 Okay, it seems to work. So today we have Tom Baden from the University of Sussex in the UK, so a very quick introduction about his background. He was trained at the University of Cambridge, working on the bipolar cell in the fish in the zebra fish, I guess. And then did a postdoc at University of Turingon in the lab of Thomas Euler, now moving to the mouse visual system and became professor and started his lab at University of Sussex in 2016. So you will see the research explore the color vision, the evolution of vision in different animal models such as mice, zebrafish, sharks, for chicken and probably all the animal you can find in the zoo. And also he's very involved in technology development. He is using many techniques such as individual to photomicroscopy, electrophysiology, BFU, computational neuroscience. His research is very well funded from the European Research Council Welcome Trust Biotechnology on Biological Science Research Council and many other agencies also got many prestigious awards such as the MeND-UP and Nature, Jung and Rupert and Investigator Awards and the Welcome Trust to the Investigator Award. So his research already have a strong impact in the community with almost 4000 citation and almost 60 publications in journals like Nature, Nature and Science, Nature Review Nature Communication and so on. I'm a board member of many organizations. And finally, I would like, I think it's important to mention also that he's very involved in some social initiative. The first one is a founder of Open Lab, where it's oriented on open source, open science projects that make science affordable for any scientist of the planet. So you can download things and really print your microscope, micro pipette, micro manipulator, whatever you want to do. And also is a co-founder of Trend in Africa. So it's a nonprofit organization dedicated to develop neuroscience education and research in African country. So they organize for example, summer school equipment donations and scholarships, et cetera. So many things. So maybe I talked too much. I will let him to continue this. And I hope you will enjoy his talk. Thank you, Thomas. Yes, thank you very much for the invite and the introduction. And I'm sure I'm going to talk too much now. So that's all fine. So yeah, so what I'm gonna talk about. So I, the lab over the recent few years has been taking a tour towards evolution and the way from looking at one model system over extended periods of time. So the question that we really want to address if we could is if we look at two animals as you're on the slide indicated with admittedly complex brains that are understood to some degree, but not a deep degree, I would argue. How did it come to be that those two brains work at all that they have similar functions? What are inside those brains? What are the circuits that make them similar? And I think that's a pretty big question which sort of goes also in the direction of, what are we in the first place and how are we here? The problem though is that if you look at those two brains that's really hard to do because I mean, I guess what the illustrator here has done has indicated a bunch of regions that they think might be functionally equivalent. Are they really equivalent? How close can we take a circuit that's in this orange area here and compare it to a circuit that's inside this orange area here? Would that be a fair comparison? Would the origin of those cells be the same? So these kinds of questions, we're not really equipped to answer and probably the answer is no anyway. So to make slightly more useful headway in this big question, what we've been doing is we've been looking at one of the circuits that most vertebrates have which tends to look more or less the same no matter which vertebrate you look at and that's the retina, right? So what we've got here is just two schematics, one from this primate, one from this pigeon that approximates roughly what the retina looks like and it kind of always looks the same, right? You've got a bunch of photoreceptors here. You've got a bunch of ganglion cells here that form the optic nerve. In between, you've got interneurons that do interesting things and they connect to each other in approximately predictable ways at least within the big picture. So they're quite similar in that sense but they're also different in the sense that, well, this circuit is not gonna do the same job as this circuit because for a start it's got way more neurons and the neurons are different, right? It starts at the input and ends at the output. Everything is a bit different. So of course, what this animal sense to the brain is gonna be different to what this animal sense to the brain. And just to make that picture a little bit more broader because he has a bird and a monkey but you can really play this game with any vertebrate you find including even the lump race, right? So if we go here to the old end in terms of phylogeny, you can see we're generally talking fairly low complexity retinas, right? We've got large cells. These are not very well understood. Then the second we get to fish we start to get high density retinas and amphibians to some extent too and reptilians also really quite high density and then in bird you've got kind of the pinnacle and then a sort of this drop again towards the mammals. And I'm gonna talk about a little bit towards the end about what I think that means and how we can maybe think about it. But fundamentally, as I say, the structure is conserved. You've got the photoreceptors, you've got the output and you've got a bunch of interneurons with similar names that all of these animals share, right? So you can start to look for homologies. You can, for example, we know that there's very strong homology for the photoreceptors, right? So for example, what we call the red photoreceptor in our own eye is the so-called LWS long wavelength sensitive photoreceptor named after its opsin. That's the same one that's the bird has. It's the same one that reptile has, frog. It's even the same one that the lumpy has, right? So that's a homologist cell. The same exercise gets harder if you go into the horizontal cells or here into the, especially in the inner retina, amacrine cells and bipolar cells, you find some homology, not as much as in photoreceptors. And then at the level of the ganglion cells, it starts to be a bit hopeless in terms of saying, even between something like a mouse and a primage that these ganglion cells are truly the same thing. So we've got a gradient of evolution in a way, right? So we've got the stuff that's very similar, which is the photoreceptors. And then the deeper we go and that trend continues into the brain, the deeper we go, the less homologists it gets and therefore we can learn something about different levels of organization across evolution. So what do you have to do? Well, in order to study evolution, then you have to study more than one species. We know quite a lot about how mammals work. So I'm not gonna talk about that. We don't know so much about any of the other ones work, but what I'm gonna focus on in this talk is specifically birds because there's only so much time in the day. And specifically I'm gonna try to understand by looking at birds and taking reference to some of studies that have been done in other species, which traits that these retinas represent, like the sort of fundamental functions that these retinas can achieve. What are these are ancestral? What are really the core principles of how these retinas work and which principles that we perhaps know from mammals are not actually the original state, but something that's derived, right? So derived isn't necessarily a bad thing. It just means it's changed from the original purpose. And hopefully I can offer some insights to that by the end of this talk. So here's a slightly simplified representation of the same thing I've just been showing you. And basically if you look at the research landscape, the retinas of mammals, especially of mice are quite well understood. One might argue that they're perhaps the best well understood complex circuit in the vertebrates. That's closely followed, I would argue by some primate retinas including human. And then there's a big gap. Then we basically know very much less about pretty much any animal. And then there's this little beacon of hope which is the zebrafish, where it happens that a few people work on zebrafish. So we've got some understanding of how that thing works. And actually the more you look into the zebrafish, you more find principles that we long thought are the case based on mammalian work. They don't really apply in the same way to the zebrafish. So from there, then we might ask the question, okay, so we've got some dissonances here. So how can we resolve them? Well, the way to resolve them is by looking at more animals. Of course, we can look at a second fish that would give some information or we can look at something that's a lot more distant or different or something that no one has looked at at much depth yet. And the really interesting candidates in this context I think are the ones in the middle here. So we've got the amphibians in the reptiles and what I'm highlighting here is photoreceptor types that they have. And the talk's gonna keep homing in on that because I guess I should have pointed out that the bony fish retain what we call the ancestral state. So they have quite a lot of photoreceptors whereas mammals have lost a bunch during the age of the dinosaurs. But all of these animals haven't, right? So the reptiles still have them, the amphibians have them, the birds still have them and actually some of them have invented new photoreceptors in the process. So the amphibians, many amphibians anyway, have a second rod, a blue rod. So the idea is that they can see color in the dark when it's not very bright anyway. And the reptiles and birds they have invented this thing called the double cone which may or may not be a duplication event from the red single cone that is under debate but what is not under debate is that they have it and that there's loads of them and that they're important for vision. And the problem is we don't really know which bit of vision they're important for but there's loads of them so clearly they're important. So what I'm going to talk about is our efforts in trying to understand how this retina works here and just to give a little bit motivation other than the evolutionary angle of why we want to understand birds. I think they're quite fascinating, right? If you think about what's the vertebrate with the best eye, you know, people are going to tell you what eagle, so falcons, you know, reptorial birds, it's always the birds. And then if you think about, okay, what's the animal that has the best color resolution as far as we understand it? It's again, the birds. So because they've got so many photoreceptors and clever tricks that add on top of that. So in a way, you might argue that birds have the best eyes, whatever best means, but in terms of high performance, they have a bunch of extra features that are quite interesting, right? So one is this high complexity thing that I talked about. So birds take sort of to the extreme what we start seeing in fish that we have large numbers of very small cells squished into the retina as opposed to small numbers of very large cells, which we have in some other systems. So clearly all these cells, they must be doing something. Birds in particular are quite unique in that they have the highest eye to grind ratio, right? So if you think about the size of a pigeon, for example, pigeon head, it's not very big, right? The eyes are actually almost as big as the brain in that animal. If you think about something like an owl, they actually have elongated eyes that can't rotate against each other anymore because they elongated and they hit each other in the skull, right? So therefore, owls need to move their head rather than their eyes for looking around, which is why they're so famously good at rotating their necks. So if you think about birds in this context, they really invest a huge fraction of their neurons that they have in their heads into the eyes rather than into their brains. So there must be something about this. A while ago, we did a little survey of animal eye complexity. And here eye complexity, I really just take a very simple measure. We just count ganglion cells in the eye and there's a lot of published studies on the anatomy of different animals, ganglion cell numbers and the size of the eye, of course. So what you would expect is that an animal that has the same density of ganglion cells, they should fall onto one of those eyes or lines here, right? And if you go further out towards this end, then you're a complex animal. And if you're further in this end, then you're a simple animal, right? So if something like a hippo has a very big eye, doesn't have so many cells, that counts as a low complexity system in that case. Where something like a hummingbird would be on the extreme edge because it has a very small eye and tiny, tiny neurons, but many of them. As you can do some fun comparisons, right? So you can, for example, you can go vertical here. You can, for example, see that the hummingbird has pretty much the same size eye as a salamander, but the number of neurons that it has is about, well, 100 times for each one neuron that the salamander has, the hummingbird has 100 in the eye, right? Zebra fish is somehow in between. The mouse is towards the low end. So there's a lot of interesting things to be said here, but one thing that stands out, I think, quite clearly, is the birds, which are the red dots. They're all in the high complexity end and towards the big eye end as well, right? So even small birds have big eyes, relatively speaking. What we can do with this data is we can sort of flip it along these isodensity lines. So now I'm just plotting the density rather than number and size. And these are the same data points just moved around. And then I've taken the liberty of just taking a histogram across all of these animals here and you get two humps, basically. And if you look at what's in the humps, on the high density, high complexity end, you've got basically most things, birds, reptiles, amphibians and fish. And on the low hump, we've got us and the sharks and the jawless fish. It's a bit sad, but that's where we are. So if we now think about, okay, so we've been studying this mammalian retina for a long time and we've got some understanding of how some of these mammalian retinas work. But if they represent the low complexity state, then we're kind of missing a large slice of the pie of potential understanding of how retinas can work. So and birds, I would argue are on the extreme end of this. So I think it's definitely worthwhile looking at them. One more reason to look at them, birds have oil droplets. They're not only pretty, but they're also kind of functionally important or at least we think they are. So these are little oil droplets that sit in the photoreceptors just under the outer segment, which is the light sensitive bit. And they're basically colored sunglasses. So they help spectral filtering of the signal that gets picked up by the photoreceptors in the first place. And one of the key interpretations of the need for these things is that they improve color vision. They basically narrow down the spectral response of each photoreceptor, which means you get a better resolution when you want to compare wavelength, which is one of the reasons why we think birds have such good color vision. Another reason for oil droplets is that they can help bundle the light. So basically you improve sensitivity. Then they have the extra cones. I call them two extra cones. They're called double cone. It's basically two red cones that are glued together. There's loads of them and we really don't know what they do, but they're there. So that must be important. And birds have foveas and some birds have two foveas. And I'm just picking up one here from, I think it's a skylock, but it doesn't really matter what it is. The point is to have a second fovea, some birds do, which supposedly is even sharper than the regular fovea that we're familiar with in our own eye, because there's an optical effect where the light coming in breaks and bends outwards. So you basically have a little magnification effect. So the idea is that that is one of the reasons that they have such high spatial acuity. At least in the fovea. Okay. And of course, birds are just cool, right? They do all these things that we don't do. They fly and they have, they're lateralized. So, you know, if you look at a parrot, the left and the right eye are not the same. And also the behavior, like the parents are right handed or footed or left footed, right? These kinds of things that go all the way into the way that these eyes are organized. So we can compare left eye and the right eye and probably find that they don't work exactly the same way. Those kind of features, we're not aware of mammals doing that. So we can't study it there. There's a lot of sexual demorphisms. So we expect the eyes of the male and the female birds to be different, mating displaces, seasonality, all of these things sort of play into the complexity that we might be able to study by looking at birds. So then the big question is which bird? And he has a bunch of very nice and colorful birds. And I am pleased to inform you that we're working on this one. Because there's many reasons. One is, it's the easiest to get. But there's a bunch of other reasons. So one key reason is that we have, even though it's not complete, we've got a partial connectome or contactome or whatever you wanna call it. Basically from this lovely study here, they did an EM serial section stack through a bunch of chicken retina and reconstructed all of the photoreceptors and bipolar cells that were within their volume. And what that gives you, and I think this is a very important thing, is it gives you a lookup table of which cone feeds into what kind of circuits that's downstream. And as I'm gonna hopefully come back to at the end of this talk, I think it's really quite fundamental to understand that particular step because it seems to, as far as we can see, link very much to how retinas work. If you've got a lot of photoreceptors, they do one thing. If you don't have so many to do another thing. So being able to look those things up is interesting. And from collected studies over the years, there is also anatomical evidence for what evidence. There's an anatomical atlas of sorts that we can use to look at what cells, where cells project, what they look like. So for the retina aficionados, for example, we've got ganglion cells that look like this, right? Stratifying in what is this five different layers. If you look at mouse cells, they tend to go at most into two layers. So how this thing works, who knows. We've got stratification already in the outer retina, right? So the photoreceptors don't will go to the same layer. They go to different layers and the bipolar cells use that to pick their targets. So there's a lot of complexity that we just don't have in other species that we get to study here. In the chicken specifically, we have another wonderful thing, which is a transcriptomic atlas, which was released by the Sains lab a couple of years ago. So we have some sort of overview of the numbers of cells that we are expecting in terms of type diversity and that sort of thing. And if any bird is gonna end up being genetically modified in a stable and useful way, probably chicken is gonna be the first. They have been modified to some extent, not very useful for our studies here. This is just GFP, for example, but this is the direction that the field is going. So once those transgenic birds hit the world stage, you wanna be ready for them basically. So, but the big problem with all bird retinas is there's really just no functional data at all that you can look at other than you can look at behavior and there's behavioral studies, of course, and they show that birds are good at seeing things, okay? And then there's electroretinograms, which is a non-invasive way of understanding roughly the sort of stuff that goes into the eye in the first place. Or what the photoreceptors might be doing to some extent by polar cells. So you really, you're lacking this big in between computational finesse thing that we wanna understand there's just no data on it. And the reason there's no data on it is, well, A, not many people have tried to get data, but even the ones that have tried have mostly failed. And the reason is it's hard. So bird retinas have this really annoying tendency of generating these pathological waves of this depression. Basically, you can take a bird retina, open up the eye and just poke it. And you can see with your naked eye how there's a wave propagating away from the place where you've pricked it. And all of the tissue that's left behind is left in a weirdly depressed stage. It doesn't want to respond to light anymore for quite a long time, it's all a bit broken. So that happens in birds, which means that if you take that retina and you dissect it and you put it in your recording rig, hoping to record something, chances are you've already killed it and that's not good. So we've been playing for a long time on this problem and we've hopefully solved it, we think we've solved it. So basically by a combination of being very particular about how we dissect and actually bathing the tissue in a basically stopping synaptic transmission during dissections so that these waves can't spread. And then we're unblocking that by washing out the drug afterwards when we have it in the chamber. And the chamber specifically here is a multi-electrode array. So this is the thing you're seeing here. All the stuff you see on top of it is just visual stimulation. And when we put a bit of bird retina here with our protocols, we can get signals that look like this. And every time it flashes, that's the retina population activity responding to a flash of light basically. And the reason that you see the structure is because the retina isn't 100% flat, it attaches better in some places than another. But bottom line is we can record electrically driven activity from a bunch of cells in this way. This is all the work that I'm gonna present now is all done by Marvin Seifert who has just finished his PhD. He's of course very excellent at all of these things. And the data I'm gonna show you if you want to look at more detail, it's on the bio archive here. So what did Marvin do? So the first thing we decided is there's a lot of things he can do once he's got recordings from this retina for the first time, right? So we wanted to just basically generate a fast basic description of the sort of stuff that's in design. So we're recording from these retina gang cells using a multi-electrode array. And for simplicity, all we're doing is full field flashes of light. So there's no spatial aspect to any of these simulations yet. So for what's in this paper anyway. And because we know that birds have very good color vision and here, for example, at the absorption spectra that we expect to see in the cones if we ignore the oil droplets, we figured if we place the LEDs like this then we should be able to cover the whole space quite reasonably. Of course, the nice thing is with having an LEDs you get to flicker it really quite fast if you want to. So what we can do is we can probe color processing and we can probe time processing but we can't really probe space processing like this. So we slept the retina on there, we get responses. This might, this is what one electrode might look like but of course, this is 4,095 here. So you need spike sorting, we do that. And then eventually we can compute these things. And this is just the sanity check really that what we've done is sensible. So what this is is from a single spike sorted cells, a back correlated electrical image. So every time the cells spikes, we check what all of the other electrodes on the array are doing and we basically reverse correlate that. And what that generates is, well, what you see is what this generates. So it's basically the action potential flying along the axon here. So it gets generated here at the hillock and cell body and then it sort of flies along the axon and it stops because we stopped computing it, right? So this is showing three milliseconds in time here, the distance is about 2.5 millimeters across the whole thing. So from this, you get to compute the speed in all of these things. But the key thing is that it looks like a cell firing. And this is our justification that we're recording and spike sorting in a sensible way, the ganglion cells of this bird. This is one, here's another one. This one is a bit different. It fires like it propagates a bit quicker. And what's actually kind of nice is that you can see the nodes of run via here. So this is a malinated cell and you can see the action potential hopping. This was something that was slightly before as shown already by the guys from Oldenburg. They've quantified this really nicely. This is sort of a bird specific thing that you can see these malinated fibers in the eye. And one of the things that achieves is it makes the eye a bit quicker. Not in terms of responding to light, but once it has responded can tell the brain a bit quicker. So that seems useful. Here's a third neuron just to show you that you get all kinds of varieties, right? You get slow neurons, fast neurons, intermediate neurons, big neurons, small neurons, all kinds of stuff. And what we don't really see is neurons that look weird, that sort of have two axons that sort of thing which would be spike sorting errors probably. So having said all of this, what do they actually do? So here is one cell. So this is just to give you a little gentle introduction of the sort of data that we get. So for example, here in the bottom left you see in a very simple experiment where we take all of the LEDs together which makes white roughly. We switch them on and then we switch them off for two seconds each. And we do that a few times. So these are the repeats, five repeats here. And every time we do this, as you can see, the cell likes to fire here during the off transition and it doesn't do much else. So this is transient off cell, okay? Then we do the same experiment here for different wavelengths at a time. So different LEDs at a time. So if you flash the red light, it looks quite similar to what you do with the white light. If you go yellow, green, blue, eventually it disappears. So clearly this is a red biased cell. It responds better to red than it responds to blue, for example. Then we can take, we go back to white light and we flicker the light slow initially and then we go faster and faster and faster until 30 Hertz here in this case which is the fastest that our projector can do. And this particular cell as it so happens doesn't like the slow flicker but it likes the quick flicker. So it gives out this band pass tuning. And then the fourth thing we can do is we can compute these, we call them spectral kernels. It's basically if you flicker all of the LEDs at the same time, but independently at a pseudo random sequence and you see when the spikes happen, then you can later reverse correlate what the stimulus was that drove the spike on average. So this is basically the average stimulus that drives the spike in the cell. So the way to read it is the light goes on, the light goes off and then it stops for a bit and then 70 milliseconds later that cell fires. So this is a red off cell in this case which is consistent with what we see here. Here's a second cell, completely different as you might expect. So this is obviously it fires a lot when you switch on the light, but then it actually, and you can see this more clearly here in the colors, it gives you a second peak when you switch off the light. So you get the sort of sluggish on, oops, sluggish on and then a sharp off transition. And if you look at the relative ratio between these two, they seem to vary in a color dependent ways. For example, by the time you're in the deep blues, the off seems to have disappeared. The response all over this chirp stimulus and then it gives you a color opponent kernel. Color opponent in this case means that on average, you drove a spike by decreasing red and blue light, but increasing the turquoise and the green light, right? So the classical interpretation of this kind of kernel would be that this is a color opponent cell. Therefore, its purpose is color processing. Whereas this one, the one in below here is a non-opponent cell and therefore its purpose is not color processing. It's gonna do something about grayscale coding. Here's another cell, just to show the diversity. This is also an on off cell, as you can see here, but it's much more transient than this one. And it doesn't like the fast flicker in this case. And one last cell, here's an on cell, is very transient on, so in that sense, it's very similar to this off cell, but what's quite different to it in this case is that it's extremely selective to wavelength, much more selective than this one. It doesn't like flicker at all, right? So you wouldn't know you have a cell if you only played a flicker and it doesn't give a kernel, which is probably related to the fact that it doesn't like the flicker because this is also a form of flicker, right? So this is a ridiculously sensitive cell, right? Unless you give it a red LED flash, basically it doesn't want to respond, okay? Very selective, okay? So we can cluster this, we get a bunch of clusters, and what's actually quite striking is that, this was done by Paul here, is you can roughly group everything into three things. You've got a very small number of off cells, which respond very nicely. Then you've got a huge number of on off cells, which respond in all kinds of complicated ways. And then you've got quite a large number of on cells, which respond in an interesting way during the steps, but they respond terribly to anything related to flicker. So that's weird, right? So if you did this experiment with a mouse, that's not what you would get. You would get much more balanced in all of these aspects because that's how we think the mouse works. So here's just a sort of summary, which is a bit more readable of the key clusters that we get, right? So what I'm showing here is this cluster one, and this is always sorted now from off to on. You get this cluster one, which is this transient off thing, which responds to high frequency flicker. It's about 5% of cells. Then you get 2 thirds of all cells do something like this. They respond on off. They do usually different things with different wavelengths. But it's never that it sort of blew on right off. It's always a mix of things that comes together to end up giving you a fairly complicated response here. And then when you look at the kernels, it always gives you something color opponent. They usually do respond to the flicker, but that usually prefer the slow flicker. And then in the ons, you get two types of two general features that you see. You get these ridiculously over selective things that only respond to one thing. I showed you a red one earlier. Here's a blue one. It only responds to blue light and nothing else. Or you get these fairly unselective sluggish on cells as a handful of those as well. Okay. Why do I spend so much time on that? Well, the reason is that this is completely different from mammals. So in mammals, the lovely thing is because we've worked on mammals for so long, we basically know what's in there more or less. And so there's this very nice paper here from the Schwartz lab last year, which made a heroic effort of basically summarizing what all of the ganglion cells in the mouse do. And you can go into that paper and download some of the key punchlines. For example, they also played white steps like this and just quantify them in terms of are these off cells? So it's the polarity minus one or are these on cells polarity one or are these on off cells, which will be polarity zero, right? And as you can quite clearly see the mouse has a bunch of off cells here and it has a bunch of on cells and a small number of on off cells. That's sort of how the mouse is organized. Now, if we do the same exact computation for the chicken, you can see that chicken fits into the hole here, right? Is diametrically opposite. You've got no cell that's perfectly off. Even this off cell has a little on hump here. You've got no cell that's perfectly on. Oops, I'm going backwards. Okay. And the vast majority of cells sit somehow in the middle as you can see from these on-offs. So clearly the way that this written output is organized in this one is just not the same. So let's compare this a little bit broader. So what we have plotted here now is the same data. So again, polarity off to on, but I've added transients. So is this a quick cell that would be one or is this a really slow cell that would be minus one? And what you can hopefully see is that these clusters, each of these blobs is one of the clusters that I've shown you. They follow this trend that the offs tend to be the quick ones and the ons tend to be the slow ones and the on-offs tend to hover in the middle. And that's actually a very strong correlation which becomes a very significant. If you do the same exercise for the mouse, you get no trend whatsoever. And that is actually one of the defining characteristics of how we think retina works is that it doesn't give us this correlated response. It gives us a de-correlated response that you've got fast on-cells and slow on-cells and you've got fast off-cells and you've got fast on-cells. Well, you've got all of the combinations possible such that the brain gets this nice pre-digested version of everything in a different channel whereas the brain, the bird brain clearly doesn't get a picture like this, right? And just to make some comparisons, we've earlier recorded some data in zebrafish and I just want to point out that zebrafish kind of do the chicken thing. They don't do it quite as extreme but they also seem to have this correlation from fast off to bit slower on. And actually, I'm not showing this here but human data looks like the mouse. So there's a difference there. Next big difference is that if we look at not just polarity against transients but polarity against whether or not the cells are color-opponent, so supposedly informative about wavelengths, color, vision, that sort of thing, we find that basically all of these on-off cells give a high-opponency index so they're important for color and everybody else gives you a low-opponency index. That's because either they don't respond or they give you a non-opponent kernel, okay? So we're supposed to learn then that you've got this on-off trend and transients this trend here but then the ones in the middle are the ones that give us color vision but that would be really weird because can this chicken really afford to send two-thirds of his retinal output into the color brain and then only one-third into the let's see the rest of the world brain? It doesn't chime. You would expect that most cells are informative about the grayscale stuff in the world because that's the most important thing to see and then some of the cells go into the day-in-form world color. That's how the primate eye seems to be organized whereas this eye isn't organized in this way. I'm gonna come back to that point. So the structure that I'm not gonna adopt and I need to check my time. Oh, I'm already quite late, aren't I? Okay, I'm gonna try to be quite quick. So I'm gonna talk about the off-cells briefly. I'm gonna talk about the on-cells and then I'm gonna try to explain to you that the on-offs do both the things that the offs and the ons do. So they're basically a multiplexer. So if we look at the CHERP response and these are representative stimuli clusters from off to on and the on-offs in the middle, the offs as discussed are quick. The middle ones, they do all kinds of stuff that can be quick, but most of them are sort of biased towards the slow end and then the on-cells that do, well, they don't respond basically. You can see this is uncorrelated activity. So we can take it for year transform of these and then basically quantify them as high and low frequency preferring or nothing. And if you plot this for year spectrum for all of the clusters sorted from off to on, you can see that the trend, it goes from fast to slow to nothing. Yeah, so a very obvious trend. So clearly the offs are for fast stuff and they're the only cells in the eye that can do that can follow the fast flicker, right? So if the chicken needs to be fast, it has to use its off channel really. This is just to show that the face looks quite nicely, right? So at all of the frequencies up to 20 Hertz in this case here, you get nice face looking and we can quantify the face looking and what's really quite interesting about this. So this is just the vector strength. So how well do they follow? And if the line is high, that means they follow well and this is against frequency. You can see this off cluster does it really well. The C2, which is the next best one, does it kind of well and then the rest ones, they don't do it really well at all, they're probably in the background. But what's kind of funky about the whole thing is that if we look at these fast clusters during fast stimulation, then their face looks to the off, meaning they do what they're supposed to do, right? You've got an off cell, you flicker the light and it responds every time the light switches off. But when you look at these cells or any other cell for the slow frequency stuff, they're actually face locked to the on. So even the off cells, which are supposedly fast, actually do slow on stuff, if you give them slow flicker, right? So they're basically the switch polarity depending on the frequency. What's also kind of fun is, given that we have these electrical images, what we can do is we can compute the speed, the conduction speed. And if you expect that the off cells do the fast stuff in the eye, they should also have the fast axons. And that seems to be the case. It's not a huge trend, but here's the off clusters. And then if you go to on off and then the different types of on clusters, you can see there's a bit of a trend there. So it seems that these fast off cells are also the ones that have the fastest axons. Okay, so what else can we say about the off cells? Well, what I can say about them is that they're really, really boring. So if you give them a different contrast steps and you just see how well they respond, they're basically linear. So that's nice if you want to inform about achromatic information in the world, but it's not very interesting. If you take the off fraction of the on off clusters, so because they respond to on and to off, you can analyze them as both. If you take the off fraction of the on off clusters, they do the same thing. If you look at the wavelength selectivity of these off clusters, they basically follow the red option, which sort of suggests very strongly that they're driven by one of the red photoreceptors. And remember in the bird, we've got the standard red and that weird red that we don't understand. So it could maybe be driven by that. That would be interesting. And this is also true for the off fraction of the on off clusters. So it seems that all of the off signals, whether in the off cluster or in the on off cluster, but the off bit, they all kind of do the same thing, right? They're basically fast homogeneous linear achromatic, right? So just think of them as a fast boring system. They basically tell you quickly what's going on without giving you much detail. Okay, let's look at the ons. The ons are completely the other way around. So here's one of the on clusters. If you look at how it responds to different wavelengths of light, this is a fairly broadly tuned cell. And we've plotted that here, right? So we can't really explain this tuning function with one option, we need more. So it sort of implies that it pools more than one photoreceptor signal into that cell. But that's not actually the interesting one. The interesting one is this one here. It's the one that responds to nothing except for blue light. And if you plot the tuning function, it's difficult to explain that with just the option. There's multiple versions of this. So this is not just one cluster, there's two clusters to do this. And there's the red version of the same thing, right? So these, and what's really important here is that these are narrower than an option, right? So we've got a system that is non-opponent, right? So it's not a color vision system in the classical sense. However, the spectral sensitivity function is, you can't explain it without having the options talk to each other in the circuit, right? So really what this is, is this is what cortex people might call a hue cell, right? One that is one step past the potency. So you go color opponent, and then you clip half the oponency. You clip basically below zero, and then what's left above zero is narrower than the original input. This is how we think color processing works in the in the primate brain, for example. It seems that at least some of the cells in the chicken do this in the ganglion cells already. Yeah, here's the primate reference. So here's this classic paper from 1980 where this guy, Zachy, showed that when you record in cortex primate, in the cortex of a primate in color areas, you get tuning functions that are basically narrower than absence and they coined this idea of hue coding. And we're seeing them in the chicken here. Yeah, this is just the explanation of how that would work. So of course we can describe this color stuff, but it's a little bit more useful to put it into an actual quantification. So what we've done here is we've quantified what I call color dominance. So that means if you can drive the cell better with a stimulus that is not white. So if you flesh a blue light, it responds well. If you flesh white light, it responds badly. That means it's a color dominant cell. It's easier to drive it with color than with white. And tuning sharpness fairly self-explanatory. If they're narrow, it'll be here. If they're broad, they're here. And what we can see is that the clusters, they sort of form an axis. You get these incredibly selective ones and you get these incredibly non-selective ones, whereas the on-off cells and the off-cells here, they're sort of, they're sitting in the middle. They don't do anything interesting with color. And just for comparison, this is what happens when you do the same exercise with the offs. Nothing interesting going on if anything goes the wrong axis. So that kind of means to summarize that the on-cells carry all the color information. The off-cells don't do anything interesting with color. The on-cells do everything interesting that's with color. So it's almost saying, okay, if I want to inform the world about grayscale stuff and about color stuff, I'm just gonna use my off-system for the grayscale and the on-system for the color. And this is just the tuning functions we get. So where that really becomes interesting, I think, is when we now consider the on-off, right? So if the off had these properties, the on-have these properties, what are the properties of the on-off system? And long story short, I'm gonna try to convince you that they do both at the same time in such a way as to save on the optic, so basically save wave bandwidth on the optic nerve, basically sending two messages at the same time. So the idea is this classic multiplexing engineers have been using this for a long time. You basically have a bunch of fairly simple messages on one end, but your cable isn't thick enough to put them all in parallel. So you combine them in some sort of clever way at a multiplexer end. And then at the other end of that cable, you de-multiplex them and you recover your original signals. So if we think of the I as the multiplexer and the brain as the de-multiplexer, then maybe the optic nerve carries these weirdly compressed messages. Wouldn't that be nice? So one evidence in that direction is if we look at the kernels that we see in the color upon themselves, they tend to be opponent, as you can see here, but before the spike, they tend to converge to a non-opponent bit. So there's always an opponent fraction and a non-opponent. So it's like color and not color at the same time, just separated by time. And we get that in a time compressed version or we get this in a time dilated version in different clusters, right? So we can throw that all together and basically just, so if we time align them, basically stretch them all to be one time unit, then on average, this is always the case. We have a long time period that is opponent and then just before the spike, it goes non-opponent, which means that in principle, they carry both the color and the non-color information if you can decode it on the other end. And just the second way of looking at the same problem, this is the step responses to all of the colors here indicated in the different colors and to the white step, which is the black line for on and off. And hopefully what you can see is that if you do this for off, they all have the same shape. So no matter which wavelength you use, you basically get the same shape and the amplitude varies as a function of which of the ones you've picked. Meaning that you can't really look at the response and say what the color was unless you knew what the intensity was. So this is not informative about color. Whereas here, for example, the red one is nice and transient, the blue one is nice and sustained. So you could look at the temporal envelope of this response and have some sort of opinion about what the color is. So to do this slightly more principled way, you can do a principle component analysis and I'm just plotting the first two principle components for each of these versions and then plot them as one does into this nice two-dimensional space, PC1, PC2. And what you can hopefully appreciate here is that all the white responses of different intensity, they just follow PC1. Whereas the color responses, they have a tendency of sitting at a fixed point in PC1 and following PC2. So basically the amplitude of these responses is informative about intensity, but the time course is informative about the wavelengths. So you basically have both kinds of information in these. That completely fails in your off responses. It doesn't work at all. Okay, and this is just to show that you can do this for intermediate cluster. So let's skip that. Okay, so then you on off to both, maybe, hopefully. That's our working theory anyway. So what does that really mean? And I'm just about finished. So don't you worry. So if we now jump back at this philogenetic tree or a variant of it that's shown from the beginning, we know that different lineages have different photoreceptor complements. And as I pointed out in the beginning, the ones in the middle here highlighted in yellow, they're the ones that have retained all of the original stuff and sometimes invented new stuff. For example, this double cone. Whereas the mammals have specifically lost the middle to here and something similar happened also in the cartilaginous fish. So sharks and rays and skates, they've lost different ones, but still have lost some. And lump rays, let's ignore them for a while. They're a little bit outside of this argument here. Then basically what you can see is that stuff, complicated stuff sort of gradually evolved over the early vertebrate lineages. For example, we're starting to see proper rods, which we don't see in the lump rays yet. And then develop these things, the oil droplets, they're actually quite old. And all of these species, there's representatives that have them. And then we invent these new photoreceptors. So we're building complexity at the input level of these retinas. And then along come the dinosaurs. The mammals get pushed into nocturnal leash for hundreds of millions of years. And many of these amazing things that have been evolved weren't very useful at the time. So they lost them, right? So the mammals lost the oil droplets and they lost two of their four cones. And as a consequence, they have now basically a spring cleaning. This is a spring cleaning event in the retina. It allows you to clean up some of the weird hangovers that must have accumulated as this complexity evolved over time. So if we think of it that way, then if we just sort of compare the overall characteristics of these retinas, maybe we can see some patterns. So one of the things you might think about, for example, is how do these retinas work for on and off, right? So for example, it's on and off well segregated as it is in mammals that gets an arrow up, or is it poorly segregated like in chicken, then it gets an arrow down. And we can actually look into existing literature and it looks like these are all down arrows here for these other species. When I said question markers, I couldn't find literature to support either way. Or are on and off balanced, right? Do you have the same number roughly of on and off cells? Or do you have like mostly on cells or mostly on cells, off cells? Again, the mouse and most mammals do it in a balanced way and the chicken is distinctly non-balanced. The fish is definitely not balanced. So it looks there's a trend. Okay, let's look at Alastair kinetic strategy. So do you have equal number of past and slow cells? Mouse has it, chicken doesn't, and again, or is polarity and kinetics correlated? Definitely, well, D correlated in this case. So the mouse, the D correlated, so gets an arrow up and these guys get an arrow down. So again, there's a lot of stuff that's just different but we have all these question marks but maybe we can fill in some of these question marks and other things. For example, let's think about color processing. All of these animals here have a lot of photoreceptors and we think they go that color processing. So they get an arrow up. For example, in terms of the number of color opponent responses in the NEI, mice not so many. Number of photoreceptors clearly it's just an anatomical marker that's low here and high for all of these. And actually now we can start integrating these guys because we know that in the sharks is not there either. Anatomical complexity, that's the thing that I started off with. Are these complicated retinas or are these simple retinas, right? Anatomically speaking, mammalian retinas are simple. All of these are really quite dense. So complicated and then if you go back down here they're simple again. So the pictures that sort of dawns at least on me is that we have this spring cleaning event here in the mammals and perhaps also in the sharks, who knows that basically allowed them to throw away a lot of the stuff that got evolved in their fuller genetic history that was useful for possibly quite specific things that animals since have taken and co-opted and turned into new visual abilities. But that means that the retina needs to work in a certain way. And the mice have thrown that away. Well, the mammals have thrown that away and basically said, okay, we're gonna get rid of this we're gonna clean up and we're gonna make our retina efficient. And that's the picture that we're getting now. We know that mammalian retinas are very efficient in terms of all of the, in terms of the way that they distribute their resources. We don't know at this time whether the chicken is efficient. Maybe it is, but certainly the mammals are. So this is sort of the direction that we're starting to think along. And just to close, we've been starting to fill in some of these question marks here just to see what that looks like. So we've got a project in the lab looking at small spotted cat sharks, again with the multi-electroderay. And here's just some examples. This is by George, PhD student in the lab. And the first thing you can see already is that we have a good number of cells or they're labeled the wrong way around. I'm sorry, these are off cells, these are on cells and these are on-offs. So not so many on-off cells, quite a lot of clean on-offs, right? So it's a bit mouse-like, right? You get this and you get this and you don't get so much in the middle. So there's a bunch of other things that we've seen but basically we can start filling in the shark arrows and they almost perfectly match this pattern with the exception of this one. There seems to be still a correlated strategy between on-off and fast and slow in the shark as far as we currently see. We don't know what the lampreys do. We haven't got this data, but maybe one day we will. So it sort of emerges that here we've got our little mouse shark that twice in evolution, this kind of spring cleaning has happened and that is how I'm going to close. So I just want to thank everyone involved and the funders and you for your attention. And yes, I've been too long again, like I always do, I apologize. Thank you, Thomas, for this wonderful talk. We learned so much in one hour and I will let people asking questions but maybe just a quick general question first about. I'm surprised that why do you think the nature never really exploited the nearer flow rate part of the spectrum to for visual processing in any species? There's a physical limit. So this is a really interesting question that people have been looking at a lot. And basically it's the thermal noise that you get in an opsin, right? So the photon hits your opsin, you get this photoisomerization event. That triggers your cascade, your amplifier and you get a response, okay? If you are a short wavelength system for UV, then that is an incredibly clean thing to do because you've got high energy UV photons, they hit that opsin. And because of the high energy involved, you don't need to set up your opsin to be particularly responsive. You can make it very reserved basically, right? It only responds if there's actually a photon. Whereas if you go to the other end to the long wavelength, the low energy end, you start, you're forced to put your opsin into such a state that it's responsive, not just the photon coming in, but to basically heat. And this is what we in mammals, for example, call the dark noise, right? So if you go to a perfectly dark room, you still have photoreceptor activity. It's that the rods are sort of trickling away. They're being thermally driven, which is why it's so difficult to see a single photon. You can see a single photon statistically, not individually, right? And there are some animals that have pushed this a little bit. So you can, because this is a thermal noise problem, if you are an animal that's warm-blooded, it's not helpful, right? You want to be cold-blooded and you want to live in cold water, ideally. So you're externally cooled. So there's some species of fish, for example, that go a lot further into the red than we can. But we're talking 20, 30 nanometers, not like deep infrared. And actually animals shift that around that particular sensitivity peak of the red opsin, even over the lifetimes, for example, salmon as they go up and down the rivers for the spawning. The water temperature is different and they will change the spectral sensitivity of the red system over that period to match, to basically be able to push into that infrared range if they can. Okay, thank you, it's very interesting. So we have questions for Elvir and Dan Serju. Okay, thank you very much for all of these. So exciting and impressive talks, thank you. And I just was wondering where the human retina fits in your big picture. Yeah, it's like the mouse. It's the extreme of the mouse, I would argue. So one thing that was quite surprising when the transcriptomic analysis of the chicken came out two years ago is that the chicken has roughly the same number of ganglion cells in terms of type diversity as the mouse, right? I would have expected it to be higher. Well, I guess that's what it is. But one thing that's never really changed and maybe people watching can correct me on this is that the primates are nowhere near the number of mouse ganglion cells, rats and primates, my current number is maybe 20 odd that I've got in my head. Because what the primates have done, of course, is they've gone for this fixated, foveated strategy. So you invest hugely into the ability of seeing extremely high resolution, especially. The only way to be able to do that in a sensible way is to always fixate on stuff that you wanna see, right? To do that, you need a cell that is high resolution. That's why we've got midgets. But the second you've got midgets is you basically, that's half your retinal real estate, right? What 40% of human ganglion cells are midget cells-ish, I think. So you're basically pushing away all these extra functions. And then you're supplementing them with parasol cells, which are actually not all that big, they're quite small as well, but they're really quick, right? So now you've got this ridiculously high resolution spatial vision system, and you've got a fast system. You've got the on and the off versions. They tell you about color because of the way that the human retina is organized, right? So you basically have most of the stuff that you need, maybe, in four cells, right? So this is sort of the extreme human strategy. And of course the human, the primate retina in general has other cell types that do other things. It's just in terms of their numbers, they're much more subdued. And I mean, people have argued that one reason we get away with this is because we've got this big, amazing brain that can take these really general inputs from things like midget and parasol cells and turn it into a perception of the world. But maybe if you don't have that luxury, then you need to process much more deeply in the eye before you can send things to the brain. I'm sure there's exceptions to this, but I think that's sort of the general direction that makes sense to think in. Thanks. Thank you, Sergio, and then, Stefano. Thanks, Tom, for the presentation. It's been really a blast and I learned a lot today. So I have actually two questions and sort of you already started introducing the answer to the first one of them, because the first one is technical. So when you are analyzing these on-off cells, do you also regard the location in the retina? Because yes, like primates, we do have phobia, but also birds have like a region that it's like sort of adapted to be a phobia. So do you find that maybe these on-off, like more versatile cells are more associated with that region, perhaps? Wouldn't that be amazing? Yeah. So we haven't done that for many reasons. So one, this is our first attempt of understanding anything at all in a bird. So you have to go for the simple bit. Going straight for the phobia would be tricky. Chickens don't have a phobia. Chickens have what we call an aerial centralis, which is basically a phobia where the actual phobia bit is missing. So phobia just means that the tissue is pushed away to expose the photoreceptors. So you've got loads of neurons packed very densely. It's just not a phobia. The punishment that you get with a phobia or an arrangement like this on a multi-electrode array is that the retina gets really thick. And that's bad news for trying to extracellularly trying to record from the axons, right? And I think that's also one of the key reasons why there is not so much phobia data even available for primates, because it's really hard to get, even though by this point, we've got a very good understanding of what the peripheral primate retina does, at least from the point of view of the gammon cells. But yeah, no, you're absolutely right. Maybe you've got loads of these on-off cells in the phobia and that's how it works. Yeah, I was just curious because this is actually linked to my second question that it's a little bit more philosophical, so to say. So have you worked with like a paleontologist or like an archaeologist? And the reason I'm saying this is because you're studying sharks and you're studying birds. So when I read the title of your presentation today, the first thing I came to my mind was dinosaurs. So just like to follow up with the thread I'm trying to open, like we know nowadays that, evolutively speaking, birds are very close to dinosaurs, but we don't know so much whether the visual system, it's also related to them. But at the same time, sharks are much like before dinosaurs appeared, like they are more ancient. However, you're showing that sharks are more similar to mouse, which are much more evolved than mammals. So do you think that then, evolutively speaking, birds have evolved more than sharks and we now have like devolved to go to a structure that was like maybe more efficient at the visual level. I mean, there's so many aspects to this. So sharks do have very low slow generation time, right? So they evolve and the genome doesn't reshuffle quite as readily as in other species. So they do evolve slower. Having said that, they've had time. I would expect that what we find in a shark eye today is going to be as close, as optimal as that shark is ever going to get it for the shark visual environment. In much the same way that that's true for probably any animal that we can get our hands on. But in terms of trying to understand what maybe the dinosaurs were thinking, my money would be on the dinosaurs having absolutely excellent vision, probably in no way inferior to what we see in modern birds today, probably better in some cases. Jurassic Park was wrong. Jurassic Park was wrong. I don't know, but I mean, think about this, right? So some dinosaurs have huge eyes, you know, like this. No, like this. So there's no bird that has that because birds, well, most birds fly and the big ones, they have big eyes actually. We should look at Lost Ridge now that I'm thinking of. But I would expect that the circuit architecture of what the dinosaurs had is maybe not all that different to what we see in some birds today. And if you look at the tree, the complexity, so the dinosaurs are where the dinosaurs are here, right? It's fairly new in the ground scheme of things. But that's why, like, looking at this, it looks like we are going back to the point of the manta ray. You see, because... But that would be convergent evolution probably, right? Exactly, because it's something like illogical at the same time, you know. But it's funny, like the way that you represented was like the first thing that caught my attention, like function-wise, we are going back to that level. And structure-wise, we are also going to that level. But you have to work with what you have, right? So if there is... I mean, it's specifically the mid-wavelength cones that I'm missing in particular in the mammals. So if those cones contribute something special to the circuit and that's gone, you can't have it back. You know, now mammals have been living in the day and being extremely reliant on vision for much time, right? If they could have easily re-evolved those circuits that would have done it. So you're actually helping me a lot with these questions, too, because that's what actually was going to ask you. Are you planning to look at retina from nocturnal dirt? Because this may be also something relevant to understand like the efficacy of these on-off cells. You're completely right. But I think the first step would be to look in the chicken with less light, because the chicken also has rods. So the first question would be if we stimulated at a level that only the rods should be able to see, do we even see the same thing? Or is it completely different? It's fascinating. Like today you like... Well, if you can help me with this, I may think about doing some research with birds. Good. So, Stefan, there's no one else for the moment. Can I take the floor? Yes. Well, thank you very much for this lovely talk. Actually, I have two questions. One, in the old days I read that the birds and even mammals have a retinopital system that originates in the brain and goes down to the retina. You didn't mention this system. Is it because it more or less disappeared or it's not particularly important? No. So, I expect that the system is very important in birds. The way that we're doing experiments means we're forced to destroy it because the only way we can get at those cells is by taking out the eye and cutting the nerve. So, it's gone. There's actually a very nice review summarizing what we know about these retinopita connections, at least in terms of which animals have them. And basically all vertebrates have them. It's just the numbering that's different. So, even in the mouse, we know that there's a few fibers going the other way. And there seem to be mostly neuromodulatory in nature and that might be linked to some studies lately coming out showing how stuff changes in vision, in a mouse vision that isn't readily explained by the brain. That must come from the retina, if that makes sense. And I would expect that the birds have this a lot. And I think we need to study it. I'm just not sure how. Well, I haven't read about this system for a long time now. I thought it was forgotten by the community. No, no, it's definitely there. And people are looking at this now also in mouse. Well, my second question is more or less just to have your opinion. The brain reconstruct the visual system. So it doesn't need to have a very complex. Actually, it's better to have a simple input, because you can use these inputs and build up like, let's say edges and et cetera. So I'm not surprised that in mouse, your system as a very, very small, very simple retina, because the method it changes, it's forward to columns and cortex must be very simple so the brain can use it. If it's already a complex image or a complex message, then the brain cannot do anything with it. But we have to think of what came first. The cortex or the eye, right? The eye is much older than the cortex. So I completely agree that one strategy, which evidently works very well, is what we do is sending fairly simple messages to a brain that can deal with that information and reconstruct. But I would also argue that another strategy is to do a lot of pre-processing to the point where you could almost wire your ganglion cells straight to your motor neurons. Maybe that's the extreme view, but like something like this, which then means that you don't need such a big brain to do good vision, to do good visual behaviors, right? So if we think about the very origins of where the vertebrate eye comes from, we're talking Cambrian shallows, right? So basically, animals had just about evolved to be able to move at all, right? Animals used to be stuck to the sea ground and then they started to get some locomotor activities. At the very same time, we see the emergence of eyes and we see the emergence of primitive brains. So those three things come together. And the big idea there is that animals started to be able to move. Therefore, they wanted to eat each other and therefore they needed to see their prey or stop being eaten by seeing your predator. And therefore you need all three things co-evolving. And those animals had primitive technology concede in the fossil record already, but of course, cortex was long way off. So the original visual strategy is a tactile one. Yeah, but once the message is in the brain, it can interact with other sensory inputs like auditory, somatosensory, whatever. Even if it stays at the retinal level, then it is sort of protected from other multi-model connections and influences. Unless you've got those fibers that go back into the eye. But it's true, yes. But I mean fundamentally the tectum, which is the first stage for most Cambrian cells, even in the mouse, we just call it superior colliculus, right? But in this tectum is a multi-model integrator, right? It's not a visual center per se. It's a sensory center, if sensory motor center, even some would argue. So the integration comes early. Thank you very much, sir. So if we don't have any more questions, so we'll leave you. So thank you very much, Thomas. Nice talk. Thank you very much for the invite. I will stop the recording.