiful paper to read because it's just like what I did, is simply passive membrane properties and shunting inhibition to give you the nonlinearity that is required for direction selectivity.
And the neat thing, or addition to his idea, is simply that for preferred direction enhancement, it's the same mechanism, but it's a release from shunting inhibition that leads to amplification.
TOMASO POGGIO: We got half [INAUDIBLE].
ALEXANDER BORST: Yeah, exactly. [LAUGHS] And so that's very rewarding, I would say. So now, of course, we were curious to see how it does work. And for that, I have to acknowledge Lukas Groschner. He's a postdoc who came in four years ago, or five years ago, to the lab. And he managed to record intracellular by whole cell patch from the T4 and the T5 cells. And that was a tremendous achievement.
And there was Eyal Gruntman in Michael Reiser's group at Janelia who also did that. But in our lab, no one could do it. And then Lukas came and did that. And that was extremely important to have a membrane potential recording because this prediction required also testing input resistance and reversal potentials and all of that, things that you could not do with calcium imaging, right?
And he also trained a PhD student, Jonatan Malis. And Jonatan and Lukas agreed on a project that sounded, like, crazy to me because they wanted to apply a certain stimulus set in all the experiments and record from all the different input neurons and as well as from T4 cells in subsequent preps under-- using the same stimuli, right? And then-- in order to test this hypothesis.
And to me, that-- I call them the incredibles because they were just crazy, providing these data. That was a tour de force. For about two years, they spent day and night in the lab. They were crazy. But they really wanted to do that. And this is what they found.
So for preferred direction motion of the edge, you see this increase here and then [CREAKS], the Mi9 signal goes down. And that is followed by an increase of the Mi1 signal. And that is followed by an increase of the Mi4 signal. So as predicted, one goes down before the other one goes up, leaving this window of opportunity for Mi1 to kick in. And that is the response in the T4 cell to that. For null direction, those signals overlap, as predicted. And as predicted, you get only a very small null direction response to that.
So this was not the end of the story. They wanted to test this model further. And first, they looked whether, indeed, glutamate is an inhibitory transmitter, so they also patched from the T4 soma and then puffed glutamate on the dendrite. And they saw a hyperpolarization. If you knock down the GluCl alpha receptor with RNAi, the cells depolarize permanently and no longer respond to glutamate. So this tells you glutamate is indeed an inhibitory transmitter.
Second question was, Does it really affect the input resistance? So for that, Lukas injected current of various amplitudes into the cells under control conditions and after glutamate has been puffed on the dendrite. And you see that the input resistance is-- drops to 50% when you puff glutamate onto the dendrite. So it's a huge decrease of input resistance due to glutamate. So glutamate can affect the input resistance of the cells.
Next question is, Does it also do so during visual motion and not transmitter puffing? And so here, Lukas measured not only the membrane potential of the T4 cell during motion of an edge in the preferred direction, but also the input resistance. And you see that the input resistance grows right before excitation kicks in and then goes down again. And so you will see that increase of the input resistance in the order of about 30% to 40% also during visual motion.
And so finally, the question is, Is that-- does this mechanism really affect the directional tuning of these T4 cells? And this is the directional tuning of control flies. And when we block again the expression of GluCl alpha with RNAi we see this reduction of directional tuning.
It's very important to know that it's still directional, but it's just reduced directional selectivity because you took away one of the two mechanisms that is responsible for that.
And so we thought that we now could go on and have a T4 cell model, a biophysical model, conductance-based model, a point neuron, because the dimensions are so small that compartmental modeling is not necessary in that case, unfortunately to me, because I like doing compartmental modeling a lot, but it's not necessary for T4 cells.
And we used as the input the measured signals from Mi9, Mi1, and Mi4 cells and also the others that I have not shown to you. If we do that, here are the data. And this is the model. And the same thing for-- now for 30 degrees per second. This is the data, and this is the model.
So it fits very tightly, I would say. And it's a model that has just 14 different parameters, so it has the reversal potential. It has the gain for the different synapses. And it has one threshold, so it's a rectilinear transfer function that we used here.
And now, for the whole data set, for gratings moving at five different velocities, for on edges at these five different velocities, off edges at five different velocities, always see the preferred direction and the null direction data in the bold, thick lines and the model in the faint lines here. And you see that they nicely agree over this whole stimulus set. Here, you see just the peak responses. And I think I'm rather satisfied with that kind of fit. There might be some minor details which you can still tweak, but I would say we're pretty much done here.
So that brings me to the next question and to the projects that we're actually working on right now in the lab. And that is the mechanism underlying direction selectivity in T5 cells, the off pathway. And the off pathway is very different from on pathway with respect to the different neurons that provide input to them.
So there are five different neurons that provide input to the T5 cells. There is Tm9 that is on the tips. So this is where Mi9 was for T4 cells. Then there are three cells providing input into the center. That's Tm1, 2, and 4. And on the proximal side, there is only one neuron contacting the T5 cell dendrite. And that's CT1.
Now, CT1 is an amacrine cell. It's not a columnar cell and it's one of the most beautiful cells that I've ever looked at. And so I want to show that reconstruction from Amy Sterling from Princeton. This is the cell. There exist only two cells in the fly brain, two CT1 cells. And with their branches, they go into every facet or every column in the medulla and every column in the lobula, making little knots there, little processes. But there is only one soma on each side, crosses over to the other side, and then builds this amazing structure.
You can enjoy the cell one more time. It's really something that-- I mean, maybe you can go to FlyWire yourself and take a look at that. It's a beautiful cell. It's an amazing cell. But for us, it, of course, was the question-- if this is the only cell that provides inhibitory input in the off pathway, how can it do it if it's an amacrine cell, right?
So I first reconstructed part of the CT1 cell and did compartmental modeling. And I realized that taking regular membrane and internal resistivities, that these different compartments here corresponding to these processes in the different columns are electrically isolated from each other. If you inject current there, it just doesn't spread to the rest of the structure, so they seem to be electrically isolated.
And there was one postdoc in the lab, Matthias Meier. He did calcium imaging in CT1. And he presented white noise stimuli and then did reverse correlation to determine the receptive field of each of these little knots in each of these columns of the CT1 cell. And what he realized is that each compartment has its own receptive field. And it's just about 6, 7 degrees wide. And it's just like a columnar cell, a separate cell, so the CT1 cell seems to act as if there are 750 of them, right? But it's one cell, but each process acts in electrical isolation from its neighboring process.
So these results suggested that the CT1 cell is well capable of providing local inhibitory input for null direction suppression in the T5 cells.
And so the next thing that Matthias Meier and a student, Amalia Braun, did was to kill CT1 by expression of hid or rpr. And when they do that, you see-- this is-- the black line is the control fly, the directional tuning. And you see after CT1 has been killed, it's broader. And it's broader because in null directions, inhibition is no longer there.
And if you then instead of killing CT1 express RNAi against the Rdl subunit of the GABA receptor, we see a very similar broadening of the directional tuning of the T5 cells.
And then finally, we also wanted to look at the time course of the different signals and confirm if this is really the delayed inhibition here. Is it really slow, as opposed to Tm1, 2, or 4? And so this shows to you the off response of Tm2, Tm9, and CT1. And you see Tm2 in the center is fast, and Tm9 is delayed, as it should be, because it also should provide preferred direction enhancement.
But CT1 on the null side is even more slow, slowly, right? So it really qualifies as being the inhibitory signal. And if we take it away, we see the broadening. So all that says that null direction suppression in T5 cells is provided by the CT1 cell, and it can do so because it acts like 750 columnar neurons and not just one cell.
So that brings me to the final question now. What about preferred direction enhancement in T5 cells? And I have to admit, I cannot give you a final answer here. What Amalia and Matthias did was block all the different cells and see how that affects the motion response in the T5 cells.
And here, you see with Shibire, blocking Tm1 has almost no effect. Tm2 and Tm4 have intermediate effects. And Tm9 has the strongest effect, right?
And the same thing is if we optogenetically hyperpolarize the cells by GtACR expression, again, there is no effect in Tm1, but blocking Tm2 or 4 has sort of an intermediate effect. And again, Tm9 has the strongest effect.
So it seems like Tm9 really is the delayed signal that is needed for this preferred direction enhancement and preferred direction enhancement-- the excitation is provided by these central neurons. But how they interact and lead to a nonlinear amplification-- we don't know at present. Maybe it's through some voltage-gated processes. We just have to do the same kind of analysis, looking at the input resistance and do that with whole cell patch clamp recording and further experiments. So we're right now doing that. But I, at present, cannot tell you the biophysics of preferred direction enhancement in T5 cells.
So that brings me to the summary. I have told you that we've been, for a very long time, looking for the cellular implementation of this Hassenstein-Reichard motion detector model in the fly optic lobe. And we found that T4T5 cells to be the primary motion sensors in the fly with the four subtypes tuned to rightward, leftward, upward, and downward motion. And that led-- that is like a beautiful experimental result.
In particular, if you think about-- that the flies' eyes start with hexagons, right? So you have-- you don't have orthogonal Cartesian coordinates, but it's sort of downsampled onto Cartesian coordinates in between, which is very interesting.
And having found and identified those cells, then, what are the mechanisms we found? That both preferred direction enhancement and null direction together lead to direction selectivity in T4 and T5 cells. And with respect to the biophysical mechanisms, we know-- we're sort of done with T4 cells. We know the null direction suppression through CT1 and T5 cells. But the mechanism for preferred direction enhancement is not done here, but hopefully, within the next two or three years, we will-- I can give you an answer.
And with that, I want to close and thank the Max Planck Society for supporting my work for many years. And my collaborators at Janelia Farm, Gerry Rubin and Aljoscha Nern-- they provided these tremendously important driver lines that are basic for my research.
And I also want to thank the Fly Connectome Team at Janelia, who provided the data about the connectivity that saved me so much time. Instead of screening through all the possible candidates, you just know which cells are connected, and you just pick those driver lines. And that's a tremendous help and led to lots of progress in this.
But most importantly, I want to thank my team, the Motion Detectives at the MPI, all these wonderful people with various backgrounds, from computer science to physiology, from molecular biology, and we all met on the dendrites of the T4 and T5 cells and joined forces there. And thanks a lot for your attention.
[APPLAUSE]
have an interactive transcript feature enabled, which appears below the video when playing. Viewers can search for keywords in the video or click on any word in the transcript to jump to that point in the video. When searching, a dark bar with white vertical lines appears below the video frame. Each white line is an occurrence of the searched term and can be clicked on to jump to that spot in the video.
