 Okay, I guess we are live. Welcome, everybody to this talk of the Sussex vision. My name is Marvin Seifert. I'm a PhD in the lab of Tom Baden at Sussex University. And before I will start to introduce our guests today, I would like to thank the organizers of the Worldwide Euro Initiative, especially Tim Vogels from the University of Oxford, for providing the general framework, which makes it possible to have these talks online. So thank you very much. Today, we are delighted to have Marianne Siles on the program. She's not only my previous supervisor, but she's also one of the leading experts in the field of Drosophila genetics and vision. She started her science career by doing a PhD at the University of Munster, which she finished in 2009. After that, she moved to the University of Stanford and did a postdoc with Thomas Glendinin. Here she studied the motion vision circuit in the Drosophila optic lobe. And after that, she moved to Götting, where she became a PI at the European Neuroscience Institute. And she continued to study vision in Drosophila. And I guess the spectrum became a bit broader, so with more vision in general. And since 2019, she's a full professor at the University of Mines. I'm really happy to have her today. And I guess most of the work that she's presenting will be from the time in Götting. Before we start the talk, I would like to say some things about the organization. So if you have any question at any time, you can post that in the live chat, which you can find either right next to the video or maybe under the video. And also, if you have questions that we will not answer in the after the video, then you are happy to join our Zoom meeting after the video or after the presentation. We will share the link in the live stream after the talk is finished and after we have answered some questions. Yeah, and I guess it would be fun to just have a casual hanging out after the talk. I guess that's everything I wanted to say. I hope I haven't forgotten anything. So yeah, please welcome Marvin Siles. Okay, thank you very much, Marvin. So let me share my screen with you. Okay, so I would of course like to thank Marvin and the Barden Lab for having me today and all of worldwide neural for making this possible and adding so much fun science to the last weeks. So I'm very excited to share with you what my lab has been doing over the past a year or a few years. And I talk about how vision can work in dynamically changing environments and more specifically how visual systems can accurately compute contrast and in rapidly changing visual contexts. And so I will start by showing you a movie that my previous grad student Katja took of her dog Yuppie. So what you see here is Yuppie and catching his toy. And what you can appreciate is Yuppie is very good at doing that. And so he can do that sort of like regardless of what the external like conditions are. And we actually learned this by going on one of the very first lab trips went into our hike where we sort of got lost in the gutting woods. And Yuppie was with us and we sort of like entertained him by throwing a ball and why we could like hardly see our path anymore. Like Yuppie was still like, you know, very constantly good at like catching in his toy. And as you can see here, and he can even do that when you are sort of confounding the experiment by shining like a bright light atom so that Katja can protect this movie. So and if I explain this to you, then that visual processing works efficiently in very different light conditions, then you will probably all think of adaptation. So the reason why visual systems can handle such different light conditions is that they are sensitive to contrast or more specifically temporary webber contrast in this case. And sort of what visual systems are sensitive to is the difference in luminance between object and background and normalized to background. So because of this term here, vision works sort of very stably and across different adapted states and therefore different light conditions. And this is a concept that was also very beautifully introduced by Jeff Diamond's talk that I hope you all had a chance to see here on the Sussex Vision Channel two weeks ago. But there is a problem or there might be a problem for visual systems when the background is quickly changing. So here I just drew a fake tree and a fake shade into this image here. And so I think there's a problem when Yuppie will try to catch a ball that we are throwing into the shade for him. And this is because the visual system is not adapting instantaneously. So adaptation is very fast, but it doesn't happen within milliseconds. And this is shown here in like older data, for example, from the retina. So this is a skate retina here on the left-hand side. So you see that if you sort of like measure adaptation, there's a fast component to adaptation that happens maybe even within tens of milliseconds and seconds. And then here there's a very slow component of adaptation that lasts up to seconds. And these different time scales of adaptation are true in both invertebrate and vertebrate visual systems and also in humans. So here are some human psychophysics experiments where you see that there's a fast adapting component in the response that works on the order of hundreds of milliseconds and then there's a slow component that lasts tens of seconds. And so the problem is that if you or if a visual system encounters a visual signal sort of like within where a background luminance is very quickly changing, then contrast sensitive neurons won't be able to compute WebR contrast correctly. And this is something that I'm showing you in this term down here. So here I'm so basically what the what a contrast sensitive neuron would then be sensitive to. So it would still sense the luminance difference between the object and the background. So in this case, like the dog the dog's toy and the shade of the tree, but it would sort of like divided by the luminance of the background. So in this case, the bright snow. And this would lead to a wrong estimation of contrast to a too small contrast response. And this is something that we sometimes call apparent contrast. And this is of course not only a challenge for dog visual systems, but for visual systems in general. So my lab uses a fruit flies Drosophila as a model to study a visual processing. And you can easily imagine that if a fruit fly is trying to escape a predator, like the praying mantis here, then for the behavior of the animal, it would be advantageous if it was able to like do this equally well under all of these conditions. So in bright daylight on the left, a dusk or dawn, or when it's for example, when the predator is hiding in a shade, for example, under a leaf here. And again, the problem here is that what adaptation can achieve is contrast computation in slowly changing environments. So for example, the contrast that the fly is calculating, sort of looking at the head of the praying mantis here in bright background will be the same. So the contrast signal will be the same at dusk or dawn because of adaptation. But if adaptation hasn't hit, then the challenge will be to compute sort of the luminance difference between this object and the background divided by a too large background from the previously adapted state. So contrast sensitivity might underestimate stimulus in quickly changing environments. And today I will and again, so what I'm like sort of illustrating in gray here is that it would be advantageous for the fly to exhibit contrast constant behavior, so to have equally equal behavior responses to visual sickness under any of these conditions. And today I am telling you about some work that we did over the past years. And I'll show you that luminance information is still present in the visual system and ensures the accurate estimation of contrast in dynamically changing visual environments. And part of this work was actually published earlier this year. So you can read about it later if you want to want to hear more. And in the published work, we basically showed that luminance can like sort of do exactly what I just showed you, namely fix this discrepancy between contrast sensitive responses that sort of underestimated visual stimulus and the actual behavioral response. But then I'll also show you unpublished data and showing new scenarios where the opposite is true. So I show you like some data where we found conditions where contrast sickness would overestimate the relevance of a visual stimulus. And I will show you that luminance information can also fix this. And if I'll have time, I will also talk about unpublished work looking at the cellular and second mechanisms of this luminance based and scaling of contrast computation. And I would just like to add here that there's a lot of data in here that I've actually never talked about before. So it's a bit scary to do this in this context for the first time. But this also means that I would really appreciate your questions and feedback and criticism. So I'm looking forward to many questions afterwards and also to a likely discussion afterwards for those of you who are interested. Okay, so let me start by telling you about our systems. So again, we work with Drosophila. This is a Drosophila eye. And it's composed of these 800 single units or omatidia. And downstream of these 800 omatidia, you will find 800 units. So basically you have 800 columns that together form a retinotropic image of the outside world in each eye. And if we take a cross-section through the system, then we basically see the structure illustrated here. So here on the outside, which I didn't show you, is the retina where the photoreceptors are housed. And then photoreceptor terminates are coming down into this first structure here, which is called the lamina. And there they synapse onto a series of so-called lamina monopolar cells called LMCs. And they are L1, L2, and L3. And you will hear a lot about L2 and L3 today. And these are the inputs through pathways that are specialized to detect contrast increments or on stimuli and contrast decrements for off stimuli. And L2 and L3 are two inputs to the off pathway. And they synapse on a series of interneurants here, which then in turn project onto the dendrites of these direction-selective cells. So down here in the lobular plate are the first direction-selective cells of the on and the off pathways. And they are called T45 introsophila. And what we know, so we basically know extremely well how these peripheral stages of visual processing respond to light and to contrast. And so they have been like extremely well characterized by the labs of Simon Loff and Roger Hardy and Miku Yusula and others. And so how a photoreceptor responds to light looks like is shown here in this orange box. So basically a photoreceptor will respond to a step change in luminance with this like transient response that basically already signals contrast. And then a sustained component, which is luminance sensitive. And then all these lamina neurons downstream or the lamina neurons downstream or together the LMCs are described to be amplified more transient and sign inverted versions compared to their photoreceptor input. And this is something that you can see here. So you see that they have the opposite sign signal in response to light. And this is because photoreceptors release histamine and then all the LMCs will express a histamine-gated chloride channel. So you have a sign inverting synapse at this stage in visual processing. And then you see that these LMCs here and these I should maybe already add that these are mostly data from a 1 and a 2 neurons here are really transient versions of the photoreceptor input. And the idea is that they are like very contrast sensitive and that the visual system sort of like got rid of luminance information at this point. And so what we asked or what we wondered if these contrast sensitive responses like this early in visual processing are fully sufficient to explain behavioral responses to contrast under dynamically changing light conditions. And to ask this we basically did behavioral experiments. So we put a fly on an air-cushed ball. It's situated in an arena in which we just can display visual stimuli. And then we showed the flies and stimuli where we had a dark edge moving on to different background luminances. And sort of every time we like show a new moving edge to the fly we show a new background luminance. And what is important is that all of these stimuli are of minus 100% Weber contrast. Because it's a dark edge the luminance will be almost zero and this means that Weber contrast will always be minus one or minus 100%. Whereas the apparent contrast, so like the contrast response of a non-adapted cell will increase with increasing a background luminance. And so we basically asked how do flies respond to these different minus 100% Weber contrast stimuli and how does this scale with contrast sensitive responses in laminate neurons. And to get at the contrast sensitive responses in laminate neurons we could go back to again all of these recordings that have been done in these cells and they have been so extensively characterized that we could basically predict the responses of these contrast sensitive responses based on existing literature. So and this is all work done by a really brilliant PhD student in my lab Madura. And so what Madura did she first measured behavioral responses to these different off edges. And what you can see is that the behavioral responses here really scale with Weber contrast. So again these are all 100 minus 100% contrast stimuli. And what you can see here is that the behavioral responses are not significantly different to each other. So the flies can respond well to each of these off edges even when the background luminance is quickly changing. Okay. And then she compared this while she first predicted the responses of these contrast sensitive LMCs based on existing data and basically found that a contrast sensitive LMCs would have a smaller response to responses onto a dark background. And if you compare this with the behavioral responses that Madura measured. So this is from the same data as shown on the previous slide only here I'm putting the peak training velocity only to the five different edges here. Then you see that there's a discrepancy where a contrast sensitive response alone cannot account for behavioral response if the background just turned dim so in sort of like contextual dim light. And this is basically illustrating the problem that I showed you in the schematic drawings in the beginning that under the conditions where the background just turned dark you basically have a contrast sensitive response that would underestimate and the behavioral relevance of the stimulus. And so what we hypothesized is that there must be a corrective signal that sort of like accounts for this discrepancy and ensures an appropriate behavioral responses to contrast even in dynamically changing and visual environments. So what could be accounting for this discrepancy? And so we got some insights from work that another PhD student Katia was doing. So Katia is a PhD student who recently graduated and just actually sadly left the lab three weeks ago. And Katia did very extensive characterization of these lamina cells that I introduced to you already. So Katia expressed the genetically encoded calcium indicator GCAMP in either L2 and L3. And I should just remind you that whenever I say L2 I mean L2 as a cell type. So there are 800 L2s in the fly visual system, one in each of these 800 omateria that we have. So basically this is how this looks like. So Katia is recording calcium signals from the axon terminates of L2. And you see that they are in one layer in the visual system. They are several axon terminates lined up next to each other. And the same for L3. And when Katia shows these flies like very simple light flashes, then what we see is that there's an increase in calcium signal in response to off in both L2 and L3 neurons. And whereas the calcium signal in L2 goes back to baseline fairly quickly, the calcium signal in L3 is really sustained. And we thought this was interesting for several reasons. One of them is that very similar properties have been described in the virtual retina. This is actually data from Tom Barton from a bit more than 10 years ago, where when he was doing very similar experiments and recording calcium signals in bipolar cell axon terminates, he also saw these transient and sustained bipolar cell types. And we also thought that this is interesting with respect to the idea that we need this corrective signal, but that all laminate neurons were thought to be contrast sensitive. So we decided to look at this carefully and designed a stimulus that allows us to really distinguish contrast and laminate sensitivity. And this is inspired by a paper from Jeff Diamond's lab, where he like discovered that there's laminate sensitivity present in the virtual retina in A2 M-acron cells, for instance. So what we did is we adapted flies to a bright background for 30 seconds. And actually, I should say that this is something that Marvin did when he was a master student in my lab. So what Marvin did, he adapted flies to a bright background for 30 seconds. And then he showed them two sequential off steps, the A and the B step. And each of them can take on one of six different variants. And so the A step will always differ in both luminance and contrast. But then the B step, and this is the important one, is designed so that it will take on one of these six different luminance values. But the B step will always be 25% contrast relative to the A step. So this really allows us to distinguish a contrast from a luminance sensitive cell. Namely, a contrast sensitive cell will respond to with equal amplitude to all of these six different 25% contrast steps. And here in blue, I'm showing you the responses of these transient A2 neurons. And when you plot the responses, the kites and signals as a function of contrast here, then what you see if you look at these six different data points for minus 25% contrast, you see where maybe you can't even see, but they are all lying on top of each other. So this is really how contrast sensitive cell looks like. If we instead plot this as a function of luminance, you see how similar they are. Okay, if we do the same for A3, then we see that this looks very different. So the responses to these six different 25% off steps already differs. And what you see is then if we plot the responses to the stimulus as a function of luminance, then at three kinds of signals respond very similarly to nearby luminances, no matter if they were coming from the A or the B steps. So this hints towards A3 being a luminance sensitive cell. We tested this explicitly by sort of doing the opposite experiment where we fix luminance. So we allowed random transition between five different luminances. And these steps are then associated with like different contrasts. And what you see is even in like responses of single L2 exon terminus is that L2 neurons have this like very, very flat baseline. And there's this transient response in L2 neurons whenever there's an off step. And this transient response is larger, the larger the amplitude of the step. This looks very different for A3. So A3 basically like doesn't have like one baseline really, but there's a constant change in like steady state whenever there's a step change in luminance. And what you see here is that even for like, you know, two or equal steps, so two equal step changes in luminance at three responses can be very different. So here you see like a large increase in calcium. And here you basically see no response. But what instead at three appears to be sensitive to sensitive to is like really the luminance of the stimulus. So for example, here in black, I'm showing you all the transitions to complete darkness on the screen. And this is always associated with the same calcium signal or the same calcium steady state response in A3 neurons. And so we can like pull all of these responses. So this will be the response to like any of these steps and landing on complete off. And if we do this also for all the other steps, then we see that at three has this nonlinear and decline in response amplitude and with increasing luminance. So it's a luminance sensitive cell that is active in dim light. And this is basically perfect. This is exactly what we need to like fix contrast computation. So we hypothesized that luminance information from this A3 pathway might provide this corrective signal that ensures these equal responses to and the same weather contrast stimuli. So that ensures contrast consistency. We tested this by going back to behavioral experiments. And so again, the idea. So what I showed you before is the discrepancy between contrast sensitive neuronal responses and behavior. And if we are correct and this A3 pathway provides sort of like this corrective signal here, then taking out the A3 pathway and this is something that we can genetically do in flies should sort of like bring these behavioral responses down or closer to these contrast sensitive responses. So this is what we did. This is what we tested. So we genetically blocked the A3 pathway. And first I'm showing you Y type. So our control behavioral responses again, this is essentially the same data that you have seen before. Only now I'm sort of like not plotting them on top of each other. But on the left hand side, you see the turning responses to off edges moving on to dark backgrounds and in the middle intermediate backgrounds and then on the right side, right backgrounds. And again, you can see that the behavior responses to these different to these off edges moving onto different backgrounds is essentially the same under all of the for all of these stimuli. If we now block the output of the contrast sensitive pathway L2, you see that L2 is required for all of these for behavioral responses to all of these conditions. So what I'm actually plotting is the turning response as a function of time and in gray is the time where the off edge is sort of moving across the visual field of the fly. And what you see is that the turning response is strong, you reduce in all of these conditions. So the contrast sensitive pathway is required under any of these conditions. When we block the outputs of the aluminum sensitive A3 neuron, instead, what we see is that the peak turning amplitude is not altered if the off edge is running or if the fly is turning to an off edge running onto a bright background, we start to see an phenotype and intermediate background luminance and you see that the turning responses severely altered when the edge is running onto a dark background. Okay, if I summarize these data and I plot them together onto this like curve that I showed you before, then you see that this behavior response of A3 silent splice now looks much more similar to contrast sensitive responses alone. And what this basically shows is that luminance information is really like the thing that enables proper contrast responses and this is provided by the system. We have this like parallel dedicated pathway that gets inputs from A3 and that is sensitive to luminance. Okay, so but so far, you know, I only showed this sort of like for one adaptation state. Yeah, so the fly is sitting there, it's like encountering these like randomly interleaved five different edges. But in principle, when we, you know, the visual system is like able to encounter or will encounter many much larger range of luminances. So how would this like basically compare to if we like put the fly into, you know, dusk or dawn or like near darkness or something. So this is something that we tested. So we characterized these behavioral responses at many different adaptation regimes. And we did this by using the same set of five stimuli where off edges are running onto different backgrounds. And then we changed the whole range of luminance is that the fly was encountering by adding neutral density filters to this and by doing this, we could cover four orders of magnitude as shown here. And what you see here on top for wild type turning and wild type turning is basically almost a continuous function of luminance. And then we measured behavior down to a regime where like basically both the flies and actually also as humans can't see anything anymore. So until we are like reaching sensitivity threshold down here. And then we asked, okay, so what's the requirement of this luminance sensitive pathway under all of these conditions? And how does this again scale with signals that would be provided by purely contrast sensitive first order internals. So Madura again went and predicted contrast sensitive responses for all of these conditions. And for any of these adaptation states indicated by one gray scale here, you see that contrast sensitive responses would always under estimate a visual stimulus in like in dim light relative to the range of luminance is that the fly is adapted to. And this is something we call contextual dim light. And then when we when we compare this with like the behavior of flies in which we block a three, then you see that for any of these conditions, this three block behavior is actually very similar to these contrast as sensitive predictions. So for any of these conditions, for any of any adaptation state, a three is required in contextual dim light. So dim light relative to the range of luminance is that the fly is adapted to again. Okay. And yeah, so this means that as we can like basically fix contrast computation over many different adaptation states and over a large range of like absolute luminance is and something that we also saw in the data. And I think we didn't point out in the paper, but I'll point this out today is that when you look at like the responses that are nearing this like, you know, dim end of where flies can actually like still see anything. So like low absolute luminance is what you see is that it appears to be that a three silencing also has an absolute effect. So basically here for for these adaptation, like these two dimmers, like or these two experiments where we use the most and defilters so where we have the lowest absolute luminance is what you see is that at three is not only required, you know, in contextual dim light, but it's required overall. So here you see that at three silence behavior doesn't match wild type behavior, even on the bright end of this adaptive state here. This is something that I come back to in a bit. Okay. So yeah, so this is all published. And so what I hope I could convince you and so far is that luminance information is required for these accurate estimation of contrast, when the world is quickly changing. And so I illustrated the problem to you that, you know, you have like two small contrast sensitive responses, if the world is just like, you know, quickly going dim, and this can be corrected by this luminance signal that we identified. Okay. And but so far, I only talked about like, sort of you can say a very special case, namely, namely 100% contrast. So a condition where like complete dark edges are moving on to like different backgrounds. So what about other contrasts? So Madhura continued to characterize behavior and showed different contrasts by having a not completely dark edge running onto different luminances. And these would be associated with WebR contrast ranging from 50% to 87%. And what you see here when in the wild type turning responses here is that the behavior responses again, scale beautifully with WebR contrast. So the larger the contrast, the larger the response. And so we then look more closely at these 50% contrast responses where you see that wild type turning is actually very low. And we design stimuli that are all of that are now all 50% 50% WebR contrast minus 50% WebR contrast, but associated with different foreground and background luminances. But again, all of them are 50% WebR contrast. And again, what you see is when you measure behavior responses to any of these stimuli is that they essentially look the same. So behavior responses, again, scale with WebR contrast, if you have different ways to achieve the same WebR contrast, you will still have the same behavioral response. And so then we ask what happens to these behavior responses if you now take out the luminance sensitive pathway. And what we saw is that under these conditions, so when you now block at three here, you actually see an increase behavior response. And what this basically means is that this is like, and I mean, maybe just so I'm not showing you like these LMC predictions here, but just so you can even grasp this intuitively, what you are seeing here is a very, very small wild type turning response to these contrasts. But 50% is half of 100% contrast. So contrast sensitive LMCs will still have like a strong contrast sensitive response to a strong response to like any of these stimuli, but apparently the behavioral relevance of these signals is very low. So you have like very low wild type turning response to these signals. And this is again something that you sort of like uncover when you silence at three. So the idea is, again, that like at three sort of like matches the behavior relevance of the discrepancy between the behavioral relevance of a stimulus and a purely contrast sensitive behavior only under these conditions, it's sort of like is doing the opposite. So luminance information here suppresses behavior if a purely contrast sensitive response would overestimate a visual stimulus. Okay. And then Madura didn't only do this for one adaptation state, but also did this for many different adaptation states, again, by using ND filters. And what you see here is that this is again something that holds true for different adaptation regimes. So here on the right hand side of this plot, you again see that at three or luminance, the luminance sensitive pathway via at three suppresses behavioral responses to very small contrast. And again, just note how similar, you know, all of these wild type responses in gray are to all of these 50% WebR contrast. Yeah. And again, this is something that that you don't see anymore when you when you when you silence at three. So you sort of need this luminance sensitive pathway to keep behavioral responses to WebR contrast constant. And what you also see is that there seems to be like a switching point on the left hand side of this plot, where you turn the behavioral phenotype of these at three block flies from having enhanced responses to having lower responses. And again, this is something where at three appears to have a role in absolute dim light. And this is the same dichotomy that I pointed out earlier in for the minus 100% WebR contrast data where I, you know, put these arrows and said, look, there also appears to be an absolute contribution there. And that is something that you see much more clearly now that we have this overestimation case, otherwise within this within the data. And so this like, you know, seemed complicated to us. So we thought, okay, they are apparently like many different things that this luminance sensitive pathway is doing, it can like fix overestimation and underestimation of contrast and in contextual luminance conditions, but then it also has this absolute requirement. So we went to collaborate with Shuaishao, a gradient student in Juliana, Georgia was left at the Max Planck Institute for Brain Research in Frankfurt, and discuss the data with them. And I should maybe like say that it was like really Shua, who like, you know, pointed out this dichotomy in our data and like, you know, showed us very clearly that this is there when he's like, just subtracting the two and behavioral conditions so controlled and at resilencing. And then, you know, this was like even before we had these data for like low contrast where this is like becoming very evident. And so based on all these like observations, he devised a model and trained this on the data. And so we basically now have one model that like really can capture all of these components that we that we see in our data. So that sort of explains what that shows this like underestimation of visual signals when a three is silenced in in for high contrast, the overestimation for low contrast and also the absolute requirement in like absolute dim light. Okay. Okay, so what I showed you so far is that luminance information is basically doing many things and sort of like the overarching theme is that it ensures the accurate estimation of contrast in dynamically changing visual environments. And for the last 10 or 50 minutes or so, I will tell you about the work that we are doing to figure out how this is done. So we are like interested to understand the cellular and circuit mechanisms of this luminance based scaling of contrast computation. And there are many open questions out there. And one of them is like how is contrast constancy achieved in visual circuitry. And this is a question the project that Burak Gure another graduate student in the lab and took on. And so so far I only showed you data sort of comparing these contrast sensitive responses in the very periphery. So in these first order laminar internurance to behavior. But ultimately, you know, and I showed you that a two and a three. So the two input neurons here have very different sensitivities. So one is sensitive to contrast, the other one is sensitive to luminance. And the question is like really where are these like two pathways or waste information from these two pathways combined. So this clearly has to be achieved somewhere premotor. And so what again, what we know is that these neurons project onto different neurons in the medulla here. And they really have sort of like their own target neurons in the medulla, but then information converges latest at the at the level of the dendrites of these directions, like 55 neurons. So we thought that this will be a good start to check. And what we now did is we used exactly the same stimuli that we used in behavior and showed them to flies, why we recorded calcium signals in these neurons here. And what you see here, so we now recorded and calcium signals in these directions like 55 sets for exactly these, you know, sort of like stimuli that I showed you where we interleaved these five different background luminances. And then we run off edges across the fly, we match them to the directional preference of these direction selective cells here. And then what you see is that T5's neurons are contrast constants. So these off direction selective cells exhibit contrast constancy already and show sort of like invariant responses or statistically non-significant responses to these different minus 100 percent global contrast conditions. And again, so far when I talked about contrast sensitive responses up here in the lamina, we generated these predictions from previous data, but here we decided it's about time that we like record these neurons ourselves and look at calcium signals under exactly the conditions that we use in behavior. And what you see here is basically what I told you before is that two neurons themselves don't exhibit this contrast constancy. So they still have this problem that they have like the too low calcium response or too low neuronal response in response to contrasts for like these large dark background luminances. Okay, so that means that, you know, this combination of contrast and luminance information must happen at the latest when information is coming together onto the dendrites of T5 cells and we are currently investigating. And if this could potentially already happen before that in pre-synaptic circuitry. And then the next question is that we spend quite a bit of time on this, the question how contrast constancy is, sorry, how contrast and luminance sensitive pathways become specialized to be so different to begin with, because these are just two first order intern neurons that are sensitive to completely different features of the visual scene. How can this be the case? And this is again work, a lot of work that Katja did. And she now teamed up with Junot Akta, a postdoc in the lab who is coming from punctual genomics. And so we first, our first hypothesis was that maybe it's already in the photoreceptor input that these intern neurons are getting. Because I mean the idea is that in these motion ejection circuits get predominant input from R1 through R6 photoreceptors. But there's also data that report that sort of photoreceptors which are people more often think are more associated with color vision like R8 and can also contribute to these circuits and that R8 might be feeding into advenions. And so we thought okay maybe it's true that sort of like less adapting R8 cells contribute to this sustained component in photoreceptors. And we did a series of experiments to show that this is not the case. So for example here's just one of them. So we know that these outer and inner photoreceptors, so R1 through R6 or R8, express different options. And so we basically tested the spectral efficiency or the spectral tuning of a two and a three neurons and used this as an indirect measurement to ask where from which photoreceptor they get the input. We did this by building lexance in collaboration with Gregor Belusic and the student Marco where we basically take a series of LEDs and send the light via a diffraction grating and through a lens onto the eye of the fly and can like sort of very precisely show them many different wavelengths. And when we do this then we see that the spectral tuning properties of a two and a three are really like fairly similar and or both peak at the point where R1 has its peak sensitivity. We see some difference in the UV range but this is not relevant for what we have been looking at here. So and we also did a lot of genetic experiments where we sort of isolated the different photoreceptor components and I can just tell you that a two and a three both rely on R1 through R6 input predominantly. So then the question is I think an even more interesting one. So how do you sort of like separate these two signals at sort of just one synapse between these photoreceptors and either a two or a three and to just sort of know what sort of like calcium signal we'd be dealing at the photoreceptor terminals Katja recorded responses from the photoreceptor terminals here. And you can see that the calcium signal so the steady state response in photoreceptors also tracks luminance but sort of has a more linear relationship with luminance than than as we knew and who had and what you also see is that the photoreceptor calcium signal increases when the light is turned on. This is the opposite of these laminar neural responses but this is what I already explained to you and this is due to these and histamine-gated originals. Okay so we are sort of dealing with two types of transformations here. So on the one hand side we need to like transform and this like sort of linear um relationship to luminance that is present in the photoreceptor exon terminus to this non-linear preference for dark in at v-nurance and then we completely have to eliminate this luminance sensitive baseline in a tool to get these purely contrast sensitive responses out. And so we asked and you know there are several possibilities um how this can be achieved um and maybe very costly this can be sort of like cell autonomous mechanisms or this can be done through circuit interactions. So to distinguish the two we started to look at L2 first and recorded calcium signals not only at the exon terminus but also in input regions and along the exon and the idea here was to like really see how the dendritic signal is that L2s are you know inheriting or getting from photoreceptors or how they respond to a photoreceptor signal at their at the post synaptic site and then to see if there are any transformations happening along the neuron which would argue for circuit or could argue for circuit interactions. So this is something that we actually saw in L2 so we saw that the dendritic calcium signal is actually not as transient as the as the um calcium signal at the pre-synapse arguing that some transformation is happening along the neurons I just said that I mean of course these can also be cell intrinsic properties but this is something that we then further tested and we can test this by genetically isolating L2 from its circuit environment and this we can do by taking ORT mutants, ORT is encoding the system you get a chloride channel and so basically in an ORT mutant no one should be talking to photoreceptors and then we can use fly genetics to put this channel selectively back only in L2 neurons so the only synapse that is intact here is the L2 as the photoreceptor to L2 synapse again so um what we what I showed you before is that when we show these like different you know steps of luminance or these different luminances to um to flies expressing GCAMP in L2 you see these like one fixed baseline and you see these like purely purely contrast-sensitive transient responses in L2 controls when you do the same experiment and you record from L2 neurons in this L2 ORT rescue condition so again this is like a situation where a tool is sort of ice or where the photoreceptor L2 synapse is the only one that's really present and what we see here is that they still have this like transient contrast-sensitive response but they also start to get this luminance sensitive plateau component and what this argues is that this the um the sort of like elimination of this luminance sensitive plateau component between photoreceptors and L2 neurons and depends on circuit interactions and um then we did the same for for L3 and neurons and so we um we again recorded kites and signals in L3 from the dendrites the exon and the exon terminates and here in contrast to L2 the responses basically look the same no matter where you where you check so dendritic kites and signals are sustained and we don't see you know any transient component in the response there so this argues that further difference between L2 and L3 are likely mediated by um cellophoneness and mechanisms to get at these um we uh yeah uh well we Junite the postdoc in the lab um started to do some functional genomic experiment and um isolated um RNA from like either L2 or L3 neurons we actually use GCAMS so we use exactly the same experimental conditions that we use for imaging for sorting and then we obtain cell test specific RNA sequencing data sets um I should also say that this is something that was previously done in Larry Cipersky's lab but during development and we initially looked at these data sets and then decided that it makes sense to us to redo that in adult flies and then what we did initially we initially we thought okay now we just you know sort of like and look at all the um molecule classes that you know can mediate these like different properties of channels transporters um anything that has to do with synapses um and look at those and um we screened like many many different candidates there so I think uh Katja and Burak tested a whole of I don't know 50 or so different um candidate transporters and channels that are differentially expressed um and we didn't find anything that's like really strikingly changing the the properties from a contrast um to aluminum sensitive cell or vice versa so what Burak found is you know some like sort of my my daffino types so for example this is um data where Burak found that um two potassium channels shaker and shawl are preferentially expressed in L2 neurons and when you um take them out in this case um using RNAi but you also use pharmacology to do that um then you see that you are sort of like changing the temporal dynamics a little bit and you make them a little bit more sustained yeah but again what we were hoping for is like really genes that really differentiate these like you know contrast sensitive at true from these women and sensitive at reinurance and so after you know not finding anything in like um yeah a lot of screening we thought okay maybe there isn't like a single a single gene like a single channel or something that like really is alone mediating and these these responses which I think makes a lot of sense so um to still get at these differences we decided to look a little bit like more upstream molecularly speaking now and instead of like looking at you know synaptic genes we looked at um transcription factors that could specify a whole cohort of um of proteins that could then like mediate L2 or L3 properties and what we found also what we first um looked at is a um transcription factor that has had already been known to be L3 specific it's called DFAS FOIRM and during development it was already shown to be um basically only expressed in L3 and both at the RNA level and at the protein level and um when we looked in our adult RNA sequencing data set we see that um so here um these are like the pink shaded areas are the um axons is that they are also expressed in L3 um in adults but no not in L2 and um so DFAS F was shown to be required for the set a specific layering of L2 and L3 during development and we thought to um ask if it maybe also has the function in mediating the the physiological properties of L2 and L3 neurons and I should say that this is um done was done in very close interaction with um the lab of med PCOT um who is was like really a fantastic person to have around and discuss science with and exchange like tools and ideas um who unfortunately lost this battle to cancer like a few months ago and um so the images that I'm showing you here um were generated by Ivan Santiago a PhD student who now radiated a postdoc in um in the PCOT lab okay so um what what we are doing here is we are basically looking at um individual clones um um at three neurons that are dominantly labeled with G-Camp and so what you are seeing here is when you record from the exon terminals of these individual clones then you see um again how um sustained at three neurons are and how they track luminance actually in this genetic context so this is a fly that has a lot of trans genes and the they are a little bit more spiky than they normally are but these responses are still totally um luminance sensitive so these are at three controls if we now mutate these and neurons to um lose a DFSF so this is now a dominantly marked homozygous mutant DFSF clone in an otherwise heterosegous background then what you see is that these um DFSF mutant at three neurons um have very different physiological properties so they sort of like become spiky and lose um lose this like shift in in baseline so they sort of like start to have one baseline and really start to resemble um at two responses and this is something that we can quantify across like many cells um in many flies here so they really start to to and look more like at two neurons here and have this like flat plateau response and if we explicitly test contrast and luminance sensitivity using these AED step stimulus that Marvin brought to the lab what Katja who now did the experiment found is that you really turn and uh previously luminance sensitive um neuron into a contrast sensitive if we learn and this is shown here so these are control data where I showed you before that at three controls and show very similar responses to nearby luminances regardless of whether this was the A or the D step this is no longer true and in DFSF mutant at threes and you see how uh very similar or indistinguishable the responses to these six different 25 percent contrasts um become in a DFSF mutant okay so to summarize um this part um we are sort of on our way to working out the different signal transformations that happen in the um lamina or at the photoreceptor to lamina synapse and what I showed you is that a large part of the difference between A3 and A2 neurons um is established in a DFSF dependent manner and that the elimination of this um baseline in A2 uh in normal A2s and is further dependent on on circuit interactions and with this I'm almost coming to the end so I just want to highlight that the the challenges I introduced today um are of course not just the challenge for like the fly visual system but a common challenge for visual systems in general so you know I like illustrated this by showing you two examples of dogs catching balls and flying flies trying to escape predators and um again I just want to like uh highlight uh data from Jeff Diamond's lab who again spoke here two weeks ago um that showed that luminance information is also present in the virtual version of post synaptic of photoreceptors so they showed that there's a transient um component in for example A2 amicron cells that um uh uh is sensitive to contrast and to sustain component that like really skates um with luminance so I think it's really um an appealing hypothesis to like I think that this luminance information could um be there to um ensure contrast consistency in in many different um visual systems so to summarize I showed you that luminance information ensures the accurate estimation of contrast and dynamically changing visual environments um and that this luminance sensitive pathway can adjust for both over estimation of contrast which is new and I just showed this for the first time today and for the underestimation of contrast which is something that we previously published and then I talked about the cellular and circuit mechanisms of this luminance space scaling of contrast computation and I showed you that within a circuitry this happens um presynaptic or onto the dendrites of these direction selective cells I told you that the luminance information from um contrast sensitive A2 is eliminated in a circuit dependent way and that the DFSF transcription factor is required to mediate these physiological A3 properties and with this I again want to like highlight um or I think all of the people that have done this work I think I mentioned Burak and Katja and Marvin and Madura um throughout my talk um I unfortunately didn't have the time to um talk about the work of um other students in the lab like Miriam and Louise and Juan and Sebastian and maybe some other time and I also just really want to thank um some our excellent technical supports of Christina Jonas and Simone um yeah for the the assistance I think I mentioned um our collaborators and I just want to thank my colleagues at the University of Mainz um so this is like a really newly forming neuroscience department and um it's a lot of fun uh to be here and so thank you all for for the great first year here and um I also want to I should probably um acknowledge funding so we were funded by the Imanuta program of the DFG Collaborative Research Center in in Göttingen in ERC and now by the University of Mainz and um I also want to say that we are hiring postdocs and PhD students so if you are interested we are now located in the beautiful city of Mainz in a newly built bio center and again I think this is a great place to do neuroscience so um shoot me an email or contact me um if you are interested in this okay thanks for your attention I'm happy to take questions thanks for the talk Miriam um it's a really nice story I would say um I think um we have yeah what 15 minutes to take questions um people have a bit shy on asking questions um so far um but um I was um thinking about something that maybe also is interesting for other people um so when you speak about these turning behaviors um maybe you could explain a bit more what the actual behavioral background is because normally flies are not sitting on a ball and seeing kind of bars moving in front of the eyes I would say um okay so um yeah I mean I mean this is like clearly a very simplified setup for like you know showing like something I mean I talk about like dynamically changing visual scenes and and stuff and I mean everything that we have been doing is like show like sort of boring edges um to fly so this is like um true what is true is that this is like I think a nice system to um to illustrate that you know how um behavior responses uh scale is contrast and um yeah so I mean I think it would be super interesting like to bring this into like more real world scene statistics soon and maybe start to look at um natural scenes and stuff and this is something we are like definitely interested in doing in the future I think I mean I think so far you know um I think it was I mean I hope I could convince you that we could like learn something from like using these uh you know very simple and visual stimuli to do this but I think now is the time to to start and look into this using more natural stimuli there's one question um this if the trend and responses that you are seeing if they are all stochastic which means like suggesting there is a spiking process underlying or perhaps physical release or ability or are they all always the same okay so what I didn't say is that these um uh non-spiking neurons so you know very similar to the periphery in the virtual visual system these are all neurons with um graded potentials so um yeah so what we are basically seeing is um yeah what we are measuring is like um the the the pre-synaptic calcium which will like turn into a release of neurotransmitter and this you know sort of will give graded information about the the the input that these cells are getting and um so in terms of stochasticity I mean these are like really robust responses I would say I mean I showed you data for um individual exon terminals in which you can like basically really see how and the responses um gave with contrast and it's fairly easy to record from these neurons and you know sort of like just by looking at these like single single cell traces even you can like really see that they gave with contrast you know if they have a luminous sensitive plateau um or not what we typically do is we record from um many neurons at the same time so we can like and usually like record 8 to 10 exon terminals um in one fly and then also across um average across many flies but you generally get a really good sense by by just looking at single single neuron responses. The question is um if there's any evidence for axo-axonic inputs at the L2 terminals and do you think that um that this is the source of differential response you observe this to L2 cells? Yeah so this is a very good question I mean it's definitely so what we have in the fly visual system we have um EMC reconstructions that basically um you know really like tell us which neurons are there and who's talking to whom and by how many neurons these um these cells are connected to each other they are um you know always like some um axonic inputs as well you know and they also get like a lot of inputs and in their dendritic regions there are some reciprocate synapses and determinants even so I think there's like a lot of um a lot of room for like circuit inputs there um what I mean what we have done so far maybe I just um talk a little bit more about this question so what we um what we have done so far is that we um have uh sort of I mean we wanted to understand where um like who's basically like you know responsible for this like elimination of the of this aluminum sensitive baseline in in L2 neurons yeah so and the most obvious candidate um seemed to us that this could like be coming from L3 neurons because L3 have like aluminum sensitive components so and they don't make direct synapses but you know they are some like intern neurons GABA-ergic intern neurons also where we thought okay you know this like would be relatively straightforward to feed the signal back onto L2 neurons and we have imaged um L2 neurons why we uh silence at V neurons and we don't see any change there so it doesn't seem to be coming from there so that's something where we still don't know if there's maybe someone else who like is grasping photoreceptor input and then like feeding this back and this information back onto L2 to eliminate this baseline or something so this is something we are I'm still investigating there are a few more candidates that we can test based on these um uh based on these uh EM data um yeah and again nothing we have done so far I'm not sure it has to be exo-exonic I mean it could totally be dendritic I think yeah so and then you can like have some cell intrinsic properties that also um do this further um to see where this baseline sort of is eliminated is something you know we would have liked to do so I mean this is something we like didn't manage to achieve by recording the dendritic signals for these stimuli as well it's like really hard to get to the dendrites because they are sort of like you know just yeah hidden underneath the retina I mean the exon terminus of these neurons you know sort of like project to the next neuro pill in the medulla it's really easy to get there but to perform calcium imaging experiments sort of like from from the lamina itself is very tricky so Katja had to try very hard and like managed to get these full field flesh responses but uh it's impossible to show like many stimuli there and get conclusive and sequence uh responses so this unfortunately we can't answer there's another question um do you think that the l3 neurons are slower because they're aluminum sensitive or is this just a coincidence um yeah so I'm not um 100 percent sure what you mean with slower so um increase like the initial increase um well I would I mean if you talk if you're talking about the initial increase in calcium then that is something that we see for some stimuli that is true so uh well observed in the data I would say then um and uh where this is coming from is um something I can't fully answer I can tell you that this is something that we were actually messing a lot with when we were modulating individual channels so when we like you know sort of like trying to find something that is um changing uh lumen and sensitive at re-response into a contrast sensitive at re-response by taking out individual channels we would often affect sort of like just the kinetics of this um initial onset response so that seems to be a cell intrinsic property um and there are actually many different um channels that sort of have this phenotype when you knock them out where you're sort of messing with the initial kinetics but yeah I think the question was if l3 neurons are slower and they're responses then to the general yeah so I think they are a bit and we it's nothing we have looked into and they're careful what I should also say I mean you know there will already be about some people who are like all you're doing is just measure calcium and how can you talk about fast as long when you're just measuring calcium and so what we also did is we um and it's in the paper so we also thought to look at this like very initial um fast response in both of these sets um because for example what you might have noticed that in our calcium recordings and photo receptors for example we only see this like lumen and sensitive plateau and we don't see like the fast you know contrast sensitive signal that I showed you in the voltage recordings from photo receptors um earlier so we wondered if um l3 neurons for example might actually also have like a fast contrast sensitive component that we have been missing whenever we talk about this lumen and sensitive calcium signal and so what we did is we expressed asap 2 um so a voltage genetically encoded voltage indicator in both at 2 and at 3 neurons and if we do this then in at 3 we also like see um a very fast uh contrast sensitive like spike yeah not spike but fast response basically uh and sort of like the functional relevance for that um was then that we wondered if you know um so it's really like it's basically a three just the pathway that scales contrast computation in other pathways or can it itself have like a more active role in in motion computation for example by having a fast contrast signal um itself that we sort of have been missing in in in these uh calcium recordings and so we try to get at that functionally by um asking if we only have a three neurons present which we can do in these odd rescue experiments that are explained to you then is this like sufficient to um respond to motion and it is so if you sort of like have a visual system where you have only photoreceptors talking two or three you can actually get behavioral responses to motion so you don't need these like dedicated contrast sensitive pathways um for that and what is also interesting is is that um this is like something that is um specifically prominent like in contextual dim light again which is like then like then totally matched with this other function of a three that you are like need this pathway to do well in these like specific uh luminance regimes okay maybe one last question um and this is something of course that we that we um also try to investigate I guess I know back from the time I was in the lab but uh what's with this hypothesis that a three gets less input from photoreceptors than a two do you think that has anything to do with this difference in behavior so this is really the question that Martin Marvin should answer because this was his master thesis that we thought hey you know maybe it's this is like the reason that they are sort of like saturated earlier and this like makes up these differences um and I guess Marvin will agree with me that we like couldn't find like any uh strong evidence for that um that's quite tricky to do so it's quite tricky to do like yeah it's true so what we and so what we were like messing with is like you know sort of like um uh the amount of receptor on the post-eneptic side you know there are some ideas on how you can like actually change um how many like um synapses a neuron is making on a on a onto a post-eneptic cell which is then hard to validate but sort of like with the different like genetic um tricks that we tried we didn't find anything so I wouldn't say we can say it's not the reason um I would say um we at least so far have tried and have haven't found evidence for it but this is a very interesting question I'm actually also maybe I should add to that I'm not 100% sure that the idea is that like at we is providing many fewer synapses into like these pathways actually holds true if you start to consider and potential white feed neurons yeah so I mean like yes how many synapses does um for photos have just make on to at three and at three on to the downstream neuron but then you know they are all these uncharacterized white feed neurons which could have a very interesting role for this um for this like contrast constancy um idea so we will see okay I guess I think again for answering all the questions and for giving the talk and to everybody who's still watching if you want to um if you want to ask more questions to Marianne you can join us on the zoom I will just post the link now and yeah feel free to join us um yeah I guess that's it thank you so I keep the live stream open because if people want to see the the link the link will disappear look at the questions myself I was sort of uh okay I should have checked to see who is asking a question yeah I should maybe I should have read the names um it was a bit challenging to to read and ask answer some more questions meanwhile so if people ask the listening I can add uh it should have told me asked at 50 percent where about contrast there was a behavioral response at this stimulus offset where can the behavioral response come from um yeah this is totally true it's something that we see um and uh we don't know exactly where this is coming from um it for sure sort of like depends on the adaptation state of the fly so it's something that we can alter if we change the interstimulus interval so for everything that I showed you we keep sort of like an interstimulus interval at one second where um it's completely dark and if we don't do that but we sort of like use any other interstimulus interval um we um we lose that um response um there was one more question um on top of the autonomous function of l cells do you think that the different input is also important for the difference between contrast luminescence function pretty sure I think on top of the cell autonomous do you think the different inputs also important for the difference between contrast and luminescence function yeah I mean for sure I mean this circuit input that l2 is getting if we eliminate that they um they are not sensitive to they start to become sensitive to luminescence as well so that is something that is um for sure also contributing there um you're all right it's very hard to read and answer questions at this point I think the last question that it says um do you think that this contrast insensitivity activity could extend well to other types of behavior the s cells may have specialized circuits for computing motion the s cells ah the s cells the s cells Marvin you have to can you read this question once more so do you think that this contrast insensitivity could extend well to other types of behavior so I'm not 100 percent sure I understand the question because what I was trying to convince you that the behavior is very contrast sensitive yeah so um even if contrast sensitive neurons alone can't handle that and that you sort of have a mechanism to fix that so the question would be like if you silence l3 and then you try different behavioral tasks yeah so this this I definitely think it will be the case I mean I I think you know there are many behaviors which visual behaviors which sort of like scale with contrast or which initially rely on contrast so I'd be very surprised if that was like an um a phenomenon that that's only there for for motion guided behaviors it's the only thing we've tested so far and because it seemed like very easy and there these like very well characterized responses to motion um yeah so Tom Barden asked is it possible to measure the actual release from l2 and l3 yes that should be possible and we haven't done that yet it would be really interesting because what we are very interested in is like sort of to see the time scale and for which you have this um luminance based scaling right because at some point adaptation will catch up so how are these two balanced yeah so when adaptation catches up you shouldn't like have an entry response anymore and we were trying to get at this by uh recording calcium signals in uh at free neurons and our idea was that you know maybe naively that their response will just go go down at the time scales um at which like adaptation goes up and we don't see any of that so and it's unclear if that just you know really is true that at free neurons aren't adapting um or if that's something we just haven't seen in um in uh in calcium imaging so we are very much interested in measuring release actually now that I say that I remember that we tried once and I don't think it's like as beautiful as you know what um as beautifully possible as in other systems so we have very high baseline um um expression of uh so for example it's sent up to Florence um that gives us very tiny delta f over f so that makes it a bit tricky but now that we have like all these um indicators for that we can express on post-synaptic sides to look for glutamate and acetylcholine and stuff that should be possible so both at 200 we are calling NERGIX so if we um potentially this is like an experiment that we could definitely do and measure um yeah acetylcholine release yeah so apparently they're working now quite well I don't know if you saw the talk by Gotem of Atamani so he's gotten that to work in the mouse the acetylcholine sensors or yeah yeah so maybe there's room for that to work that would be great yeah and we have them and stuff we just haven't done the experiment so yeah I think those were all the questions from the YouTube yep I will keep bugging you though so you know invertebrates we're we're quite obsessed about counting and now I see everyone hey Michael hey man hi Simon so you know invertebrates we're really obsessed with counting cell types and then we're giving them names and then but I guess in slice you're a little bit further with this right so if we now mutate that's Simon I think that's someone's YouTube is on maybe it's I think it's Simon maybe you could mute I can mute yeah so sorry so um so what's the situation in the flies in terms of cell types so how many L is there a one a five L seven where does it end it ends at five and a one two and three are the ones that are getting direct input from photoreceptors so okay manageable it's kind of weird right that the other ones like l4 and l5 are still weird right like like evolutionarily they're weird like what why are they in the lemon in the first place when they're basically medulla neurons right there's they're just weird yeah Simon looks skeptical mutaton so I can't say anything at the moment so marion can I ask again just just about timing because you know it's like my own personal obsession so this is in no way like a shortcoming of your beautiful work but I just you know I've always been interested in like you know the the like at the limit right when it seems like the thing that makes vision so cool is that it's actually useful on the timescale of milliseconds right especially for like escape behaviors right and so I just wonder it's not obvious to me which is harder right to make an accurate contrast estimate or an accurate luminance estimate they both actually seem kind of hard right if you have you know 50 milliseconds something is sweeping by and you've got to do it right it just seems weird to me that l3 just seems slower right at steady state it's definitely slower right it can't track changes and in in our original lamina paper when we science l3 like the deficits we saw were on slow stuff so that's why in my head I just have in mind like l3 is a slow neuron it's a immediate slow vision it's not obvious does it make sense though like yeah I mean so yes and no so I mean I agree that it's like so I mean yeah it maybe doesn't track like fast changes as much what what we have seen is for example how long does it take to get to steady state and exactly the same and I mean yeah you guys have like looked at this more carefully where we where you guys have shown that it's required for slow motion so that that would be something that you would be looking for in this like contrast sensitive component in l3 right so which is something that we basically don't see in calcium so I can't really answer this question like at the timescales that you are interested in I think so then we would like have to do voltage imaging for all of these right no I think you can still do I mean I think if you just do like I mean this is if I were like you know to do follow-up experiments and your experiments in my lab which I won't do don't worry I would just do everything over like two or three orders of magnitude in timescale right just make all the stimuli super fast you can still look at the you know slower calcium response to a fast event right that's not a problem so you would just be looking at the filters and see how they look like it just seems like the hardest thing about vision is not the steady state stuff right it's like doing things over like super short time the steady state stuff is doing cool stuff you know like yes right no but yeah no no but your protocol is like it's a cool mix of the steady state and the transient right because it's just this edge sweeping through right it pushes on both yeah it's true yeah so I mean I can't answer I mean we have we actually like really never like you know measured filters or like very fast step responses of these neurons so I really maybe should we should really add that to like the standard step of stimuli when we are testing more like you know of our I mean we still like hope to get at the mechanisms that are downstream of this deep as F transcription factor and now that we have like this handle sort of go back to the membrane and maybe you know modulate a few things where we know that they are now there I mean we like sort of you know that this RNA sequencing now also in the indifference epi clones that are asked we can like go back so maybe when we like test candidates they are and we are we like include that and so maybe I can answer this question next time is it easier to ask questions about timing with the behavior it'd be really cool to do maybe we've already done speed different speed sweeps but across these true you should like yeah I mean that's a good idea because that's something that Michael could pull out in the in the touch study right like these differences in like requirement for different speeds for a two versus a three so that would be interesting yeah it's not so as you know it's not so easy to screen a lot of candidates doing behavior and I mean and then look at speed tuning right I mean the beautiful thing about these like calcium responses that they are so robust yeah so there you just look at a few exon terminates and you know if you have a phenotype or you don't so yeah but now that I have armies of like undergarments here in mines I can do that can you say something else about do you have like so for going for targets downstream of a transcription factor that's presumably like I don't know hundreds of genes or something or many dozens yeah like do you have uh is idea to go for like top candidates like so what we what we did yeah so what we did is my interest so we have still things that are normally differentiated expressed between L2 and L3 and now we did RNA-seq between L3 neurons with or without DFSF so we know which of the genes are differentiated and regulated in these neurons by DFSF and we are currently complemented this with chip seek to just like see who is like a direct target and then what we did is we sort of like group them functionally so now you know we are still interested in like voltage gated channels and channels in general and transporters and now we sort of like have clusters of um of potassium channels where we see okay you know there's like six three of them are up-regulated in L3 and three are up-regulated in L2 and generally the ones that are in L3 are more of the ka type or something like that so you know that I think gives us a better handle so we are like yeah so they are like a few c and g channels which would make a lot of sense for like having these very sustained responses and stuff so that's what we are now looking into to sort of like see based on their huge large lists are they like groups of genes where they put like function together and make sense and then yeah I don't know it will be hard to get there right I mean I don't I don't know even with flip stop and stuff like to do a quadruple flip stop potassium channels or something that we see but it also just seems like we love right our methods are best if there was like one or two channels that explain huge effects we're totally set up for that but if it's like you know 15 channels together like how do you do that experiment so we actually try to do the opposite but haven't succeeded so far and this is like maybe the craziest idea I've like ever followed I'm not sure I should tell you before we even started with this experiment but I tell you anyways so like uh you know what I was thinking is that this might be a very cool test bet for like you know sort of like Yves Marta type of ideas where you have these like huge parameter of like M channels that like mediate certain properties so what we were thinking now that everyone is doing single cell sequencing to get at cell types what we should do we should like do single cell sequencing only of L2 neurons and so sort of like see what are sort of like the functional clusters that you get within a certain cell type and you know maybe I don't know can you guys get at the depth of of uh yeah sort of like and it's actually something that wasn't only inspired by like you know us not being able to figure out you know what is like mediating these differences but also by the fact if you for example for these um you know shawl and shaker things when we either do pharmacology or we do RNAi it's really like you see a phenotype or you don't see a phenotype and this is completely different from fly to fly so the variability there is much higher than for like any other genes so if we like take our DFSF out of L3 neurons they are always contrast sensitive if we do the same for any of these we now in the lab call them terminal genes yeah then they might or might not be required which makes a lot of sense if you like look at Eve's work right so that's like sort of maybe an experiment system in which you can actually get at this sort of variability. Hi everybody I'm sorry I was late joining in I had a lot of trouble with my computer um I always go back to the fact that um at least in our hands in caliphora it looked as if the on transients and off transients in L1 and L2 were produced by modulating the chloride current the postsynaptic chloride coming into the L1 and L2 and this was done by measuring conductances and reversing the potentials in interest of the recordings and so really robust robust result and the puzzling and that what that suggests is that the background subtraction which isolates contrast is being done by either inactivating progressively the histamine gated chloride channels or by regulating transmitter release from the presynaptic terminal now the interesting thing is that L3, L1 and L2 share the same neurotransmitter at tetradic synapses so if I was going to look for something in L3 I would look for something about a something that would modulate for example its aught receptor I thought the aught receptor experiment was really cool by the way I really like that yeah so the answer to this is that Marvin really should have stayed in the lab because this was exactly like what he was like after when he started you know that we also thought I mean what we did is we like played with like these aught receptors we overexpressed them see you know do levels do something and then we thought okay what we really should do is like look for things that interact with this and I think you are absolutely right so maybe that's also like a good um a good idea to like look into this again when we are now like scanning our candidates to sort of see you know what could potentially interact with with these histamine gated chloride channels you're absolutely right yeah it's very good value of the comment I also agreed with Michael's comment I have a cut I think you should try some other stimuli to get behavioral responses and to get physiological responses because these these transients are just dreadful things they're incredibly non-linear particularly in monopolar cells and they've got all sorts of different components in yeah and so it's really it's really difficult to interpret what's going on yeah the deeper you dig the more confusing it all becomes and in the end I gave up but with these transients if you gave up I'm not sure I'm going to try but so like I mean with these transients you mean like these like very fast on transients for example that you only see when you measure voltage right or yes well I mean it's partly that but it's partly the fact that um you're on and off isolating stimuli which are receding edges which actually cover the whole receptive field the hedge goes down all the way down the receptive field and on one side it's white and on the other side it's black or gray they are very unusual stimuli in space and time and they may do you may get some quite weird behavioral responses to that not to say physiological responses and because they're wide field they tend to confound um changes in spatial receptive field with all the other changes because sorry whose wide field your stimulus is very wide field yeah yeah and and we know that at very low light levels the there's spatial pooling in the lamina of I think I mean I think that is exactly the reason why we see this dichotomy you know where I think that that that could be it and so there's there's some different mode of processing going on at really low light levels yeah I fully agree which is very interesting in its own right yeah I mean what we are definitely doing at the moment is to try this with more local stimuli and also see you know under which I mean sort of how how wide this spatial pooling is and I mean like really the idea would be that you um I mean also like you know what you don't want to see you don't want to use luminance information to just like even out all contrast computation that would also you know then you basically like yeah so what you want is you want like something that like where your luminance sort of phase scaling um is a little bit more global right so and that's I hope something that we can get it by using like stimuli of different sizes and then maybe also and this is I mentioned the DM neurons like at the end right so they could be like really the circuit implementation of this where you have these laws in medulla neurons that spend several columns so there you could like combine signals from several atrees to like use that to um compute contrast constancy basically and then you still keep your local contrast signals um from a tool yeah but I see that you are there and all these things are possible all these things are possible that the the trick is to design an experiment which can tease them apart and that's a big challenge but when it works it's fun I mean can I can I just suggest like a fun I mean one of the challenges with the behavioral silencing experiments is for the earlier neurons like the monopolar cells it just it's hard to interpret and even make use of a particularly subtle phenotype right so I think you revert to pretty you know like things like just large luminance contrast changes maybe a little speed tuning right you you can try a few classes of stimuli but it's it's it would be hard to do much with more subtle things like an experiment that I think would be really beautiful is find some if you wanted to get away from motion detection right find like a different output pathway like a looming sensitive neuron which is pretty close to the kind of stimulus right that Marion is already showing right but the readout is turning so then do a looming sensitive and then ask how that you know feature detecting neuron like say like an l4 or an l6 or something is modulated by manipulation style 3 like that would be very much in line with the kind of stimulus regime you're already working on you can imagine flies really want to be able to detect lumen stimuli at all ranges of you know luminance without the like very particular effect of you know things which are optimized for motion processing right like because otherwise I think just using behavior alone it's going to be quite hard I think if you're like you know mutating a channel in an l3 right like that's going to be very hard to read out yeah I mean I think all of these things I was more looking at thinking of physiology but yeah then it's hard to get at this very fast component so yeah I think it's a good idea I think Norma got a question Norma you wanted to say something yeah I wanted to ask something because you showed these like really nice like attenuation of the on responses in the l2 cells I think it wasn't an attenuation so basically like in the beginning you have like a response to the off but also to the on and then the on response attenuates right do you know anything like why that happens or like is there some kind of inhibition in the in these very upper layers what do you mean with attenuates like so the response is there but then it kind of flattens out so I mean why do these l2 neurons have transient responses in the first place or no no no no more like um so uh more that you you basically you have a transient response to the to the on but like later on it's not there anymore right so they so how do these cells become rectified or yes yes well I mean like so l2 cells are like well somewhat linear so you know this has been like sort of like fairly extensively looked at and like sort of re- rectification in these on and off pathways is happening like a synapse further down and and yeah so so the reason why you sort of like see only a tiny um uh on response like sort of negative you know like never negative calcium signal there or decreasing calcium might partially be cause of like calcium indicators and low baselines in these neurons and yeah partially that there's already some non-linearity and within these cells and then sets but does someone else want to add something to that okay so so basically it's not not really known what what makes them like like because they are linear in the beginning right but then like the on response kind of is I mean rectification you often get like sort of for free at at synapses yeah I mean the question is I think oftentimes more how do you keep like a how do you keep a system linear so if you want to do this like but I mean like maybe I'm also reading the plot wrong but it looks more like it actually at the synapse it's still like non-rectified and then it becomes rectified like wow so all of the L2 data I showed um uh like all at the exon terminus so you're meaning like when I like then writes exon exon terminus yeah so I mean there I think well actually I'm not 100% sure how well you can tell from this what is actually like the on response or like a non non-complete decay to off yet or something so that's yeah so maybe then that one would have to do was like more like an intermediate grade I mean here with these are just full feed fleshes so I think from there it's hard to very hard to talk about rectification so I don't know okay thanks really nice talk I don't have any specific questions I just wanted to say hi and we've had congratulations for your awesome job thank you and I guess for almost getting married last weekend oh yeah I know it's so crazy it's been good though we still celebrated we made a cake yeah and nice to meet you Madura very nice data same year I saw you on the first day I only heard about you so far so this is Madura who did I mean I'm pointing where Madura is in my screen but different for you so this Madura who did all the um LMC predictions and behavior experiment that's beautiful okay okay I actually have to run but I wanted to say hi fantastic talk I really enjoyed it thank you hi everyone hi hi okay any other questions comments yes a few people which don't show their videos Natalia do you have something to say Natalia has the good spirit of our lab okay hello Mario and it was a great talk actually yeah I'm not like I'm not showing off today so it's okay next time but thank you for your talk and thank you to Marvin for hosting yeah it's actually nice I mean I haven't met most of you people so it's nice to see some faces maybe nice to hear your voice Natalia thank you so much bye bye okay Tom disappeared Marvin you should try these things in the word professional yeah first time to get my own recordings oh but I guess with the MEA it's much more easy to get the um the faster sponsor yeah I mean I guess if there are no further questions and well I'd just like to point out one that you probably know this anyway sorry to jump in but in in the hardy and vextrum paper which is the only paper which really discriminates between L1 and L2 and L3 using intracellular responses and where they mark themselves it's a really fantastically accomplished paper as you'd expect from Roger and in there they found very consistently that L3 always had a larger plateau response than L1 and L2 which fits in very nicely with with what you're saying it's exactly what you would predict if L3 had a bigger luminous component it's absolutely true and you know I mean they also like you know sort of assign different like potassium conductances to these neurons so all of that makes sense and actually you even see that like in the and in like you know these um which paper like usitola like you know sort of like some of the like previous papers where you like sort of you know see L3 type neurons in other species you actually if it's like never described but if you look at the voltage traces and you sort of like draw a baseline in there you always see that like even the L3 voltage response never returns completely um to to baseline yeah so I actually had like a backup slide where I was like going to show all the like you know um previous examples where people had like assigned L2 and L3 types to to their data and it's it's it's always there like you know it's not that we found something that like wasn't in the data before just you know but I mean they definitely like in voltage recordings always have these um past uh contrast sensitive peak which is like clearly dominating these responses but yeah it's not something that we are now reinventing in drosophila it has always been there in the data but there's always this big problem I think we discussed this before that there's this big extracellular potential in the laminar that's induced by light so you never really know what the baseline is and really the only baseline that counts is the baseline that's seen by the presynaptic terminal of the monopolar cell in the medulla and that's where we are recording electrical compartment so this is a really interesting tricky problem yeah because you actually don't know what the membrane potential is and when you correct for what is your best estimate of the membrane potential in l1 and l2 recorded in the laminar by subtracting away your best estimate of the extracellular bias potential you find there is quite a reasonable plateau component ah okay so that's in the if you look in the vextrum Laughlin paper from 2010 it's in there right so it's a good thing that we are recording a big deal of this because it's all sort of dirty secrets all coming out you know stuff we don't know that we ought to I'm glad this is all consistent with your previous recordings and I guess you know it's one of the selling points for measuring presynaptic calcium then because we are past all of these like potentials in the laminar I'm just a skeptical old-fashioned voltage person you know but but as you point out there's not like one voltage right you'd like voltage in all compartments and you'd like all the references so you know one electrode is also an imperfect measurement right all right I don't I don't want to cut this off but I think no I'm ready for if they're not okay thank you all for coming yeah thank you and thank you thank you very interesting talk thank you bye bye okay bye